SARS-CoV-2 replication induces profound cytopathic effects in host cells
To image and investigate SARS-CoV-2 replication in near-native cell context, we infected Vero cells grown on indexed EM grids with SARS-CoV-2 at 0, 0.1 and 0.5 multiplicity of infection (MOI). At 24 hours post infection (hpi), the cells were fixed with 4% paraformaldehyde and plunge frozen in liquid ethane. As illustrated in the workflow (Figure S1), cryoEM grids containing SARS-CoV-2 infected cells were first imaged in a Titan Krios to identify each individual infected cell (39.2 % for MOI of 0.1 and 45.4% for MOI 0.5) where cryoET tilt series were collected first at the cell periphery. The grids were then transferred to a FIB/SEM dualbeam instrument and the exact same infected cells were imaged with serial cryoFIB/SEM volume imaging 36 or cryoFIB milling of cellular lamellae at the target region where additional cryoET tilt series were collected 37. Alternatively, we imaged the same infected cells on cryoEM grids by soft X-ray cryo-tomography 38. These imaging modalities provide the necessary structural and ultrastructural information at different length scales to visualise the infecting viruses in their cellular context and are highly complementary. Indeed, such a unique approach enabled the direct visualisation of the SARS-CoV-2 replication and cytopathic effects in a multi-modal, multi-scale and correlative manner.
Compared to uninfected cells (Figure S2, Movie 1), serial cryoFIB/SEM images of SARS-CoV-2 infected cells display an extensive array of cytopathological alterations throughout the entire cell, as illustrated in Figure 1 and Movies 2-5. At the cell surface, there were many virus-containing membrane tunnels extending deep into the cell (Figure 1A, “T”, Movie 2), resembling those in HIV-1 infected microphages 39. CryoET confirms the presence of SARS-CoV-2 particles just outside of cells and in membrane tunnels. In addition, virus particles were also found within intracellular membrane vesicles that are not connected to cell membrane (Figure 1A, red arrow). Deep into the cell, we found that much of the cytoplasm (Fig. 1B), especially the paranuclear region (Fig. 1C), is occupied with abundant membrane compartments of different morphologies, including numerous vesicles (“V”), the endoplasmic reticulum (ER) and the nucleus (“Nuc”) (Movies 2-4). CryoET of cell lamellae containing these vesicles confirmed that they are the so-called “double membrane vesicles” (DMVs) (Fig. 2A and B) where viral genome replication takes place 6. Nuclear pores are clearly distinguishable in both SARS-CoV-2 infected and uninfected cells (Figure 1C, Figure S2A and C, blue arrows). We frequently found electron-dense complex membrane compartments in infected cells (Figure 1B, pink arrows). A more striking feature observed in infected cells is the cytopathic damage to the nucleus compared to the control cells (Figure 1D, Figure S3G), where, in extreme cases, nearly a half of the nucleus has been taken up by the invaginated cytoplasm (Movie 5). Such cytoplasm invagination was also noticed in a conventional EM image of stained plastic sections of SARS-CoV-2 infected cells 40.
Exploiting high throughput whole-cell imaging capability offered by soft X-ray cryo-tomography, we analysed many targeted infected and uninfected cells that had been identified and imaged by cryoEM/ET at the cell periphery. At the whole cell level, soft X-ray images show substantial mitochondria morphological changes. We observed that long tubular shaped mitochondria in the uninfected cells (Figure S3A, C, yellow arrows) have been mostly disrupted in the infected cells (Figure S3B, D). Consistently, we observe numerous vesicles at perinuclear regions (Figure S3F) and cytoplasmic invaginations (Figure S3G) in the infected cells.
SARS-CoV-2 RNA synthesis and transport
The first step in SARS-CoV-2 production is viral genome replication. Coronaviruses have evolved a sophisticated RNA replication strategy for the generation of the genomic negative-sense and subgenomic RNAs, which relies heavily on double stranded RNA intermediaries, a potent activator of RIG-I and MDA-5 41-43. Thus, cellular compartmentalization of RNA transcripts serves as an innate immune evasion strategy. DMVs are induced during the replication of a variety of RNA viruses 4,5,7-9 and were identified as the sole compartment where viral RNA transcription occurs for coronaviruses 6. Indeed, cryoET of cell lamella revealed that abundant intracellular vesicles observed in the 3D volume of infected cell (Figure 1B, Movies 2-4) are DMVs likely containing viral RNA transcripts as previously suggested 6,44 (Figure 2A-C, Movie 6). There are also a substantial amount of vesicle packets (VPs) (Figure 2A) 29, apparently resulting from the fusion of the outer membranes of DMVs. Since the sample was cryofixed 24 hours post infection, this is consistent with a previous observation that the number of VPs increases with the time of infection 6. Until very recently, DMVs were thought to be completely enclosed, which raised the question of how the viral mRNAs could gain access to the cytoplasm to be translated. We observed several double-membrane-spanning pore complexes in DMVs (Figure 2B-D, yellow arrow), resembling the RNA transport portal observed in DMVs of murine hepatitis coronavirus (MHV) infected cells in a recent study 26. However, the portal appears rare in DMVs of SARS-CoV-2 (total 9 portals from 24 DMVs) compared to those of MHV (average 8 portals per DMV, or 192 for 24 DMVs) (Wolff et al., 2020a), signifying a difference between coronaviruses.
SARS-CoV-2 assembly and budding
The translation of the subgenomic vRNAs gives rise, amongst others, to the structural proteins N, M, E and S, which are required for assembly. M, E and S are membrane-associated proteins and are localized to the ER, Golgi and the ERGIC 30,44. The N protein associates with the genomic vRNA and M protein, which presumably drives vRNA packaging and genome encapsidation 45,46. The main assembly and budding site of other coronaviruses has been previously described at the ERGIC by conventional EM of stained plastic sections 7,28,30,44. In serial cryoFIB/SEM images of SARS-CoV-2 infected cells, we observed vesicles containing virus particles (Figure 3A, black arrows), along with a string of small dense vesicles (Figure 3B, pink arrow) lining along the vesicle membrane. The same architecture was captured by high-resolution cryoET of cell lamella from a similar perinuclear region, which shows these are in fact SARS-CoV-2 assembly and budding sites (Figure 3C-E). CryoET and subtomogram averaging further revealed that the small dense vesicles are SARS-CoV-2 spike containing transport vesicles (Figure S4), possibly supplying newly synthesized spikes and other viral components via fusion with the single membrane vesicles (SMVs) where viral assembly takes place (Figure 3C-E pink arrows, Movie 7). Indeed, spikes are observed on SMV membranes sparsely distributed or otherwise clustered at the assembly sites (Figure 2C, Figure 3C-E, red arrows, Movie 7). Interestingly, several SARS-CoV-2 assembly intermediates were observed within a single tomogram from a cell lamella (Figure 3C-E, blue arrows, Movie 7), along with fully assembled virus particles released into SMVs (Figure 3C-E, black arrows, Movie 7), thus capturing the assembly and budding process of SARS-CoV-2. It is conceivable that upon fusion of transport vesicles with the SMV, spikes are readily diffused on the SMV membrane. They cluster when interacting with N-associated vRNA, possibly via M protein 19,45, which initiates the assembly and budding process that finally releases the viral particle into the SMV. Consistent with this, spike clusters are observed exclusively associated with the agglutination/gathering of electron dense material, which presumably represents viral genome. Noticeably, the virus assembly site is frequently present in the vicinity of RNA portals in DMVs (Figure 2B-D, yellow arrows), potentially facilitating the assembly process.
Most virus particles are found in SMVs, some contain a single virion, while others encompass multiple virions (Figure 3E-F). CryoET and subtomogram averaging of 450 spikes from these particles yielded a density map at 11 Å resolution (at 0.143 FSC cut-off) by emClarity 47 (Figure 3G, Figure S5). The averaged density map resolves the overall spike structure, which overlaps very well with prefusion spike atomic models 12-17 (Figure 3G). Some virus particles were also observed in electron-dense complex membrane compartments (CMC) (Figure 4). There were two types of virus particles: Viruses protected by single membrane vesicles (SMV) in CMC show prefusion spikes (Figure 4D, G-H); viruses in the lumen of CMC, however, have either no spikes (Figure 4B) or a few postfusion spikes on their surfaces (Figure 4E-F). These virus particles could be off-pathway viral assemblies (in the case of SMV-protected viruses displaying prefusion spikes), or remnants of late endosomes from viral entry or lysosomes for viral degradation. The fact that the spike proteins are in the postfusion state suggests that proteolytic processing has taken place in these compartments resulting in S1 shedding. Therefore, we suggest that assembly at the SMVs is the only pathway which will lead to infectious viral progeny.
There has not been much detailed studies on how SARS-CoV-2 viruses are released from cell. We investigated SARS-CoV-2 egress using both serial cryoFIB/SEM volume imaging and cryoET. CryoFIB/SEM images reveal abundant virus exiting tunnels in 3D at the cell periphery connecting to cell membrane (Figure 5A-B, Movie 2). This likely resulted from the fusion of very large multi-virus containing vesicles with cell membrane, i.e. egress through exocytosis-like mechanism. Consistent with cryoFIB/SEM analysis, we also observed virus exiting tunnels in cryo-tomograms (Figure 5C). The fact that these compartments often contained many viral particles suggests that this is a snapshot of viral exit, rather than cellular entry.
In addition to exiting through tunnels, we also frequently found plasma membrane discontinuities, or membrane lesions, next to virus particles outside the cell (Figure 5E). The membrane lesions are mostly discrete in appearance, which argues against them being an artifact of sample preparation. There were 116 membrane lesion sites in 74 tomograms, and 44.6% of tomograms show cell membrane lesions in infected cells (Figure 5D-E, Movies 8-9), whereas 18.7 % tomograms from uninfected cells display similar but more sparse membrane lesions (10 membrane lesion sites from 16 tomograms). Close inspection of individual membrane lesions indicates that the underlying cytoskeleton, such as actin filaments, is largely retained (Figure 5E inset). The fact that we observed similar membrane lesions, but to a lesser extent in control cells, suggests that SARS-CoV-2 may exploit the host cell machinery for its egress. It is unclear whether the cell can recover from such membrane wounds, or if exit through membrane lesion is a sign of late infection and will eventually lead to cell lysis and death.
CryoET subtomogram averaging of 7090 spikes from extracellular virus particles yielded a density map at 8.7 Å resolution (at 0.143 FSC cut-off), which represents the prefusion state (Figure 5F, Figure S5). Spike structures from intracellular and extracellular viruses agree with each other very well (Figure 5G), suggesting that no further substantial structure rearrangement takes place for viral spikes from assembly to egress, given the current resolution. While all previous spike structures are either from recombinant proteins or from purified inactivated virus particles 10-17,32-34, the two spike structures presented here were derived directly from infected cells in the cellular context, and thus represent the closest to the native condition.