Identification and Characterization of a Small-Molecule Rabies Virus Entry Inhibitor

Rabies PEP depends on anti-RABV IgG, which is expensive and in limited supply in geographical areas with the highest disease burden. Replacing the IgG component with a cost-effective and shelf-stable small-molecule antiviral could address this unmet clinical need by expanding access to life-saving medication. This study has established a robust protocol for high-throughput anti-RABV drug screens and identified a chemically well-behaved, first-in-class hit with nanomolar anti-RABV potency that blocks RABV G protein-mediated viral entry. Resistance mapping revealed a druggable site formed by the G protein fusion loops that has not previously emerged as a target for neutralizing antibodies. Discovery of this RABV entry inhibitor establishes a new molecular probe to advance further mechanistic and structural characterization of RABV G that may aid in the design of a next-generation clinical candidate against RABV.

ABSTRACT

Rabies virus (RABV) causes a severe and fatal neurological disease, but morbidity is vaccine preventable and treatable prior to the onset of clinical symptoms. However, immunoglobulin (IgG)-based rabies postexposure prophylaxis (PEP) is expensive, restricting access to life-saving treatment, especially for patients in low-income countries where the clinical need is greatest, and does not confer cross-protection against newly emerging phylogroup II lyssaviruses. Toward identifying a cost-effective replacement for the IgG component of rabies PEP, we developed and implemented a high-throughput screening protocol utilizing a single-cycle RABV reporter strain. A large-scale screen and subsequent direct and orthogonal counterscreens identified a first-in-class direct-acting RABV inhibitor, GRP-60367, with a specificity index (SI) of >100,000. Mechanistic characterization through time-of-addition studies, transient cell-to-cell fusion assays, and chimeric vesicular stomatitis virus (VSV) recombinants expressing the RABV glycoprotein (G) demonstrated that GRP-60367 inhibits entry of a subset of RABV strains. Resistance profiling of the chemotype revealed hot spots in conserved hydrophobic positions of the RABV G protein fusion loop that were confirmed in transient cell-to-cell fusion assays. Transfer of RABV G genes with signature resistance mutations into a recombinant VSV backbone resulted in the recovery of replication-competent virions with low susceptibility to the inhibitor. This work outlines a tangible strategy for mechanistic characterization and resistance profiling of RABV drug candidates and identified a novel, well-behaved molecular probe chemotype that specifically targets the RABV G protein and prevents G-mediated viral entry.

IMPORTANCE Rabies PEP depends on anti-RABV IgG, which is expensive and in limited supply in geographical areas with the highest disease burden. Replacing the IgG component with a cost-effective and shelf-stable small-molecule antiviral could address this unmet clinical need by expanding access to life-saving medication. This study has established a robust protocol for high-throughput anti-RABV drug screens and identified a chemically well-behaved, first-in-class hit with nanomolar anti-RABV potency that blocks RABV G protein-mediated viral entry. Resistance mapping revealed a druggable site formed by the G protein fusion loops that has not previously emerged as a target for neutralizing antibodies. Discovery of this RABV entry inhibitor establishes a new molecular probe to advance further mechanistic and structural characterization of RABV G that may aid in the design of a next-generation clinical candidate against RABV.

INTRODUCTION

Rabies virus (RABV) is an enveloped, nonsegmented, and single-stranded RNA virus that belongs to the genus Lyssavirus within the family Rhabdoviridae (1). The virus is the causative agent of rabies, which continues to be a global public health concern that causes approximately 60,000 deaths annually (2).

Considered the most lethal vaccine-preventable viral disease, spillover of RABV into the human population occurs from a worldwide distribution of animal reservoirs. The majority of cases occur in Asia and Africa, where access to vaccination and postexposure prophylaxis (PEP) is most limited (3). RABV PEP consists of passive immunization with human-derived anti-RABV IgG (hRIG) combined with RABV vaccination, followed by three additional doses of vaccine administered on days 3, 7, and 28 of exposure (4). A frequently short supply and the high cost of hRIG has led some countries to explore alternative passive immunization with equine RIG (eRIG). Although effective, eRIG PEP has resulted in severe allergic reactions, including anaphylaxis and serum sickness (5–7). After the onset of clinical symptoms, RABV disease nearly inevitably progresses to death of the patient. Limited availability, high cost, and a continuous cold-chain requirement of the RIG component of RABV PEP create an urgent need for the development of safe and cost-effective replacements.

Small-molecule antiviral compounds offer some fundamental advantages over biologics, such as high shelf stability, manufacture in existing facilities, and scalable production (8). Furthermore, RIG replacements could broaden the antiviral indication spectrum, potentially also providing protection against recently emerged zoonotic phylogroup II lyssaviruses (9–11). However, automated large-scale discovery of novel RABV inhibitors has been impaired by the biocontainment requirements for work with live RABV and the lack of a discernible cytopathic effect associated with RABV infection in cell culture (12). Most modern antiviral high-throughput screening (HTS) campaigns use reporter viruses for a quantitative readout of viral replication (13–18). Bioluminescence-based reporters in particular provide a wide dynamic range, high sensitivity, and a linear signal correlation over several orders of magnitude, making them preferable to fluorescence-based reporters such as green fluorescent protein (GFP) in primary screens (19). A recombinant RABV nanoluciferase reporter strain established proof of concept for successful application of the technology to anti-RABV screens, but the impact of this fully replication-competent reporter strain has remained limited due to biosafety requirements (20). Single-cycle reporter viruses that are capable of proceeding through an initial round of replication yet unable to produce pathogenic progeny (21) could offer a solution to the problem, as demonstrated by the precedent established with human immunodeficiency virus and hepatitis C virus reporter strains (22–24).

RABV G, the only protein expressed on the virion surface, is responsible for both receptor binding and fusion of the viral envelope with target cell membranes. Accordingly, the trimeric G homo-oligomer is the primary target for neutralizing antibodies used for RIG treatment (25–29). Receptor tropism of RABV is complex, since the acetylcholine receptor subunit alpha (30), the neural cell adhesion molecule (31), the low-affinity nerve growth factor receptor (32), and the metabotropic glutamate receptor 2 (33) in addition to carbohydrates, gangliosides, and lipids (34–36) have all been implicated in virus entry. RABV may depend on different receptors to accomplish distinct stages of central nervous system (CNS) invasion (37). Once attached to target cells, RABV virions are taken up endocytotically, and the increasing acidification of the maturing endosome triggers major conformational changes in the prefusion G trimer that ultimately mediate merger of the viral envelope with target cell membranes (38–40). Reflecting the critical role of G in viral entry, virions lacking G are noninfectious, although the absence of G does not prevent the assembly of naked virions (41). Providing G in trans to a RABV vector lacking G (ΔG RABV) restores infectivity of the first-generation progeny, but viral amplification is limited to a single round of replication (41, 42).

In this study, we pioneered a large-scale HTS campaign using a single-cycle RABV reporter virus with G deleted and expressing nanoluciferase as a primary readout. The assay was miniaturized to a 384-well plate format, validated, and applied to open-discovery compound libraries curated to minimize compounds with undesirable substructures. Hit candidates emerging from the primary campaign were subjected to primary and orthogonal counterscreens, and the mechanism of action of confirmed hits was determined. Having identified a first-in-class RABV entry inhibitor, GRP-60367, with nanomolar potency against some RABV strains and a specificity index (SI) of >100,000, we resistance profiled the chemotype, confirming escape mutation candidates in the context of transient cell-to-cell fusion assays and replication-competent recombinant viruses. These studies yielded a highly specific molecular probe directed against RABV G and identified a novel candidate druggable site that can be exploited without adverse effects on host cell viability.

RESULTS

RABV HTS assay development.

To generate a RABV reporter virus suitable for use in our HTS facility, we replaced the G protein-encoding gene in a cDNA copy of the RABV genome with a nano-luciferase gene (Fig. 1A). The resulting RABV-ΔG-nanoLuc virus was recovered and amplified on BSR-T7/5 cells stably transfected with an inducible RABV G protein derived from strain SAD-B19 under the control of a tetracycline-dependent promoter (Tet on). In preparation for large-scale library screening, we employed this reporter virus and a previously described broad-spectrum RNA virus blocker, JMN3-003 (15), to establish and quantitatively validate an automated drug discovery protocol in a 384-well plate format. After automated seeding of BEAS-2B target cells, plates were stamped with a set of three validation source plates that contained alternating columns harboring vehicle (V), JMN3-003 at 0.5 times the 50% effective dose equivalent (0.5× EC₅₀; referred to as low dose [L_D]), and JMN3-003 at 10× EC₉₀ (high dose [H_D]). These columns were arranged in a continuous V-L_D-H_D, L_D-V-H_D, or V-H_D-L_D scheme on three individual plates. Substance transfer was carried out with a 384-needle pin tool (20-nl transfer volume/needle). Stamped target plates were infected with resulting RABV-ΔG-nanoLuc, and relative bioluminescence was determined after 48 h incubation. Results of individual wells were alternatively plotted by raw and then column or by column and then raw (Fig. 1B) to scan for signal patterns arising from edge, column, or raw effects. After optimization of target cell density and virus inoculum dose, the assay returned under fully automated conditions a mean dynamic range (signal-to-background [S/B] ratio) of 76.7, a coefficient of variance (CV) of 17.4%, mean inhibition at 0.5× EC₅₀ of 17.6%, and overall robustness (Z′) of 0.51 across the three test plates. Visual inspection of plate heat maps and alternative plotting of results did not reveal undesirable signal patterns. This performance was within our predefined thresholds (S/B > 10; CV < 20%; Z′ >0.5), confirming suitability of the protocol for automated drug screening (16, 43).

FIG 1.

HTS campaign to identify small-molecule RABV inhibitors. (A) Schematic of the genome organization of RABV and the single-cycle RABV-ΔG-nanoLuc reporter virus generated for screening (N, nucleoprotein; P, phosphoprotein; M, matrix protein; G, glycoprotein; L, large [polymerase] protein). For screening, RABV-ΔG-nanoLuc virions were pseudotyped with G protein derived from the RABV SAD B-19 strain. (B) Validation of the RABV-ΔG-nanoLuc HTS protocol in 384-well plate format. Three source plates containing in alternating columns vehicle (V; DMSO) or the previously identified host-directed broad-spectrum inhibitor JMN3-003 (85) in low-dose (L_D; 20 nM final concentration) and high-dose (H_D; 10 μM) concentrations in distinct patterns were stamped against BEAS-2B cells in target plates, followed by infection with pseudotyped RABV-ΔG-nanoLuc as for panel A. Values from individual wells in a representative plate are plotted by column and then raw. Statistical analysis shows means ± SD of all plates. CV, coefficient of variation at L_D; % inhib, percent inhibition at L_D; plate S/B, plate signal-to-background ratio. Plate Z′ was calculated as [(AVR_V – 3×SD_V) − (AVR_HD + 3×SD_HD)]/SD_V. (C) Rocket plot of primary HTS results, ordered by robust Z-score and percent inhibition. Dotted lines denote hit candidate cutoffs (≥70% inhibition and ≥2.0 robust Z-score). The inset shows in silico, direct, and orthogonal counterscreening results for 380 primary hit candidates. Only GRP-60367 (red) passed all counterscreens. (D) Summary of counterscreen types and number of hit candidates eliminated by each filter. Host cell specificity was tested by comparing antiviral activity in BEAS-2B, Hep2, and 293T cell lines.

We next applied the assay screen to our in-house open discovery collections of nearly 150,000 compounds, each tested at a final concentration of 5 μM. For hit candidate identification, we subjected raw data to two-pronged control-dependent or -independent statistical evaluation that we developed previously (16, 18, 43), using >70% inhibition and robust Z-score of >2.0 as predefined thresholds. This strategy yielded 380 hit candidates (Fig. 1C). In silico cross-referencing with previous HTS campaigns that we have carried out with the same libraries against different biotargets (16, 18, 44) allowed elimination of known cytotoxic compounds, undevelopable promiscuous frequent hitters, known reporter-interfering compounds, and other undesirable compounds (in total, 337 compounds). The remaining 43 candidates were picked and tested in direct and orthogonal dose-response counterscreens. In these increasingly stringent filter assays, three hit candidates emerged as false positives, 36 were rejected due to a starting EC₅₀ of >5 μM, two did not pass a predefined SI (50% cytotoxic concentration [CC₅₀]/EC₅₀) threshold of >10, and one showed strong host cell line dependence of anti-RABV activity (Fig. 1C [inset] and Fig. 1D). Only hit candidate GRP-60367 met all predefined objectives.

Activity validation of GRP-60367.

In silico assessment of past performance of GRP-60367 in five antiviral campaigns with distinct targets revealed that this chemotype (Fig. 2A) resides in the dark matter of our collections, a chemical space that had been completely unresponsive in previous screens. Sourced GRP-60367 confirmed specific and potent anti-RABV activity of the scaffold, returning in dose-response assays EC₅₀s ranging from 2 to 52 nM on different host cell lines (Fig. 2B). We furthermore did not detect morphological signs of cytotoxicity or altered cell metabolic activity after prolonged 48-h exposure of cells to up to 300 μM GRP-60367, translating to initial SI values of >5,792 to >150,000. Confirming RABV specificity and low cytotoxicity, a vesicular stomatitis virus (VSV)-nanoLuc recombinant was not inhibited by the compound (Fig. 2B).

FIG 2.

Activity profiling of GRP-60367. (A) Chemical structure of GRP-60367. (B) GRP-60367 dose-response inhibition curves of RABV-ΔG-nanoLuc (Fig. 1A) in BEAS-2B, HEp2, and 293T host cells. RABV-ΔG-nanoLuc was pseudotyped with different strain origin RABV G proteins, as specified. A recombinant recVSV-nanoLuc reporter virus was tested in parallel on BEAS-2B cells only. Solid lines show 4-parameter variable slope regression curves where possible, and dashed lines show cytotoxicity for each cell line. Numbers are active concentrations (EC₅₀ and CC₅₀), with 95% confidence intervals in parentheses. Symbols represent means for at least three biological repeats; error bars indicate SD. (C) GRP-60367 dose-response inhibition curves of replication-competent RABV-SAD-B19, RABV-CVS-N2c, and RABV-Ef-T2 and of recombinant RABV-SAD-B19 expressing VSV G. Virus yields were determined as FFU by titration and are expressed normalized for vehicle-treated cells (left axis). Solid lines show 4-parameter variable slope regression models where possible; the dashed line shows cytotoxicity on N2a cells (right axis). Symbols represent means for three biological repeats ± SD, or two repeats ± range (rRABV-G_VSV only).

To explore the indication spectrum of GRP-60367 against different rhabdoviruses, we examined antiviral activity of GRP-60367 against RABV-SAD-B19, RABV-CVS-N2c, and a big brown bat isolate (RABV-Ef-T2) on disease-relevant Neuro-2a cells. Inhibitory activity against fully replication-competent RABV-SAD-B19 was consistently submicromolar (EC₅₀, 0.27 μM), while rRABV-CVS-N2c was approximately 10-fold less sensitive (EC₅₀, 2.63 μM) and the bat isolate was only marginally inhibited in the compound concentration range tested (10 μM, highest) (Fig. 2C). By comparison, no inhibitory activity was detected in dose-response assays against a recombinant RABV-SAD-B19 expressing VSV-derived G instead of homotypic RABV G. Cytotoxicity assessment demonstrated that compound concentrations of 300 μM were well tolerated by the neural cells also, without negative impact on cell metabolic activity (Fig. 2C). These data demonstrate specific, submicromolar activity of GRP-60367 against RABV-SAD-B19.

Mechanism of action of GRP-60367.

Capitalizing on insensitivity of VSV to GRP-60367 for target evaluation, we pseudotyped the single-cycle RABV-ΔG-nanoLuc with VSV G and generated a replication-competent recombinant VSV that expressed enhanced GFP (eGFP) from an additional transcription unit in the post-M position and encoded RABV G in place of homotypic VSV G (Fig. 3A). In dose-response assays, GRP-60367 was >6,000-fold less potent against the resulting RABV-ΔG-nanoLuc-G_VSV than RABV-ΔG-nanoLuc-G_RABV virions (Fig. 3B). Providing additional positive-confirmation of RABV G specificity of GRP-60367, the chimeric recVSV-ΔG-eGFP-G_RABV virus was also efficiently (EC₅₀, 0.005 μM) inhibited by GRP-60367 (Fig. 3B), returning activity concentrations similar to those observed before for RABV-SAD-B19. This G origin dependence of the inhibitory effect suggests that GRP-60367 functions as a viral entry inhibitor and argues against compound interference with a host pathway, since entry of RABV and VSV follows a common overall strategy (45).

FIG 3.

Mechanistic characterization of GRP-60367. (A) Genome schematic of recVSV-ΔG-eGFP reporter virus expressing RABV G. (B) Dose-response testing against single-cycle RABV-ΔG-nanoLuc pseudotyped with RABV G or VSV G and replication-competent recVSV-ΔG-eGFP-G_RABV from panel A or parental recVSV-ΔG-eGFP-G_VSV. Virus replication was determined based on relative luciferase units (RLU) (for single cycle RABV-ΔG-nanoLuc) or through TCID₅₀ titration (recVSV-ΔG-eGFP) and is expressed normalized for vehicle-treated samples. Regression modeling was done as for Fig. 2B; symbols represent means of three biological repeats ± SD. (C) Time-of-addition testing of GRP-60367 (final concentration of 10 μM) against RABV-ΔG-nanoLuc displaying RABV G or VSV G, as for panel A. Reporter activities were determined 24 h after infection, and values are normalized for signals obtained after infection in the presence of vehicle (DMSO). Symbols represent means of three biological repeats ± SD. Two-way ANOVA with Dunnett’s multiple comparisons post hoc test relative to vehicle-treated was done. ***, P = 0.0009; **, P = 0.0019; *, P = 0.0403; NS, not significant.

To verify the mechanism of action of GRP-60367 independently, we subjected the compound to a time-of-addition (TOA) variation study, using single-cycle RABV-ΔG-nanoLuc-G_RABV and RABV-ΔG-nanoLuc-G_VSV as viral targets (Fig. 3C). BEAS-2B cells were preinfected with either virus at a multiplicity of infection (MOI) of 0.1 particle/cell each, followed by addition of compound at the indicated time points (symbols in Fig. 3C) at a sterilizing 10 μM concentration. Relative bioluminescence activity was determined in all wells 24 h after infection. GRP-60367 was most effective when added within 60 min of infection, whereas no significant reduction of relative viral replication was observed when the compound was added later than 2 h after infection.

This TOA profile is consistent with that of previously described viral entry inhibitors (17, 18, 46, 47). Taken together, these results reveal that GRP-60367 specifically blocks RABV entry and identify the RABV G protein as the molecular target of the compound.

recVSV-eGFP-G_RABV adaptation to GRP-60367.

To better understand the molecular nature of GRP-60367 activity, we used the recVSV-eGFP-G_RABV chimera for resistance profiling through a dose escalation viral adaptation strategy. BEAS-2B cells were originally infected at an MOI of 0.1 particle/cell, followed by exposure to a starting concentration of GRP-60367 of 1.1 μM (1 EC₉₀) 30 min after infection (Fig. 4A). Eight independent recVSV-eGFP-G_RABV adaptation lineages were initiated in parallel, and the viruses were passaged approximately every 2 days based on microscopic assessment of green fluorescence. The inhibitor concentration was raised 3-fold at each passage, until a concentration of 110 μM (10 EC₉₀s) was tolerated. All eight lineages were subjected to a final 24-h infection cycle in the presence of fully sterilizing 1 mM GRP-60367, followed by microscopic verification of virus replication (Fig. 4B) and extraction of total RNA from all cultures. Sanger sequencing of the G open reading frames (ORFs) after reverse transcription-PCRs revealed in each lineage at least one coding point mutation in the bipartite fusion loops of the G protein, located at positions Y77/V78 and Y119, respectively (Fig. 4C), and thus in immediate proximity to the conserved hydrophobic residues identified in VSV G (48), which was postulated to mediate rhabdovirus G penetration into target membranes (Fig. 4D). Noteworthily, a Y119S substitution emerged in several lineages (L2, L3, L5, L6, and L7), while all other mutations occurred in only one lineage each. Only lineage 1 featured in addition to a Y119C substitution a premature stop codon at G position R466, resulting in truncation of the G cytoplasmic tail.

FIG 4.

Adaptation of recVSV-ΔG-eGFP-GRABV (Fig. 3C) to GRP-60367. (A) Escalating-dose adaptation strategy applied to eight distinct adaptation lineages. Starting at an MOI of 0.1 TCID₅₀/cell, initial compound application was 2 h after infection. Virus replication was monitored noninvasively by following the development of GFP fluorescence. (B) Fluorescence microphotographs of adaptation lineages from panel A after completion of the adaptation protocol. Cells were infected in the presence of 1 mM GRP-60367. “Vehicle” denotes cells infected with recVSV-ΔG-eGFP-GRABV passaged in the presence of vehicle (DMSO) equivalents instead of GRP-60367. (C) Schematic of the RABV G protein organization, color coded by functional domain. Candidate resistance mutations are highlighted. (D) Sequence alignment of the fusion loop sections of VSV (Indiana strain) and RABV (SAD strain) G proteins. Residues in red were demonstrated to be part of the VSV G membrane attack group, bold black denotes the predicted homologous residues in RABV G, and purple underlining specifies candidate resistance sites. Yellow lines specify β-sheets and helical regions in VSV G.

Effect of candidate resistance mutations on G protein fusion activity.

To probe for causality between individual G mutations and resistance to GRP-60367, we rebuilt each exchange individually in a RABV G expression plasmid. Assessment of intracellular and plasma membrane G protein steady-state levels after transient transfections of cells with the resulting mutant plasmids demonstrated that all RABV G variants carrying amino acid substitutions were expressed and intracellularly transported comparably to standard RABV G (Fig. 5A).

FIG 5.

Confirmation of resistance candidates. (A) Intracellular and plasma membrane steady-state levels of transiently expressed RABV G proteins. SDS-PAGE and immunoblotting were carried out after cell surface biotinylation and streptavidin affinity precipitation. Host cell GAPDH and TFR served as internal standards for quantitation. Blots of a representative biological repeat are shown. The graph shows densitometry results for four biological repeats. Columns show means and SD; symbols represent results for individual biological repeats. One-way ANOVA with Tukey’s multiple-comparisons post hoc test relative to standard RABV G (wt RABV G) was done. **, P = 0.0023; NS, not significant; WCL, whole-cell lysates; SF biotinylation, cell surface protein fraction. (B) Schematic of the kinetic cell-to-cell fusion assay used to determine bioactivity of transiently expressed RABV G variants. DSP, dual-split protein. (C) Kinetic fusion assay applied to transiently expressed standard RABV G in the absence (vehicle) or presence of GRP-60367. Values were normalized for the highest luciferase signal obtained in the absence of GRP-60367 (wtRABV G max); symbols represent means for at least five biological repeats ± SD. Two-way ANOVA with Geisser-Greenhouse correction and Sidak’s multiple-comparisons post hoc test between treated and untreated samples was done for each mutant. (D and E) Effect of candidate resistance mutations on cell-to-cell fusion kinetics, assessed as for panels B and C. Values were normalized for wtRABV G max (gray lines); symbols represent at least four biological repeats ± SD. Two-way ANOVA was done as for panel C. NS, not significant.

By adapting to RABV G a dual-split eGFP-Renilla luciferase protein (DSP)-based transient cell-to-cell fusion reporter assay that we developed previously to monitor paramyxovirus glycoprotein-mediated cell-to-cell fusion kinetics (17, 49–52), we investigated the impact of the individual candidate resistance mutations on G-mediated fusion kinetics in the presence and absence of 20 μM GRP-60367 (Fig. 5B). We first established the assay for standard RABV G, quantifying content mixing of BEAS-2B effector and target cell population that were transiently cotransfected with RABV G and DSP_1-7 expression plasmids or transfected with the DSP_8-11 plasmid, respectively. The pH of the culture media was lowered to 5.8 after preloading of cells with EnduRen luciferase substrate and mixing of the two cell populations, and relative bioluminescence was monitored at 15-min intervals over a 4-h period. In the absence of GRP-60367, signal intensity increased gradually until a plateau was reached approximately 200 min after pH drop (Fig. 5C). In contrast, no significant increase in bioluminescence was detected when the inhibitor was present, directly demonstrating inhibition of RABV G fusion activity by GRP-60367.

When applied to the panel of RABV G resistance mutant candidates, the fusion assay revealed that escape from GRP-60367 is associated with two distinct fusion phenotypes. A subset of three candidate mutations (V78I, Y119N, and Y119S) caused G hyperfusogenicity, resulting in signal plateaus 200% to 300% or higher that of standard G (Fig. 5D). No significant differences in fusion activities of these G protein mutants were detected in the presence or absence of GRP-60367, confirming that these substitutions mediate resistance. A second subset of three other candidate mutants (Y77S, Y119C, and R466*) showed fusion activities equivalent to that of standard RABV G in the absence of GRP-60367 (Fig. 5E). Fusion activities of each of these three G mutants was furthermore unaffected by the presence and absence of GRP-60367, demonstrating that also these mutations caused resistance to the compound. These data confirm that each of the adaptation lineages acquired at least one mutation in RABV G that mediated escape from inhibition by GRP-60367 in the transient cell-to-cell fusion assay. Whereas a subset of mutants showed enhanced fusion activity, hyperfusogenicity was not a requirement for escape from GRP-60367.

Confirmed resistance hot spots cluster in the RABV G fusion loop.

To validate escape from inhibition by GRP-60367 in the context of viral infection, we rebuilt a subset of substitutions affecting G residues V78 and Y119 in the context of recVSV-eGFP-G_RABV (Fig. 6A). These recombinants were recovered successfully and subjected to dose-response inhibition tests after amplification. Each of the substitutions tested resulted in a >2,000-fold increase in EC₅₀, confirming causality for the resistance phenotype in replication-competent viruses.

FIG 6.

Resistance confirmation in recVSV-ΔG-eGFP-G_RABV. (A) Dose-response testing of recVSV-ΔG-eGFP-G_RABV expressing standard RABV G or RABV G variants with individual resistance mutations. Regression modeling was done as for Fig. 2B where possible. Virus titers were normalized for those obtained in the presence of vehicle (DMSO) for each recombinant. Symbols represent means for three biological repeats ± SD. (B) Homology models of pre- and postfusion RABV G based on the coordinates reported for prefusion (PDB 5I2S [86]) and postfusion (PDB 5I2M [48]) VSV G. Ribbon models of the G trimers are colored by monomer. Resistance sites in the G fusion loops are shown as red spheres. (C) Homology model of a prefusion RABV G monomer. The inset shows the position of confirmed resistance sites (red) and key residues of the bipartite fusion patch (gray). (D) Side views of surface models of the fusion loops from panel C, color coded by degree of residue conservation across the genus Lyssavirus from most (gray) to least (red) conserved. Confirmed resistance sites are labeled. (E) Sequence alignment of selected members of the genus Lyssavirus, color coded by degree of conservation. Sites of resistance to GRP-60367 are indicated (arrowheads).

For a structural appreciation of confirmed hot spots of escape from GRP-60367, we generated homology models of RABV G in pre- and postfusion conformations based on the coordinates reported for VSV G. The resistance sites, Y77/V78 and Y119, are predicted to be located in close proximity to each other near the tips of the rhabdovirus G protein loops cd and Pe, respectively, which together form the bipartite membrane attack group (Fig. 6B and C). Resistance sites Y77 and V78 are fully conserved across pathogens in the genus Lyssavirus, while residue Y119 is located in a more variable patch (Fig. 6D and E). These observations are consistent with direct physical targeting of RABV G by GRP-60367, which in turn interferes with the ability of G to mediate membrane merger for viral entry.

DISCUSSION

hRIG is required for life-saving passive immunization after exposure to RABV. However, limited supply and high costs of the IgG component restrict distribution in low-income countries where the clinical need is greatest, making the development of a cost-effective substitute a priority. Although RABV does not require high biocontainment conditions, all laboratory personnel working with live virus must have confirmed rabies vaccine protection, complicating automated drug discovery. Minimal cytopathicity associated with RABV infection in cell culture is furthermore incompatible with screening protocols that depend on host cell morphology changes as the readout. We have addressed these obstacles with the development of the single-cycle RABV-ΔG-nanoLuc reporter virus, which supports monitoring of RABV replication under standard biosafety level 2 (BSL2) conditions when pseudotyped with RABV G expressed in trans. Our assessment of HTS assay robustness has revealed consistent performance parameters that meet accepted quantitative targets for automated drug discovery (53, 54). By design, the screening protocol favors the identification of entry and polymerase inhibitors, whereas late-acting compounds interfering with particle assembly and egress are less likely to be discovered.

Application of this assay to a collection of nearly 150,000 small-molecule compounds returned a primary hit discovery rate of approximately 0.26%, which is on par with other antiviral screens that were carried out with well-curated libraries and at low screening concentrations similar to those in our campaign. In subsequent in silico and physical counterscreens, however, we implemented an aggressive elimination strategy that rejected all known undesirables in our libraries (i.e., promiscuous frequent hitters, known cytotoxic compounds, and covalently reactive compounds) based on the results of our previous drug screens with this collection and demanded a viable starting SI. Ultimately, this strategy reduced the candidate pool to a single compound, GRP-60367, that passed all counterscreening filters. Higher primary HTS hit rates are often reported, but in our opinion, these rates typically reflect a high representation of promiscuous—and ultimately undevelopable—chemotypes in the libraries examined (55). The best approach to hit candidate identification is certainly a subject of debate in HTS-based drug discovery, and more relaxed initial exclusion filters may be suitable for some indications. However, the intended clinical use of a successful anti-RABV candidate for rabies PEP in adults and pediatric patients after a suspected exposure calls for an extremely stringent safety profile, which is incompatible with all broad-spectrum and/or host-directed candidates that we have examined thus far.

GRP-60367 was highly potent against the RABV screening virus, returning low-nanomolar EC₅₀s. This compound had not shown activity in any of our previous screens against unrelated biotargets, placing it into the dark matter chemical space of the libraries. In counterscreens, GRP-60367 showed no cytotoxicity or other morphologically appreciable off-target activities after prolonged exposure of cells to concentrations of up to 300 μM, underscoring the experience that a hit in the dark matter, when it occurs, is typically chemically well behaved and highly target specific (56). Based on TOA studies, desensitizing RABV particles to the inhibitor by pseudotyping with VSV G, and inducing susceptibility of a recombinant VSV to the compound by replacing VSV G with RABV G, we conclude that GRP-60367 specifically targets the RABV G protein. Suppression of RABV G-mediated cell-to-cell fusion in the transient DSP-based fusion assay confirmed entry inhibition as the underlying mechanism of action.

Promiscuous receptor usage makes entry inhibition a challenging target for antivirals. GRP-60367 was indeed approximately 10-fold less potent against a mouse-adapted RABV strain and largely inactive against a bat-derived RABV isolate, underscoring our concerns that entry inhibition may not be the most fruitful approach to pharmacologically controlling mononegavirus infection (57, 58). However, drug discovery efforts directed against other pathogens of the order Mononegavirales, namely, respiratory syncytial virus (RSV) and members of the family Paramyxoviridae (17, 59–63), have yielded entry inhibitors with fundamental features reminiscent of GRP-60367 that in some cases have advanced to clinical development (64, 65). Several of these entry inhibitors were exceptionally potent, often with nanomolar or even subnanomolar active concentrations, and very well tolerated (49, 66, 67). However, all suffer from a low genetic barrier to viral resistance, which emerged as a liability both in animal models and in early clinical trials (17, 64, 65). Therefore, not unexpectedly, adaptation of a sensitized recVSV-eGFP-G_RABV virus to GRP-60367 consistently resulted in viral escape after seven or eight passages in cell culture. Two resistance hot spots in G emerged, at residues Y77/V78 and Y119, which are predicted to be exposed on the G protein loops that together from the bipartite rhabdovirus fusion patch (48). In particular, tyrosine side chains frequently line the interface between the carbohydrate chains and head groups of lipid bilayers (68). Although the exact motifs are not fully conserved between VSV G and RABV G, these aromatic residues were proposed to destabilize the organization of target membranes during G protein-mediated fusion pore formation (48).

Substitution of fusion loop Y77 and V78 was unexpected, since alignment of G protein sequences derived from diverse lyssaviruses revealed that the cd loop is fully conserved in all members of the genus from position 70 to 82 (Fig. 6E). Furthermore, the tyrosine at position 73 in VSV G could only be changed to another aromatic residue without abolishing G fusion activity (69). Although the Pe loop is also overall highly conserved in the genus Lyssavirus, residue 119 is part of a 2-residue variable patch (Fig. 6E), providing a ready explanation as to why different substitutions at this position were well tolerated in RABV G. Engaging the membrane attack groups of viral glycoproteins is biochemically challenging due to structural flexibility and the high content of hydrophobic side chains. However, several RSV entry inhibitors stabilize a prefusion conformation of the RSV fusion protein through tethering of the fusion peptide to the prefusion structure (70), providing a precedent for productive direct engagement of viral membrane attack groups by small molecule inhibitors. Considering the overall very high degree of lyssavirus fusion loop conservation, this domain emerges as an attractive target for G neutralization. It is out of reach, however, of the host humoral response, since the dense G protein shield displayed on the virion surface prevents antibody access. Accordingly, known neutralizing antibody epitopes are located distal to the viral envelope on the prefusion G protein (Fig. 7A to C). It will be an important synthetic objective to explore whether the GRP-60367 scaffold can be further directed toward the cd loop or conserved residues in the Pe loop, provided that the compound indeed interacts with the fusion loop domain.

FIG 7.

Predicted location of confirmed resistance sites on the viral surface. (A) Localization of confirmed resistance sites (red) and epitopes (blue) of known neutralizing antibodies (87–90) on the homology model of prefusion RABV G. (B and C) Positioning of resistance sites and neutralizing-antibody epitopes relative to the viral envelope (yellow plane) in side (B) and top (C) views.

At present, though, we have no final confirmation that the resistance spots in the fusion domain are part of the primary binding site of the compound. Alternatively, the substitutions identified in our study could mediate viral escape through long-range structural effects or altered fusion kinetics. In fact, narrowing the window of opportunity for pharmacological interference with viral entry through accelerating glycoprotein refolding has been identified as a viable escape mechanism employed by diverse viral targets (17, 71–73). Fusion profiling in our study revealed that a subset of escape mutants indeed caused G hyperfusogenicity, which is considered to indicate altered refolding kinetics. Fusion activity of some of these mutant G proteins (i.e., G-V78I and G-Y119N) was reduced back to approximately standard RABV G-like levels by GRP-60367, possibly suggesting a kinetic resistance effect. However, hyperfusogenic G-Y119S retained enhanced fusion activity in the presence of the compound, and three other substitutions causing resistance to GRP-60367 showed no difference in cell-to-cell fusion rate compared to that of standard RABV G. Since one of these changes (Y119C) affected the same residue as two (Y119S and Y119N) associated with hyperfusogenicity and the other (Y77S) was a direct neighbor of a substitution (V78I) causing G hyperfusion, we favor the view that enhanced G fusion activity may aid escape but does not represent the root cause of resistance. In support of this interpretation, previous work has established that RABV G residues 392 to 396 function as important regulators of G protein fusion activity (27–29, 74–76), but none of these sites emerged in our adaptation study and, inversely, the residues identified in our work have not been previously implicated in a fusion regulatory capacity. Furthermore, resistance through premature truncation of RABV G at residue R466 did not coincide with an enhanced rate of cell-to-cell fusion. However, long-range conformational effects arising through inside-out signaling from altered viral glycoprotein cytosolic tails have been well documented (77) and have been shown, for instance, to cause escape from neutralizing antibodies (78). Future studies aimed at extracting the physical docking pose of GRP-60367 will be required to address whether this novel entry inhibitor class specifically interacts with the RABV G fusion loops.

In summary, we have established a tangible framework for rigorous anti-RABV drug screens that are capable of delivering hit candidates meeting the safety requirements of an intended use for rabies PEP. The first-in-class confirmed hit GRP-60367 specifically targets the RABV G protein, blocking entry of some RABV strains with nanomolar potency. Resistance hot spots map predominantly to the G protein fusion loop and have not been previously characterized through mapping of escape from neutralizing antibodies, indicating that the compound interacts with a novel druggable site on the G protein. Based on target specificity, potency, and resistance profile, GRP-60367 may represent an exciting new molecular probe that will advance further mechanistic and structural characterization of RABV G, aiding in next-generation vaccine and antiviral-drug design.

MATERIALS AND METHODS

Cells.

Human bronchial epithelial cells (BEAS-2B; ATCC CRL-9609), human embryonic kidney cells (HEK-293T; ATCC CRL-3216), baby hamster kidney cells stably expressing T7 polymerase (BSR-T7/5), BSR-T7/5 cells expressing the G protein of SAD-B19 (42), and African green monkey kidney epithelial cells (Vero; ATCC CCL1-81) were maintained at 37°C in 5% CO₂ in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 7.5% fetal bovine serum. The BSR-T7/5 cell line was supplemented with G-148 at every fifth passage, and BSR-T7/5 cells stably expressing SAD-B19 G were grown in the presence of 2 μM doxycycline.

Molecular biology.

The RABV-ΔG-nanoLuc was generated by introduction of a nanoluciferase-encoding fragment between the SmaI and NheI sites of the SPBN vector (42). RABV-ΔG-G_VSV was generated in the same backbone by grafting the VSV-G (Indiana strain) ectodomain and transmembrane domain onto the cytoplasmic tail of RABV G (SAD-B19 strain) and inserting the resulting chimeric gene into a cDNA copy of the RABV genome between SmaI and NheI sites (79). Point mutations emerging from viral adaptation were rebuilt in a RABV G (SAD-B19 strain) plasmid, followed by transfer of the G open reading frame into recVSV-ΔG-eGFP-G_RABV, using MluI and NheI restriction sites.

Compounds.

Screening libraries consisted of a compilation of commercially available subsets (sourced from ChemDiv, ChemBridge, Prestwick, and Asinex) and proprietary collections originating from previous drug development campaigns. All chemical sets were curated against compounds with pan-assay-interfering or other known undesirable chemical structures. Compounds were dissolved in dimethyl sulfoxide (DMSO) to a concentration of 10 μM, reformatted into 384-well plate daughter libraries, and stored at −80°C. Each screening source plate contained 64 positive (cycloheximide) and negative (DMSO) control wells, arranged in a checkerboard pattern in the two lateral columns on each side of the plate. Chemical inventory management, screen execution, and data analysis was carried out using the MScreen software package.

HTS protocol.

BEAS-2B cells (3.5 × 10³/well) were seeded in barcoded 384-well, white-wall, clear-bottom plates using a MultiFlo automated dispenser (BioTek) equipped with dual 10-μl peristaltic pump manifolds. After brief sedimentation (150 × g for 90 s at 25°C), plates were incubated for 14 h at 37°C in a 5% CO₂ environment. Test compounds were added to a final concentration of 5 μM (20 nl/well) with a Nimbus96 liquid handler outfitted with a high-density pin tool (V&P Scientific) attached to the pipetting head. Both source and assay plates were barcode-read by the Nimbus96 unit at the time of compound stamping. Cells were infected with recRABVΔG-nanoLuc pseudotyped with RABV G (MOI = 0.01 50% tissue culture infective dose [TCID₅₀]/cell in 10 μl/well) using the MultiFlo dispenser followed by spin collection and incubation under standard conditions for 48 h. Plates were loaded into an H1 synergy plate reader (BioTek) equipped with a stacker, integrated barcode reader, and substrate injectors delivering nano-Glo substrate (10 μl/well, delivered at 225 μl/s). For each well, bioluminescence was recorded after a 3-min signal stabilization period (endpoint reading; instrument gain, 130; 400-ms integration time).

HTS data normalization and analysis.

For automated data analysis, bioluminescence raw data were imported automatically from the plate reader into the MScreen software package, and normalized relative inhibition values were calculated for each compound by subtracting the measured value for each well from the mean of the plate vehicle controls and then dividing the results by the difference between vehicle and positive-control means. Plate control-independent robust Z-scores were calculated as [S_i − median(S_all)]/MAD(S_all), and MAD(S_all) was calculated as 1.4826 × median[|S_i − median(S_all)|], where S_i is an individual compound value and S_all are values for all compounds on the plate analyzed. Hit candidates were defined as compounds with ≥70% inhibition of normalized signal intensity against RABV-ΔG-nanoLuc, a robust Z-score of ≥2.0, and <35% inhibition against recRSV-fireSMASh (80), influenza A virus WSN-nanoLuc (18), and a human parainfluenzavirus 3 expressing nanoLuc. Emerging hit candidates were queried using the SwissADME (81) and Badapple (82) first- and second-generation filter algorithms to identify undesirable substructures and the SciFinder database package to determine known bioactivities.

Automated dose-response counterscreening.

BEAS-2B cells (3.5 × 10³ cells/well) were seeded in barcoded white-wall, clear-bottom, 384-well plates using the automated dispenser as described above. The Nimbus96 unit was used to pick hit candidates from the stock library and prepare counterscreening stock plates in 384-well format containing 3-fold serial dilutions of hit compounds (0.078 to 10 μM range) for subsequent dose-response and cytotoxicity counterscreening. The stock plates were stamped against the cell-containing target plates using the high-density pin tool, followed by infection with reporter viruses and readout of reporter activity as described above. To determine cytotoxicity, cell viability was measured after 48 h incubation in the presence of compound using PrestoBlue substrate (5 μl/well) (Life Technologies). After a 90-min incubation period in the presence of substrate, top-read fluorescence (excitation at 560 nm, emission at 590 nm; instrument gain, 85) was determined in the H1 synergy multimode plate reader.

Dose-response reporter assays.

The Nimbus96 unit was used to prepare and transfer 3-fold or 5-fold serial dilutions (as indicated in the figure legends) of hit candidate compounds to 96-well plates seeded with BEAS-2B, 293T, or Hep-2 cell lines (1.5 × 10⁴ cells/well). Controls included four positive (1 mg/ml cycloheximide) and four negative (0.05% vehicle [DMSO]) wells on each plate, arranged in an alternating pattern. Cells were infected with the different target viruses at specified in the figure legends (MOI = 0.1 TCID₅₀/cell), and bioluminescence reporter activity was read after a 24-h (recVSV-nanoLuc) or 48-h (all other target viruses) incubation period. Dose-response inhibition curves were generated according to the following formula: % inhibition = [(X_sample − X_min)/(X_max − X_min)] × 100, where X_min represents the mean of the positive-control and X_max the mean of the negative-control wells. Cell viability was determined using PrestoBlue substrate after a 48-h exposure of uninfected cells as described above. Active concentrations (EC₅₀ and CC₅₀) were determined through four-parameter variable slope regression modeling when possible. All dose-response activity assays were performed in at least three biological repeats.

Virus yield reduction assays.

Mouse neuronal N2a cells (1.5 × 10⁴ cells/well) were seeded in a 96-well plate format and incubated for 24 h. GRP-60367 was added in 3-fold serial dilutions, followed by infection of 6,000 focus-forming units (FFU) (MOI, 0.2 FFU/cell) of virus. Control wells without compound were infected in parallel with the same amount of inoculum. After 60 min, the inoculum was removed and replaced with 100 μl fresh DMEM with (treated) or without (controls) compound. Culture supernatants were collected after 48 h incubation at 34°C, and progeny virus titers were determined on N2a cells.

Virus adaptation.

BEAS-2B cells were infected with recVSV-ΔG-eGFP-G_RABV at an MOI of 0.1 TCID₅₀/cell and incubated for 30 min at 37°C and 5% CO₂, followed by addition of GRP-60367 to a final concentration of 1.1 μM. Fresh BEAS-2B cells were infected with a 10-fold dilution of released virions in cleared culture supernatants every 2 days, and compound concentrations were gradually increased until 110 μM was tolerated without collapse of the virus population. Total RNA was extracted from eight individual adaptation lineages using the RNeasy minikit (Qiagen) and cDNAs generated with Superscript III reverse transcriptase and random hexamer primers of first-strand synthesis. The G protein-encoding ORF was PCR amplified, and purified amplicons were subjected to Sanger sequencing. Candidate resistance mutations were introduced into G protein expression plasmids using QuikChange site-directed mutagenesis. Selected mutations were rebuilt in the full-length cDNA copy of the recVSV-ΔG-eGFP-G_RABV genome for virus recovery.

Kinetic cell-cell fusion assay.

BEAS-2B cells (1.5 × 10⁵ cells/well) were cotransfected with 0.5 μg of DSP_1-7 plasmid DNA (83) and 1.0 μg of plasmid DNA/well encoding RABV G or RABV G harboring individual candidate resistance mutants. A separate cell population was independently transfected with 0.5 μg DSP_8-11 plasmid DNA/well. Cells were harvested 24 h after transfection, mixed at equal ratios, and reseeded in 96-well plates at a total density of 4 × 10³ cells/well. Reseeded cells were pretreated with EnduRen substrate at 4°C for 1.5 h according to the manufacturer’s instructions. Fusion was triggered by replacing the media with 100 μl of DMEM pH 5.8, and EnduRen bioluminescence was measured every 15 min for 4 h in a Cytation5 automated high-content imager with phase-contrast microscopy capacity (BioTek), set to 37°C and 5% CO₂. Values were normalized for maximal bioluminescence signal intensity observed in the presence of standard RABV G (wtRABV G_max).

Recombinant virus recovery.

recVSV-ΔG-eGFP-G_RABV variants were recovered following a previously described protocol (84). Briefly, expression plasmids encoding VSV-N (0.5 μg plasmid DNA), VSV-P (0.4 μg plasmid DNA), and VSV-L (0.2 μg plasmid DNA) were cotransfected with plasmids containing the cDNA copies of parental recVSV-ΔG-eGFP-G_RABV or recVSV-ΔG-eGFP-G_RABV with mutant RABV G (1.0 μg plasmid DNA) into BSR-T7/5 cells (2.5 × 10⁵ cells/well) stably expressing T7 polymerase. Transfected cells were incubated at 37°C and monitored microscopically for the appearance of green fluorescence. Culture supernatants containing recovered progeny virions were transferred to Vero-E6 cells for generation of a P₀ stock. After amplification, cleared supernatants containing stocks of recovered viruses were stored in aliquots at −80°C, titers were determined by TCID₅₀ titration using eGFP fluorescence as the readout, and the integrity of the G protein-encoding open reading frame was verified through reverse transcription-PCR (RT-PCR) and Sanger sequencing.

Surface biotinylation, SDS-PAGE, and immunoblotting.

BSR-T7/5 cells were seeded in 6-well plates (4 × 10⁵ cells/well) and transfected with 1 μg plasmid DNA encoding standard RABV G or RABV G with a confirmed resistance mutation. Twenty-four hours after transfection, cells were washed twice with cold phosphate-buffered saline (PBS), followed by biotinylation of surface proteins with 0.5 mg/ml sulfosuccinimidyl-2-(biotinamido)ethyl-1,3-dithiopropionate (Pierce) for 30 min on ice, quenching with 1 M Tris (pH 7.5), three washes with cold PBS, and lysis in radioimmunoprecipitation assay (RIPA) buffer (1% sodium deoxycholate, 1% NP-40, 150 mM NaCl, 50 mM Tris-Cl [pH 7.2], 10 mM EDTA, 50 mM NaF, and protease inhibitors [Pierce]). Streptavidin bead slurry was added to cleared lysate (20,000 × g, 30 min, 4°C), and samples were incubated on a rotator for 2 h at 4°C. Washed precipitates (3 times in RIPA buffer at 2,000 rpm for 2 min, followed by 3 times in PBS at 2,000 rpm for 2 min) and total lysates prepared in parallel were fractioned through sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), followed by semidry transfer to polyvinylidene difluoride (PVDF) membranes and immunodetection of G protein antigenic material using anti-RABV G protein rabbit antiserum raised against purified SAD-B19 G (Pacific Immunology) or detection of host cell glyceraldehyde 3-phosphate dehydrogenase (GAPDH; 6C; Ambion) and transferrin receptor (TFR; Zymed) as internal standards for intracellular and plasma membrane proteins, respectively. Immunoblots were developed using appropriate anti-species IgG horseradish peroxidase (HRP)-conjugated secondary antibodies and a ChemiDoc digital imaging system (Bio-Rad). Quantitations were performed on four biological repeats, analyzing nonsaturated images with the Image Lab software package (Bio-Rad) and applying global background correction. Positive (standard RABV G) and negative (equivalent amounts of vector DNA replacing the RABV G expression plasmid) controls were present on each immunoblot, and no across-blot normalizations were performed.

Statistical analyses.

The MScreen, Prism 8 for Mac (GraphPad), and Excel (Microsoft) software packages were used for data analysis. The statistical significance of differences between groups was determined through one-way or two-way analysis of variance (ANOVA) in combination with post hoc multiple-comparison tests as specified in the figure legends. Experimental variation is identified in the figures by error bars, representing standard deviation (SD), as specified in the figure legends.