key: cord-0784842-b2wkm03g
authors: Yao, Hangping; Song, Yutong; Chen, Yong; Wu, Nanping; Xu, Jialu; Sun, Chujie; Zhang, Jiaxing; Weng, Tianhao; Zhang, Zheyuan; Wu, Zhigang; Cheng, Linfang; Shi, Danrong; Lu, Xiangyun; Lei, Jianlin; Crispin, Max; Shi, Yigong; Li, Lanjuan; Li, Sai
title: Molecular architecture of the SARS-CoV-2 virus
date: 2020-09-06
journal: Cell
DOI: 10.1016/j.cell.2020.09.018
sha: c7cb53fbdf637f9820f0c18417ccbb57118274e1
doc_id: 784842
cord_uid: b2wkm03g

SARS-CoV-2 is an enveloped virus responsible for the COVID-19 pandemic. Despite recent advances in the structural elucidation of SARS-CoV-2 proteins, detailed architecture of the intact virus remains to be unveiled. Here we report the molecular assembly of the authentic SARS-CoV-2 virus using cryo-electron tomography (cryo-ET) and subtomogram averaging (STA). Native structures of the S proteins in both pre- and postfusion conformations were determined to average resolutions of 8.7-11 Å. Compositions of the N-linked glycans from the native spikes were analyzed by mass-spectrometry, which revealed highly similar overall processing states of the native glycans to that of the recombinant glycoprotein glycans. The native conformation of the ribonucleoproteins (RNP) and its higher-order assemblies were revealed. Overall, these characterizations have revealed the architecture of the SARS-CoV-2 virus in exceptional detail, and shed lights on how the virus packs its ∼30 kb long single-segmented RNA in the ∼80 nm diameter lumen.

As of August 31 st , 2020, a total of over 25 million cases of COVID-19 were reported and more than 850 thousand lives were claimed globally (https://covid19.who.int). The causative pathogen, Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is a novel βcoronavirus Wu et al., 2020; Zhou et al., 2020) . SARS-CoV-2 encodes at least 29 proteins in its (+) RNA genome, four of which are structural proteins: the spike (S), membrane (M), envelope (E) and nucleocapsid (N) proteins (Kim et al., 2020) .

The ~600 kDa, trimeric S protein, one of the largest known class-I fusion proteins, is heavily glycosylated with 66 N-linked glycans Watanabe et al., 2020a; Wrapp et al., 2020) . Each S protomer comprises the S1 and S2 subunits, and a single transmembrane (TM) anchor (Wrapp et al., 2020) . The S protein binds to the cellular surface receptor angiotensin-converting enzyme-2 (ACE2) through the receptor binding domain (RBD), an essential step for membrane fusion (Hoffmann et al., 2020; Lan et al., 2020; Shang et al., 2020; Wang et al., 2020; Yan et al., 2020; Zhou et al., 2020) . The activation of S requires cleavage of S1/S2 by furin-like protease and undergoes the conformational change from prefusion to postfusion (Belouzard et al., 2009; Kirchdoerfer et al., 2018; Simmons et al., 2004; Simmons et al., 2013; Song et al., 2018) . Several prefusion conformations have been resolved for the S protein, wherein the three RBDs display distinct orientations, "up" or "down" Wrapp et al., 2020) . The receptor binding sites expose, only when the RBDs adopt an 'up' conformation. The "RBD down", "one RBD up" and "two-RBD up" conformations have been observed in recombinantly expressed S proteins of SARS-CoV-2 (Henderson et al., 2020; Walls et al., 2020; Wrapp et al., 2020) . Upon activation, S follows a classic pathway among class-I fusion proteins (Rey and Lok, 2018) : it undergoes dramatic structural rearrangements involving shedding its S1 subunit and inserting the fusion peptide (FP) into the target cell membrane (Cai et al., 2020) . Following membrane fusion, S transforms to a needle-shaped postfusion form, having three helixes entwining coaxially (Cai et al., 2020) . Despite the efforts in elucidating the SARS-CoV-2 virus host recognition and entry mechanism at near-atomic resolution using recombinant proteins, highresolution information regarding the in situ structures and landscape of the authentic virus is in-demand.

Coronavirus has the largest genome among all RNA viruses. It is enigmatic how the N protein oligomerizes, organizes, and packs the ~30 kb long single-stranded RNA in the viral J o u r n a l P r e -p r o o f lumen. Early negative-staining electron microscopy of coronaviruses showed single-strand helical RNPs with a diameter of ~15 nm (Caul and Eggleston, 1979) . Cryo-ET of SARS-CoV revealed that RNPs organized into lattices underneath the envelope at ~4-5 nm resolution (Neuman et al., 2006) . However, such ultrastructure was not observed in the Mouse hepatitis virus (MHV), the prototypic β-coronavirus (Barcena et al., 2009) . No molecular model exists so far for the coronavirus RNP, and little is known about the architecture, assembly and RNA packaging of the RNPs of other (+) RNA viruses.

To address these questions, we combined cryo-ET and STA for the imaging analysis of 2,294 intact virions propagated from an early viral strain (Yao et al., 2020) . To our knowledge, this is the largest cryo-ET data set of SARS-CoV-2 virus to date. Here we report the architecture and assembly of the authentic SARS-CoV-2 virus.

SARS-CoV-2 virions (ID: ZJU_5) were collected on January 22 nd , 2020, from a patient with severe symptoms, and were propagated in Vero cells. The patient was infected during a conference with attendees from Wuhan (Yao et al., 2020) . For cryo-EM analysis, the viral sample was fixed by paraformaldehyde, which has minor effects on protein structure at 7-20 Å resolution (Li et al., 2016; Wan et al., 2017) . Intact In total, 56,832 spikes were manually identified from the virions, approximately 97% of which are in the prefusion conformation, 3% in the postfusion conformation (Method details).

An average of 26±15 prefusion S were found randomly distributed on each virion ( Figures   1B-C) . The spike copy number per virion is comparable to HIV (Liu et al., 2008) , but ~5 times less than the Lassa virus (LASV) (Li et al., 2016) or ~10 times less than the Influenza J o u r n a l P r e -p r o o f virus (Harris et al., 2006) . 18,500 RNPs were manually identified in the viral lumen (Table   S1 ), giving an average of 26±11 RNPs per virion. However, since the viral lumen is tightly packed with RNPs and electron opaque, the actual number of RNPs per virion was estimated to be 20-30% more, i.e. 30-35 RNPs per virion. Regularly ordered RNP ultrastructures were occasionally observed ( Figure S5A ), indicating the RNPs could form local assemblies.

Two conformations of the prefusion S, namely the RBD down and one RBD up conformations from inactivated SARS-CoV-2 virions were classified and reconstructed to 8.7 Å and 10.9 Å resolution by STA, with local resolution reaching 7.8 Å ( Figures S3A-S3C ).

The heptad repeat 1 (HR1) and central helix (CH) domains of the S2 subunits represent the best resolved domains ( Figure S3D ). The proportion of RBD down conformation among all prefusion S was estimated to be 54% per virion ( Figure 1C ). The membrane proximal stalk of S represented the poorest resolved region with a local resolution of ~20 Å, showing no trace of the TM or membrane in the structure ( Figure 1D ). When the tomogram slices were scrutinized, spike populations that either stand perpendicular to or lean towards the envelope were observed, suggesting that the TM was averaged out in the map.

Refined orientations of the prefusion S showed they rotate around their stalks almost freely outside the envelope, leaning at an average angle of 40°±20° relative to the normal axis of the envelope ( Figures 1B and 1D ). The rotational freedom of spikes is allowed by its low population density, which is prominently distinct from other enveloped viruses possessing class-I fusion proteins (Harris et al., 2006; Li et al., 2016; Liu et al., 2008) . Interestingly, a minor population of Y-shaped spikes pair having two heads and one combined stem were observed ( Figures S2A and S2C ), which possibly represent spikes intertwined with their stems. These observations suggest that the SARS-CoV-2 spikes possess unusual freedom on the viral envelope. Such unique features may facilitate the virus in exploring the surrounding environment and better engaging with the cellular receptor ACE2, allowing multiple spikes to bind with one ACE2 or one spike with multiple ACE2 simultaneously. However, the sparsely packed spikes on the viral envelope are also more vulnerable to neutralizing antibodies that bind the otherwise less accessible domains (Chi et al., 2020) or glycan holes (Walls et al., 2019) . Our observations on the structures and landscape of the intact SARS-CoV-2 are consistent with two other cryo-ET studies appeared at the same period (Turoňová, 2020; Ke et al., 2020) .

J o u r n a l P r e -p r o o f

The native structures of S in the RBD down and one RBD up conformations were similar to the rigidly fitted recombinant protein structures (PDB: 6XR8 and 6VYB) (Cai et al., 2020; Walls et al., 2020) , except for the N terminus domain (NTD) . Comparison between the rigidly and flexibly fitted PDB: 6XR8 suggested that the NTD on the native spike structure shifted 9 Å (centroid distance) away from S2 ( Figure S3E ). The slight dilation and lower local resolution ( Figure S3A and S3B) of the NTD on the native spike against recombinant spike structures was also observed on the other cryo-ET structures . It is known that NTD exhibits certain mobility as a rigid body (Cai et al., 2020; Walls et al., 2020; Wrapp et al., 2020; Xiong et al., 2020) . It is possible that through large date set and classification, the near-atomic resolution cryo-EM reconstruction of the spike represents its metastable conformation, while the cryo-ET reconstructions represent an average of various dynamic states of the NTD.

Ten N-linked glycans are visible on the RBD down and seven on the one RBD up conformations, of which N61, N282, N801, N1098 and N1134 were best resolved.

Interestingly, densities for glycans N1158 and N1173/N1194 are visible on the stem of the spike ( Figures 2B and S4C ). In general, the glycan densities observed on the native spike fit well with the full-length recombinant structure (Cai et al., 2020) , however they are bulkier than observed in the TM truncated recombinant spikes Wrapp et al., 2020) . Similar observations were reported by two studies of the native spike structures appeared at the same period (Turoňová, 2020; Ke et al., 2020) .

We further determined the native glycan identity by analyzing the virus sample using massspectrometry (MS). The viral particles, with or without PNGase F digestion, were resolved on SDS-PAGE. After PNGase F treatment, the S1 and S2 subunits were reduced by ~30 kDa and ~20 kDa, respectively, in weight ( Figure 2C ). The bands corresponding to S1 and S2 before PNGase F treatment were analyzed by MS to reveal the glycan compositions at each of the 22 glycosylation sites ( Figures 2D and S4B ). The overall processing states of the native glycans are highly similar to that of the recombinant glycoprotein glycans ( Figures   S4A and S4B ) (Watanabe et al., 2020a) , a feature shared with MERS and SARS-CoV (Walls et al., 2019) . Populations of under-processed oligomannose-type glycans are found at the same sites as seen in the recombinant material, including at N234 where the glycan is suggested to have a structural role (Casalino et al., 2020) . However, as is observed at many sites on HIV (Cao et al., 2018; Struwe et al., 2018) , the virus exhibits somewhat lower levels J o u r n a l P r e -p r o o f of oligomannose-type glycosylation compared to the recombinant, soluble mimetic. Overall, the presence of substantial population of complex-type glycosylation suggests that the budding route of SARS-CoV-2 into the lumen of endoplasmic reticulum-Golgi intermediate compartments (ERGIC) is not an impediment to glycan maturation and is consistent with both analysis of SARS-CoV glycans (Ritchie et al., 2010) and the identification of neutralizing antibodies targeting the fucose at the N343 glycan on SARS-CoV-2 (Pinto et al., 2020) . Furthermore, the lower levels of oligomannose-type glycans compared to HIV and LASV are also consistent with lower glycan density (Watanabe et al., 2020b; Watanabe et al., 2018) . Comparing the structures between our native prefusion S to the recombinant ones, we conclude that 1) the N-linked glycans present on the native spike are bulkier and contain elevated levels of complex-type glycans; 2) the recapitulation of the main features of native viral glycosylation by soluble, trimeric recombinant S glycoprotein is encouraging for vaccine strategies utilizing recombinant S protein.

Apart from the triangular prefusion S, needle-like densities were occasionally observed on the viral envelope ( Figure S2B (Cai et al., 2020; Fan et al., 2020) . In comparison to the prefusion conformation, the fixation in orientation to the envelope suggests a dramatic conformational reordering of the stem region to achieve the postfusion conformation. The postfusion S were found only on a small Figure 3B ), compared to ~15 nm average distance between the nearest prefusion S.

Distinguished from the SARS-CoV virus, which was estimated to possess an average of ~50-100 spikes per virion (Neuman et al., 2011) , the SARS-CoV-2 virus possesses approximately J o u r n a l P r e -p r o o f half of the prefusion S and occasionally some postfusion S. The postfusion S observed on the SARS-CoV-2 virus may come from 1) products of occasional, spontaneous dissociation of S1 (Cai et al., 2020) , which was cleaved by host proteinases; 2) syncytium naturally formed on infected cells , when budding progeny virions carried a few residual postfusion S from the cell surface; 3) sample preparation procedure, as cryo-EM images of ßpropiolactone fixed viruses showed most spikes present on the virus are postfusion-like Gao et al., 2020) . Such instability of the prefusion S was reported on the other βcoronaviruses (Pallesen et al., 2017) . In addition, the distribution graph ( Figure 3C ) implies that the kinetically trapped prefusion S is more fragile than the postfusion S, and could even dissociate from the virus. The speculation is based on the fact that intracellular virions possess more spikes on average (Klein et al., 2020) than the extracellular virions reported by us and the others (Turoňová, 2020; Ke et al., 2020) , and the occasional observation of the spike-less "bald" virus in our data.

In summary, we speculate that the SARS-CoV-2 prefusion S are unstable, indicating that the distribution of solvent exposed epitopes on the virions is more complicated than the observations on the recombinant proteins. Our observation has implications for efficient vaccine design and neutralizing antibody development, which prefer a sufficient number of stable antigens.

It remains enigmatic how coronaviruses pack the ~30 kb RNA within the ~80 nm diameter viral lumen; if the RNPs are ordered relative to each other to avoid RNA entangling, knotting or even damaging; or if the RNPs are involved in virus assembly. When raw tomogram slices were inspected, densely packed, bucket-like densities were discernible throughout the virus lumen, some of which appeared to be locally ordered ( Figure S5A ). Combining previous cryo-ET observation on coronaviruses (Barcena et al., 2009 ) and SDS-PAGE/MS analysis ( Figure 2C ), the densities most likely belonged to the RNPs.

In total 18,500 RNPs were picked in the viral lumen, and initially aligned using a sphere as the template and a large spherical mask. A bucket-like conformation with little structural feature emerged adjacent to the density for lipid bilayer ( Figure S5D ), suggesting that a significant number of RNPs were membrane proximal. Alignment using a small spherical mask revealed a 13.1 Å resolution reverse G-shaped architecture of the RNP, measuring 15 J o u r n a l P r e -p r o o f nm in diameter and 16 nm in height ( Figure 4B ). Its shape is in good comparison to a recently reported SARS-CoV-2 RNP conformation (Klein et al., 2020) , as well as the in situ conformation of the chikungunya viral RNP, which is also positive stranded (Jin et al., 2018) , but different from the MHV RNP released using detergent (Gui et al., 2017) .

The map was segmented into five head-to-tail reverse L-shaped densities, each fitted with a pair of N proteins (N terminus domain (N_NTD): 6WKP, C terminus domain (N_CTD): 6WJI) dimerized by the N_CTD (Chen et al., 2007) (Figure S6A and S6C). We analyzed the electrostatic potential distribution on the surface of the decamer, and suggested a tentative structural model of RNA winded RNP ( Figures S6B and S6D) . Interestingly, an early observation on the MHV RNPs showed ~15 nm diameter helices with five subunits per turn (Caul and Eggleston, 1979) . Due to the limited resolution and little previous structural knowledge about the (+) RNA virus RNPs, our model shall be interpreted with caution.

Further 2D classification of the RNPs revealed three classes: 1) closely packed against the envelope, 2) hexagonally and 3) triangularly packed RNPs ( Figure 4A ). Following 3D refinement, a membrane proximal, "eggs-in-a-nest" shaped RNP assembly (referred to as the "hexon"), and a membrane-free, "pyramid" shaped RNP assembly (referred to as the "tetrahedron") emerged . Projection of the two class averages back onto their refined coordinates revealed that the majority of hexons came from spherical virions, while more tetrahedrons from ellipsoidal virions ( Figure 4D ). This was quantified by statistics: ellipsoidal virions tend to pack more RNP tetrahedrons ( Figure 4E ). Furthermore, the spacing between two neighboring RNPs (~18 nm) is the same for both the tetrahedrons and hexons, and some tetrahedrons could assemble into hexons when projected onto their in situ coordinates ( Figure S5B ), suggesting that the RNP triangle is a key and basic packing unit throughout the virus. We further propose that the RNPs are involved in coronavirus assembly and help strengthen the virus against environmental and physical challenges, as purified virions remained intact after five cycles of freeze-and-thaw treatment ( Figure S7 ).

Such involvement of RNPs in viral assembly was also reported by (Neuman et al., 2006) , who showed RNPs form a lattice underneath the envelope; as well as seen in intracellular virions (Steffen Klein et al., 2020) . However, it remains unanswered if the ultrastructures of RNPs are assembled by RNA, the transmembrane M or E proteins, the RNP itself, or multiples of the above.

Solving RNPs to subnanometer resolution was hindered by the crowding of RNPs against each other ( Figures S5E-S5G) . Furthermore, structural features of the RNPs on higher order assemblies smeared, possibly due to the symmetry mis-match between individual RNPs and the assembly (Figures 4C and 4D) . No virus with strictly ordered RNPs throughout the lumen was found by projections. We conclude that the native RNPs are highly heterogeneous,

densely packed yet locally ordered in the virus, possibly interacting with the RNA in a beadson-a-string stoichiometry. wrote the manuscript. All authors critically revised the manuscript.

The authors declare no competing interests. (2) purified ZJU_5; (3) ZJU_5 treated with PNGase F and (4) PNGase F as control. After PNGase F treatment, the molecular weight of the S1 subunit is reduced by ~30 kDa, and S2 by ~20 kDa in weight. 

illustrated by projecting the refined structures onto their coordinates and overlaying with the raw tomogram (lowpassed to 80 Å resolution). 

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Sai Li (sai@tsinghua.edu.cn).

This study did not generate new unique reagents. 

The Purification, concentration, biochemical analysis and sample preparation for electron microscopy of inactivated virions were carried out in a BSL-2 lab. For cryo-ET, the fixed virions were pelleted through 30% sucrose cushion by ultracentrifugation (Beckman, IN) at 100,000 g for 3 hours at 4 °C and resuspended in 60 µl HEPES-saline buffer containing 10 mM HEPES, pH 7.3, 150 mM NaCl at 4 °C overnight (Neuman et al., 2008) .

To deglycosylate S protein of virions, 20 µg purified virion sample was treated with 500 units 

Glycopeptide fragmentation data were extracted from the raw file using Byonic™ (Version 2.8.2). The MS data was searched using the Protein Metrics 309 N-glycan library. The search criteria were as follows: Non-specificity; carbamidomethylation (C) was set as the fixed modifications; the oxidation (M) was set as the variable modification; precursor ion mass tolerances were set at 20 ppm for all MS acquired in an orbitrap mass analyzer; and the fragment ion mass tolerances were set at 0.02 Da for all MS2 spectra acquired.

The intensities of each glycan type in identical site were combined and analyzed for proportion. Data with score less than 30 were discarded. The glycans were classfied into oligomannose, hybrid and complex type based on composition. Hybrid and complex type glycan were subdivided according to fucose component and antenna.

7 µl virus sample was applied onto a glow discharged copper grid coated with holey carbon (R 2/2; Quantifoil, Jena, Germany), and subsequently dipped onto 500 µl HEPES-saline buffer for 1 second to clear the sucrose. A drop of 3 µl gold fiducial beads (10 nm diameter;

Aurion, The Netherlands) was applied and the grid was blotted for 4.5 s, vitrified by plungefreezing into liquid ethane using a Cryo-plunger 3 (Gatan, CA). Fixed cells cultured on grids were applied with 2 µl fiducial beads (10 nm diameter; Aurion, The Netherlands) prior to single-sided plunge-frozen.

The grids were imaged on a Titan Krios microscope (Thermo Fisher Scientific, Hillsboro, OR) operated at a voltage of 300 kV equipped with an energy filter (slit width 20 eV; GIF Quantum LS, Gatan, CA) and K3 direct electron detector (Gatan, CA). Virions were recorded in super-resolution mode at a nominal magnification of 64,000×, resulting in a calibrated pixel size of 0.68 Å. 361 sets of tilt-series data were collected using the dose-symmetric scheme (Hagen et al., 2017) from -60° to 60° at 3° steps and at various defocus between -1.7

and -5 µm in SerialEM (Mastronarde, 2005) . At each tilt, a movie consisting of 8 frames was recorded with 0.0265 s/frame exposure, giving a total dose of 131.2 e -/Å 2 per tilt series.

For the freeze-and-thaw test, 1.5 µl purified virions were diluted in 6 µl HEPES-saline buffer at 4 °C, then subjected to five cycles of freezing in liquid nitrogen and thawing in water bath at 37 °C. For the negative staining microscopy, 4 µl freeze-and-thaw sample was applied on copper grids (Zhongjingkeyi Technology, Beijing, China), stained using 2% Uranium acetate and imaged using a Tecnai Spirit TEM (Thermo Fisher Scientific, Hillsboro, OR).

Tilt series data was analysed in a high-throughput pre-processing suite developed in our lab.

The electron beam induced motion was corrected using a combination of MotionCor (Li et al., 2013) and MotionCor2 (Zheng et al., 2017) by averaging eight frames for each tilt. Defocuses of the tilt series were measured using Gctf (Zhang, 2016) . The tilt series were contrast transfer function corrected using Novactf (Turonova et al., 2017) , 319 tilt-series with good fiducial alignment and relative thin ice thickness were reconstructed to tomograms by weighted back projection in IMOD (Kremer et al., 1996) , resulting in a final pixel size of Subtomogram averaging was done using Dynamo (Castano-Diez et al., 2012) . For the prefusion S reconstruction, 54,878 subtomograms were extracted from 4 × binned tomograms into boxes of 96×96×96 voxels and EMD-21452 was used as the template for their alignment. The resolution was restricted to 40 Å and C3 symmetry was applied at this stage. 8,562 spikes present at the edges of the tomograms were removed to minimize the impact of air-water interface effect and incomplete signal on the structure. The remaining particles were subjected to multi-reference alignment imposing C1 symmetry using EMD-21452 and EMD-21457 lowpassed to 30 Å resolution as the templates, resulting in 25,236 spikes (54.5%) classified into RBD down conformation and 21,080 spikes (45.5%) into one RBD up conformation. Coordinates of the two spike conformations were used to extract boxes of 160×160×160 voxels from the 2 × binned tomograms for further alignment. To prevent overfitting, a customized 'gold-standard adaptive bandpass filter' method was used for the alignment at this stage, and a criterion of 0.143 for the Fourier shell correlation were used to estimate the resolution. The 2 × binned spikes in the RBD down and one RBD up conformations were independently further aligned imposing C3 or C1 symmetry respectively, to 9.5 and 10.9 Å resolution. Finally, the RBD down spike subtomograms were extracted from unbined tomograms into boxes of 256×256×256 voxels and aligned to 8.7 Å resolution.

The prefusion S maps were lowpassed according to the estimated local resolutions of the reconstructed subunits. Universal empirical B-factors of -1200 and -2000 were applied to sharpen the RBD down and one RBD up spikes, respectively .

For the postfusion S reconstruction, 2,010 subtomograms were extracted from 4 × binned tomograms into boxes of 96×96×96 voxels, which were averaged to give an initial template for their alignment. The resolution was restricted to 30 Å and C3 symmetry was applied at this stage. Next, the refined coordinates were used to extract 1,954 postfusion S from the 2 × binned tomograms into boxes of 160×160×160 voxels for gold-standard alignment.

Subsequent alignment achieved 15.3 Å resolution.

For the RNP reconstruction, 18,500 manually picked RNPs were extracted into subtomograms of 80×80×80 voxels from 4 × binned tomograms and globally aligned using a large sphere (radius 36 pixels) as the template. The resolution was restricted to 40 Å and no symmetry was applied at this stage. Lipid bilayers were visible in the aligned maps, suggesting part of the RNPs are relatively packed with the membrane. The alignment was repeated using a small spherical mask (radius 18 pixels). A reverse "G"-shaped structure appeared after this stage and the refined coordinates were used to extract particles from the 2 × binned tomograms into boxes of 128×128×128 voxels. Gold-standard was applied to align the RNP to a final resolution at 13.1 Å.

To analyze the local pattern of the RNP assembly, the picked RNPs' coordinates were imported into the Relion subtomogram averaging pipeline (Bharat and Scheres, 2016) . The RNP particles were extracted and projected into 2D images. Three characteristic patterns of the 2D classification are selected and subjected to 3D initial model generation and 3D classification: 1) closely packed towards the envelope, 2) hexagonally packed and 3) triangularly packed RNPs. Following 3D refinement, the first class converged only on the membrane. The second class aligned into a hexagonally packed, membrane proximal RNP assembly, and the third class aligned into a tetrahedrally packed, membrane-free assembly.

Refined coordinates and orientations of the hexagonal particles ( 

Three representative SARS-CoV-2 virus (Figures 1B and 4D ) and a bundle of postfusion S ( Figure 3B ) were reconstructed by projecting all spikes and RNPs onto their refined coordinates and merging the structures using Jsubtomo (Huiskonen et al., 2014) . For other map-projection to coordinates ( Figure 1D , S2C and S5B), the 'dtplot' function in Dynamo was used. UCSF Chimera (Pettersen et al., 2004) and ChimeraX (Goddard et al., 2018) were used for rendering the graphics.

Atomic models (PDB accession code 6XR8, 6VYB, 6XRA) of the pre-and postfusion S were rigidly fitted to the corresponding densities using the Fit in Map tool (Pettersen et al., 2004) .

The RNP map was segmented into five reverse L-shaped units, which can be further ungrouped into 7 segments above and 10 segments on the base. According to the previous Small-angle X-ray scattering (SAXS) (Chang et al., 2009 ) and cryo-EM (Gui et al., 2017) reports, the N_NTD and N_CTD possibly form a reverse L-shaped unit; the N_CTD dimer was suggested to be an assembly unit of the RNP (Chen et al., 2007) . One segment above and two on the base forming a reverse "L" from the best solved region were selected, and were fitted with a N_CTD dimer (6WJI) using the 'fit to segments' tool (Pintilie et al., 2010) in UCSF Chimera. The segment with the best fitting score (0.78 against 0.74 and 0.72) were adopted as the N_CTD. The other two segments were fitted with the N_NTD monomer (6WKP, score 0.92 and 0.90). With the reverse "L"-shaped N_NTD-CTD pair formed, we fitted the rest of RNP with four such units, leaving two upper segments unoccupied. Together, the map was interpreted as a decamer of N.

Molecular dynamic flexible fitting (MDFF) (Trabuco et al., 2009 ) was applied to improve the fitting of the atomic model to the S in RBD down conformation. PDB: 6XR8 was prepared in VMD (Humphrey et al., 1996) for the MDFF, which was performed in vacuum until convergence using NAMD 2.12 (Phillips et al., 2005) Statistics was performed using Python package Scipy.

The aligned location and orientation were used for the statistics of the spike tilt angles. For a given spike on a given virus, the nearest envelope mesh to the spike stem end was found. The angle between vector A (the spike's Z axis) and vector B (normal to the envelope mesh) was calculated. For the statistics in Fig S1C, envelopes of 113 virions before ultracentrifugation and 157 after ultracentrifugation were fitted by ellipse to estimate their diameters.

382 virions with more than 5 tetrahedron/hexon RNP assemblies were included for the statistics shown in Figure 4E . They were sorted into three classes according to the ratio of the longest to the shortest axis: spherical (long/short axis: J o u r n a l P r e -p r o o f

Cryo-electron tomography of mouse hepatitis virus: Insights into the structure of the coronavirion

In situ structural analysis of SARS-CoV-2 spike reveals flexibility mediated by three hinges

Activation of the SARS coronavirus spike protein via sequential proteolytic cleavage at two distinct sites

Resolving macromolecular structures from electron cryotomography data using subtomogram averaging in RELION

Distinct conformational states of SARS-CoV-2 spike protein. Science

Differential processing of HIV envelope glycans on the virus and soluble recombinant trimer

Dynamo: a flexible, user-friendly development tool for subtomogram averaging of cryo-EM data in high-performance computing environments

Dynamo Catalogue: Geometrical tools and data management for particle picking in subtomogram averaging of cryo-electron tomograms

Coronavirus-like particles present in simian faeces

Multiple nucleic acid binding sites and intrinsic disorder of severe acute respiratory syndrome coronavirus nucleocapsid protein: implications for ribonucleocapsid protein packaging

Structure of the SARS coronavirus nucleocapsid protein RNA-binding dimerization domain suggests a mechanism for helical packaging of viral RNA

A neutralizing human antibody binds to the N-terminal domain of the Spike protein of SARS-CoV-2

Viral Architecture of SARS-CoV-2 with Post-Fusion Spike Revealed by

Cryo-EM analysis of the post-fusion structure of the SARS-CoV spike glycoprotein

Development of an inactivated vaccine candidate for SARS-CoV-2

Meeting modern challenges in visualization and analysis

Electron microscopy studies of the coronavirus ribonucleoprotein complex

Implementation of a cryo-electron tomography tilt-scheme optimized for high resolution subtomogram averaging

Patientderived mutations impact pathogenicity of SARS-CoV-2. medRxiv

Influenza virus pleiomorphy characterized by cryoelectron tomography

Controlling the SARS-CoV-2 spike glycoprotein conformation

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Averaging of Viral Envelope Glycoprotein Spikes from Electron Cryotomography Reconstructions using Jsubtomo

VMD: visual molecular dynamics

Neutralizing Antibodies Inhibit Chikungunya Virus Budding at the Plasma Membrane

The Architecture of SARS-CoV-2 Transcriptome

Stabilized coronavirus spikes are resistant to conformational changes induced by receptor recognition or proteolysis

Computer visualization of three-dimensional image data using IMOD

Structure of the SARS-CoV-2 spike receptor-binding domain bound to the ACE2 receptor

Acidic pH-Induced Conformations and LAMP1 Binding of the Lassa Virus Glycoprotein Spike

Electron counting and beam-induced motion correction enable near-atomic-resolution single-particle cryo-EM

Molecular architecture of native HIV-1 gp120 trimers

Shielding and Beyond: The Roles of Glycans in SARS-CoV-2 Spike Protein

Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding

All-atom empirical potential for molecular modeling and dynamics studies of proteins

Automated electron microscope tomography using robust prediction of specimen movements

Purification and electron cryomicroscopy of coronavirus particles

Supramolecular architecture of severe acute respiratory syndrome coronavirus revealed by electron cryomicroscopy

A structural analysis of M protein in coronavirus assembly and morphology

Immunogenicity and structures of a rationally designed prefusion MERS-CoV spike antigen

UCSF chimera -A visualization system for exploratory research and analysis

Scalable molecular dynamics with NAMD

Quantitative analysis of cryo-EM density map segmentation by watershed and scale-space filtering, and fitting of structures by alignment to regions

Structural and functional analysis of a potent sarbecovirus neutralizing antibody

Common Features of Enveloped Viruses and Implications for Immunogen Design for Next-Generation Vaccines

Identification of N-linked carbohydrates from severe acute respiratory syndrome (SARS) spike glycoprotein

Structural basis of receptor recognition by SARS-CoV-2

Characterization of severe acute respiratory syndrome-associated coronavirus (SARS-CoV) spike glycoproteinmediated viral entry

Proteolytic activation of the SARScoronavirus spike protein: cutting enzymes at the cutting edge of antiviral research

Cryo-EM structure of the SARS coronavirus spike glycoprotein in complex with its host cell receptor ACE2

SARS-CoV-2 structure and replication characterized by in situ cryo-electron tomography

Site-Specific Glycosylation of Virion-Derived HIV-1 Env Is Mimicked by a Soluble Trimeric Immunogen

Molecular dynamics flexible fitting: a practical guide to combine cryo-electron microscopy and X-ray crystallography

Efficient 3D-CTF correction for cryoelectron tomography using NovaCTF improves subtomogram averaging resolution to 3.4A

Structure, Function, and Antigenicity of the SARS-CoV-2 Spike Glycoprotein

Unexpected Receptor Functional Mimicry Elucidates Activation of Coronavirus Fusion

Structure and assembly of the Ebola virus nucleocapsid

Structural and Functional Basis of SARS-CoV-2 Entry by Using Human ACE2

Site-specific glycan analysis of the SARS-CoV-2 spike

Vulnerabilities in coronavirus glycan shields despite extensive glycosylation

Structure of the Lassa virus glycan shield provides a model for immunological resistance

Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation

A new coronavirus associated with human respiratory disease in China

Inhibition of SARS-CoV-2 (previously 2019-nCoV) infection by a highly potent pan-coronavirus fusion inhibitor targeting its spike protein that harbors a high capacity to mediate membrane fusion

A thermostable, closed SARS-CoV-2 spike protein trimer

Structural basis for the recognition of SARS-CoV-2 by full-length human ACE2

Gctf: Real-time CTF determination and correction

MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy

A pneumonia outbreak associated with a new coronavirus of probable bat origin

Structures and distributions of SARS-CoV-2 spike proteins on intact virions