key: cord-0970099-4r11i0r4
authors: Olia, Adam S.; Tsybovsky, Yaroslav; Chen, Steven J.; Liu, Cuiping; Nazzari, Alexandra F.; Ou, Li; Wang, Lingshu; Kong, Wing-Pui; Leung, Kwan; Liu, Tracy; Stephens, Tyler; Teng, I-Ting; Wang, Shuishu; Yang, Eun Sung; Zhang, Baoshan; Zhang, Yi; Zhou, Tongqing; Mascola, John R.; Kwong, Peter D.
title: SARS-CoV-2 S2P spike ages through distinct states with altered immunogenicity
date: 2021-08-27
journal: J Biol Chem
DOI: 10.1016/j.jbc.2021.101127
sha: 7832f34273ffe86f2e45aa1157ff14b9b64fa1e6
doc_id: 970099
cord_uid: 4r11i0r4

The SARS-CoV-2 spike is the primary target of virus-neutralizing antibodies and critical to the development of effective vaccines against COVID-19. Here, we demonstrate that the prefusion-stabilized two-proline “S2P” spike –widely employed for laboratory work and clinical studies– unfolds when stored at 4 °C, physiological pH, as observed by electron microscopy (EM) and differential scanning calorimetry, but that its trimeric, native-like conformation can be reacquired by low pH treatment. When stored for approximately one week, this unfolding does not significantly alter antigenic characteristics; however, longer storage diminishes antibody binding, and month-old spike elicits virtually no neutralization in mice despite inducing high ELISA-binding titers. Cryo-EM structures reveal the folded fraction of spike to decrease with aging, though its structure remains largely similar, although with varying mobility of the receptor-binding domain. Thus, the SARS-CoV-2 spike is susceptible to unfolding which affects immunogenicity, highlighting the need to monitor its integrity.

SARS-CoV-2, the etiological cause of the COVID-19 pandemic, has been the focus of a worldwide effort both to more fully understand its biology and to develop clinical products to treat or prevent infection (1) (2) (3) (4) . A major focus of SARS-CoV-2 research involves the trimeric spike glycoprotein, a type-1 fusion machine responsible for receptor engagement and viral entry into target cells (5) (6) (7) (8) (9) . Antibodies targeting the receptor-binding domain (RBD) as well as the Nterminal domain (NTD) have been identified from convalescent donors that potently neutralize SARS-CoV-2 (10) (11) (12) (13) (14) (15) (16) (17) (18) and several of these have been shown clinically to ameliorate infection (19, 20) .

As is the general case for type-1 fusion proteins, the mature spike is embedded in the viral membrane in a metastable prefusion conformation, which undergoes a large structural rearrangement as it transitions to its postfusion form, a transition which facilitates merging of viral and target cell membranes and virus entry into the cell. Two proline mutations, originally developed for the MERS coronavirus spike (21) , have been shown to stabilize the soluble trimeric ectodomain in its prefusion conformation. When combined with the T4-bacteriophage foldon trimerization domain and a mutated furin site to prevent cleavage (9) , the "S2P" construct has become the de facto protein format for research into the biological function of the SARS-CoV-2 spike. Related di-proline constructs have been utilized for approved Moderna and Pfizer mRNA vaccines (22, 23) as well as for the proteinaceous Sanofi and Novavax vaccines, and the spike component of the Johnson & Johnson adenovirus-based vaccine (24) (25) (26) . While further stabilized constructs have been developed, including Hexapro with six proline mutations (27) , the soluble, trimeric S2P remains widely studied and utilized.

The effect of refrigerated storage on S2P has been documented as resulting in global unfolding, which can be counteracted by a heat-shock procedure and potentially mitigated by low-pH storage (28) . However, details of unfolding and its impact on S2P immunogenicityespecially the elicitation of neutralizing responseshave remained unclear. Here, we investigate the cold storage-induced unfolding of S2P by measuring both its antigenicity and its immunogenicity and by comparing negative-stain electron microscopy (EM) images and cryo-EM structures of fresh and aged S2P. Notably, upon cold-storage for 30 days, S2P still elicits substantial ELISA responses in mice, but no longer elicits neutralizing responses. Further, we demonstrate that a simple low-pH treatment procedure reverses the storage-induced unfolding.

The S2P spike glycoprotein (Fig. 1a) was expressed in mammalian cells and purified using metal affinity and size exclusion chromatography, following the general protocol established previously (9) , maintaining the protein at neutral pH and room temperature.

Following purification, the protein was stored at 4°C, to mimic typical storage conditions. We then investigated this S2P construct at several time points to determine the stability of the protein as a function of storage time. Notably, storage for more than 30 days at 4 °C in phosphate buffered saline (PBS) did not cause cleavage or other major peptide modifications to the protein, as indicated by SDS-PAGE (Fig. 1b) . However, differential scanning calorimetry (DSC) showed that storage at 4 °C in PBS for 8 days (8d) resulted in a drop of melting temperature from 65.1 °C for freshly purified protein to 47.9 °C, although the overall energy of unfolding, as calculated from the peak area, remained similar at ~250 kcal/mol (Fig. 1c) . This 8d-aged S2P sample also J o u r n a l P r e -p r o o f exhibited changes in the conformational state as detected by negative-stain electron microscopy (NS-EM) (Fig. 1d) . While molecules of freshly purified S2P appeared primarily as compact, trimeric particles, unfolded species prevailed in the 8d-aged S2P, with the estimated fraction of compact trimeric particles reduced from 89% to 47%. It is likely that the NS-EM captured the states of the protein in equilibrium between the folded and unfolded conformations, with individual domains still intact in the unfolded spike.

After approximately one month of storage in PBS at 4 °C, the aged S2P (~30d) appeared fully unfolded. This was indicated by the lack of a DSC transition (Fig. 1c) and near complete disappearance of compact trimeric particles in the NS-EM images (estimated fraction of folded molecules: 5%) (Fig. 1d ). In addition to the unfolding of S2P spike trimers, the appearance of smaller particles was also noticed, suggesting complete dissociation of the trimer. Overall, the results indicate that the SARS-CoV-2 S2P trimer is unstable when stored in PBS at 4 °C.

One potential mechanism for the observed loss of the ordered trimeric S2P structure was that the trimeric core of the protein slowly dissociated during storage, with the individual domains retaining their structural state and the protomers remaining associated via the foldon trimerization domain (Fig. 1e) . To test this possibility, we subjected the 8d-aged S2P to lower pH conditions, as this was previously observed to stabilize the all RBD-down conformation of the spike (29) . We found that transferring 8d-aged S2P protein into a buffer of pH 5.5 for 10 minutes restored its DSC-melting profile to a shape nearly identical to that of the fresh protein (Fig. 2a) and restored a fully folded conformation of compact trimeric particles as revealed by NS-EM ( Fig. 2b) . Similar results could be achieved by exposing the S2P spike to acidic pH in the 4.0 -J o u r n a l P r e -p r o o f 5.5 range, whereas treatment at pH 6.0 failed to fully restore the trimer (Fig. S1 ). Of note, we did not observe unfolding of the S2P trimer by NS-EM when it was stored at 4°C in pH 4.0 for 23 days (Fig. S2a ).

To determine if the impact of low-pH-induced refolding could be maintained upon returning the protein to neutral pH, the 8d-aged protein that was subjected to the 10-minute low-pH treatment was transferred back into PBS using buffer exchange and then reassessed by DSC and NS-EM. Such pH-treated protein exhibited essentially identical melting temperature to that of the freshly purified protein (Fig. 2c ) and retained its folded, trimeric conformation (Fig. 2d ).

This effect was also observed by NS-EM when the protein was refolded at pH 4.0-5.0 (Fig. S1 ).

Furthermore, subsequent storage at 4 °C in PBS resulted in gradual unfolding similar to that observed in protein samples not subjected to low-pH treatment, indicating that transient exposure to low pH "resets" the S2P spike to its original state (Fig. S2b) . We attempted the same pH treatments on the ~30d-aged S2P protein, but were unable to restore the trimeric particle conformation (Fig. S2c) , indicating that the unfolding of S2P becomes irreversible after longer storage in PBS at 4 °C. Consistent with the above observations, while we confirmed that the 8daged protein refolded successfully in response to raising the temperature to 37 °C, as reported by Edwards et al. (28) , such temperature treatment failed to rescue the ~30d-aged S2P spike (data not shown). Overall, the S2P protein appeared to be more stable at low pH, with unfolding induced by short-term storage in PBS at 4 °C reversible by low-pH treatment.

To address the immunological implications of the aging of the spike S2P protein, we immunized BALB/cJ mice at weeks 0 and 3 with 1 g of fresh S2P (0d) or S2P after extended J o u r n a l P r e -p r o o f storage in PBS (pH 7.4) at 4°C for ~30 days (~30d-aged) (Fig. 3a) . No significant difference in serum ELISA titers against S2P was observed between the two groups of mice, and sera from two weeks after the last immunization from both groups showed >10 5 endpoint ELISA titer (Fig.   3b ). However, serum neutralizing titers against SARS-CoV-2 pseudovirus from these two groups were dramatically different (Fig. 3c ). Mice immunized with fresh S2P showed SARS-CoV-2 neutralizing titers of ~1000 mean ID50, whereas almost no neutralization was observed with sera from ~30d-aged S2P-immunized mice.

To gain insight into the lack of correlation between the ELISA titers, which measured the amount of serum antibodies capable of binding S2P, and neutralization titers, which measured the ability of the immunized sera to neutralize SARS-CoV-2, we assessed the mouse sera for reactivity to S2P as well as to isolated RBD and NTD domains ( Fig. 3d-f, Fig. S3 , Table S1 ) by Bio-Layer Interferometry (BLI). The sera from the fresh S2P-immunized group bound well to immobilized S2P, RBD and NTD, with an average response between 0.25 and 1.0 nm. However, the sera from ~30d-aged S2P-immunized animals showed reactivity only to S2P, but no response against the isolated RDB or NTD domains. As most known SARS-CoV-2 neutralizing antibodies target the RBD or NTD regions of the spike, the BLI results indicate the high ELISA titers, in combination with lack of neutralization from sera of ~30d-aged S2P immunized mice, could be attributed to elicited antibody responses being directed to non-neutralizing epitopes. Partially corroborating for this attribution is the fact that the single outlier in the ~30-day aged neutralization, which showed an ID50 of ~900 (Fig. 3c) , was the same animal that had a small BLI-measured binding response to RBD (Fig. 3e ), indicating this animal was able to generate antibodies against the RBD, allowing for the observed neutralization.

To better understand the impact of aging on the epitope presentation of the S2P protein, we performed BLI binding studies using a panel of antibodies and ACE2 to probe recognition of S2P protein of different ages (Fig. 4) . The fresh, 8d-aged, and 8d-aged low-pH treated S2P bound well to the immobilized ACE2 receptor, whereas the ~30d-aged S2P did not bind (Fig.   4a ). Similarly, in nearly every case the BLI signals were virtually identical between fresh (0d), 8d-aged, and 8d-aged low-pH treated S2P for binding to RBD-, NTD-, and S2-targeting antibodies ( Fig. 4a-c) . One of the RBD-targeting antibodies, 2-43, had ~50% decrease in BLI signal for binding to 8d-aged spike relative to the fresh and the low-pH restored samples ( Fig.   4a ), consistent with the fact that the 2-43 epitope is quaternary, and therefore its binding is likely disrupted by trimer unfolding even if the individual domains remain uncompromised (10) . As expected, the ~30d-aged S2P had greatly reduced binding to all the antibodies tested.

These results supported the idea that the individual domains of the spike were largely intact in the 8d-aged sample, and only the quaternary organization of the trimer was affected. By contrast, the structure of the individual domains was likely disrupted in the ~30d-aged S2P spike, which was completely disordered in solution.

To determine if prolonged storage and transient exposure to mild acidic conditions are associated with structural changes in the folded spike, we used cryo-electron microscopy (cryo-EM) and single particle analysis to solve structures of the spike immediately after purification (0d), following storage at neutral pH and 4 °C for about 30 days (~30d), and after a brief J o u r n a l P r e -p r o o f incubation of 12-days aged spike at pH 5.5 followed by returning it to neutral pH (12d pH 5.5 to pH 7.4). The three cryo-EM specimens were prepared at identical protein concentrations and vitrification conditions, and the same protocol was followed during the computational analysis.

We noticed that the number of intact spike molecules per micrograph was substantially reduced for the aged protein versus the fresh protein (average particles picked per micrograph using crYOLO software (30): 63 vs. 243) (Fig. 5) , and refolding via a transient exposure to pH 5.5

increased the average number of particles per micrograph to 174. While variations in ice thickness and other factors could contribute to the observed differences, these results were consistent with the negative-stain EM data, indicating that the number of compact, fully assembled trimers dropped during storage and could be restored by low-pH treatment. The final cryo-EM datasets for fresh, aged, and refolded spike contained 403,167, 95,641, and 252,067 particles, respectively, and the corresponding maps were refined to resolutions of 2.6 Å, 3.5 Å, and 3.2 Å (Table S2) . While there are ten unique combinations of RBD positions for each experimental state (e.g., all down, two down and one undefined, one down and two undefined, etc.), we have only deposited a single structure to represent experimental state, as there were an insufficient number of particles to determine the full ensemble of RBD structures for the aged state and this variation in RBD conformation is well defined in the published literature (29) .

Overall, the structures of the three spike proteins were very similar (r.m.s.d. < 1 Å).

However, examination of the maps revealed weaker cryo-EM density for all the RBDs in the "aged" protein in the absence of noticeable differences in other regions (Fig. 5) . We quantified the occupancies of the RBDs in the up and down position using 3D classification within a mask encompassing both conformations (Fig. 5d, Fig. S4 ). The position of the RBD was categorized as undefined if the corresponding 3D class did not contain cryo-EM density for this RBD at the set J o u r n a l P r e -p r o o f map level (Fig. S4) . Such "empty" classes could represent intact but highly mobile RBDs or, alternatively, RBDs that underwent unfolding. We found that, for each RBD, the fractions of the up and down conformations decreased substantially in the aged protein, while the fraction of the RBD without a defined position invariably increased (Fig. 5d) . Importantly, pH-induced refolding restored the fractions of the up and down conformations to near-original values. Based on these findings we hypothesize that unfolding of the spike could begin with increased mobility of the RBDs, which could destabilize the S1 subunit of the spike trimer.

While it is expected that long storage of most proteins will result in unfolding and degradation, understanding the rates and pathways of unfolding of the SARS-CoV-2 S2P spike protein and determining the optimal storage conditions can benefit both laboratory investigations and development of clinical products. It was previously reported (28) that cold storage of the S2P spike resulted in unfolding of the protein, with less than 10% of native-like molecules remaining after 7 days at 4 °C, according to NS-EM. Although in our experiments the S2P spike demonstrated slower unfolding (~50% intact molecules after 8 days), the overall trend was similar. We found that changes in biophysical and structural characteristics of the S2P spike are detectable after just one week of storage at 4 °C in neutral pH, the conditions that are often used by default in laboratory research. Moreover, the antigenicity of the S2P protein becomes affected by one month of storage at these conditions, which can have direct implications not only in research but also for development and deployment of protein-based vaccines for COVID-19.

Given that after a week of cold storage the antigenicity of the S2P spike is largely unchanged, yet the thermal stability and assembly as assessed by negative-stain EM are J o u r n a l P r e -p r o o f significantly decreased, it is likely that this state of the protein is characterized by weakening of the inter-protomer contacts followed by protomer separation, with the trimer being held together by the foldon domain at the C-terminus of the construct, and the RBDs and NTDs remaining intact. Based on the drop in the occupancy of the "up" and "down" conformations of the RBDs detected by cryo-EM in the fraction of aged protein molecules that remained folded, the disassembly process likely begins with increased mobility of the RBDs. This growing mobility weakens the inter-protomer interactions, increasing the likelihood of protomer separation. In this regard, we observed earlier that reducing the pH increases the fraction of the RBDs in the down position, with all three RBDs down at pH 4.0-4.5 (29) . Here we found that prolonged storage of the S2P spike at pH 4.0 prevents unfolding, supporting a link between RBD mobility and trimer stability.

It has been proposed that an ELISA readout could be a useful correlate of vaccine protection (31) . In this regard, if only serum ELISA were performed in an immunization experiment with one-month old S2P spike, the results could be misleading. While the storage time of one month is beyond what most would consider tolerable, it is possible that the loss of the expected correlation between ELISA readout and neutralization potency occurs gradually beyond the eight-day time point as the storage-induced unfolding progresses and becomes irreversible. The aged S2P protein thus provides an example for which ELISA did not predict neutralization, indicating that, while these values are often correlated, it is clearly not always the case.

The S2P construct of the SARS-CoV-2 spike protein is being employed as a clinical product in the Sanofi vaccine candidate (24) (25) (26) as well as a research tool in numerous labs around the world. It is therefore imperative that the quality and state of the protein are well J o u r n a l P r e -p r o o f understood during its use. While cold storage-induced denaturation of the S2P spike was reported earlier (28), we show that the unfolding of the S2P protein involves at least two states with distinct biophysical and structural properties and capacities to elicit neutralizing antibody.

Furthermore, lowering the pH ameliorates and prevents the degradation of the folded spike trimer, potentially providing a convenient way of long-term storage at 4 °C. As such, if flash freezing and storage at -80°C is not compatible with study designs or real-world usage, a protein formulation near pH 5.0 allows for refrigerator-temperature storage. Finally, this study highlights the need for thorough biochemical characterization of protein-based immunogens including detailed assessment of stability and structure.

Expression and preparation of SARS-CoV2 S2P SARS-CoV2 S2P protein was produced as previously described (9) . Briefly, one liter of 

Differential scanning calorimetry measurements were performed on the spike samples using a VP-DSC (GE Healthcare/MicroCal). The protein was diluted to 0.25 mg/ml in PBS and measured from 20-95°C at a rate of 1° per minute. Thermal denaturation temperature (Tm) and enthalpy of unfolding were calculated from fits provided by Origin software.

The S2P spike sample was diluted to a concentration of 0.01 -0.02 mg/ml with buffer containing 10 mM HEPES, pH 7.4, 150 mM NaCl (for samples at neutral pH) or buffer containing 10 mM Na-acetate, 150 mM NaCl (for samples at acidic pH). A 4.7-µl drop of the J o u r n a l P r e -p r o o f diluted sample was applied to a freshly glow-discharged carbon-coated copper grid for 15 s.

Filter paper was used to remove the drop, and the grid was washed by applying consecutively several 4.7-µl drops of the buffer that was used for sample dilution. After the last drop was removed, three drops of 0.75% uranyl formate were applied in the same manner, and the grid was allowed to dry. Datasets were collected with SerlialEM (35) The ELISA assays were performed as described (22) . Briefly, clear flat-bottom 96-well plates (ThermoFisher, Catalogue # 442404) were coated with 100 ng of SARS-CoV-2 S-2P without any tags per well in 1× PBS, pH 7.4 at 4 °C for 16 h. After standard washes and 2 h of blocking at room temperature, plates were incubated with serial dilutions of heat-inactivated sera for 1 h at room temperature. Following three washes, the plates were incubated with anti-mouse IgGhorseradish peroxidase conjugate (Invitrogen, Catalogue # G21040) for 1 h at room temperature.

After standard washes, the plates were developed using 3,5,3′5′-tetramethylbenzidine (TMB) (KPL) as the substrate to detect antibody responses. The reaction was quenched using 1 N H2SO4. The plates were read at 450 nm on a SpectraMax Plus plate reader. End-point titers were calculated as the serum dilution that yielded an optical density exceeding 4× background (secondary antibody alone).

Pseudovirus neutralization assay SARS-CoV-2 spike pseudotyped lentiviruses that harbor a luciferase reporter gene were produced and neutralization assay was performed as described previously (22, 36) 

Antibody binding to fresh, aged and pH-shocked SARS-CoV-2 spike was assessed using a FortéBio Octet HTX instrument (FortéBio). Experiments were conducted in tilted black 384well plates (Geiger Bio-One). PBS + 1% BSA was used as buffer. Plates were agitated at a speed of 1,000 rpm and kept at 30°C for the duration of the binding assays. Antibody IgGs (B38, S309, 2-43, CR3022, P2B-2F6, 2-51, 4-18, S652-112) or Fc-tagged dimeric ACE2 (SinoBio) were loaded at 2-3ug/mL onto anti-human Fc biosensors (FortéBio) for 1 minute to achieve a capture level of 0.5-0.7nm. The load was consistent with variability between sensors for a given antibody being below 0.13nm; however, one exception was S309, which loaded ~0.2nm higher in the fresh S2P sample than the subsequent S309 trials. Immobilized IgG was dipped into each spike sample (50nM to 3.125nM). Binding was measured for 5 minutes and dissociation recorded for 30 minutes. Octet data analysis software version 12.0 (FortéBio) was used to process the data, perform reference well subtraction to correct for systematic baseline drift, and fit curves. The data was fit using a global fitting and a 1:1 binding interaction model. Plots were generated using GraphPad Prism.

Mouse sera were assessed for binding to SARS-CoV2 S2P protein, RBD and NTD domain. Biotinylated SARS-CoV2 S2P protein, RBD and NTD domain were immobilized to SA biosensors, the sensor tips were subsequently equilibrated in 1% BSA/PBS for 60 s. Sera were diluted 100-fold in PBS + 1% BSA, and binding to the immobilized proteins was assessed for J o u r n a l P r e -p r o o f 300 seconds. The naive prebleed sera response for each group of immunogens was used as a reference.

Cryo-EM specimen preparation and data collection S2P spike samples that had not been subjected to low-pH treatment were vitrified at 0.5 mg/ml in PBS. To prepare the low pH-treated specimen, Na-acetate, pH 5.5, was added to the S2P sample that had been stored for 12 days at 4°C in PBS to a final concentration of 100 mM. After a 10min incubation at 4°C, Trizma, pH 8.2, was added to a final concentration of 50 mM. This brought the pH value back to 7.4 as determined by measuring pH after mixing proportional volumes of the same buffers in a separate experiment. The final protein concentration was 0.5 mg/ml. Immediately before vitrification, Quantifoil R 2/2 gold grids were glow-discharged using a PELCO easiGlow device (air pressure: 0.39 mBar, current: 20 mA, duration: 30 s). Cryo-EM specimens were prepared using an FEI Vitrobot Mark IV plunger at 4°C and 95% humidity; the drop volume was 2.7 µl. Datasets were collected at the National CryoEM Facility (NCEF), National Cancer Institute, using two Thermo Scientific Titan Krios G3 electron microscopes equipped with K3 direct electron detectors and Gatan Quantum GIF energy filters (slit width: 20 eV) (Table S2 ). For the pH-treated S2P specimen, an additional dataset was collected with the stage tilted to 30° to improve angular coverage; this dataset was merged with the original dataset during computational processing.

Single particle analysis of cryo-EM data Single particle analysis was performed using the Frederick Research Computing Environment (FRCE) computing cluster (0d and ~30d specimens) and the Biowulf computing cluster (pH-J o u r n a l P r e -p r o o f treated specimen). MotionCorr2 was used for patch-based movie frame alignment (37) . Contrast transfer function (CTF) parameters were estimated with ctffind4 (38) (FRCE) or GCTF (39) (Biowulf). Particle picking was performed with cryOLO 1.4 (FRCE) or 1.5 (Biowulf) using a general network model (30) . The following steps were performed using Relion 3.1 (40) unless otherwise stated. Particles were extracted with 4x binning and subjected to 2D classification. For the ~30d dataset, the initial 3D reference was obtained using EMAN 2.1 (41) ; this 3D reference was used to perform 3D classification for all three datasets. Particles contributing to high-quality 3D classes were re-extracted without binning, followed by an additional round of 3D classification. For the 0d dataset, this 3D classification produced two distinct conformations, with one (93% of particles) and two (7% of particles) RBDs in the up position; the corresponding sub-datasets were then processed separately. Only the one-up conformation was resolved in the remaining two datasets. All the structures were gradually improved with rounds of 3D autorefinement, CTF refinement, and Bayesian polishing. Local resolution was determined with RBD occupancy quantification was performed using local 3D classification without particle re-alignment. A mask encompassing both the up and down conformations of the RBD was first created (Fig. S4a) . To do this, two copies of the 0d cryo-EM map were aligned in UCSF Chimera (43) such that RBD A (predominantly up position) of one copy was in register with RBD B (predominantly down position) of the other copy. The densities corresponding to the two RBDs were isolated by segmentation and combined. A soft mask was then created from this composite density. This mask was used to perform local 3D classification into 5 classes without particle alignment for each RBD of each cryo-EM map. Before evaluating RBD occupancy, J o u r n a l P r e -p r o o f equivalent map levels were first determined by aligning the three refined maps (before postprocessing) and visually comparing the density in the stable S2 region. The RBD 3D classes were then examined at these equivalent map levels (0.008 for one-up 0d map; 0.007 for two-up 0d map; 0.0068 for the ~30d map; 0.0073 for the pH-treated spike map) and categorized as belonging to the up or down conformation or lacking a defined conformation (Fig. S4b) . The fractions of the up, down, and undefined states were calculated based on the numbers of particles that contributed to each class.

The cryo-EM structure of the S2P spike at pH 5.5 (PDB ID 6xm3) was docked into the one-up 0d map using UCSF Chimera and refined by alternating rounds of model building in Coot (44) and real-space refinement in Phenix (45) . The refined structure was used as a starting model for the ~30d and pH-treated S2P spike structures, which were refined in the same way. Structure validation was performed with Molprobity (46) . Map-model correlations were evaluated with phenix.mtriage (47) . Figures were prepared with UCSF Chimera.

Cryo-EM maps of the freshly purified (0d), aged (~30d) and refolded by low-pH treatment SARS-CoV-2 S2P have been deposited to the EMDB with accession codes EMD-23982, EMD-23983 and EMD-23984, respectively, and fitted coordinates have been deposited to the PDB with accession codes 7MTC, 7MTD and 7MTE, respectively.

This article contains supporting information. 

2020) A strategic approach to COVID-19 vaccine R&D

2020) Safety, tolerability, and immunogenicity of a recombinant adenovirus type-5 vectored COVID-19 vaccine: a doseescalation, open-label, non-randomised

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

Cell entry mechanisms of SARS-CoV-2

Structural basis of receptor recognition by SARS-CoV-2

2020) Structure, Function, and Antigenicity of the SARS-CoV-2 Spike Glycoprotein

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

Potent neutralizing antibodies against multiple epitopes on SARS-CoV-2 spike

Potent Neutralizing Antibodies against SARS-CoV-2 Identified by High-Throughput Single-Cell Sequencing of Convalescent Patients' B Cells

Human monoclonal antibodies block the binding of SARS-CoV-2 spike protein to angiotensin converting enzyme 2 receptor

Human neutralizing antibodies elicited by SARS-CoV-2 infection

Isolation of potent SARS-CoV-2 neutralizing antibodies and protection from disease in a small animal model

REGN-COV2, a Neutralizing Antibody Cocktail

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

Safety and Immunogenicity of Two RNA-Based Covid-19 Vaccine Candidates

Safety and immunogenicity of SARS-CoV-2 recombinant protein vaccine formulations in healthy adults: interim results of a randomised

NVX-CoV2373 vaccine protects cynomolgus macaque upper and lower airways against SARS-CoV-2 challenge

Cold sensitivity of the SARS-CoV-2 spike ectodomain

Cryo-EM Structures of SARS-CoV-2 Spike without and with ACE2 Reveal a pH-Dependent Switch to Mediate Endosomal Positioning of Receptor-Binding Domains

A highly conserved cryptic epitope in the receptor binding domains of SARS-CoV-2 and SARS-CoV

A noncompeting pair of human neutralizing antibodies block COVID-19 virus binding to its receptor ACE2

Focused evolution of HIV-1 neutralizing antibodies revealed by structures and deep sequencing

Automated electron microscope tomography using robust prediction of specimen movements

In vivo gene delivery and stable transduction of nondividing cells by a lentiviral vector

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

CTFFIND4: Fast and accurate defocus estimation from electron micrographs

Gctf: Real-time CTF determination and correction

RELION: implementation of a Bayesian approach to cryo-EM structure determination

EMAN2: an extensible image processing suite for electron microscopy

Quantifying the local resolution of cryo-EM density maps

UCSF Chimera--a visualization system for exploratory research and analysis

Coot: model-building tools for molecular graphics

Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix

MOLPROBITY: structure validation and all-atom contact analysis for nucleic acids and their complexes

New tools for the analysis and validation of cryo-EM maps and atomic models