Nucleocapsid and Glycoprotein Organization in an Enveloped Virus

Cryo-EM and Image Reconstruction

Electron micrographs of RR embedded in vitreous ice (Figure 1) provided a set of 68 particle images from which a 3D reconstruction was computed at 25 Å . The correct hand of the structure was determined from experiment s in which RR was tilted in the microscope, and resulting images were compared by cross-correlation with corresponding projections of left- and right-hand versions of the reconstructions. The improved quality of the RR map, when compared to maps obtained for SIN (Fuller, 1987; Paredes et al., 1993a; N. H. O. and T. J. S., unpublished data) or SF (Vogel et al., 1986; Vénien-Bryan and Fuller, 1994; Kenney et al., 1994), indicates that the icosahedral symmetry of RR is better preserved in the specimen. The difference might be related to RR being propagated in mosquito cells rather than mammalian cells. Alternatively, RR may be more stable during specimen preparation for cryo-EM.

Surface-shaded representations of the RR reconstruction, viewed along twofold and threefold axes, clearly show the threefold nature of the surface spikes and the bilobal nature of each of the spike petals (Figure 2). The surface of the virus is almost exclusively covered by protein, except for prominent holes (50 Å long × 40 Å wide) that expose the surface of the lipid bilayer at the icosahedral twofold axes, 20 Å diameter holes at the fivefold axes, and 15 Å diameter holes at the base of each spike. The organization of the spikes within the T = 4 icosahedral lattice (Figure 2c) is discussed in greater detail below.

Equatorial sections of the reconstruction depict the layered structure of the virion (Figure 3). The density map on its own cannot differentiate between nucleic acid, protein, and lipid components of the virion, but low-angle neutron (Jacrot, 1987) and X-ray (Harrison et al., 1971, 1974; Stubbs et al., 1991) scattering results permit us to ascribe specific features to these components. The scattering data suggested that the glycoproteins are located between radii of 350 Å and 260 Å and that the inner and outer surfaces of the membrane are at 210 Å and between 250 Å (Jacrot, 1987) and 258 Å radius (Harrison et al., 1971), respectively (Figure 3a). Radii between 210 Å and 170 Å mainly contain protein, whereas a mixture of protein and RNA occurs at lower radii. The low-angle scattering results permitted us to identify features in the RR reconstruction and ascribe these to the various viral components.

The El-E2 Glycoprotein Spike

The RR spikes (Figure 4) are similar in size to those previously reported for SF and SIN (Vogel et al., 1986; Fuller, 1987; Paredes et al., 1993a), although they are seen in more detail. They extend 95 Å radially outwards from 255–350 Å. The glycoprotein spike head group has a large, flower-like morphology with bilobal petals the surfaces of which lie within 15 Å of the surfaces of neighboring spikes (see Figure 2; Figures 4a and 4b). The head group has a diameter of ~ 125 Å. The stem immediately below the head group is ~ 85 Å in diameter with a roughly triangular cross-section and a 25 Å hole in the center (Figure 4d, cross-section B), demonstrating the trimeric clustering of the spike glycoproteins. At radii less than 290 Å, the glycoproteins skew outwards as separate E1–E2 heterodimers to a diameter of ~ 125 Å. At a viral radius of 290 Å, all 240 heterodimers are separated, and each one then appears to traverse the membrane to form a specific interaction with a nucleocapsid monomer (see Figure 3). Separation of the individual heterodimers from the stem at the base of the spike opens up lateral holes, about 15 Å in diameter, creating a large cavity that extends 30 Å radially from the membrane to the base of the head group. The current data do not discriminate between the E1 and E2 glycoproteins, although recent studies of alphaviruses bound with neutralizing antibody to E2 localize an epitope to the larger of the two lobes in the head group at the most distal portion of the spike (unpublished data).

Spikes on the icosahedral threefold axes contact monomers in three adjacent nucleocapsid hexamers, whereas spikes on the quasi-threefold axes contact two hexamers and one pentamer. However, at the current resolution, there is little difference in the structures of the two types of spikes (Figures 4a, 4b, and 4c). Nevertheless, the different environments at the base portion of the spikes and the potential different accessabilities of solvent and solute molecules to the interior regions of these portions of the spikes are consistent with the observations that the two spikes exhibit different sensitivity to detergent extraction (Helenius and Kartenbeck, 1980; Vénien-Bryan and Fuller, 1994). This differential extraction may be due to the greater accessibility of detergent to spikes on the threefold axes via the large holes at the three adjacent icosahedral twofold axes. In contrast, the spikes situated on the quasi-threefold axes are surrounded only by two large holes on the icosahedral twofold axes and one much smaller hole on the icosahedral fivefold axes.

Secondary structure predictions (Rost, 1993), based on the amino acid sequences of El and E2, suggest that both are devoid of α-helices, except in the hydrophobic regions near the C-termini. Presumably, the α-helices penetrate the lipid bilayer. The absence of helical structure in the alphavirus spikes contrasts with the largely α-helical stems of the glycoprotein spikes found in influenza virus (Bullough et al., 1994).

The Nucleocapsid Shell

The component of the nucleocapsid structure between 210 Å and 170 Å radius is organized as pentamers about the icosahedral fivefold axes and is organized as hexamers about the icosahedral twofold axes (Figures 5a and 5b). This corresponds to 240 subunits that associate in a classic quasiequivalent fashion (Caspar and Klug, 1962). At radii greater than 170 Å, the monomer subunits form contacts only within the clearly defined capsorneres (pentamers and hexamers). These capsomeres “rest” on a continuous core of density below 170 Å, radius, consisting of both protein and RNA, as suggested by neutron scattering results (Jacrot, 1987). Putative protein–RNA interactions are responsible for maintaining the core structure when virions are disassembled. Viral RNA inside the core is sensitive to ribonucleases, thus suggesting it lies at the core surface exposed at sites between the pentamers and hexamers or at the center of these units (Kääriäinen and Söderlund, 1971). The density inside the ~ 170 Å radius must consist of the genomic RNA and the basic residues at the N-terminus of the capsid protein. Association of single-strand RNA with disordered, basic N-terminal domains of capsid proteins is prevalent in plant viruses (Harrison, 1984; Rossmann and Johnson, 1989).

The concept of quasisymmetry dictates similar monomer–monomer contacts in the formation of pentameric and hexameric capsomeres (Caspar and Klug, 1962) and, as a consequence, a quasi-twofold axis relates pentamer and hexamer subunits (A₁ and C₁ in Figure 5b). However, attempts to fit the crystallographic SCP dimer structure (Choi et al., 1991) into the 3D map of the core at the observed T = 4 quasi-twofold axes (dimers A₁C₁ and B₁D₂ in Figure 5b) showed that the dimer is not present in the core. The dimer is too short in length and cannot fill the distance representing the B₁ and D₂ subunits (Figure 5c). Furthermore, a model based on the packing of core dimers would require considerable density in the region between capsomeres, but such density is clearly absent in the reconstruction (Figure 5c). The absence of the crystallographic dimer in the virion is consistent with site-directed mutagenesis of the residues involved in stabilizing the dimer. Amino acid substitution of these residues resulted in production of mostly wild-type viruses and suggested that the dimer may not be biologically relevant (K. E. Owen and R. J. K., unpublished data).

Instead of using the crystallographically observed dimer structure to fit into the RR virus density, the structure of a single C-terminal domain of the SCP monomer was fit into the nucleocapsid portion of the reconstructed density. The latter was isolated from the whole RR virion density by zeroing all density outside a radius of 210 Å. A 6D search (three angles and three positions) produced a unique fit of all the atoms of the ordered region of the SCP atomic structure (residues 114–264) to the high density features of the EM reconstruction (Figure 5d). Sequence identity between RR and SF in their C-terminal domains is 93%, whereas RR and SIN are 68% identical (Dalgarno et al., 1983). The atomic structures of the SIN (Choi et al., 1991) and SF (H.-K. C., G. Wengler, and M. G. R., unpublished data) nucleocapsid proteins are similar, with a root mean square (rms) deviation of 0.7 Å in the Cα atoms. Hence, at the limited resolution of the EM reconstruction, the core protein structures of RR and SIN are likely to be indistinguishable.

The core proteins within pentamers and hexamers interdigitate each other with little steric interference between residues in neighboring monomers, and they form a much more extensive intersubunit interface compared with that observed in the crystallographic dimer. The monomer–monomer interface in the fitted structure involves contacts between residues 179– ~ 191 and 219–227 on one subunit and residues 120– ~ 126 and 212 on the other.

This core model concurs with biochemical data that identify the spatial orientation of specific residues within the intact core. Iodination of isolated cores showed that only 1 of the 4 possible tyrosine residues (Tyr-180) was accessible and, therefore, was presumed to be on the outside of the core particle (Coombs and Brown, 1987b), in agreement with the model. In addition, cross-linking studies by K. Coombs and D. T. Brown (unpublished data) suggest that lysine residues (Lys-250) on adjacent monomers are ~ 12 Å apart. Indeed, in agreement with the cross-linking data, the core model places Lys-250 on subunit B₁ of the hexamer ~ 12 Å from Lys-250 on monomer A₁ of the pentamer (Figure 5d; asterisks mark the lysine residues). The N-terminus of the ordered domain of the SCP (residue 114) lies exposed on the inner wall of the capsomere at a radius of 192 Å and points outwards. Presumably, the polypeptide reverses its direction near this point because the basic, N-terminal domain is associated with the RNA.

Membrane Structure and Glycoprotein-Nucleocapsid Interactions

The bilayer membrane is seen in cross-section (see Figure 3a) as a distinct pair of concentric, electron-dense features (phosphate head groups), with a center-to-center separation of ~ 32 Å and a total thickness of ~ 45 Å (radius from 255–210 Å), which sandwiches a lower density region (lipid hydrocarbon chains). These features are in complete agreement with small-angle X-ray solution scattering results (Harrison et al., 1971, 1974). The membrane adopts a roughly spherical shape, except where it bulges inward by 5–10 Å in the regions centered above the nucleocapsid capsomeres (see Figure 3a). At this point, the bilayer maintains its 32 Å head group separation.

Nucleocapsid–glycoprotein associations are involved during virus assembly prior to budding and also during the uncoating events that follow virus attachment and penetration. The E1 and E2 giycoproteins both contain transmembrane domains. E2 has a 31-residue cytoplasmic domain that interacts with the nucleocapsid (Garoff and Simons, 1974; Lopez et al., 1994), whereas E1 contains only two residues (Arg-437 and Arg-438) on the inner surface of the membrane, which are dispensible for assembly and replication in SF (Barth et al., 1992). Therefore, E1 probably plays little if any role in forming the associations between spikes and the nucleocapsid core.

The RR reconstruction permits the unambiguous assignment of specific spike proteins to specific nucleocapsid protein monomers (see Figures 3a and 3b; Figure 6a). Each pentamer of the nucleocapsid is associated with five El–E2 heterodimers that originate in five separate spikes, each of which is located on a quasi-threefold axis. Each nucleocapsid hexamer is associated with six spikes, four on quasi-threefold axes and two on icosahedral threefold axes (Figure 6a). Thus, the El–E2 heterodimers in the 20 threefold spikes only associate with hexamer core proteins, whereas one of the El–E2 heterodimers in each of the 60 quasi-threefold spikes associates with a pentamer protein (Figure 6a).

A close-up, cross-section view of the RR reconstruction, contoured at three different density levels, illustrates the overall organization within the multilayered RR structure and shows the putative association between the El–E2 heterodimer and the core protein (Figure 6b). Each El–E2 heterodimer (blue) traverses the membrane immediately above the center of each capsid monomer (yellow), where the membrane appears to be pinched and adopts a smaller thickness, presumably owing to the presence of protein on both of its sides. Here, Asn-172 in the fitted SCP model lies at the inner surface of the membrane and is juxtaposed with the glycoprotein on the external membrane surface. Thus, residues near Asn-172 may be involved in binding the 31 C-terminal, cytoplasmic residues of E2 (Helenius and Kartenbeck, 1980; Metsikkö and Garoff, 1990). Asn-172 is one of the residues believed to be involved in E2 binding on the basis of sequence comparisons between western equine encephalitis virus, eastern equine encephalitis virus, and SIN (Hahn et al., 1988). The interaction of the cytoplasmic domain of E2 with the nucleocapsid appears to be altered by site-specific mutation of Tyr-180 and Glu-183 in the nucleocapsid of SIN (Lee and Brown, 1994). This is consistent with Tyr-180 being accessible on the nucleocapsid surface and with its being near the apparent site of penetration of the membrane by the glycoprotein spike. The charge distribution on the surface of the pentameric and hexameric capsomeres is noteworthy. The outer rim of the capsomere close to the membrane is positively charged, which may complement the negatively charged head groups of the bilayer. The surface closest to the RNA is negatively charged and suggests the interface between capsomere and RNA maybe occupied by the basic, N-terminal domain of the nucleocapsid protein.

The capsid core associates with the glycoprotein spikes at the plasma membrane after the spikes have independently assembled in the endoplasmic reticulum and Golgi and after cores have assembled in the cytoplasm (Strauss and Strauss, 1994). The T = 4 arrangement of the nucleocapsid proteins might direct 80 El–E2 trimers to assemble as a T = 4 lattice as a consequence of the one-to-one association between capsid monomers and glycoprotein heterodimers. In agreement with previous observations that virions are devoid of host cell proteins (Strauss, 1978), the presence of 80 spikes leaves little room for any additional proteins.

Virus Preparation

The T48 strain of RR, rescued from the full-length cDNA clone pRR64, was used as the source of virus (Kuhn et al., 1991). Mosquito cell line C6/36 (A. albopictus) was used for virus propagation. Cells were grown at 30°C in roller bottles using Eagle minimal essential medium supplemented with 10% fetal calf serum, and they were infected with virus at a multiplicity of one. Following a 36 hr incubation, cell supernatant was harvested, and virus was concentrated by precipitation in 10% PEG and 0.5 M NaCI. The virus was then resuspended and purified by sucrose density centrifugation.

Cryo-EM

The cryo-EM and image analysis procedures that were used to obtain the 3D reconstructions of RR were essentially similar to those previously reported (Baker et al., 1988; Cheng et al., 1994). The sample was maintained at near liquid nitrogen temperature with a Gatan 626 cryotransfer stage in a Philips EM420 transmission electron microscope. Micrographs were recorded under minimal dose conditions (~ 20 e^–/Ų), at an objective lens defocus of under 1 μm and at 80 kV, with an instrument magnification setting of 36,000×. Enantiomorphic information (handedness) is lost in the 2D images of the particles. To determine the correct hand, image pairs were recorded in the microscope of the same field. For the second image of a pair, the goniometer stage was tilted 5° relative to the first. Two enantiomorphic reconstructions were then computed from the untilted data, and projections of each were calculated based on the orientation parameters of each particle image. Each particle image of the tilted data was then cross-correlated with the corresponding projections of the two reconstructions of the untilted data. The reconstruction of the correct hand had an average correlation coefficient of 0.41 ± 0.07, compared with 0.26 ± 0.06 for the reconstruction with the incorrect hand. Micrographs were digitized at 25 μm intervals corresponding to spacing of 7.29 Å in the specimen. Bacteriophage ΦX174 was mixed with the RR sample and was used as an internal magnification calibration standard (Olson and Baker, 1989; Cheng et al., 1994). Particle image selection was performed as previously described (Baker et al., 1991; Cheng et al., 1992).

Image Analysis

Particle images (Figure 1) were screened using a model-based approach that helped determine the relative orientations and provided quantitative criteria to select the best set of self-consistent image data (Cheng et al., 1994). Initial estimates of the phase origins (particle centers) were made by means of a cross-correlation method, with a circular-averaged image from all the digitized particle projections serving as the reference (Olson and Baker, 1989). A preliminary, 35 Å resolution, 3D reconstruction was obtained from eight particle images, at a resolution of 35 Å, with common-lines and Fourier-Bessel procedures (Crowther, 1971; Fuller, 1987; Baker et al., 1988). The resulting model was used to refine further the orientation and origin parameters of the particle images. Projected images of the 3D model, representing all unique views of the structure in 1° angular increments, were used as a reference data base for determining the orientation of newly selected particles (T. S. B. and R. H. C., unpublished data). The orientation search procedure was optimized by correlating data that had been filtered in both real space (only data between radii of 110–350 Å were used) and reciprocal space (band pass between 1/25 Å and 1/90 Å). The orientation and origin of each image was refined by maximizing the correlation coefficient between that image and the corresponding projected view of an interim reconstruction. The quality of the reconstructions significantly improved when corrections were made to adjust for slight magnification variations in the individual images (Aldroubi et al., 1992). These methods, along with cross-common lines procedures, were used to select a data set of 68 particle images to compute the final 3D reconstruction. Complete icosahedral symmetry was imposed on the final density map (Fuller, 1987). The effective resolution was estimated to be ~ 25 Å, based on comparing independent reconstructions (Baker et al., 1991; Cheng et al., 1992).

Model Fitting

The known atomic structure of the C-terminal domain of the SCP, being homologous to RR core protein, was oriented and positioned into each of the four independent monomeric sites of the image reconstruction that was truncated at a radius of 210 Å. Two sequential procedures were used to determine the best fit of the monomer into the EM density, An initial, coarse 3D orientational search was followed by a systematic “climb” to obtain the best fit using six refinable parameters (three orientational and three positional). The quality of fit was defined either as the average electron density or as the rms electron density taken over all the predicted atomic sites in the independent monomers. The density at each atom position was determined by an 8-point interpolation in the RR map, The number of steric clashes between neighboring monomers was assessed in terms of the number of grid points closest to each atom of one monomer that were also closest to atoms in a neighboring monomer. No attempt was made to avoid such clashes, but steric hindrance was checked graphically and found to be small for the better fits.

As a start, the center of the gravity of the SCP monomer was placed onto a designated approximate center of one of the electron density lumps (A₁ in Figure 5b) representing the RR core structure. Symmetry-related monomers within a pentamer were generated to give A₂, A₃, A₄, and A₅ by means of the nearest fivefold icosahedral axis. The monomer at A₁ was rotated into the equivalent monomer position C₁ by means of the nearest quasi-twofold axis assumed to be radial. Monomer C₁ was then reproduced to give C₁, D₁, B₆, C₆, D₆, and B₁ by virtue of the appropriate quasi-sixfold axis (Figure 5b; this sixfold axis is coincident with an icosahedral twofold axis). The position of the quasi-twofold axis is not known precisely and was, therefore, systematically adjusted to obtain the best fit. In principal, it should be necessary to refine separately the orientational and positional parameters of the four quasiequivalent monomers A₁, B₁, C₁, and D₁. However, to keep the number of refinable parameters to a minimum and, hence, to increase the significance of the resultant fit, the quasi-sixfold axis was treated as an exact sixfold axis. This gave six refinabie parameters for the monomer at A₁ (three positional and three orientational), alternated with the refinement of two parameters defining the quasi-twofold axis orientation, as well as the hand of the map relative to the known absolute hand of the monomer structure.

The search procedure started by rotating the atoms of the C-terminal domain of the SCP monomer from their standard orientation by the Eulerian angles (θ1, θ2, θ3). The center of gravity of the rotated monomer was placed onto the approximate center at A₁ in the electron density from which the complete pentameric and hexameric units were then generated. The search for the best fit was pursued over all Eulerian angles (0 ≤ θ_l < 2π; 0 ≤ θ₂ < π; 0 ≤ θ₃ < 2π) in increments of 30° or 20°, retaining the top 100 fits.

Each of the top fits was then subjected to exploration by “climbing” to the nearest best fit, while varying the six refinable parameters in turn. Eventually, the best fit was accepted when the angular increments had decreased from an initial 10° to a final 0.25° and when the positional increments had decreased from an initial 5 Å to a final 0.5 Å. Many of the final, best fits were the same. That is, although the climb started at a different point, the best orientation and position was found to be the same, independent of the starting parameters. The original top 100 fits, obtained by the complete Eulerian angle search, reduced itself down to about 10 distinct best fits. Two other criteria of quality of fit were, therefore, developed to differentiate them. A count was kept of atoms that were in steric collision between monomers (nclash) and of the number of atoms placed into negative density (–den). Although neither measurement was able to select clearly among the remaining best fits, a product of (nclash) × (–den) appeared to be the most sensitive quantity.

Complete explorations were performed for positions of the quasi- twofold axis using roughly 1° intervals for either hand (± 1) of the EM map. The best position of the quasi-twofold axis gave a clear preference for a unique solution, consistent with the experimentally determined hand of the EM reconstruction (Figure 5d).