|Home | About | Journals | Submit | Contact Us | Français|
Hantaviruses (family Bunyaviridae) are rodent-borne emerging viruses that cause a serious, worldwide threat to human health. Hantavirus diseases include hemorrhagic fever with renal syndrome and hantavirus cardiopulmonary syndrome. Virions are enveloped and contain a tripartite single-stranded negative-sense RNA genome. Two types of glycoproteins, GN and GC, are embedded in the viral membrane and form protrusions, or “spikes.” The membrane encloses a ribonucleoprotein core, which consists of the RNA segments, the nucleocapsid protein, and the RNA-dependent RNA polymerase. Detailed information on hantavirus virion structure and glycoprotein spike composition is scarce. Here, we have studied the structures of Tula hantavirus virions using electron cryomicroscopy and tomography. Three-dimensional density maps show how the hantavirus surface glycoproteins, membrane, and ribonucleoprotein are organized. The structure of the GN-GC spike complex was solved to 3.6-nm resolution by averaging tomographic subvolumes. Each spike complex is a square-shaped assembly with 4-fold symmetry. Spike complexes formed ordered patches on the viral membrane by means of specific lateral interactions. These interactions may be sufficient for creating membrane curvature during virus budding. In conclusion, the structure and assembly principles of Tula hantavirus exemplify a unique assembly paradigm for enveloped viruses.
Hantaviruses constitute a genus in the Bunyaviridae, a large family of viruses with more than 350 characterized members (6, 22, 24, 32). In contrast to the arthropod-borne members of the other genera (Orthobunya-, Nairo-, Phlebo-, and Tospovirus), hantaviruses are rodent-borne (24). Recently, hantaviruses have been detected also in insectivores (36). Hantaviruses cause persistent infections in their host animals worldwide. Climate change may cause unexpected changes in the host populations and thus in the frequency of hantavirus outbreaks (19). Hantaviruses are transmitted to humans by inhalation of infected rodent aerosolized excretions. Infections can lead to two different severe diseases: hemorrhagic fever with renal syndrome and hantavirus cardiopulmonary syndrome. Hemorrhagic fever with renal syndrome, caused by Old World hantaviruses, such as Hantaan and Seoul virus, has a mortality rate of <15%. In Europe, Puumala virus causes a mild form of the disease, nephropathia epidemica, with a mortality rate of <1%. Hantavirus cardiopulmonary syndrome caused by New World hantaviruses, such as Sin Nombre and Andes virus, is more severe, with a mortality rate of >40% (31). Currently, no specific antiviral drugs or approved vaccines exist (17, 21). Rational design of antivirals relies on structural information of viral particles and proteins. In addition, knowledge of virus structure is a cornerstone in understanding the crucial events in viral replication cycles, such as virion entry into the host cell and virus assembly. Here, we have taken the first steps to understand the ultrastructure of these viruses and show that their ultrastructure is remarkably different from that of other enveloped viruses studied.
Hantaviruses, like all other members of the Bunyaviridae family, are enveloped viruses with segmented, negative-sense, single-stranded RNA genomes (7, 24, 30). The genome encodes four structural proteins: a nucleocapsid protein (N), two glycoproteins (GN and GC), and an RNA-dependent RNA polymerase (L) (29). The two glycoproteins are responsible for binding to the β3-integrin receptor (11) and membrane fusion (38). They are synthesized as an M-segment-encoded precursor protein that is cleaved to N- and C-terminal fragments during translation, which eventually mature into GN and GC. Maturation and transport to the Golgi apparatus require the presence of both glycoproteins, and the process is thus assumed to be driven by a signal formed by the GN-GC heterocomplex (5, 34). The viral RNA segments are coated by the nucleocapsid protein to form ribonucleoprotein (RNP) complexes (32). The cytoplasmic tails of one or both glycoproteins are thought to interact with the RNP complexes (8, 26, 35). The virions bud into the lumen of the Golgi apparatus and are transported in vesicles to the plasma membrane, where they are released from the cell (9).
The shape of the mature hantavirus virions, as seen in negatively stained electron microscopy (EM) specimens, varies from spherical (diameter of ~100 nm) to elongated (~110 nm by ~170 nm) (22). Recently, structures of two other bunyaviruses, Uukuniemi (25) and Rift Valley fever phlebovirus (10, 16, 33), have been studied in detail. In contrast to supposedly pleomorphic hantavirus virions, these structures revealed an icosahedrally symmetric organization of the glycoproteins on the virion surface. Furthermore, the structure of Rift Valley fever virus suggested that the glycoprotein layer is composed of pentamers and hexamers of GN-GC heterodimers (16). In contrast, hantavirus GN and GC were recently reported to favor homooligomers over heterooligomers, and the virion surface was hypothesized to be composed of homotetrameric GN complexes interconnected with homodimeric GC complexes (13). However, in the absence of experimental three-dimensional structural models, the organization of the viral glycoproteins and the packaged RNP in the virion remains poorly understood.
Here, we have studied the structure of apathogenic Tula hantavirus virions (28, 39). The pleomorphic morphology of hantavirus virions has impeded their structural characterization using conventional electron cryomicroscopy and single-particle reconstruction techniques (1, 2). To overcome this challenge, we used a combination of electron cryomicroscopy, computational three-dimensional tomographic reconstruction, and averaging of tomographic subvolumes (1, 12, 37). The averaging of tomographic subvolumes has become a useful tool in the analysis of glycoprotein structures of pleomorphic membrane viruses (37). Our tomographic reconstructions revealed the virion ultrastructure, including the organization of the surface glycoprotein layer, of the membrane and of the RNP. The structure of the GN-GC spike complex in situ, at 3.6-nm resolution, suggested that each spike is composed of a tetramer of GN-GC heterodimers. Tetrameric glycoprotein complexes were organized as ordered patches on the viral membrane. Taken together, the novel features of the Tula hantavirus structure exemplify a unique assembly paradigm for enveloped viruses.
Vero E6 green monkey kidney epithelial cells (ATCC 94 CRL-1586) were grown in minimal essential medium (MEM) supplemented with 10% heat-inactivated fetal calf serum (FCS), glutamine, 100 IU/ml of penicillin, and 100 μg/ml of streptomycin at 37°C in a humidified atmosphere containing 5% CO2. Tula virus Moravia strain 5302 was used to infect an ~80% confluent monolayer of cells in 75-cm2 cell culture flasks by adsorption with 1 ml of 1:50 to 1:100 diluted virus supernatant (collected 10 days postinfection) for 1 h at 37°C. This equals a multiplicity of infection of between 0.01 and 0.1. After adsorption, the infected cells were grown in supplemented MEM as described above. After 7 days, the 10% FCS was replaced with 2 to 3% FCS filtered through a 100-kDa cutoff filter (Millipore). The medium containing excreted viral particles was collected at 3-day intervals for 7 to 10 days in total and stored on ice at +4°C. All collected media were pooled prior to virus purification.
Collected growth medium containing virus particles was passed through a 0.22-μm-pore-size filter (Millipore) to discard larger assemblies. Medium was concentrated 10-fold using a 100-kDa cutoff concentrator (Millipore). A density gradient was formed by layering 1.5 ml of 50, 42, 35, 28, 21, 14, and 7% solutions of OptiPrep (OptiPrep stock solution diluted in 10 mM HEPES-135 mM NaCl) sequentially into SW41 tubes (Beckman) and allowing the gradient to form by diffusion at +4°C overnight. The concentrated medium was layered on top of the preformed OptiPrep density gradient and centrifuged for 15 h at 25,000 × g at 4°C. Fractions of approximately 600 μl were collected from the bottom by puncturing the tube with a needle. Virus titers were determined as described previously (18). Protein concentration in each step was determined using bicinchoninic acid protein assay reagent (Thermo Scientific, Pierce Protein Research Products) according to the manufacturer's instructions. Sample purity was analyzed by separating proteins in SDS-PAGE and by Coomassie staining. To remove the OptiPrep prior to electron cryomicroscopy, fractions containing virus were dialyzed against buffer.
For electron cryotomography, a 3-μl aliquot of concentrated virus suspension was pipetted on a glow-discharged holey carbon-coated EM grid. Colloidal 10-nm gold particles, coupled to bovine serum albumin (BSA), were added. Excess suspension was removed with filter paper prior to vitrification by plunging the sample into liquid ethane. Electron cryotomography was performed at liquid nitrogen temperature using a 300-keV electron microscope (Polara; FEI, Eindhoven, Netherlands) equipped with an energy filter (GIF 2002; Gatan, Pleasanton, CA) operated in a zero-energy-loss mode with a slit width of 10 eV.
Twenty-five single-axis tomographic tilt series were collected covering approximately the angular range of −60 to 60° at 2° increments by using SerialEM (23) under low-dose conditions, keeping the total dose less than 100 electrons/Å2 per series. The images were acquired approximately at −5 to −6 μm defocus and at a calibrated magnification of ×67,300, corresponding to a pixel size of 0.45 nm.
Three-dimensional reconstructions were calculated in IMOD using 10-nm gold particles as markers to align the images of the tilt series with respect to each other (20). A low-pass filter was imposed during tomographic reconstruction to remove spatial frequencies higher than the first zero (on average at 3.3 nm−1) in the contrast transfer function of the microscope. To visualize slices through the tomographic reconstructions (tomograms), the tomograms were first denoised using nonisotropic diffusion in Bsoft software package (14).
Further image processing was carried out using the Bsoft software package (14) unless stated otherwise. Forty-five large spherical virions and five tubular virions were extracted from the tomographic reconstructions for further analysis. The diameters of spherical virions were measured from radially averaged densities (25), and the sizes of the tubular particles were measured manually.
The structure of the GN-GC membrane-bound complex was solved using iterative template-matching alignment and averaging of aligned subvolumes. A program based on the Bsoft function library (14), Jsubtomo, was written to carry out these tasks (available upon request from the authors). The analysis concentrated on the spikes in 45 large spherical virions since these virions were the most abundant and were not aggregated. Initial analysis showed the clear presence of at least 2-fold, if not 4-fold, symmetry in the structure of the spike, and thus 2-fold symmetry was applied during the alignment and averaging. The alignment was based on cross-correlation between a template volume for the spike and the tomographic volume of a virion. The missing information in the tomographic volumes (“missing wedge”) was taken into account. First, a missing wedge mask was applied during the cross-correlation search. Second, the amplitudes in the Fourier transform of the averaged volume were normalized to restore underrepresented amplitudes. The template volume for the first round of template matching was generated by manually averaging 10 subvolumes. In further rounds, the average of all the subvolumes provided a template volume for the next round. Before the averaging of the subvolumes, false positives (e.g., complexes not found to locate on the viral membrane or complexes belonging to a neighboring viral particle) were discarded interactively in UCSF Chimera (27). The process was iterated for seven rounds. Finally, orientations of the true positives were refined to 1° accuracy, and the final average was calculated using 2,353 subvolumes. The resolution of the model was assessed by Fourier shell correlation. Two averages were calculated from two half-sets of the data, and 2-fold symmetry was applied. The spatial frequency at which the correlation dropped below a threshold of 0.5 defined the resolution of the averaged structure.
The average was correlated with the five tubular virions to determine the number of spikes also in these particles. To calculate the packing efficiency of the spikes, the membrane area was measured in IMOD (20). The packing efficiency was calculated by dividing the summed area of all spikes by the measured area of the membrane.
The structure of the GN-GC membrane-bound complex was submitted to the Electron Microscopy Data Bank (EMDB) under accession number EMD-1704.
We first optimized the hantavirus purification protocol to improve the yield of intact hantavirus particles for structural analysis (see Materials and Methods) (13). The efficiency of the virus purification was analyzed by determining the total number of infectious viruses, the protein concentration, and the specific infectivity at each step (Table (Table1).1). The gradient fraction with the highest yield of infectious particles was chosen for electron cryotomography.
We performed electron cryotomography of purified virions and computed 25 three-dimensional reconstructions, each covering a field of several virus particles (Fig. (Fig.1).1). Because no heavy metal staining was used, the reconstructions resolved the true three-dimensional density distribution arising from the virions themselves. In addition, possible artifacts due to drying were avoided (1, 12).
The reconstructions revealed virions with different morphologies. Unlike icosahedrally symmetric viruses with a defined size and shape, the size and shape of Tula virions varied. Large spherical particles were most abundant (Fig. (Fig.1A).1A). The diameter of these particles was 120 to 160 nm, and the median diameter was 135 nm. Occasionally smaller particles (diameter, 60 nm) (Fig. (Fig.1,1, asterisks) and aggregates of tubular particles (Fig. (Fig.1B)1B) were observed. The most elongated tubular particles were ~350 nm long and ~80 nm in diameter.
The surfaces of the particles were nearly fully covered by spike structures. Only small areas of the membrane were exposed (Fig. (Fig.1A,1A, arrowheads). In large spherical and tubular particles, all of the spikes appeared similar, displaying four globular features. We assign these spike structures to glycoproteins GN and GC. The membrane appeared as one 5-nm-thick continuous layer. Threads of RNP (8 nm wide) filled the particles nearly evenly and made connections to the membrane occasionally (Fig. (Fig.2).2). Sometimes layers of parallel RNP threads were evident. In some of the particles, straight rod-shaped densities distinct from the RNP densities were present (Fig. (Fig.1B,1B, arrows). The small 60-nm particles were devoid of RNP density altogether and revealed a different surface pattern from that of the large spherical and tubular particles (data not shown). Thus, the 60-nm particles may not, in fact, be of viral origin.
To locate the GN-GC complexes on the virions and to average all the complexes together for detailed structural analysis (see below), we used a computational approach based on the Bsoft software package (14) (see Materials and Methods for details). Briefly, a template structure of the spike was correlated with the virion densities (Fig. (Fig.3A).3A). Cross-correlation defined the most likely positions and orientations of each spike. The method was sensitive enough to detect the spikes, as shown by the peaks in the cross-correlation maps (Fig. (Fig.3B).3B). All spikes could then be aligned for averaging the spike densities together to create a new template, and the process was iterated. Finally, for visualization, the average density of all spikes was placed on the initial positions to generate nearly noise-free models of the viral spike layer (Fig. (Fig.3C).3C). These models revealed ordered patches of four-lobed, square-shaped glycoprotein spikes and resembled the grid-like patterns observed in negatively stained EM images earlier (22).
To gain better understanding of the interactions between the GN-GC spikes in these locally ordered patches, we performed quantification of interspike distances and angles in 45 nearly spherical virions. We calculated the pairwise distances between all spikes in each virion (Fig. (Fig.4A).4A). The minimum spike-to-spike distance was 15 nm, corresponding to edge-to-edge packing of spikes. The second smallest preferred spike-to-spike distance was 20 nm, corresponding to diagonally related spike neighbors in the ordered spike lattice. The rest of the preferred distances reflected a longer-range order in the lattice.
We calculated the angles between every spike and its neighbors (defined as spikes within a 24-nm radius). The neighboring spikes had a preferred out-of-plane angle difference of 15°, reflecting a tendency to adopt a specific curvature on the viral membrane (Fig. (Fig.4B).4B). The histogram of the in-plane angle differences showed a peak centered at 0°, indicating that within the patches, all the spikes tend to point in the same direction (Fig. (Fig.4C).4C). To conclude, the defined distance and angles between a spike and its neighbors argue for specific spike-to-spike interactions, as opposed to random assortment of spikes in the viral membrane.
The subtomogram averaging approach yielded an average structure of 2,353 GN-GC spike complexes (Fig. (Fig.5).5). The resolution of the structure was 3.6 nm as determined by Fourier shell correlation (using a correlation threshold of 0.5). Because the spikes exhibited nearly even distribution of orientations on the spherical particles, the distorting effect of the missing wedge of tomographic data collection was nearly averaged out in the structure. Further care was taken to remove any remaining effect by up-weighting the underrepresented amplitudes in the averaged structure (see Materials and Methods).
The spike complex was a square-shaped assembly, measuring 15 nm by 15 nm and 12 nm in height and displayed a clear 4-fold rotational symmetry although only 2-fold symmetry was imposed during the averaging. The complex had four globular domains distal to the membrane (Fig. (Fig.5A).5A). These globular domains were connected to the membrane by thinner stalk regions (Fig. (Fig.5B).5B). The membrane was 5 nm thick. Additional densities were apparent on the intraviral surface of the membrane, right under the center of the spike (Fig. (Fig.5C5C).
In the averaged structure, the neighboring spike complexes were also resolved since they exhibited a preferred orientation relative to the complex aligned in the middle (Fig. (Fig.5A).5A). The spikes appeared to be lined up in rows. Spikes made more extensive contacts to their neighbors within a row and less extensive contacts between the rows. Thus, although each single spike seems to have a 4-fold symmetry, the lattice seems to have only a local 2-fold symmetry, possibly to account for the curvature in the lattice.
Slices through the averaged density revealed internal density variation in the spike complex (Fig. 5D and E). Each spike can be divided conceptually into five different structures: (i) membrane distal part, consisting of four globular densities; (ii) a central stalk; (iii) four peripheral stalks; (iv) membrane part; and (v) intraviral density, residing under the membrane. The four membrane distal globular densities showed a clear 4-fold symmetry (Fig. (Fig.5D,5D, slice 3, blue). The two types of stalks were evident in the middle part of the spike (Fig. (Fig.5D,5D, slices 7 to 9). Pairs of peripheral stalks formed structures with local 2-fold symmetry at the spike-spike interfaces (Fig. (Fig.5D,5D, slice 8, blue, and E, slice 3, arrows). These pairs formed the main spike-to-spike contacts, and they surrounded the central stalk. Together with the central stalk (Fig. (Fig.5E,5E, slice 8, arrow) they connected the membrane distal part of the spike to the membrane. Due to limited resolution, no transmembrane densities were observed spanning the membrane (Fig. (Fig.5D,5D, slice 11). However, an intraviral density protruding from the membrane was found to locate right under the central stalk (Fig. (Fig.5D,5D, slice 13, yellow, and E, slice 8, arrowhead).
Recently Hepojoki et al. (13) assessed the protein composition of hantavirus spike complexes by cross-linking GN and GC on the surface of purified Tula virus virions and by studying the oligomerization of isolated GN and GC. The association of GN mainly as homotetramers and of GC mainly as homodimers suggested that the stoichiometry of each complex in the virion is (GNGC)4. In addition, the predicted masses of the GN and GC ectodomains from the sequence are 54 kDa and 51 kDa, respectively. These results fit well with our observations. First, the mass predicted from the structure for the complete spike complex ectodomain was 480 kDa (Fig. 5A to C). Second, we observed a 4-fold symmetry in the spike complex. These observations are consistent with the idea that one spike contains four copies of both of the glycoproteins. Furthermore, it is feasible that the membrane-distal part of the spike with four globular features corresponds mainly to a GN tetramer. In addition, the only features with local 2-fold symmetry, the pairs of peripheral stalks, are likely to correspond to GC homodimers. However, at the current resolution, we cannot exclude the possibility that some domains of GN and GC are intertwined in the structure. Thus, both GN and GC could contribute, for example, to the 4-fold symmetrically distal part of the spike. To conclude, we suggest that the spike complex is most likely a tetramer of GN-GC heterodimers, but further studies are required to determine the exact domain organization.
How do the hantavirions assemble and how do the glycoprotein spikes form larger assemblies on the viral membrane? We suggest that most of the membrane curvature-creating forces required during budding arise through the interactions between the ectodomains of the two glycoproteins for three reasons. First, there is no matrix protein present. Second, our data excluded the possibility that the endodomains of the glycoproteins or RNP could form an organized shell under the membrane and carry out a matrix-like function. We observed only a small ordered density under the membrane, likely corresponding to an ordered intraviral glycoprotein density, but no interactions were seen between these densities. Third, the spikes were seen to form lattices in a nonrandom fashion and to interact in a ~15° angle (Fig. (Fig.4B).4B). This interaction could be sufficient to create enough membrane curvature for budding. Interactions between preassembled RNP particles and the glycoprotein tails or cellular factors may still contribute to the budding process.
How much membrane-curving potential the spike ectodomains have depends not only on their shape but also on how tightly they can pack on the membrane. Each spike occupies an estimated area of 225 nm2 on the viral membrane. The membrane area for an average-sized spherical virion was 35,000 nm2. With 100% packing efficiency, this area could accommodate ~150 spikes. However, the actual packing efficiencies were expected to be much lower for two reasons. First, we observed membrane areas devoid of spikes. Second, covering a sphere fully with square-shaped tiles is a geometrical impossibility. In our analysis we observed packing efficiencies of up to ~70%. This corresponds to ~100 spikes for a 135-nm diameter virion. Some spikes have likely remained undetected in our computational analysis, and thus the actual maximum packing efficiency may be slightly higher. In the few observed tubular virions, packing efficiencies up to ~90% were observed.
How are the RNPs incorporated in the virions? Interactions between the RNP and the glycoprotein intraviral tails have been reported for other bunyaviruses recently (26, 35). Both of the Tula virus glycoproteins, GN and GC, contain intraviral tails whose predicted lengths are 120 and 10 amino acid residues, respectively. However, to the best of our knowledge, neither the GN nor GC of hantaviruses has been demonstrated to bind RNP. The GN intraviral domain has recently been shown to adopt a novel fold consisting of two classical zinc fingers. However, this domain did not bind RNA and was suggested to bind the nucleocapsid protein instead (8). In slices through our Tula virus tomograms, connections from the RNP to the membrane were occasionally observed (Fig. (Fig.2).2). These connections seem to be mainly disordered since we detected only a small intraviral density under the spikes (Fig. (Fig.5C).5C). We suggest that these connections, whose nature remains unknown, may be sufficient for genome incorporation. Thread-like RNP density was seen to fill most of the particles evenly, despite their various sizes. This suggests that different amounts of RNP, for example, different copy numbers of each segment, can be incorporated in the virions.
To summarize our model on hantavirus assembly, we suggest that lateral growth of the glycoprotein ectodomain lattice on the Golgi membrane drives the budding. Additional force may arise from glycoprotein-RNP interactions. In the other enveloped viruses, where an ordered lattice, either intraviral or extraviral, drives budding, the building blocks are pentagon and hexagon shaped (3, 4, 15). In contrast, the building blocks described here are square shaped. Taking these considerations together, we suggest that Tula hantavirus exemplifies a unique assembly paradigm for enveloped viruses.
This work was supported by the Academy of Finland (Post Doctoral Researcher's Project 2007-2009 grant 114649 to J.T.H. and grant 102371 to H.L. and Centre of Excellence Program in Virus Research 2006-2011 grant 1129684 to S.J.B.), by the European Molecular Biology Organization (Long-Term Fellowship ALTF 820-2006 to J.T.H.), by Tekes, the Finnish Funding Agency for Technology and Innovation (grant 40264/07 to H.L.), by the Jusélius Foundation (to S.J.B. and A.V.), and by the Deutsche Forschungsgemeinschaft (GR1990/1-3,4 and -2-1,2 to K.G.).
John Briggs and Thomas Bowden are thanked for their valuable comments on the manuscript.
Published ahead of print on 10 March 2010.