Peptide Deformylase Inhibitors as Antibacterial Agents: Identification of VRC3375, a Proline-3-Alkylsuccinyl Hydroxamate Derivative, by Using an Integrated Combinatorial and Medicinal Chemistry Approach

Abstract

Peptide deformylase (PDF), a metallohydrolase essential for bacterial growth, is an attractive target for use in the discovery of novel antibiotics. Focused chelator-based chemical libraries were constructed and screened for inhibition of enzymatic activity, inhibition of Staphylococcus aureus growth, and cytotoxicity. Positive compounds were selected based on the results of all three assays. VRC3375 [N-hydroxy-3-R-butyl-3-(2-S-(tert-butoxycarbonyl)-pyrrolidin-1-ylcarbonyl)propionamide] was identified as having the most favorable properties through an integrated combinatorial and medicinal chemistry effort. This compound is a potent PDF inhibitor with a K_i of 0.24 nM against the Escherichia coli Ni²⁺ enzyme, possesses activity against gram-positive and gram-negative bacterial pathogens, and has a low cytotoxicity. Mechanistic experiments demonstrate that the compound inhibits bacterial growth through PDF inhibition. Pharmacokinetic studies of this drug in mice indicate that VRC3375 is orally bioavailable and rapidly distributed among various tissues. VRC3375 has in vivo activity against S. aureus in a murine septicemia model, with 50% effective doses of 32, 17, and 21 mg/kg of body weight after dosing by intravenous (i.v.), subcutaneous (s.c.), and oral (p.o.) administration, respectively. In murine single-dose toxicity studies, no adverse effects were observed after dosing with more than 400 mg of VRC3375 per kg by i.v., p.o., or s.c. administration. The in vivo efficacy and low toxicity of VRC3375 suggest the potential for developing this class of compounds to be used in future antibacterial drugs.

Although the study of microbial genomes has revealed an abundance of potentially useful targets, so far little has resulted from this much-heralded effort. Instead, most newly disclosed antibacterial agents target known enzymes that were identified through the use of classical biological methods. Witness the selection of posters recently presented at the “New Antimicrobial Agents and New Research Technology” section of the 42nd Interscience Conference on Antimicrobial Agents and Chemotherapy: 18 proteins were discussed as targets for antibacterial drug discovery; all of these were discovered through standard microbiology and biochemistry methods. One such target that has received much attention lately is the bacterial peptide deformylase (PDF) (EC 3.5.1.31).

The protein synthesis processes for bacterial and mammalian cells are very similar. Both utilize the same amino acids and codons and share the same mechanism for elongation. However, a major difference between bacterial protein synthesis and mammalian cytosol protein synthesis is the use of formylmethionine as the initiator (19, 20). Unlike cytosol protein synthesis in mammalian cells, which is initiated with methionine, protein synthesis in bacteria is initiated with N-formylmethionine (2, 16), which is generated through enzymatic transformylation of methionyl-tRNA by formylmethionine tRNA transferase. The N-formyl methionine of the nascent protein in bacteria is removed by the sequential action of PDF and a methionine amino peptidase in order to afford the mature protein (2, 21). This formylation-deformylation cycle is essential for bacterial growth and is conserved among all studied bacterial species. Previous reports indicate that this cycle is not required for mammalian cells (for reviews, see references 7 and 31). The specific bacterial requirement for PDF in protein synthesis provides a rational basis for selectivity, making it an attractive drug discovery target. Although the in vitro development of bypass-resistant mutants in some bacterial organisms casts a shadow on the potential of using PDF inhibitors as antibacterial drugs, in vivo studies suggest that such mutants are much less virulent (for a detailed review, refer to Yuan et al. [31]).

PDF activity was first reported by Adams in 1968 (1). This enzyme was not fully characterized until the early 1990s, when the Escherichia coli deformylase gene, def, was cloned (23) and when PDF was subsequently overproduced in E. coli (22, 23). Bacterial PDF utilizes a Fe²⁺ ion as the catalytic metal ion (8, 25, 26), but the ferrous ion in PDF is very unstable and can be quickly and irreversibly oxidized to the ferric species, resulting in an inactive enzyme (27). However, the ferrous ion can be replaced with a divalent nickel ion in vitro, resulting in much greater enzyme stability with little loss of enzyme activity (8).

Since PDF is a metalloprotease, one possible approach to designing inhibitors consists of having a nonspecific chelating pharmacophore that binds to the catalytic metal ion and is coupled with a second moiety that binds to the active site, thus correctly positioning the chelator and providing the necessary selectivity and physicochemical properties. Such an approach has been successfully applied to the design of inhibitors of many other therapeutically important metalloproteases, the prime example being the angiotensin converting enzyme, followed more recently by the matrix metalloproteases (15). Recently, actinonin, a naturally occurring antibiotic with a hydroxamate moiety and a tripeptide binding domain, was shown to be a potent PDF inhibitor (5). Several three-dimensional studies of PDF-actinonin complexes suggest that the hydroxamate moiety of the inhibitor binds to and chelates the active center metal ion, while the tripeptide domain fits into the S₁′-S₃′ pocket of the enzyme (9).

Using mechanistic information about the reaction catalyzed by PDF, together with an understanding of the general principles of metalloprotease inhibition, others have constructed several chelator-based inhibitor libraries according to the generic PDF inhibitor structure (Fig. 1) (31). In these libraries, X represents a chelating pharmacophore element that can bind to the metal ion at the active center of PDF, the N-butyl group mimics the methionine side chain, and P₂′ and P₃′ are domains of the inhibitor that can provide additional binding energy, selectivity, and favorable pharmacokinetic properties. These libraries were tested for PDF inhibition, antibacterial activity, and cytotoxicity. In this report, we describe the construction of such libraries, explore the inferred structure-activity relationships (SARs) of the resulting compounds, and report the subsequent identification of VRC3375, a unique PDF inhibitor-based broad-spectrum antibacterial agent with oral efficacy.

FIG. 1.

A generic PDF inhibitor structure. In the structure, X corresponds to a pharmacophore element capable of chelating metal ions. X attaches to a 2-substituted hexanoyl, which mimics the transition state of the hydrolysis of formyl-methionine.

MATERIALS AND METHODS

Materials.

Actinonin, formate dehydrogenase, catalase, and NAD⁺ were obtained from Sigma. N-Formyl-methionine-alanine-serine (fMAS) was obtained from Bachem. Hydroxypropyl-β-cyclodextrin was purchased from Aldrich. All other chemicals used were of the highest commercial grade and are available from common vendors.

Preparation of chelator-based chemical libraries.

Three 528-member (22 × 24) chelator-based libraries were prepared on solid phase in a parallel fashion as indicated in Fig. 2A. Twenty-two amines (Fig. 2B) for P₃′ substitution were immobilized through a 5-(4-formyl-3,5-dimethoxyphenoxy)valeric (BAL) aldehyde linker on a polyethylene glycol resin through reductive amination (trimethyl orthoformate-NaBH₃CN). Each amine resin was further divided and coupled to 24 different 9-fluorenylmethoxycarbonyl-protected natural and unnatural amino acids (P₂′) 2-(1-H-9-azobenzyltriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate-diisopropylethylamine (HATU-DIEA), followed by removing the 9-fluorenylmethoxycarbonyl protection group (20% piperidine in dimethylformate [DMF]) to give 528 dipeptides immobilized on solid phase. The 528-member thiol library was prepared by reacting the amine of the 528 dipeptides on resin with 2-acetylsulfanylmethylhexanoic acid followed by acid cleavage from resin (trifluoroacetic acid [TFA]). Alternatively, the dipeptides were coupled with 4-mono-methyl 2-(R)-butylsuccinic ester by using the HATU-DIEA method. The methyl ester was cleaved from resin (TFA) and converted into the corresponding hydroxamate (NH₂OH-dioxane-water) to give a 528-member hydroxamate library. The carboxylate library was prepared by the alkaline hydrolysis (LiOH-THF-H₂O) of the corresponding methyl ester on resin, followed by TFA cleavage of the corresponding carboxylate dipeptide from the resin. The yield of each library member was estimated based on high-pressure liquid chromatography (HPLC) analysis.

FIG. 2.

Synthesis of focused chemical libraries used in this study. (A) All libraries were prepared using solid-phase parallel synthesis as outlined; (B) the building blocks used for preparing these libraries are summarized.

Preparation and purification of selected compounds from hydroxamate library.

To a mixture of 4-mono-methyl 2-(R)-n-butylsuccinic acid (1 M in DMF, 1 ml, 1 mmol) and corresponding amine (1 mmol), which represents P₂′ and P₃′ substitution, and benzotrizole-1-yl-oxy-tris-pyrrolidino-phosphonium hexafluorophosphate (PyBOP) (520 mg, 1 mmol) in DMF (2 ml), was added DIEA (0.4 ml, 2.3 mmol). The mixture was stirred overnight; diluted with ethyl acetate; washed with aqueous HCl (1 N) (three times), water, and saturated sodium bicarbonate (two times) and brine; and dried over sodium sulfate. The filtrate was concentrated to give a residue that was purified on silica gel and eluted with ethyl acetate-hexane to afford a pure ester. The ester (0.1 mmol) was treated with dioxane (1 ml) and hydroxylamine (2 ml, 50% in water) for 1 to 3 days and subjected to preparative HPLC to yield the corresponding hydroxamate-containing compounds.

Preparation of VRC3375.

To a mixture of 4-mono-methyl 2-(R)-n-butylsuccinic acid (1 M in DMF, 5.0 ml, 5.0 mmol), t-butyl l-proline hydrochloride (6 mmol), and PyBOP (3.12 g, 6.0 mmol) in DMF (20 ml), DIEA (2.08 ml, 12 mmol) was added, and the mixture was stirred at room temperature overnight; diluted with ethyl acetate; washed with aqueous HCl (1 N) (three times), water, and saturated sodium bicarbonate (two times) and brine; and dried over sodium sulfate. The filtrate was concentrated and purified on silica gel and eluted with ethyl acetate-hexane (1:1) to give an ester (1.49 g, 87% yield). To a solution of the ester (1.0 g) in dioxane (10 ml) was added a solution of hydroxylamine in water (50%, 20 ml). The solution was stirred at room temperature for 2 days, diluted with ethyl acetate, washed with water (2×) and brine, and dried over sodium sulfate; the mixture was then concentrated and purified to afford VRC3375 (0.49 g, 50%). ¹H nuclear magnetic resonance: (300 MHz, CDCl₃) δ 0.89 (t, J = 7.2 Hz, 3 H), 1.43 (s, 9H), 1.24 to 1.71 (m, 6H), 1.86 to 2.43 (m, 5H), 2.53 (dd, J = 10.5 and 13.2 Hz, 1 H), 3.06 (m, 1H), 3.45 to 3.80 (m, 2H), 4.28 to 4.40 (m, 1H).

Enzyme assays.

All absorption measurements were carried out by using a SpectraMax plate reader (Molecular Devices). E. coli Ni-PDF and Streptococcus pneumoniae Zn-PDF were overexpressed and purified as previously described (5, 18).

Deformylase activity was assayed by a PDF-formate dehydrogenase (FDH) coupled assay (5, 10, 14). Briefly, the assay was carried out at room temperature with 5 nM E. coli Ni-PDF or 10 nM S. pneumoniae Zn-PDF (18) in a buffer of 50 mM HEPES (pH 7.2), 10 mM NaCl, and 0.2 mg of bovine serum albumin per ml, in half-area 96-well microtiter plates (Corning). The reaction was initiated by adding a mixture with final concentration of 0.5 U of FDH per ml, 1 mM NAD⁺, and 4 mM fMAS at each individual well. To assess the ability of testing compounds to inhibit PDF activity, the enzyme was preincubated for 10 min with various concentrations of testing compounds prior to the addition of reaction mixture. The initial reaction velocity, y, was measured as the initial rate of absorption increase at 340 nm.

The concentration of inhibitor that can inhibit 50% of enzyme activity, IC₅₀, was calculated based on equation 1:

(1)

To determine the dissociation constant of PDF inhibitor VRC3375, the initial reaction velocities were measured with various concentrations of fMAS at several actinonin concentrations. The data were then calculated according to the method of Henderson, which can be used to determine the dissociation constant of tight-binding competitive enzyme inhibitors (11):

(2)

For these equations, [In], [En], and [S] represent the total concentrations of inhibitor, enzyme, and substrate, respectively; v_i and v₀ are reaction rates in the presence and absence of inhibitor, respectively; and K_m is the Michaelis constant for the substrate fMAS.

All data fitting was carried out with nonlinear least-square regression using the commercial software package DeltaGraph 4.0 (Deltapoint, Inc.).

Cytotoxicity assay.

Compounds were screened for their in vitro cytotoxicity effect against the human leukemia K562 (ATCC CCL-243) cell line. In short, the assays were performed in 96-well microtiter plates as previously described (10). The cells were exposed to the testing compounds for 72 h at 37°C in the presence of 5% CO₂. On day 4, an indicator solution containing 1 mg of XTT per ml and 7.7 μg of phenazine methosulfate (Sigma) per ml in phosphate-buffered saline (PBS) was added, and the suspension was reincubated for 4 h. The XTT cleavage product was detected by recording the absorbance change at 450 nm; the percentage of cell growth compared with that of cells in the corresponding control well was used to calculate IC₅₀ according to equation 1.

Susceptibility studies.

The MICs were determined by the broth microdilution method per NCCLS guidelines (24) with a bacterial inoculum size of 1 × 10⁵ to 5 × 10⁵ CFU/ml. The organisms used are all part of the Vicuron strain collection (5), consisting of ATCC strains and clinical isolates of relevant pathogens, including methicillin-resistant Staphylococcus aureus, vancomycin-resistant enterococci, and penicillin-resistant S. pneumoniae. The E. coli acr and the Haemophilus influenzae acr strains are efflux pump mutants kindly provided by H. Nikaido. The MIC was the lowest concentration that yielded no visible growth after 24 h of incubation at 35°C; endpoints were obtained by measuring the optical density at 600 nm in a SpectraMax plate reader.

Library screening.

The high-throughput screening of the libraries was carried out with flat-bottom polypropylene 96-well plates on a fully integrated Automated Robotic System (Scitec, Inc., Wilmington, Del.). Each master plate contained 80 testing compounds, and the remaining 16 wells were various controls. The enzymatic screening of the library was performed using E. coli Zn-PDF, which was obtained as previously described (5). The compounds were dissolved in dimethyl sulfoxide to an estimated concentration of 10 mM and diluted 10-fold into water; 10 μl from each well was transferred to an assay plate containing 80 μl of E. coli Zn-PDF enzyme solution. The enzyme-compound mixture was preincubated for 10 min prior to adding 10 μl of substrate mixture as described in the above-mentioned PDF assay. Assay for microbial growth inhibition was carried out against S. aureus ATCC 25923. The estimated final inhibitor concentration was 100 μM for both enzyme assays and the antimicrobial assay. The cytotoxicity screening was carried out with a similar protocol as that used for human K562 cells, except that the inhibitor concentration in the final growth medium was 150 μM. The percentage of inhibition of each compound at the testing concentration was determined by comparing the levels with those of the corresponding control wells on each plate.

P_BAD-def-regulated E. coli strain.

The deformylase gene was placed under arabinose control in a tolC-deficient E. coli strain as previously described (5). The resulting bacteria were more sensitive to PDF inhibitors when def expression was down-regulated at a low arabinose concentration. To evaluate potential PDF inhibitors, bacteria were grown overnight at 35°C in Luria-Bertani medium supplemented with 0.2% arabinose. This cell suspension was further diluted 1:1,000 and used as inoculum. Checkerboard assays were performed to determine the association between susceptibility and arabinose concentration. Progressively lower MICs at decreasing arabinose concentrations indicate that the compound inhibits bacterial growth through PDF inhibition.

Molecular docking.

Reference protein coordinates used for docking were taken from the X-ray structure of E. coli Ni-PDF in complex with actinonin (PDB entry 1g2a, monomer A) (6). All water molecules were removed. The bound ligand actinonin structure was extracted from the complex and served as the reference for docking. The protein and its Ni²⁺ ion were used as the target for docking. Hydrogen atoms were added to both the ligand and protein. Energy minimization of the protein target was performed with restraints of the heavy atoms at their X-ray positions. Two-dimensional structures of actinonin, VRC3375, and VRC3376 were prepared with ISIS (MDL, Inc., San Leandro, Calif.), converted into three-dimensional structures with the CONCORD module in the SYBYL software suite (Tripos, Inc., St. Louis, Mo.), and relaxed by geometry optimization using the MMFF94 force field and net atomic charges, a dielectric constant of 1.0, a nonbonded cutoff value of 8.0 Å, and a convergence criterion of 0.035 kcal/(mol · Å) for the gradient.

Genetic optimization for ligand docking (GOLD) version 2.0 (Cambridge Crystallographic Data Centre, Cambridge, United Kingdom) was used for protein-ligand docking with actinonin, VRC3375, and VRC3376. The active site was defined as the collection of protein residues enclosed within a 15.0-Å radius sphere centered on the metal ion. The default speed selection was used to avoid a potential reduction in docking accuracy. Fifty genetic algorithm runs with default parameter settings were performed without early termination, and the 10 best solutions were kept for each ligand. GoldScore was chosen for the fitness function.

Pharmacokinetics study.

All in vivo studies were approved by the Animal Care and Use Committee of Comparative Biosciences (Mountain View, Calif.), where all such studies were carried out. CD1 female outbred mice (Charles River Laboratories) were used for pharmacokinetic analysis of PDF inhibitor VRC3375. The compound was formulated in PBS (pH 6.5) and filter sterilized. After 1 week of acclimation, mice (weight, 20 to 25 g each) were administered 100 mg of VRC3375 per kg of body weight at an administration volume of 10 ml/kg by intravenous (i.v.), subcutaneous (s.c.), or oral (p.o.) route. Blood samples were collected from anesthetized mice via cardiac puncture at 3, 6, 10, 20, 40, 60, 90, 180, and 360 min after dosing. Groups of four mice were used for each time point. The blood was allowed to clot, and the serum samples were collected and stored immediately at −80°C. For analysis, standards were prepared by spiking VRC3375 in control serum, and such samples were extracted with equal volume of acetonitrile; the supernatant was analyzed by HPLC (4.6 × 50 mm pro-C₁₈ column from YMC on HP1100 ChemStation from Agilent with UV detection at 220 nm). The quantification limit of this HPLC method for VRC3375 in serum sample is 5 μg/ml. The pharmacokinetics parameters, including T_max (time to maximum concentration), C_max (maximum concentration measured), t_1/2 (terminal half-life), and area under the curve (AUC) were calculated using WinNorlin. The oral bioavailability was calculated as the ratio of AUC of p.o. versus the AUC for i.v. administration.

Tissue distribution of VRC3375 in mice.

Various mouse tissues were collected 3 min after 100-mg/kg i.v. administration. The tissues were homogenized in 1 to 2 ml of cold PBS. The homogenates were extracted with equal volumes of acetonitrile, and the compound concentration was determined by HPLC as described above.

Protection from infection in mouse septicemia model (S. aureus, Smith strain).

A standard peritonitis model of infection was used to evaluate the efficacy of VRC3375. Briefly, outbred CD1 female mice (weight, 20 to 25 g each; Charles River) were inoculated intraperitoneally with a 90-to-100%-lethal dose (LD_90-100) (5.0 × 10⁷ CFU/mouse) of S. aureus (Smith, ATCC 19636; MIC, 8 μg/ml for VRC3375) in 0.5 ml of brain heart infusion broth containing 5% mucin. VRC3375 was dissolved in 20% cyclodextrin and administered at doses of 100, 50, 25, 12.5, 6.25, and 3.125 mg/kg at a dosing volume of 10 ml/mg via i.v., s.c., or p.o. route at 1 and 5 h after infection. Vancomycin was included as a control antibiotic, and animals in this antibiotic regimen were dosed subcutaneously. Groups of six mice were used for each dose level. Mice were monitored daily for 6 days, and cumulative mortality was used to determine the 50% effective dose (ED₅₀), which was determined by the method of Reed and Muench (29).

Acute toxicity study in mice.

Outbred CD1 mice (weight, 20 to 25 g each) were given a single dose (125, 250, 400, or 500 mg/kg; six mice/group) of VRC3375 by i.v., s.c., or p.o. administration. The compound was formulated in PBS at 23.1 mg/ml (pH 6.5) and was given in single-dose volumes of 5.4, 10.8, 17.3, and 21.6 mg/kg for dose levels of 125, 250, 400, and 500 mg/kg, respectively. Intravenous dosing was given via the tail vein over approximately 60 s, and the subcutaneous dose of 500 mg/kg was divided in 2 equal volumes that were injected simultaneously at two separate sites. The animals were monitored for 7 days. The LD₅₀s were calculated by the method of Reed and Muench (29). The gross observations of skin, lung, heart, liver, kidney, and spleen were performed upon sacrifice at day 8.

RESULTS

Preparing and screening potential PDF inhibitor libraries.

Since PDF is a metallohydrolase, it is reasonable to envision an inhibitor structure with a chelator as its “warhead,” which binds to the active site metal ion, together with a “delivery” system of a peptidomimetic group, which, in addition to supplying additional binding affinity, provides the selectivity and physicochemical properties of the molecule. Based on the mechanism of the reaction catalyzed by PDF, three low-molecular-weight chemical libraries with the general structure depicted in Fig. 1 were constructed. The building blocks used in the construction of these libraries were selected based on our prior experience and on commercial availability.

All three libraries were screened for deformylase inhibition, antibacterial activity, and cytotoxic liability. We defined the initial leads as those molecules that were inhibitors of the target enzyme with measurable antibacterial activity but without significant cytotoxicity. The screening results for the hydroxamate, carboxylate, and mercaptan libraries are compiled and summarized in Fig. 3A, B, and C, respectively. Leads were selected based on a review of all assay results.

FIG. 3.

Library screening results. Summary of screening results for three libraries with chelator of hydroxamate (A), carboxylate (B), and mercaptan (C). Each square represents one library member with the corresponding P₂′ and P₃′ substitution as indicated in Fig. 2B. Each library compound, with its corresponding P₂′ and P₃′ substitution, is represented by one square composed of three color-coded positions. Each of these positions represents (left to right) the percentage of PDF enzyme inhibition, the percentage of inhibition of S. aureus growth, and the percentage of growth for K562 cells compared to that of control. Library compounds with optimal results appear red in all three assays.

In Fig. 3, each square represents one particular library compound with its corresponding P₂′ and P₃′ substitution as specified in Fig. 2B. For each library compound, the square includes three color-coded positions. Each of these positions represents (from left to right) the percentage of PDF enzyme inhibition, the percentage of inhibition of S. aureus growth, and the percentage of growth of K562 cells. Library compounds with optimal results appear red in all three assays.

A quick scan of the data indicates that each of the three libraries has compounds capable of inhibiting PDF. Within the hydroxamate library (Fig. 3A), antibacterial activity was detected for numerous compounds at the screening concentration of 100 μM. In contrast, the carboxylate (Fig. 3B) and mercaptan (Fig. 3C) libraries appear relatively noncytotoxic in this screen, but they also lack antibacterial activity.

Further examination of the data for the hydroxamate library suggests that many of the compounds, while able to inhibit the growth of S. aureus, are also toxic to the mammalian cells under the conditions tested. For example, compounds with p-chlorobenzyl (6′), 2,4-dimethoxlbenzyl (17′), or m-trifluoromethylbenzyl (21′) at the P₃′ position have excellent antibacterial activity, but they are associated with cytotoxicity, suggesting that such substitutions may not be suitable for this class of compounds.

Among all P₂′ substitutions, it is interesting that proline (12) gave the most promising overall activity profiles. Many compounds with proline at the P₂′ site afford adequate antibacterial activity while showing very little cytotoxicity. This finding prompted us to focus our efforts on a peptidomimetic scaffold with proline at the P₂′ position.

Confirmation and initial characterization of library screening “hits.”

Using information provided from the initial hydroxamate library, we constructed smaller, more focused libraries with various cyclic or linear groups at the P₂′ position to derive a more detailed SAR. These results confirmed that proline at the P₂′ position is the preferred substitution (data not shown). Nine compounds were synthesized through traditional medicinal chemistry to confirm the findings of library screening, and biological assay results for these selected compounds are summarized in Table 1.

TABLE 1.

Biological activities of select PDF inhibitors

Most of the compounds described in Table 1 have excellent antibacterial activity. For example, VRC3324, a compound identified from one of the focused libraries, is an excellent PDF inhibitor with potent antibacterial activity against the three pathogens tested. However, VRC3324, which combines isoleucine at the P₂′ position with 2-naphthyl amide at P₃′, is also very toxic to mammalian cells in the cytotoxicity screen.

Results in Table 1 further confirm the initial observation that compounds with proline at the P₂′ position, such as VRC3375, afford excellent antibacterial activity with a reduced cytotoxic liability. For comparison, d-proline was also substituted at the P₂′ position (VRC3376), and this compound weakly inhibited the E. coli Zn-PDF and had no antibacterial activity, presumably reflecting a poor fit in the PDF active site (see Discussion). VRC3375 was selected to scale up for further biological evaluation.

The IC₅₀ of 4 nM for VRC3375 against E. coli Ni-PDF is very close to the actual enzyme concentration used in the assay (5 nM), suggesting that the binding of the inhibitor to the enzyme is very tight. Under such conditions, conventional steady-state kinetics (Lineweaver-Burke plot) cannot be used to determine the dissociation constant (K_i). In order to obtain an accurate dissociation constant for the enzyme-inhibitor complex, the PDF-FDH coupled assay was carried out at different substrate fMAS concentrations (0.21 to 10 mM) and at a reduced enzyme concentration (1.6 nM). At each substrate concentration, the concentration of VRC3375 was varied in a range of 0 to 9 nM. The initial reaction rates obtained in these experiments were plotted in Fig. 4A. The data were recalculated according to equation 2 and plotted in Fig. 4B, where the line represents the linear least-squares fit of data with a dissociation constant of 0.3 nM, which is similar to that of actinonin (5).

FIG. 4.

Inhibition of E. coli Ni-PDF by VRC3375. (A) The enzyme activity of 1.6 nM E. coli Ni-PDF was plotted with increasing concentration of substrate fMAS in the presence of 0 (▪), 0.57 (•), 1.14 (▴), 2.28 (♦), 4.56 (□), and 9.13 (○) nM concentrations of VRC3375. The curves represent the best fits of the data at a fixed inhibitor concentration to the Michaelis-Menten equation. (B) Data in panel A were reanalyzed according to equation 2 and replotted as indicated. The line represents the linear least-squares fit of the data. On the basis of this method, the dissociation constant for VRC3375 against E. coli Ni-PDF is 0.3 nM.

The antibacterial activities of VRC3324 and VRC3375 were further tested against an expanded panel of clinically relevant bacterial pathogens; the results are summarized in Table 2. Both compounds have antibacterial activity against gram-positive and fastidious gram-negative bacteria. VRC3324 has notably better activity than either VRC3375 or actinonin against S. pneumoniae or Enterococcus. Both compounds have good activity against H. influenzae and Moraxella catarrhalis.

TABLE 2.

Antimicrobial activity comparison of actinonin, VRC3324, and VRC3375

Organism (no. of strains tested)	MIC (μg/ml)
	Actinonin	VRC3324	VRC3375
S. aureus (3)	8-16	1-2	1-4
Staphylococcus epidermidis (1)	4	2	1
Enterococcus faecalis (3)	32-64	2	32-64
S. pneumoniae (3)	8-32	1	8-32
Streptococcus pyogenes (1)	8	0.25	64
H. influenzae (3)	1-2	1-4	2-4
H. influenzae acr (1)	0.13	0.06	0.13
M. catarrhalis (1)	0.5	0.25	0.25
E. coli (1)	>64	>64	>64
E. coli acr (1)	0.25	0.25	0.25
Bacteroides fragilis (1)	0.25	≤0.06	1

To establish that VRC3375 inhibits bacterial growth through the inhibition of PDF activity, a P_BAD-def construct in E. coli tolC was used to assess the compound's mechanism of action. Since the level of deformylase expression inside cells of this strain depends on, and is proportional to, the inducer (arabinose) concentration in the growth medium (5), dependence of MIC on the concentration of arabinose in the medium will indicate that the compound inhibits bacterial growth through the inhibition of deformylase. As shown in Fig. 5, while the susceptibility of the P_BAD-def bacteria to the control antibiotics fosfomycin and ciprofloxacin remains constant at different concentrations of arabinose, the MIC of VRC3375 gradually decreases with decreasing arabinose concentration. These data confirm that VRC3375 inhibits bacterial growth through the inhibition of PDF, the intended target of the compound.

FIG. 5.

PDF is the main target of VRC3375. Efflux-deficient P_BAD-def E. coli was used to measure the MIC under conditions of different arabinose concentrations. The PDF level in this organism increases with increasing arabinose concentration. The MICs for VRC3375 (•) increase with increasing concentration of arabinose, while the MICs for control antibiotics ciprofloxacin (♦) and fosfomycin (▪) remained constant, suggesting that VRC3375 inhibits bacterial growth through the inhibition of PDF.

Docking analysis of binding interactions of VRC3375.

The accuracy of the docking method was established by a comparison of the best GOLD complex of actinonin in E. coli Ni-PDF to the experimentally observed binding mode. The root mean square deviation of heavy atoms was 1.65 Å for the entire molecule and 0.35 Å for selected atoms (excluding the pyrrolidine ring). Thus, GOLD provides high accuracy when docking the flexible molecule of actinonin, if one ignores the pyrrolidine ring at the P₃′ position of actinonin. VRC3375, having an l-proline at the P₂′ position, and its d-proline diastereomer (VRC3376) were docked into the protein target using the same settings as those used for actinonin. The 10 top-ranked binding conformations of VRC3375 superimpose very well, and the one with the highest GoldScore was chosen to represent the ligand's binding mode. For VRC3376, 7 out of 10 top-ranked solutions adopted similar conformations, and the one with the highest GoldScore was selected as the binding mode. The hydroxamate group of VRC3375 superimposes well on that of actinonin (Fig. 6A), indicating that VRC3375 adopts a similar metal chelation with the Ni²⁺ ion, which is consistent with the enzymatic activity against PDF as shown by the data in Table 1. The excellent superimposition of the n-butyl side chain of VRC3375 and the n-pentyl side chain of actinonin reveals a similar hydrophobic occupancy in the S₁′ pocket of the enzyme, having van der Waals interactions with Ile44, Ile86, Glu88, Leu125, Ile128, and His132 (6). Though the carbonyl oxygen atom (circled in Fig. 6A) of VRC3375 has a different orientation from that in actinonin, it retains hydrogen bonding with Ile44. After adding hydrogen atoms to the protein residues, the distance between the carbonyl oxygen atom of actinonin and the amide hydrogen atom of residue Ile44 is 2.68 Å, similar to the distance reported for the X-ray structure of E. coli Ni-PDF complexed with VRC4307 (10). The tert-butyl ester group of VRC3375 is directed toward the side chains of hydrophobic residues Ile44, Ile86, and Leu125, and the l-proline is closer to Cys90 to form better van der Waals interactions (Fig. 6B). The structural variation of VRC3375 and actinonin at the P₂′ and P₃′ positions does not generally affect the binding mode, because the analogues are able to achieve a relatively good superimposition. As shown in Fig. 6C, bound VRC3375 is well accommodated in the active site of the enzyme, with an excellent fit of its n-butyl chain in the S₁′ pocket, as observed in the other reported PDF X-ray complex structures (6, 9, 10). Figure 6A shows the difference of the docking results for VRC3376 and VRC3375, including different orientations of the carbonyl group (circled) and the tert-butyl group, with slight changes in the positions of the hydroxamate and n-butyl groups. As discussed later, the tert-butyl group of VRC3376 is distant from hydrophobic residues Ile44, Ile86, and Leu125, resulting in the lack of van der Waals interactions with those residues.

FIG. 6.

Docking result for VRC3375 in E. coli PDF. (A) Overlay of the actinonin X-ray binding mode (carbon atoms in green) and the docking results for VRC3375 (carbon atoms in orange) and VRC3376 (carbon atoms in magenta). The Ni²⁺ metal ion is shown in purple. (B) Binding environment of VRC3375 (carbon atoms in orange) in PDF. (C) Docked VRC3375 in PDF represented by a surface. The surface of the protein was created by MOLCAD using Fast Connolly in the SYBYL software application. The carbon atoms are shown in orange, and the metal ion is indicated by a magenta portion of the surface.

Hydrogen bonding is important for the binding of the natural PDF inhibitor actinonin. Three carbonyl oxygen atoms in actinonin are hydrogen bond acceptors for the amide hydrogen atoms of Ile44, Gln50 (side chain), Gly89, and Leu91 (Fig. 7A). As hydrogen bond donors, all four polar hydrogen atoms in actinonin have interactions with the carbonyl oxygen atoms of Gly45, Glu87, Gly89, and Glu133 (side chain) to form six hydrogen bonds. VRC3375 has a very similar hydrogen bonding network with PDF except in the P₂′ or P₃′ position, due to the replacement of the large group in actinonin by a small group in VRC3375. However, two additional hydrogen bonds are observed between VRC3375 and the side chain amide hydrogen atoms of Gln50 and Arg97 (Fig. 7B), which presumably compensate for the loss of a hydrogen bond with Gly89. The lack of three hydrogen bonds in the docking solution of VRC3376, relative to that of VRC3375 (Fig. 7C), indicates a less favorable binding environment for VRC3376 in PDF.

FIG. 7.

Illustration of the intermolecular hydrogen bonding network involving PDF inhibitors actinonin (A), VRC3375 (B), and VRC3376 (C). Hydrogen bonds formed between the ligand's oxygen atom and the enzyme's hydrogen atom are shown in red, and hydrogen bonds formed between the ligand's hydrogen atom and enzyme's oxygen atom are shown in green. Bonds in magenta represent the additional hydrogen bonds, which are absent in the PDF-actinonin complex.

In vivo evaluation of VRC3375.

The in vivo efficacy of VRC3375 was evaluated in a mouse S. aureus septicemia infection model. VRC3375 was administered by i.v., s.c., or p.o. routes to mice that had been infected intraperitoneally with S. aureus (Smith strain). VRC3375 had comparable protective effects by i.v., s.c., or p.o. routes of administration, with ED₅₀s of 32, 17, and 21 mg/kg, respectively (Table 3). Control antibiotic vancomycin has an ED₅₀ of 1 mg/kg under these conditions (s.c.).

TABLE 3.

Summary of in vivo efficacy, LD₅₀, and pharmacokinetic parameters of VRC3375 in mice

Administration route	ED50 (mg/kg)a	LD50 (mg/kg)b	t1/2 (min)c	Cmax (μg/ml)c	Tmax (min)c	AUCd (μg/ min/ml)	F%e
i.v.	32	447	16	147		2,456
s.c.	17	>500	15	65.7	20	2,794
p.o.	21	>500	15	43	10	1,568	64

^aMurine septicemia model (S. aureus Smith strain).

^bLD₅₀ determined in mice after single administration.

^cKinetics parameters determined with single 100-mg/kg dosing in mice.

^dAUC, area under the curve.

^eF%, oral bioavailability.

In vivo pharmacokinetic studies were also carried out for VRC3375. Mice were dosed with 100 mg of VRC3375 per kg via i.v., s.c., and p.o. routes. Sera were collected at different time points after dosing, and concentrations of the compound were determined by HPLC analysis. The results are summarized in Fig. 8 and Table 3. As shown in Fig. 8A, VRC3375 was rapidly absorbed after oral administration and reached maximum concentration 10 min after dosing. In comparison, the absorption through subcutaneous administration is relatively slow, reaching maximum concentration in serum 20 min after dosing. Based on these data, the absolute oral bioavailability of VRC3375 was calculated to be 64%. Note that formulation in the in vivo efficacy study was different from that in the pharmacokinetics study (cyclodextrin versus PBS). Although the effect of different formulations on drug exposure for this class of compounds is unknown, the pharmacokinetic parameters from three administration routes agree with the efficacy data.

FIG. 8.

Pharmacokinetic study of VRC3375 in CD1 female outbred mice. (A) VRC3375 concentrations in serum were measured at various times after the mice received 100 mg/kg of the compound by i.v. (•), s.c. (▪), and p.o. (▴) administration. (B) The tissue concentrations of VRC3375 3 minutes after the mice received 100 mg/kg of the compound by i.v. administration.

Several mouse tissues were also collected at 3 minutes after i.v. dosing with the compound, and VRC3375 concentrations were determined in these samples to assess the tissue distribution of the compound. As shown in Fig. 8B, the compound had already distributed among all tissues studied 3 minutes after the i.v. administration.

The acute toxicity of VRC3375 was determined for mice by all three routes of administration. The mice were monitored daily for 1 week. Even at the highest dose of 500 mg/kg, no mortality or adverse effects were observed for mice that received VRC3375 by s.c. and p.o. routes. In the i.v. group, three mice died immediately after receiving the 500-mg/kg dose. The three remaining mice from that dosage group were given 400 mg/kg of VRC3375. Immediately after receiving that dose, the mice had temporary adverse reactions, such as heavy breathing and lethargy; the mice appeared normal the following day. An LD₅₀ of 447 mg/kg was calculated for the i.v. group (Table 3). At day 8, all animals were sacrificed, and a gross necropsy was performed for skin, lung, heart, liver, kidney, and spleen. No abnormalities were observed, except that three mice in the 500-mg/kg s.c. group had dark red skin plaques at the injection sites. No differences in body weight were observed during the course of the experiments for any of the groups.

DISCUSSION

In this study, we provide a rare example of the identification of in vivo active antibacterial agents through the construction of focused chemical libraries based on a target enzyme mechanism. Chemical libraries constructed in this study were based on the recognition that PDF is a metallohydrolase and that the S₁′ site of the enzyme prefers a methionine-like side chain, since all natural deformylase substrates invariably possess methionine at their P₁′ sites.

Although Huntington et al. reported that a series of peptide thiols were potent PDF inhibitors with whole-cell activity (12), the majority of PDF inhibitors reported to have significant antibacterial activity contain either a hydroxamate or reverse hydroxamate as the metal chelator (31). In fact, in a recent study for which compounds of the same P₁′-P₃′ peptidomimetic but with different metal binding groups were prepared, only hydroxamate- and N-formyl hydroxylamine-containing compounds showed antibacterial activity (12a, 30). The screening of the libraries described in this study revealed that all libraries contain members with inhibitory activity against E. coli PDF. However, as shown in Fig. 3, only the hydroxamate library contains compounds that also have antibacterial activity at the concentrations tested. Since there is no clear strategy yet for engineering whole-cell activity into enzyme inhibitor leads, only compounds of the hydroxamate library were pursued further.

In antibacterial drug discovery, a common pitfall is to pursue antibacterial potency (low MICs) without paying enough attention to the potential toxicity and pharmacokinetics properties early in the lead selection process. Indeed, most antibacterial compounds that fail do so in the early development stages due to problems related to toxicity. Our main goal at the early discovery stage was to identify structural features that can provide sufficient potency but with the least toxicity liability. Therefore, hit selection criteria from our primary screenings included low cytotoxicity for the lead structure(s), even if this meant only modest antibacterial activity. Our past experience taught us that it is often easier to improve the antibacterial activity than to remove the cytotoxicity liability.

On the basis of this philosophy, all three libraries were assayed for deformylase inhibition, antibacterial activity, and cytotoxicity. Initial screening of the hydroxamate library revealed that several members with hydrophobic substitutions at the P₂′ and P₃′ positions possessed potent antibacterial activity. These results were further confirmed by the subsequent synthesis of VRC3324 and related compounds. In addition to compounds with very hydrophobic moieties, several compounds with proline at the P₂′ position were identified that also afford antibacterial activity. Significantly, these compounds have the least cytotoxicity liability at the concentrations tested. As shown in Table 1, the purified compounds with proline at the P₂′ position had the highest IC₅₀s against the K562 human cell line. Table 1 also reveals that VRC3324, which has linear P₂′ structure, had better activity against S. pneumoniae than did any of the other compounds tested. This may be due in part to improved activity against the S. pneumoniae target PDF.

Table 2 compares the antibacterial activity of actinonin, VRC3324, and VRC3375 against an expanded panel of pathogens. Among these PDF inhibitors, the hydrophobic VRC3324 is the most active compound against the bacteria tested. Note that the MICs of all three compounds for efflux-deficient mutants of H. influenzae and E. coli are significantly lower than those for the corresponding wild-type strains, suggesting that these PDF inhibitors are substrates for the acr efflux pump.

In selecting the lead structure, emphasis was given to those compounds that had measurable antibacterial activity but low cytotoxicity. Although compounds of the VRC3324 structural class are the most potent antibacterial agents among the leads identified, examination of the cytotoxicity data clearly suggests that such compounds will also have major toxicity liabilities. In addition, initial studies indicated that these compounds have very low solubility, which may also result in poor pharmacokinetic properties. By contrast, proline-containing lead structures such as VRC3375, though not the most potent in terms of MIC, still provide sufficient antibacterial activity accompanied by a low cytotoxicity liability. Therefore, efforts were focused on the lead structure with proline at the P₂′ position, which resulted in the identification of VRC3375.

In screening chemical libraries for early leads, it is quite common for the major emphasis to be placed on the potency against the primary target, such as the target enzyme. Due to the limited resources, only a limited number of lead structures can be monitored. Hit selection based on potency often leads to structures from which it proves very difficult to eliminate toxicity or other undesirable properties such as low solubility. In fact, many good drugs are not very potent against their targets but are very selective (low toxicity) and have excellent pharmacokinetic properties. For example, flurbiprofen, a nonsteroidal anti-inflammatory drug, has an IC₅₀ of 10⁻⁶ M against its target, cyclo-oxygenase, which would most likely not be considered a hit in high-throughput screening (13). Nonetheless, the drug is very effective and has been on the market for many years. Such a disconnect between potency against intended target and the actual clinical endpoint mandates considering other parameters in addition to specificity for primary targets when selecting the hits from library screening.

The docking of VRC3375 into E. coli PDF reveals a stabilizing binding environment achieved through various protein-ligand interactions: (i) the involvement of the hydroxamate group in hydrogen bonding and metal chelation has been shown experimentally to be important for antibiotic activity against PDF (17, 30; S. Sundaram, R. Jain, S. Lopez, C. Wu, D. Chen, C. J. Hackbarth, W. Wang, G. Withers, D. V. Patel, J. Trias, R. White, and Z. Yuan, Abstr. 42nd Intersci. Conf. Antimicrob. Agents Chemother., abstr. F-1669, 2002) and is supported by the docking performed in this study; (ii) proline substitution at the P₂′ position not only results in a very tight fit of the inhibitor in the active site of the enzyme, but it also maximizes its selectivity with minimum cytotoxicity (Table 1); and (iii) the bound orientation of the tert-butyl group favors van der Waals interactions with the side chains of Ile44, Ile86, and Leu125, a hydrogen atom of one methyl group of the tert-butyl group having the shortest distances of 2.07, 3.27, and 2.85 Å to a methyl hydrogen atom of Ile44, Ile86, and Leu125, respectively. In contrast to VRC3375, the d-proline analog VRC3376 fits less favorably in the PDF active site due to reduced hydrogen bonding and poor van der Waals interactions involving the tert-butyl group and the hydrophobic residues of the enzyme. This binding preference for l-proline (VRC3375) over d-proline (VRC3376) is demonstrated by the more than 10-fold difference in IC₅₀s for Zn-PDF (Table 1). However, this difference was not observed for Ni-PDF. This apparent contradiction reflects the fact that the IC₅₀ for Ni-PDF approaches half that of the enzyme concentration; therefore, this parameter can no longer differentiate among inhibitors with intrinsic binding constants (K_i) less than or equal to half of the enzyme concentration. On the other hand, the IC₅₀ for Zn-PDF is well above half that of the corresponding enzyme concentration; hence, this parameter can be used to differentiate the binding affinity of the corresponding inhibitors. It has been reported that the identity of the bound metal ion is crucial for the catalytic efficiency of the enzyme (3, 28). The Zn²⁺ ion causes a striking loss of the catalytic activity relative to that of Fe²⁺ (natural form) and Ni²⁺ ions, but the differences in the protein structures due to the identity of the bound metal are extremely small (9). As anticipated, when we superimposed the X-ray structures of actinonin in complex with E. coli Ni-PDF (PDB entry 1g2a, monomer A) and Zn-PDF (PDB entry 1lru, monomer A), the overall conformation of the residues in the active site and the bound ligand was similar in both complexes.

VRC3375 showed modest in vivo efficacy, even with oral administration, in a mouse S. aureus septicemia infection model (ED₅₀s of 32, 17, and 21 mg/kg by i.v., s.c., and p.o. administration, respectively). By comparison, actinonin, a naturally occurring PDF inhibitor, lacked in vivo efficacy at dose levels of up to 500 mg/kg (4), whereas both actinonin and VRC3375 display comparable in vitro antibacterial activity. These data clearly demonstrate that the current lead structure, VRC3375, which was identified through screening-focused chemical libraries designed on the bases of the enzyme reaction mechanism, has a dramatic improvement of in vivo efficacy. This improvement in efficacy most likely reflects the improved pharmacokinetic properties of the compound, as suggested by the absolute bioavailability of 64% for VRC3375 when orally dosed at 100 mg/kg. On the other hand, the relatively short half-life of 15 min, seen for all three administration routes, suggests that the compound may not be stable in vivo. This instability is most likely due to the conversion of the hydroxamate moiety to the corresponding carboxylic acid or amide. Indeed, using the authentic corresponding carboxylic acid and amide as standards, we found a time-dependent appearance and subsequent disappearance of both compounds in the serum samples taken at different times after dosing with VRC3375 (data not shown). Thus, similar to hydroxamate-based matrix metalloprotease inhibitors, the stability of the hydroxamate in PDF inhibitors will be a key factor in achieving better in vivo efficacy.

Among all PDF inhibitors reported previously, only two types of compounds, represented by BB-3479 and VRC4307, demonstrated measurable in vivo antibacterial activity (6, 10). VRC3375 represents the third example of PDF inhibitors with in vivo antibacterial activity. A common structural feature of these compounds is that the metal chelating group has been confined to a hydroxamic acid or N-formyl hydroxylamine group; in contrast, diverse P₂′ and P₃′ peptidomimetic substituents are tolerated. The P₂′ varies from an unnatural tert-leucine fragment in BB-3479 to the cyclic amino acid proline fragment in VRC3375, and VRC4307 bears a urea backbone in place of the customary P₁′-P₂′ amide group. In fact, it was proposed that the superior in vivo activity of BB-3497 (in comparison to actinonin) is due to the compound's tert-leucine at the P₂′ position (6).

As mentioned before, the key reason for focusing on VRC3375 as the lead structure was its reduced in vitro toxicity liability. The validity of this screen was borne out with the results from the in vivo acute toxicity study. The compound did not show any toxic effects up to 500 mg/kg by oral or subcutaneous administration and had an LD₅₀ of 447 mg/kg after i.v. administration. In a separate study, a related compound with severe in vitro cytotoxicity (IC₅₀, 10⁻⁶ g/ml against K562 cell line) demonstrated significant adverse reactions when mice were dosed at 30 mg/kg (data not shown). Although the currently used cytotoxicity assays will not be able to predict many types of unexpected toxicity in humans, it is reasonable to use such assays as primary screening tools to filter out compounds that may subsequently display general toxicity in vivo. Examples provided in this study demonstrate the potential advantage of using such filters early in the discovery stages so as to focus on lead structures that will have a higher likelihood of success in subsequent preclinical studies.

In summary, VRC3375 was identified as a highly selective PDF inhibitor that is orally active in a mouse septicemia model. This compound is equally active against drug-susceptible and -resistant bacteria such as methicillin-resistant S. aureus and vancomycin-resistant enterococci, suggesting the possibility of using this class of compounds for treating infections caused by bacteria that are resistant to currently available antibacterial drugs. Further optimization of this lead structure is needed in order to identify compounds that are suitable as antibacterial drugs. Results of these studies will be published elsewhere.