Encoded Library Technology Screening of Hepatitis C Virus NS4B Yields a Small-Molecule Compound Series with In Vitro Replicon Activity

Abstract

To identify novel antivirals to the hepatitis C virus (HCV) NS4B protein, we utilized encoded library technology (ELT), which enables purified proteins not amenable to standard biochemical screening methods to be tested against large combinatorial libraries in a short period of time. We tested NS4B against several DNA-encoded combinatorial libraries (DEL) and identified a single DEL feature that was subsequently progressed to off-DNA synthesis. The most active of the initial synthesized compounds had 50% inhibitory concentrations (IC₅₀s) of 50 to 130 nM in a NS4B radioligand binding assay and 300 to 500 nM in an HCV replicon assay. Chemical optimization yielded compounds with potencies as low as 20 nM in an HCV genotype 1b replicon assay, 500 nM against genotype 1a, and 5 μM against genotype 2a. Through testing against other genotypes and genotype 2a-1b chimeric replicons and from resistance passage using the genotype 1b replicon, we confirmed that these compounds were acting on the proposed first transmembrane region of NS4B. A single sequence change (F98L) was identified as responsible for resistance, and it was thought to largely explain the relative lack of potency of this series against genotype 2a. Unlike other published series that appear to interact with this region, we did not observe sensitivity to amino acid substitutions at positions 94 and 105. The discovery of this novel compound series highlights ELT as a valuable approach for identifying direct-acting antivirals to nonenzymatic targets.

INTRODUCTION

Significant progress has been made in recent years in the discovery and development of novel direct-acting antivirals (DAAs) to treat hepatitis C virus (HCV) infection. However, the ability of HCV to rapidly evolve and develop resistance is such that there is a continued need to discover DAAs that act through novel mechanisms. The standard of care for chronic HCV infection was a combination of pegylated alpha interferon and ribavirin up until the approval of the NS3 protease inhibitors telaprevir and boceprevir in 2011 (1, 2). Other classes of DAAs showing clinical efficacy target the NS5A protein (3, 4) or viral polymerase (NS5B; reviewed in reference 5). Of the other viral proteins, several small-molecule DAAs to nonstructural protein NS4B have been identified (6–14), but to date, there are no examples of clinically active NS4B-targeting compounds.

NS4B is a small (27-kDa), hydrophobic, membrane-associated protein that is derived through processing of the HCV polyprotein by the viral protease NS3/4A (reviewed in references 15, 16, and 17). It complexes with other HCV nonstructural and possibly a host cell factor(s) to form the viral replicase (18–22), which serves to replicate the viral RNA and is likely coupled to the assembly process (20, 23). Besides its structural role in the replicase, NS4B is thought to be involved in the induction of membranous vesicles that provide a platform for HCV RNA replication (24, 25), and it may also participate in virion assembly and release (26, 27). NS4B is also reported to hydrolyze nucleoside triphosphate (NTP) and nucleoside diphosphate (NDP) substrates and possess adenylate kinase activity (28).

The N-terminal domain of NS4B is cytosolic and contains a structurally resolved amphipathic helix (AH2) involved in oligomerization, endoplasmic reticulum (ER) targeting, and replication (6, 29–31). AH2 extends from amino acids 42 to 66 and has the ability to traverse the lipid bilayer following oligomerization (30). The central region of NS4B comprises several transmembrane (TM) domains which are involved in membrane anchoring and ER retention (32–34). The precise limits of each TM region have not been determined, but they are postulated to consist of a series of approximately 20-amino-acid stretches following the N-terminal AH2 (34). The C-terminal region of NS4B is palmitoylated (35) and contains two helical structures (H1 and H2) that self-interact and are essential for replication (25). The C-terminal domain is also thought to have an important role in the encapsidation process (23).

NS4B is not an obvious target for DAAs, because its enzymatic activity is difficult to measure biochemically. However, small molecules that bind NS4B have been discovered by exploiting the properties of AH2 to cause vesicle aggregation (6) or through targeting posttranscriptional mechanisms (9). We recently described a series of compounds that directly bound purified NS4B protein and were active against the HCV replicon (7, 10, 11). These compounds were structurally related to compounds shown to affect the subcellular distribution of NS4B (36, 37). We wanted to discover complementary scaffolds that would act through another mechanism or have a different resistance profile. For this, we took a nontraditional approach to screening using encoded library technology (ELT) (38–41).

ELT is a technology platform that uses DNA-tagged combinatorial libraries to identify small-molecule compounds that bind protein targets (Fig. 1). Each molecule in the ELT library comprises a drug-like moiety attached to a double-stranded DNA (dsDNA) coding region through an adapter module (the DNA headpiece). Each cycle of synthetic chemistry employed in the construction of the drug-like moiety is encoded (i.e., recorded) by ligation of a short dsDNA tag that identifies the building block added. Using split/mix methods, chemical diversity on the order of 10⁶ to 10⁹ drug-like moieties can be readily achieved. The libraries are then screened by affinity selection. Compounds that bind are separated from nonbinding molecules by multiple rounds of capture to an immobilized target protein of interest followed by heat elution. The identity of the binding compounds is then decoded by translation of the DNA tagging sequences. Because of the high sensitivity and throughput of modern DNA sequencers, nanomolar quantities of input library and microgram amounts of target protein are sufficient for selection experiments. The low material consumption allows multiple selection conditions to be tested in parallel. The technology is also particularly suited to targets like NS4B where novel mechanisms of action are desired.

FIG 1.

Schematic of ELT technology. (A) DNA-encoded chemical libraries are synthesized. (B) Affinity selection is performed to separate DEL molecules bound to a protein target from unbound DEL molecules. (C) DNA tags of bound DEL molecules are amplified by PCR. (D) Amplified DNA tags are sequenced by a high-throughput sequencing method. (E) Sequence information is processed and translated into chemical structures. (F) Chemotypes are identified through analysis of structural information. (G) Exemplar compounds from identified chemotypes are synthesized. (H) Assays are performed on synthesized compounds to confirm activity. C1, C2, and C3 indicate cycles 1, 2, and 3, respectively.

Through the use of ELT with purified NS4B protein to screen several DNA-encoded combinatorial libraries, we identified a novel class of compounds that bound NS4B in a radioligand displacement assay. Upon optimization, compounds were identified with <100 nM activity against several genotypes of HCV in the replicon system. These compounds were chemically distinct from our previously reported imidazo[1,2-α]pyridine series (7) and had a different resistance profile in tissue culture. We describe the discovery of this series and the biochemical and genetic methods used to confirm NS4B as its target and understand its mechanism of action.

MATERIALS AND METHODS

Cell lines.

Stable cell lines carrying a bicistronic genotype 1a (H77), genotype 1b (Con1 strain with cell culture-adapted mutations) or genotype 2a (JFH-1) replicon were created in-house, licensed from ReBLikon GmbH (Mainz, Germany) or constructed in-house from HCVcc virus licensed from Apath, LLC (Brooklyn, NY), respectively (42, 43). All three replicons express firefly luciferase and neomycin phosphotransferase and were maintained and propagated as described previously (4). ET-cured cells are a derivative of replicon-containing genotype 1b (Con1 strain) cells generated by treating Con1 cells with alpha interferon for several passages until HCV RNA levels were undetectable.

Preparation of purified NS4B protein and radioligand binding assay.

NS4B was expressed and purified for use in binding assays and ELT and tested in a binding assay as described previously (7). The binding assay essentially measured the displacement of a radiolabeled compound known to bind NS4B (compound 1d in reference 7; dissociation constant [K_d] of 30 nM) and was performed using a 125 nM concentration of the C-terminally His-tagged NS4B protein of genotype 1b and a 20 nM concentration of the radioligand.

ELT selection.

ELT selection was performed as previously described with some modifications (41). FLAG-His-tagged NS4B (5 μg) was immobilized on 5 μl of anti-FLAG resins (Phynexus). Twenty-eight DNA-encoded combinatorial libraries (DEL) (2.5 nmol) in 60 μl of selection buffer (50 mM Tris-HCl [pH 7.5], 150 mM NaCl, 0.1% Tween 20, 1 mg/ml sheared salmon sperm DNA [Ambion], 1 mg/ml bovine serum albumin [BSA] [Ambion]) were incubated with the immobilized NS4B for 1 h at room temperature and then washed five times with 100 μl of selection buffer to remove unbound DEL molecules. To elute bound molecules, resins were incubated in 60 μl of selection buffer at 80°C for 10 min. The eluent was incubated with 5 μl of anti-FLAG resins for 22 min to remove denatured NS4B before the next round of selection. This process was repeated two additional times. The same procedure was followed for the no-target control except naked anti-FLAG resins were used instead of anti-FLAG resins with NS4B immobilized. The output of selection was quantified by quantitative PCR and sequenced by 454 technology (Roche). Based on sequence information of DNA tags in eluted DEL molecules, chemical structures of DEL moieties were identified. Individual compounds representing enriched families of hits were synthesized off DNA using O-(7-azabenzotriazole-1-yl)-N,N,N,N′-tetramethyluronium hexafluorophosphate (HATU)/diisopropylethylamine (DIEA) for amide bond formation, and NaOH-mediated SNAr addition. Final compounds were purified by preparative liquid chromatography (LC).

Stable HCV replicon assay.

Replicon assays were performed as described previously (4). Briefly, stable-replicon-containing cell lines were seeded at a density of 2 × 10⁴ cells per well in 96-well assay plates in Dulbecco modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and penicillin-streptomycin. Compounds dissolved in 100% dimethyl sulfoxide (DMSO) were tested either by seeding the cells onto predosed plates or adding the compounds after the cells were plated. The plates were incubated for 1 to 3 days, and replication was monitored by determining firefly luciferase activity using Steady-Glo (Promega). Luminescence was measured in an EnVision 2103 multilabel reader (PerkinElmer). Cytotoxicity of compounds was measured on parallel plates using CellTiter-Glo (Promega). Replicon 50% effective concentration (EC₅₀) and CC₅₀ values, the concentration of compound required to inhibit 50% of the assay response, were calculated by curve fitting data to the Hill equation, using a nonlinear least-squares curve-fitting program.

Transient HCV replicon assay.

Transient transfections of replicon RNA derived from cDNA clones of wild-type, chimeric, or mutant constructs were performed as described previously (4). Essentially, RNA was generated from cDNA clones using the T7 RiboMAX express large-scale production system (Promega) and was transfected by electroporation into ET-cured Huh7 cells. For each transfection, 5 × 10⁶ cells were washed once with phosphate-buffered saline (PBS), resuspended in PBS, and electroporated in 0.4-cm cuvettes with 5 μg RNA, using a Bio-Rad Gene Pulser II set at 270 V, 950 μF, and infinite resistance. Electroporated cells were resuspended in prewarmed DMEM with 10% FBS. A total of 2 × 10⁴ cells were transferred to each well of a 96-well plate containing compounds or DMSO. Plates were incubated at 37°C and 5% CO₂ for 72 h, and firefly luciferase expression was measured as described above for stable replicons but with Bright Glo reagent (Promega).

Generation of HCV replicon chimeras and mutants.

Replicon genotype chimeras were constructed in which most of the NS4B gene was replaced with that from each of the major genotypes. We used both the HCV genotype 1b and 2a replicon cDNAs as backbones for the constructs to improve the possibility that we would generate chimeras that were capable of replicating. To retain the parental protease cleavage sites, the nucleosides encoding the N-terminal four amino acids and C-terminal six amino acids of NS4B were not substituted. Initially, for each of the major genotypes (2b, 3a, 4a, and 5a), we selected two sequences for cloning: a genotype consensus sequence derived from all available NS4B data from public databases and the individual database sequence most similar to the consensus sequence. For genotype 6, two different clones designated 6a and 6o were chosen to represent subsets of the available genotype 6 sequences. In addition for genotype 6, we constructed chimeras in which only amino acids 75 to 130 were substituted. Each NS4B sequence was artificially synthesized and substituted into both the genotype 1b and 2a transient-replicon backbones (GenScript USA Inc., Piscataway, NJ).

The following HCV strain and genotypes (GenBank accession numbers are shown in parentheses) were used in the construction of chimeras: JCH6, genotype 2a (AB047645.1), genotype 2b (AY232741.1), genotype 3a (AF046866.1), genotype 4a (DQ418788.1), genotype 5a (AF064490.1), genotype 6a (AY859526.1), and genotype 6o (EU246934.1). Individual mutations were synthesized in the replicons of interest by GenScript USA Inc., Piscataway, NJ.

Resistance passage using HCV replicon.

Stable-replicon cell lines that contained either HCV genotype 1a or 1b replicon were treated with GSK2188 at 10× EC₅₀ in 10-cm dishes. The medium and compound were changed every 3 or 4 days until colonies started to form (approximately 2 weeks). Total RNA from each plate was isolated with TRIzol (Invitrogen), and cDNAs were generated with a high-capacity cDNA reverse transcription kit (Applied Biosciences). The NS4B gene was amplified by PCR for population sequencing.

RESULTS

Identification of ELT features bound to HCV NS4B.

Purified NS4B protein was subjected to selection with 28 ELT libraries, which contain 1 million to 8 billion compounds with diverse chemical structures. Putative binders were identified from a library (DEL 54) containing 4 cycles of chemistry, wherein the first cycle consisted of 111 amino acids, the second of 16 dicarboxylates, the third of 134 diamines, and the fourth of 1,434 capping groups (carboxylic acids, isocyanates, sulfonyl chlorides, and heteroaryl chlorides). The data for this library can be summarized in a three-dimensional (3-D) scatterplot (Fig. 2), where building blocks corresponding to cycles 1, 3, and 4 are arrayed along the x, y, and z axes, and cycle 2 building blocks are coded by color. In this visualization, a threshold of two unique sequence reads was set, ensuring that points were significantly enriched. Two families of vertical lines were seen, corresponding to families of putative ligands defined by cycle 2, 3, and 4 building blocks but insensitive to the building block at cycle 1. One family was dominated by a bipiperidyl-triazine scaffold, whereas the other contained a spiro-diazaundecane pyrimidine core. Using standard synthetic methods, compounds representing the enriched families were synthesized off DNA for testing.

FIG 2.

Identification of chemotypes bound to HCV NS4B. (A) Graph of enrichment ratio (relative to no-target control) versus DEL library showing the generic structure of DEL 54. (B) Spotfire visualization of putative hit families from DEL 54, with chemistry cycle 2 depicted by color. Structures corresponding to the selected cycle 2–cycle 3–cycle 4 chemotypes are shown below along with minimized derivatives.

Compounds synthesized from DEL 54 are active in NS4B binding and HCV replicon assays.

Compounds derived from DEL 54 were tested for NS4B binding activity with a displacement assay that has previously been used to rank order the potency of NS4B inhibitors (7) using a radiolabeled inhibitor known to be active against this protein. One of the more potent of these first compounds, GSK2189 from the pyrimidine family, had a K_d of 53 nM in this assay (Fig. 3A and Table 1). It was subsequently tested for activity in an HCV genotype 1b replicon assay and was active with an EC₅₀ of 172 nM and toxicity (CC₅₀) of 39 μM (Table 2). It thus had a selectivity (ratio of activity to toxicity) of more than 200-fold. Compounds from the triazine family displayed only modest replicon activity and were not studied further.

FIG 3.

Structures of compounds derived from DEL 54. Et, ethyl group; tBu, tert-butyl group; Ph, phenyl group.

TABLE 1.

HCV NS4B binding activity of ELT-derived compounds

Compound	IC50 (nM) [95% CIa]	nb
GSK2189	52.9 [47.3–59.1]	11
GSK0109	45.8 [36.6–57.3]	11
GSK4809	42.1 [34.7–51.2]	15
GSK5874	61.0 [51.8–71.7]	4

^a95% CI, 95% confidence interval.

^bn is the number of replicates.

TABLE 2.

HCV stable-replicon activity of ELT-derived compounds

Compound	Genotype 1a assay		Genotype 1b assay		Genotype 2a assay		CC50 (nM) [95% CI]	n
	EC50 (nM) [95% CIa]	nb	EC50 (nM) [95% CI]	n	EC50 (nM) [95% CI]	n
GSK2189	2,222 [825–5982]	6	172 [117–252]	7	NDc		38,681 [37,979–39,396]	4
GSK0109	803 [699–923]	83	20 [18–22]	88	7,586 [4,926–11,683]	3	>43,469 [35,103–53,829]	11
GSK4809	782 [703–870]	101	60 [57–64]	106	5,168 [4,681–5,706]	21	>23,798 [19,534–28,993]	52
GSK5874	477 [264–864]	9	16 [13–20]	9	10,633 [7,760–14,571]	3	>50,119d	7

^a95% CI, 95% confidence interval.

^bn is the number of replicates.

^cND, not determined.

^dAll values were >50,000 nM.

Based on these initial findings, additional compounds were synthesized using the DEL 54 scaffolds as the templates to guide chemistry. Figure 3A shows the structure of one of the more potent compounds from this early optimization effort, GSK0109. While potency in the radioligand displacement assay was not improved significantly (Table 1; K_d of 46 nM), activity in the HCV genotype 1b replicon improved to 20 nM, with toxicity of >43 μM (Table 2; selectivity of >2,000). We also looked at replicon RNA levels in these assays using TaqMan as an alternative readout and observed similar EC₅₀s (data not shown).

HCV replicon activity of ELT-derived compounds differed according to genotype.

One of the goals of this project was to discover compounds that were efficacious across different genotypes of HCV, so we tested compounds against two additional stable replicons derived from genotypes 1a and 2a. Compounds were generally at least 10-fold less active against genotype 1a compared to genotype 1b (Table 2), with GSK2189 having an EC₅₀ of 2,222 nM and GSK0109 having an EC₅₀ of 803 nM. Against genotype 2a, they were even less potent, with an EC₅₀ for GSK0109 of 7,586 nM. While we were encouraged because the different potencies according to genotype suggested that the compounds were acting as direct antivirals (as opposed to interacting with host cellular functions), we needed to refocus the optimization efforts in order to generate compounds that were efficacious across different genotypes.

Subsequent work aimed to identify novel compounds with improved potency against genotypes 1a and 2a. Despite exhaustive efforts, we were not able to significantly improve upon the potency against HCV genotype 2a. Two of the better compounds from this optimization effort, GSK4809 and GSK5874 (Fig. 3B), showed increased activity against at least one of the genotypes but did not have overall improved profiles. Although GSK4809 had an EC₅₀ of 5,168 nM against genotype 2a and had about the same activity against genotype 1a as GSK0109, it was 3-fold less potent against genotype 1b (Table 2). Likewise, GSK5874 had slightly improved potency against genotypes 1a and 1b but was less active against genotype 2a (Table 2).

Activity of ELT-derived compounds against a comprehensive genotype panel.

We decided to obtain a better picture of the potency of these compounds across different genotypes by constructing chimeras of each major genotype using either the genotype 1b or 2a replicon as the backbone. NS4B sequences from these backbones were switched with the corresponding sequences from other genotypes, while preserving the original backbone sequence encoding the first 4 and last 6 amino acids of NS4B. The chimeras were tested for replication in a transient-replicon assay, and viable chimeras were subsequently used to test compound activity. Figure 4 shows a comparison of the sequences that were switched to create chimeras that replicated following transient transfection of Huh7 cells. Most of the chimeras made using the genotype 2a (JFH1) backbone were viable. The genotype 4a chimera did not replicate in this backbone, but it replicated weakly using the genotype 1b (Con1 strain) backbone. Genotypes 6a and 6o did not replicate at all in either backbone, but we were able to insert truncated versions of NS4B from these genotypes (amino acids 75 to 130) into the genotype 2a backbone to create viable chimeras.

FIG 4.

Alignment of NS4B amino acid sequences used in the construction of chimeric replicons. Chimeras of genotypes 1a (H77) and 4a were constructed using the genotype 1b (Con1 strain) backbone; hence, their sequences are aligned beneath that of genotype 1b. The remaining chimeras were constructed using the JFH1 (genotype 2a) backbone, and their sequences are shown beneath that of JFH1. Shading of amino acids in the JFH1 sequence indicates differences with the genotype 1b sequence. Lettering above the sequences denotes the 21-amino-acid blocks that were switched to create the genotype 1b-2a domain chimeras.

Table 3 shows the potency of two of the lead ELT-derived compounds, GSK0109 and GSK4809, against this panel of chimeric replicons and compares it to a previously reported imidazo[1,2-α]pyridine compound, GSK8853 (7). The ELT-derived compounds were reasonably potent (<500 nM) against genotypes 3a, 4a, and 5a, and against genotype 2b, they all had EC₅₀s of approximately 2,100 nM. GSK0109 was not active against the two genotype 2a replicons (JFH1 and JCH6) up to 10 μM, and only marginal activity was observed with GSK4809. Against genotype 6a, the ELT-derived compounds had potencies of <1,000 nM, and they were weakly active against genotype 6o. In comparison, GSK8853 was more potent against genotypes 1a, 1b, 4a, and 5a, had similar potency against genotype 2b, but was less potent against genotypes 6a and 6o.

TABLE 3.

Potency of compounds against different genotypes in a transient-replicon assay^a

Genotype (strain)	EC50 (nM) [95% CIb]
	GSK8853c	GSK0109	GSK4809
1a (H77)	0.88 [0.81–0.96]	965 [812–1,147]	1,525 [1,340–1,734]
1b (Con1)	0.77 [0.56–1.07]	9.17 [8.07–10.4]	50.1 [46.9–53.5]
2a (JFH1)	>10,000	>10,000	9,870 [9,660–10,084]
2a (JCH6)	>10,000	>10,000	9,557 [9,143–9,990]
2b	2,437 [2,049–2,899]	2,158 [1,770–2,632]	2,109 [1,992–2,233]
3a	42.6 [34.6–52.4]	43.8 [31.3–61.2]	102 [89–116]
4a	6.98 [3.13–15.6]	205 [168–251]	494 [392–623]
5a	2.53 [2.19–2.92]	26.6 [23.6–30.1]	72.9 [63.6–83.6]
6a	4,479 [4,194–4,783]	413 [363–472]	947 [853–1,050]
6o	>10,000	9,550 [8,977–10,159]	6,621 [5,290–8,288]

^aThe number of replicates was 9 in all cases.

^b95% CI, 95% confidence interval.

^cThe structure shown is compound 21 in Shotwell et al. (7).

L98 is a key amino acid involved in genotype 2a resistance to the ELT-derived compounds.

The genotype 1b replicon was passaged in the presence of approximately 10× EC₅₀ of one of the initial ELT hits, GSK2188 (a larger variant of GSK2189) (Fig. 3A), and 0.5 mg/ml G418 to select for replicon that was resistant to the compound. After about 2 weeks, resistant colonies were expanded, and replicon RNA was characterized by reverse transcription-PCR (RT-PCR) of the NS4B gene followed by sequencing. A single amino acid change, F98L, was identified. This was put back into the original genotype 1b replicon, and compounds were tested in a transient-replicon assay against this mutant.

Table 4 shows the activity of the ELT-derived compounds and a previously reported imidazo[1,2-α]pyridine compound, GSK8853, against the F98L amino acid substitution and two other substitutions, H94N and V105M, that were responsible for resistance to GSK8853 (7). Against H94N and V105M, there was no decrease in potency of the ELT-derived compounds, whereas F98L caused a 40-fold decrease in the potency of GSK0109 and a 120-fold decrease in the potency of GSK4809. We also determined the effect of an L98F substitution in the genotype 2a replicon and found that the potency improved from inactive (>10 μM) to 1.4 μM (data not shown).

TABLE 4.

Potency of compounds against transient NS4B mutants^a

Compound	EC50 (nM) for the wild typeb	H94N mutant		V105M mutant		F98L mutant
		EC50 (nM)	Fold change	EC50 (nM)b	Fold change	EC50 (nM)b	Fold change
GSK8853c	1.1	38.0	35	178	162	27.5	25
GSK0109	6.1	10.4	1.7	6.5	1.1	242	40
GSK4809	38.7	47.9	1.2	41.9	1.1	4,786	124

^aThe EC₅₀s for the transient NS4B mutants are shown in bold type.

^bData are averages of at least two replicates.

^cThe structure shown is compound 21 in Shotwell et al. (7).

Lack of genotype 2a potency of ELT-derived compounds maps to amino acids 76 to 109 of NS4B.

Although the presence of residue L98 in genotype 2a was determined to be a key determinant of sensitivity to this compound series, it alone did not account for all the loss of potency against genotype 2a. Furthermore, it was not present in genotype 1a, so it could not account for the loss of potency against this genotype.

To understand better the lack of potency of the ELT-derived compounds to genotype 2a, we generated a series of transient-replicon chimeras in which 21 amino acid sequence blocks in the genotype 1b replicon were replaced with the corresponding sequences from genotype 2a (Fig. 4). The five blocks from positions 68 to 193 roughly correspond to five transmembrane domains that have been proposed (33). This enabled us to look sequentially and rapidly across NS4B at the amino acid differences between genotypes 1b and 2a that may have contributed to the observed potency differences. Chimeras of regions A, G, and L, corresponding to amino acids 5 to 25, 131 to 151, and 236 to 256, respectively, did not give a sufficiently high luciferase signal in a transient-replicon assay and were not tested further. Region J was not changed, because the sequence is the same in genotypes 1b and 2a.

Table 5 shows the potency of GSK4809 against the genotype 2a chimera panel. Against all chimeras but chimeras D and E, it had EC₅₀s of <50 nM, similar to that of the genotype 1b replicon. Against chimera D (comprising amino acids 68 to 88 of genotype 2a), there was a 5-fold decrease in potency compared with genotype 1b, and against chimera E (amino acids 89 to 109 of genotype 2a), there was a 120-fold decrease. Regions D and E contain 5 and 9 amino acid differences, respectively, between genotypes 1b and 2a, spanning the region from amino acids 76 to 109. On the basis of these data, it was likely that all or some of these amino acid differences were responsible for the lack of potency against genotype 2a relative to genotype 1b. These results, combined with the binding data, suggested that the site of interaction of the ELT-derived compounds with NS4B was between amino acids 76 and 109.

TABLE 5.

Potency of compounds against genotype 2a chimera panel

Compound	Parameter	Value for parameter for strain or chimeraa
		Con1	2a-B	2a-C	2a-D	2a-E	2a-F	2a-H	2a-I	2a-K
GSK8853b	EC50 (nM)c	1.1	1.78	0.89	0.51	1,273	4.27	2.85	0.61	0.71
	Fold change	1	1.62	0.81	0.46	1,157	3.88	2.59	0.55	0.65
GSK4809	EC50 (nM)	41	21	33	202	4,868	36	37	18	21
	Fold change	1	0.51	0.8	4.92	119	0.88	0.9	0.44	0.51

^aThe values for parameters for strain Con1 or genotype 2a chimera panel are shown. The genotype 2a chimera panel are shown in the following format: 2a-B for genotype 2a chimera B.

^bThe structure shown is compound 21 in Shotwell et al. (7).

^cThe EC₅₀ data for the two compounds are shown in bold type. The data were derived from a single experiment performed in duplicate. EC₅₀s for chimeras 2a-D and 2a-E were confirmed in two additional experiments (data not shown).

Effects of individual amino substitutions in the NS4B region from amino acids 76 to 109 on the potency of GSK4809.

To map the contribution of individual amino acids to the shift in potency of this compound series against genotypes 1a and 2a, we also looked at the effects of various substitutions in the region of NS4B from amino acids 76 to 109 (Table 6). Between genotypes 1a and 1b, there are only 4 differences in this region. Of these, only S85A appeared to affect potency, causing a 3-fold loss of potency on its own or 3.6-fold as part of a double substitution with I86V.

TABLE 6.

Potency of GSK4809 against various amino acid substitutions in NS4B region from amino acids 76 to 109

Location and position(s)	Amino acid in the following genotype:			Substitution(s) testeda	Fold shift from wtb
	1b	2a	1a
2a-D					4.9
76	I	V		ND
79	L	M		ND
83	T	S		ND
85	S	A	A	S85A	3.0
86	I	L	V	I86V	1.7
85–86				S85A + I86V	3.6
2a-E					119
91	T	S		T91S	1.4
93	Q	S	G	Q93G	0.9
94	H	T	Q	H94N	1.2
93–94				Q93G + H94Q	0.6
96	L	I		ND
98	F	L		F98L	124
101	L	M		ND
105	V	L		V105M	1.1
107	A	S		ND
109	L	I		ND

^aND, not determined.

^bThe fold shift data from the wild-type (wt) value were derived from single experiments performed in duplicate. The data for chimeras 2a-D and 2a-E are shown in bold type.

In the region from amino acids 91 to 109 (region E in Fig. 4), which had shown the greatest effect on potency in the genotype 2a chimera panel (Table 5), only F98L caused a reduction in potency (of the substitutions tested), of similar magnitude to the shift seen with the 2a-E chimera. We did not test substitutions at positions 96, 101, 107, and 109, because the same amino acid is present at these positions in genotype 5a as in genotype 2a and we had not observed a shift in potency in this genotype (Table 3).

In summary, the resistance to genotype 2a compared to that to genotype 1b appears to derive largely from amino acid differences at positions 85 and 98. For genotype 1a, in which we observed a 30- or 100-fold shift in potency with GSK4809 or GSK0109, respectively (Table 3), we were not able to find amino acid substitutions, alone or in combination, that caused this degree of shift in potency in the NS4B region from amino acids 79 to 109. We did not investigate the loss of potency against genotype 6a or 6o further; however, both contain leucine at position 98, which is likely a key determinant based on the genotype 2a data.

DISCUSSION

We describe through the use of encoded library technology (ELT) the discovery of a novel chemical series of compounds with members such as GSK0109 and GSK4809 that bind to the HCV NS4B protein and have <100 nM activity against the HCV replicon in tissue culture. While this series is chemically distinct from previously reported series (6–9, 12–14), GSK0109 and GSK4809 appear to interact at a similar site on NS4B, although they give a partially different resistance profile in tissue culture. Notably, these compounds are not sensitive to amino acid substitutions at H94 and V105 in the genotype 1b replicon, which has been observed with inhibitors targeting NS4B (7–9).

We employed the approach of encoded library technology, which has not been reported previously for screening membrane-associated or viral proteins. ELT is based on affinity binding, so prior knowledge of mechanism of action is not required when working on a protein target, and discovery of active compounds is not limited to a certain binding site or a certain function. When we initiated the NS4B project, we did not know how many binding sites might exist for small-molecule compounds and whether a particular binding site would be relevant to its biological function. From our work, only one chemical series was identified from a collection of 28 DNA-encoded chemical libraries ranging from 1 million to 8 billion compounds with diverse structures. All data indicate that this lead series binds at the same site and therefore likely has a similar effect on biological function as our previously reported imidazo[1,2-α]pyridine compounds. Therefore, it is possible that only one binding site of NS4B exists for small-molecule active compounds. We ran similar screens with NS4B from genotypes 1a and 2a but did not identify additional ELT features (data not shown).

In the last few years, several research groups, including our group, have published inhibitors of NS4B (Fig. 5A). While several of the scaffolds are structurally related to the prototypical compound anguizole (6), the ELT series is clearly distinct from it and the other chemotypes shown. This outcome highlights the value of diversity screening in searching for novel chemical matter. It also raises the question of whether the new series binds to a distinct site and/or acts through an alternative mechanism. We showed through a displacement assay using a radiolabeled compound (7) that the ELT compound series likely bound the same site on NS4B as previously reported compounds, although displacement could also be caused by an allosteric effect on the binding pocket. The binding activity of different compounds appeared to have similar potency, which did not correlate with the observed differences in replicon activity. It is probable that the activity in this assay is related to the binding limit of the assay, because below a certain threshold for displacement, there is no differentiation in the IC₅₀ data. Flexible alignment of the known ligands shown in Fig. 5A using Molecular Operating Environment (Chemical Computing Group, Inc.) with default settings indicates that the ELT series can present in a geometry that is very similar to the other inhibitors (Fig. 5B). The pyrimidine core aligns with the indole/quinoline type cores, placing the aryl groups in positions similar to those of the substituents on the reported molecules. Whether these different series have similar mechanistic activities is not known. Effects on subcellular localization of NS4B were described by Bryson et al. (37), but were not observed for PTC725 (9). More in-depth studies for these different compounds are needed to define their activities more clearly.

FIG 5.

Recently published NS4B inhibitors. (A) Chemical structures. The structure shown for Shotwell et al. (7) is GSK8853. (B) Overlay of all structures except Gu et al. (9) using Molecular Operating Environment (Chemical Computing Group, Inc., Montreal, Canada). GSK0109 is highlighted in orange.

In a stable genotype 1b HCV replicon, we observed <100 nM potency against genotype 1b for several compounds. However, there was a significant reduction in potency against genotype 1a (Table 2: GSK0109 potency reduced 40-fold), and even more so against genotype 2a (GSK0109 potency reduced 380-fold). Despite a fairly thorough chemistry effort (>200 compounds synthesized), we were unable to find compounds that were more potent than the ones described against HCV 1b or that had a relative improvement of potency against genotype 1a or 2a. Reports from other chemical series have shown a similar loss of potency against genotype 2a (6, 8, 9, 13, 14), although in the case of Phillips et al. (14), while the potency against genotype 2a shifted about 30-fold compared with that against genotype 1b, it was still in the low nanomolar range.

To determine the resistance profile of our series and to better understand the underlying reasons for the observed shifts in potency against different genotypes, we performed serial passage of the genotype 1b replicon against one of the compounds (GSK2189). In addition, we determined the shift in potency against a genotype 2a chimera panel and looked at the effects of individual amino acid substitutions. Like reports from other series, the amino acid at position 98 was found to be a key determinant of potency (9, 12–14). This resides in the proposed first transmembrane region of NS4B (34). With respect to the PTC725 compound, a shift of >50-fold was observed with a substitution from F to C or L in the genotype 1b replicon (9). However, in our case F98L was the only amino acid substitution causing resistance identified from resistance passage, and we did not observe a change at V105, which had been noted for our imidazo[1,2-α]pyridine series and for PTC725. F98 is highly conserved (>97%) in genotypes 1 and 3, as is L98 in genotype 2 (9). It is possible that resistance passage with other more potent compounds in the series would have uncovered additional substitutions, but still not V105M, as this was subsequently shown as having no impact on the potency of GSK0109 or GSK4809 using site-directed mutagenesis (Table 4).

The presence of Leu at position 98 was likely the major reason for the loss of potency of GSK0109 and GSK4809 against genotype 2a. To determine whether other regions of genotype 2a were important to this shift in potency, we tested GSK4809 against a genotype 2a chimera panel and found that most of the shift was accounted for by the region comprising amino acids 89 to 109 (Table 5). Subsequent mutagenesis experiments suggested that in this region, Leu98 was likely the sole cause of this potency shift. Interestingly, we also observed a smaller but reproducible shift in potency from the region comprising genotype 2a amino acids 68 to 88. S85A was the only substitution found to have an effect on potency in this region and had not previously been identified in resistance passage. It is likely that the structure of the binding site of this and other series is also determined in part by interactions with other viral and host proteins, hence amino acids that are not directly involved in binding may influence the site via other mechanisms.

To better understand the mechanism of action of this series would require a more detailed investigation of how it affects interactions with other host and viral proteins and RNA and membrane association. It is possible that the series could act by disrupting oligomerization of NS4B, as has been described for another series that may be binding the same site (44). The second amphipathic helix of NS4B was shown to be important for the oligomerization process, as well as for binding NS5A and determining subcellular localization. At this time, we are unable to describe the mechanism of action of the ELT series more precisely.

We did not study the ELT compound series further because of lack of potency against genotype 2a. However, it could still be a valuable tool to understand NS4B biology, and it would be interesting to understand why it has a different resistance profile to other compound series that appear to be binding to the same site. NS4B is attractive as a target because it has been demonstrated by several groups that it is possible to identify small molecules that apparently bind this protein and inhibit HCV replication. Whether these compounds would have clinical efficacy if they had been studied further is not known. For most of these studies, it is not stated why they were discontinued, but likely it was from lack of cross genotype potency as in our case or for issues that arose in animal studies.

We made some attempts to improve cross genotype potency by performing ELT in which successive rounds of enrichment were run against different genotypes (data not shown). For example, the initial selection performed against NS4B genotype 1b was then mixed with genotype 2a protein to enrich for compounds that bound both. Unfortunately, we were not able to identify other scaffolds in these experiments. Overall, however, our work demonstrated how ELT, an alternative approach for drug discovery, could be effectively applied to a membrane-associated viral protein to identify inhibitors of replication. Hence, ELT is a versatile screening method and can be used to screen a wider range of potential drug targets than previously thought.