Structure of Hepatitis E Virion-sized Particle Reveals an RNA-dependent Viral Assembly Pathway*

Hepatitis E virus (HEV) induces acute hepatitis in humans with a high fatality rate in pregnant women. There is a need for anti-HEV research to understand the assembly process of HEV native capsid. Here, we produced a large virion-sized and a small T=1 capsid by expressing the HEV capsid protein in insect cells with and without the N-terminal 111 residues, respectively, for comparative structural analysis. The virion-sized capsid demonstrates a T=3 icosahedral lattice and contains RNA fragment in contrast to the RNA-free T=1 capsid. However, both capsids shared common decameric organization. The in vitro assembly further demonstrated that HEV capsid protein had the intrinsic ability to form decameric intermediate. Our data suggest that RNA binding is the extrinsic factor essential for the assembly of HEV native capsids.

Hepatitis E virus (HEV), 3 the causative agent of acute hepatitis in humans, is primarily transmitted through contaminated water and generally results in epidemic outbreaks in many developing countries. Sporadic cases have also been reported between outbreaks in HEV-endemic regions as well as in nonendemic areas, and these cases are transmitted through zoonotic route. The overall mortality rates of HEV during outbreaks range from 1 to 15% in general, and the highest mortality occurs in pregnant women, with fatality rates of up to 30% (1).
HEV consists of a non-enveloped icosahedral capsid and a single-stranded, positive-strand RNA genome of ϳ7.2 kb that encodes three open reading frames (ORFs) (2). The capsid protein, encoded by the ORF2, is composed of 660 amino acids and responsible for most capsid-related functions, such as virion assembly, host interaction, and immunogenicity. Like other hepatitis viruses, HEV is unable to propagate in currently available cell culture systems, and the research of HEV relies largely on the recombinant HEV capsid proteins (3)(4)(5)(6). Virus-like particle (VLP) was obtained when the truncated HEV capsid protein was expressed in insect Tn5 cells with deletion of 52 residues from the C terminus and 111 residues from the N terminus (PORF2) (7). Our previous structural analysis of this HEV-VLP by cryo-electron microscopy (cryo-EM) provided a basic understanding of the quaternary arrangement of PORF2, where the reconstructed VLP displayed a Tϭ1 icosahedral particle composed of 60 copies of PORF2 (8). The essential element of PORF2 protein for Tϭ1 VLP assembly includes amino acids 125-600 (9). Recently, the structural information was further refined by the crystal structures of genotype-3 Tϭ1 VLP (10) and genotype-4 Tϭ1 VLP (11), which revealed the tertiary structure of PORF2 to the level of amino acids. However, the Tϭ1 VLPs used in these experiments were much smaller than that of the native virion, which has a diameter of 320 -340 Å, as determined by immuno-EM (12). There is still a need to investigate the assembly pathway of HEV capsid.
HEV virion is hypothesized to be made of 180 copies of the capsid protein (8,11). To test this hypothesis, we screened for HEV genotype expression and successfully produced a virionsized VLP from the HEV genotype-3 ORF2 protein after deleting 52 residues from the C terminus. This VLP allowed us to investigate the molecular interactions that govern HEV virion assembly.

EXPERIMENTAL PROCEDURES
Production of HEV-VLPs and in Vitro Disassembly and Reassembly-HEV-VLPs were produced and purified according to the protocol described previously (7,13). Briefly, the recombinant baculovirus Ac(G3n13ORF2), which harbored the genome of the N-terminal 13 amino-acid-deleted genotype-3 HEV ORF2, was infected into Tn5 cells with recombinant baculovirus at a multiplicity of infection of 10 (20 mM). Scanning Transmission Electron Microscopy Analysis of HEV-VLPs-Scanning TEM (STEM) was performed at the Brookhaven National Laboratory STEM facility, with tobacco mosaic virus (TMV) as an internal control. The mixture of VLP and TMV was quickly frozen in liquid nitrogen and then maintained at Ϫ150°C during data collection to eliminate contamination and reduce mass loss. The specimen was scanned by a 40-keV electron beam of 0.25 nm in size, and images were collected with a preamp gain of 10 for both large and small angle detectors (14). The image was recorded with a pixel size of 10 Å and was analyzed with the PCMass29 program. After normalizing the background, the mass of the VLPs was selected with the murine sarcoma virus (MSV) shell model provided by the program. Mass measurements for TMV and HEV-VLPs were always performed from the same image. The HEV-VLP mass was measured in MDa (mass per particle), and the TMV mass was measured in KDa/Å (mass per unit length) (15).
Cryo-electron Microscopic Structure Determination of HEV Tϭ3 VLP-The collection of cryo-EM data for image reconstruction was performed on a JEOL JEM-2100F TEM operating at 200 kV according to the procedure described in detail previously (8). Briefly, a 3-l solution containing VLP or reassembled ORF2 complex was placed on holey carbon film-coated copper grids and then quickly plunged into liquid ethane after the removal of excess solution. The VLPs were embedded into a thin layer of vitrified ice and transferred into the EM using a Gatan 626 cryo-transferring system. The specimen was observed under 50,000ϫ magnification, and the area of interest was recorded on a TVIPS CCD camera (TemCam-F415). The micrographs were recorded with a pixel size of 2.0 Å at a specimen space and defocus level of 0.7-3.5 Å (supplemental Fig.  1A). Digital images with no stigmatism or drift were selected for later image processing. Images of individual HEV Tϭ3 VLPs were then boxed out and processed with an established software package for icosahedral particles (16,17). In total, 7,720 individual images were included in the process, and their defocus levels were distributed mainly within 1.0 -2.5 Å.
To correct contrast transfer function effect, we applied phase flipping on each image with an in-house program. The density maps were initially reconstructed by combining 1,812 individ-ual images to an effective resolution of 14 Å. Next, amplitude correction was applied during map reconstruction while new data were added. The final density map was reconstructed by combining images of 4,348 individual particles, and the final resolution was assessed as 10.6 Å by Fourier shell correlation with a cutoff of 0.5 (supplemental Fig. 1B).
Docking of the Tϭ1 crystal structure into the Tϭ3 cryo-EM density map was first done manually with the program O (18) and then refined with the Situs software package (19). The PORF2 monomer was treated as a rigid body during the initial fitting and refinement processes.
X-ray Crystallographic Structure Determination of Tϭ1 HEV-VLP-Crystallization of the VLPs was performed according to a previously described method (20). Crystals were directly flashfrozen in liquid nitrogen, and x-ray diffraction experiments were performed. All x-ray experiments of the HEV-VLP crystals were performed at SPring-8 in Hyogo, Japan. Particle orientation in the unit cell was determined with a self-rotation function (21), and the particle position was determined by a translation search with the cryo-EM structure as the model. The asymmetric crystal unit contains one particle; as a result, 60-fold non-crystallographic symmetry averaging was enforced. The cryo-EM structure (8) was used to obtain the initial phases of Data I (supplemental Table 1) and generated the envelope (mask) used for non-crystallographic symmetry averaging. The phases were refined by real space electron density averaging with icosahedral symmetry elements and solvent flattening. The resolution was gradually extended to 8.3 Å (R-factor ϭ 0.21, correlation coefficient ϭ 0.92). This structure was used for the phasing of Data II (supplemental Table 1), and the phases were refined and extended to a 3.8 Å resolution (the overall R-factor and correlation coefficient were 0.18 and 0.97, respectively). The electron density map revealed a clear main chain structure. Thus, we built an atomic model into this electron density map using the program O (18) without difficulty.
Sixty icosahedrally related S-subunits were treated as identical, and strict non-crystallographic symmetry constraints were applied during refinement. The data with the resolution range of 20 -3.8 Å were included in the refinement (supplemental Table 1 and supplemental Fig. 2) using the CNS program (22). The subunit position and orientation were again clarified by rigid body refinement (R-factor of 0.391 (R free ϭ 0.395)). After the first cycle of simulated annealing refinement, positional and B-factor refinement followed, and the model was enhanced to an R-factor of 0.261 (R free ϭ 0.264). Further positional and B-factor refinement, followed by manual revision of the model, resulted in an R-factor of 0.242 (R free ϭ 0.245) with reasonable stereochemistry (root mean square deviations in bond lengths and bond angles were 0.010 Å and 1.68°, respectively). Because of the high non-crystallographic symmetry, the R-factor and R free factor were almost identical. After refinement, the stereochemistry of the structure was checked with Procheck (23). 98.1% of the non-glycine residues were within the most favored and the additional allowed regions of the Ramachandran plot, and none of the residues were in the additional regions. Atomic structure representations were generated using MolScript (24) and Raster3D (25).
Nucleic Acid Extraction and Characterization-100 g of purified large and small VLPs were treated with DNase I (final concentration; 0.01 mg/ml) (Sigma) and RNase A (final concentration; 0.5 g/ml) at 37°C for 1 h and then followed by centrifugation at 50,000 rpm for 2 h at 4°C in a Beckman TLA55Ti rotor. After removal of the supernatant, the pellets were resuspended in 200 l of PBS minus buffer. The nucleic acids were extracted with RNAzol-LS reagent (Tel-test, Inc., Friendswood, TX) and analyzed on 1% agarose gels.
For detection of HEV RNA in VLPs, the extracted RNAs from VLPs were mixed with sample buffer (final concentration: 2% formaldehyde, 50% formamide in MOPS buffer containing 20 mM MOPS, 5 mM sodium acetate, and 1 mM EDTA (pH 7.0), Sigma-Aldrich, Tokyo, Japan). The solution was denatured at 65°C for 10 min and cooled immediately in ice-cold water and then mixed with 6ϫ loading buffer (80% formamide, 0.25% bromphenol blue, 0.25% xylene cyanol, 6 mM EDTA, Sigma-Aldrich). After separation on the formaldehyde-denatured agarose gel (18% formaldehyde, 1% agarose, Sigma-Aldrich) in 1ϫ MOPS buffer, the gel was washed twice for 15 min with 20ϫ SSC (3 M NaCl, 0.3 M trisodium citrate dehydrate (pH 7.0), Roche Applied Science, Tokyo, Japan) and transferred onto a Hybond N ϩ membrane filter (GE Healthcare, Tokyo, Japan) by the capillary method. RNAs were fixed on the membrane by irradiation with UV light (Stratalinker UV crosslinker, Stratagene, Tokyo, Japan) and dried in the air. RNAs on the membrane were hybridized with DIG-labeled probe. DIG-labeled RNAs were detected by using a DIG Northern starter kit (Roche Applied Science) according to the manufacturer's protocol and visualized by the LAS-3000 imaging analyzer, Fujifilm, Tokyo, Japan). The purified template DNA for the probe RNA was prepared by digesting the plasmid carrying 584-base cDNA of HEV ORF2 (5,903-6,486 nucleotides) under the T7 promoter with BamHI. The probe RNA was prepared using the DIG Northern starter kit according to the manufacturer's manual (incorporation of DIG-UTP during RNA transcription). The norovirus-like particles that were produced by the recombinant baculovirus were used as a control, and the same procedure was used with the HEV-VLPs.

STEM-
The virion-sized HEV-VLP was recovered when the genotype-3 ORF2 sequence was expressed in insect cells. This VLP projected as a spherical image with a diameter of ϳ40 nm, larger than the Tϭ1 VLP (27 nm in diameter) (Fig. 1A). In Both large and small Tϭ1 HEV-VLPs are projected as spherical images, and their corresponding particle mass was calculated. The long straight rod is TMV, which was added as an internal mass standard. (Bar ϭ 1,000 Å.) B, the large HEV-VLP appeared as an intact particle decorated with a dot-like pattern on the surface on cryo-EM of the large HEV-VLP. (Bar ϭ 500 Å.) C, mass spectrum of the large HEV-VLP showing that the molecular mass of the ORF2 protein is 65.5 kDa. D, plot showing the observed mass/length of TMV against the observed mass of the large HEV-VLP from different image conditions. For a known TMV mass/length of 13.1 kDa/Å, the mass of the large HEV-VLP was calculated as 11.8 MDa (asterisk). MD, megadalton. OCTOBER 22, 2010 • VOLUME 285 • NUMBER 43 cryo-electron micrographs, the images of HEV-VLP are decorated with spike-like features and are homogeneous in contrast (Fig. 1B).

Assembly of HEV T‫3؍‬ Virion-sized Particle
To determine the composition of the large HEV-VLP, we performed mass measurements by using STEM, a technique that measures the amount of electrons scattered from the objects, such as VLPs, on an EM grid. A mixture of purified large and small HEV-VLPs was freeze-dried onto EM grids for STEM mass measurement. TMV with a known mass-to-length ratio was used as an internal standard. The HEV-VLPs appeared as spherical projections with white contrast on the dark field STEM images (Fig. 1A). White cloud-like objects were present in the background, which might be the VLPs broken during sample preservation. The mean mass of large VLP and TMV in the images was measured to generate a plot of the mean TMV mass per unit length versus mean VLP mass per particle (Fig. 1A). A first-order fit was calculated, and the mass of the large HEV-VLP was determined to be 11.8 MDa (Fig. 1D). The mass of the genotype-3 ORF2 protein, which was recovered from the large VLP, was measured as 65.5 KDa by mass spectrometry (Fig. 1C). Therefore, the large HEV-VLP contains 180 copies of ORF2 proteins, suggesting that the large HEV-VLP is a Tϭ3 icosahedral particle (Tϭ3 VLP). We further performed element analysis with x-ray photoelectron spectroscopy. X-ray photoelectron spectroscopy, also known as electron spectroscopy for chemical analysis, determines the chemical composition of the sample with a depth of 50 -70 nm. Phosphorus element, as a characteristic element of nucleic acid, was detected from the G3-VLP that was applied to the carbon-coated copper grid (supplemental Fig. 1C). Although the signal of phosphorus element is weak when compared with that of carbon, the phosphorus peak is not detectable in the control grid without VLP materials. This result suggests the co-existence of nucleic acid within the large G3-VLPs.
Three-dimensional Reconstruction of the HEV Virion-sized Particle-The cryo-EM structure of the large HEV-VLP revealed 90 protruding spikes on a complete icosahedral shell ( Fig. 2A), which is consistent with the Tϭ3 icosahedral symmetry and the results of the STEM mass measurements. The VLP had an overall diameter of 410 Å and a central cavity of 170 Å in radius as measured from the three-dimensional density map (Fig. 2C). The single-layer capsid contained 180 copies of the ORF2 protein, which were grouped into three unique monomers according to their geometric environments. Although monomers A and B formed dimeric spikes (A-B dimers) around each of the five-fold axes, two two-fold related C monomers formed a spike (C-C dimer) at each of the icosahedral two-fold axes (Fig. 2B). The surface lattices of ORF2 proteins in HEV Tϭ3 VLP were similar to the capsid arrangement of caliciviruses. When compared with the A-B dimer, the morphology of the HEV C-C dimer was less well defined, perhaps due to flexibility in the angle of the protruding domain toward the icosahedral shell.
The density map of the Tϭ3 VLP displayed four discrete domains, designated from the outside inward as P, M, S, and N, on a section 52 Å from the equatorial plane (Fig. 2C). The density profile of the P, M, and S domains displayed less variation from that observed in Tϭ1 HEV-VLP, and the docking of the crystal structure of the Tϭ1 PORF2 protein to the density map of Tϭ3 HEV-VLP showed a very good agreement between the two structures (Fig. 2D). The docking positioned N-terminal tail of the PORF2 protein at the capsid inner surface aligned well with the density linker in Tϭ3 VLP (Fig. 2D). The linker density served as a tag to connect the N domain with the icosahedral capsid, indicating the FIGURE 2. Three-dimensional structure of HEV T‫3؍‬ VLP. A, overall structure of the large HEV-VLP reveals the Tϭ3 icosahedral lattice of the ORF2 proteins. One icosahedral facet is defined as the triangular area within the three adjacent five-fold axes. B, there are two unique dimeric ORF2 spikes on the HEV Tϭ3 VLP surface. The A-B dimer is located around the five-fold axis, and the C-C dimer is located at the two-fold axis. C, HEV Tϭ3 VLP has a radius of 205 Å and contains a low density cavity with a radius of 85 Å in the particle center. The distribution of the cryo-EM density revealed four ORF2 domains, P, M, S, and N, at 50 Å from the equatorial section. D, the crystal structure of the HEV subunit from Tϭ1 VLP docks well with the cryo-EM density in the shell region of HEV Tϭ3 VLP, with the N-terminal loop pointing toward the center.
location of the N-terminal 111 amino acids of the ORF2 protein in Tϭ3 HEV-VLP.
Crystal Structure of the Genotype-1 Tϭ1 HEV-VLP-The crystal structure of the truncated genotype-1 capsid protein (PORF2, containing residues 112-608) can be separated into three domains, S, M, and P, with a less resolved region covering residues 555-560. The S domain formed by residues 118 -317 folds into a classical eight-stranded ␤-barrel with a jelly roll motif (Fig. 3A), as observed in many Tϭ3 viral capsid proteins (26,27). Uniquely, three additional short ␣-helixes were observed in the S domain between strands E and F and strands G and H. The capsid shell was mainly stabilized by intersubunit interactions between the S domains. The folded M domain, consisting of residues 318 -451, was a twisted antiparallel ␤-sheet with an ␣-helix between the BЈ and CЈ strands (Fig. 3A). The P domain, composed of residues 452-606, folded into a ␤-barrel composed of antiparallel ␤-sheets, FЉAЉBbЉ and BaЉEЉDЉCЉ (Fig. 3A), and was connected with the M domain through a long proline-rich hinge (PTPSPAPSRP of residues 452-461) (Fig. 3A). Although both the M and the P domains existed above the S domain, the protruding spikes in the HEV cryo-EM map contain only the P domain density, which is a clear difference to those caliciviruses (supplemental Fig. 3). The PORF2 dimers have the largest buried surface area between monomers (5,900 Å 2 ) mainly due to the interface between P domains (Fig. 3B). The buried surface area is 3,400 and 1,600 Å 2 for the two adjacent PORF2 subunits around a three-fold axis and a five-fold axis, respectively (Fig. 3B). Moreover, the buried surface area of three molecules around a three-fold axis (9,500 Å 2 ) is much wider than that around a five-fold axis (4,700 Å 2 ).
Sequence alignment of genotype-1 PORF2 with the sequences of genotype-3 (10) and genotype-4 (11) revealed that the S domain is the most conserved region among HEV genotypes, whereas greater divergence was seen in the N-terminal region (supplemental Fig. 4A). Among the solved structures, genotype-3 appeared flexible at the N-terminal end and was 11 amino acids shorter than the others, whereas the structure deviation from genotype-1 PORF2 is very small (total root mean square deviation is 0.62 for the 472 equivalent amino acids) (supplemental Fig. 4B). Because amino acids 118 -129 play an important role in bridging the N-domain to the S-domain in Tϭ3 VLP and serve as a docking register, we used the crystal structure of genotype-1 to decipher the Tϭ3 cryo-EM density map.

Consistent Interdimeric Interactions between Tϭ3 and Tϭ1
HEV-VLPs-To understand the mechanism of ORF2 protein transition between Tϭ1 and Tϭ3 assemblies, we docked the Tϭ1 decamer and hexamer into the Tϭ3 cryo-EM density map. The decamer of Tϭ1 VLP consisted of 10 adjacent PORF2 monomers corresponding to five dimers around a five-fold axis, whereas the Tϭ1 hexamer corresponded to three adjacent dimers around a three-fold axis. Unlike the hexamer, the coordinates of the PORF2 decamer fitted very well with the curvature of the Tϭ3 density map at the five-fold vertex (Fig. 4A) and with the domain separation (Fig.  4B). The curvature of Tϭ3 capsid at the three-fold axis did not agree with the coordinates of the PORF2 hexamer as one of the dimers appeared to be sticking out of the cryo-EM density map (data not shown). Besides, the orientation of the P domain of the C-C dimer relative to its M/S domains was 90°different from that of the A-B dimer (Fig. 5). This suggests that the molecular interactions among A-B dimers in the Tϭ3 icosahedron are consistent with the dimer-dimer interactions in the Tϭ1 icosahedral assembly, whereas the interaction between the A-B dimer and C-C dimer is unique to the Tϭ3 assembly.
In Vitro Reassembly of the ORF2 Protein-To understand the role of ORF2 decamer in Tϭ3 VLP assembly, we analyzed the self-assembly process of HEV-VLP in vitro. A combination of chelating (EDTA) and reducing (DTT) agents was found to disassemble Tϭ3 VLP in a high alkaline environment (pH 10) without denaturating the ORF2 protein (data not shown). The addition of 20 mM CaCl 2 into the disassembly solution led to the association of the ORF2 dimers into star-shaped complexes, and no refolded VLP was found (Fig. 4C). When we examined the star-shaped complexes, we found that the distance between two opposite vertices was ϳ18 nm, close to the diameter of TMV (Fig. 4D). This size was consistent with that measured from PORF2 decamers. Thus, the star-shaped complexes resembled not only the appearance but also the size of the ORF2 One subunit is colored red, and the other subunit is colored according to its domain structure (blue, S domain; green, M domain; orange, P domain). C, surface areas that buried at the interfaces between two adjacent subunits are overlapped with a PORF2 hexamer (left) and at the PORF2 dimeric interface (right). One icosahedral facet is defined as the triangular area within the three adjacent five-fold axes.
decamer (a pentamer of dimers). Although the overall buried surface area around the three-fold axis is larger than that around the five-fold axis, we did not find any complexes that could fit with the PORF2 hexamer.
The in vitro disassembly and reassembly suggested that other factors than ORF2 protein contribute to Tϭ3 VLP assembly. Considering the electropositivity of the ORF2 N-terminal 111 amino acids, we performed nucleic acid extraction from both the Tϭ3 and the Tϭ1 VLPs. Electrophoresis results demonstrated the presence of nucleic acids in the Tϭ3 extract, whereas the Tϭ1 VLP extract was negative for nucleic acids (Fig. 4E). The extracted nucleic acids were sensitive to RNase treatment and resistant to DNase treatment. To further characterize the nucleic acid, we performed Northern blotting according to the protocol described under "Experimental Procedures." An RNA band about 2 kb in size was detected from large particles, whereas no band was detected from the small particles and the control norovirus-like particles (Fig. 4F). This result is consistent with the VLP profiles observed from the cryo-electron micrographs and further indicated that the Tϭ3 particles encapsulated the HEV gene encoding the ORF2 protein.

DISCUSSION
Hepatitis E virus is a human pathogen that causes acute liver failure. Like other hepatitis viruses, HEV cannot be propagated with currently available cell culture techniques. The capsid protein of genotype-3 HEV can be expressed in insect cells as PORF2 protein, including amino acids 112-608 that self-assemble into Tϭ1 VLP, and as ORF2 protein, including amino acids 14 -608 that form Tϭ3 VLP.
The crystal structures of PORF2 revealed three functional domains, S, M, and P, and the function of each domain constrained its sequence flexibility. The S domain formed an icosahedral shell that served as the base for arranging M and P domains; hence, the subunit surface should be highly conserved among genotypes. Sequence alignment agreed very well with this function, identifying the S domain as the most conserved region among HEV genotypes (28). The P domain serves as the putative binding site for both neutralizing antibody and cellular receptor (29) and contains 19 divergent amino acids across four genotypes (supplemental Fig. 4). Only nine of these amino acids were exposed at the surface of the P domain. Inspection of the binding footprint of antibodies on the cryo-EM density map indicated that only one amino acid was buried within the antibody-binding interface 4 . This explains why the HEV serotype is non-divergent despite sequence variation among HEV genotypes. The direct correlation between sequence variability and domain functionality may be necessary for the HEV capsid to carry multiple functions and to ensure error-free assembly. It also explains why the transition of HEV-VLP from the Tϭ3 to the Tϭ1 lattice does not disturb its antigenicity and why Tϭ1 VLP can be disassembled and reassembled in vitro to carry foreign antigenic epitopes (30) or DNA plasmids (31).
The Tϭ3 HEV-VLP has a similar morphology to that of calicivirus; however, the crystal structures of PORF2 revealed a distinctive M domain arrangement, although the folding of the HEV M domain is similar to the folding of the P1 domain in caliciviruses (supplemental Fig. 3). In HEV, the P domain is located at the C-terminal end of the M domain, whereas the P2 domain of caliciviruses is inserted into the P1 domain at the region between the AЈ and BЈ strands (32,33). Furthermore, the M domain of HEV interacts strongly with the S domain and connects to the P domain via a long proline-rich hinge, whereas the P1 domain in caliciviruses is a subdomain of the protrusion spike (supplemental Fig. 3). This seems to have an impact on VLP stability; the spike of the HEV C-subunits appeared weakly defined when compared with that in the A-B dimer, whereas the spike of the Norwalk virus C-subunit appeared rigid and similar to that in the A-B dimer in the cryo-EM structure (34). Additionally, deletion of the N-terminal positively charged amino acids from the Norwalk virus capsid protein does not induce Tϭ1 VLP because the Norwalk virus capsid protein only contains a short N-terminal tail of 20 amino acids (35).The HEV C-C dimer is profoundly different from the HEV A-B dimer in the orientation of the P domain relative to the M/S domain (Fig. 5). Conformational difference between the A-B dimer and C-C dimer has been reported earlier Tϭ3 viruses. In tomato bushy stunt virus, binding of RNA plays an important role to differentiate the C-C dimer from the A-B dimer. The N-terminal arm of the C-C dimer is well ordered and interacts with the RNA genome, whereas the A-B dimer is disordered and free from RNA interactions (36). In the flock house virus, the C-C dimeric contact acquires a flat conformation to accom-modate the RNA duplex, whereas the A-B dimer is in a bent conformation and involves no RNA (37). The different orientation observed between the HEV C-C dimer and A-B dimer may result from the difference in RNA occupancy. The A-B dimers do not interact with RNA and have a bent conformation. As a result, the angled contact of the M/S domains accommodated the proline-rich hinge within the V-shaped cleft, similar to that in the Tϭ1 VLP, thus solidifying the orientation of the P domain (Fig. 5C). In contrast, the contact with the RNA led the C-C dimer to a flat conformation that pushes the hinge out of the cleft. Thus, the P domain in the C-C dimer is flexible and could take a 90°rotation from the orientation in the A-B dimer (Fig. 5D).
The native HEV capsid was predicted to possess Tϭ3 icosahedral symmetry (8,11), and Guu et al. (11) suggested that such Tϭ3 capsids require the dimer to have both flat and inwardly bent conformation at quasi-two-fold and icosahedral two-fold positions, respectively. Here, with detailed features, our cryo-EM structure of Tϭ3 VLP provided direct observation of the two unique conformations between the A/B and C/C dimer at the foothold of S-domain contact but also in the arrangement of the protruding P-domain.
It is suggested by molecular simulation that Tϭ3 icosahedral capsid assembly utilizes a mechanism in which preformed aggregates of intermediates combine in contrast to the formation of the Tϭ1 icosahedral capsid that includes the addition of predominately monomers (38). The ORF2 decamer is therefore the assembly intermediate of Tϭ3 HEV capsid and is located at each of the five-fold vertexes. The appearance of a hexameric ring at icosahedral three-fold positions is the critical step in Tϭ3 capsid assembly and depends on the C-C dimer. The in vitro disassembly and reassembly also indicates the involvement of an extrinsic factor other than the ORF2 protein in the assembly of Tϭ3 VLP and the C-C dimer is in a flat conformation that is concomitant with RNA binding. The induction of C-C conformation has been reported with bacteriophage MS2, where the complete assembly of capsid requires the presence of synthetic RNA fragment (39). Therefore, the pentamer of dimer (Fig. 4C) may be the assembly intermediate that is common for the Tϭ3 virus. The interaction of RNA with the N-terminal end of ORF2 is the driving force leading the C-C dimer to the flat formation and ultimately full capsid formation through the integration of 30 copies of C-C dimers with 12 copies of A-B decamers (Fig. 6).
The existence of the N-terminal amino acids 14 -111 prevents ORF2 proteins from forming Tϭ1 VLP. The capsid of Tϭ1 VLP encloses a central cavity with a volume allowing a maximum of 55 additional residues on each copy of PORF2 protein, if the average protein density is considered to be 1.30 g/ml. The maximum diameter of the central cavity is about 340 Å for the Tϭ3 VLP and thus is sufficient to accommodate both the HEV genome and the ORF2 N-terminal domains. By characterizing the size and the sequence of the encapsidated RNA, we found that the Tϭ3 HEV VLP selectively encapsidated the RNA fragment that encodes the ORF2 protein. Thus, it is very possible that the native HEV capsid is the Tϭ3 icosahedron. There, the encapsulated genomic RNA may play a direct role in the assembly of the HEV infectious virion. Our data demonstrated here that HEV was different from caliciviruses in its assembly pathway, protein domain arrangement, and genome organization, although both viruses are Tϭ3 icosahedral particles with dimeric spikes. Hepatitis E virus showed a high similarity to some plant viruses in its assembly pathway: the utilization of a long electropositive N-terminal domain to interact with genomic RNA. Although the evolutionary origin for such similarities requires further investigation, our data place the HEV structure in a unique position, deviating from that of human caliciviruses and approaching that of Tϭ3 small plant viruses.