Structural Changes in Dengue Virus When Exposed to a Temperature of 37°C

Virus sample preparation for cryo-EM studies.

The methods for virus production and purification are similar to those described elsewhere (8). Briefly, the DENV2 NGC strain was grown in C6/36 mosquito cells. The cells were infected at a multiplicity of infection of 0.1 at 28°C under 5% CO₂. After 2 h of infection, the inoculant was replaced with fresh RPMI medium supplemented with 2% fetal bovine serum (FBS) and was further incubated for 4 days. The virus-containing supernatant was clarified from cell debris by centrifugation, and virus particles were collected by precipitation with 8% polyethylene glycol 8000 (PEG 8000) in NTE buffer (10 mM Tris-HCl [pH 8.0], 120 mM NaCl, 1 mM EDTA). Virus particles were resuspended and then purified by centrifugation using a 30% sucrose cushion and then a 10 to 30% potassium tartrate gradient. The virus band was collected and concentrated, and buffer was exchanged into NTE buffer using an Amicon Ultra-4 centrifugal concentrator (Millipore) with a 100-kDa molecular-mass cutoff filter. The virus preparation contained little or no prM, indicating low levels of contamination by immature virus, as determined by Coomassie blue-stained SDS-PAGE. To estimate the envelope protein concentration, the corresponding band was compared with bovine serum albumin (BSA) at different concentrations.

Cryo-EM image collection.

The virus sample was grown in C6/36 cells at 28°C and kept at 4°C throughout the purification steps. The sample was incubated for 30 min at 37°C, followed by another 2 h of incubation at 4°C prior to freezing. The step involving 30 min of incubation at 37°C was omitted for the control. A virus sample of 2.5 µl was applied onto ultrathin carbon-coated lacey carbon grids (Ted Pella), and blotted with filter paper for ∼2.5 s before being plunged into liquid ethane using the FEI Vitrobot Mark IV. The frozen grid was stored at liquid nitrogen temperature. The virus particles were imaged using a 300-kV FEI Titan Krios electron microscope at a nominal magnification of 47,000 with an electron dose of 18.5 e/Ų. The images were captured on a charge-coupled device (CCD) camera (giving a pixel size of 1.9 Å) at an underfocus range of 1.5 to 3 μm. Images showing drift and strong astigmatism were discarded, and 302 images were selected for further processing.

Cryo-EM reconstruction.

The sample incubated at 37°C had mostly particles with a rough surface, and there were only a few smooth particles (Fig. 1). Reconstruction of the particles was initiated by picking the rough-looking unbroken particles using the e2boxer tool in the EMAN2 software package (13). In total, 15,111 particles were selected. Contrast transfer function (CTF) parameters were determined using EMAN2. The orientation search was done by using the multipath simulated annealing (MPSA) procedure (14). The selected virus particles were initially treated as a homogenous population, and thus, only one search model was used: a three-dimensional (3D) map of the mature DENV (15). Following the orientation search, a 3D reconstruction was done with make3d from EMAN (16) with the result used as the reference model for the next cycle. Default parameters were used in the orientation search using the MPSA program, with the exception of the numbers of reference projections and the data resolution range in the Fourier space. Three reference projections and a data resolution range of 53 to 160 Å were used for the first 3 cycles; once the model was improving, we increased the number of projection references as well as the resolution range.

The 3D reconstruction of the rough-looking virus particles, by using the known mature dengue virus structure as the initial model (15), resulted in a map showing a smooth surface at a resolution of about 30 Å (result not shown). This map showed that the particles expanded from ∼240 Å to ∼ 280 Å with a small protrusion corresponding to the position of the dimer near the icosahedral 2-fold vertex. Since it was a very-low-resolution map, we had to check for the correctness of this reconstruction. Two separate tests were done. In the first test, we did the reconstruction using another starting model (i.e., a poor model obtained from the program starticos in the EMAN suite). A similar cryo-EM map was obtained, indicating the reliability of the previous result. In the second test, the data set was split into two halves and reconstructions were done separately using the previously published mature DENV map for the initial models. The reconstruction of the first half gave a similar map as before. However, the other half resulted in a map that was very similar to the mature virus model. This result suggested that the particle population is not homogeneous and contained some unexpanded mature virus and, possibly, some other particle classes.

To improve the reconstruction, we decided to sort particles into classes and then use the 3D maps generated from these classes as initial models for reconstructions. Using the K-mean classification method implemented in EMAN2, the particle images were sorted by performing iterative reference-free classification to get a set of representative class averages. There were three visibly different classes of particles: particles with rough edges (called class II), particles with relatively smooth edges (class III), and particles with apparently smaller diameters (Fig. 2) (class IV). The initial models were generated for each class using e2initialmodel from EMAN2. Two separate 3D reconstructions were done. For the first reconstruction, 1,584 particle images corresponding to class IV were extracted from the initial particle set and the reconstruction was done by using the initial model created from this particle class. The particles from classes II and III were combined (total of 13,527 particles), and an orientation search was done using four starting models simultaneously. The four models were the published mature dengue virus structure (named class I) (15), two initial models generated from the class II and III particles, and immature DENV models (7). The projections were compared to all four models and grouped into the model class that had the highest correlation. Even though the control sample (virus not exposed to a temperature of 37°C) indicated only low levels of immature virus in the virus preparation, we included the immature DENV model in the multimodel orientation search to eliminate possible interference by the incidentally picked immature particles. In the final cycles of reconstruction of class IV, 1,327 particles were selected for the reconstruction, resulting in an ∼23-Å resolution map (as measured using Fourier shell correlation of independent half-data sets, with a cutoff at 0.5). The final resolutions of the class I, II, and III maps are 13, 14, and 14 Å, respectively. The particle numbers selected for the final reconstructions are 2,366, 5,085, and 5,861 particles, respectively. Approximately 200 of the 13,527 particles were immature dengue virus particles.

To validate these structures, we used a different method of classification. Using the entire data set and mature virus as a model for reconstruction by MPSA, a structure with a larger diameter than that in the mature virus model was generated. (This was later used as the first model.) However, not all of the particles were selected from this reconstruction, indicating that there are other classes of particles. We then generated a random model by simply averaging the remaining projections. (This was later used as the second model.) We then started another reconstruction using the first two models and the original mature virus maps as initial models running simultaneously. Maps generated from these three models were similar to the class I, II, and III maps obtained when classifications were done by K-mean classification.

Model fitting.

The cryo-EM structure of the DENV2 E protein (Protein Data Bank [PDB] accession code 1P58) (15) was used for fitting into the cryo-EM maps. The structure fits well into the class I map since it is similar to the previously published cryo-EM DENV2 map. For the class III map, the handedness of the map was determined by superimposing it with the envelope protein structure mentioned above. The E proteins were manually fitted into the map density using the program O (17). Briefly, the three E proteins in the icosahedral asymmetric unit had to be moved to a higher radius in order to fit into the density. The dimer (molecules A and C) that lies closer to 5-fold axis required translation and rotation adjustments, whereas the 2-fold E protein dimer (molecules B and B′) required some rotational adjustments. The final fitting results showed no clashes between the E proteins, and all densities were interpreted. The Fit in Map tool implemented in the program Chimera (18) was then used to optimize the fit. Although the class II map is estimated to have a 14-Å resolution, visual inspection of the map showed that it did not have enough visible features for the interpretation of the map. One possibility is there is a heterogenous population of virus in this class, resulting in smeared densities. Thus, the class II map was not interpreted. Similarly, no model fitting was done for the class IV map as the map only shows the lipid bilayer with only small patches of density left for the E proteins.

Comparison of the infectivities of DENV preincubated at 28 and 37°C.

Briefly, BHK cells were cultured in RPMI supplemented with 10% FBS at 37°C in 5% CO₂. When cells reached approximately 80% confluence, virus infection was carried out. Virus was serially diluted 10-fold from 1:10² to 1:10⁸ using RPMI supplemented with 2% FBS. For the 28°C experiment, the virus was preincubated at 28°C for 30 min or 1 h and then 100 μl of virus was layered onto the BHK cells, which were preequilibrated at 28°C for 2 h. For the 37°C experiment, the virus and cells were treated in a similar manner to the 28°C experiment, except the temperature was set at 37°C. The cells were then further incubated at their respective temperatures for 2 h. The infected cells were overlaid with RPMI supplemented with 2% FBS and 1% Aquacide. The plates were further incubated at 37°C in 5% CO₂ for 4 days. The cells were then stained by first discarding the supernatant and then adding 1 ml of 0.5% (wt/vol) crystal violet–25% formaldehyde to the cell monolayer and further incubated for 1 h. The cell monolayers were then washed with water, and plaques were counted by visual inspection. The experiments were repeated twice.

Protein structure accession number.

The cryo-EM map reconstructed from class III particles was deposited in the Electron Microscopy Database under accession no. EMD-2296. The modeled protein structure was deposited in the PDB under accession no. 3ZKO.

Virus morphology changed when the virus was incubated at 37°C.

The cryo-EM images of the control sample of the DENV2 NGC strain that was not exposed to 37°C (Fig. 1, left) showed about ∼80% of the particles had a smooth surface and round shape indicative of the mature virion. The rest of the virus population consisted of either broken nonrounded particles or the spiky immature virus. When incubated at 37°C, the majority of the virus particles showed rough surface (Fig. 1, right). The same observation was also found with DENV-2 WRAIR strain S16803 (data not shown).

DENV incubated at 28 and 37°C showed similar infectivities in BHK cells.

When tissue culture supernatant containing virus was incubated at 28 or 37°C for 30 min and then layered onto BHK cells for another 2 h at their respective temperatures, the number of plaques obtained were 3.6 ×10⁶ PFU/ml and 3.1 × 10⁶ PFU/ml, respectively. Increasing the incubation time of the virus to 1 h prior to infection showed plaque counts of 2.4 × 10⁶ PFU/ml and 2.6 × 10⁶ PFU/ml for 28 and 37°C, respectively. This suggests that increasing the temperature to 37°C does not alter the infectivity of the virus.

Reconstruction of four different classes of DENV particles observed at 37°C.

The cryo-EM maps of classes I, II, and III are shown in Fig. 3. The class I map is similar to the previously published mature DENV map (7–9, 15). Briefly, the particle has a radius of ∼240 Å, and the surface is relatively smooth with small bumps at positions corresponding to all E protein domain IIIs. The central cross-section shows the well-resolved lipid bilayer inside the envelope protein shell. The class II particle is slightly larger than the class I virion and has protrusions between the 5- and 3-fold vertices. The E protein layer is located at a higher radius of ∼250 Å compared to class I particles, indicating an expansion in the outer protein shell. Class III particles are the largest among the three classes, with a radius of ∼260 Å. The class III particle shows more pronounced protrusions between 5- and 3-fold vertices that reach out to 280 Å from the center. There is a hole at all of the 3-fold vertices (Fig. 4, right).

The cryo-EM map reconstructed from class IV particles showed a radius of only ∼180 Å with some irregular densities around the 5-fold vertices (Fig. 3). The cross-section shows that most of the densities corresponding to the E protein shell have disappeared, indicating that the E protein layer has lost its icosahedral symmetry.

Structure fitting to the cryo-EM map.

The organization of E proteins in the class I particle is identical to those of previously reported DENV structures (7–9, 15). This class has the most compact structure on its outer layer compared to the other classes. There are 180 copies of envelope proteins on the virus surface forming the herringbone pattern (Fig. 4, left).

The map generated from the class II particles cannot be interpreted as the surface is very smooth and continuous, with no obvious features or boundaries, except at the protrusion between the 5- and 3-fold vertices (Fig. 3). Attempts to fit the map produced numerous possibilities with no clear optimal fit. In contrast to the class II map, the class III map had more features and showed clearer boundaries guiding the fit of the E protein molecules. The fit was done by manually fitting the E protein dimer as a rigid body into the densities near the 5-fold vertex and assigning the remaining density to the E proteins at the icosahedral 2-fold vertex (Fig. 4, right). The final structure has a correlation coefficient of 0.84 as calculated by the Fit-in-Map tool in Chimera (18). In the final structure, domains I and III of molecules C of the dimers near the 5-fold vertices define the periphery of the hole at the 3-fold vertices (Fig. 4 and 5B), whereas the protrusions between the 5- and 3-fold vertices (Fig. 4) are filled by domains I and III of molecule B (Fig. 4 and 5B).

Conformational changes of DENV at 37°C.

Increasing the temperature to 37°C induced structural rearrangement of the E proteins on the surface of DENV. We observed in the class III particles that the E protein layer moves outward, while the membrane (lipid bilayer) remains at the same radii as the unexpanded particles, thus creating a gap of about 15 Å (Fig. 5A). The 5-fold dimer (molecules A and C) undergoes translation about ∼14 Å along the dimer axis and also rotates about 7° anticlockwise (Fig. 5B). The dimeric interactions of the E proteins at the icosahedral 2-fold vertices loosen (molecules B and B′), and the two E protein molecules move in opposite directions perpendicular to the dimer axis, creating a distance of ∼6 Å between the molecules. Furthermore, each of the E proteins rotates clockwise by about 14° using the fusion loop as a pivot point (Fig. 5C). The conformational changes of the icosahedral 2-fold E proteins break the raft structure.