Dengue virus protease activity modulated by dynamics of protease cofactor

Abstract

The viral protease domain (NS3pro) of dengue virus is essential for virus replication, and its cofactor NS2B is indispensable for the proteolytic function. Although several NS3pro-NS2B complex structures have been obtained, the dynamic property of the complex remains poorly understood. Using NMR relaxation techniques, here we found that NS3pro-NS2B exists in both closed and open conformations that are in dynamic equilibrium on a submillisecond timescale in aqueous solution. Our structural information indicates that the C-terminal region of NS2B is disordered in the minor open conformation but folded in the major closed conformation. Using mutagenesis, we showed that the closed-open conformational equilibrium can be shifted by changing NS2B stability. Moreover, we revealed that the proteolytic activity of NS3pro-NS2B correlates well with the population of the closed conformation. Our results suggest that the closed-open conformational equilibrium can be used by both nature and humanity to control the replication of dengue virus.

Significance

The dengue virus protease is an attractive target for drug development against dengue fever because it is essential for virus replication. However, its structure-based drug development has been unsuccessful because of the shallow substrate-binding pocket. The study presented here demonstrates for the first time, to our knowledge, that the protease activity can be reduced dramatically by shifting the closed-open conformational equilibrium of the protease in complex with its cofactor from the majority of a closed conformation to the majority of an open conformation. Moreover, our work clarifies the structure of the open conformation which has been elusive for a long time. Our results also suggest an alternative method for designing protease inhibitors based on the closed-open conformational equilibrium.

Introduction

Dengue fever is a disease that is caused by the dengue virus, which is transmitted by the mosquito species Aedes aegypti. It has been estimated that up to four billion people are at risk of dengue fever globally (1), with as many as 400 million new infections annually (2). Symptoms of dengue fever include fever, headache, muscle and joint pains, nausea, vomiting, swollen glands, and rashes (3). More severe cases manifest as dengue hemorrhagic fever or dengue shock syndrome and can be fatal (4). The dengue vaccine Dengvaxia has been approved recently, and a few more are undergoing clinical trials. However, Dengvaxia can cause antibody-dependent enhancement and is therefore not recommended for people who have never been infected with dengue before (5). Meanwhile, there are no approved drugs for treating dengue yet. It is therefore of great importance and urgency to understand more about the dengue virus, so that effective treatments can be developed against it.

The genome of the dengue virus consists of a positive-sense single-stranded RNA ∼11 kb long (6). It contains an open reading frame that encodes for a single polyprotein containing three structural proteins and seven nonstructural proteins (3,7). For each of the proteins to perform their respective functions, the polyprotein needs to be cleaved at various sites by both the host proteases and the viral protease (8). If the proteolytic processing of the viral polyprotein can be disrupted, the dengue virus will no longer be able to replicate and propagate, effectively stopping the progression of dengue fever on its tracks. This makes the viral protease an attractive target for drug development. To do so, sufficient structural and dynamic information of the dengue protease is required.

The N-terminal protease domain of nonstructural protein 3 (NS3pro) is a chymotrypsin-like serine protease responsible for the cleavage of the polyprotein at various sites (9, 10, 11, 12). NS3pro also cleaves within some viral proteins like the capsid (C) protein (13), NS2A (14), NS4A (15), and even NS3 itself (16,17). NS3pro is unable to fold correctly by itself and requires the hydrophilic region of nonstructural protein 2B (NS2B) as a cofactor for its correct folding and protease activity (9,18,19). The N-terminal 18 amino acids of the hydrophilic region of NS2B (residues 49–66) allow NS3pro to adopt its correct structure by contributing a β-strand to complete the β-barrel of NS3pro (18). However, this region of NS2B is still insufficient for the protease activity of NS3pro (18,19). The C-terminal portion of the hydrophilic region of NS2B (residues 63–100, denoted as NS2Bc hereafter) contributes a β-hairpin that completes the substrate-binding pocket of NS3pro and is therefore necessary for its activity (Fig. 1 a; (20)).

Figure 1.

Comparison of NS2B-NS3pro crystal structures. NS2B is colored red, NS3pro is colored green, and the catalytic triad is colored purple. (a) DENV3, closed state (PDB: 3U1I) (20). (b) DENV2, open state (PDB: 2FOM) (18). To see this figure in color, go online.

Existing crystal and solution structures of the NS2B-NS3pro complexes from dengue virus, West Nile virus, and Zika virus show that NS2B can adopt drastically different conformations (18, 19, 20, 21, 22, 23, 24, 25, 26). In the presence of inhibitor, NS2Bc is in close proximity to the catalytic triad of NS3pro and completes the substrate-binding pocket (Fig. 1 a). This structure is denoted as the closed conformation, which is shown to be the enzymatically active form (20). In the crystal structure of NS2B-NS3pro with aprotinin (Protein Data Bank, PDB: 3U1J), the electron density of most residues in NS2Bc (residues 70 to the end) was missing (20), suggesting a highly dynamic nature of this region. In the absence of an inhibitor, except for one of the Zika virus NS2B-NS3pro structures, NS2Bc turns away from the catalytic triad of NS3pro in the crystal state, and this structure is denoted as the open conformation (Fig. 1 b; (18)). However, in the crystal structure of the inhibitor-free NS2B-NS3pro from Zika virus (PDB: 5GPI), NS2B-NS3pro adopts the closed conformation (26). In all the reported open conformation structures, NS2Bc adopts different conformations with various secondary structure contents in different structures, and some residues are missing in some structures (Table S1), implying that this region is highly dynamic in the open state. In the solution state, no inhibitor-free structure is available, probably because of the dynamic nature of NS2B (27,28). Nevertheless, previous NMR studies showed that the inhibitor-free NS2B-NS3pro complexes from dengue and West Nile viruses exist mainly in a conformation highly similar to the closed conformation (28, 29, 30), which undergoes conformational exchange with a minor open conformation (29,30). All the previous studies indicate the dynamic nature of the protein. So far, however, no detailed studies on the dynamics of the NS2B-NS3pro complex have been reported. Moreover, apart from a study that showed a decrease in protease activity of NS2B-NS3pro upon cross-linking into the open conformation (31), the relationship between the dynamics and protease activity of NS2B-NS3pro remains unclear.

In this study, we investigated the role of NS2B dynamics in the catalytic activity of NS2B-NS3pro using NMR relaxation techniques, mutagenesis, and an activity assay. Structural information of the elusive open conformation was revealed by NMR dynamics experiments, and the activity level of the dengue protease was found to be correlated to the population size of the closed conformation.

Materials and methods

Cloning of DENV2 NS2B and NS3pro

The codon-optimized gene encoding DENV2 NS3pro (residues 14–185, based on the accession number ARO84675.1) was synthesized by GenScript (Piscataway, NJ) and ligated into pSKDuet01 using the sites NcoI and XhoI. DENV2 NS2B (residues 48–100) was assembled from four primers using PCR, after which an N-terminal 6×His-Smt3 tag was added by PCR. The resulting PCR product was ligated into pSKBAD2 using the sites NdeI and HindIII. pSKDuet01 and pSKBAD2 were gifts from Hideo Iwai (Addgene plasmids #12172 and #15335; Watertown, MA) (32).

Point mutants of DENV2 NS2B-NS3pro were generated through site-directed mutagenesis. The mutant inserts were generated in a first PCR step as two fragments. The two fragments were then linked by a second PCR step to generate a single fragment. PCR products of NS2B mutants were ligated into pSKBAD2 using the sites NdeI and HindIII, and PCR products of NS3pro mutants were ligated into pSKDuet01 using the sites NcoI and XhoI.

Expression of DENV2 NS2B-NS3pro complex

BL21 (DE3) cells with both plasmids containing 6×His-Smt3-NS2B and NS3pro were grown in lysogeny broth medium with 100 μg/mL ampicillin and 25 μg/mL kanamycin at 37°C and 200 rpm until an OD₆₀₀ of 0.6 was reached. Expression of 6×His-Smt3-NS2B was induced with 2 g/L arabinose for 1 h at 37°C and 200 rpm, after which 0.1 mM isopropyl β-D-1-thiogalactopyranoside was added to induce the expression of NS3pro at 30°C and 200 rpm for 3 h.

For the expression of isotopically labeled NS2B complexed with unlabeled NS3pro, a protocol derived from Muona et al. was used (33). BL21 (DE3) cells with both plasmids containing 6×His-Smt3-NS2B and NS3pro were grown in the same lysogeny broth medium and condition as that for the nonlabeled protein until an OD₆₀₀ of 0.6 was reached. The cells were harvested by centrifugation and then transferred to M9 medium. After growing for 1 h, 2 g/L arabinose was added to induce the expression of 6×His-Smt3-NS2B for 3 h at 37°C and 200 rpm. Subsequently, the cells were collected by centrifugation. After washing the cells with Terrific Broth, the cell pellet was resuspended in Terrific Broth medium containing 100 μg/mL ampicillin and 25 μg/mL kanamycin. Finally, 0.1 mM isopropyl β-D-1-thiogalactopyranoside was added to induce the expression of NS3pro for 3 h at 30°C and 200 rpm.

Purification of DENV2 NS2B-NS3pro complex

The cells were harvested by centrifugation and lysed by sonication. The protein was purified from the supernatant of the lysate using a 5 mL HisTrap HP column (GE Healthcare, Chicago, IL). The Smt3 tag was cleaved off before further purification by a HiLoad 16/600 Superdex 75 size exclusion column (GE Healthcare) in NMR buffer (20 mM sodium phosphate (pH 6.5), 10 mM sodium chloride, 1 mM EDTA). Aliquots were prepared and stored at −80°C for activity assays.

Activity assay of DENV2 NS2B-NS3pro

Activity assays were performed in 100 μL reaction volumes on a 96-well plate as described in a previous study (34). Briefly, 10 nM NS2B-NS3pro was incubated with the substrate benzoyl-Nle-Lys-Arg-Arg-7-amino-4-methylcoumarin (Bz-nKRR-AMC) (GenScript) at concentrations varying from 2 to 200 μM. The assay buffer used was 50 mM Tris (pH 8.5), 20% glycerol (v/v), and 1 mM 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate. The fluorescence was then monitored in a Tecan Infinite 200 PRO microplate reader (Tecan, Männedorf, Switzerland) at 25°C, with excitation and emission wavelengths of 380 and 450 nm, respectively. Fluorescence readings were converted to AMC concentrations using a standard AMC curve. Data points from the activity assay were fitted to the Michaelis-Menten equation (35).

CD spectroscopy

Circular dichroism (CD) spectra were recorded on samples containing 10 μM protein in the NMR buffer using a Jasco J-1100 CD spectrophotometer (Jasco, Tokyo, Japan) at 20°C.

NMR spectroscopy

NMR samples were prepared at concentrations of 0.5–0.7 mM protein in the NMR buffer plus 5% D₂O. All NMR experiments were performed on a Bruker AVANCE 800 MHz spectrometer (Billerica, MA) equipped with a cryoprobe at 25°C. NMR data were processed using NMRPipe and analyzed using SPARKY (36).

Three-dimensional HNCOCA, HNCA, and ¹⁵N-edited NOESY-HSQC spectra were acquired on ∼0.7 mM of ¹³C,¹⁵N-labeled NS2B in complex with unlabeled NS3pro. Heteronuclear ¹⁵N nuclear Overhauser effects (NOEs) were measured on a ¹⁵N-labeled NS2B complexed with unlabeled NS3pro (∼0.5 mM) from two sets of data obtained in the presence and absence of proton saturation. In the measurements, the recycle delay was 8 s, and the proton saturation time was 4 s. The ¹⁵N NOE values were derived from ratios of the peak intensities with and without proton saturation (37).

CPMG relaxation dispersion (RD) experiments were performed on a ¹⁵N-labeled NS2B-unlabeled NS3pro complex (∼0.5 mM) using a pulse sequence established previously (38,39). A constant CPMG delay (T_CPMG) of 20 ms was used for each CPMG experiment. The CPMG frequencies (v_CPMG) used were 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 800, 900, 1000, 1200, 1400, 1600, and 2000 Hz. The CPMG experiment at v_CPMG = 100 Hz was performed twice to estimate the experimental error. The total experimental time was ∼18 h.

¹⁵N chemical exchange saturation (CEST) experiments were performed on a ¹⁵N-labeled NS2B-unlabeled NS3pro complex (∼0.5 mM) using two weak RF fields of 13 and 26 Hz, with a CEST mixing time of 0.5 s. At the higher RF field, 27 ¹H-¹⁵N HSQC spectra were collected with ¹⁵N carrier frequencies from 107 to 133 ppm at 1 ppm intervals. On the other hand, at the lower RF frequency, 38 ¹H-¹⁵N HSQC spectra were collected with ¹⁵N carrier frequencies from 106 to 135 ppm at 0.5 ppm intervals between 115 and 124 ppm and 1 ppm intervals otherwise. The total experimental time was 44 h. For peaks with ¹⁵N chemical shift greater than 120 ppm, data points before 111 ppm were used to estimate the error. On the other hand, for peaks with ¹⁵N chemical shifts smaller than 120 ppm, data points after 129 ppm were used to estimate the error.

Analyses of NMR data

Backbone ¹⁵N and ¹H_N assignments of NS2B were achieved using three-dimensional HNCA and HNCOCA spectra because of a small number of residues. In the cases in which ambiguities occurred because of similar ¹³Cα chemical shifts, ¹⁵N-edited NOESY-HSQC was used.

For a system without conformational exchange, each ¹⁵N CEST profile displays a single dip that is symmetric about the ¹⁵N resonant frequency. For a system with conformational exchange that can be described by a two-state model, the CEST profile of one ¹⁵N spin shows two dips when the exchange rate (k_ex) is similar to or smaller than the resonant frequency difference between the two states (k_ex < Δν or k_ex ∼Δν). In this case, kinetics parameters and Δν can be obtained from the CEST data. When k_ex > 2Δν, however, the CEST profile displays a single dip that is asymmetric. In this, it is difficult to obtain kinetics parameters and Δν by using only the CEST data. Analyzing RD data alone, we can get |Δν| (the absolute value) for an exchange system but cannot obtain the sign. As demonstrated previously, the sign can be determined by analyzing RD and CEST data simultaneously, even in the case in which there is a single CEST dip, because the asymmetry of the dip depends on the sign and magnitude of Δν (40). With the value of Δν (including the sign), it is possible to obtain the chemical shifts of the minor state (δ_I = δ_N + Δν, δ_I and δ_N are the chemical shifts of the minor state and major native state, respectively).

To obtain conformational exchange parameters and chemical shifts of the minor state of NS2B-NS3pro, RD and CEST data of residues displaying obvious relaxation dispersion were simultaneously fitted to a two-state exchange model (N ⇌ I, where N and I represent the major native state and minor non-native state, respectively) as described previously (40). Briefly, the data for each residue (j) were fitted to estimate individual exchange rate (k_ex(j)), population (p_I(j)), chemical shift (δ_I(j)), intrinsic transverse relaxation rate (R₂(j)), and longitudinal relaxation rate (R₁(j)). From these estimations, initial values of the fitting parameters were determined roughly. Next, the data for all the residues were fitted globally to extract global kinetics parameters (exchange rate and population) and residue-specific parameters (chemical shifts and relaxation rates). In the fitting, the R₂- and R₁-values for each residue were assumed to be independent of conformational states. The assumption of equal intrinsic relaxation rates for different conformational states does not introduce appreciable errors into the extracted exchange parameters even if the exchange is between folded and unfolded states, as demonstrated previously (41).

Error estimation was done using a Monte Carlo method described previously (42). Simulated RD and CEST data for each residue were generated using the global exchange parameters (k_ex, p) and residue-specific parameters (δ_I, R₂, and R₁) obtained from the global fitting. The synthetic data were obtained by adding a Gaussian error of 5%, or the same as the experimental error if the experimental one is larger than 5%, to each of the profiles. To obtain a set of exchange parameters, the synthetic profiles were fitted in the identical way as the experimental data. The errors were calculated from the distributions of 200 sets of parameters obtained from 200 sets of synthetic data.

Chemical shift perturbation

For a given residue, the chemical shift perturbation was calculated using the equation

$$\Delta \delta =\sqrt{\Delta {\delta }_H^2+{\left(\Delta {\delta }_N\times {\alpha }_N\right)}^2},$$

where Δδ_H (Δδ_N) is the difference of ¹H (¹⁵N) chemical shifts between the wild-type (WT) protein and its mutant and α_N is the scaling factor for the ¹⁵N chemical shift difference (43). A value of 0.17 was used for α_N in this study.

Results

Dynamics of NS2B in NS2B-NS3pro complex on ns–ps timescale

The NS2B-NS3pro complex consists of 225 amino acid residues. A previous study showed that ∼50% of the NS2B ¹H-¹⁵N correlations (peaks) are unavailable for providing dynamic information, mainly because of peak overlap (28,44). To overcome this overlapping problem, we obtained a sample of ¹³C,¹⁵N-labeled NS2B (S48-R100) complexed with unlabeled NS3pro using a sequential expression approach described previously (33). Except for S48 and D81, backbone ¹H-¹⁵N correlations for all other residues were observable in the HSQC spectrum with good peak dispersion (Fig. 2). The assignments were achieved using triple resonance experiments.

Figure 2.

HSQC spectrum of ¹⁵N-labeled DENV2 NS2B complexed with unlabeled NS3pro, with peak assignments included. The peaks of W61s (side chain HεNε), G82, and K87 are aliased by 20 ppm in the ¹⁵N dimension. To see this figure in color, go online.

The rigidity of NS2B in the wild-type complex was probed through the measurement of heteronuclear ¹⁵N NOE (HetNOE) values. High HetNOE values (>0.5) were observed for residues E52–Q64, E66–S68, and I73–K87 of NS2B and relatively low values (<0.45) for the C-terminal region (N88–R100), the N-terminal end (A49–L51), Q64, A65, E66, G69, and S70 (Fig. 3). The result indicates that the C-terminal region N88–R100 and N-terminal end are flexible on the nanosecond to picosecond timescale. This is consistent with previous x-ray crystallography studies, which showed that the C-terminal region is undetectable in both closed and open conformations because of high flexibility and lack of regular helical or β-strand structure (18,20,21). In the closed conformation of NS2B-NS3pro complex, NS2B consists of four short β-strands (β1: D50–A57, β2: E66–G69, β3: I73–I78, and β4: S83–I86) and three loops (loop 1: D58–A65, loop 2: S70–P72, and loop 3: S79–G82) between the strands. Interestingly, most of residues in these loops are as rigid as those residues in the β-strands. Previous NMR studies found that the residues in the β-strands of NS2B have similar rigidity as those in the β-strands of NS3pro (28). In the open conformations of NS2B-NS3pro complexes that were solved previously by x-ray crystallography, NS2Bc adopts different conformations with various secondary structure contents in different structures (Table S1). This suggests that NS2Bc is likely disordered in the open state in aqueous solution. Based on previous crystallography and NMR studies, the inhibitor-free NS2B-NS3pro complex can adopt both the closed and open conformations, but the populations of the two conformations are unknown in aqueous solution. If the region of E63–K87 is disordered, it should be significantly more flexible than the first β-strand (D50–A57). On the basis of HetNOEs, except for a few residues in loop 1, β2, and loop 2 (Q64, A65, E66, G69, and S70), all other residues within E52–K87 have similar rigidity. Therefore, most residues in the region of E52–K87 should be in contact with NS3pro, and the wild-type NS2B-NS3pro complex is mainly in the closed conformational state. This is consistent with the conclusion from previous NMR studies (28, 29, 30). The biased open conformation structures solved may be caused by using an artificial linker G₄SG₄ between NS2B and NS3pro, as all the reported open conformation structures were solved by using a single-chain construct of NS2B and NS3pro connected by an artificial G₄SG₄ linker, but the closed conformation structure was obtained by using a naturally unlinked NS2B-NS3pro in the absence of inhibitor (26). The other possible reason is the used crystallization conditions unfavorable to the closed conformation.

Figure 3.

HetNOE values of NS2B residues for the wild-type (black asterisks), the NS2B M84P mutant (red diamonds), and the NS2B I73C + NS3pro L115C disulfide bond mutant (blue circles). The secondary structure of NS2B is included within the plot. To see this figure in color, go online.

Dynamics of NS2B in NS2B-NS3pro complex on millisecond to microsecond timescale

To probe conformational exchange on a millisecond to microsecond timescale, we carried out RD experiments. Residues located in the N- (A49–E62) and C-terminal (E89–R100) regions displayed nearly no ¹⁵N RD with R_ex-values smaller than 1 s⁻¹ (Fig. 4, a and g). On the other hand, 21 out of 24 residues with available assignments in the region of E63–N88 exhibited significant RD with R_ex-values larger than 3 s⁻¹ (Fig. 4, c and e). Here, R_ex is defined as R₂^eff(2000) − R₂^eff(50), where R₂^eff(2000) and R₂^eff(50) were the relaxation rates measured at CPMG fields of 2000 and 50 Hz, respectively. The RD data indicate that at least one “invisible” minor state (I) is in dynamic equilibrium with the observed major native state (N) on a millisecond timescale for the region of E63–N88. To examine whether slow conformational exchange processes on subsecond timescales exist, CEST experiments were also performed. No residues displayed two obvious dips, indicating that the conformational exchange is too fast to be detected by CEST. To obtain structural information of the “invisible” state and kinetics parameters for the conformational exchange process, a two-state model (N ⇌ I) was used to fit both the CEST and RD data simultaneously.

Figure 4.

Representative ¹⁵N relaxation dispersion profiles of DENV2 NS2B residues (a, c, e, and g), with their corresponding CEST profiles (b, d, f, and h). In the CPMG profiles, open circles (○) represent data points collected for the wild-type, and asterisks (^∗) represent data points collected for the NS2B I73C + NS3pro L115C disulfide bond mutant. In the CEST profiles, open circles (○) represent data points collected on the wild-type sample in the higher saturation field of 26 Hz, and asterisks (^∗) represent data points collected in the lower saturation field of 13 Hz. Solid lines for wild-type sample data were the best fits.

Fitting the data from 13 residues with R_ex-values larger than 6 s⁻¹ globally, we obtained populations of states I and N (p_I = 4.0 ± 0.2% and p_N = 96.0 ± 0.2%) and the total exchange rate (k_ex = 3257 ± 106 s⁻¹). ¹⁵N chemical shifts of state I determined from the data fitting are listed in Table S2. For the other residues with 3 < R_ex ≤ 6 s⁻¹, their chemical shifts in state I were obtained from data analysis by fixing k_ex, p_I, and p_N at the values derived from the global fitting, which are also listed in Table S2. The chemical shifts in state I were obtained from the chemical shift difference (Δν, both magnitude and sign) and the chemical shifts in state N (δ_I = δ_N + Δν). As shown in our previous work, the sign could be determined when |Δν| > 1 ppm for relatively fast exchange (k_ex ∼1500–3000 s⁻¹) by comparing the χ²-values for the fits with +|Δν| and −|Δν|. For instance, χ² = 31 when Δν = 1.57 ppm, whereas χ² = 91 when Δν = −1.57 ppm for Q64. Based on the Akaike information criterion, Δν = 1.57 ppm is ∼10⁺²⁰ times more probable than Δν = −1.57 ppm.

Because the exchange is close to the fast exchange regime, we examined whether the chemical shift difference (Δν) and population of the minor state (p) can be determined independently. From the result of 200 Monte Carlo simulations, we found that Δν and p distributed in a relatively narrow range without an obvious correlation (Fig. S3, a and b). When residue S71, which had the largest chemical shift difference (Δv = 6.6 ppm or 513 Hz), was excluded, the correlations become more obvious (Fig. S3, c and d). Therefore, the p- and Δν-values extracted from our data are reliable because of the presence of residues with large Δν-values.

Structure of state I

Past studies have not been able to obtain structural information of the open conformation because of its low population size (∼4% as shown here). In this study, we combined the fitting of both the CPMG and CEST data to obtain both the magnitudes and signs of the chemical shift differences between states N and I, by following the method demonstrated previously (40). From this, ¹⁵N chemical shifts of the minor state I were calculated and compared with those of both the major state N and unfolded (or disordered) state U (Table S2). In terms of ¹⁵N chemical shifts, which are different for folded and disordered proteins, the region of E63–N88 in state I is very similar to that in the disordered state but significantly different from that in the major native state N (Fig. 5). The result suggests that state I of NS2B adopts a partially unfolded structure; D50–E62 has a native conformation with a regular β-strand (D50–D58), E63–N88 exists in disordered conformations, and E89–R100 remains disordered. In fact, state I resembles the open conformation of NS2B-NS3pro shown in Fig. 1 b, but it contains no regular secondary structure elements in E63-R100. Note that the secondary structure elements in the region E63–R100 observed in the open structures very likely result from crystal contacts as they vary from one structure to another in the absence of inhibitors (Tables S1 and S3). Therefore, the conformational exchange between states N and I involves partial unfolding of NS2B, and the unfolding rate was estimated to be ∼130 s⁻¹ (k_ex × p_I).

Figure 5.

¹⁵N chemical shift comparisons between the minor state (I) of DENV2 NS2B complexed with NS3pro and its native state (N) (a) and between the state I and the unfolded state (U) (b).

Effect of mutation on the open conformation population

With the information obtained about the equilibrium between the closed and open conformations of the NS2B-NS3pro complex, we further investigated the effects of perturbing this equilibrium on the activity level of the protease. This was achieved through the generation of point mutants. First, a proline residue was introduced as a single point mutation on β4 of NS2B (M84P) to destabilize the structure and shift the equilibrium toward the open conformation. For the wild-type NS2B, the backbone ¹H-¹⁵N peaks were well dispersed without significant overlap in the HSQC spectrum (Fig. 2). For the M84P mutant, on the other hand, the correlations became much less dispersed (Fig. 6). This point mutation caused significant changes in peak positions of most residues in the region of E63–N88. Peaks with little chemical shift perturbations belong to the terminal regions (A49–E62 and E89–R100), including β1 (Fig. 6). The result suggests that the region of E63–N88 adopts a conformation different from the native conformation. Because of the poor ¹H chemical shift dispersion characteristic for unfolded protein, this region most likely adopts a disordered form, and the mutant mainly exists in an open conformation.

Figure 6.

(a) ¹H-¹⁵N HSQC spectrum of the M84P mutant (red), overlaid with the ¹H-¹⁵N HSQC spectrum of the wild-type DENV2 NS2B-NS3pro (black). Residues with significant chemical shift perturbations are labeled. (b) Combined chemical shift perturbation (Δδ) for each residue of NS2B upon the introduction of the M84P mutation, with the secondary structure of NS2B included within the chart. The region that undergoes conformational exchange in the wild-type is indicated by a red line, and the position of the mutation is indicated by a black triangle. The lower dotted line indicates the average of the chemical shift perturbations, and the higher dotted line indicates the average plus one standard deviation. To see this figure in color, go online.

To further examine the structural feature of the region of E66–N88 in the M84P mutant, a ¹⁵N HetNOE experiment was performed. For the mutant, the region of G69–M84 was similar to the disordered C-terminal region of N88–L98 with small HetNOE values (<0.25), but the region of D50–Q64 had large HetNOE values (>0.4) (Fig. 3). In comparison with the WT NS2B-NS3pro, the region of G69–M84 in the mutant had significantly smaller HetNOE values, and other regions remained nearly unchanged. Because the region of G69–R100 in the mutant is as flexible as the disordered region of 95–R100 in the WT protein and has poor ¹H_N chemical shift dispersion, the M84P mutant should be mainly disordered in the region of G69–R100 or exist mainly in an open conformation, consistent with the poor peak dispersion feature. The result indicates that destabilization of β4 by mutation (M84P) shifts the closed-open conformational equilibrium from a closed conformation to an open conformation.

Next, a pair of cysteine residues were introduced as point mutations at both I73 of NS2B and L115 of NS3pro. The cysteine pair is positioned such that the disulfide bond formed under oxidizing conditions will lock the C-terminal region of NS2B in place, shifting the equilibrium toward the closed conformation. The HSQC spectra of the WT protein and its disulfide bond mutant (NS2B I73C + NS3pro L115C) were very similar in peak dispersion and positions (Fig. S1 a). This mutation caused chemical shift perturbations on a few residues either sequentially or spatially near to the mutation point (Fig. S1 b). Additionally, peak intensities were quite uniform, and the peak of residue D81, which was missing in the HSQC spectrum of the wild-type, was visible for the disulfide bond mutant. The disulfide bond mutant had similar HetNOE values to the wild-type (Fig. 3). In fact, loop2 between β2 and β3 became more rigid upon the introduction of the disulfide bond. The results show that the disulfide bond mutant exists mainly in a structure of the state N of the wild-type protein, which is the closed conformation.

To examine the effect of the introduction of the disulfide on conformational exchange of the complex, RD and CEST experiments were performed. The disulfide bond mutant displayed no relaxation dispersion, even for the residues with large R_ex-values and far away from I73 in the WT protein (Fig. 4). In addition, the mutant displayed one dip for each ¹H-¹⁵N peak. The results show that there are no more conformational exchanges on millisecond and subsecond timescales or that the population of the open conformation is too small (<0.5%) to be detected for this mutant. The result also demonstrates that stabilization of the complex shifts the closed-open conformational equilibrium toward the closed conformation by the disulfide bond.

Effect of mutation on the activity of NS2B-NS3pro

The above results demonstrate that mutations can shift the dynamic equilibrium between the open and closed forms. To see how the change of this equilibrium affects the protease activity, activity assays were performed to compare the mutants with the wild-type. The M84P mutant had no activity, whereas the WT protein was active (Fig. 7; Table 1). The CD spectra of this mutant and the WT protein are very similar (Fig. S2), indicating that the loss of the activity is not caused by the misfolding of the NS3pro but by shifting the equilibrium from dominance of the closed NS2B conformation to dominance of the open conformation.

Figure 7.

Michaelis-Menten plot for the proline mutant M84P and the disulfide bond mutant NS2B I73C + NS3pro L115C (denoted as I73CL115C), with comparison to the wild-type (WT) DENV2 NS2B-NS3pro. To see this figure in color, go online.

Table 1.

Kinetic parameters of the mutants compared with the wild-type DENV2 NS2B-NS3pro

Construct	Vmax (nM/s)	Km (μM)	kcat (s−1)	kcat/Km (M−1 s−1)
WT	8.0 ± 0.2	104 ± 4	0.80 ± 0.02	7800 ± 300
M84P	ND	ND	ND	ND
NS2B I73C+ NS3pro L115C	8.3 ± 0.1	85 ± 2	0.83 ± 0.01	9800 ± 400

ND, data not available.

On the other hand, the disulfide bond mutant NS2B I73C + NS3pro L115C had a slight increase in enzymatic activity (Fig. 7; Table 1). As the population size of the closed conformation increased by only 4% by introduction of the disulfide bond, the enzymatic activity enhancement should be insignificant by stabilization of the wild-type NS2B-NS3pro complex.

Discussion

The existence of the open conformation of the NS2B-NS3pro complex has baffled researchers since its discovery in 2006 (18). It was initially thought that the dengue NS2B-NS3pro complex is predominantly in the open conformation, in contrast to the West Nile NS2B-NS3pro. This was due to the lack of a crystal structure showing dengue NS2B-NS3pro in the closed conformation. This has been changed when a closed conformation of dengue NS2B-NS3pro was obtained in the presence of a peptide inhibitor (20). The closed conformation of inhibitor-free Zika NS2B-NS3pro (26) and previous NMR studies (28,30,44,45) have provided evidence to support that the closed conformation is the dominant and enzymatically active form and therefore biologically relevant. Recent NMR studies have given an estimate of not more than 10% of the complex in the open conformation in the absence of inhibitor (30,44). Our results are in agreement with these studies, showing that the dengue NS2B-NS3pro complex is mainly in the closed conformation (p_N = 96%), which is the active form.

Having the open conformation as an alternative inactive form seems pointless, but for the open conformation to persist through the selective pressures of evolution, it must play a significant role in nature. To understand its role in the protease function, structural information of the open conformation is required. Previous attempts to study the open conformation have been unsuccessful because of its small population size, as well as severe NMR peak broadening and overlapping (28,30,44). In this study, we overcame these problems using two approaches. First, we expressed NS2B and NS3pro sequentially such that a complex between isotopically labeled NS2B and unlabeled NS3pro was obtained. This has reduced the number of NMR peaks on the HSQC spectrum from at least 200 to around 50, reducing the amount of peak overlapping. This allowed us to achieve ∼95% amide peak assignments for NS2B (Fig. 2). Next, we performed a combined fitting of both the CPMG and CEST data to a two-state model to obtain the chemical shifts of the open conformation (Table S2). These allowed us to obtain structural information of the open conformation, which revealed that the C-terminal region of NS2B is unfolded or disordered in the open conformation of NS2B-NS3pro (Fig. 4). The open conformation obtained here is in contrast to some of the crystal structures, which suggests the existence of the β-hairpin (residues 73–87) in the open conformation (21,22). The β-hairpin structure is likely to be an artifact that occurred during the crystallization process because of crystal contacts (Table S3). Better understanding of the functional role of the open conformation needs further studies on structures, dynamics, and protease activities of NS2B-NS3pro complexes from other dengue virus serotypes.

Our study has revealed through activity assays of various mutants that shifting the closed-open conformational equilibrium to either side causes an effect on the proteolytic activity of the protease (Fig. 7; Table 1). This shifting can occur either through natural means like pH and salt concentration (30), or artificial means like site-directed mutagenesis and small molecule inhibitors (46,47). The protease activity can therefore be controlled through this equilibrium by both nature and man. We speculate that the open conformation is used by the virus as an autoregulatory mechanism to tune down its protease activity when the environmental conditions are not favorable for its replication. The functional implication of this is that the open conformation can be exploited by man during drug design to inhibit the viral protease. Although the closed conformation is more biologically relevant, the open conformation may be the key to developing drugs with great efficacy and specificity to inhibit the dengue protease.

Conclusion

Our results show that the dengue NS2B-NS3pro complex is predominantly in the closed conformation and the open conformation is unfolded in region E63–K87 of NS2B. The shifting of the equilibrium between the open and closed conformations has also been shown to change the activity level of the protease. This closed-open conformational equilibrium is possibly used by the virus for self-regulation and at the same time can be exploited in future drug design to treat dengue.

Author contributions

D.Y. conceived the project. W.H.K.L., W.L., and J.-S.F. performed experiments and data analysis. W.H.K.L. and D.Y. wrote the manuscript with input from all authors.

Editor: Elizabeth Komives.