key: cord-1026149-ngg4wsa2
authors: Huo, Jiandong; Zhao, Yuguang; Ren, Jingshan; Zhou, Daming; Duyvesteyn, Helen ME.; Ginn, Helen M.; Carrique, Loic; Malinauskas, Tomas; Ruza, Reinis R.; Shah, Pranav NM.; Tan, Tiong Kit; Rijal, Pramila; Coombes, Naomi; Bewley, Kevin R.; Tree, Julia A.; Radecke, Julika; Paterson, Neil G.; Supasa, Piyasa; Mongkolsapaya, Juthathip; Screaton, Gavin R.; Carroll, Miles; Townsend, Alain; Fry, Elizabeth E.; Owens, Raymond J.; Stuart, David I.
title: Neutralisation of SARS-CoV-2 by destruction of the prefusion Spike
date: 2020-06-19
journal: Cell Host Microbe
DOI: 10.1016/j.chom.2020.06.010
sha: 228b195466781d3a04673bd5b5c3528b392e5507
doc_id: 1026149
cord_uid: ngg4wsa2

Summary There are as yet no licenced therapeutics for the COVID-19 pandemic. The causal coronavirus (SARS-CoV-2) binds host cells via a trimeric Spike whose receptor binding domain (RBD) recognises angiotensin-converting enzyme 2 (ACE2), initiating conformational changes that drive membrane fusion. We find that the monoclonal antibody CR3022 binds the RBD tightly, neutralising SARS-CoV-2 and report the crystal structure at 2.4 Å of the Fab/RBD complex. Some crystals are suitable for screening for entry-blocking inhibitors. The highly conserved, structure-stabilising, CR3022 epitope is inaccessible in the prefusion Spike, suggesting that CR3022 binding facilitates conversion to the fusion-incompetent post-fusion state. Cryo-EM analysis confirms that incubation of Spike with CR3022 Fab leads to destruction of the prefusion trimer. Presentation of this cryptic epitope in an RBD-based vaccine might advantageously focus immune responses. Binders at this epitope may be useful therapeutically, possibly in synergy with an antibody blocking receptor attachment.

Incursion of animal (usually bat)-derived coronaviruses into the human population has caused several outbreaks of severe disease, starting with severe acute respiratory syndrome (SARS) in 2002 (Menachery et al., 2015) . In late 2019 a highly infectious illness, with cold-like symptoms progressing to pneumonia and acute respiratory failure, resulting in an estimated 6% overall death rate (Baud et al., 2020) , with higher mortality among the elderly and immunocompromised populations, was identified and confirmed as a pandemic by the WHO on 11 th March 2020. The etiological agent is a novel coronavirus (SARS-CoV-2) belonging to lineage B betacoronavirus and sharing 88% sequence identity with bat coronaviruses (Lu et al., 2020a) . The heavily glycosylated trimeric surface Spike protein mediates viral entry into the host cell. It is a large type I transmembrane glycoprotein (the ectodomain alone comprises over 1200 residues) (Wrapp et al., 2020) . It is made as a single polypeptide and then cleaved by host proteases to yield an N-terminal S1 region and the C-terminal S2 region.

Spike exists initially in a pre-fusion state where the domains of S1 cloak the upper portion of the Spike with the relatively small (~22 kDa) S1 RBD nestled at the tip. The RBD is predominantly in a 'down' state where the receptor binding site is inaccessible, however it appears that it stochastically flips up with a hinge-like motion transiently presenting the ACE2 receptor binding site (Roy, 2020; Song et al., 2018; Walls et al., 2020; Wrapp et al., 2020) . ACE2 acts as a functional receptor for both SARS-CoV-1 and SARS-CoV-2, binding to the latter with a 10 to 20-fold higher affinity (K D of ~15 nM), possibly contributing to its ease of transmission (Song et al., 2018; Wrapp et al., 2020) . There is 73% sequence identity between the RBDs of SARS-CoV-1 and SARS-CoV-2 ( Figure 1 ). When ACE2 locks on it holds the RBD 'up', destabilising the S1 cloak and possibly favouring conversion to a postfusion form where the S2 subunit, through massive conformational changes, propels its fusion domain upwards to engage with the host membrane, casting off S1 in the process (Song et al., 2018; Wrapp et al., 2020) . Structural studies of the RBD in complex with ACE2 (Lan et al., 2020; Wang et al., 2020; Yan et al., 2020) show that it is recognised by the extracellular peptidase domain (PD) of ACE2 through mainly polar interactions. The Spike protein is an attractive candidate for both vaccine development and immunotherapy. Potent nanomolar affinity neutralising human monoclonal antibodies against the SARS-CoV-1 RBD have been identified that attach at the ACE2 receptor binding site (including M396, CR3014 and 80R (Ter Meulen et al., 2006; Sui et al., 2004; Zhu et al., 2007) ). For example 80R binds with nanomolar affinity, prevents binding to ACE2 and the formation of syncytia in vitro, and inhibits viral replication in vivo (Sui et al., 2004) . However, despite the two viruses sharing the same ACE2 receptor these ACE2 blocking antibodies do not bind SARS-CoV-2 RBD (Wrapp et al., 2020) . In contrast CR3022, a SARS-CoV-1-specific monoclonal selected from a single chain Fv phage display library constructed from lymphocytes of a convalescent SARS patient and reconstructed into IgG1 format (Ter Meulen et al., 2006) , has been reported to cross-react strongly, binding to the RBD of SARS-CoV-2 with a K D of 6.3 nM (Tian et al., 2020) , whilst not competing with the binding of ACE2 (Ter Meulen et al., 2006) . Furthermore, although SARS-CoV-1 escape mutations could be readily generated for ACE2-blocking CR3014, no escape mutations could be generated for CR3022, preventing mapping of its epitope (Ter Meulen et al., 2006) . Furthermore a natural mutation of SARS-CoV-2 has now been detected at residue 495 (Y N) (GISAID (Shu and McCauley, 2017) :

Accession ID: EPI_ISL_429783 Wienecke-Baldacchino et al., 2020), which forms part of the ACE2 binding epitope. Finally, CR3022 and CR3014 act synergistically to neutralise SARS-CoV-1 with extreme potency (Ter Meulen et al., 2006) . Whilst this work was being prepared for publication a paper reporting that CR3022 does not neutralise SARS-CoV-2 and describing the structure of the complex with the RBD at 3.1 Å resolution was published (Yuan et al., 2020) . Here we report crystallographic analysis to significantly higher resolution, use a different neutralisation assay to show that CR3022 does neutralise SARS-CoV-2 and use cryo-EM analysis of the interaction of CR3022 with the full Spike ectodomain to demonstrate a mechanism of neutralisation not seen before for coronaviruses.

Taken together these observations suggest that the CR3022 epitope should be a major target for therapeutic antibodies.

To understand how CR3022 works we first investigated the interaction of CR3022 Fab with isolated recombinant SARS-CoV-2 RBD, both alone and in the presence of ACE2. Surface plasmon resonance (SPR) measurements (Methods and Figure S1 ) confirmed that CR3022 binding to RBD is strong (although weaker than the binding reported to SARS-CoV-1 (Ter Meulen et al., 2006) ), with a slight variation according to whether CR3022 or RBD is used as the analyte (K D = 30 nM and 15 nM respectively, derived from the kinetic data in Table S1 ).

An independent measure using Bio-Layer Interferometry (BLI) with RBD as analyte gave a K D of 19 nM (Methods and Figure S1 ). These values are quite similar to those reported by Tian et al. (Tian et al., 2020 ) (6.6 nM), whereas weaker binding (K D ~ 115 nM) was reported recently by Yuan et al. (Yuan et al., 2020) . Using SPR to perform a competition assay revealed that the binding of ACE2 to the RBD is perturbed by the presence of CR3022 ( Figure S1 ). The presence of ACE2 slows the binding of CR3022 to RBD and accelerates the dissociation. Similarly, the release of ACE2 from RBD is accelerated by the presence of CR3022. These observations are suggestive of an allosteric effect between ACE2 and CR3022.

A plaque reduction neutralisation test using SARS-CoV-2 virus and CR3022 showed a probit mid-point PRNT 50 of 1:11,966 (95% confidence interval 5,297-23,038) for a starting concentration of 1.36mg/mL (calculated according to Grist (Grist, 1966) ), superior to that of the NIBSC international standard positive control used by Public Health England (MERS convalescent serum which gives a PRNT 50 of 1:874 (95% confidence interval 663-1220), see Methods, Figure 2 and Table S2 ). This corresponds to 50% neutralisation at ~0.114 µg/mL (~ 1 nM) exceeding the 11 µg/mL reported by Ter Meulen et al. (Ter Meulen et al., 2006) for SARS-CoV-1, however, as discussed below, it is in apparent disagreement with the result reported recently by Yuan et al. (Yuan et al., 2020) . In light of this discrepancy, further neutralisation tests were performed to rule out differences in the assay with regard to antibody/virus contact time. Repeated PRNT tests deliberately using three different batches of CR3022 gave similar results (Table S2 ) and leaving the virus/antibody mix in place throughout the incubation on the plate and removing the antibody after one hour also gave similar results (PRNT 500 values of 1:4,666 and 1:6,504 respectively, Methods, Figure S2 ). In summary all of these results, taking into consideration the different CR3022 starting concentrations, were within the same confidence levels. Following these experiments a commercial source of antibody CR3022 (Creative Biolabs) was tested (using the same method and on the same date as the above wash and leave experiment, with a starting concentration of 1 mg/mL). This gave markedly weaker neutralisation: PRNT 50 1:27 leaving the antibody on the plate and 1:285 washing it off. Note that in both cases the neutralisation was slightly higher when the antibody was washed off. Although the differences were within the confidence levels of the experiments, it is possible that this reflects unbound virus remaining in the inoculum being washed off.

We determined the crystal structure of the SARS-CoV-2 RBD-CR3022 Fab complex (see Methods and Table S3 ) to investigate the relationship between the binding epitopes of ACE2 and CR3022. Crystals grew rapidly and consistently. Two crystal forms grew in the same drop. The solvent content of the crystal form solved first was unusually high (ca 87%) with the ACE2 binding site exposed to large continuous solvent channels within the crystal lattice ( Figure S3 ). These crystals therefore offer a promising vehicle for crystallographic screening to identify potential therapeutics that could act to block virus attachment. The current analysis of this crystal form is at 4.4 Å resolution and so, to avoid overfitting, refinement used a real-space refinement algorithm to optimise the phases (Vagabond, HMG unpublished, see Methods) . This, together with the favourable observation to parameter ratio resulting from the exceptionally high solvent content, meant that the map was of very high quality, allowing reliable structural interpretation ( Figure S4 , Methods). Full interpretation of the detailed interactions between CR3022 and the RBD was enabled by the second crystal form which diffracted to high resolution, 2.4 Å, and the structure of which was refined to give an R-work/R-free of 0.213/0.239 and good stereochemistry (Methods, Table S3, Figure S4 ). The structure is similar to that reported by Yuan et al., (2020) , the RMSD in Cαs for the RBD is 0.5 Å, whilst for the CR3022 heavy chain it is 1.1 Å and for the light chain 0.7 Å. There are also some differences in the overall interaction compared to that structure, after overlapping the RBD, the angular differences for the variable domains are 5.5 and 8º.

The high-resolution structure is shown in Figure 3 . There are two complexes in the crystal asymmetric unit with residues 331-529 in one RBD, 332-445 and 448-532 in the other RBD well defined, whilst residues 133-136 of the CR3022 heavy chains are disordered. The RBD has a very similar structure to that seen in the complex of SARS-CoV-2 RBD with ACE2, rmsd for 194 Cα atoms of 0.6 Å 2 (PDB, 6M0J (Lan et al., 2020) ), and an rmsd of 1.1 Å 2 compared to the SARS CoV-1 RBD (PDB, 2AJF (Li et al., 2005) ). Only minor conformational changes are introduced by binding to CR3022, at residues 381-390. The RBD was deglycosylated (Methods) to leave a single saccharide unit at each of the N-linked glycosylation sites clearly seen at N331 and N343 ( Figure S4 ). CR3022 attaches to the RBD surface orthogonal to the ACE2 receptor binding site. There is no overlap between the epitopes and indeed both the Fab and ACE2 ectodomain can bind without clashing ( Figure   3D ) (Tian et al., 2020) . Such independence of the ACE2 binding site has been reported recently for another SARS-CoV-2 neutralising antibody, 47D11 (Okba et al., 2019) . The Fab complex interface buries 990 Å 2 of surface area (600 and 390 Å 2 by the heavy and light chains respectively, Figure 4 and Figure S5 ), somewhat more than the RBD-ACE2 interface which covers 850 Å 2 (PDB 6M0J (Lan et al., 2020) ). Typical of a Fab complex, the interaction is mediated by the antibody CDR loops, which fit well into the rather sculpted surface of the RBD ( Figure 3B , C). The heavy chain CDR1, 2 and 3 make contacts to residues from α2, β2 and α3 (residues 369-386), while two of the light chain CDRs (1 and 2) interact mainly with residues from the β2-α3 loop, α3 (380-392) and the α5-β4 loop (427-430) ( Figures 1, 3 & 5) . A total of 16 residues from the heavy chain and 14 from the light chain cement the interaction with 26 residues from the RBD. For the heavy chain these potentially form 7 H-bonds and 3 salt bridges, the latter from D55 and E57 (CDR2) to K378 of the RBD. The light chain interface comprises 6 H-bonds and a single salt bridge between E61 (CDR2) and K386 of the RBD. The binding is consolidated by a number of hydrophobic interactions ( Figure 5 ). There are slight differences in the interactions between these and those reported by Yuan et al. (2020) , for instance the contact area for the light chain-RBD differs by ~12.5% between the two structures. Of the 26 residues involved in the interaction, 23 are conserved between SARS-CoV-1 and SARS-CoV-2 (Figures 1 & 4) . The CR30222 epitope is much more conserved than that of the receptor blocking anti-SARS-CoV-1 antibody 80R for which only 13 of the 29 interacting residues are conserved (Hwang et al., 2006) , in-line with the lack of cross reactivity observed for the latter.

The reason for the conservation of the CR3022 epitope becomes clear in the context of the complete pre-fusion S structure (PDB IDs: 6VSB (Wrapp et al., 2020) , 6VXX, 6VYB (Walls et al., 2020) ) where the epitope is inaccessible ( Figure 6 ). When the RBD is in the 'down' configuration the CR3022 epitope is packed tightly against another RBD of the trimer and the N-terminal domain (NTD) of the neighbouring protomer. In the structure of the pre-fusion form of trimeric Spike the majority of RBDs are 'down', although presumably stochastically one may be 'up' (Walls et al., 2020; Wrapp et al., 2020) . The structure of a SARS-CoV-1 complex with ACE2 ectodomain shows that this 'up' configuration is competent to bind receptor, and that there are a family of 'up' orientations with significantly different hinge angles (Song et al., 2018) . However, the CR3022 epitope remains largely inaccessible even in the 'up' configuration. Modelling the rotation of the RBD required to enable Fab interaction in the context of the Spike trimer, showed a rotation corresponding to a > 60° further declination from the central vertical axis was required, beyond that observed previously (Walls et al., 2020; Wrapp et al., 2020) (Figure 6I ), although this might be partly mitigated by more complex movements of the RBD and if more than one RBD is in the 'up' configuration this requirement would be relaxed somewhat. Since locking the up state by receptor blocking antibodies is thought to destabilise the pre-fusion state (Walls et al., 2019) binding of CR3022 presumably introduces further destabilisation, leading to a premature conversion to the post-fusion state, inactivating the virus. CR3022 and ACE2 blocking antibodies can bind independently but both induce an 'up' conformation, presumably explaining the observed synergy between binding at the two sites (Ter Meulen et al., 2006) .

To test if CR3022 binding destabilises the prefusion state of Spike, the ectodomain construct described previously (Wrapp et al., 2020) was used to produce glycosylated protein in HEK cells (Methods). Cryo-EM screening showed that the protein was in the trimeric prefusion conformation. Spike was then mixed with an excess of CR3022 Fab and incubated at room temperature, with aliquots being taken at 50 minutes and 3 h. Aliquots were immediately applied to cryo-EM grids and frozen (Methods). For the 50 minutes incubation, collection of a substantial amount of data allowed unbiased particle picking and 2D classification which revealed two major structural classes with a similar number in each, (i) the prefusion conformation, and (ii) a radically different structure (Methods, Table S4 and Figure S6 ).

Detailed analysis of the prefusion conformation led to a structure at a nominal resolution of 3.4 Å (FSC = 0.143), based on a broad distribution of orientations, that revealed the same predominant RBD pattern (one 'up' and two 'down') previously seen (Wrapp et al., 2020) with no evidence of CR3022 binding ( Figure 7A ). Analysis of the other major particle class revealed strong preferential orientation of the particles on the grid ( Figure S6C ). Despite this a reconstruction with a nominal resolution of 3.9 Å within the plane of the grid, and perhaps 7 Å resolution in the perpendicular direction ( Figure S6G ), could be produced which allowed the unambiguous fitting of the CR3022-RBD complex ( Figure 7B ). Note that in addition there is less well defined density attached to the RBD, in a suitable position to correspond to the Spike N-terminal domain (Wrapp et al., 2020) . These structures are no longer trimeric, rather two complexes associate to form an approximately symmetric dimer (however, application of this symmetry in the reconstruction process did not improve the resolution).

The interactions responsible for dimerisation involve the ACE2 binding site on the RBD and the elbow of the Fab, however the interaction does not occur in our low-resolution crystal form and is therefore probably extremely weak and not biologically significant. Since conversion to the post-fusion conformation leads to dissociation of S1 (which includes the Nterminal domain and RBD) these results confirm that CR3022 destabilises the prefusion Spike conformation. Further evidence of this is provided by analysis of data collected after 3 h incubation. By this point there were no intact trimers remaining and a heterogeneous range of oligomeric assemblies had appeared, which we were not able to interpret in detail but which are consistent with the lateral assembly of Fab/RBD complexes ( Figure S7 ). Note that the relatively slow kinetics will not be representative of events in vivo, where the conversion might be accelerated by the elevated temperature and the absence of the mutations which were added to this construct to stabilise the prefusion state (Kirchdoerfer et al., 2018; Pallesen et al., 2017; Wrapp et al., 2020) .

Until now the only documented mechanism of neutralisation of coronaviruses has been through blocking receptor attachment. In the case of SARS-CoV-1 this is achieved by presentation of the RBD of the Spike in an 'up' conformation.

Although not yet confirmed for SARS-CoV-2 it is very likely that a similar mechanism can apply. Here we define a second class of neutralisers, that bind a highly conserved epitope ( Figure 1 ) and can therefore act against both SARS-CoV-1 and SARS-CoV-2 (CR3022 was first identified as a neutralising antibody against SARS-CoV-1 (Ter Meulen et al., 2006) ). We find that binding of CR3022 to the isolated RBD is tight (~20 nM) and the crystal structure of the complex reveals the atomic detail of the interaction. Despite the spatial separation of the CR3022 and ACE2 epitopes we find an allosteric effect between the two binding events. The role of the CR3022 epitope in stabilising the prefusion Spike trimer explains why it has, to date, proved impossible to generate mutations that escape binding of the antibody (Ter Meulen et al., 2006) .

Whilst in our assay CR3022 strongly neutralises SARS-CoV-2, a recent paper (Yuan et al., 2020) reported an alternative assay that did not detect neutralisation. We tested whether the removal of the antibody/virus mix after adsorption to the indicator cells, performed by Yuan et al, before incubating to allow cytopathic effect (CPE) to develop, would explain this difference. This would be in-line with the distinction previously seen between neutralisation tests for influenza virus by antibodies which bind the stem of hemagglutinin and therefore do not block receptor binding (Thomson et al., 2012) . These antibodies did not appear to be neutralising when tested with the standard WHO neutralisation assay, in which similar to Yuan et al, the inoculum of virus/antibody is washed out before development of CPE.

Neutralisation was observed, however, when the antibodies were left in the assay during incubation to produce CPE. Performing side-by-side PRNT experiments leaving the antibody/virus mix in place and washing it off did not however show a significant difference.

In fact, the neutralisation was marginally stronger when excess antibody and virus was washed off. To check if there were issues related to the reproducibility we performed neutralisation tests on three separate batches of CR3022. All gave essentially indistinguishable results (Table S2 ), however when we tested commercially sourced CR3022 (Creative BioLabs, USA; CAT#: MRO-1214LC) the neutralisation was markedly reduced, perhaps improperly folded antibody. It is possible the loss of neutralisation ability with commercial antibody may be related to the report that CR3022 does not neutralise SARS-CoV-2 (Yuan et al., 2020) . In addition, we note that in all the PRNT tests performed, CR3022 appears to give strong but incomplete (90% plaque reduction) neutralisation. Such partial neutralisation has been reported before, for antibodies against Ebola virus, which nonetheless confer profound protection (Rijal et al., 2019; Saphire et al., 2018) . Given the mechanism of neutralisation we rationalise that this arises from the kinetic limitation of antibody binding and Spike destruction, as seen by cryo-EM, where in the absence of ACE2, the CR3022 Fab destroys the prefusion-stabilised trimer with T 1/2 ~1h at room temperature. This might also lead to slightly higher neutralisation when antibody and (non-inactivated) virus is washed off the cells after one hour. In summary CR3022 neutralises SARS-CoV-2, but via an unusual mechanism that some assays appear to detect poorly, as observed by Yuan et al, 2020 . It is now important to establish how effective this mechanism is at controlling viral infection.

With monoclonal antibodies now recognised as potential antivirals (Lu et al., 2020b; Qiu et al., 2014; Salazar et al., 2017) our results suggest that CR3022 may be of immediate utility, since the mechanism of neutralisation will be unusually resistant to virus escape. In contrast antibodies which compete with ACE2 (whose epitope on SARS-CoV-2 is reported to have already shown mutation at residue 495 (GISAID: Accession ID: EPI_ISL_429783 Wienecke-Baldacchino et al., 2020 (Shu and McCauley, 2017) ), are likely to be susceptible to escape. Furthermore, with knowledge of the detailed structure of the epitope presented here a higher affinity version of CR3022 might be engineered. Alternatively, since the same mechanism of neutralisation is likely to be used by other antibodies, a more potent monoclonal antibody targeting the same epitope might be found (for instance by screening for competition with CR3022). Additionally, since this epitope is sterically and functionally independent of the well-established receptor-blocking neutralising antibody epitope there is considerable scope for therapeutic synergy between antibodies targeting the two epitopes (indeed this type of synergy has been described for SARS-CoV-1 (Ter Meulen et al., 2006) ). Moreover, it has been reported (Wan et al., 2019) Residue numbers are those of SARS-CoV-2 RBD, conserved amino acids have a red background, secondary structures are labelled on the top of the sequence, and the glycosylation site is marked with a blue hexagon. Residues involved in receptor binding are marked with magenta disks. Blue disks mark the residues involved in interactions with the CR3022 heavy chain (Vh), cyan disks mark the residues interacting with the CR3022 light chain (Vl) and green disks with both chains.

For CR3022 at a starting concentration of 1.36mg/mL, the dilutions used were from 1:160 to 1:327,680. The probit mid-point is 1:11,966 (95% confidence intervals: 5,297 -23,038). 

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, David I Stuart (dave@strubi.ox.ac.uk).

Plasmids generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

The high resolution and lower resolution coordinates and structure factors of the SARS-CoV-2 RBD/CR3022 complex are available from the PDB with accession codes 6YLA and 6YM0 respectively (https://www.rcsb.org/). EM maps and structure models are deposited in EMDB and PDB with accession codes EMD-11119 and 6Z97 for the prefusion Spike, and EMD-10863 and 6YOR for the dimeric RBD/CR3022 complex respectively (https://www.emdataresource.org/). The data that support the findings of this study are available from the corresponding authors on request. 

To further validate the SPR results the K D of Fab CR3022 for RBD was also measured by bio-layer interferometry. Kinetic assays were performed on an Octet Red 96e (ForteBio) at 30 ℃ with a shake speed of 1000 rpm. Fab CR3022 was immobilized onto amine reactive 2nd generation (AR2G) biosensors (ForteBio) and serially diluted RBD (80，40，20，10 and 5 nM) was used as analyte. PBS (pH 7.4) was used as the assay buffer. Recorded data were analysed using the Data Analysis Software HT v11.1 (Fortebio), with a global 1:1 fitting model. The plate was incubated at 37 o C in a humidified box for 1 h to allow neutralisation to take place, before the virus-antibody mixture was transferred into the wells of a twice DPBSwashed 24-well plate containing confluent monolayers of Vero E6 cells. Virus was allowed to adsorb onto cells at 37 o C for a further hour in a humidified box before being overlaid with MEM containing 1.5% carboxymethylcellulose (Sigma, Dorset, UK), 4% (v/v) FCS and 25mM HEPES buffer. After 5 days incubation at 37 o C in a humidified box, the plates were fixed overnight with 20% formalin/PBS (v/v), washed with tap water and then stained with 0.2% crystal violet solution (Sigma) and plaques were counted. A mid-point probit analysis (written in R programming language for statistical computing and graphics) was used to determine the dilution of antibody required to reduce SARS-CoV-2 viral plaques by 50%

(PRNT50) compared with the virus only control (n=5). The script used in R was based on a source script from (Johnson et al., 2013) . Antibody dilutions were run in duplicate and an internal positive control for the PRNT assay was also run in duplicate using a sample of heatinactivated (56 ○ C for 30 min) human MERS convalescent serum known to neutralise SARS-CoV-2 (National Institute for Biological Standards and Control, UK). This protocol was repeated in two further experiments with CR3022 (from a different batch) at a starting concentration of 1 mg/mL to compare leaving the virus/antibody mixture on the plate and in parallel with washing it off before the addition of overlay media.

Purified and deglycosylated RBD and CR3022 Fab were concentrated to 8.3 mg/mL and 11 mg/mL respectively, and then mixed in an approximate molar ratio of 1:1. Crystallization screen experiments were carried out using the nanolitre sitting-drop vapour diffusion method in 96-well plates as previously described (Walter et al., 2003 (Walter et al., , 2005 . Crystals were initially Crystals were mounted in loops and frozen in liquid nitrogen prior to data collection.

Diffraction data were collected at 100 K at beamline I03 of Diamond Light Source, UK.

Diffraction images of 0.1° rotation were recorded on an Eiger2 XE 16M detector (exposure time of either 0.002 s or 0.01 s per frame, beam size 80×20 μm and 100% beam transmission). Data were indexed, integrated and scaled with the automated data processing program Xia2-dials (Winter, 2010; Winter et al., 2018) . The data set of 720° was collected from a single frozen crystal to 4.4 Å resolution with 52-fold redundancy. The crystal belongs to space group P4 1 2 1 2 with unit cell dimensions a = b = 150.5 Å and c = 241.6 Å. The structure was determined by molecular replacement with PHASER (McCoy et al., 2007) using search models of human germline antibody Fabs 5-51/O12 (PDB ID, 4KMT (Teplyakov et al., 2014) ) heavy chain and IGHV3-23/IGK4-1 (PDB ID, 5I1D (Teplyakov et al., 2016) ) light chain, and RBD of SARS-CoV-2 RBD/ACE2 complex (PDB ID, 6M0J (Lan et al., 2020) ). There is one RBD/CR3022 complex in the crystal asymmetric unit, resulting in a crystal solvent content of ~87%.

During optimisation of the crystallisation conditions, a second crystal form was found to grow in the same condition with similar morphology. A data set of 720° rotation with data extending to 2.4 Å was collected on beamline I03 of Diamond from one of these crystals (exposure time 0.004 s per 0.1° frame, beam size 80×20 μm and 100% beam transmission).

The crystal also belongs to space group P4 1 2 1 2 but with significantly different unit cell dimensions (a = b = 163.1 Å and c = 189.1 Å). There were two RBD/CR3022 complexes in the asymmetric unit and a solvent content of ~74%.

The initial structure was determined using the lower resolution data from the first crystal form. Data were excluded at a resolution below 35 Å as these fell under the beamstop shadow. One cycle of REFMAC5 (Murshudov et al., 2011) was used to refine atomic coordinates after manual correction in COOT (Emsley and Cowtan, 2004) Data collection and structure refinement statistics are given in Table S3 . Structural comparisons used SHP (Stuart et al., 1979) , residues forming the RBD/Fab interface were identified with PISA (Krissinel and Henrick, 2007) , figures were prepared with PyMOL (The PyMOL Molecular Graphics System, Version 1.2r3pre, Schrödinger, LLC).

Purified spike protein was buffer exchanged into 2 Table S4 .

For both the 50 minute and 3 h incubation datasets, motion correction and alignment of 2x binned super-resolution movies was performed using Relion3.1. CTF-estimation with GCTF (v1.06) (Zhang, 2016) and non-template-driven particle picking was then performed within cryoSPARC v2.14.1-live (https://cryosparc.com/) followed by multiple rounds of 2D classification (Punjani et al., 2017) .

For the 50 minutes dataset, 2D class averages for particle groups A and B were used separately for template-driven classification before further rounds of 2D and 3D classification with C1 symmetry. Both structures were then sharpened in cryoSPARC. Data processing and refinement statistics are given in Table S4 .

An initial model for the Spike (group A) was generated using PDB ID, 6VYB (Walls et al., 2020) and rigid body fitted into the final map using COOT (Emsley and Cowtan, 2004) . The For the 3 h incubation dataset, particles were extracted with a larger box size (686 pixels as compared to 540 pixels), and, following multiple rounds of 2D classification, 2D class averages from 'blob-picked' particles showing signs of complete 'flower-like' structures were selected for ab initio reconstruction (in some classes, petals from these flower-like particles were missing, Figure S7 ). For the 3 h data no detailed fitting was attempted.

SPR kinetic data were fitted using Biacore T200 Evaluation software 3.1 (www.cytivalifesciences.com). BLI data were analysed using data analysis software HT V 11.1 (www.fortebio.com). PRNT neutralisation data were subjected to mid-point probit 

The authors declare no competing interests.

Binding affinity between RBD and CR3022 Fab, related to STAR Methods, (Surface plasmon resonance and Bio-layer Interferometry).

(A-B) Surface plasmon resonance binding sensorgrams measured with a Biacore T200. Biotinylated (Bio-) RBD was immobilised as the ligand and CR3022 Fab was used as analyte at five concentrations (5.9, 11.9, 23.8, 47.5 and 95 nM). (C-D) CR3022 IgG was immobilised as the ligand and RBD-His was used as analyte at five concentrations (6.25, 12.5, 25, 50, 100 nM). Data were fitted to a 1:1 binding model using the Biacore T200 Evaluation Software 3.1. The average kinetic values from these two sets of experiment are listed in Extended Table 1 

Table S3 X-ray data collection and refinement statistics, related to Figure 3 and STAR Methods (Crystallisation, data collection and X-ray structure determination and X-ray crystallographic refinement and electron density map generation). 

A time-and cost-efficient system for highlevel protein production in mammalian cells

Real estimates of mortality following COVID-19 infection

Isolation and rapid sharing of the 2019 novel coronavirus (SAR-CoV-2) from the first patient diagnosed with COVID-19 in Australia

Coot: Model-building tools for molecular graphics

Diagnostic methods in clinical virology. x

Structural basis of neutralization by a human anti-severe acute respiratory syndrome spike protein antibody, 80R

Acaricide, Fungicide and Drug Interactions in Honey Bees (Apis mellifera)

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

Inference of Macromolecular Assemblies from Crystalline State

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

LigPlot+: Multiple ligand-protein interaction diagrams for drug discovery

Structural biology: Structure of SARS coronavirus spike receptor-binding domain complexed with receptor

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

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

Development of therapeutic antibodies for the treatment of diseases

Phaser crystallographic software

A SARS-like cluster of circulating bat coronaviruses shows potential for human emergence

Human monoclonal antibody combination against SARS coronavirus: Synergy and coverage of escape mutants

REFMAC5 for the refinement of macromolecular crystal structures

A pipeline for the production of antibody fragments for structural studies using transient expression in HEK 293T cells

The production of glycoproteins by transient expression in mammalian cells

Transient expression in HEK 293 cells: An alternative to E. coli for the production of secreted and intracellular mammalian proteins

Early Release-Severe Acute Respiratory Syndrome Coronavirus 2−Specific Antibody Responses in Coronavirus Disease

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

Immunopathogenesis of coronavirus infections: Implications for SARS

CryoSPARC: Algorithms for rapid unsupervised cryo-EM structure determination

Reversion of advanced Ebola virus disease in nonhuman primates with ZMapp

Therapeutic Monoclonal Antibodies for Ebola Virus Infection Derived from Vaccinated Humans

Dynamical asymmetry exposes 2019-nCoV prefusion spike

Antibody therapies for the prevention and treatment of viral infections

Systematic Analysis of Monoclonal Antibodies against Ebola Virus GP Defines Features that Contribute to Protection

GISAID: Global initiative on sharing all influenza datafrom vision to reality

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

Crystal structure of cat muscle pyruvate kinase at a resolution of 2.6 Å

Potent neutralization of severe acute respiratory syndrome (SARS) coronavirus by a human mAb to S1 protein that blocks receptor association

Antibody modeling assessment II

Structural diversity in a human antibody germline library

Pandemic H1N1 Influenza Infection and Vaccination in Humans Induces Cross-Protective Antibodies that

Potent binding of 2019 novel coronavirus spike protein by a SARS coronavirusspecific human monoclonal antibody

Immunization with SARS coronavirus vaccines leads to pulmonary immunopathology on challenge with the SARS virus

Unexpected Receptor Functional Mimicry Elucidates Activation of Coronavirus Fusion

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

A procedure for setting up high-throughput nanolitre crystallization experiments. I. Protocol design and validation

A procedure for setting up highthroughput nanolitre crystallization experiments. Crystallization workflow for initial screening, automated storage, imaging and optimization

Molecular Mechanism for Antibody-Dependent Enhancement of Coronavirus Entry

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

Xia2: An expert system for macromolecular crystallography data reduction

DIALS: Implementation and evaluation of a new integration package

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

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

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

Gctf: Real-time CTF determination and correction

Potent cross-reactive neutralization of SARS coronavirus isolates by human monoclonal antibodies

• CR3022 binds the RBD of SARS-CoV-2 and shows strong neutralization • Neutralisation is by destroying the prefusion SPIKE conformation • CR3022 binds a highly conserved epitope that is inaccessible in prefusion S protein • CR3022 may have therapeutic potential alone or in synergy with a receptor blocker