A cocrystal structure of dengue capsid protein in complex of inhibitor

Significance

DENV represents the most important mosquito-borne viral pathogen in humans. Although one dengue vaccine has been approved for clinical use, its efficacies against serotypes 1 and 2 remain to be improved. No effective anti-DENV therapy is available for patient treatment. In this study, we solved the cocrystal structure of the DENV-2 capsid protein in complex with an inhibitor that preferentially inhibits serotype 2 (not the other three serotypes) in vitro and in vivo. Mechanistically, the compound is incorporated into virions by inducing capsid tetramerization. Such inhibitor-containing virions are defective in nucleocapsid uncoating when infecting new cells. Our results have also uncovered the resistance mechanism and provided a high-resolution structure for rational design of capsid inhibitors with pan-serotype DENV activity.

Abstract

Dengue virus (DENV) was designated as a top 10 public health threat by the World Health Organization in 2019. No clinically approved anti-DENV drug is currently available. Here we report the high-resolution cocrystal structure (1.5 Å) of the DENV-2 capsid protein in complex with an inhibitor that potently suppresses DENV-2 but not other DENV serotypes. The inhibitor induces a “kissing” interaction between two capsid dimers. The inhibitor-bound capsid tetramers are assembled inside virions, resulting in defective uncoating of nucleocapsid when infecting new cells. Resistant DENV-2 emerges through one mutation that abolishes hydrogen bonds in the capsid structure, leading to a loss of compound binding. Structure-based analysis has defined the amino acids responsible for the inhibitor’s inefficacy against other DENV serotypes. The results have uncovered an antiviral mechanism through inhibitor-induced tetramerization of the viral capsid and provided essential structural and functional knowledge for rational design of panserotype DENV capsid inhibitors.

The mosquito-borne dengue virus (DENV) affects over 3 billion people living in the tropics and subtropics, causing 390 million human infections annually (1). DENV belongs to the Flavivirus genus from the Flaviviridae family. Besides DENV, many other flaviviruses are significant human pathogens, including yellow fever, Zika (ZIKV), West Nile (WNV), Japanese encephalitis, and tick-borne encephalitis viruses. Flaviviruses consist of an envelope outer layer, formed by envelope (E) and membrane (M) proteins on a lipid bilayer, and an internal nucleocapsid, formed by the capsid (C) protein and genomic RNA. The viruses infect cells through receptor-mediated endocytosis. The acidic environment in the late endosome induces membrane fusion between the virus and endosome, releasing the nucleocapsid. After uncoating the C protein from the nucleocapsid, the genomic RNA translates three structural proteins (C, prM/M, and E) and seven nonstructural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5). The nonstructural proteins, together with host proteins and lipids, form replication complexes in the endoplasmic reticulum (ER). Virion assembly requires the C protein binding to progeny viral RNA to form the nucleocapsid, which buds into the ER lumen as an immature virion consisting of icosahedral prM/E heterotrimers. During virion exit, the acidic pH of the Golgi network triggers a conformational change in prM to allow host furin protease cleavage, resulting in infectious mature virions (2).

Vaccine development for DENV has been challenging because of the potential risk of antibody-dependent enhancement among the four serotypes of DENV (3–5). This has been illustrated by the increased hospitalization in children who were seronegative when vaccinated with Dengvaxia (6–8). The lack of an effective DENV vaccine underscores the importance of antiviral development. Although several repurposed compounds have been tested in dengue clinical trials, none of them has shown clinical benefits in patients (9–13). No bona fide inhibitor designed for DENV has been advanced to clinical trials. Despite direct antiviral agents (DAAs) successfully developed for HIV, hepatitis C virus (HCV), and other viruses, development of DENV DAAs has proven challenging. After >15 y of effort from academia and industry, only a few DENV DAAs have shown efficacy in mice (14). These DAAs include nucleoside analogs (15–17), NS4B inhibitors (18), and capsid inhibitors (19). However, these compounds were confronted with in vivo toxicity (for nucleoside analogs) and a narrow antiviral spectrum (for NS4B and capsid inhibitors). New knowledge is urgently needed to overcome these limitations for further development of those inhibitors.

ST148 (SI Appendix, Fig. S1A) was identified through a cell-based DENV-2 infection screen (19). It potently inhibits DENV-2 but not DENV-1, -3, or -4. Resistance to ST148 was mapped to the DENV-2 C gene. Detailed antiviral and resistance mechanisms of ST148 remain largely unknown (20, 21). Here we took structural, biochemical, and virologic approaches to demonstrate that ST148 induces a “kissing” interaction between two C dimers. Such a ST148-induced C tetramer is assembled inside virions. When infecting new cells, the compound-containing viruses are defective in nucleocapsid uncoating. Structure-guided analysis has defined the determinants of drug sensitivity among the four serotypes of DENV C proteins. Our results have uncovered an antiviral mechanism and knowledge for rational design of DENV capsid inhibitors.

Results

DENV-2 C Protein Structure.

The C protein, representing amino acids 21 to 100 of the DENV-2 NGC strain, was expressed and purified (SI Appendix, Fig. S1 B–D). The DENV-2 apo-C protein crystals formed in space group P3₁21 with six molecules in the asymmetric unit (au) and diffracted to 2.6 Å. Attempts to solve the structure using the NMR structure of the DENV-2 C protein (22) as a search model failed. Instead, the structure was solved using a high-resolution cocrystal structure (1.5 Å) of the DENV-2 C protein in complex with inhibitor ST148. The cocrystal structure was determined using the WNV C crystal structure as a search model (23). Our DENV-2 apo-C crystal structure contains four α-helices, with the six N-terminal amino acids disordered (SI Appendix, Fig. S2A). Comparison of the DENV-2 apo-C crystal structure with the NMR structure (Protein Data Bank [PDB] ID 1R6R) showed an ∼130° rotation of N-terminal helix 1 (rmsd = 9.38 Å [7.52 and 7.97 Å for chain A and chain B, respectively] for all Cα atoms; SI Appendix, Fig. S2B). The DENV-2 apo-C crystal structure is similar to the WNV C crystal structure (rmsd = 2.04 Å [1.73 and 1.86 Å for chain A and chain B, respectively] for all Cα atoms; SI Appendix, Fig. S3A). In the WNV structure, helix 1 from both dimeric molecules forms a single long helix (SI Appendix, Fig. S3 A and B). In the DENV-2 apo-C crystal structure, helix 1 from the dimeric molecules exhibits an asymmetric conformation (SI Appendix, Fig. S3C): Helix 1 from one molecule is long and well ordered, whereas helix 1 from another molecule is split into helix 1A and helix 1B at Phe33 and Ser34, causing helix 1A to rotate relative to the full helix. In the DENV-2 apo-C crystal structure, two dimers form a kissing interface through the helix 1 hydrophobic surface (SI Appendix, Fig. S3D). A similar crystal contact was observed in the WNV C crystal structure, but with a different angle between the dimers’ twofold axis (23). The surface of the DENV-2 apo-C dimer is highly positively charged, except for the helix 1 surface (SI Appendix, Fig. S4A). This positively charged patch may interact with viral RNA, possibly through bending the viral RNA around the helix 4 surface as previously shown with bacterial HU histone-like proteins (24).

Cocrystal Structure of the DENV-2 C Protein in Complex with an Inhibitor.

The addition of ST148 to the DENV-2 C protein formed cocrystals that diffracted to 1.5 Å. The cocrystals were in space group P2 and contained a symmetric C dimer in the au (Table 1), which forms a tetramer via the P2 crystallographic symmetry (Fig. 1A). Two inhibitors bind to a pocket formed by the C tetramer across the kissing interface between dimers. The pocket has P22 symmetry due to the dimer’s twofold noncrystallographic symmetry and the tetramer’s crystallographic twofold symmetry. Thus, the inhibitor dimer could bind the pocket in either of the two orientations (Fig. 1B). Seven residues (Thr30, Phe33, Leu35, Met37, Leu38, Leu50, and Phe53) from each tetrameric protein collectively constitute the inhibitor pocket (Fig. 1C). The structure shows the following inhibitor–protein interactions: 1) the two amine groups of the ST148 dimer form H bonds with one pair of Phe33’s carboxylates, leaving another pair of Phe33 carboxylates unbound in the C tetramer; 2) one pair of Leu38’s side chain stacks onto the triple ring of ST148; and 3) the remainder of ST148 is bound through hydrophobic interaction (Fig. 1C).

Table 1.

X-ray data collection, processing, and refinement statistics

	DENV-2 apo-capsid	DENV-2 capsid + ST148
PDB ID	6VSO	6VG5
Space group	P3121	P2
Cell, Å; °	142.0, 67.0	40.4, 25.5, 68.5; 92.55
Resolution, Å	2.6 (2.64–2.60)	1.5 (1.53–1.5)
Rmeas†	0.205 (4.1)	0.103 (0.91)
Rpim†	0.062 (1.3)	0.042 (0.37)
Redundancy	11.1 (10.3)	6.1 (5.8)
Completeness, %	100 (99.8)	93.4 (76)
CC1/2†	1.0 (0.450)	1.0 (0.695)
Wavelength, Å	0.97857	1.07812
Beamline	APS LS-CAT 21-ID-G	APS LS-CAT 21-ID-D
Rwork†	20.8	14.5
Rfree†	24.4	19.0
rmsd bonds, Å	0.009	0.009
Mean B factor, Å2	55	18.0
Protein chains	54, 54, 68, 68, 42, 43	18.7, 17.4
Solvent	69	30.8
Ligands	NA†	9.6, 11.7
Number of chains	6	2
Protein atoms	3,651	1,294
Solvent atoms	76	122
Ligand atoms	NA†	58
Ramachandran
Favored, %	99.78	98.10
Allowed, %	0.22	1.90
Outliers, %	0.0	0.0

^†Rmeas, residual factor for individual data; Rpim, residual factor for merged data, normalized to account for measurement multiplicity; CC1/2, half correlation coefficient; Rwork, residual factor for diffraction data used in crystallographic refinement; Rfree, residual factor for diffraction data omitted from crystallographic refinement; NA, not available.

Fig. 1.

Cocrystal structure of the DENV-2 C protein in complex with ST148. (A) Structure of a DENV-2 C tetramer in complex with ST148. Two C dimers form a kissing tetramer. Two ST148 molecules bind to a pocket at the tetramer interface. ST148 and the C protein tetramer are presented as Corey-Pauling-Koltun (CPK) spheres and a cartoon, respectively. (B) Electron density of the ST148 dimer in the cocrystal. (C) Amino acids that form the compound-binding pocket. (Left) Seven residues from an individual C protein that form the ST148-binding pocket. Residues from different C proteins are indicated by different colors. H bonds (yellow) are formed between F33 (Phe33) and ST148. (Right) The schematic between ST148 and the C protein (using the MolBridge program). L38 (Leu38) stacks on the aromatic ring of ST148. (D) Inhibitor-induced compound-binding pocket. The ST148-binding site in the C protein alone (Left) and in the cocrystal structure (Center) is shown. Upon ST148 binding, F33_A (Phe33_A) rotates 2° and moves 2.7 Å out of the pocket (Right). The opposite F33_B (Phe33_B) rotates 58° and moves 3.3 Å away to form the inhibitor pocket.

Compared with the apo-C structure, the cocrystal structure shows two major conformational changes to create the ST148-binding pocket. First, the cocrystal structure shifts the C tetramer interface to a symmetrical conformation (Fig. 1A), with the P2 axes of both dimers becoming coaxial and the dimers rotated 47° from each other along their twofold coaxes (SI Appendix, Fig. S5A). To form the coaxial tetramers, one dimer needs to shift 10 Å from the apo to the inhibitor-bound conformation (SI Appendix, Fig. S5B). Second, the cocrystal structure changes the conformation of Phe33. In the apo-C structure, the side chain of Phe33 intrudes into the compound pocket (Fig. 1 D, Left). In the cocrystal structure, each helix 1 from the C tetramer is split at Phe33, and the rotation of this residue creates the ST148 pocket (Fig. 1 D, Middle). Specifically, Phe33 from one capsid molecule (labeled F33_B [Phe33_B] in Fig. 1 D, Right) rotates 58° out of the pocket and shifts 3.3 Å, allowing its carboxylate to form a H bond with the amine group of ST148 (Fig. 1C). Additionally, Phe33 from the other capsid molecule (labeled F33_A [Phe33_A] in Fig. 1 D, Right) shifts 2.7 Å away to form the compound-binding pocket. These conformational changes are essential for ST148 binding and C tetramer formation. Such a ST148-mediated C tetramer is assembled into a virion, leading to defective uncoating of a nucleocapsid when progeny virus infects new cells (see below in Figs. 5 and 6).

Fig. 5.

Incorporation of ST148 into DENV-2 particles. (A) Experimental scheme to examine the effects of ST148 on virion assembly (see text for details). (B–D) Effect of ST148 on virion assembly and infectivity. WT and resistant S34L DENV-2 were prepared from infected Vero cells in the presence of 10 µM ST148. Extracellular progeny viruses were quantified for viral RNA (B) and infectivity (C) using qRT-PCR and plaque assay, respectively. The viral RNA/PFU ratios are also presented (D). Error bars represent mean ± SD. Results are representative of three independent experiments with each one in triplicate. Statistical significance was determined by Student’s two-tailed t test, *P < 0.05, **P < 0.01, ***P < 0.001. (E) Effect of ST148 on viral protein level in supernatant and infected cells. Vero cells were infected with WT and S34L DENV-2 and treated with ST148 as described in (A). Culture supernatants and cell lysates were analyzed by Western blot. (F) Experimental scheme for LC-MS/MS measurement of ST148 in virions. (G) Incorporation of ST148 in the WT DENV-2 virion. As negative controls, WT ZIKV and DENV-2 were directly incubated with 10 µM ST148 for 2 h; the level of ST148 in virions was quantified by LC-MS/MS, and the amount of ST148 recovered from the WT ZIKV group was defined as 1. LC-MS/MS was performed on equal amounts of viruses (viral RNA). The recovered compound amounts (in fold changes) were normalized to the ST148 level in a control ZIKV sample.

Fig. 6.

Defective nucleocapsid uncoating of ST148-containing virion. (A) Reduced viral RNA translation after ST148-containing virion enters cells. Vero cells were infected with equal amounts (2 × 10⁷ RNA copies) of WT Rluc-D2 without integrated ST148 (Rluc-D2) or Rluc-D2 with integrated ST148 (Rluc-D2-ST148). Luciferase activities were measured at 1 to 6 h postinfection to indicate virus entry and initial RNA translation. Resistant virus Rluc-D2-S34L and Rluc-D2-S34L-ST148 were used as controls. (B) Experimental scheme for quantification of virus attachment and entry. (C) Intracellular level of viral RNA after virus entry. Vero cells were infected with WT DENV-2 (D2) or D2 with integrated ST148 (D2-ST148) and processed following B. Intracellular RNA was extracted, and viral RNA was quantified by qRT-PCR. Resistant virus D2-S34L and D2-S34L-ST148 were used as controls. (D) Effect of MG132 on intracellular viral RNA after virus entry. Vero cells were infected with D2 or D2-ST148 with or without MG132 (20 µM). The infected cells were processed following B, and viral RNA was quantified by RT-qPCR. Error bars represent mean ± SD from three independent experiments. Statistical significance was determined by Student’s two-tailed t test and two-way ANOVA multiple comparisons with correction using Tukey’s test, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.001. (E) Effect of MG132 on capsid stability after virus entry. Vero cells were infected with D2 or D-ST148 for 1 h at 4 °C. After removing virus inocula and washing in PBS three times, the cells were replenished with media in the presence or absence of 20 µM MG132. Intracellular capsid protein was detected at indicated time points by Western blot. (F) A model for the antiviral mechanism of ST148. See text for details.

DENV-2 C Structure in Solution.

We performed small-angle X-ray scattering (SAXS) experiments to examine the solution structure of DENV-2 C in the presence and absence of ST148. The SAXS curves of the apo-C protein fit the crystal dimer model (χ² = 1.4) with no evidence of tetramer formation (SI Appendix, Fig. S6 and Table S1). The fitting to the SAXS data was improved to χ² = 1.0 after modeling the six missing disordered N-terminal residues of the apo-C crystal structure. In contrast, the apo-C SAXS curve does not match the NMR model (χ² = 2.5; SI Appendix, Fig. S6). In the presence of ST148, the SAXS data fit a mixed sample with 22% of the C protein forming a tetramer (SI Appendix, Table S1). Collectively, the SAXS results suggest that apo-C forms a dimer in solution and the addition of ST148 induces a C tetramer.

Resistance Mechanism.

A single mutation S34L was reported to confer DENV-2 resistance to ST148 (19). Engineering S34L into a Renilla luciferase DENV-2 (Rluc-D2; Fig. 2A) did not attenuate viral replication (Fig. 2B) and conferred resistance to ST148 inhibition (50% effective concentration [EC₅₀] shift > 38 folds; Fig. 2C). Structurally, the hydroxyl group of Ser34 forms a weak electrostatic H bond with the amine of the Gly36 main chain (Fig. 2D). This stabilizes another weak electrostatic H bond between Ser34’s carboxylate and Met37’s amine. The S34L mutation abolishes these interactions, leading to a conformational change in the adjacent Phe33 whose carboxylate forms a H bond with the amine group of ST148. The loss of the Phe33–ST148 H bond results in the loss of compound binding. In support of this structural interpretation, ultracentrifugation experiments showed that wild-type (WT) apo-C formed dimers in solution; the addition of ST148 converted the majority of C into tetramers (Fig. 2E). Mutation S34L completely abolished C tetramer formation in the presence of ST148 (Fig. 2F). To further validate the resistance mechanism, we prepared two more mutant viruses: Mutant S34T maintains the weak H bond between Thr34 and Gly36, whereas mutant S34V abolishes this interaction. As expected, the S34T virus was inhibited by ST148 (EC₅₀ = 149 nM), whereas the S34V virus was partially resistant to compound inhibition (EC₅₀ = 4,119 nM; Fig. 2C). Altogether, the results support a resistance mechanism that mutation S34L abolishes two H bonds, leading to the loss of compound binding or C tetramer formation.

Fig. 2.

Resistance mechanism. (A) WT and mutant Rluc-D2. The mutant Rluc-D2 contained capsid mutations at amino acid 34. (B) Effects of resistant mutations on viral replication. Ten micrograms of WT or mutant Rluc-D2 RNA were electroporated into BHK-21 cells. The cells were quantified for luciferase activity at the indicated time posttransfection. (C) Antiviral activity. Vero cells were infected with WT or mutant Rluc-D2 (MOI = 0.5). The infected cells were treated with ST148 or DMSO as a control. EC₅₀ values are calculated from the luciferase activity measured 48 h postinfection. Data were normalized to DMSO treatment samples which were set to 100%. Error bars represent mean ± SD. Results are representative of three independent experiments with each one analyzed in triplicate. (D) The role of Ser34 in inhibitor binding. Ser34 forms H bonds with Gly36 (cyan) and Met37 (yellow). These interactions are critical to maintain Phe33’s H bond (yellow) with ST148. (E and F) Analytical ultracentrifugation analysis. Analytical ultracentrifugation assays were performed for the DENV-2 WT capsid (E) and resistant S34L capsid (F) in the presence (Right) or absence (Left) of ST148. The addition of ST148 converts the WT capsid from a dimer (18.6 kDa) to tetramer (37.2 kDa) conformation (E) but not the resistant S34L capsid (F).

Determinants of Drug Sensitivity of DENV-1, -3, and -4 Capsids.

To understand why ST148 does not potently inhibit DENV-1, -3, or -4, we prepared three chimeric Rluc-D2 viruses in which the DENV-2 C was substituted with the counterparts from other serotypes, resulting in Rluc-D2-D1capsid, Rluc-D2-D3capsid, and Rluc-D2-D4capsid (SI Appendix, Fig. S7A). The chimeric viruses did not significantly attenuate viral replication (SI Appendix, Fig. S7B). ST148 inhibited DENV-2 (EC₅₀ = 108 nM) but not DENV-1, -3, or -4 (EC₅₀ > 5 µM; SI Appendix, Fig. S7C). Notably, these EC₅₀ values are higher than those previously reported (19), possibly due to different experimental conditions. We did not test concentrations >10 µM because the solubility of ST148 is <10 µM.

Sequence alignment showed that, among the seven residues forming the compound pocket, two are fully conserved among the four serotypes of the C protein (Fig. 3A). To identify the determinants for ST148 sensitivity, we made three panels of mutant Rluc-D2 that contained individual variants from DENV-1, -3, or -4. These mutant viruses were tested for ST148 inhibition. Panel I Rluc-D2 mutants contained four DENV-1 variants (T30A, L35K, M37L, and L50F). None of these variants affected viral replication (Fig. 3B). The EC₅₀s were 110, 441, >5,000, 310, and 138 nM for the Rluc-D2, T30A, L35K, M37L, and L50F viruses, respectively (Fig. 3C), suggesting that L35K is responsible for DENV-1 resistance to ST148. To confirm this result, we reverse engineered the K35L mutation into the Rluc-D2-D1capsid (Fig. 3D). Remarkably, the single K35L mutation sensitized the Rluc-D2-D1capsid to ST148 inhibition (EC₅₀ = 41 nM; Fig. 3E), confirming the role of Leu35 in compound sensitivity. In the cocrystal structure, Leu35 forms a hydrophobic interaction at the interface between two kissing C dimers (4.0 to 4.3 Å apart; Fig. 3F); the longer and positively charged Lys35 would impose a steric clash and an electrostatic repulsion between these residues, resulting in weakened tetramer formation (Fig. 3G). Indeed, ultracentrifugation data showed that L35K abolished the tetramer formation of DENV-2 C in the presence of ST148 (Fig. 3H). Conversely, ST148 did not induce tetramer formation of WT DENV-1 C (Fig. 3I), whereas the mutation K35L conferred tetramer formation (Fig. 3J). Since T30A also increased EC₅₀ fourfold (from 110 to 441 nM), we examined the structural basis for this EC₅₀ change. As shown in Fig. 3K, Thr30 forms a backbone H bond with Val26; its hydroxyl group forms a strong H bond with the Gln27 carbonyl, and its methyl group (CG2) forms a hydrophobic packing against Leu38, Leu35, and ST148. The mutation T30A abolishes the strong H bond with Gln27 and hydrophobic packing but maintains the H bond with Val26 (Fig. 3L). The loss of the Ala30/Gln27 H bond and the hydrophobic packing interactions could account for the EC₅₀ change. Overall, the results demonstrate that Lys35 and, to a minor extent, Ala30 determine the resistance of DENV-1 to compound inhibition.

Fig. 3.

Determinants of drug sensitivity of DENV-1. (A) Sequence alignment of the ST148-binding pocket among the four DENV serotypes. Seven residues that form the ST148 pocket are indicated by arrows. Red and blue arrows represent identical and different residues among the four serotypes, respectively. (B and C) Viral replication and drug sensitivity of DENV-1 variants. DENV-1 variants were individually introduced to Rluc-D2 to evaluate their role in viral replication (B) and drug sensitivity to ST148 (C). (D and E) Reverse analysis. K35L was reverse engineered to Rluc-D2-D1capsid to evaluate its role in viral replication (D) and ST148 sensitivity (E). (F and G) Structural effect of the L35K variant on tetramer formation. In the cocrystal structure, WT L35 forms a hydrophobic interaction to facilitate tetramer formation (F). In modeled L35K mutant, Lys35 sterically clashes and repulses to disfavor the tetramer formation (G). The space available for the Leu or Lys side chains is outlined in red. (H–J) Analytical ultracentrifugation analysis. Analytical ultracentrifugation assays were performed to analyze C dimer and tetramer formation of DENV-2 L35K capsid (H), DENV-1 capsid (I), and DENV-1 K35L capsid (J) in the presence (Right) or absence (Left) of ST148. (K and L) Structural effect of the T30A variant on tetramer formation. In the cocrystal structure, Thr30 forms H bonds with Val26 and Gln27; its methyl group (CG2) forms hydrophobic interactions with ST148 (K). In the modeled T30A mutant, Ala30 loses the above interactions except the H bond with Val26 (L).

Panel II Rluc-D2 mutants contained four DENV-3 variants (T30A, L35R, M37L, and L50F), among which T30A, M37L, and L50F overlap with the panel I mutants (Fig. 3A). Mutation L35R did not compromise viral replication (SI Appendix, Fig. S8A) but increased the EC₅₀ to >5 µM (SI Appendix, Fig. S8B). Reverse engineering of R35L to Rluc-D2-D3capsid conferred ST148 inhibition (EC₅₀ = 262 nM; SI Appendix, Fig. S8B). Mechanistically, L35R may function through repulsion of C tetramer formation, as described for the DENV-1 L35K variant (Fig. 3 C–J). Collectively, the results demonstrate that L35R accounts for DENV-3 resistance to ST148 inhibition.

Panel III Rluc-D2 mutants contained five DENV-4 variants (T30V, L35T, M37L, L38F, and L50F), among which M37L and L50F mutations did not significantly change the EC₅₀ (Fig. 3 A–C). Rluc-D2 mutants containing T30V, L35T, or L38F did not attenuate viral replication (Fig. 4A) but increased EC₅₀s to 568, 613, and >5,000 nM, respectively (Fig. 4B), suggesting that L38F is the most important variant in conferring resistance. Reverse engineering of F37L (DENV-4 C is one residue shorter than DENV-2) to Rluc-D2-D4capsid virus did not substantially affect viral replication (Fig. 4C) but imparted compound inhibition (EC₅₀ = 235 nM; Fig. 4D). In the cocrystal structure, Leu38 locates on the outer layer of the compound pocket (Fig. 4E); the aromatic ring of Phe38 intrudes into the pocket and weakens compound binding (Fig. 4F). In addition to L38F, T30V and L35T also increased EC₅₀ by 6.2- and 6.7-fold, respectively; the structural changes that account for the EC₅₀ increase may be similar to the DENV-1 T30A and L35K variants (Fig. 3). Altogether, the results demonstrate that Phe38 and, to a lesser extent, Val30 and Thr35 confer DENV-4 resistance to ST148.

Fig. 4.

Determinants of drug sensitivity of DENV-4. (A and B) Viral replication and drug sensitivity of DENV-4 variants. The DENV-4 variants were individually engineered to Rluc-D2 to evaluate their effects on viral replication (A) and drug sensitivities (B). The EC₅₀ values are presented. (C and D) Reverse analysis. Mutation F37L was reverse engineered to Rluc-D2-D4 capsid to evaluate its effect on viral replication (C) and antiviral activity (D). (E and F) Structural effect of L38F on ST148 binding. In the cocrystal structure, Leu38 locates on the outer layer of the ST148 pocket (E). The inhibitor-binding pocket is shown in gray. In the modeled L38F mutant, the aromatic side chain of Phe38 intrudes into the pocket (F).

Inhibition of Nucleocapsid Uncoating.

The above results prompted us to hypothesize that 1) ST148 is integrated into progeny virions in complex with the C tetramer and 2) the compound-containing virions are defective in nucleocapsid uncoating when infecting new cells. Four sets of experiments were performed to test these hypotheses. First, we excluded the possibility that ST148 directly blocks WT DENV-2 infection. Vero cells were infected with WT or resistant S34L DENV-2 (multiplicity of infection [MOI] = 2) in the presence of 5 µM ST148. The infected cells were quantified for intracellular viral RNA at 1 to 6 h postinfection. ST148 treatment did not change the intracellular levels of WT or S34L viral RNA (SI Appendix, Fig. S9A), indicating that the compound does not directly block virus attachment or entry. We also infected cells with WT Rluc-D2 (MOI = 2) in the presence of ST148; no difference in luciferase activity was observed at 2 or 4 h postinfection between the compound-treated and mock-treated groups (SI Appendix, Fig. S9B). The results indicate that, when directly treating WT DENV-2 infection, ST148 does not affect nucleocapsid uncoating and viral RNA translation.

Second, we examined whether the compound affects virion assembly (Fig. 5A). Vero cells were infected with WT DENV-2 (MOI = 1) without ST148. At 12, 24, or 36 h postinfection, cells were washed three times with phosphate buffered saline (PBS) to remove extracellular virus and replenished with medium containing ST148 (10 µM). After incubation for another 12 h, extracellular levels of viral RNA and infectious virus were measured by qRT-PCR and plaque assay, respectively. Surprisingly, ST148 treatment did not significantly reduce extracellular viral RNA (Fig. 5B), but reduced extracellular infectious virus by ∼10-fold (Fig. 5C), leading to an increased RNA copy/(plaque-forming unit [PFU]) ratio (Fig. 5D). As controls, ST148 treatment did not affect extracellular levels of viral RNA, infectious virus, RNA/PFU, or E protein for resistant S34L DENV-2 (Fig. 5 B–E). These results suggest that ST148 does not affect virion assembly and release but reduces the infectivity of progeny virus.

Third, we directly measured the incorporation of ST148 into virions using targeted liquid chromatography mass spectrometry (LC-MS/MS; Fig. 5F). As a control, ZIKV, which is not inhibited by ST148 (SI Appendix, Fig. S10), was incubated with ST148 for 2 h, precipitated, and subjected to LC-MS/MS analysis. The amount of ST148 detected from ZIKV was defined as adventitious and was set as the background level (Fig. 5F). After direct incubation of WT DENV-2 with ST148, only background levels of compound were measured (Fig. 5G), suggesting that ST148 does not penetrate into WT DENV-2. Next, we measured the amount of ST148 from the virions produced from ST148-treated, DENV-2-infected cells (i.e., the amount of compound recovered from the virions that were formed in the presence of ST148 during infection). The WT DENV-2 contained ST148 that was 8.93-fold higher than the background level, whereas resistant S34L DENV-2 contained only a background level of ST148 (Fig. 5G). The results indicate that ST148 is incorporated into WT progeny DENV-2 but not the resistant virion. We conclude that 1) ST148 is unable to bind capsid once the protein has already been incorporated into a virion and 2) ST148-mediated tetramer formation occurs before or during virion assembly.

Fourth, we examined the effect of incorporated ST148 on virus entry and nucleocapsid uncoating. We prepared a batch of WT Rluc-D2 containing ST148 (Rluc-D2-ST148; compound added at 36 h postinfection, as shown in Fig. 5A). As a control, we also prepared a batch of resistant S34L Rluc-D2 (Rluc-D2-S34L-ST148) using an identical protocol. After infecting Vero cells with equal amounts of viruses (i.e., equal viral RNA levels), Rluc-D2-S34L-ST148 (virus containing ST148) and Rluc-D2-S34L produced similar luciferase profiles at 1 to 6 h postinfection, whereas Rluc-D2 (virus without ST148) generated significantly higher luciferase activity than Rluc-D2-ST148 (virus containing ST148; Fig. 6A). Corroboratively, equivalent amounts of intracellular viral RNAs were detected from the D2-S34L-ST148- and D2-S34L-infected cells, whereas significantly lower levels of viral RNAs were detected from the D2-ST148-infected cells than those from the D2-infected cells (Fig. 6 B and C). These results imply that 1) incorporated ST148 may block nucleocapsid uncoating of D2-ST148 and 2) the nonproductive nucleocapsid (from the D2-ST148 infection) may be degraded by the host proteasome. To test the latter possibility, we treated the D2-ST148- and D2-infected cells with MG132, a proteasome inhibitor. Indeed, MG132 restored the intracellular viral RNA levels of D2-ST148 to those of D2 (Fig. 6D). Western blot showed that, in the absence of MG132, intracellular C from the D2-ST148 infection was degraded at a faster rate than that from the D2 infection (Fig. 6 E, Top); treatment with MG132 significantly increased the stability of C from both D2-ST148 and D2 infections (Fig. 6 E, Bottom). Collectively, the results suggest that virions containing ST148 are defective in nucleocapsid uncoating, leading to C degradation through the proteasome pathway. However, our results do not exclude the possibility that the capsid-ST148-RNA complex may make it a strong target for the proteasome degradation so that the nucleocapsid does not have a chance to disassociate.

Discussion

The current study has uncovered an antiviral mechanism through inhibitor-induced tetramerization of the DENV-2 C protein. ST148-bound C tetramers are productively incorporated into progeny virions during virus assembly. However, the resulting virions are defective in nucleocapsid uncoating when infecting new cells. This mechanism is distinct from those previously reported for capsid inhibitors of human rhinovirus, hepatitis B virus (HBV), or HIV. The human rhinovirus inhibitors (e.g., pleconaril) block virus attachment to cells through binding to a hydrophobic pocket in the VP1 protein (25–27). The HBV C inhibitors (e.g., ABI-H0731 and RO7049389) interfere with pregenomic RNA packaging (28–30) or induce aggregated or disrupt capsid structure (31–34). The HIV inhibitor (e.g., GS-CA1) binds directly to the capsid and interferes with capsid-mediated nuclear import of viral DNA, capsid assembly, and virus production (25).

Recent studies suggested that flavivirus NS2A recruits structural C/prM/E proteins, NS2B/NS3 protease, and genomic RNA for virion assembly in a highly regulated manner (35, 36). The arrangement of C protein and genomic RNA to form nucleocapsid is poorly understood. A beads-on-string model was proposed for ZIKV nucleocapsid formation through viral RNA wrapping around individual C dimers (37). Since DENV-2 C exists as a dimer in solution (as indicated by NMR, SAXS, and ultracentrifugation analysis), it is likely that the dimer, rather than the tetramer, is the functional C unit for nucleocapsid formation. Due to distinct charge distribution on the C dimer surface (SI Appendix, Fig. S4A), the hydrophobic helix 1 side was proposed to interact with the membrane, while the positively charged helix 4 side binds viral RNA (22). During virion assembly, the hydrophobic helix 1 surface of individual C dimers is likely positioned to the cytosolic side of the budding prM/E membrane, while the positively charged helix 4 surface binds genomic RNA (Fig. 6F). Virion formation could be driven by the prM/E-mediated membrane curvature (38, 39), long-distance viral RNA base pairing (40, 41), and unidentified capsid-mediated interactions (e.g., through the 26 N-terminal amino acids). Our results showed that the ST148-induced C tetramer could be incorporated into virions during assembly; this is not surprising because the outside surface of the C tetramer is positively charged with the hydrophobic helix 1 forming the dimer–dimer interface (SI Appendix, Fig. S4B). However, the tetramer-containing virus is defective in uncoating, leading to proteasome- and RNase-mediated degradation of the C protein and genomic RNA, respectively (Fig. 6F). We currently do not know the copy numbers of the “poisoned” tetrameric C in those ST148-induced noninfectious particles. Future studies are required to define the minimal numbers of the poisoned tetrameric C to confer defective nucleocapsid uncoating. Nevertheless, our results suggest that flavivirus nucleocapsid uncoating is also a highly regulated process. The conformational changes between the apo-C structure and the ST148-C complex structure clearly illustrate the underlying mechanism of the observed antiviral activity.

Although ST148 showed promising efficacy in the cell culture and mouse model (19), its development has been hampered by two major hurdles: 1) a narrow antiviral spectrum with efficacy against only DENV-2 and 2) low aqueous solubility that curbs formulation and pharmacokinetics. The antiviral spectrum could be improved through rational design of analogs that have more interactions with the compound-binding pocket. Specifically, the compound could be extended to interact with the unoccupied space in the cocrystal structure or to induce expansion of the current binding pocket. The increased compound–protein interactions may overcome the repulsion effect (e.g., DENV-1 Lys35 and DENV-3 Arg35) or side chain intrusion (e.g., DENV-4 Phe37) that currently limits the potency against other DENV serotypes. Alternatively, chemically linking two ST148 molecules into one compound may improve potency to achieve panserotype activity. Active inhibitors from both approaches could be cocrystallized to guide iterative compound synthesis. During the above rational design effort, compounds with improved solubility and other drug-like properties could be prioritized for further development.

In summary, the current study has provided the important structural, biochemical, and virological mechanism of the DENV capsid inhibitor. Such information is essential for further development of the current inhibitor for treatment of DENV. This antiviral approach may be expanded to inhibit other closely related flaviviruses that frequently pose threats to global public health.

Materials and Methods

Crystal Structure Determination.

The DENV-2 C protein was purified and concentrated to ∼40 mg/mL for crystallization. Cocrystals of the DENV-2 C/ST148 complex were grown in 2.0 µL crystallization drops consisting of 1.0 µL of protein inhibitor solution (1 mM capsid and 2 mM ST148) and 1.0 µL of reservoir solution. The crystallization drops were suspended over 500 µL of reservoir solution (3.0 M NaNO₃, 100 mM 2-(N-morpholino)ethanesulfonic acid [MES] [pH 6.8], and 5% glycerol) in a 24-well plate (Hampton Research) at 19 °C for 20 h. For crystallization of DENV-2 apo-C, 40 µg of protein were seeded, and the crystals were grown with reservoir solution containing 3.0 M NaNO₃, 100 mM Hepes (pH 7.2), and 5% glycerol at 19 °C for 3 d. Crystals of DENV-2 apo-C or C/ST148 complex were soaked in mother liquor with 30% glycerol added as a cryoprotectant prior to flash freezing. The apo-C crystal and C/ST148 cocrystal data were collected at the Advanced Photon Source (APS) Life Sciences Collaborative Access Team (LS-CAT) 21-ID-G and LS-CAT 21-ID-D beamlines, respectively. The data were processed using HKL2000 (Table 1). The cocrystals diffracted to 1.5 Å and were in the P2 space group. The cocrystal structure was solved using molecular replacement in Phaser (42) with the WNV C structure (PDB ID code 1SFK) (23). The model was refined in Refmac (43) with restrained anisotropic temperature factors. The geometry restraints for the inhibitor were generated using ProDRG (44). This 1.5 Å resolution cocrystal structure was used as a search model in Phaser to determine the apo-C structure. The 2.6 Å apo-C structure was refined using Phenix (45), with non-crystallographic symmetry (NCS) restraints and seven anisotropic atomic displacement parameter groups (TLS groups) per chain. Model building was performed using Coot (46). The figures were generated using PyMOL (47).

Construction of Recombinant Viruses.

Recombinant viruses were constructed using an infectious complementary DNA (cDNA) clone of Renilla luciferase DENV-2 (48). The mutations were introduced by overlap PCR, and the amplicons were inserted in the cDNA clones using standard molecular cloning methods (49). Plasmids were linearized, and viral RNAs were in vitro transcribed using a T7 mMESSAGE mMACHINE kit (Ambion) as described previously (49). Ten micrograms of RNA were electroporated into BHK-21 cells, and cells were incubated with 10% fetal bovine serum (FBS) Dulbecco's modified Eagle’s medium (DMEM) at 37 °C. Cells were transferred to a 30 °C incubator with 2% FBS fresh DMEM at 24 h postelectroporation. Viruses were harvested on day 5 postelectroporation. All primers are listed in SI Appendix, Table S2.

Luciferase Reporter Virus Transient Transfection Assay.

The RNAs of Renilla luciferase DENV-2 and mutations described above were electroporated into BHK-21 cells. The electroporated cells were seeded into a 12-well plate (3.2 × 10⁵ cells/well). At various time points, cells were washed twice with PBS and lysed in 200 µL of lysis buffer. Luciferase signals were measured by mixing with Renilla luciferase substrates (Promega) and read by Cytation 5 (Biotek) according to the manufacturer's instructions.

Antiviral Assay.

Vero cells were used to study the antiviral activity of ST148 against Rluc-D2 and DENV-1, -3, -4 chimeric Rluc-D2 viruses. Vero cells were seeded at 10⁴ cells/well in a white opaque 96-well plate (Corning) with 50 µL of medium containing 2% FBS without phenol red. ST148 was twofold diluted from 5 mM in dimethyl sulfoxide (DMSO), and the same amount of compound dilutions was mixed with virus aliquots. The cells were infected with 50 µL of viruses (Rluc-D2 or DENV-1, -3, -4 chimeric Rluc-D2) in the presence of serial dilutions of ST148 or DMSO as control. At 48 h postinfection, luciferase activity was quantified using ViviRen Live Cell Substrate (Promega) and read by Cytation 5 (Biotek) according to the manufacturer’s protocol. EC₅₀s of ST148 were determined by a nonlinear regression curve; the bottom and top were constrained to 0% and 100%, respectively.

Data Analysis.

Data were analyzed with GraphPad Prism 7 software. Data are expressed as the mean ± SD. Comparisons of groups were performed using Student’s t test and two-way ANOVA multiple comparisons with correction using Tukey’s test. A P value of <0.05 indicates statistically significant.

Data Availability.

Data for the structural studies were deposited in the Protein Data Bank (PDB ID codes 6VSO and 6VG5); all pertinent data are available in the main text and SI Appendix.