Two distinct trimeric conformations of natively membrane-anchored full-length herpes simplex virus 1 glycoprotein B

To study full-length HSV-1 gB in a native membrane, it was displayed on cell-derived membrane vesicles. In these extracellular vesicles, proteins are natively anchored with correct topology in genuine biological membranes (18). Vesicles with full-length HSV-1 gB were collected from the culture medium of adherent mammalian cells at 48 h posttransfection with the gene for HSV-1 gB and then purified by differential centrifugation (details are provided in Materials and Methods), resulting in high yields of vesicles. The specific enrichment and the identity of the proteins incorporated were validated by SDS/PAGE, Western blot analysis (Fig. S1 A–C), MS using an exponentially modified protein abundance index (emPAI; Table S1), and immunogold labeling (Fig. S1 D and E). Under denaturing and reducing conditions, gB on the vesicles had an apparent molecular mass of ∼120 kDa, and under seminative conditions, it had an apparent molecular mass of ∼250 kDa (Fig. S1), consistent with the molecular mass observed for mature gB extracted from virions and infected cells (19). The emPAI not only verified the SDS/PAGE results that HSV-1 gB was the most abundant protein in the vesicle preparation but also, importantly, detected no other integral membrane proteins (Table S1). Immunogold labeling using a polyclonal anti-gB Ab readily detected the protein on the vesicle membranes (Fig. S1 D and E).

Electron cryomicroscopy (cryoEM) imaging of the gB-enriched vesicles showed that they were mainly spherical and studded with proteins, which appeared as discrete densities (Fig. 1A). At least two kinds of protein densities of different lengths and shapes were observed, being clearly discernible in cryoET slices (Fig. 1 B and C). Together with the MS results, these findings indicated that, for the first time to our knowledge, a conformation other than the postfusion trimer of gB had become amenable to further characterization.

The mean diameter of the vesicles was related to the predominant protein conformation on its surface (Fig. 1D). The majority (>80%) of the vesicles had a normal distribution of sizes, with a mean diameter of ∼110 nm. On those vesicles, the protein densities were roughly globular, with a height of ∼10 nm. A smaller fraction of the vesicles (<5%) had a smaller mean diameter, and the protein densities on the surface had a height of ∼16 nm (Fig. 1 A and B). On some of the vesicles, both types of densities were found.

Chemical cross-linking was used to determine the oligomeric state of gB on the vesicle membrane. The gB vesicles were incubated with either the membrane-impermeable cross-linker bis[sulfosuccinimidyl] suberate, which has a spacer arm length of 11.4 Å, or the membrane-permeable cross-linkers disuccinimidyl tartrate and 1,5-difluoro-2,4- dinitrobenzene (DFDNB), which have a spacer arm length of 6.4 Å and 3.0 Å, respectively. All three of them cross-link primary amines (and thiol, imidazolyl, and phenolate groups, in the case of DFDNB). A band, corresponding to a gB trimer, was observed for all of the cross-linkers, albeit with different efficiency depending on the cross-linker concentration (Fig. S1C).

To characterize the two conformations of gB at higher resolution, we applied cryoET and averaging of tomographic subvolumes (details are provided in Materials and Methods). For both forms, subvolumes were manually picked at the vesicle surfaces. These volumes were then each iteratively aligned and averaged in a reference-free manner with 1,909 (short form) and 395 (long form) subvolumes contributing to the final averages (Figs. S2 and S3). The resulting 3D reconstructions of the short and long forms (Fig. 2) were determined to have a resolution of 23 Å and 31 Å, respectively, at the 0.143 threshold based on the gold-standard criterion (Fig. S4). The lower resolution of the long-form reconstruction is likely a result of a considerably smaller dataset. Although the two reconstructions were markedly different (Fig. 2), both showed clear threefold symmetry with an axis normal to the membrane. This symmetry was already apparent in the early steps of the refinement. During later steps, threefold symmetry was applied (Figs. S2 and S3).

The short and more condensed form had an ectodomain measuring 10 nm in height and 8 nm in width (Fig. 2 A–C). It was anchored to the membrane by three densities ∼6 nm apart, forming a tripod (Fig. 2B), whereas on the inside of the membrane, a central density ∼1.5 nm in height and ∼3 nm in width was observed (Fig. 2C). The ectodomain had its maximal lateral extension at a height of 4 nm from the membrane. At this height, neighboring trimers contacted each other, forming a pseudolattice (Fig. S5) with variable neighbor occupancies and relative position inconsistencies. Most distal to the membrane were three oblong densities, with their tips pointing away from the membrane in an angle to the bilayer and creating a central cavity between them (Fig. 2 A and B).

The long-form ectodomain was ∼16 nm in height and was anchored to the membrane through a single central density (Fig. 2 D and E). This form was highly similar in shape to the postfusion crystal structure of gB (8, 10, 13, 14) and the 3D reconstruction of the ectodomain bound to liposomes (20). This conformation appeared predominantly on the smaller vesicles, indicating that the postfusion form of gB prefers higher curvature, as observed previously on HSV-1 virions (20). The trimeric postfusion crystal structure could be fitted as one rigid body in the EM map (Fig. 2F).

To derive the model of the short-form ectodomain, we applied a hierarchical rigid domain-fitting approach based on large ensembles generated by four independent fitting methods. The postfusion crystal structure of HSV-1 gB defined five domains (8, 10, 13, 14). Domain I is the largest and most distinct in shape, containing the FLs that are vital for inducing membrane merging. It is connected to the globular domain II by two short linkers. To position domain I in the map, we generated independent ensembles using four different global search rigid domain-fitting methods (Fig. S6). Ranking, clustering, and filtering of these ensembles resulted in two groups of possible orientations for domain I, both at the membrane distal part of the EM map but oriented roughly 180° relative to each other (Fig. S6). Applying domain II distance constraints allowed us to discard one group of solutions, leaving only those solutions with the FLs pointing away from the membrane (Fig. 3 and Movie S1).

The positioning of domain I most distal to the membrane and the FLs pointing away was unexpected, representing a fundamentally different arrangement from the arrangement found in the membrane-bound postfusion trimer (20) (Fig. 2F) that presents the molecule in a conformation more prone for interaction with a target membrane.

The most distinct difference between the condensed form of gB and the postfusion one is the FLs between protomers being ∼5 nm apart. Notably, the orientation of domains I and II relative to the threefold symmetry axis between the two conformations is different, thus leading to a different surface of domains I and II being exposed (Fig. 4 and Movie S2). Moreover, glycosylation site N398 is on the surface and directly facing neighboring gB trimers, suggesting that it might be involved in interprotein interactions (Fig. 4 and Fig. S5).

Studies of Abs that neutralize HSV-1 infection have identified several antigenic sites on gB (21, 22). A major antigenic site on gB is on domain II, as is the case for the rhabdovirus glycoprotein (23). This domain presents an extensive exposed surface in the condensed conformation model, with epitopes for the neutralizing Abs H1781 (residues 454–473) and H1838 (residues 390–410) being exposed. Of particular interest is a recently raised anti-FL2 polyclonal Ab (residues 258–260). Reported data for this Ab indicated that the FLs are accessible on the virus (22). Although this observation could not be reconciled with the previously available postfusion structure of gB, these data are consistent with the condensed conformation on the membrane with the FL exposed presented here. Taken together, all of the available data on neutralizing Abs mapped to domains I and II, which could not be explained previously on the basis of the gB postfusion structure alone, are consistent with the new conformation of gB described here and support its assignment as a prefusion conformation or an intermediate.

Extensive mutational analyses of HSV-1 gB have been previously reported. Specifically, random linker-insertion mutagenesis (24) and fluorescence protein insertion mutagenesis (25) identified positions of the protein that tolerated insertions, thus permitting processing and transport of mutated gB to the plasma membrane and positions that did not tolerate insertions. All of the reported positions tolerating insertions are on the surface of the condensed gB conformation model (Fig. 4A). The successful insertion at position A261 of FL2, and its subsequent expression on cell surfaces (24), is consistent with the FLs being accessible in the metastable state of gB before fusion and in agreement with the antigenic data mentioned above. Furthermore, the two residues I185 and E187 in the β-sheet leading to FL1 are surface-exposed in the condensed conformation (Fig. 4A), thus permitting the reported expression of insertion mutations in these positions (24). However, these residues are buried in the postfusion conformation (Fig. 4B), making these mutants nonfunctional (24) (i.e., prohibiting the conformational change to the latter). Linker insertions can potentially interfere in any level of protein folding and processing (i.e., secondary structure, tertiary structure, quaternary structure). Of the many positions not tolerating insertions, a few are intriguing because they could not be explained by the postfusion structure alone but can now be explained by the swap in exposed surfaces observed between the two conformations (Fig. 4). For example, the insertion at position A334 was expected to be tolerated based on the postfusion structure because it is in a big loop on the surface; however, in the condensed form, this loop is much more centrally positioned and faces a helix from an adjacent protomer that, likewise, did not tolerate insertions (positions 193–203). Together, these experimental data further validate the condensed conformation as relevant and being significantly different in domain topology from the postfusion form.

The core of the gB postfusion conformation is composed of a trimeric coiled coil (domain III) with the long C-terminal arms (domain V) packed against it in an antiparallel manner, a feature that is typical for class I viral fusion proteins (26). Domain IV is a globular domain located between domains III and V, and it is known to be significantly repositioned between the pre- and postfusion conformations of vesicular stomatitis virus glycoprotein (VSV G) (27). Domain IV also shows high variability between class III fusion proteins and has intrinsic flexibility (15). To consider the position of this domain in the map, a large ensemble was again generated, scored, clustered, and filtered (Fig. S8). The results of these analyses indicated that domain IV is located in the map’s central density area (domain IV, colored orange in Fig. 3) next to domain II and at the same height from the membrane. However, the specific orientation was not fully conclusive (Fig. S8). Domain III could not be confidently fitted in our map. Although the limited resolution is a partial cause, the expected substantial refolding of domain III based on the VSV G pre- and postfusion structures presents a major obstacle for placing them in the map. However, because they are expected to form the trimerization interface, they are most likely to be centrally located in the map, a possibility that our model accommodates (Fig. 3, yellow densities and Movie S1). Finally, because the C terminus of the ectodomain consists of domain V and the as yet structurally uncharacterized highly hydrophobic membrane proximal region, we assigned them to the density forming the tripod (Fig. 3, red densities and Movie S1). Together, domain fitting and further domain assignments based on connectivity and previous information account very well for the overall volume of the short conformation map.

HSV-1 gB is the largest and most complex member of the class III viral fusion proteins (15). Unlike VSV G, it is part of a multicomponent fusion machinery and its transition from pre- to postfusion conformation is not assumed to be at least solely pH-dependent or reversible on the viral membrane. Despite the functional and structural differences observed between the postfusion forms of gB and VSV G, a comparison between the condensed conformation of gB and the prefusion form of VSV G (27) highlights some common features (Fig. S9). Both are about two-thirds the height of their postfusion forms but are wider, and the FLs of both are ∼5 nm apart.

The two conformational states of gB in the membrane reported here (Figs. 1 and 2), combined with previous knowledge, allow us to propose the following mechanism for how conformational changes in gB promote membrane fusion (Fig. 5). Upon receptor binding to gD, gH/gL triggers a prefusion form of gB to undergo a significant rearrangement. The short conformation reported here, stretching out from the membrane only two-thirds of the height compared with the postfusion form, likely facilitates interaction with the other components of the herpesvirus fusion machinery that share the same approximate height. Assuming that the trimer is preserved throughout the conformational change, the most likely route would be, concomitant to the formation of the coiled-coil core of domain III, a rotation of domain I combined with a petal-like “closing” movement. This movement leads to close proximity of the FLs from all three protomers (from an ∼5 nm pairwise distance to direct contact) and results in the formation of a hydrophobic fusion patch. Elevated by the coiled-coil formation of domain III, this fusion patch would contact the target membrane, which is held in a reachable distance by the receptor interaction of gD, and enable insertion into the outer leaflet of the bilayer (20) (Fig. 5), resulting in a gB trimer anchored by the transmembrane region on the viral membrane and with its FLs in the host cell membrane. Formation of the hairpin by zippering of the long arms of domain V along the trimer would conclude the conformational change and complete the membrane merging as suggested for class II viral fusion proteins (28).

Regardless whether the new conformation described here is a prefusion conformation or an intermediate, it is only the second viral class III protein to be reported as not being in a postfusion conformation (27), and, to our knowledge, the first to have been determined in detail fully intact and membrane-anchored. Both this context and resolution allowed the orientation of the protein, and, in particular, the position of the FLs to be deduced. This unexpected and crucial insight represents a significant breakthrough in understanding the class III viral fusion protein mechanism, and provides new opportunities for probing new functional interfaces in the complex herpesvirus fusion machinery and for structure-guided approaches to intervene in herpesvirus entry.

Vesicles enriched with HSV-1 gB in the membrane were produced as recently described, with small modifications in the protocol (18, 29). In brief, baby hamster kidney (BHK-21) cells were transiently transfected with the plasmid for gB [pPEP98 (30)] using Lipofectamine (Invitrogen). Transfected cells were cultured at 37 °C at 5% CO₂ in Glasgow minimum essential medium (GMEM), 20 mM Hepes (pH 7.4), and 2% (vol/vol) tryptose phosphate broth supplemented with 2% (vol/vol) FBS. After 24 h, the culture medium was changed to serum-free medium. Vesicles were collected and purified by differential centrifugation 48 h posttransfection as follows. The medium was collected and cleared from cell debris by centrifugation at 3,000 × g for 20 min and at 4 °C; vesicles were then pelleted through a 20% sucrose cushion at 100,000 × g and resuspended after careful removal of any supernatant and wall-adherent material in 25 mM Hepes (pH 7.4) and 130 mM NaCl. The vesicle preparations were independently repeated 15 times, each starting from a fresh tissue culture. For every vesicle preparation, at least two aliquots were analyzed by cryoEM.

Cross-linkers (Pierce) were dissolved to a stock concentration in 0.2 M triethanolamine (pH 8.0) or DMSO. For each reaction, 5 μL of gB vesicle sample was mixed with final concentrations of cross-linker at 5 mM, 2.5 mM, 1.25 mM, and 0 mM (control). Reactions were incubated for 30 min at room temperature and were quenched using SDS/PAGE sample buffer. The results were analyzed by SDS/ PAGE and silver-stained.

Microscopy was performed at 300 keV using a TF30 “Polara” electron microscope (FEI) equipped with a Quantum postcolumn energy filter (Gatan) operated in zero-loss imaging mode with a 20-eV energy-selecting slit. Images were recorded on a postfilter ∼4,000 × 4,000 K2-summit direct electron detector (Gatan) operated in counting mode with dose fractionation, with a calibrated pixel size of 0.23 nm at the specimen level. Tilt series were collected at 300 keV using SerialEM (31) at defocus ranges of −6 to −3 μm. During data collection, the autofocusing routine was iterated to achieve a very stable defocus through the tilt series with 100-nm accuracy. Tomographic reconstructions were calculated in IMOD using weighted back-projection (32). Defocus determination and phase flip according to the contrast transfer function were done in IMOD.

Particle subvolumes were picked manually and aligned and averaged using PEET 1.9 (33). Gold-standard Fourier shell correlation (FSC) calculation was performed by splitting the dataset into odd and even particles, and applying the same refinement procedure to both independently. The split averages were aligned and masked using a custom mask based on segmentation of the merged average. To avoid correlation between the masks, one average was masked with a hard-edged mask and the other with a soft-edged one. The FSC was calculated using Bsoft (34).

Details are provided in SI Materials and Methods. A detailed description of the model building protocol is also found in SI Materials and Methods.

Plasma-cleaned, formvar carbon film-coated, 300 mesh, copper grids (Electron Microscopy Sciences) were incubated 2 min on 5-μL drops containing gB vesicles. Dilutions were optimized for each vesicle preparation. After two washing steps on 50-μL drops of wash buffer (0.1% BSA in PBS), grids were incubated on 50-μL drops containing buffer A (1% BSA in PBS) or R218 anti-gB polyclonal Ab (a gift from Gary H. Cohen and Roselyn Eisenberg, University of Pennsylvania, Philadelphia) (8 µg/mL in buffer A) in a wet chamber for 3 h at room temperature. After five washes on wash buffer, grids were incubated for 30 min on 20-μL drops of 1:40 dilutions of anti-rabbit IgG-gold conjugate (5-nm particle size) in wash buffer. Before negative staining, grids were washed on five drops of wash buffer, followed by five washes on water drops. Staining was done on 2% NanoW (Nanoprobes) drops for 10 s before drying.

Imaging was performed using a TF30 electron microscope (FEI) operated at 200 keV. Images were recorded on a US4000 CCD ∼4,000 × 4,000 camera (Gatan) with a calibrated pixel size of 0.295 nm at specimen level and a defocus of −5.5 μm using SerialEM (31). After recording of overview images in lower magnification on various regions of the grid, random spots containing vesicles were marked and imaged at the indicated magnification.

For the reconstruction of the short form of gB (Fig. S2), particle subvolumes were picked manually from two sets of tomograms. Each was from independent (i.e., different cell culture) preparation of gB vesicles, comprising eight tomograms with 881 particles and four tomograms with 1,381 particles, respectively. Both sets of tomograms were binned fourfold, and the particles were aligned separately to confirm that each dataset converged to similar structures. The averages from both datasets showed strong threefold symmetry. For removal of duplicate particles, groups of particles with center-to-center distances of <80 Å were removed, except for the one representative with the highest cross-correlation value to the reference. The remaining particles were merged into one dataset of 1,909 particles. Tilt orientations (axes normal to the membrane) and positions were kept from the previous alignments, but the twist angle (angle around symmetry axis) for each particle was randomized. Ten iterations of twist alignment were performed on fourfold binned particles, and six iterations of further refinement were then performed around all axes on twofold binned particles. This structure showed the same strong threefold symmetry as seen before. Thus, two extra particles were added for each particle already picked, each with the same tilt angle and position as the original but with either 120° or 240° of twist rotation applied. A further three iterations of refinement were run. The resulting map was resymmetrized (C3) and filtered to the resolution given by the FSC (Fig. S4; details are provided below). All alignments were done using PEET 1.9 (33).

Gold-standard FSC calculation was performed by splitting the merged 1,909-particle dataset into odd and even particles, and applying the same refinement procedure to both independently. The split averages were aligned and masked using a custom mask based on segmentation of the merged average. To avoid correlation between the masks, one average was masked with a hard-edged mask and the other with a soft-edged one. The FSC, calculated using Bsoft (34), is 27 Å at the 0.5 criterion and 23 Å at the 0.143 criterion.

For the reconstruction of the long form of gB (Fig. S3), 457 particles were picked manually from a total of nine tomograms. All tomograms were fourfold binned. Five iterations of tilt and shift alignment were applied to the entire dataset. Duplicate particles were removed as before, leaving 395 particles. Five iterations of twist refinement were then applied on twofold binned particles, followed by five iterations of minor refinement around every axis. C3 symmetry was applied as in the short-form refinement, and a further four iterations of refinement were applied. The resulting map was resymmetrized (C3) and filtered to the resolution given by the FSC.

The FSC calculation was made by splitting the 395-particle dataset into two halves and refining them separately. The two halves were aligned and masked with a short cylindrical mask that contains only the protein, one soft-edged and one sharp-edged. The FSC was 36 Å at the 0.5 criterion and 31 Å at the 0.143 criterion.

A model combining Protein Data Bank ID codes 2GUM and 3NWF was built using MODELER (35). It contains the coordinates of the FLs at neutral pH (resolved in 2GUM, residues 261 and 262) with the additional regions resolved at acidic pH (present in 3NWF, residues 331–337 and 460–476).

The density corresponding to the membrane bilayer was first removed using the crop tool in Chimera (36). The resulting density map was segmented using the Segger tool (37) with default values. The segmented densities were visually inspected and manually grouped to include only segments from the gB trimer.

For the fitting of domain I (residues 154–363), four independent ensembles of fits were generated using different rigid fitting methods that vary on the space-sampling method and/or criteria used to assess the quality of fit of the resulting models. Ensemble 1 was created using the Fit_in_map global search function in Chimera (36), sampling 1,000 initial random points within the density map. Ensemble 2 was generated with ADP_EM (38) setting a bandwidth of 32 and using translational sampling of 4.6 Å (single-voxel size), a density threshold of 1.63, and the exhaustive search mode. GMFit (39) was used to generate Ensemble 3 by fitting three copies of domain I into the density map. The number of Gaussians for each individual domain and the density map was set to eight and 15, respectively. The number of initial configurations, the number of configurations for the local search, and the number of output configurations were set to 1,000, 100, and 50 respectively. Ensemble 4 consisting of 50 fits was generated using the MultiFit tool (40) implemented in Chimera. The density threshold, the resolution of the map and the sampling rate were set to 1.63, 22 Å, and 15°, respectively. The latter two ensembles were generated using the threefold rotational symmetry option available for the respective program. For the first two ensembles, two threefold symmetrical copies were added to the final results using the sym command in Chimera. From each method, the 30 top-ranking models were collected, resulting in a new ensemble of 120 models in total.

Next, two complementary scores were computed for each domain I fit, the segment-based cross-correlation coefficient (SCCC) (41) and an atom protrusion score (APS), we have implemented in TEMPy (42). The APS calculates the fraction of atoms in a given model/protein that protrude out of the envelope of the EM density map as follows:where N is the total number of backbone atoms and n is the number of backbone atoms protruding outside the envelope. The envelope is defined by a set of voxels for which the density values are above a given threshold ε:where voxel_density_bb is the density value of the voxel occupied by the backbone atom bb and ε is the threshold value. Here, the value of ε was set to 1.63 (1 SD from the mean of the density map).

A combined score integrating both measures was used for calculating the goodness of fit by taking the difference between SCCC and APS scores (ranging between −2 and 1). Using this score, the 120 domain I fits were then clustered by a hierarchical clustering method based on both the score and the Cα-rmsd between the fits. The cluster cutoffs between 7 Å and 17 Å were tested, and the value of 13 Å was selected based on the number of clusters and cluster size distribution (Fig. S6), resulting in 50 clusters. Within each cluster, the top-scoring fit was chosen as a representative. The 10 top-scoring representatives were selected for further assessment. These fits were filtered based on backbone clashes between protomeric units using Chimera, resulting in seven fits positioning domain I in the membrane-distal region of the map. Next, the postfusion crystal structure of both domains I and II (residues 142–153 and 354–460) was superimposed on each fit of domain I. Using the superimposed domain II as the starting position, an ensemble of 250 configurations of domain II was generated (allowing a ±60° rotation and 3-Å translation) and symmetrized (threefold) using TEMPy (42) in the context of the map. The resulting 1,750 domain II fits were scored (as before) and filtered using backbone clashes and distance constraints between domain I and domain II. Distance constraints were based on the maximum length possible for the linker regions between residues 152–157 and 355–366, respectively. The scores of the remaining models clearly discarded all domain I models oriented with the FLs pointing toward the membrane (Fig. 3). To complete the fit, we ran four iterations of Flex-EM (43) on the top four models to refine the position of domain II in the map further and confirm that the linkers were capable of joining the two domains without breaking bond restraints.

For domain IV (residues 111–116 and 573–660) positioning, an ensemble of 210 fits was generated with Chimera as described earlier, and the threefold symmetrized fits were scored (as above), clustered, and filtered using backbone clashes between the domain IV fits and the previously obtained domain I and II models. After testing (Fig. S7), a Cα-rmsd cutoff of 9 Å was chosen for clustering, yielding 151 clusters. Only 50 fits passed this filtering step. The four best-scoring fits were selected (the score distribution showed a drastic drop after model 4). Although the selected fits did not converge to a single orientation of domain IV, they were all located in the midlevel density, neighboring domain II (Fig. S7).