A 62-year-old male was admitted to Shenzhen Third People's Hospital on January 15, 2020, with pneumonia and was further diagnosed as COVID-19 positive. An epidemiological investigation confirmed a Wuhan travel history between January 1 and January 14 for this patient, and symptoms began on January 11, 2020, including fever and cough. Testing for common respiratory viruses, including influenza A virus, influenza B virus, adenoviruses, human parainfluenza virus, and other human coronaviruses, was negative. Lymphopenia, elevated C-reactive protein, and elevated interleukin-6 were found upon admission (Table S1). Computed tomography scans showed multiple ground-glass opacities in bilateral lungs at the early stage, and lung consolidation occurred during hospitalization (Figure 1A).
The bronchoalveolar lavage fluid sample was collected and subjected to next-generation sequencing. The genome sequence of this virus was submitted to the Global Initiative on Sharing Avian Flu Data under accession number EPI_ISL_406594 and designated as “BetaCoV/Shenzhen/SZTH-003/2020.” Phylogenetic analyses showed that the virus possessed a high homology to two isolates, BetaCoV/Wuhan/IPBCAMS-WH-04/2019 from Wuhan and SARS-CoV-2/NTU01/2020/TWN from Taiwan (Figure S1).
Virus from the patient was propagated and isolated using Vero cells in the BSL-3 (biosafety level 3) laboratory. Typical cytopathic effects were observed 4 days postinoculation in Vero cells, including cell rounding, shrinkage, lysis, and detachment throughout the cell monolayers (Figure 1B). Using the patient's plasma, virus could be detected in cultured cells by immunofluorescence (Figure 1C). Viral RNAs were observed in the cell culture supernatant when probed using a China Food and Drug Administration-approved commercial kit targeting the open reading frame (ORF) 1ab (red in Figure 1D) and N (blue) genes of SARS-CoV-2 with low cycle threshold values (Figure 1D). The purified SARS-CoV-2 particles showed specific reactivity to convalescent plasmas from SARS-CoV-2-infected patients (Figure 1E) and to specific human monoclonal antibodies against the RBD of the S protein using ELISA (Figure 1F). Although both hmAb-1 and hmAb-2 are RBD-specific human monoclonal antibodies, hmAb-1 (named B38) has a relatively higher binding affinity to RBD, with a Kd of 4.48 nM (
), compared with hmAb-2 (named P1A-1D1), with a Kd of 260.50 nM (
). The ELISA data suggest a certain amount of RBD (or S1 spike) is present in the purified virus particles.
The isolated and concentrated SARS-CoV-2 virus particles were inactivated with β-propiolactone for structural characterization, initially by negative staining EM followed by cryo-EM and cryo-ET. SARS-CoV-2 displays typical morphology of a coronavirus, where spike proteins decorate the surface of the viral particles (Figure S2A). A close examination of the spikes suggested the presence of two types of spikes; one appears larger, with a club head, resembling the prefusion spike trimer, whereas the other is much thinner and nail shaped, and resembles the postfusion spike containing only the S2 trimer (
) (Figure S2C). Both types were observed in individual virus particles. Cryo-EM images of purified SARS-CoV-2 virus particles showed that the virus particles are roughly spherical or moderately pleiomorphic, with an average diameter of 108 ± 8 nm, ranging from 84 to 126 nm. The viral spikes and viral membranes are clearly discerned. About 20%–30% of the virions have abundant spikes around the envelope, whereas many other virus particles display few spikes (Figure S2). Consistent with the observation in negative-stain EM, cryo-EM images showed that most spikes are nail shaped, presumably in a postfusion state (Figure 2). Projection images from negative-staining virus particles may not allow for a quantitative measure of the number of spikes. They are, nevertheless, a quick and easy way to assess sample quality. One can already raise a flag if postfusion spikes are observed in negatively stained images, which would prompt a more careful 3D analysis.
Cryo-ET and Subtomogram Averaging of SARS-CoV-2 Postfusion Spike
The 3D nature of cryo-ET allows analysis of features that would normally be obstructed in cryo-EM projection images (
). Detailed parameters for tomography data collection and processing are summarized in Table S2. The spikes in the tomograms confirmed the findings from negative staining and cryo-EM, showing that the spikes adopt a conformation consistent with the postfusion state of the S2 protein (Figures 3A and 3B ). Frequently, the spikes were arranged in clusters (Figures S2B and 2D). Among 125 spikes from 19 virus particles analyzed from high defocused tomograms, we observed 32 prefusion and 93 postfusion spikes from which S1 had dissociated. The postfusion spikes account for 74% of total spikes in the β-propiolactone-treated SARS-CoV-2 particles (Figure 3C). Virus tomograms also showed the ribonucleoprotein complexes organized inside the viruses (Figures 3A–3C).
Despite the determination of many structures of prefusion SARS-CoV-2 spike alone and in complex with its receptor and antibodies, the structure of the postfusion SARS-CoV-2 spike is still lacking. A recent study showed the structure of postfusion SARS-CoV spikes from purified recombinant S proteins using single-particle cryo-EM (PDB: 6M3W) (
). To obtain an in situ 3D structure of the postfusion SARS-CoV-2 spike on the native virion, we carried out cryo-ET and subtomogram alignment and averaging (
). A set of 956 manually picked nail-shaped spike subvolumes from 242 tomograms were aligned and averaged iteratively with a 3-fold symmetry applied using the programs Dynamo and emClarity (
). The resulting map at 11 Å resolution showed the characteristic postfusion conformation of coronavirus spikes (Figures 3D and S3): a 240 Å tall structure, 66 Å at its widest point, where the connector domains and core β sheets are, and 27 Å at its narrowest point, near heptad repeat 1 (HR1). The atomic model of postfusion SARS-CoV (PDB: 6M3W) (
), which shares 89.97% sequence identity with SARS-CoV-2 at the S2 region (Figure S4), fit well into the subtomogram averaged density map. A significant portion of density at the bottom of the spike close to the membrane was not modeled (Figure 3C). This region could account for about 224 residues that were not resolved in the SARS-CoV structure (Figure S4). Protrusions on the side and top of the map are consistent with glycosylation moieties, in particular at the N1194 and N1173 sites (Figure 3D). It should be noted that these glycan densities are more extensive than what was modeled in PDB: 6M3W (just one sugar) (
). Glycosylation is expected to be complex and variable and often contains a chain of multiple sugars. Although the subtomogram average shows a continuous density that includes the viral membrane, the lipid membrane bilayer is not well resolved. This indicates that the postfusion spikes are flexible and have some confined variable orientations relative to the viral membrane. Due to the limited number of prefusion spikes available on the virus particles, a high-resolution subtomogram average of the prefusion spike was not obtained.
The relationship between receptor binding, priming proteolytic cleavage, and the pre- to postfusion spike transition is still not clear. Here, we have shown that isolated SARS-CoV-2 viruses treated with the inactivating agent β-propiolactone exhibit most of the spikes in the postfusion conformation, forming clusters. The density map of postfusion spikes derived from in situ subtomogram averaging of native virus particles shows a structure similar to that of the SARS-CoV S2 spikes previously determined by cryo-EM single-particle analysis of recombinant proteins. The anticipated glycosylation sites in the subtomogram average of SARS-CoV-2 postfusion spikes match well with the atomic model of the SARS-CoV S2 spike.
The use of chemically inactivated pathogens is one of the most common vaccine strategies. It has a good track record of generating long-lasting immunity for many different viral diseases, such as flu, polio, and yellow fever. Nonetheless, this strategy is not universally suited to all viruses and can even have disastrous consequences if a molecular and structural understanding of the antigen is lacking. One such unfortunate example is the formalin-inactivated respiratory syncytial virus (FI-RSV) vaccine trial of the 1960s, which led to enhancement of disease symptoms in vaccinated children after natural exposure to RSV, with two fatal cases. The molecular mechanism of this negative effect of the RSV vaccine was not fully understood until the structures of formalin-inactivated RSV were determined (
). The RSV spike is formed by the F protein, a class I fusion protein similar to coronaviruses' S protein, and can adopt a prefusion or a postfusion state. Structural studies revealed that one contributing factor to the vaccine failure was that the prefusion state of the RSV spike was absent and the postfusion state was primarily represented in the FI-RSV vaccine formula (
). This highlights the need to understand the differential roles of neutralizing and non-neutralizing antibodies elicited by vaccines and the challenge of avoiding antibody-dependent enhancement.
β-propiolactone is the chemical inactivating agent successfully used in rabies and other vaccines. Here we have shown that β-propiolactone-treated SARS-CoV-2 viruses exhibit most of their spikes in the postfusion conformation. It is possible that β-propiolactone could induce this conformation change, yet we cannot rule out the effect of purification and concentration procedures. Most COVID-19 vaccine candidates rely on the S protein as its antigen, since this is the primary exposed protein on the surface of the SARS-CoV-2 viral particle. As of July 7, 2020, 21 COVID-19 vaccine candidates were in clinical evaluation, 4 of them utilizing inactivated viruses (
). β-propiolactone was used in three of four candidates as the inactivation reagent (
) (Table S3). One of the vaccine candidates, PiCoVacc, was purified and inactivated the same way as in this study, and not surprisingly also showed substantial postfusion spikes, even though a prefusion state was incorrectly claimed (
). Therefore, structural studies, particularly on the conformational state of the viral spike in situ in intact virus particles, are paramount for these vaccine candidates, especially when the antigenicity of a vaccinal antigen is not predictive of the protective immunity elicited by it, as the FI-RSV trial shows.