Self-assembly of dengue virus empty capsid-like particles in solution

Summary Nucleocapsid (NC) assembly is an essential step of the virus replication cycle. It ensures genome protection and transmission among hosts. Flaviviruses are human viruses for which envelope structure is well known, whereas no information on NC organization is available. Here we designed a dengue virus capsid protein (DENVC) mutant in which a highly positive spot conferred by arginine 85 in α4-helix was replaced by a cysteine residue, simultaneously removing the positive charge and restricting the intermolecular motion through the formation of a disulfide cross-link. We showed that the mutant self-assembles into capsid-like particles (CLP) in solution without nucleic acids. Using biophysical techniques, we investigated capsid assembly thermodynamics, showing that an efficient assembly is related to an increased DENVC stability due to α4/α4′ motion restriction. To our knowledge, this is the first time that flaviviruses’ empty capsid assembly is obtained in solution, revealing the R85C mutant as a powerful tool to understand the NC assembly mechanism.


INTRODUCTION
One of the main challenges that viruses face to be perpetuated in nature is to maintain their genome tightly protected while outside the host cell, but turn it promptly exposed after reaching the cellular environment where virus replication occurs. Capsid proteins are the viral components responsible for ensuring this task is properly performed. Capsid proteins of different icosahedral viruses contain, at their C-or N-terminal regions, positively charged arginine-rich stretches, whose net charge correlates with the size of the respective viral genome. 1 Interestingly, the number negative charges on viral nucleic acid exceeds the number of positive charges on the capsid proteins in a ratio of order 2:1, a phenomenon known as overcharging. 2,3 The fact that the capsid proteins' positive charges are nonuniformly distributed, being concentrated in the protein terminal regions, which are directed toward the interior of the virion, suggests that electrostatic interactions between the capsid protein and the viral nucleic acid would be the driving force for viral NC assembly and stability. In this context, of particular interest are the flaviviruses' capsid proteins, which, among all viruses' capsid proteins, show the highest net charge:molecular mass ratio. 4 The Flaviviridae family comprise important human pathogens, including dengue virus (DENV), Zika virus (ZIKV), West Nile virus (WNV), yellow fever virus (YFV), Japanese encephalitis virus (JEV) and Tick-born encephalitis virus (TBEV). DENV infection is a major public health problem worldwide, causing a broad spectrum of clinical manifestations, from mild symptoms to a severe hypovolemic shock that can be fatal, with 400 million people estimated to be infected every year. 5 DENV is an enveloped spherical particle of 50 nm diameter. In the viral envelope, two structural proteins associate with the lipid bilayer with icosahedral symmetry, the envelope (E) and the membrane (M) proteins. Inside the shell, the viral genome, a positive-sense single-stranded RNA molecule, is packaged by multiple copies of a single protein, the capsid (C) protein.
Several cryo-EM structures have been determined for mature and immature flaviviruses, including DENV, ZIKV, WNV, JEV, TBEV, and Spondweni virus (SPOV). [6][7][8][9][10][11][12][13][14] Although these studies provided structural details of the proteins on the virus surface, no information could be obtained for the NC structure of the mature particles, suggesting either a poorly ordered NC structure or a lack of symmetry between NC and heterotetramers formed by E and M proteins in the viral envelope. [13][14][15] DENVC forms homodimers in solution, with each polypeptide chain containing an N-terminal intrinsically disordered region (IDR) followed by 4 a-helices connected by short loops. 16,17 DENVC, as other flaviviruses' C proteins, shows unique structural features, such as the predominance of quaternary contacts between a2/a2 0 and a4/a4 0 maintaining the dimeric structure, flexible a1 helices, and a highly electropositive surface throughout the protein, 18 contrasting with the nonuniform charge distribution observed for most icosahedral viruses' capsid proteins. This, together with the fact that there is no information on DENVC orientation in the capsid, makes DENVC assembly an intriguing process.
Recent work by our group showed that the in vitro assembly of DENV nucleocapsid-like particles (NCLPs) requires a coordinated neutralization of C protein positive charges through interaction with either size-specific nucleic acids or negatively charged surfaces. 19 Based on structural analysis of the electropositive DENVC surface, we hypothesized that a highly positive spot conferred by arginine 85 (R85) and lysine 86 (K86) residues in DENVC a4-helix would be the first point neutralized by RNA binding, triggering NCLP assembly.
To validate this hypothesis, we designed a mutant in which DENVC R85 was replaced by a cysteine residue (R85C mutant). The rationale behind this mutation is that it would vanish the a4/a4 0 positive spot (Figure 1), as well as restrict the a4/a4 0 intermolecular motion due to the formation of a disulfide cross-link between a4 and a4'. Here we show by dynamic light scattering and transmission electron microscopy imaging that DENVC R85C self-assembles into capsid-like particles (CLPs) in solution, in the absence of nucleic acids, corroborating our hypothesis. Additionally, to gain insights into the thermodynamics of the DENVC/CLP dissociation/denaturation process, we performed differential scanning calorimetry, circular dichroism, and fluorescence spectroscopy experiments. We found that limiting a4/a4 0 motion increases DENVC stability, which could be associated with a more efficient capsid assembly. In summary, this is the first study establishing an efficient in vitro self-assembly of DENVC in solution in the absence of nucleic acids, revealing the R85C mutant as a powerful tool to understand the DENV NC assembly mechanism.

DENVC R85C mutant
R85C is a DENVC mutant in which the arginine at position 85 in a4-helix was replaced by a cysteine residue (Figure 2A), removing the highly positive spot in the center region of the a4/a4 0 helices' (see Figure 1). Additionally, the mutation allows the introduction of an interchain disulfide bond, linking covalently DENVC monomers ( Figure 2B) and restricting the a4/a4 0 helices' motion. To ensure the formation of the correct disulfide bond in the mutant protein, two steps during the protein purification protocol were necessary: (i) obtaining the purified protein as a non-covalent dimer by using the reducing agent dithiothreitol Ribbon structure and electrostatic potential surface of DENVC wild-type (WT) and R85C mutant proteins in two different orientations. The structures were generated with PyMol using mutation insertion in PDB ID 1R6R, with red sticks highlighting the disulfide bond. The electrostatic potential values were calculated in APBS software with the protonation states and charge values determined by the PDB2PQR server along with the PROPKA program (pH 7.0, 200 mM NaCl, 25 C). The electrostatic potential ranges from À10 (red) to +10 kT (blue). Note that the highly positive spot in a4/a4 0 is vanished in R85C mutant. iScience Article (DTT) in all steps of the purification protocol; and (ii) oxidizing the reduced protein with diamide to promote the disulfide bond formation (R85C ox ). The analysis of the protein sample obtained after the first purification step in a non-reducing SDS-PAGE showed the major presence of the reduced R85C ($12 kDa) and traces of the oxidized protein, which runs with the molecular mass of the dimer ($25 kDa) ( Figure 2C). After incubation with 3 mM diamide, only the band corresponding to R85C ox was observed, indicating that all the protein was converted into covalently linked dimers ( Figure 2D). To ensure the efficiency of the covalent dimer formation, we incubated R85C ox with DTNB to evaluate the presence of remaining free thiol groups by the appearance of TNB À2 signal at 412 nm. 20 The absence of a 412 nm band confirmed the complete formation of the disulfide bonds (data not shown). The introduction of the disulfide bond did not alter DENVC secondary structure profile, as shown by the comparison of the CD spectra of WT and R85C proteins ( Figure 2E). To return the protein to its reduced form, we titrated R85C ox with DTT, establishing that the incubation of the protein with 15 mM DTT is the best condition to obtain the fully reduced R85C (R85C red ; Figure 2F).

R85C assembles in large oligomeric particles in solution
To evaluate whether R85C mutant spontaneously self-assembles into capsids, we performed DLS experiments ( Figure 3). The size distribution profile obtained for DENVC WT, R85C ox, and R85C red showed a well-defined population with apparent hydrodynamic diameters of 7.0 G 0.1 nm, 6.4 G 0.2 nm, and 7.0 G 0.1 nm, respectively. The slightly more compact size observed for R85C ox could be explained by the motion restriction caused by the disulfide bond. Theoretical calculation of DENVC iScience Article hydrodynamic diameter, using the software HYDROPRO 21 and considering a spherical shape for the protein, gives a value of 5.1 nm. The difference of $ 1-2 nm between the experimental and the calculated values may be explained by the high protein concentration used in the experiments (200 mM). At this condition, the light scattered by one particle would itself be scattered by adjacent particles, increasing the sample apparent size, a phenomenon known as multiple scattering. 22 The high protein concentration also increases the probability of transient particle-particle interactions, affecting the sample diffusion coefficient and resulting in an increased particle apparent size. Additionally, the contribution of the high mobility of the intrinsically disordered N-terminal tails may not be ruled out.
For the WT protein, in addition to the dimer, we also observed a poorly defined population with hydrodynamic diameters ranging from $100 to $1500 nm, suggesting the presence of high molecular weight aggregates ( Figure 3A). In contrast, for R85C ox , a second well-defined population with an apparent hydrodynamic diameter of 60 G 2 nm was observed ( Figure 3B). Different from the heterogeneous aggregates observed for the WT protein, the distribution profile of this population was very reproducible, supporting our hypothesis that the mutant self-assembles in regular structures in vitro. A hydrodynamic diameter of $60 nm is larger than that expected for the viral NC (although flaviviruses' NC could never be observed, the viral internal area is $33 nm; see Figure S1). However, considering that these experiments were performed in high protein concentration and that the multiple scattering phenomena is expected to be stronger for larger particles, it is plausible to hypothesize that R85C mutant forms CLPs in solution. For R85C red , the second population (hydrodynamic diameter of 80 G 3 nm) suggests that similar oligomeric structures are formed by the reduced proteins ( Figure 3C). Additionally, a third population with a hydrodynamic diameter of 1459 G 169 nm was also seen. It is interesting to note that the formation of R85C red capsids suggests that neutralizing the R85 charge is sufficient to drive capsid self-assembly.
To further investigate the conditions necessary for R85C assembly, we performed the DLS analysis of protein samples after each of the two chromatography steps of the purification protocol ( Figure S2). In the first step, Escherichia coli lysate is applied in a HiTrap Heparin HP column and the protein is purified using a step gradient with increasing NaCl concentrations. R85C is eluted from the column predominantly as a non-covalent dimer (see SDS-PAGE analysis in Figure 2C). The DLS profile obtained for this sample (Figure S2A) was very similar to that of the complete oxidized sample, indicating that either oxidized or reduced proteins were self-assembled into CLPs. This agrees with the result presented in Figure 3C, which shows that when the oxidized sample is completely reduced by incubation with 15 mM DTT (as shown in Figure 2F), the CLP population in the DLS profile is similar to that observed for the oxidized sample. Thus, the proportion of oxidized or reduced dimers does not seem to be a determinant for capsid formation. It is important to note that the sample analyzed in Figure S2A contains 1.5 M NaCl, but even in this high salt concentration, CLPs are formed. To evaluate whether the fully oxidized R85C also forms CLPs in the presence of 1.5 M NaCl, we performed a DLS analysis of the sample after the second affinity chromatography step of the purification protocol. At this step, the protein has already been completely oxidized by diamide treatment, but the salt necessary for protein elution has not been removed yet from the sample ( Figure S2B). The result showed that, as for the reduced protein, R85C ox also self-assembles in the presence of high salt concentrations, suggesting that DENV capsid structure is not exclusively maintained by electrostatic interactions.

R85C self-assembles into CLPs
To obtain additional information on R85C self-assembled oligomers, we carried out TEM imaging experiments. For R85C ox samples, we observed well-defined structures, compatible with the size and shape expected for DENV NC, with an average Feret's diameter of 26 G 14 nm ( Figure 4A). This result further supports our hypothesis that the neutralization of R85 positive charge in the DENVC due to RNA binding triggers DENVC NC assembly. It is noteworthy that, in contrast to the DLS size distribution profiles, particles' size histograms obtained in the analysis of TEM images did not result in normal distributions. The DLS data likely reflect the behavior of an ensemble in solution, for which a Gaussian distribution is expected. On the other hand, since for TEM, the particles are adsorbed on the grid surface, the observed image is a sample snapshot. However, despite these differences, the particle sizes sampled by both methods are of the same magnitude and comparable, considering that DLS size measurements are affected by the high protein concentration used in the experiments, which favors the multiple scattering and transient particle interaction (see previous section). iScience Article R85C red also formed CLPs, but the observed particles were on average smaller than those seen for R85C ox , with an average Feret's diameter of 18 G 5 nm ( Figure 4B). This behavior is different from that observed in DLS experiments, in which slightly larger diameters were found for R85C red particles ( Figures 3B and 3C). These differences would be explained by the surface properties of the particles generated by oxidized or reduced R85C mutant, which would influence particle adsorption on the grids. Although this is an interesting hypothesis, further experiments are necessary to conclusively explain these results. Additionally, differently from the observed for R85C ox , heterogeneous filamentous structures were also seen for the reduced protein ( Figure 4B). It is known that microscopy techniques favor surface-driven oligomerization, which indeed occurs with DENVC. 19 Since DLS results indicate that these filaments occur in negligible amounts in solution, we would speculate that surface-driven assembly of filament-type structures would be favored for the reduced protein. Whether these structures play a role in the virus life cycle requires further investigation.

Thermodynamics and structural stability of DENVC WT and R85C mutant
The R85C mutant was designed both to remove the positively charged site and to restrict the intermolecular motion in DENVC a4/a4 0 helices. DLS and TEM results strongly support that the neutralization of R85 is crucial for capsid assembly since structures compatible in size and shape with DENV capsids were formed either with R85C ox or with R85C red . Nonetheless, when the cysteine residues are not covalently linked (R85C red ), assembly seems to be less efficient, with the formation of filamentous structures, possibly due to less coordinated protein interactions caused by a4/a4 0 helices' dynamics. Thus, to investigate the effects of restricting a4/a4 0 helices motion on DENVC oligomerization reaction, we performed thermal and chemical denaturation experiments aiming to characterize the thermodynamics of the dissociation/denaturation processes of DENVC WT, R85C ox , and R85C red , and the respective CLPs.

Thermal denaturation studies
To analyze the structural stability of DENVC WT and mutant proteins, we used circular dichroism spectroscopy (CD) to compare their thermal dissociation/denaturation processes ( Figure 5). For the interpretation of the results, it is important to bear in mind that the flaviviruses' C proteins form intertwined homodimers stabilized predominantly by quaternary contacts, making it very unlike the existence of a folded monomer in solution. 18 Thus, the DENVC dimer dissociation process cannot be analyzed separately from monomer denaturation, so we should interpret the data as resulting from the transition from a folded dimer to an unfolded/disordered monomer or dimer.
As expected, at room temperature, WT, R85C ox , and R85C red showed a typical CD spectral pattern of a-helical proteins, with negative bands at 208 and 222 nm (Figures 5A, 5C, and 5E, respectively). Thermal dissociation/denaturation profiles for each protein were analyzed by following the decrease in ellipticity at 222 nm (i.e., decrease in a-helix content) from 298 to 368 K (25-95 C). The denaturation curves obtained for WT and R85C red proteins were very similar, reaching a disordered state with remaining residual a-helix content, as observed at 368 K ( Figures 5A and 5E). Additionally, the curves showed a sigmoidal profile typical of a cooperative process (Figures 5B and 5F -filled circles). On the order hand, for the R85C ox , a non-cooperative process ( Figure 5D) that did not reach the same degree of disorder as observed for DENVC-WT and the R85C red ( Figure 5C) was observed, indicating that the immobilization of a4/a4 0 helices stabilize DENVC. We decided not to calculate the thermodynamic parameters from these data for the following reasons: (i) complete protein unfolding was not reached, (ii) the unfolding process was not completely reversible, and (iii) the samples did not present the same initial conformational states (dimer for WT and a mixture of dimer and CLPs for the R85C). Although we did not observe a complete refolding for any of the proteins, the reversibility was improved upon dilution ( Figure S3).
To gain more insight into the energetics of the dissociation/denaturation processes of DENVC WT and mutant proteins, we performed DSC experiments. The thermograms obtained for DENVC WT, R85C ox , and R85C red are shown in Figures 6A, 6B, and 6C, respectively. Thermogram deconvolution of R85C mutant was carried out to generate the Gaussian fits from which the T m , the calorimetric enthalpy variation (DH cal ), and the Van't Hoff enthalpy variation (DH VH Þ were obtained for each transition ( Figure S4).
The thermogram obtained for WT protein showed a single transition ( Figure 6A), which can be interpreted as the denaturation of the folded dimer (DÞ to a monomer (MÞ.  Figure 6B). Curiously, for R85C red this transition occurred at a lower temperature ( T m = 337.19 G 0.01 K), which would be explained by a direct effect of the R85 substitution on protein stability ( Figure 6C). The DH VH =DH cal of $1 for WT and R85C red proteins (Table 1) indicates that the transition from (DÞ to (MÞ obeys a two-state equilibrium, which is not valid for R85C ox . Consistent with DLS and TEM results, which showed the formation of CLPs by the mutant proteins, R85C ox , and R85C red thermograms presented an additional transition at lower temperatures ( Figures 6B and 6C), probably corresponding to CLP dissociation to folded species since no changes in the protein structures were observed at the respective temperature range (see Figures 5D and 5F). The comparison of the T m of these transitions (319.2 G 0.1 K for R85C ox and 325.0 G 0.9 K for R85C red ) suggests that the stabilization of the dimer through the immobilization of the a4/a4 0 helices is an important step in capsid assembly.  Figure 6B). The helix content variation within this temperature range (from 315 to 340 K; see Figure 5D) suggests that the protein is still folded, so we represented this transition as the equilibrium between two folded states, an unknown state (Un), which could be a dimer of a capsomer iScience Article that is in equilibrium with the disordered dimer (D d ). Additionally, it is interesting to note that this extra transition is compatible with the non-cooperative denaturation process observed R85C ox (see Figure 5D).

Chemical denaturation studies
Since it was not possible to completely denature R85C ox using temperature, we performed a chemical denaturation assay using guanidine hydrochloride (Gd:HCl) within concentrations ranging from 0 to 7.4 M (Figure 7). Denaturation was monitored by measuring the redshift in the center of mass of proteins' intrinsic fluorescence emission spectra. In agreement with the thermal denaturation results, WT and R85C red chemical denaturation profiles were very similar, both showing sigmoidal curves, confirming the cooperative nature of the process. The Gd:HCl concentrations that resulted in 50% denaturation were 4 and 4.2 M for WT and R85C red , respectively. R85C ox showed a different behavior, displaying two transitions, the first occurring between 0 and 3 M Gd:HCl and the second between 3 and 7.4 M. Furthermore, a considerable increase in stability was observed for the R85C ox , with the denaturation midpoint at 5.6 M Gd:HCl, confirming that the immobilization of a4/a4 0 helices plays an important role in DENVC stabilization.

DISCUSSION
Flaviviruses' NC formation is an intriguing issue. Although the steps of virus morphogenesis and maturation involving E and prM/M proteins are relatively well understood, there is no information on how the C protein recognizes and encapsidate the viral genome as well as on how NC is structured into de virion. The currently available cryo-EM structures for several flaviviruses showed details on the structural organization of E and M proteins on the virus surface. [13][14][15][23][24][25][26] For mature particles, a smooth icosahedral outer surface is formed by 90 E protein dimers (Figures S1A and S1B). M proteins, which are mostly transmembrane, associate with the transmembrane helices of E proteins. For all mature cryo-EM structures obtained, no densities were observed between the inner membrane layer and the virus (Figures S1A and S1B). Interestingly, for immature particles of ZIKV, a residual density was observed in contact with the transmembrane domains of the E and M proteins ( Figure S1C), but even in this case, it was not possible to obtain conclusive information on NC structural organization.
To contribute to the understanding of flaviviruses' NC assembly process, we designed a DENVC mutant that self-assembles into regular empty CLPs. To our knowledge, this is the first time that flaviviruses CLP assembly is obtained in solution in the absence of any nucleic acid. Our strategy was to mimic in the mutant (DENVC R85C) two possible effects of RNA binding to DENVC: specific charge neutralization and protein immobilization. This was achieved by (i) removing the positively charged spot located in the central region of the a4/a4 0 helices; and (ii) restricting helices' motion by introducing an interchain a4/a4 0 covalent bond. TEM imaging showed that the average diameter of R85C CLPs is quite consistent with the internal area of the virion shown in the flaviviruses' cryo-EM structures ( Figure S1), supporting the hypothesis that they are structurally similar to the actual DENV NCs.
The fact that both the oxidized (R85C ox ) and reduced R85C (R85C red ) DENVC mutants assemble into CLPs suggests that charge neutralization is the major trigger DENV NC formation. However, as shown by both DLS and TEM, R85C red formed large aggregates not observed for R85C ox , suggesting that the motion iScience Article restriction of a4/a4 0 makes the capsid assembly process more efficient. Our recent observation that DENVC-WT formed empty CLPs when the protein was incubated over negatively charged surfaces 19 reinforces the importance of restricting local protein mobility for NC assembly. Here we showed using different approaches that the immobilization of a4/a4 0 helices plays an important role in DENVC dimer stabilization. Thermal denaturation experiments revealed an increased T m for R85C ox dimer denaturation when compared to the reduced form or the WT protein (347.9 G 0.1 K for R85C ox compared to 337.19 G 0.01 K and 344.1 G 0.2 for R85C red and WT proteins, respectively). The same was found in chemical denaturation experiments (denaturation midpoint at 5.6 M Gd:HCl for R85C ox compared to 4.0 M and 4.2 M Gd:HCl for WT and R85C red , respectively). Altogether, these results demonstrate that DENVC a4/a4 0 helices' immobilization increases the dimer stability, which would be a role played by the viral RNA during the assembly process. However, based on theoretical studies on viral capsid assembly, 27 we can also speculate that the interchain covalent bond would change the capsid proteins' stiffness and thus increase the capsid stability favoring the assembly process. To further address this point, we used JPRED and PONDR VSL2 servers to simulate the tendencies for secondary structure and order/disorder along with WT and mutant DENVC sequences ( Figure 8). PONDR VSL2 predictions were consistent with the known ordered regions from the structure of the dimer. 17 Interestingly, VSL2 prediction suggested an increased tendency of disorder at a4/a4 0 when compared to the a2/a2 0 hydrophobic cleft, which showed the highest predicted order. Accordingly, JPRED secondary structure prediction showed lower confidences at its N-terminal portion, while the highest confidence was observed at its central region, where the conserved signature F++-h (FRKEI for DENVC-WT) is located. 18 Remarkably, when PONDR VSL2 prediction was applied to R85C mutant, a significant increase in a4/a4 0 predicted order was found. Thus, one can speculate that, during the virus replication cycle, the relative mobility in a4/a4 0 would be essential for C protein interaction with the viral genome, which increases helices' order and stability, triggering NC assembly and preventing the formation of empty capsids or defective virus-like particles. The intrinsic motion of the capsid proteins may also be present in the assembled virus particle and may contribute to the absence of electron density related to NC in the cryo-EM reconstructed structures. 6,13,14 Curiously, a PONDR VSL2 prediction behavior similar to that found for R85C was observed for the DENVC R85A/K86A double mutant previously characterized by Teoh and cols, 28 suggesting that the removal of the a4/a4 0 positively charged spot itself would be sufficient to restrict a4/a4 0 mobility.
Charge neutralization is thought to be the major trigger of NC assembly of RNA viruses. In contrast to the double-stranded DNA viruses, which assemble their capsids before genome packaging, single-stranded RNA viruses generally package their genomes simultaneously to capsid assembly. 29 Indeed, either for flaviviruses or for alphaviruses, which are also small enveloped +ssRNA viruses, attempts for reproducing NC assembly in vitro required capsid protein charge neutralization. 28,30-32 Thus, it is expected that genomic iScience Article RNA elements, namely negative charges, and long length, drive the orientation of capsid proteins to allow the protein-protein interaction necessary to build the capsid. Our TEM results showed an average diameter size of $30 nm for the CLPs formed by the R85C ox mutant. Considering a DENVC dimer area of 15 nm 2 (estimated from the protein structure; 1R6R), we would speculate that approximately 180 dimers are necessary to stabilize the capsid by C protein-C protein interactions.
Flaviviruses' NC assembly seems to be a very coordinated process, making it difficult to reproduce NC formation in vitro, as reflected by the few examples described in the literature. For DENVC and TBEVC, NCLPs could be observed by TEM after protein incubation with single-stranded nucleic acids. 28,33,34 The particles were more homogeneous in size and shape for larger nucleic acids, especially the full-length viral RNA, while no particles were obtained in the absence of nucleic acid. Recently, we further characterized DENVC NCLP in vitro assembly. 19 We showed a stoichiometry of 1:1 (DENVC:oligonucleotide) molar ratio for 5-mer or 25-mer oligonucleotides, and that DENVC empty capsids could be formed in a surface-driven assembly process. Additionally, we hypothesized that DENV NC formation would require the neutralization of a specific positively charged spot in a4-helix (R85 and K86 residues) instead of an unspecific electrostatic effect on the entire a4/a4 0 protein face. 19 Here we confirmed this hypothesis by showing that the removal of the positive charge at position 85 in R85C DENVC mutant is to be the major trigger driving to empty capsid assembly since the reduced protein (R85C red ) also assembled as empty capsids. Interestingly, a DENV infectious clone expressing the DENVC double mutant R85A/K86A produced a functional C protein and generated infectious virus particles, although with decreased infectivity. 28 This supports the hypothesis that the removal of the positively charged spot in DENVC a4/a4 0 helices doesn't affect the overall protein structure but increases the probability of the generation of viral defective particles.
The assembly process of alphaviruses' NC is much better understood than that of flaviviruses. In contrast to the observed for the flaviviruses, cryo-EM structures of the alphaviruses clearly showed icosahedral capsids in which the C-terminal chymotrypsin-like domain of the capsid proteins (known as core protein, CP) interacts with neighboring CPs and with the envelope E2 proteins. [35][36][37][38][39][40][41][42][43][44] The disordered CP N-terminal positively charged domain (about 120 residues, for which $30 are positively charged) is thought to interact with the viral genome in the virion core. 45 Alphaviruses' CP assembles into core-like particles (CLPs) in vitro through the interaction with different polyanions, such as single-stranded DNA or RNA oligonucleotides of at least 14 bases in length, heparan sulfate, or PEG-coated nanoparticles. 46 Remarkably, mutations in Ross River Virus (RRV) CP disordered N-terminal domain that replaced 4 lysine residues (K104-107) by aspartic residues (4D mutant) led CP to assemble in empty capsids, 31 confirming that CP N-terminal charge neutralization is the trigger for NC assembly. Additionally, when an anionic RRV CP mutant, in which all the N-terminal positively charged residues were replaced by negatively charged ones, was mixed with the WT protein, iScience Article they assembled into CLPs, 32 further supporting the CP N-terminal charge neutralization mechanism for alphaviruses NC assembly.
Like alphaviruses' CP, flaviviruses' C proteins also contain a disordered positively charged N-terminal region. Nevertheless, this region has never been considered the main site for viral RNA interaction during NC assembly. This is possibly a consequence of the excessive emphasis given to the asymmetric charge distribution model for flaviviruses C protein, which considers that the positive charges are concentrated in the a4 helix despite the higher proportion of positively charged residues in the N-terminal region (8 of 24 residues). Interestingly, a TBEVC mutant truncated at the N-terminal end (TBEV-D16-C) did not form NCLPs under the same conditions that other flaviviruses' C proteins, such as full-length DENVC, assemble. 47 Thus, although not considered so far, a different orientation of flaviviruses' C proteins in the virion in which the N-terminal region interacts with the viral RNA in the central core cannot be ruled out.

Limitations of the study
We were successful in determining the formation of empty CLPs, but we do not have information on the dimer/CLP equilibrium so far. DLS informed on the coexistence of CLP and dimer of R85C ox but this technique does not allow to determine the concentrations of each species. Thus, further studies are necessary to quantify the CLP/dimer equilibrium for the mutant without nucleic acid and the wild-type DENVC in the presence of nucleic acids, 19 which will contribute to the understanding of capsid dynamics in the context of the virus particle.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Recombinant proteins Expression and purification
The coding sequences of DENVC (serotype 2; residues 1-100) and its mutant R85C were cloned into pET3a by GenScript (Piscataway, NJ, EUA) and transformed into E. coli BL21-DE3-pLysS. For cultivation, the cells were streaked out on LB plates and incubated overnight at 37 C. Single colonies were picked and inoculated into a M9 preculture (20 mL in 100-ml flasks) and cultivated for 5 h to be subsequently diluted into a second preculture. The second preculture was used to inoculate the main culture (1L), which was cultivated at 37 C, at 200 rpm. Growth was monitored by measuring the optical density at 600 nm (OD 600 ) and recombinant protein expression was induced with 0.5 mM isopropyl-D-1-thiogalactopyranoside (IPTG), overnight, at 30 C. The cells were then centrifuged ($ 5,000 g for 30 min at 4 C) and the pellets were resuspended in lysis buffer consisting of 25 mM HEPES, pH 7.4, containing 0.2 M NaCl, 1 mM EDTA, glycerol 5% (v/v) and protease inhibitor cocktail (P8465, Sigma-Aldrich). The cells were disrupted by ultrasonication. After this step, the lysate was incubated with NaCl at a final concentration of 2 M and left on agitation for 60 min, at 4 C. The lysate was ultracentrifuged at 70,400 g for 50 min at 4 C. The supernatant was applied onto a HiTrap Heparin HP column and DENVC was purified using a step gradient with an increasing NaCl concentration (0.5 -2 M). Fractions containing DENVC protein were confirmed by 18% SDS-PAGE gel, concentrated, and stored at -20 C. To prevent nonspecific disulfide bond formation during R85C expression, 1 mM DTT was added to all buffer solutions used.

Preparation of the covalent-linked dimer of R85C (R85C ox )
To generate a disulfide bond linking R85C subunits, the protein was incubated with 3 mM diamide at 4 C, with agitation, for 40 min, and applied in a HiTrap Heparin HP column coupled to an AKTA Start instrument (GE Healthcare). The protein was eluted from the column with a NaCl gradient (0.5 to 2 M) in 25 mM HEPES buffer (pH 7.4), containing 1 mM EDTA and 5% (v/v) glycerol, without DTT, with a flow of 5 ml/min. Purified R85C ox was concentrated using an Amiconâ centrifuge filter of 10 kDa cut-off (Merck-Millipore, USA) at 6,0003g, 4 C, in 55 mM NaH 2 PO 4 buffer (pH 7.4), 200 mM NaCl, 2 mM PMSF, 5 mM EDTA, and 5 mM azide, and stored at À20 C. Protein concentrations were determined spectrophotometrically at 280 nm using the extinction coefficient of 11,000 M -1 $cm -1 .

Confirmation of disulfide bond formation
The formation of the disulfide bond covalently linking the mutant monomers (R85C ox ) was confirmed by the observation of a $ 25 kDa band (dimer molecular mass) in a non-reducing 18% SDS-PAGE. Additionally, to quantify the remaining free sulfhydryl groups in the sample, 10 mM R85C ox in 100 mM Tris-HCl buffer (pH 7.4), was incubated with 5,5'-dithiobis (2-nitrobenzoic acid) (DTNB) in a final concentration of 10 mM. 20 The presence of free sulfhydryl groups was determined by measuring the absorbance at 412 nm using a BioMATE 3S UV-Vis spectrophotometer (Thermo Scientific, USA).
Preparation of the reduced (non-covalent dimer) R85C (R85C red ) where q being the measured ellipticity at wavelength l (deg), c is the protein concentration (mol/L), l the path length of the cuvette (in cm), and n the number of amino acid residues in the protein.

Differential scanning calorimetry (DSC)
DSC experiments were performed using an N-DSC III apparatus (TA Instruments, USA) in the range from 20 to 95 C with scan rates of 1.0 C/min. Initially, both calorimeter cells were loaded with the buffer solution, equilibrated at 20 C for 10 min, and scanned repeatedly until the baseline was reproducible and stable. Then, the sample cell was loaded with the DENVC WT, R85C ox or R85C red at final concentrations of 42.5 mM in 55 mM NaH 2 PO 4 buffer (pH 7.4) containing 200 mM NaCl. The baseline correction was obtained by subtracting the buffer scan from the protein scan using the software Launch NanoAnalyze, supplied by TA Instruments. The thermograms were fitted using a two-state model with the program Origin 7.0. The Van't Hoff equation was used to calculate the DH VH involved in the transition: where DH VH is the enthalpy change, R is the universal gas constant, and K u is the denaturation constant at the correspondent temperature, T. iScience Article Calorimetric enthalpy change (DH cal Þ was calculated from the thermogram area. Gaussian fits and the deconvolution analysis were performed using the program Origin 7.0 by adjusting a minimum number of Gaussian curves to the DSC thermograms, and Gaussian curves were treated as described above. Chemical denaturation studies Fluorescence spectroscopy Secondary structure and order/disorder prediction Secondary structure prediction was performed using JPRED4 server (https://www.compbio.dundee.ac.uk/ jpred/). JPRED4/JNET assigns a confidence score varying from 0 (low) to 9 (high). JNET uses a neural network that includes learning and training algorithms based on the increasing number of known protein secondary structures. The confidence level of a given prediction gives information on the tendency of a given residue to be in a secondary structure. 48 Order/disorder content in protein structure was predicted using PONDR server (http://www.pondr.com/). PONDR uses neural networks than can be trained using different inputs. VSL2 is an PONDR predictor trained by combining two predictors, for long sequences (> 40 residues) and short 8 to 9 residues sequences, 49 using not only crystallographic structures but also NMR structures and circular dichroism secondary structure data. VSL2 showed the best match to experimentally determined structural and dynamical features of DENVC. 17,52,53 QUANTIFICATION AND STATISTICAL ANALYSIS Quantification, data analysis, and plotted graphs were performed using Origin 7.0. For transmission electron microscopy images, n = 598 is the number of particles analyzed for R85C ox and, n = 177 is the number of particles analyzed for R85C red . The mean Feret's diameters values were used to calculate the standard deviation (SD) between the diameter measurements.
The electrostatic potential values were calculated in APBS software with the protonation states and charge values determined by the PDB2PQR server along with PROPKA program (pH 7.0, 200 mM NaCl, 25 C). 54 ll OPEN ACCESS