1,5-Benzodiazepines, a Novel Class of Hepatitis C Virus Polymerase Nonnucleoside Inhibitors

Abstract

The exogenous control of hepatitis C virus (HCV) replication can be mediated through the inhibition of the RNA-dependent RNA polymerase (RdRp) activity of NS5B. Small-molecule inhibitors of NS5B include nucleoside and nonnucleoside analogs. Here, we report the discovery of a novel class of HCV polymerase nonnucleoside inhibitors, 1,5-benzodiazepines (1,5-BZDs), identified by high-throughput screening of a library of small molecules. A fluorescence-quenching assay and X-ray crystallography revealed that 1,5-BZD 4a bound stereospecifically to NS5B next to the catalytic site. When introduced into replicons, mutations known to confer resistance against chemotypes that bind at this site were detrimental to inhibition by 1,5-BZD 7a. Using a panel of enzyme isolates that covered genotypes 1 to 6, we showed that compound 4a inhibited genotype 1 only. In mechanistic studies, 4a was found to inhibit the RdRp activity of NS5B noncompetitively with GTP and to inhibit the formation of the first phosphodiester bond during the polymerization cycle. The specificity for the HCV target was evaluated by profiling the 1,5-BZDs against other viral and human polymerases, as well as BZD receptors.

The global scope of hepatitis C virus (HCV) infection is a major concern for human health. The disease can lead to liver fibrosis, cirrhosis, hepatocellular carcinoma, and death if treatment is not provided. Although the current standard of care, comprising interferon and ribavirin, can eradicate the virus, many treatment failures arise due to the variability of the response rate observed across genotypes (19, 34) and tolerability issues. In addition to these challenges, factors that decrease the efficiency of the immune system, such as age, alcoholism, and human immunodeficiency virus (HIV) coinfection, also play a role in the disease progression. For these reasons, major efforts are directed toward developing novel therapeutics that include improved interferons, novel immunomodulators, and both direct and indirect antivirals (33).

The HCV polymerase (NS5B) is a focus of HCV drug discovery efforts. The main functional role of NS5B in the virus life cycle is the assembly of the replicase complex at the endoplasmic reticulum membrane and the amplification of the genetic material through RNA-dependent RNA polymerase (RdRp) activity (1). NS5B has also been shown previously to interact with the chaperone cyclophilin B to enhance the binding of the polymerase to RNA (49), to downregulate the expression of the retinoblastoma tumor suppressor (36), and to be targeted to the endoplasmic reticulum membrane through interaction with the estrogen receptor (48). Direct antivirals that are capable of inhibiting the polymerase are classified as nucleoside inhibitors and nonnucleoside inhibitors (NNIs) (26). Nucleoside analogs bind at the active site, and NNIs bind to one of four previously identified sites, NNI-1, NNI-2, and NNI-3 (40) and NNI-4 (46). Examples of antivirals that have progressed into clinical development are the nucleoside inhibitors NM283, R1626, and R7128 and the NNIs BILB 1941, VCH-759, GSK625433, and HCV-796 for NNI-1, NNI-2, NNI-3, and NNI-4, respectively (13, 17, 21, 26, 33). The short-term clinical efficacy of these compounds varies, and that of R1626 was recently shown to be the most potent; this nucleoside analog reduced the level of HCV RNA by 3.7 log₁₀ IU/ml from the baseline when 4,500 mg was administered twice a day (b.i.d.) for 14 days as a monotherapy (25) and by 5.2 log₁₀ IU/ml when 1,500 mg was coadministered b.i.d. with pegylated interferon and ribavirin for 4 weeks (42). These results demonstrate that polymerase inhibitors can match the antiviral effect previously reported for the HCV NS3/4A protease antivirals (28). To date, the clinical efficacy of the NNI class has been more modest. Preliminary data reported for monotherapy with VCH-759 (13), a thiophene analog, showed a 2.5 log₁₀ IU/ml decline in HCV RNA when 800 mg was administered b.i.d. for 10 days.

Despite the dramatic progress achieved in the field, both in terms of cellular potency and clinical efficacy, the development of polymerase antivirals has suffered from a high attrition rate due to toxicity issues. These failures highlight the need to develop other chemical scaffolds that offer the potential to inhibit HCV replication. Here, we report the discovery of a novel class of HCV polymerase NNIs, 1,5-benzodiazepines (1,5-BZDs), and we provide the biological characterization of a 1,5-BZD analog that includes genotypic profiling, X-ray crystallography, profiling against a replicon NS5B NNI site mutant panel, and kinetic and mechanistic studies.

MATERIALS AND METHODS

Purification of NS5B.

Recombinant NS5BΔ21 (from an HCV J4 genotype 1b strain [hereinafter referred to as 1b J4]) was overexpressed in Escherichia coli BL21(DE3) and purified to homogeneity as described previously (40).

RdRp assay.

The RdRp primer-dependent transcription assay was performed as described previously (40). The 50% inhibitory concentrations (IC₅₀s) in the RdRp primer-independent de novo transcription assay were determined as follows: 190 nM purified NS5BΔ21 (1b J4) enzyme was incubated with 86 nM 3′ untranslated region template derived from plasmid pCV-H77 (GenBank accession no. AF011751; bases 9218 to 9558), 0.5 mM ATP, 0.5 mM GTP, 0.5 mM CTP, 10 μM UTP, and 2.5 μCi of [α-³³P]UTP in a solution of 20 mM Tris-HCl, pH 7.5, 1 mM dithiothreitol, 5 mM MgCl₂, and 20 mM NaCl. The 40-μl reaction mixture was incubated at room temperature for 2 h, and the reaction was stopped by the addition of 20 μl of 0.5 M EDTA containing 10 mg/ml of tRNA (Roche; catalog no. 10109541001). The mixture was transferred onto filter binding plates (Millipore; catalog no. MSFBN6B50), and the precipitate was washed twice with 200 μl of ice-cold 10% trichloroacetic acid, twice with 200 μl of 1 M sodium phosphate-5% trichloroacetic acid, and twice with 200 μl of 95% ethanol. Following the addition of 80 μl of a scintillation cocktail, the 96-well plates were subjected to counting by scintillation proximity analysis on a Packard TopCount reader. The gel-based de novo initiation assay was performed as follows: 1 μM purified NS5B enzyme was incubated with 1 μM RNA template; 5 μl of 1,5-BZD 4a (in 6% dimethyl sulfoxide [DMSO]); 10 μM (each) ATP, CTP, GTP, and UTP; and 2 μCi of α-³³P-labeled nucleoside triphosphate (NTP) in a solution of 50 mM HEPES, pH 7.5, 5 mM MgCl₂, 5 mM MnCl₂, 10 mM KCl, and 1 mM dithiothreitol. The 50-μl reaction mixture was incubated at room temperature for 30 min and then subjected to a phenol extraction step and ethanol precipitation with 7.5 μg of GlycoBlue (Ambion; catalog no. AM9515). The precipitate was washed, dissolved in 5 μl of gel loading buffer II (Ambion; catalog no. AM8546G), denatured for 3 min at 95°C, and loaded onto an 8 M urea-20% polyacrylamide gel with TTE buffer (89 mM Tris, pH 8.0, 28 mM taurine [2-aminoethanesulfonic acid], 0.5 mM EDTA [National Diagnostics; catalog no. EC-871]). Following electrophoresis, the gel was fixed in the presence of 10% glycerol, 40% ethanol, and 10% acetic acid, dried on a Whatman paper, and visualized on a Typhoon imager (GE Healthcare).

Determination of the kinetic behavior of 1,5-BZD 4a.

The mode of inhibition of compound 4a was determined using the previously described RdRp primer-dependent transcription assay (40). Reaction velocities were measured at different GTP and inhibitor concentrations ranging from 0.0084 to 4.3 μM and 0 to 8 μM, respectively. The concentration of the NS5BΔ21 (1b J4) polymerase used was 100 nM. Kinetic data were graphically represented either as a double-reciprocal plot or as a direct plot. For the double-reciprocal plot, data points were fitted by linear regression to identify the mode of inhibition. Based on this outcome, a nonlinear regression was used to fit the data in the direct plot according to a noncompetitive model of inhibition, as expressed by formula 1.

(1)

where v is velocity, [S] is the concentration of substrate, [I] is the concentration of the inhibitor, and α is the factor for the modification of K_i by substrate.

Fluorescence-quenching assay.

The fluorescence-quenching assay was performed with a 96-well UV transparent microplate (Corning; catalog no. 3635) and a total volume of 200 μl per well as described previously (23). The NS5BΔ21 (1b J4) polymerase was diluted to a final concentration of 1 μM in RNase-free water containing 20 mM Tris-HCl (pH 7.5), 25 mM KCl, and 7 mM MgCl₂. Compounds were serially diluted in 50% DMSO, and 2 μl per well was added to the enzyme solution. After 5 min of incubation of the compound-enzyme complex, the fluorescence emission spectrum from 310 to 400 nm was scanned with an excitation wavelength of 280 nm by using a SpectraMax Gemini fluorescence reader. Under these conditions, the tryptophan residues in the enzyme are selectively excited to emit at the maximum wavelength of ca. 335 nm. Due to changes in the local environment of the enzyme or to direct interactions of a compound with the enzyme, the intrinsic fluorescence of this enzyme is quenched.

Human polymerase, HIV-1 RT, and dengue virus polymerase assays.

The human and dengue virus polymerase assays were performed by Replizyme Ltd. (Heslington, United Kingdom). Briefly, each enzyme-compound combination was tested in duplicate over a range of concentrations from 0.8 to 100 μM. The compounds were run alongside a control (no inhibitor), a solvent dilution (0.016 to 2% DMSO), and the relevant reference inhibitor. For the HIV type 1 (HIV-1) reverse transcriptase (RT) assay, E. coli BL21(DE3) was transformed with the N-terminal six-His HIV-1 RT expression construct, after which protein expression was induced overnight at 37°C with 0.4 mM IPTG (isopropyl-β-d-thiogalactopyranoside). The homodimer was purified by Ni-nitrilotriacetic acid and heparin chromatography to apparent homogeneity on a sodium dodecyl sulfate-polyacrylamide gel. A scintillation proximity assay was used to measure the inhibition of HIV RT activity at 2.5 nM enzyme by using the Amersham scintillation proximity assay kit and efavirenz as a reference inhibitor.

Replicon luciferase assay and counterscreen assays.

The HCV 1b subgenomic luciferase reporter replicon (replicon clone ET, obtained from R. Bartenschlager and adapted from a study by Lohmann et al. [31] with adaptive mutations E1202G, T1280I, and K1846) was used to measure anti-HCV activity. The counterscreen cell lines were a Huh7 hepatoma cell line (Huh7-CMV-Luc) containing a human cytomegalovirus major immediate-early promoter-Luc construct and an MT4 T-cell line (MT4-LTR-Luc) containing a long terminal repeat-Luc reporter. For all assays, 2,500 cells and 5,000 cells for the MT4 cell line were incubated with compounds plated in a 384-well nine-point dilution format (1/4 dilutions) for 3 days; cellular activity was then detected by the measurement of luciferase activity.

Cytotoxicity assays.

Compounds (eight serial fourfold dilutions) were tested with cells carrying HCV 1b subgenomic replicons by using subconfluent cells (500 cells/per 96 wells, grown in the absence of G418) incubated for 5 days, and toxicity was measured by an indicator for cell viability (resazurin). The same procedure was performed with HepG2 cells (4,000 cells/per 96 wells) and HEK293T cells (3,000 cells/per 96 wells), except that the cells were incubated for 72 h. For the ATP assay, MT4 cells (5,000 cells/per 96 wells) were incubated with compounds (diluted as described above) for 3 days at 37°C and then 40 μl of ATPlite 1step luminescence assay system reagent (PerkinElmer) was added to measure the intracellular level of ATP according to the manufacturer's protocol.

Protein crystallography.

Crystallography was performed by Proteros Biostructures GmbH (Martinsried, Germany). HCV (1b J4) NS5BΔ21 was crystallized by the vapor diffusion method using 5 to 8% polyethylene glycol 6000 and 100 mM Mg salts buffered at pH 6.75. Stick-shaped crystals grew within 2 days, to a maximum extension of 20 by 20 by 80 μm. Protein-ligand complexes were formed from inhibitor solutions in DMSO which were added to crystallization buffers or soaked into preformed crystals.

Data collection and processing.

Crystals were flash-frozen using glycerol as a cryoprotectant and measured at a temperature of 100 K. X-ray diffraction data were collected from crystals of HCV NS5B complexes with compound 4a under cryogenic conditions. The crystals belonged to space group P2₁2₁2₁, with two molecules per asymmetric unit. Data were processed using the programs XDS and XSCALE. Data collection statistics are provided in the supplemental material.

Structure modeling and refinement.

The phase information necessary to determine and analyze the structure was obtained by molecular replacement. The previously resolved structure of HCV NS5B was used as a search model (38). Subsequent model building and refinement were performed according to standard protocols with the CCP4 and COOT software packages. For the calculation of the free R-factor, a measure to cross-validate the correctness of the final model, about 5% of measured reflections were excluded from the refinement procedure (see the supplemental material). Refinement was carried out using REFMAC, including TLS refinement. Ligand parameterization was carried out with the program CHEMSKETCH, and LIBCHECK (CCP4) was used for the generation of the corresponding library files. The water model was built with the “Find waters…” algorithm of COOT by putting water molecules in peaks of the F₀-F_c map contoured at 3.0 σ and then refining with REFMAC5. All waters were checked with the validation tool in COOT. The criteria for the list of suspect waters were as follows: a B-factor greater than 80, a 2F₀-F_c map of less than 1.2 σ, and a distance from the closest contact of less than 2.3 Å or more than 3.5 Å. Suspect water molecules and those in the active site (with a distance from the inhibitor of less than 10 Å) were checked manually. The occupancy of side chains, which were in negative peaks in the F₀-F_c map (contoured at −3.0 σ), were set to zero and subsequently to 0.5 if a positive peak occurred after the next refinement cycle. The Ramachandran plot of the final model showed no residues in the disallowed region (for details, see the supplemental material). Statistics of the final structures and the refinement process are listed in the supplemental material. Figure 4B was prepared with MOE (version 2007.09; CCG, Montreal, Canada) and Fig. 4A and C were prepared with PyMOL (Delano Scientific, Palo Alto, CA).

FIG. 4.

X-ray structure of 1,5-BZD 4a in a complex with HCV polymerase. (A) Overall structure of NS5B in a complex with 4a. Palm, finger, thumb, and β-flap subdomains of NS5B are color coded in red, blue, green, and orange, respectively. The tubular structure of 4a is color coded in cyan. (B) Binding interaction map for the 4a-NS5B complex. “Ligand exposure” denotes ligand atoms that are exposed to the solvent when bound to the enzyme, whereas “receptor exposure” denotes enzyme atoms that are significantly buried by the ligand in the complex. The terms polar, acidic, basic, and greasy refer to the nature of each residue. The hydrogen bond with Tyr448 is shown as a blue arrow. (C) Atomic representation of the BZD binding pocket in two different orientations. Compound 4a is color coded in green, the β-flap is orange, and NS5B residues are gray. The hydrogen bond with Tyr448 is shown as a dotted magenta line.

Protein structure accession number.

The atomic coordinates and structure factors have been deposited in the Protein Data Bank, Research Collaboratory for Structural Bioinformatics, Rutgers University, New Brunswick, NJ (http://www.rcsb.org/), under accession number 3CSO.

RESULTS

Identification of 1,5-BZDs as inhibitors of the HCV polymerase.

A high-throughput screen of a library of small molecules was performed to search for novel inhibitors of HCV NS5B polymerase by using the enzymatic assay described previously (14). Among the hits identified, the 1,5-BZDs 1 and 2 were selected for further studies (Table 1). Compounds 1 and 2 exhibited micromolar activity in the biochemical assay (IC₅₀s, 3.1 and 7.9 μM, respectively) and were found to be less active in inhibiting HCV replication in HCV replicon-carrying cells (50% effective concentrations [EC₅₀s], >32 and 12.3 μM, respectively). The parallel screening of replicon-carrying cells resulted in the identification of compound 3, a 1,5-BZD analog that displayed better cellular activity than compounds 1 and 2 (EC₅₀ = 5.8 μM). This analog was confirmed to inhibit NS5B in the enzymatic assay (IC₅₀ = 3 μM). The structural similarity observed between BZDs 1 and 3 prompted us to investigate compounds 4 and 5 (Table 1), two analogs that combine the various features of 1 and 3, i.e., a chloro atom at the ortho or meta position of the phenyl ring and the O-benzyl moiety at the para position, respectively. As expected, compounds 4 and 5 were found to be more potent than 1 and 3 (IC₅₀s, 1.8 and 0.9 μM, respectively).

TABLE 1.

IC₅₀s, EC₅₀s, and 50% cytotoxic concentrations (CC₅₀s) of compounds 1 to 7^a

^aIC₅₀s for the NS5BΔ21 enzyme, EC₅₀s for cell-based replicons, and 50% cytotoxic concentrations (CC₅₀s) in the replicon counterscreen are expressed as mean values ± standard deviations (SD) from 2 to 11 independent experiments performed in quadruplicate. Ac, acetyl; Me, methyl.

The in vitro binding of compounds 3a, 4a, and 5a to the polymerase and polymerase inhibition is stereospecific.

The enantiomers of compounds 3, 4, and 5 were separated by chiral high-pressure liquid chromatography, yielding the pure enantiomeric forms 3a, 4a, and 5a and 3b, 4b, and 5b, which were then tested in the biochemical assay. Only 3a, 4a, and 5a inhibited the NS5B polymerase, indicating that inhibition by these molecules is stereospecific (Table 2). However, no clear difference could be observed in the cellular assay, raising the question of whether the observed inhibition by the enantiomers 3a, 4a, and 5a is specific with replicons. To confirm binding specificity in vitro, we evaluated the propensity of these enantiomers to bind to the polymerase in an assay measuring the quenching of intrinsic tryptophan fluorescence. Consistent with the biochemical data, we found that 3a, 4a, and 5a considerably quenched the intrinsic fluorescence of the polymerase and that the effect observed with 3b, 4b, and 5b was minor and similar to that observed with derivative 6, an analog found to be inactive in the RdRp assay (Fig. 1). These data suggest that the enantiomers 3a to 5a, and not 3b to 5b, bind stereospecifically to NS5B.

TABLE 2.

IC₅₀s, K_i values, and EC₅₀s of enantiomers of compounds 3, 4, 5, and 7^a

Compoundb	Primer-dependent assay IC50 (μM)	Primer-independent assay IC50 (μM)	Ki (μM)	EC50 (μM) for Huh7-CMV-Luc cells	CC50 (μM) for MT4-LTR-Luc cells	CC50 (μM) for Huh7-CMV-Luc cells
3a	1 ± 0.7	0.54	NDc	15.3 ± 1.6	22.2 ± 7.3	19.2 ± 8.3
3b	86.7 ± 36.9	79.1	ND	2 ± 0.9	18.3 ± 4.3	19.7 ± 3.2
4a	1.0 ± 0.6	0.9 ± 0.2	1.7 ± 1.2	1.9 ± 0.5	16.4 ± 3.0	>32.0
4b	35.92	>32	NAd	0.6 ± 0.4	12.9 ± 1.8	21.2 ± 9.3
5a	0.33 ± 0.32	ND	ND	9.3 ± 2.2	19.5 ± 6.8	49 ± 57
5b	83.1	ND	ND	3.9	17.3	53.7
7a	0.05 ± 0.04	0.05	ND	1.2 ± 0.6	16.2 ± 3.9	12.8 ± 4.7
7b	12.5 ± 7.2	18.5	ND	2.1 ± 1.3	14.4 ± 1.3	10.1 ± 3.2

^aIC₅₀s and K_i values of enantiomers of compounds 3, 4, 5, and 7 for the NS5BΔ21 enzyme and EC₅₀s determined in cell-based replicon and replicon counterscreen analyses are expressed as mean values ± SD from 2 to 23 independent experiments performed in quadruplicate. Data without SD were tested only once in quadruplicate.

^bAll “a” eutomers (6) bear the benzyloxy-chlorophenyl moiety below the plane as it is illustrated in Fig. 4B. According to the Cahn-Ingold-Prelog nomenclature rules, the configuration at the chiral center is S for 3a and 4a and R for 5a and 7a.

^cND, not determined.

^dNA, not applicable.

FIG. 1.

Results of the fluorescence-quenching assay. Shown are isotherms for the binding of 1,5-BZDs with the NS5BΔ21 (1b J4) enzyme. The level of quenching of intrinsic tryptophan fluorescence, F₀-F, as a function of the compound concentration at an emission maximum of 335 nm was calculated by the subtraction of the fluorescence (F) of the polymerase-inhibitor complex from the initial fluorescence (F₀) of the polymerase only. ♦, compound 4; ▴, compound 4a; ▾, compound 4b; •, compound 6. The isotherm patterns for the binding of compounds 3, 3a, and 3b and 5, 5a, and 5b with NS5BΔ21 are similar to the pattern for compounds 4, 4a, and 4b (data not shown).

It has been speculated previously that NS5B replicates HCV RNA in cells through a primer-independent de novo mechanism (8, 24, 39, 50). To investigate whether the discrepancy observed in the cellular assay was due to the fact that the enantiomers 3b to 5b inhibited NS5B only by a de novo mechanism, we tested the BZD analogs 4a and 4b in a primer-independent assay using a transcript corresponding to the 3′ untranslated region of the HCV genome. Consistent with the results of the primer-dependent assay, we found that only 4a inhibited the polymerase in the de novo assay (Table 2). These data indicate that the observed activities of the 3b to 5b enantiomers in cells were not due to differences in the RdRp mechanism. Next, we used medicinal chemistry to improve upon the potency of the initial hits, which resulted in the identification of the 1,5-BZD 7 (IC₅₀ = 0.040 μM) (Table 1). Consistent with the data on compounds 3a to 5a, we found that the activity of the enantiomers of compound 7, 7a and 7b, could be clearly differentiated in the biochemical RdRp assay but not with replicons (Table 2) (IC₅₀s, 0.05 and 12.5 μM; EC₅₀s, 1.2 and 2.1 μM, respectively). No cytotoxicity of 7a and 7b in the replicon, MT4, HepG2, and HEK293T cell lines was observed (Table 3). We also assessed the selectivity of 7a and 7b against BZD receptors in a radioligand binding assay and found no binding to the γ-aminobutyric acid type A (GABA_A) receptor and the peripheral BZD receptor (PBR) (Table 3).

TABLE 3.

Compound 7a and 7b CC₅₀s and IC₅₀s for BZD receptors^a

Compound	CC50 (μM) for:				IC50 (mM)d for:
	MT4 cellsb	Huh7 cellsc	HEK293T cellsc	HepG2 cellsc	GABAA CBR	PBR
7a	56.5 ± 6.6	46.2 ± 2.4	>25	31.2 ± 7.4	>30	>30
7b	38.4 ± 23.4	23.8 ± 11.5	>25	>25	NAe	>30

^aData are mean values ± SD from two to eight independent experiments performed in quadruplicate. Data without SD were tested only once in quadruplicate.

^bCytotoxicity was measured by determining intracellular levels of ATP.

^cCytotoxicity was measured by resazurin staining.

^dThe BZD receptor assays were performed at MDS Pharma Services, Ltd. (Taiwan) in a single experiment in duplicate.

^eNA, not applicable. Slight stimulation (20%) instead of inhibition was observed.

1,5-BZD 4a is noncompetitive with regard to GTP.

To investigate the mode of inhibition, the capacity of 1,5-BZD 4a to inhibit replication in the presence of high or low concentrations of radioactively labeled GTP was assessed. The dose-response curves were not affected by various concentrations of the nucleotide substrate (Fig. 2). IC₅₀s were 1.0 and 1.3 μM for GTP concentrations of 20 nM and 200 μM, respectively. The K_m for GTP (mean ± standard error of the mean) of 0.3 ± 0.1 μM was consistent with previous findings (40) and was not affected by increasing concentrations of 4a. Furthermore, a double-reciprocal plot of the initial velocities revealed a noncompetitive mechanism of inhibition (Fig. 3B). Note that the intercept was shifted below the x axis, indicating a deviation from true noncompetitive inhibition, also referred to as mixed inhibition (2). Consistent with this pattern, α is less than 1 (0.5 ± 0.1). In the direct plot (Fig. 3A), all hyperbolic curves flatten and reach lower saturation values. According to formula 1 (see Materials and Methods), the K_i for 4a was 1.7 ± 1.2 μM.

FIG. 2.

Influence of GTP on the inhibitory activity of 1,5-BZD 4a on HCV polymerase activity. The NS5BΔ21 (1b J4) HCV polymerase was incubated together with a poly(rC)/oligo(rG₁₃) template, different concentrations of the inhibitor, and either 20 nM or 200 μM GTP. Dose-response curves were generated by nonlinear regression to determine the IC₅₀s. The experiment was performed once in quadruplicate. The data are presented as means ± standard errors of the means.

FIG. 3.

Determination of the mode of inhibition and K_i of 4a. (A) Reaction velocities at different GTP concentrations in the presence of either 0 (○), 0.25 (•), 0.5 (▴), 1 (▾), 2 (♦), 4 (▪), or 8 (×) μM inhibitor and a fixed primer/template concentration were measured. (B) Lineweaver-Burk plot. The lines on the double-reciprocal plot intersect in the third quadrant, reflecting a case of (mixed) noncompetitive inhibition. Nonlinear regression according to formula 1 gives a K_i of 1.7 ± 1.2 μM and α of ≠1. These graphical representations resulted from one experiment performed in quadruplicate. The data are presented as means ± standard errors of the means. The reported K_i was calculated from the results of two experiments performed in quadruplicate and is expressed as the mean ± standard error of the mean.

X-ray structure analysis of NS5B in a complex with 4a.

To guide our medicinal chemistry effort aimed at optimizing the 1,5-BZDs, we determined the X-ray structure of 4a bound to the recombinant C-terminally truncated NS5BΔ21 (1b J4) enzyme. A detailed analysis of this and related cocrystal structures will be published elsewhere, in conjunction with more extensive medicinal chemistry results related to this series. Here, we present a brief overview of the structure of the 4a-NS5B complex. The structure shows the typical right-hand shape of an RNA polymerase (Fig. 4A), with the palm, thumb, and finger subdomains organized around a central cleft that defines the active site (37). Compound 4a is observed to be bound at the NNI-3 binding site (40), adjacent to the polymerase active site. There are two molecules in the asymmetric unit of the new structure reported here, and the superposition of all common backbone atoms upon the previously reported apopolymerase structure of NS5B (38) gave root mean square deviation values of 0.45 and 0.51 Å for the various monomer-monomer pairs, respectively. This difference is similar to the value of 0.46 Å observed for the superposition of the two monomers of our new structure. Thus, the binding of 4a does not involve any significant conformational adjustment of the protein.

In our previous study, we provided detailed descriptions of the residues constituting each of the three NNI binding sites (40). Here, we present a slightly revised list of the residues constituting the NNI-3 site, as specifically defined by the binding of 4a to NS5B (Table 4). The amino acid residues forming the binding site and the bound ligand 4a are well-defined in the electron density map (data not shown), and the difference map of bound 4a supports the S configuration for the chiral center, consistent with small-molecule crystallography results for related analogs in this chemical series (data not shown). Figure 4B provides a two-dimensional schematic of the inhibitor-enzyme binding contacts in the 4a-NS5B complex. Compound 4a interacts with the NNI-3 site primarily through hydrophobic and aromatic binding contacts. Much of the BZD fused-ring system, as well as the N-acetyl substituent, is relatively exposed to the solvent (Fig. 4B). A large network of aromatic-aromatic interactions, involving Tyr191, Phe193, Tyr415, Trp420, Tyr448, and Phe551, also includes the two exocyclic phenyl rings of bound 4a in the complex (Fig. 4B and C). The benzyloxy-chlorophenyl moiety is the most extensively buried part of the bound inhibitor. The terminal benzyl group is deeply buried in the NNI-3 site in a pocket defined by Pro197, Arg200, Leu384, Met414, Tyr415, and Tyr448 (Fig. 4B and C). There is only one intermolecular hydrogen bond observed, between the exocyclic carbonyl O of 4a and Tyr448-N (Fig. 4B and C). In addition, the oxygen of the benzyl ether makes three fairly close (3.5- to 4.0-Å) orthogonal contacts with the guanidine group of Arg200 (Fig. 4B; contacts are not shown explicitly in Fig. 4C).

TABLE 4.

1,5-BZD binding site NNI-3 on NS5B^a

Genotype	Residue at position:
	193	197	200	316	319	366	367	368	384	407	410	411	414	415	446	447	448	449
1b J4	F	P	R	N	D	C	S	S	L	S	G	N	M	Y	Q	I	Y	G
1a_2	−	−	−	C	−	−	−	−	−	−	−	−	−	F	E	−	−	−
1a_6	−	−	−	C	−	−	−	−	−	−	−	−	−	F	E	−	−	−
1b_9	−	−	−	C	−	−	−	−	−	−	−	−	−	−	−	−	−	−
1b_10	−	−	−	C	−	−	−	−	−	−	−	−	−	−	−	−	−	−
2a_11	−	−	−	C	−	−	−	−	−	−	−	−	Q	−	E	M	−	−
2b_12	−	−	−	C	−	−	−	−	−	−	−	−	Q	−	E	M	−	−
3a_13	−	−	−	C	−	−	−	−	−	−	−	−	−	−	E	M	−	−
3a_14	−	−	−	C	−	−	−	−	−	−	−	−	−	−	E	M	−	−
4a_26	−	−	−	C	−	−	−	−	−	−	−	−	V	−	D	M	−	−
5a_29	−	−	−	C	−	−	−	−	−	−	−	−	−	−	E	M	−	−
6a_28	−	−	−	C	−	−	−	−	−	−	−	−	−	F	D	−	−	−
Conservation (%)	99.5	99.3	99.8	76.5 (C), 23.0 (N)	99.7	99.5	99.8	99.5	99.8	99.5	99.8	99.7	88.4	78.2 (Y), 21.6 (F)	59.6 (Q), 29.5 (E)	83.2 (I), 16.4 (M)	99.7	99.3

^aThe discontinuous BZD binding site was defined in the context of the NS5B-compound 4a complex as any NS5B residue with one or more atoms within 4 Å of any atom of bound 4a. The binding site alignment across genotypes is reported relative to the 1b J4 enzyme and was obtained from the European HCV Database (euHCVdb; 597 sequences were retrieved on accession date 23 November 2007) (12). Dashes represent residues identical to that in the 1b J4 sequence. The percent conservation reported for each position is that of the 1b J4 residue unless otherwise stated. The percent conservation of the most abundant variant is also listed for those positions at which the level of conservation of the 1b J4 residue is low.

Specific inhibition of HCV replication by 1,5-BZD 7a in replicon-carrying cells.

We have recently described the use of a biochemical assay as a means to elucidate the sites of action of novel inhibitors (40). In the present work, we extended this panel to include a replicon NS5B NNI site mutant panel as an alternative to direct resistance selection and used this tool to investigate if the inhibition of 7a and 7b in cells was mediated through binding to NS5B. We selected and synthesized indole 55 (20), thiophene 2 (40), GSK625433 (21), and HCV-796 (26) as reference compounds for NNI-1, NNI-2, NNI-3, and NNI-4, respectively. Replicons carrying the P495L, M423T, M414Q, and C316Y mutations were engineered for resistance to NNI-1, NNI-2, NNI-3, and NNI-4 inhibitors, respectively. In addition, we included H95R, a previously described mutation conferring resistance to the NNI-3 benzothiadiazine class (47), and engineered the NNI-3 inhibitor-resistant Y448A mutant on the basis of the results obtained with our X-ray structure (Fig. 4B). Using a transient-replication assay, we determined the EC₅₀s of selected inhibitors for each replicon mutant, and the EC₅₀s were then compared with the value measured for the wild-type ET replicon and reported as n-fold changes in the EC₅₀ (Fig. 5). As expected, each reference compound inhibited each replicon except for the cognate mutated NS5B replicon. The NNI-3 reference compound GSK625433 also lost potency against the NNI-4 C316Y mutant. This result is not surprising since the NNI-3 and NNI-4 binding sites overlap with each other. Interestingly, HCV-796 showed exquisite specificity to NNI-4 in this panel. Next, we used our replicon mutant panel to characterize the 1,5-BZD enantiomers 7a and 7b. Figure 5 shows that 7a inhibition was affected by the NNI-3 and NNI-4 mutations, as was also observed for GSK625433. However, 7b was affected by none of the mutations, suggesting that only 1,5-BZD 7a, and not 7b, binds to NS5B in the cell. Altogether, these data suggest enantiomer-specific off-target activity of 7b in cells.

FIG. 5.

Inhibitory activities of 1,5 BZDs and NNI reference compounds toward the replicon NS5B NNI site mutant panel. The levels of in vitro activity of compounds 7a and 7b and the thiophene, benzofuran, acyl pyrrolidine, and indole reference compounds toward the NS5B P495L, M423T, M414Q, Y448A, H95R, and C316Y replicon mutants are shown. The mutations were engineered by site-directed mutagenesis, and the in vitro transcription and transient-replication assays were performed as described previously (27). EC₅₀ changes (n-fold) were calculated with respect to the EC₅₀ of each class of compounds for the ET replicon. Compounds with an EC₅₀ change of greater than 10-fold (dotted line) were considered to be inactive. The experiment was performed three times in quadruplicate. The data are presented as means ± standard errors of the means.

Compound 4a inhibits the formation of the first phosphodiester bond during the polymerization reaction.

Although the mechanisms of inhibition of NNIs differ across the binding sites (3, 4, 15, 43), NNIs generally block the initial steps of the polymerization cycle that take place during the transition of the polymerase closed active form to the open inactive form. It has been proposed previously that these steps can be broken down into an initiation phase for the formation of dinucleotide products (P2) and a transition phase for the formation of polynucleotide products containing up to 5 nucleotides (P5) before NS5B proceeds to the processive elongation mode, which is associated with conformational changes that yield the open form (16). The transitions of P2 to P3 and P5 to P6 were found to be the rate-limiting steps. To assess the molecular mechanism of inhibition of a pyranoindole analog, Howe et al. designed an RNA template that enables the differentiation of di-, tri-, tetra-, and pentanucleotide products among all the products that are generated from the 20-mer RNA on a polyacrylamide gel (23). Due to the sequence of the RNA template, the NS5B-transcribed products, from P5, P4, P3, or P2 to full-length products, are visible on a gel when radioactive UTP, CTP, ATP, or GTP, respectively, is added to the RNA-NS5BΔ21-NTP mix (Fig. 6A, lanes 1 to 4). We used the same gel-based assay to study the mechanism of action of 4a. When increasing dose concentrations of 4a were added to the polymerization mixture spiked with radioactive GTP, we observed a gradual product decrease from full-length to dinucleotide in a dose-dependent manner, but not to the degree of complete inhibition at the highest inhibitor concentration (Fig. 6A, lanes 5 to 12). This result was due probably to the enzyme concentration and the incubation time that were necessary to generate quantifiable products (see Materials and Methods). To address this issue, we performed a time course experiment that included the following modifications: the incubation time was gradually increased, and the concentration of 4a was kept constant. Under these conditions, we found that the extended products could be fully inhibited after a shorter incubation time (Fig. 6B, compare lanes with and without compound 4a). Altogether, these results indicate that 4a inhibits the formation of the first phosphodiester bond during the polymerization reaction.

FIG. 6.

Results of the gel-based de novo initiation assay. (A) The NS5BΔ21 (1b J4) enzyme was incubated with substrate RNA and increasing concentrations of inhibitor 4a (1.0, 2.5, 5.0, 7.5, 10.0, 12.5, 15.0, and 20.0 μM in lanes 5 to 12, respectively). The template and primer sequences are indicated above the panel and are as described previously (23). The reaction mixtures in lanes 1, 2, 3, and 4 were spiked with UTP, CTP, ATP, and GTP, respectively. The P2, P3, P4, and P5 polymerization steps that correspond to the di-, tri-, tetra-, and pentanucleotide products, respectively, are indicated to the left of the panel, together with the products that are synthesized. The bands observed above the full-length product (FL) result most likely from template switching (30). (B) Results of the time course experiment. The reaction time points were 0 and 45 s and 1, 2.5, 5.0, 10.0, 15.0, 30.0, and 60 min, from left to right, for lanes both with (+) and without (−) compound 4a. The concentration of 4a was 10 μM.

Genotypic profiling of 4a.

We investigated the genotypic coverage of 4a by using a panel of recombinant NS5B enzymes, as we have reported previously for other compound classes (40). This panel of enzyme isolates was constructed from clinical sera and covered genotypes 1a, 1b, 2a, 2b, 3a, 4a, 5a, and 6a. Increasing concentrations of 4a were tested against all these enzymes in the RdRp assay, and the degree of inhibition is presented as the change (n-fold) in the IC₅₀ relative to that for the NS5B 1b J4 enzyme. We found that 4a inhibits only genotypes 1a and 1b (Fig. 7A.). This profile is reminiscent of that observed for a benzothiadiazine analog, another chemotype that binds at NNI-3 (45). Previously, we have shown that the systematic mutation of the NNI-3 residues that vary across the enzyme isolates back to the 1b J4 sequence can be used to identify the binding site determinants that elicit the different inhibition profiles observed for the benzothiadiazine analog against genotypes 2a and 3a. This panel of mutant enzymes was also used in this study, since the 1,5-BZD 4a binding site overlaps with the benzothiadiazine binding site (Table 4). When 4a was tested against this mutant enzyme panel, we found that the Q414M mutation (in genotype 2a_11) allowed the 4a analog to inhibit genotype 2a only partially (Fig. 7B, lane 1 and 2); this outcome is consistent with what was observed previously with the benzothiadiazine analog (40) and indicates that a polar change in this lipophilic pocket around position 414 is not tolerated by the 1,5-BZD analog 4a. Similarly, for genotype 3a, we found that the E446Q and G556S mutations permitted only modest BZD inhibition. On the other hand, 4a lost even more inhibitory activity toward the 3a_13 M447I mutant enzyme, unlike the benzothiadiazine analog. Altogether, these data do not fully explain the loss of inhibition observed for genotypes 2a and 3a and suggest that 4a cannot adapt to the β-flap walls shallower than that of the genotype 1b enzyme, as we suggested previously for an acyl pyrrolidine analog (40).

FIG. 7.

Genotypic profiling of 4a. (A) Inhibitory activity of 4a against a panel of enzyme isolates. The NS5B sequences of these clinical isolates were previously deposited in GenBank and assigned accession numbers (40). IC₅₀ changes (n-fold) are reported relative to the IC₅₀ for the NS5BΔ21 (1b J4) enzyme. (B) Inhibitory activity of 4a toward mutant enzyme isolates. Mutant forms of 1b J4, 2a_11, and 3a_13 enzymes engineered by site-directed mutagenesis were profiled against 4a. The experiment was performed at least twice in quadruplicate. The data are presented as means ± standard errors of the means.

Selectivity.

To define the specificity of this series for the HCV target, compounds 1 and 2 were profiled against a panel of recombinant human polymerases, HIV-1 RT, and dengue virus polymerase (Table 5). These analogs were found not to inhibit any of these polymerases except for a minor inhibitory activity of compound 1 against the human polymerase γ (30% inhibition at 100 μM).

TABLE 5.

IC₅₀s of compounds 1 and 2 for human polymerases, HIV-1 RT, and dengue virus polymerase

Compound	IC50 (μM) for:
	Human polymerase α	Human polymerase β	Human polymerase γ	HIV RT	Dengue virus polymerase
1	>100	>100	>100a	>32	>100
2	>100	>100	>100	>32	>100

^aAt 100 μM, the highest dose with which the assay was performed, the level of inhibition was 30%.

DISCUSSION

BZDs are widely prescribed drugs in the Western world in part because of their efficacy, safety, and low cost (29). They are used for the treatment of epilepsy, schizophrenia, insomnia, anxiety, alcohol withdrawal, and muscle cramps. The therapeutic effect is mediated through binding to the central BZD receptor (CBR) located on the GABA_A supramolecular complex, a brain neurotransmitter (7). The role of the second receptor, the PBR, in the clinical action of the drug is not clear (9). BZDs can also mediate effects independent of binding through the CBR and PBR, as illustrated by the findings for inhibitors of HIV-1 RT (41), respiratory syncytial virus (10), and B-cell proliferation (44). Here, we report BZDs capable of inhibiting the HCV polymerase. Unlike the inhibitors described above that are all 1,4-BZDs, the HCV polymerase inhibitors are 1,5-BZDs. Related compounds within this class do not bind to the CBR and PBR. We have shown that 1,5-BZD 4a is an NNI that binds next to the active site, in NNI-3. Using a panel of enzymes from clinical isolates of genotypes 1 to 6, we found that inhibition by compound 4a is limited to genotype 1, consistent with what has been reported previously for other NNI-3 chemotypes (40, 45). Finally, in an effort to understand the mechanism of action of NNI-3 analogs, we have shown that 4a inhibits RdRp activity by preventing the formation of the first phosphodiester bond.

The mechanism of nucleotide polymerization by NS5B remains controversial. The structural analysis of NS5B in a complex with ribonucleotides has revealed the presence of priming (P), catalytic (C), and interrogating (I) sites (5). In a process similar to the primer-independent de novo mechanism reported previously for bacteriophage φ6 (8), HCV polymerase initiates transcription by positioning the template into the RNA channel and two complementary ribonucleoside triphosphates at the P and C sites. This quaternary initiation complex must be stabilized in order for the formation of the first phosphodiester bond and subsequent template translocation to occur. It has been suggested previously that the binding of GTP to an allosteric site located 30 Å away from the catalytic site in close proximity to the fingertip Λ1 loop may stabilize the active conformation (5). This hypothesis is supported by the observation that part of the Λ1 loop which is α-helical in the closed active form refolds into a β-hairpin in the open form and that this change may disturb the integrity of the allosteric GTP binding site (3, 15). Consistent with this scenario, the binding of thiophene analogs to NNI-2 induces a shift of helix T which most likely also affects the integrity of the GTP binding site (4). A second key element that has been speculated to stabilize the initiation complex is the β-flap domain (Leu443 to Ile454). The β-flap domain protrudes toward the catalytic site and prevents the entry of double-stranded RNA. A β-flap deletion mutant is able to polymerize RNA by using a primer-dependent mechanism (22), and the β-flap domain was found previously to be altered in the open form of NS5B (3). Based on the fact that the β-flap is conserved across pestivirus, hepacivirus, and flavivirus polymerases and on the finding of another GTP binding site next to the P site which is required for de novo initiation in bovine viral diarrhea virus (11), Ferron et al. proposed a model in which the β-flap and the GTP at i-1, the position adjacent to the priming site, would provide a stacking platform to stabilize the initiation complex (18). Such a GTP cavity potentially exists in HCV polymerase and was proposed to involve residues Tyr195, Tyr448, Ser368, Met414, Arg386, Asn394, and Asn411 (18). Interestingly, this putative GTP site overlaps with the NNI-3 binding site. According to this model, the binding of NNI-3 ligands would prevent the assembly of the ribonucleotides at the P and C sites. Alternatively, NNI-3 ligands may prevent initiation by locking a closed conformation of the β-flap (43). Regardless of the mechanism, our data show that compound 4a inhibits the formation of the first phosphodiester bond, in contrast to an NNI-2 analog that has been shown to inhibit NS5B only after the synthesis of the first 5 nucleotides (23). Thus, the NNI-2 pyranoindole and the NNI-3 1,5-BZD 4a can be best described as transition and initiation inhibitors, respectively.

The binding of NNI-3 inhibitors to the β-flap wall appears to set a threshold in terms of genotypic coverage and the potential for resistance generation. All NNI-3 chemotypes described to date inhibit only genotype 1. Consistent with what we observed previously for a benzothiadiazine and an acyl pyrrolidine analog (40), we could not identify the determinant that causes the decreased inhibition of other genotypes. The loss of inhibition observed with the 3a_13 M447I mutant and the modest gain seen with the 3a_13 E446Q mutant cannot be fully explained from our X-ray structure, since Ile447 and Glu446 are pointing toward and away from the ligand, respectively. Since both residues are located within the β-flap, and the β-flap wall was found to be shallower in the X-ray structure of the genotype 2a enzyme than in that of the genotype 1b enzyme, it is tempting to speculate that the NNI-3 inhibitors cannot bind to a shallower β-flap wall in other genotypes. Based on the van der Waals contacts observed in the NS5B-4a bound complex, the replicon NS5B NNI site mutant panel data obtained with 7a, and the resistance patterns observed previously in replicon selection experiments with another NNI-3 analog (32, 35, 47), it may be expected that the 1,5-BZDs investigated here will select several resistant variants. The preliminary structure-activity relationship and X-ray analyses presented in this study provide the means to improve the inhibitory activity of 1,5-BZD. Future studies will address if the present 1,5-BZDs can provide a compound with improved potency, selectivity, and pharmacokinetic properties for consideration as a clinical candidate.