Introduction

Herpes simplex virus 1 (HSV-1) is a large, enveloped DNA virus from the family Orthoherpesviridae that infects mucosal epithelial cells in oral or genital regions and subsequently establishes life-long latent infections in trigeminal and sacral ganglia, respectively^1,2. HSV-1 is composed of three layers: a proteinaceous capsid that contains the DNA genome, an amorphous layer of at least 23 different proteins known as the tegument, and a lipid bilayer envelope studded with viral glycoproteins^3–8. Virion assembly involves a complex multi-step pathway in the cell, beginning with the expression of immediate-early viral genes⁹. This is followed by replication of viral DNA, assembly of capsids, and packaging of DNA genomes to form nucleocapsids, all within the nucleoplasm¹⁰. Given that nucleocapsids are too large to pass through the channels of nuclear pore complexes, they cross the nuclear envelope in a process known as nuclear egress. This begins with budding of nucleocapsids into the perinuclear space through the inner-nuclear membrane (primary envelopment). Once in the perinuclear space, the nucleocapsids possess a temporary envelope, acquired from the inner-nuclear membrane, which will then fuse with the outer-nuclear membrane to release the capsid into the cytoplasm^11–14. Tegument proteins begin being deposited on the nucleocapsid in the nucleus and this process continues in the cytoplasm⁴. Cytoplasmic nucleocapsids undergo (secondary) envelopment at cytoplasmic endomembranes, such as trans-Golgi network vesicles and/or recycling endosomes^5,15, hereafter referred to as cytoplasmic envelopment. This results in the enveloped virion being located in the lumen of a carrier vesicle, which fuses with the plasma membrane to release the virion⁴.

The large DNA genome of HSV-1 encodes at least 84 proteins, many of which have uncharacterised functions in virus assembly and differ in the degree of their importance during morphogenesis⁵. HSV-1 mutants lacking expression of specific viral proteins or domains thereof have been used to study the role of viral genes in assembly using 2D transmission electron microscopy (TEM). Several phenotypes associated with attenuation have been observed with this technique, including varying numbers of capsids between the nucleus and cytoplasm, clusters of perinuclear virus particles, and stalled envelopment of cytoplasmic nucleocapsids^16–21. However, the extensive sample processing associated with TEM (i.e. chemical fixation, dehydration, hardening by resin-embedding or high-pressure freezing, sectioning, and staining) can distort ultrastructure and complicate interpretation of features that appear to show attenuation in virus assembly^22,23. Thin sectioning also limits our understanding of the 3D nature of phenotypes associated with attenuation in virus assembly²⁴. Volumetric ultrastructural imaging of numerous HSV-1 mutants in near-native conditions is needed to draw direct comparisons between viral proteins and to understand the 3D nature of stalled assembly.

In this study, the roles of nine HSV-1 genes in virus assembly were investigated using an emerging 3D correlative imaging strategy consisting of structured illumination microscopy on cryopreserved samples (cryoSIM) and cryo-soft-X-ray tomography (cryoSXT)²⁵. HSV-1 assembly intermediates have previously been observed by cryoSXT without performing cryoSIM, which limited characterisation of virus assembly²⁶. Using this correlative light X-ray tomography (CLXT) approach, we employed recombinant viruses expressing fluorescently labelled viral proteins of the capsid (VP26) and the envelope glycoprotein M (gM) to distinguish between unenveloped and enveloped particles in order to study different stages of virion assembly^27–29. Our study provides an extensive and comparative ultrastructural analysis of herpesvirus morphogenesis, revealing new insights into the virion assembly pathway of these structurally complex viruses.

Results

Generation of mutants to study HSV-1 assembly

HSV-1 particles contain three layers: capsid, tegument, and envelope. A dual-fluorescent KOS strain HSV-1 mutant containing eYFP-tagged capsids and mCherry-tagged envelopes — hereafter referred to as the dual-fluorescent parental (dfParental) virus — was used to enable the independent identification of capsids and envelope and thus distinguish between unenveloped and enveloped particles. The tagged virus encodes eYFP conjugated to the N terminus of the small capsid protein VP26^27–29. This tagging theoretically allows identification of all virus particles except capsid-free “light” particles³⁰ in the cytoplasm and immature procapsids in the nucleus that have yet to acquire VP26³¹. The C terminus of the envelope glycoprotein gM was conjugated to mCherry, enabling identification of gM-containing endomembranes plus viral envelopes²⁹ (Fig. 1A). The dfParental virus was used as the template strain to generate nine HSV-1 mutants lacking specific HSV-1 structural proteins. Mutants were generated to study phenotypes associated with an attenuation of virus assembly caused by the loss of specific viral proteins (Fig. 1A).

Figure 1.

Nine mutants containing single-gene “knockouts” were generated either by introducing stop codons or by sequence deletion (Fig. 1B). As pUL34, VP16 and gK are all essential for virion assembly, the ΔUL34, ΔVP16, and ΔgK viruses were cultured using complementing cell lines to generate infectious virions that carry the corresponding protein but are unable to synthesise it, such that virus entry and gene expression could occur as normal, but the protein would be absent during virus assembly. Protein levels of the corresponding knockout genes were undetectable in Vero cells infected with each of the HSV-1 mutants by immunoblot (Fig. 1C–D). No suitable antibody was available to detect gK expression and so the presence of the inserted stop codons was confirmed by PCR amplification of the relevant region of the HSV-1 genome and Sanger sequencing (Supp. Fig. 1A–B). Furthermore, immunoblotting for pUL20, the obligate binding partner of gK, revealed substantially reduced pUL20 expression in cells infected with the ΔgK mutant (Fig. 1C) as previously observed for independent gK deletion mutants³². The intensity of the pUS3 bands were reduced for the ΔpUL21 and ΔUL34 mutants (Fig. 1C). The US3 gene has a propensity to mutate in response to pUL21 deletion³³, but this was not known at the time these virus stocks were generated, and the ΔpUL21 virus was not grown in a complementing cell line.

Replication kinetics and cell-to-cell spread of HSV-1 mutants

Replication kinetics for all the dual-fluorescent viruses were compared over 24 hours in single-step growth curves (Fig. 2A and Supp. Fig. 2). These data were compared with measurements of plaque size 72 hours post infection (hpi) that served as collective indicators of attenuation in both replication and cell-to-cell spread (Fig. 2B–C). In line with previous work showing that gE is more important in cell-to-cell spread than in virus assembly^7,34–36, defects in replication of the ΔgE virus were not detected in the replication curves (Fig. 2A), but the plaque sizes of this mutant were significantly reduced compared with dfParental (Fig. 2B–C). gE interacts with a complex of the tegument proteins pUL11, pUL16 and pUL21, providing one route to link the tegument and envelope layers during virus assembly^37–41. Replication of the ΔpUL11, ΔpUL16, and ΔpUL21 mutants was generally reduced with respect to dfParental HSV-1 (10–100-fold reduction), suggesting these proteins have more important roles than gE during virion assembly and could perform these roles as a complex (Fig. 2A). ΔpUL11 and ΔgE plaques were significantly smaller than the ΔpUL16 and ΔpUL21 plaques, suggesting the interactions formed by pUL11 and gE are more important for cell-to-cell spread (Fig. 2B–C)⁴². However, viruses lacking functional pUL21 are known to form extremely small plaques and compensatory mutations in the US3 gene can arise when pUL21 is inactivated³³. The sequence of US3 in the ΔpUL21 virus was therefore analysed by PCR and Sanger sequencing, revealing that 84.3% of ΔpUL21 genomes encoded an amino acid substitution (C456T) in the US3 gene (Supp. Fig. 3A–C). Residues with similar physicochemical properties (e.g. serine) are present at this position in US3 from other alphaherpesviruses, suggesting that this substitution is unlikely to severely alter US3 activity (Supp. Fig. 3D–E). Although the residue is surface exposed (Supp. Fig. 3F), it is not located near the active site, suggesting it is unlikely to affect kinase activity (Supp. Fig. 3G). Nonetheless, potential alterations to pUS3 activity in the ΔpUL21 mutant should be taken into account when the results for this mutant are interpreted.

Figure 2.

After capsid assembly in the nucleus, capsids migrate into the cytoplasm by budding into the perinuclear space through the inner-nuclear membrane, forming a temporarily enveloped particle⁴³. The envelope of the perinuclear particles fuses with the outer-nuclear membrane to release the capsid into the cytoplasm. pUL34 is required for budding into the perinuclear space, and pUS3 is thought to regulate this process via phosphorylation of the pUL31/pUL34 nuclear egress complex (NEC)^44–46. While replication of the ΔUS3 virus was 10-fold reduced with respect to dfParental at 24 hpi, the plaques formed by the ΔUS3 mutant were closer in size to dfParental plaques than any other mutants at 72 hpi, suggesting that pUS3 had the least important role of all nine mutants for capsid migration into the cytoplasm and subsequent viral spread (Fig. 2A–C). In contrast, the replication kinetics of the ΔUL34 mutant were approximately 10⁵-fold reduced, and the mutant did not form plaques, which was expected for an essential component of the NEC (Fig. 2A–C)^11,12.

VP16, pUL51, and gK are highly important for cytoplasmic envelopment of HSV-1 capsids^8,47–50. By 24 hpi, replication of the ΔpUL51 virus was 10 to 500-fold reduced and plaque sizes were greatly reduced as well (Fig. 2A–C and Supp. Fig. 2). Replication kinetics of the ΔgK virus were 10³ to 10⁵-fold reduced at 24 hpi, and no plaques were visible at 72 hpi (Fig. 2A–C and Supp. Fig. 2). This suggests gK is important in cell-to-cell spread as well as virion assembly. It is possible that infectious ΔgK virions were produced but unable to egress from the cell. In this scenario, infectious particles could be detected in cell lysates generated for titration of the replication curves despite an inability of ΔgK HSV-1 to form plaques. In contrast to the ΔgK mutant, the ΔVP16 mutant produced detectable plaques by 72 hpi even though infectious virions could not be detected by 24 hpi with the replication curves, suggesting the lack of VP16 could delay to beyond 24 hours the production of detectable titres of infectious virus (Fig. 2A–C). Alternatively, it is possible that loss of VP16 leads to low levels of cell-cell fusion to form small syncytia, enabling cell-to-cell spread of viral genome and subsequent plaque formation²¹.

Correlative light X-ray tomography of HSV-1 assembly

After generating and characterising the mutants, defects in virus assembly were explored with high-resolution imaging techniques under cryogenic conditions. Cryogenic imaging using soft-X-ray tomography captures the ultrastructure of the cell in 3D and has the advantage of allowing the study of samples in a near-native state without the need for chemical fixation^26,51. U2OS osteosarcoma cells were used in this study because they have been used for ultrastructural HSV-1 research and have been demonstrated to be durable under X-ray exposure (Fig. 3A)^26,52,53. U2OS cells were infected at MOI = 2 with dfParental or mutants for 15.5 hours and were stained with MitoTracker Deep Red (Thermo Fisher Scientific) for 30 minutes to label the mitochondria⁵⁴. Gold fiducials were added to the surface of the cells to facilitate alignment of tomographic projections⁵⁴. Samples were cryopreserved by plunge cryocooling in liquid ethane at 16 hpi. In a synchronously-infected population, infected cells will progress through stages of virus assembly at different rates^26,29,55. Cells were cryopreserved at a relatively late timepoint (16 hpi) to increase the proportion of cells at late stages of infection. Infected cells were first imaged by cryoSIM to capture viral eYFP-VP26 and gM-mCherry fluorescence plus MitoTracker mitochondrial fluorescence (Fig. 3B)^25,56. These same infected cells were later imaged by cryoSXT. Given that mitochondria produce high contrast in cryoSXT tomograms and are easily distinguishable based on their complex shapes and arrangements²⁶, the MitoTracker stain in the cryoSIM and the mitochondria in the tomograms were used as landmarks to guide 3D image alignment for correlation²⁵. CLXT was used to identify virus particles at various stages of assembly. The nucleus is the site of capsid assembly and nuclear capsids have previously been observed in the tomograms as dark puncta²⁶. These capsids correlated with the fluorescently tagged capsid protein eYFP-VP26 (Fig. 3C). gM-mCherry⁺ vesicles and virus particles were detected in the cytoplasm, enabling the study of cytoplasmic envelopment. Finally, eYFP-VP26+ and gM-mCherry+ particles were detected in spaces between cells, allowing study of extracellular virions (Fig. 3C).

Figure 3.

Nuclear egress attenuation of ΔpUL16, ΔpUL21, ΔUL34, ΔVP16, and ΔUS3 HSV-1

Nuclei of samples infected with dfParental, ΔpUL16, ΔpUL21, ΔUL34, ΔVP16, and ΔUS3 viruses were labelled with Hoechst stain to distinguish between nuclear and cytoplasmic capsids by cryoSIM (Fig. 4A). The plasma membrane of infected cells was delineated by digitally oversaturating the gM-mCherry fluorescence. The number of eYFP-VP26⁺ pixels in maximum Z projections of the cryoSIM data were quantitated using these nuclear and plasma membrane borders. The nuclear:cytoplasmic (N:C) ratio of capsids was used as measures of nuclear egress attenuation, which could manifest from a defect in nuclear egress or a delay in the replication kinetics of mutants (Fig. 4B). Infected cells were included in the analysis if the nucleus and cell borders could be easily distinguished, if most of the cell was in the field of view, and if most of the gM-mCherry⁺ areas of the cytoplasm did not overlap with the nucleus. However, partial overlap between gM-mCherry fluorescence and the nucleus was common (e.g. ΔpUL21 and ΔUL34 in Fig. 4C), which could have produced an overestimated N:C ratio. These data demonstrated that ΔpUL16, ΔpUL21, ΔUL34, ΔVP16, and ΔUS3 viruses experienced a defect or a delay in nuclear egress when compared with the dfParental virus (Fig. 4B–C). The ΔgE virus was included in the analysis as a negative control because this virus did not have delayed replication kinetics (Fig. 2A), and gE is not suspected to be involved in nuclear egress. The ΔgE virus-infected cells were not stained with Hoescht, and instead the nucleus was identified using images of nuclei collected from tiled X-ray projection images (X-ray mosaics). The N:C ratio of the dfParental (N=14) and ΔgE (N=12) viruses did not differ significantly, suggesting gE does not play a role in nuclear egress. pUL34 is essential for nuclear egress and served as a positive control for nuclear capsid retention in this study^11,12. The ΔUL34 virus had the highest N:C ratios (N=10; Fig. 4B–C), which was commensurate with the high level of attenuation observed for this virus in terms of replication kinetics and its inability to form plaques (Fig. 2A–C). For the other mutants, significance in difference was assessed with respect to the N:C ratios of both the dfParental and ΔUL34 viruses. Although the N:C ratios of the ΔUS3 virus (N=13) were significantly higher than those of the dfParental virus, they were also significantly lower than those of the ΔUL34 virus. The ΔUS3 virus had the lowest N:C ratios of the mutants other than the ΔgE negative control (Fig. 4B–C), suggesting ΔUS3 was least attenuated in nuclear egress, which is commensurate with its modest attenuation in the replication curves and plaque size assay (Fig. 2A–C). ΔpUL16 (N=13) and ΔpUL21 (N=11) N:C ratios were significantly higher than those of the dfParental virus but were not significantly different from those of the ΔUL34 virus (Fig. 4B–C), suggesting a greater defect or delay in nuclear egress than observed for ΔUS3 and consistent with known roles for both proteins in nuclear egress of HSV-2^18,20,57. N:C ratios of ΔVP16 (N=24) were also significantly greater than those of dfParental, suggesting a defect or delay in nuclear egress (Fig. 4B–C), consistent with TEM studies implicating VP16 in nuclear egress^21,58. CryoSIM data from infected cells were correlated onto X-ray tomograms, revealing an absence of, or reduction in, cytoplasmic capsids for each of the five mutants (Fig. 4D).

Figure 4.

Differential capsid clustering and gM-mCherry⁺ endomembrane association of ΔpUL11, ΔpUL51, ΔgE, ΔgK, and ΔVP16 mutants

CryoSIM imaging showed that interspersed eYFP-VP26+ capsids and gM-mCherry+ endomembranes could be observed at juxtanuclear assembly compartments (JACs) in cells infected with dfParental (Fig. 5A). Varying degrees of attenuation in cytoplasmic envelopment have been previously reported for HSV-1 mutants lacking some of the tegument proteins or glycoproteins investigated in this study (i.e. pUL11^16,17,59, VP16²¹, pUL51^48,49,60, gK^47,61, and gE^36,42), and some proteins have been proposed to participate in cytoplasmic envelopment (i.e. pUL16³⁷ and pUL21^62,63). For cells infected with ΔpUL11, ΔpUL51, ΔgK, or ΔgE, clusters of capsids could be observed at the JACs (Fig. 5C). Smaller clusters of capsids were observed in the JACs for the ΔpUL11 and ΔgE mutants (Fig. 5C), whereas more extensive clusters of capsids were observed in the JACs for the ΔpUL51 and ΔgK. Capsid clustering was not observed in cells infected with ΔVP16 virus, and cytoplasmic capsids appeared to associate less with gM-mCherry+ endomembranes compared with capsids of dfParental, suggesting VP16 is important in capsid recruitment to envelopment compartments (Fig. 5B–C). The spatial distribution of capsids in the cytoplasm of cells infected with ΔpUL16 and ΔpUL21 could not be reliably assessed due to a paucity of cytoplasmic capsids detected, presumably arising from their observed defect/delay in nuclear egress (Fig. 4B-D).

Figure 5.

To quantitate cytoplasmic distribution of virus particles, the intensity of gM-mCherry fluorescence was used to delineate the approximate borders of JACs (Fig. 5A). dfParental capsids in these assembly compartments associated closely with gM-mCherry⁺ endomembranes. However, cells infected with the ΔVP16 virus contained fewer capsids in the JAC and they were less closely associated with gM-mCherry⁺ endomembranes (Fig. 5A–C). Thresholds on intensity were applied to the eYFP-VP26 and gM-mCherry cryoSIM fluorescence to produce binary masks where noise and background were filtered out. The ratio of eYFP-VP26⁺ pixels to gM-mCherry⁺ pixels in the binary masks was measured for each JAC and were significantly lower for the ΔVP16 virus when compared with dfParental (Fig. 5D), indicating fewer capsids at the JAC for this virus. This is consistent with the lack of overlap between capsids and gM-mCherry+ endomembranes observed (Fig. 5B) plus the observed defect/delay in nuclear egress (Fig. 4B–D). ΔUS3 was included in the analysis as a negative control for a virus with impaired nuclear egress that is not expected to be attenuated in cytoplasmic envelopment. The ratio of eYFP-VP26⁺ pixels to gM-mCherry⁺ pixels at the JACs was not significantly different between the dfParental and ΔUS3 viruses (Fig. 5D), suggesting that the lower ratio of eYFP-VP26⁺ pixels to gM-mCherry⁺ pixels in the JAC of ΔVP16 infected cells reflects poor capsid recruitment to these compartments in the absence of VP16. The eYFP-VP26⁺ pixels to gM-mCherry⁺ pixels ratio for the ΔpUL11, ΔpUL51, and ΔgE mutants was not significantly different from dfParental, suggesting capsid recruitment was unimpaired, and for ΔgK this ratio was significantly increased, suggesting that capsid recruitment to or retention at JACs is enhanced in cells infected with this mutant (Fig. 5D).

Colocalization analyses were performed to determine the overlap between capsids and gM-mCherry⁺ endomembranes at the JACs. Manders colocalization coefficients, which are based on spatial coincidence, were chosen over intensity-based measures, such as Pearson’s correlation coefficients, to minimize the influence of noise and cryoSIM reconstruction artefacts. Manders coefficient 1 (M₁) was calculated for each JAC and represented the number of eYFP-VP26⁺ pixels that overlapped with gM-mCherry⁺ pixels as a proportion of all eYFP-VP26⁺ pixels. Manders coefficient 2 (M₂) was also calculated for each JAC and represented the number of gM-mCherry⁺ pixels that overlapped with eYFP-VP26⁺ pixels as a proportion of all gM-mCherry⁺ pixels. These two measurements reflect the association of capsids with gM⁺ endomembranes. Both metrics were significantly lower for the ΔVP16 virus (N=22) compared with dfParental (N=16; Fig. 5E, Supp. Fig. 4). The M₁ and M₂ coefficients for the mutants that formed small capsid clusters (i.e. ΔpUL11 [N=25] and ΔgE [N=23]) did not significantly differ from dfParental. For the mutants that form large capsid clusters, the M₁ coefficient was larger than dfParental for both ΔpUL51 (N=17) and ΔgK (N=18) but only reached statistical significance for ΔpUL51, and the M₂ coefficient was increased for both ΔgK and ΔpUL51 but only reached statistical significance for ΔgK. This suggests enhanced association of capsid and gM⁺ endomembranes in the JAC for both these mutants but with different distributions of the components (Fig. 5E).

CLXT imaging provided higher resolution details of the relative capsid and membrane arrangement for these mutants. As observed by cryoSIM (Fig. 5A), cytoplasmic clusters of virus particles were not observed for dfParental HSV-1 by CLXT (Fig. 6A). Small clusters of virus particles (<10 μm³) were observed in the JACs of the ΔpUL11 and ΔgE mutants (Fig. 6B), and larger clusters of virus particles (≥10 μm³) were observed in the JACs of the ΔpUL51 and ΔgK viruses (Fig. 6C). Additionally, linear arrays of capsids were observed in the cytoplasm of the ΔgK virus (Fig. 6D) that were not observed in dfParental-infected cells (Fig. 3C). CLXT revealed that these linear capsid arrays are not associated with gM-mCherry⁺ endomembranes. From their linear organisation we suspect they are located along filaments that could not be reliably resolved by cryoSXT, potentially microtubules as these are known to be important for intracellular capsid transport⁶⁴.

Figure 6.

3D envelopment mechanism of HSV-1

The high penetrating power of X-rays enables the entire depth of the cell in each field of view to be imaged by cryoSXT, allowing rare or transient events to be captured²⁶. Combining this with the HSV-1 mutant strains that have impaired envelopment allows us to enrich intermediate stages of virus assembly that are otherwise extremely rapid and thus difficult to visualise. Numerous independent assembly intermediates of the ΔpUL51 virus were detected by CLXT (Fig. 7A–B, Supp. Fig. 5A–B). These included unenveloped capsids in the cytoplasm and capsids embedded in the surface of gM-mCherry⁺ vesicles. Note that the anterior and posterior faces of these vesicles cannot be reliably segmented owing to their lower contrast, a result of the ‘missing wedge’ in the tomographic data acquisition. Such vesicles were enriched in gM-mCherry at the pole near the capsids, indicating that gM-mCherry and potentially other viral proteins become concentrated at microdomains on vesicles prior to envelopment (Fig. 7A). This observation is consistent with a previous hypothesis suggesting concentration of tegument and glycoproteins at the ‘assembly pole’ of the virion⁶⁵. Virus particles budding into vesicles and fully enveloped virions were also observed (Fig. 7A–B). Fully enveloped virions were distinguished from budding intermediates by observing differences in voxel intensity across the vesicles that contain them: the virions were separated from the vesicle membrane by a narrow volume of lumenal space that appeared brighter (Supp. Fig. 5B), whereas budding events displayed a continuous drop in intensity between the membrane and the budding intermediate (Supp. Fig. 5A). A 2D cross-section of a budding event (Fig. 7C–D) appears topologically similar to envelopment events observed by 2D TEM (Fig. 7E). This 3D view of cytoplasmic budding suggests envelopment is driven by capsid budding into spherical/ellipsoidal vesicles rather than by thin tubular endomembranes forming projections to wrap around capsids.

Figure 7.

Capsid arrays were frequently observed near cytoplasmic vesicles for the ΔgK (Fig. 8A), ΔpUL11 (Fig. 8B), ΔpUL51 (Fig. 8C), and ΔgE (Fig. 8D) viruses. These features were not observed in dfParental-infected cells (Fig. 3C), and they share four phenotypes regardless of the type of mutant. First, an array of capsids is located around one pole of the vesicle but not the antipole. Second, the capsids appear to be near the vesicles but generally do not appear embedded in the surface. Third, the pole of the vesicle near to the capsid arrays appears darker in the tomograms, indicating a greater presence of X-ray absorbing material. Fourth, CLXT revealed the vesicle pole near to the capsid arrays is enriched in gM-mCherry, suggesting gM and potentially other viral proteins accumulate at the pole nearer the capsid arrays, which could account for the greater abundance of X-ray absorbing material. Based on these four shared phenotypes, we interpret these features to represent capsids interacting with appropriate target membranes but experiencing a delay or defect in their ability to bud into the vesicle lumen to acquire an envelope. To our knowledge, stalled envelopment events with these details have not been described by 2D TEM. The 3D nature of cryoSXT increases the likelihood that ultrastructural features such as these could be captured regardless of orientation with respect to the XY projection plane⁶⁶.

Figure 8.

To quantitate the variation in X-ray absorbing material along the membrane from a stalled envelopment event for each virus, the voxel intensities of 30 points on the relevant vesicles were measured (Fig. 8E–H). X-ray tomograms reconstructed by weighted back projection (WBP) were used for the analysis without applying any noise-averaging simultaneous iterations reconstruction technique (SIRT)-like filters⁶⁷. Given that WBP tomograms are noisy⁶⁷, voxel intensities were sampled from a 3ξ3 voxel matrix in the XY plane at each point on the vesicle and the minimum value was used. Voxel intensities were collected and averaged from three tandem tomographic projections spanning a depth of 30 nm. For all four mutants, the voxel intensity was lower on the vesicle pole nearer to the capsid arrays (Fig. 8E–H).

Voxel intensity is proportional to the X-ray radiation transmitted through the sample to the detector. A low voxel intensity suggests lower X-ray transmittance and greater X-ray absorption, indicating the vesicle pole nearer the capsid arrays contained a greater amount of carbon-rich X-ray-absorbing material. To visualise the variation in X-ray absorption around the vesicle from Fig. 8D in 3D, the voxel intensity of the tomogram was superimposed and false-coloured onto a segmentation of the vesicle (Fig. 8I). This illustrated that the vesicle pole nearer the capsid arrays had greater X-ray absorption than the antipole. The width of vesicles associated with capsid arrays were measured using the quantitation program Contour⁶⁸ and these vesicles had a mean width of 704.9 ± 206.5 nm (mean ± SD, n = 34) (Supp. Fig. 6A). Stalled envelopment features were not observed in dfParental-infected cells, and we wondered whether the enrichment of gM-mCherry at one vesicle pole was an artefact of deleting tegument or envelope proteins that could directly or indirectly interact with gM (e.g. pUL11 or gE)⁴. However, polarisation of vesicular gM-mCherry was observed in dfParental-infected cells in the absence of capsid arrays (Supp. Fig. 6B). In this case, the vesicle appeared to be constricted in between the gM-mCherry⁺ and gM-mCherry⁻ poles, which could arise for numerous reasons, such as fusion, fission, or pressure imposed by microtubules.

Stalled envelopment was frequently observed for the ΔgE virus (Supp. Fig. 7A). This was surprising because no attenuation in replication kinetics was detected for the ΔgE virus when compared with dfParental (Fig. 2A). Previous research indicated HSV-1 lacking both gE and gM was more severely attenuated in replication kinetics than HSV-1 lacking one of either protein⁶⁹. This could be related to the indirect interaction between gE and gM via the VP22 tegument protein (Supp. Fig. 7B)⁶⁹. Untagged and eYFP-VP26/gM-mCherry-tagged forms of the virus displayed similar replication kinetics, suggesting no detectable impact on replication of mCherry conjugation to the C terminus of gM (Supp. Fig. 7C). Plaque sizes were measured for each virus to determine if the tagging affected cell-to-cell spread. The ΔgE virus was on average 0.65ξ smaller than the WT virus for the untagged form but 0.35ξ smaller for the tagged form, suggesting the tagging contributed to an attenuation in cell-to-cell spread (Supp. Fig. 7D–E).

Width measurements of virus assembly intermediates

Electron cryo-tomography (cryoET) has been used to measure the width of the tegument layer and an asymmetric distribution of tegument around the capsid has been demonstrated, with widths of 5 nm and 35 nm at opposing poles⁷⁰. Widths of viral assembly intermediates were measured for the dfParental, ΔpUL51, and ΔgE viruses by cryoSXT to determine if the added width of the tegument layer around the capsid could be detected (Fig. 9 & Table 1)²⁶. No difference was observed in the width of nuclear capsids and unenveloped membrane-proximal capsids by cryoSXT (Fig. 9 & Table 1). This could indicate that recruitment of tegument to capsids does not occur until cytoplasmic envelopment. However, ICP0, ICP4, pUL36, pUL37, and pUS3 are thought to condense on capsids in the nucleus, demonstrating tegument condensation begins early in assembly^71,72. Alternatively, it is possible that the tegument layer is too diffuse around cytoplasmic capsids such that it does not produce detectable X-ray absorption by cryoSXT. Under this scenario, X-ray absorption may become detectable if the tegument layer were to compress around capsids upon membrane-embedding and budding. Embedded particles widths were measured normal (at 90°) to the membrane to limit distortion of the measurements by the membrane. The widths of embedded particles (184.12 ± 4.86 nm SEM; N = 17; range 150–220 nm; SD 20.02 nm) were significantly greater than nuclear capsids (122.48 ± 0.94 nm SEM; N = 149; range 100–150 nm; SD 11.44 nm; Mann-Whitney U-test p-value = 4.27 × 10^-12) or membrane-proximal capsids (124.95 ± 1.15 nm SEM; N = 105; range 100–150 nm; SD 11.78 nm; Mann-Whitney U-test p-value = 1.79 × 10^-11), suggesting the tegument layer condenses and compresses around the capsid upon membrane-embedding and budding (Fig. 9 & Table 1). The widths of embedded particles were also significantly lower than intracellular enveloped particles (216.85 ± 4.65 nm SEM; N = 54; range 160–300 nm; SD 34.19 nm; Mann-Whitney U-test p-value = 6.8 × 10^-4) and extracellular particles (210.00 ± 5.65 nm SEM; N = 21; range 180–280 nm; SD 25.88 nm; Mann-Whitney U-test p-value = 3.5 × 10^-3), consistent with incomplete membrane acquisition.

Figure 9.

Table 1.

Discussion

Application of correlative cryoSIM and cryoSXT imaging to the study of viral infection has thus far been limited to visualising the entry of reovirus during infection and has not yet been used to study other stages of virus assembly, nor has it been used to study phenotypes of mutant viruses²⁵. In this study, nine mutants of HSV-1, each lacking a protein involved in virus assembly, were generated to study attenuation in nuclear egress and cytoplasmic envelopment with correlative cryoSIM and cryoSXT imaging. Each mutant genetically encoded fluorescent capsid (eYFP-VP26) and envelope (gM-mCherry) proteins that allowed identification of viral assembly intermediates without the need for chemical fixation, cell permeabilization or immunostaining, all of which could introduce ultrastructural artefacts. Instead, samples were vitrified in a physiologically-relevant near-native state by plunge-cryocooling⁶⁶. Lack of protein expression for each mutant gene was verified (Fig. 1C–D); the replication kinetics and plaque sizes for each virus were assessed (Fig. 2A–C); roles of pUL16, pUL21, pUL34, VP16, and pUS3 in nuclear egress of HSV-1 were observed (Fig. 4); CLXT revealed different phenotypes associated with attenuation for cytoplasmic envelopment with the ΔpUL11, ΔVP16, ΔpUL51, ΔgK, and ΔgE viruses, suggesting the corresponding proteins possess different functions in envelopment (Fig. 5–8); and the widths of virus assembly intermediates identified by CLXT were measured (Fig. 9 & Table 1).

The large defect in nuclear egress of HSV-1 lacking pUL34 as assessed by cryoSIM and CLXT observation (Fig. 4B–D) is consistent with its role as an essential component of the viral NEC. A nuclear egress defect was apparent but was significantly less pronounced for the ΔUS3 mutant, consistent with the impact of pUS3 kinase-regulated NEC activity and nuclear lamina dispersal being variable in magnitude and cell-type specific^46,73–75. While both pUL16 and pUL21 have been shown to promote nuclear egress in HSV-2^18,57, previous studies have not confirmed a role for either protein in nuclear egress in HSV-1^19,76. Our cryoSIM and CLXT data (Fig. 4B–D) show that both pUL16 and pUL21 promote nuclear egress in U2OS cells. Of note, the ΔpUL21 mutant was found to contain a non-synonymous point mutation in pUS3 that may independently influence nuclear egress (Supp. Fig. 3), consistent with our previous data showing positive selection of pUS3 mutations during passage of pUL21 mutant viruses that rescues HSV-1 replication and spread³³. While the presence of the pUS3 C463T mutation in our eYFP-VP26 and gM-mCherry tagged ΔpUL21 virus may be expected to compensate for any defect in nuclear egress in order to restore viral fitness, the nuclear egress defect observed for the ΔpUL21 mutant is greater than for the ΔUS3 mutant (Fig. 4B). This suggests that, if anything, our cryoSIM and CLXT data may underestimate the importance of pUL21 in this process. In addition, we note that our CLXT studies cannot differentiate mature genome-containing ‘C’ capsids from the genome-free ‘A’ and ‘B’ capsids that are known to accumulate in the absence of HSV-1 pUL16⁷⁶ and pUL21^19,77. The reduced extent of nuclear egress at 16 hpi for the ΔVP16 virus (Fig. 4B–D) could be attributed to a slower rate of replication in the absence of VP16 (either via diminished immediate-early gene trans-activation⁷⁸ or via loss of regulation of the HSV-1 viral host shutoff protein⁷⁹) rather than an attenuation in nuclear egress. However, the results are consistent with previous research that showed VP16 is incorporated on perinuclear virus particles in HSV-1⁵⁸ and that cells infected with a ΔVP16 virus produce clusters of virus particles in the perinuclear space²¹. The significant defect or delay in nuclear egress observed in this study (Fig. 4B–D) indicates that VP16 could play a larger role in nuclear egress than previously supposed.

In addition to nuclear egress, the CLXT workflow presented here represents a powerful platform for comparative investigation of viral cytoplasmic envelopment. Substantial cytoplasmic envelopment defects were observed for the ΔpUL11, ΔpUL51, ΔgE, ΔgK, and ΔVP16 mutants (Fig. 5–6), but the precise phenotype of the defect differed between mutants. Capsids were largely absent from juxtanuclear gM-mCherry⁺ endomembranes in cells infected with the ΔVP16 virus. This is consistent with the highly-abundant tegument protein VP16 playing a key role in bridging inner tegument proteins with the outer tegument and viral glycoproteins during assembly as well as the severe replication deficiencies of this mutant at 24 hpi (Fig. 2A)^50,80–86. Cells infected with ΔpUL11 and ΔgE HSV-1 had small clusters of capsids in JACs (Fig. 5 & Fig. 6), consistent with previous TEM studies^{16,17,36,87,88}, and both had moderate defects in virus replication and/or spread (Fig. 2A–C), suggesting that both play modest roles in cytoplasmic envelopment. Removal of pUL51 or gK gave rise to much larger clusters of capsids (Fig. 5 & Fig. 6) and more extensive replication and spread defects (Fig. 2A–C), suggesting more severe deficiencies in capsid envelopment. pUL51 and its binding partner pUL7 have established roles in stimulating the cytoplasmic wrapping of nascent virions^49,89, and the 3D structure of pUL51 closely resembles components of the cellular endosomal sorting complex required for transport (ESCRT)-III machinery⁴⁸, but the precise molecular roles pUL7 and pUL51 play in capsid envelopment remain unclear. Two replication curves reveal that removal of pUL51 expression causes a wide-ranging 10- to 500-fold reduction in virus replication (Fig. 2A and Supp. Fig. 2), which is consistent with data showing that inhibition of ESCRT activity via expression of a dominant negative form of the ESCRT-associated cellular ATPase VPS4A effectively prevents HSV-1 budding⁹⁰. CLXT reveals large clusters of cytoplasmic virus particles and stalled envelopment events for ΔpUL51 HSV-1 (Fig. 5 & Fig. 6), suggesting that pUL51 acts as a catalyst to accelerate cytoplasmic envelopment but that additional (redundant) cellular and/or viral mechanisms support envelopment in the absence of pUL51. Disruption of the pUL20-gK complex causes substantial defects in cytoplasmic envelopment and the cytoplasm of cells infected with HSV-1 ΔgK or ΔUL20 mutants has been shown to harbour a greater number of unenveloped membrane-associated capsids^17,47. The gK/pUL20 complex has been shown to interact with pUL37, and TEM analysis revealed that unenveloped capsids accumulate in the cytoplasm if the interacting residues in pUL37 are mutated^91,92, suggesting that interactions between gK/pUL20 and pUL37 are required for the formation of virus assembly compartments. However, our study revealed that capsids still associate with membranes in the absence of gK, indicating an interaction of gK/pUL20 with pUL37 is not required for recruiting capsids to assembly compartment membranes (Fig. 5–6). What our CLXT data does demonstrate is that the activity of both gK and pUL51 (and by association the gK/pUL20 and pUL7/pUL51 complexes) are important for virion assembly to progress beyond association of capsids with cytoplasmic membranes. Whether both these complexes function to regulate the similar cellular machinery involved in membrane curvature and/or scission events that are required to complete the cytoplasmic envelopment process, for example ESCRT activity, remains to be established. In addition to extensive membrane-associated cytoplasmic capsids being observed in ΔgK infected cells, linear arrays of unenveloped capsids were observed in the cytoplasm for this virus (Fig. 6D). Cytoplasmic capsids migrate along microtubules and these linear arrays could represent capsids stalled on microtubules when envelopment compartments become saturated with capsids in the absence of gK⁶⁴. One surprising result was the observation of numerous stalled envelopment events in ΔgE infected cells, despite no defects in replication kinetics being observed for the ΔgE virus (Fig. 8). This could arise from subtle differences in the kinetics of envelopment, whereby a minor defect in ΔgE budding slows the rapid process of envelopment sufficiently for stalled budding profiles to be observed, but not for long enough to cause a significant defect in virus replication.

Previous ultrastructural analysis of HSV-1 has largely been performed using TEM, which retains some advantages over cryoSXT. TEM offers higher resolution, allowing different components of the virus assembly intermediates to be unambiguously identified, such as the capsid, tegument, and envelope, thereby negating the need for fluorescence correlation with genomically-encoded tags. This allows HSV strains lacking fluorescent fusion proteins to be imaged, reducing the risk that the fluorescent proteins attached to structural proteins could interfere with virus assembly. However, 2D cross-sections of cytoplasmic envelopment events visualised by TEM, which appear as a capsid encircled by a tubular C-shaped endomembrane, are compatible with more than one 3D model of envelopment; techniques such as cryoSXT that allow greater volumes of the cell to be imaged than TEM are required to clarify the envelopment mechanism. Our 3D CryoSXT analysis reveals numerous budding events occurring in spherical/ellipsoidal vesicles and fully enveloped virions within spherical/ellipsoidal carrier vesicles (Fig. 7). We therefore propose a model for HSV-1 envelopment wherein the apparently C-shaped endomembranes observed in TEM represent the cross-sectional appearance of a spherical or ellipsoidal vesicle deformed by budding of a capsid into the vesicle lumen (Fig. 7). We also observed several instances of polarised arrays of capsids around spherical vesicles for mutant viruses with delayed envelopment (ΔpUL11, ΔpUL51, ΔgK, and ΔgE) (Fig. 8). To the best of our knowledge such observations have not been made in TEM studies of HSV-1 assembly with wild type or mutant viruses. This discrepancy may result from the low probability of capturing multiple capsids bordering a vesicle within a single TEM ultrathin section. Our cryoSXT imaging shows a greater density of material on the side of the vesicle membrane nearer the capsid (Fig. 8), a feature that has also been observed in TEM¹⁵. One advantage of our CLXT imaging is the ability to extend such insights by revealing the enrichment of specific viral proteins (exemplified in our data by gM-mCherry) at the capsid-proximal pole of the vesicle. The observations are consistent with the hypothesis that HSV-1 budding is asymmetric, being initiated at a ‘budding pole’ that is rich in tegument and glycoproteins⁶⁵. In the future, probes for different viral and host membrane proteins and specific lipid species in CLXT could shed light on membrane partitioning processes that occur during herpesvirus assembly.

An advantage of the relatively high-throughput analysis afforded by cryoSXT (compared to TEM and cryoET) is the ability to obtain robust particle size data in cellulo. No difference was observed in the measured width of nuclear capsids compared with membrane-proximal cytoplasmic capsids (Fig. 9 and Table 1), suggesting that there is only minimal recruitment of tegument around cytosolic capsids and/or that the tegument is too diffuse to produce detectable X-ray absorption. However, the average width of envelopment intermediates for ΔgE and ΔpUL51, where capsids are engaged with the membrane (‘embedded’), is only approximately 32.7 nm lower than that of fully enveloped extracellular enveloped virions (Fig. 9 and Table 1). This suggests that that the tegument layer becomes compressed around capsids immediately prior to, or concomitantly with, virus budding.

In conclusion, our multi-modal imaging strategy has provided novel ultrastructural insight into HSV-1 assembly, allowing the assembly trajectory of wild-type and mutant viruses to be observed in 3D. This revealed that envelopment occurs by the lumenal budding of capsids at spherical/ellipsoidal vesicles, rather than by wrapping of tubular membranes around capsids, and that tegument compression is concomitant with budding. Polarised arrays of capsids at cytoplasmic vesicles were observed for several mutants, suggesting envelopment is focused to one side of the endomembrane, and CLXT imaging suggested that these capsid-proximal surfaces are enriched in viral glycoproteins. Previously uncharacterised defects in nuclear egress were observed for HSV-1 lacking VP16, pUL16 and pUL21. Furthermore, comparative analysis reveals that deletion of VP16, pUL11, gE, pUL51 or gK cause distinct defects in cytoplasmic envelopment. Our data highlight the contributions that key HSV-1 envelope and tegument proteins make to virus assembly and underscore the power of correlative fluorescence and X-ray tomography cryo-imaging for interrogating virus assembly.

Materials and Methods

Reagents

Quantifoil^® 3 mm gold TEM grids with a holey carbon film (R 2/2, 200 mesh) were used as a substrate for cells prepared for cryopreservation. TEM grids were treated with poly-L-lysine (Sigma Aldrich). 150 nm gold fiducials were used to align cryoSXT projections (Creative Diagnostics Nanoparticle AF647).

Cell lines

U2OS cells (ATCC HTB-96; RRID CVCL_0042) and African green monkey kidney (Vero) cells (ATCC #CRL-1586) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM; Thermo Fisher Scientific) containing 10% (v/v) fetal bovine serum (FBS; Capricorn), 4mM L-glutamine (Thermo Fisher Scientific), and penicillin/streptomycin (10000 U/mL; Thermo Fisher Scientific). Hanks’ Balanced Salt Solution (HBSS; Thermo Fisher Scientific) and 0.25% Trypsin-EDTA (Thermo Fisher Scientific) were used to wash and detach adherent cells, respectively. Cells were maintained in a humidified 5% CO₂ atmosphere at 37°C.

Generation of mutants

Mutants were generated from a bacterial artificial chromosome carrying the KOS HSV strain⁹³. A mutant was previously reconstituted from this system containing eYFP-VP26 and gM-mCherry²⁹. This was used as the dfParental virus and the BAC from which it was derived was used as template for the mutagenesis of other mutants. A two-step red recombination system⁹⁴ was used to generate the mutants with the primers shown in Supp. Table 1. The ΔpUL11 virus was generated by introducing one stop codon at residue 3 of the UL11 open reading frame (ORF). The ΔpUL16 virus was generated by introducing three stop codons at residue 15 of the UL16 ORF. The ΔpUL21 virus was generated by introducing three stop codons at residue 23 of the UL21 ORF. The ΔUL34 virus was generated by deleting codons 1-228 from the UL34 ORF. The ΔVP16 virus was generated by deleting codons 1-478 from the UL48 ORF. The ΔpUL51 virus was generated by introducing three stop codons at residue 21 of the UL51 ORF. The ΔgK virus was generated by introducing three stop codons at residue 54 of the UL53 ORF. The ΔUS3 virus was generated by deletion of the entire coding sequence of US3. The ΔgE virus was generated by introducing three stop codons at residue 21 of the US8 ORF. Mutant BACs were transfected, together with a Cre recombinase expression plasmid (pGS403) to excise the BAC cassette from the KOS genome, into 6-well plates containing 70-80% confluent Vero cells to reconstitute the viruses, other than ΔUL34, ΔVP16, and ΔgK, which were transfected into the respective complementing Vero-modified cell lines: UL34CX, 16_8, and VK302^61,95,96. Successive stocks were generated by infecting Vero cells or complementing cells at MOI = 0.01. Cells were harvested once all cells demonstrated cytopathic effect, freeze-thawed and sonicated at 50% amplitude for 40 seconds in a cuphorn sonicator. Final stocks were clarified by centrifugation at 3,200×g for 5 minutes in a benchtop centrifuge, divided into 10–20 μL aliquots and were stored at -70°C. All virus stocks were titrated on U2OS cells, Vero cells, or Vero-modified complementing cells.

Infection assays

For the immunoblots, Vero cells (Fig. 1C–D) and U2OS cells (Supp. Fig. 7B) were seeded in 6-well plates, were left to reach 70–80% confluency, were infected at MOI = 5 with indicated viruses in a minimal volume of medium (500 μL) and were incubated in a 5% CO₂, 37°C incubator. After 1 hour, the inoculum was diluted to 2 mL with fresh medium and cells were incubated overnight. For the single-step replication curves (Fig. 2A, Supp. Fig. 2, and Supp. Fig. 7C), U2OS cells were seeded in 24-well plates at a density of 1×10⁵ cells per well and were infected the next day at MOI = 2 with the indicated viruses in a minimal volume of medium (250 μL) for 1 hour in a 5% CO₂, 37°C incubator. After 1 hour, the inoculum was aspirated off the cells and the cells were treated with citric acid solution (40 mM citric acid pH 3, 135 mM NaCl, and 10 mM KCl) for 1 minute to inactivate unabsorbed virus. Citric acid solution was subsequently aspirated off the cells and the samples were washed thrice with 500 μL PBS before adding 500 μL fresh medium. For the plaque size assays (Fig. 2B–C and Supp. Fig. 7D–E), U2OS cells were seeded on 6-well plates, were left to reach ∼90% confluency, and were infected with a low titer of the indicated viruses (calculated to produce an average of 30 plaques per well). Cells were incubated with reduced serum (2% v/v) medium supplemented with 0.3% high viscosity carboxymethyl cellulose and 0.3% low viscosity carboxymethyl cellulose for 72 hours. For cryoSIM and cryoSXT experiments (Fig. 3–8 and Supp. Fig. 4–5), TEM grids were treated by glow discharge and were incubated in filtered poly-L-lysine for 10 minutes in 6-well plates as described previously^26,54. Poly-L-lysine was aspirated off and U2OS cells were seeded onto the holey-carbon coated side of the grids at a density of 3×10⁵ cells per well. After overnight culture in a 5% CO₂ and 37°C incubator, the cells were infected at MOI = 2 with indicated viruses in a minimal volume of medium (500 μL) and were incubated in a 5% CO₂, 37°C incubator. After 1 hour, the inoculum was diluted to 2 mL with fresh medium and were incubated for an additional 14.5 hours. Samples were washed with twice with 1 mL serum-free medium and were then overlain with a staining solution containing 50 nM MitoTracker Deep Red (Thermo Fisher Scientific) in serum-free medium. For samples infected with the dfParental, ΔpUL16, ΔpUL21, ΔUL34, ΔVP16, and ΔUS3 viruses, the staining solution also contained 2 μg/mL Hoescht33342 (Thermo Fisher Scientific). Samples were washed twice with serum-free medium after 30 minutes of staining and the grids were loaded into a Leica EM GP2 plunge freezer set to 80% humidity. A working solution of gold fiducials was prepared by centrifugation of a 1 mL stock (provided in the Reagents section) at 12×g for 5 minutes at RT. The pellet was resuspended in HBSS and the working solution was sonicated at 80 kHz (100% power) and 6°C to disperse clumps of gold fiducials. A 2 μL solution of the working solution was overlain onto the holey-carbon coated side of the grids, which were then blotted for 0.5–1 seconds on the opposite side using Whatman paper. Grids were then immediately plunged into liquid nitrogen-cooled liquid ethane and were transferred into storage containers maintained under liquid nitrogen.

For immunoblots, infected cells were washed twice with 1 mL PBS, scraped off 6-well plates, and were centrifuged at 2000×g for 5 minutes. Pellets were isolated and lysed with a solution of Complete Protease Inhibitor without EDTA (Roche) diluted 1 in 10 in a lysis buffer (Sigma Aldrich) on ice for 20 minutes. Insoluble material was removed by centrifugation at 20,000×g, 4°C for 10 minutes. Supernatants were resuspended in SDS-PAGE loading buffer supplemented with 2-mercaptoethanol. To immunoblot pUL20 and gE, the mixtures were heated to 42°C for 20 minutes. To immunoblot the other proteins, separate mixtures were boiled in a water bath for 5 minutes. Samples were resolved on SDS-PAGE gels alongside a Blue Protein Standard Broad Range ladder (New England BioLabs). Bands were transferred onto nitrocellulose membranes and were blocked with a 5% (w/v) solution of milk powder in PBS. The following primary antibodies were used: anti-pUL11⁵⁹ at 1:1000, anti-pUL16⁹⁷ at 1:1000, anti-pUL21³³ at 1:1, anti-pUL34⁹⁸ at 1:500, anti-VP16⁹⁹ at 1:10, anti-pUL51 (3D3)⁴⁹ at 1:1, anti-pUL20⁴⁷ at 1:1000, anti-pUS3¹⁰⁰ at 1:1000, anti-gE¹⁰¹ at 1:10, anti-VP5¹⁰² at 1:10, and anti-GAPDH at 1:1000 (GeneTex, GTX28245). Membranes were stained with the following secondary antibodies: IRDye 680T conjugated goat anti-rat (926–68029), donkey anti-rabbit (926–68023) or goat anti-mouse (926–68020), or LI-COR IRDye 800CW conjugated donkey anti-rabbit (926–32213), donkey anti-chicken (926–32218), or goat anti-mouse (926–32210). Primary and secondary antibody solutions were generated in PBS-T supplemented with 0.5% milk powder, membranes were washed 4 times for 5 minutes on a rocking platform with PBS-T after each round of antibody staining. For membranes immunoblotted for pUL16 or pUL21, TBS and TBS-T were used instead of PBS and PBS-T. An Odyssey CLx Imaging System (LI-COR) and Image Studio Lite Software (LI-COR) were used to visualise immunoblots.

Single-step replication curves

Infected samples were prepared as described in Infection Assays and were transferred to -70°C storage at 2, 6, 9, 12, and 24 hours post infection. After at least two hours of storage at -70°C, the samples were thawed at 37°C. This process of freeze-thawing was repeated and the thawed samples were scraped off the 24-well plates using the blunt end of a 1mL plunger and were transferred into 1.5 mL microcentrifuge tubes for -70°C storage for at least two hours. 10-fold serial dilutions of the samples were performed and titrated on Vero cells as previously described⁴⁹. To logarithmically transform the PFU data, 0 values on the exponential scale were converted to 1.

Sequencing and alignment

Working stock solutions of ΔgK and dfParental HSV-1 were prepared for Sanger sequencing by adding 10 μL of virus stock to 200 μL of a Quantilyse solution¹⁰³. Samples were heated to 55°C in a thermal cycler for 2.5 hours, 85°C for 45 minutes, and were left on hold at 10°C. 2 μL of these solutions were used as templates for PCR amplification with custom primers (Supp. Table 2). Amplicon products were purified (EconoSpin, Epoch Life Science) and analysed by Sanger Sequencing with custom sequencing primers (Supp. Table 3). The HSV-1 strain KOS UL53 sequence (encoding gK) was aligned with the sequencing results using Clustal Omega and were visualised using Jalview^104–106.

Plaque size assays

Infected cells were prepared as described in Infection Assays and were washed with 1 mL PBS at 72 hpi. Samples were fixed with 1 mL 4% formaldehyde for 10 minutes and were washed with 1 mL PBS. Samples were blocked for 30 minutes with a solution of 5% FBS and 0.1% Tween-20 in PBS on a rocking platform. Samples were incubated for 1 hour with a 1:10 dilution of anti-gD (LP2) in blocking buffer on a rocking platform¹⁰⁷. Samples were washed with blocking buffer 3 times and were incubated for 1 hour with a goat-derived anti-mouse IgG (H+L) secondary antibody conjugated to horseradish peroxidase (Thermo Fisher Scientific; product no. 31430) at 1:1000 in blocking buffer. Samples were washed with blocking buffer 3 times and once with PBS. Plaques were detected using either TrueBlue peroxidase substrate (Seracare) according to the manufacturer’s instructions (Supp. Fig. 7D–E) or with ImmPACT DAB peroxidase substrate (Vector SK105) according to the manufacturer’s instructions (Supp. Fig. 3). Images were captured using an EPSON scanner V600 at 1200 dpi. Plaque area (in pixels) was measured using Fiji by applying an intensity threshold to images of each plaque, creating binary masks with a value of 0 for background and 1 for plaque pixels^108,109. Plaque area was determined by automated counting of each pixel within a plaque.

CryoSIM

Cryopreserved TEM grids were placed onto a liquid nitrogen cryostage (Linkam) and were imaged by cryoSIM as previously described⁵⁶. The cryoSIM was developed in-house as previously described¹¹⁰. Hoescht stain fluorescence was excited using a 405 nm laser and was detected using an EM-452-45 filter (452 ± 22.5 nm). eYFP-VP26 fluorescence was excited using a 488 nm laser and was detected using an EM-525-50 filter (525 ± 25 nm). gM-mCherry fluorescence was excited using a 561 nm laser and was detected using an EM-605-70 filter (605 ± 35 nm). MitoTracker Deep Red fluorescence was excited using a 647 nm laser and was detected using an EM-655-lp filter (≥ 655 nm). SIM data were reconstructed using SoftWoRx (AppliedPrecision Inc., Issaquah, WA) and the fluorescent channels were aligned using Chromagnon¹¹¹.

CryoSXT

Cryopreserved TEM grids were loaded into a liquid nitrogen-cooled vacuum chamber of an UltraXRM-S/L220c X-ray microscope (Carl Zeiss X-Ray Microscopy) at beamline B24 of the Diamond Light Source⁶⁶. Incident soft X-rays generated at the synchrotron (500 eV, λ = 2.48 nm) were used to illuminate the samples and transmitted X-rays were detected using a 1024B Pixis CCD camera (Princeton Instruments). Transmitted light was focused using a 25 nm zone plate objective with a nominal resolution limit of 25 nm. Samples were focused by Z translations of the zone plate and samples were centred along a rotational axis by Z translations of the sample grid. Individual X-ray projections (9.46×9.46 μm) were captured and tiled together in 7×7 montages known as X-ray mosaics to inspect sample quality and identify regions of interest for tomography. Tomographic data were collected at fields of view measuring 9.46×9.46 μm. In each case, a collection of X-ray projections known as a tilt series were acquired by rotating the sample with maximum tilt angles of –60°/+60° and acquiring images at increments of 0.2° or 0.5°. A 0.5 or 1 second exposure time was used depending on the intensity of the transmitted X-rays. Tilt series were reconstructed using IMOD (version 4.9.2)¹¹² as previously described²⁶.

Correlation of cryoSIM and cryoSXT

CryoSIM data was correlated onto CryoSXT data using easyCLEMv0¹¹³ as previously described¹¹⁴ with a few differences. CryoSXT tomograms were used as the target for the correlation and a frame was added around the tomogram, increasing the XY dimensions to 1200×1200 voxels. This increased the canvas size of the transformed cryoSIM data and reduced the probability that edges of the cryoSIM data would overlap with the tomogram at the centre of the frame during XY translations and rotations of the cryoSIM data. To avoid the need for an affine transformation model, maximum Z projections of cryoSIM data were used to generate transformation files using the rigid transformation model. The X-ray mosaic and a minimum Z projection of the tomogram were used as targets for these transformations. Rigid-transformation files generated using the 2D maximum Z projections of the cryoSIM data were applied to the 3D Z stacks to prevent the anisotropic stretching of signal generated by affine transformations in lieu of rotation around the X or Y axis. Before conducting a 3D rigid transformation of cryoSIM Z stacks onto the target tomogram, the number of slices in the cryoSIM image was increased 5-fold (reducing the voxel depth from 125 nm to 25 nm). This made it easier to correlate the front and back edges of mitochondrial fluorescence and tomographic mitochondria together. The TransformJ plugin in Fiji was used to further fine-tune the correlation where needed¹⁰⁹.

Quantitation of nuclear egress

3D CryoSIM images were collected from TEM grids containing U2OS cells infected at MOI = 2 with mutants as described in Infection Assays. A Hoechst 33342 stain was used to label the nucleus. Reconstructed images of eYFP-VP26, gM-mCherry, and Hoescht stain were registered together and maximum intensity projections in Z were generated using Fiji. Edges of the nucleus and plasma membranes were delineated using the Hoescht and gM-mCherry fluorescence, respectively, to determine nuclear and cytoplasmic regions of interest. Cells infected with the ΔgE virus were used as a negative control for attenuation in nuclear egress. As these cells did not contain a Hoescht stain, the borders of the nuclei in the X-ray mosaics were used to demarcate the nuclei in the fluorescence. A threshold was applied to the eYFP-VP26 maximum projection to filter out background and isolate pixels containing capsid fluorescence. The ratio of eYFP-VP26⁺ pixels was measured between the nuclear and cytoplasmic regions of interest. A total of 95 infected cells were included in the analysis. Infected cells were included in the analysis if the JAC (determined by a concentration of gM-mCherry vesicles) was not located over or under the nucleus with respect to the XY plane. Exclusion criteria included cells with indistinguishable boundaries from each other, cells with a faint Hoescht nuclear stain, or cells that had a majority of their area located outside the field of view.

Quantitation of cytoplasmic clustering

3D CryoSIM images were collected from TEM grids containing U2OS cells infected at MOI = 2 with mutants as described in Infection Assays. Reconstructed images of eYFP-VP26 and gM-mCherry were registered together using Chromagnon¹¹¹. Borders of the JAC were determined based on the relative intensity of the gM-mCherry fluorescence. Binary masks of eYFP-VP26 and gM-mCherry fluorescence were generated by applying thresholds that filter out background and noise. The total number of retained pixels in the borders of the JAC were counted for the two channels and an eYFP-VP26 / gM- mCherry ratio was produced. In order to calculate colocalization between these channels, a spatial coincidence-based method known as Manders correlation was used instead of intensity-based methods, such as Pearson’s correlation, to minimize the influence of noise, background, and cryoSIM reconstruction artefacts. Thresholds on intensity were applied to the eYFP-VP26 and gM-mCherry fluorescent channels using Fiji, and the number of fluorescent pixels were quantitated from masks^108,109. Overlap between the channels was determined using the subtract function in Fiji. To calculate M1 values, masks of gM-mCherry fluorescence were subtracted from eYFP-VP26 masks, and the remaining pixels were quantitated and subtracted from the total pixels in the eYFP-VP26 mask to determine the number of eYFP-VP26 pixels that overlapped with gM-mCherry. M1 values were calculated in Excel (Microsoft) by dividing the number of overlapping pixels by the total number of eYFP-VP26 pixels in the region of interest. To calculate M2 values, the same process was performed but with the two channels switched. Exclusion criteria included cells where the JAC overlapped with the nucleus or cells where the JAC was partially excluded from the field of view.

Quantitation of membrane intensity

The membrane intensity of vesicles surrounded by capsid arrays was measured from signed 16-bit tomograms generated by WBP without reducing noise by applying a SIRT-like filter. The minimum voxel intensity was measured from 3×3 voxel matrices at 30 points around the vesicle. Data were collected from 3 tandem projection planes spanning 30 nm in depth using the same 30 XY coordinates and the mean ± SD were graphed (Fig. 8E–F and Supp. Fig. 7B). Of the 30 coordinates assessed, those nearest to proximal capsids were marked and this was used to draw a boundary between capsid-proximal and capsid-distal sides of the vesicle.

Segmentation

Segmented volumes were generated manually using Segmentation Editor in Fiji¹⁰⁹. The X-ray absorbance heatmap (Fig. 8G–H) was generated by applying the Fire lookup table to the tomogram in Fiji and superimposing it onto the segmented volume of the vesicles¹⁰⁹. Segmented volumes were rendered in 3D using 3D Viewer in Fiji¹⁰⁹. Segmented vesicles (Fig. 7B–C, Fig. 8A–D, I and Supp. Fig. 7A) appeared open-ended at the anterior and posterior faces with respect to the imaging plane because cryoSXT produced less contrast in those regions compared with side regions perpendicular to the imaging plane, making it impossible to fully segment the vesicles without extrapolation. This contrast discrepancy was not due to a limitation in the tomogram’s depth as cryoSXT captured the entire volume of cells within each field of view but was due to a mechanical constraint in the X-ray microscope. During tilt series collection, the sample could only be rotated by a 120° range (–60° to +60°) rather than a desired 180° range (–90° to +90°) to avoid collisions with other components in the microscope. During acquisition, the vesicle’s side regions were positioned parallel with the X-ray beam at 0°, maximising X-ray absorption and contrast. As the anterior and posterior faces perpendicular to the side regions were never positioned at ±90°, they were never parallel with the X-ray beam, reducing absorption and contrast in those regions. The open ends all share the same orientation, confirming that they are artefacts of the imaging setup rather than bona fide features.

Graphs and statistics

Single-step replication curves (Fig. 2A, Supp. Fig. 2, and Supp. Fig. 7) and graphs of membrane intensity (Fig. 7E–F and Supp. Fig. 7B) were generated in Excel (Microsoft). Relative plaque area graphs (Fig.2B and Supp. Fig. 7E), the nuclear/cytoplasmic capsid ratio graph (Fig. 4B), the eYFP-VP26/gM-mChery ratio graph (Fig. 5D), the Manders correlation graphs (Fig. 5E and Supp. Fig. 4) and the graph showing the width of virus particle intermediates (Fig. 9) were generated using the ggplot2 package¹¹⁵ in R studio¹¹⁶. The graph showing the width of vesicles surrounded by capsid arrays was generated using SuperPlots¹¹⁷ (Supp. Fig. 6C). Single-step replication curves report the mean Log₁₀(PFU) values from two technical repeats and the error bars indicate range. For the graphs of membrane intensity for which the data was normally-distributed, a two-tailed t-test was used to assess the significance of differences between voxel intensity between the capsid-proximal and capsid-distal sides of the vesicle (Fig. 8E–F and Supp. Fig. 7E). Mann-Whitney U tests were used to assess significant differences for non-normally distributed data (Fig. 4B, Fig. 5D–E, Fig. 9, Fig. 2B, Supp. Fig. 4, and Supp. Fig. 7E). The boundary was determined by the location of the capsids. The width of vesicles surrounded by capsid arrays was measured using Contour⁶⁸.

Data availability

Original imaging data for tomograms illustrated in the manuscript will be deposited in the University of Cambridge Apollo Repository and representative tomograms will be published in the EMPIAR repository (EMBL-EBI) upon acceptance.

Acknowledgements

We thank Diamond Light Source for access to beamline B24 (proposals MX18925, MX19958, BI21485, BI23508, BI25247, BI26657 and BI30442) and the experimental hall coordinators for helpful support. We thank members of beamline B24 at the Diamond Light Source (Thomas Fish, Archana Jadhav, Mohamed Koronfel, Ilias Kounatidis, Chidinma Okolo, Matt Spink, and Nina Vyas) for technical support with cryoSXT and cryoSIM. We thank the DNA Sequencing Facility at the Department of Biochemistry, University of Cambridge for their support. We thank Mike Hollinshead (University of Cambridge) for assistance with electron microscopy of HSV-1 infection. This work was supported by a PhD studentship co-funded by Diamond Light Source and the Department of Pathology, University of Cambridge, to KLN, by a Sir Henry Dale Fellowship, jointly funded by the Wellcome Trust and the Royal Society (098406/Z/12/B) to SCG, and by a Biotechnology and Biological Sciences Research Council (BBSRC) Research Grant (BB/M021424/1) to CMC. For the purpose of Open Access, the authors have applied a CC BY public copyright licence to any Author Accepted Manuscript (AAM) version arising from this submission.

Additional files

Supplementary data