key: cord-1008595-sbaq8dbr
authors: Asarnow, Daniel; Wang, Bei; Lee, Wen-Hsin; Hu, Yuanyu; Huang, Ching-Wen; Faust, Bryan; Lon Ng, Patricia Miang; Xian Ngoh, Eve Zi; Bohn, Markus; Bulkley, David; Pizzorno, Andrés; Ary, Beatrice; Tan, Hwee Ching; Lee, Chia Yin; Minhat, Rabiatul Adawiyah; Terrier, Olivier; Soh, Mun Kuen; Teo, Frannie Jiuyi; Chin Yeap, Yvonne Yee; Kheng Seah, Shirley Gek; Zuo Chan, Conrad En; Connelly, Emily; Young, Nicholas J.; Maurer-Stroh, Sebastian; Renia, Laurent; Hanson, Brendon John; Rosa-Calatrava, Manuel; Manglik, Aashish; Cheng, Yifan; Craik, Charles S.; Wang, Cheng-I
title: Structural insight into SARS-CoV-2 neutralizing antibodies and modulation of syncytia
date: 2021-04-24
journal: Cell
DOI: 10.1016/j.cell.2021.04.033
sha: e2bd2c5f2157864ee19de06f40a9f0d182fe82eb
doc_id: 1008595
cord_uid: sbaq8dbr

Infection by SARS-CoV-2 is initiated by binding of viral Spike protein to host receptor angiotensin-converting enzyme 2 (ACE2), followed by fusion of viral and host membranes. While antibodies that block this interaction are in emergency use as early COVID-19 therapies, precise determinants of neutralization potency remain unknown. We discovered a series of antibodies that all potently block ACE2 binding, yet exhibit divergent neutralization efficacy against live virus. Strikingly, these neutralizing antibodies can either inhibit or enhance Spike-mediated membrane fusion and formation of syncytia, which are associated with chronic tissue damage in COVID-19 patients. Multiple cryogenic electron microscopy structures of Spike-antibody complexes reveal distinct binding modes that not only block ACE2 binding, but also alter the Spike protein conformational cycle triggered by ACE2 binding. We show that stabilization of different Spike conformations leads to modulation of Spike-mediated membrane fusion, with profound implications in COVID-19 pathology and immunity.

The first step of viral infection by coronaviruses such as SARS-CoV and 2 SARS-CoV-2 is the binding of a Spike protein on the virion to a specific receptor in 3 the membrane of a host cell (Tortorici and Veesler, 2019) . The virus enters cells by 4 fusion of the viral envelope with cellular plasma membranes and alternatively by 5 endocytosis and subsequent fusion of the viral envelope with endosomal 6 membranes. The SARS-CoV-2 Spike protein, similar to that of other coronaviruses, 7 comprises two subunits, S1 and S2, and is responsible for target recognition and 8 mediating viral entry (Tortorici and Veesler, 2019) . Upon binding to the host cell 9

receptor through the receptor binding domain (RBD) at the tip of the S1 subunit, the 10 Spike protein undergoes dramatic conformational changes and proteolytic 11

processing. Further shedding of the S1 subunit exposes the S2 subunit fusion 12 peptide, which inserts into the host cell membrane and induces viral fusion (Benton 13 J o u r n a l P r e -p r o o f replication in HAE was reduced 1000-fold by 5A6 at 75 ng/mL and 10,000-fold at 150 1 ng/mL, and 5A6 also helped maintain epithelium integrity (represented by trans-2 epithelia electrical resistance), supporting its activity in a physiologically relevant in 3 vitro model ( Figure 2C ). 4

All IgGs effectively block ACE2-RBD binding, with IC 50 values below 50 5 nanomolar ( Figure S1A ). Therefore, the relative viral neutralization potencies of 6 these antibodies cannot be ascribed to competitive receptor blocking alone. In order 7

to interrogate other determinants of neutralization, we next compared the potency of 8 each IgG antibody to their respective monomeric Fab fragments. A bivalent ACE2-Fc 9 fusion protein was included as a reference for multivalent receptor blockade. All 10 antibodies show dramatically increased potency against live virus compared to Fab, 11 which is consistent with bivalent engagement of the Spike trimer by the IgG 12 compared to the monovalent Fab fragment. The affinity or avidity for RBD is 13 generally predictive of viral neutralization IC 50 , with two striking exceptions ( Figure  14 2D, Table S2 ). Antibody 5A6 exhibits far greater viral neutralization potency than 15 other antibodies with superior avidity. Conversely, 3D11 is among the least potent in 16 viral neutralization despite displaying the strongest binding. 17 We asked if the discrepancy could arise from the differences in the structural 18 arrangement of IgGs bound to the RBDs of intact, trimeric Spike on the virion. We 19 used surface plasmon resonance (SPR) to measure antibody binding to immobilized 20

Spike trimers, as opposed to immobilized RBD ( Figure S2B -S2D, Table S2 ). 21

Although kinetics of binding to trimer were largely similar, we noted that 5A6 IgG 22 binds somewhat more tightly (3.6x) to the intact trimer than to the flexible Fc-RBD 23 construct used for BLI, while 3D11 IgG binds much (18.7x) more weakly (but still 24 apparently bivalently, with 21.7x tighter binding than 3D11 Fab). To further 25 J o u r n a l P r e -p r o o f investigate the relationship of the antibodies to intact Spike assemblies, we purified 1 SARS-CoV-2 pseudovirus by gradient centrifugation and immobilized the viral 2 particles on ELISA plates. The higher optical signal at saturation in concentration-3 dependent binding curves reveals that 5A6 IgG likely packs with higher density on 4 the viral surface than the other four tested IgG antibodies or 5A6 Fab ( Figure 2E ). It 5 has been proposed that effective viral neutralization requires antibody packing 6 density exceeding a critical threshold (Burton et al., 2001; Dowd and Pierson, 2011; 7 Flamand et al., 1993) and 5A6 may possess a unique binding mode that 8 accommodates a denser structural arrangement on the viral surface. Notably, 3D11 9 IgG exhibits a similarly high signal at saturation, but with lower affinity for the 10 pseudoviral particles, despite having higher affinity than 5A6 for immobilized RBD or 11

Spike trimer ( Figure 2E , Figure S2 ). Finally, 2H4 and 1F4 IgGs saturate immobilized 12 pseudovirus at 1/3 the density of 5A6 or 3D11, while neutralizing live virus slightly 13 more effectively than 3D11. These results suggest at least three different classes of 14 receptor-blocking antibodies with distinct structural relationships to RBDs on viral 15 particles. 16

It is widely appreciated that viral proteins often possess multiple critical (Koot et al., 1993; Sylwester et al., 1997) , and now 1 widely observed in late-stage COVID-19 (Bussani et al., 2020) . We therefore 2 assessed whether antibodies discovered in our campaign inhibit Spike-mediated 3 syncytia formation. To directly examine syncytia formation by Spike alone, we 4 expressed Spike protein with a C-terminal fluorescent tag in Vero E6 cells. Addition 5 of trypsin as an exogenous Spike-processing enzyme resulted in cells with a diffuse 6 fluorescent signal and multiple nuclei, indicative of syncytia formation. ( Figure 3A) . 7

We assayed the impact of receptor-blocking antibodies on this trypsin-induced cell-8 cell fusion, using antibodies 2H4, 5A6 and 3D11, which represent different modes of 9 viral neutralization. The 5A6 IgG has a dose-dependent inhibitory effect on syncytia 10 ( Figure 3B ). By contrast 2H4 IgG has no significant effect. Surprisingly, 3D11 11 potentiates cell-cell fusion. Furthermore, 5A6 Fabs also enhanced syncytial fusion, 12 albeit weakly ( Figure 3A -B). We conclude that 5A6 IgG directly inhibits Spike-13 mediated fusion, while other receptor blocking antibodies fail to inhibit or even 14 accelerate this process. 15

The Spike trimer exists in equilibrium between the closed conformation, with 17 all RBDs nestled closely around the S2 subunit, and "receptor seeking" states trimer, and bind with distinct geometries relative to the RBD and ACE2 interface 3 ( Figure 4 ). This form of three-dimensional epitope mapping provides atomic level 4 understanding of the determinants of viral inhibition. 5 2H4 is an orthosteric receptor-mimetic antibody 6 We determined multiple structures of 2H4 Fab bound to Spike with resolution 7 sufficient for unambiguously docking a model of 2H4, but precluding precise 8 modelling of the epitope and complementarity determining regions (CDRs) (Figure  9 S3E). Three major conformational states were identified by 3D classification, 10 revealing either one, two, or three 2H4 Fabs bound to the Spike trimer. The receptor 11

blocking activity of 2H4 is straightforward, as it recognizes an epitope that overlaps 12 much of the ACE2 interface ( Figure 4A ). Binding of the 2H4 Fab is compatible with 13 both major RBD conformations, and the structures are drawn from an ensemble of 14 quaternary states reminiscent of those that follow ACE2 binding, and lead to S1 15 shedding and spike-mediated membrane fusion (Benton et al., 2020) ( Figure 4B ). 16 Figure 4C ). These observations suggest 2H4 1 directly blocks receptor binding, but also acts as a receptor-mimetic that admits the 2 same cycle of Spike conformations as does ACE2. A neutralizing antibody against 3 SARS-CoV has also been reported to engage in orthosteric receptor mimicry (Walls 4 et al., 2019) , suggesting activation of fusion-associated conformational changes may 5 be an intrinsic consequence of direct receptor interface binding in betacoronaviruses. 6 3D11 allosterically blocks ACE2 binding and triggers Spike opening 7

The Spike:3D11 complex is relatively homogeneous, with only one major 8 state ( Figure 4D ), and we determined its structure to ~3.0 Å resolution ( Figure S3C ). 2020). We therefore term 3D11 an allosteric receptor-mimetic antibody, which does 2 not directly target the ACE2 interface, yet prevents ACE2 binding and enhances 3 Spike-mediated fusion by rapidly advancing the Spike conformational cycle to its 4 final stages. 5 5A6 traps a pre-fusion conformation to inhibit spike-mediated fusion 6 Multiple states of the Spike:5A6 complex were resolved to better than 3.0 Å, 7 with local resolution sufficient for accurate modelling of the Fab-RBD interface 8 ( Figure S3G ). 5A6 recognizes surface loops near the tip of the RBD, which are 9 solvent exposed even when all RBDs are closed. The binding geometry is 10 permissive of any trimer configuration and any stoichiometry, without steric 11 constraints from Spike or Fab. Yet despite this complete conformational freedom, all 12 Figure 5B) , and the C H1 and C L domains of 5A6 at the cryptic quaternary epitope 24 induce an even more severe clash with ACE2 ( Figure 4G ). Two Fabs thus act 25 synergistically to block ACE2 binding, while one Fab is capable of blocking ACE2 at 1 two RBDs simultaneously. The secondary interface must be released in order for the 2

bound RBD to open, and we hypothesize that 5A6 at its quaternary epitope locks 3 one RBD closed, thereby arresting the trimer in its pre-fusion state. Precise 4 conservation of binding mode and the cryptic quaternary epitope from free Fab to 5

IgG is confirmed by a structure of 5A6 IgG complexed with Spike trimer at ~15 Å 6 resolution ( Figure 5C ). Although the Fc domain is not well resolved due to the 7 flexibility of the hinge region, the structures suggest 5A6 IgG may bind to two RBDs 8 from the same trimer ( Figure S4C ), and shows that no steric effects preclude binding 9

of three IgGs at sufficient concentration. Noting the weak potentiation of syncytia 10 formation by 5A6 Fab, we deduce that the cryptic epitope likely appears following 11 initial Fab binding, and leads to cooperative action against SARS-CoV-2 by imbuing 12 a second binding event with enhanced affinity and receptor blockade. We also 13 conclude that the geometry of the quaternary epitope and avidity of 5A6 IgG drive 14 robust pre-fusion conformational trapping and potent inhibition of Spike-mediated 15 fusion and syncytia formation. 16

Receptor engagement to a Spike RBD locks it in the open conformation and 18 triggers a cooperative process in which the Spike conformational ensemble is driven 19 towards further opening by successive rounds of receptor binding (Benton et al., 20 2020 ). The process culminates in unsheathing of the S2 subunit, and following 21 proteolytic cleavage, shedding of the S1-ACE2 subcomplex. S1 shedding in turn 22

facilitates the post-fusion state transition, leading to membrane fusion and viral entry. We identified six receptor-blocking antibodies that exhibit differences in 7 avidity, binding mode, and neutralization of live virus, despite the fact that all of them stages of S2 unsheathing. In contrast, 5A6 possesses a unique binding mode that 23 stabilizes a Spike conformation that prohibits S1 shedding and traps the pre-fusion 24 state. We hypothesize that synergy between receptor blockade and pre-fusion to 5A6, 3D11 and 2H4 Spike:IgG complexes were not tractable for single-particle 15 cryo-EM ( Figure S4E ). Qualitative image analysis suggests these species do not trap 16 defined conformational states of the Spike trimer, and that 3D11 in particular may 17 achieve viral neutralization by destabilizing the Spike, as does the CR3022 antibody 18 with a similar epitope (Huo et al., 2020) . Intriguingly, we found that 3D11 has greatly 19 reduced potency against pseudovirus bearing D614G Spike, while that of 5A6 is 20 slightly improved, though position 614 is outside the RBD and far from the epitope of 21 either antibody in the RBD ( Figure S6 ). The D614G mutant Spike is known to occupy 22

states with multiple open RBDs (Yurkovetskiy et al., 2020) , and has been found to 23

shed the S1 subunit less readily than the original SARS-CoV-2 Spike protein (Zhang 24 et al., 2020). We can thus understand altered neutralization of D614G pseudovirus 25 by 3D11 and 5A6 in terms of the model of Figure 6 , because the effects of both 1 antibodies are mediated by open RBD conformations that represent immediately 2 available binding sites for 3D11, and present the full quaternary epitope of 5A6. The 3 reduced S1 shedding of the more stable D614G Spike may also assist the pre-fusion 4 trapping activity of 5A6, while conveying resistance against trimer denaturation by 5

The quaternary epitope recognized by 5A6 conveys cooperative binding as 7 well as avidity, and both aspects may hinder viral escape via mutations in Spike during the on-going COVID-19 pandemic. 20

Despite providing structural insights into Spike function in mediating cell-cell 22 fusion, our study does not address how antibodies may influence specific 23 biochemical events preceding fusion, such as proteolytic cleavage. Furthermore, a 24 number of Spike sequences with mutations in addition to D614G have now been 25 reported, and the specific interplay between Spike sequence variation and the 1 conformational changes elicited by antibody binding still remains to be understood. 2

We thank Professor Yee-Joo Tan NIH Grants S10 S10OD020054 (Y.C.) and S10OD021741 (Y.C.). This work was 17 Tagless RBD is introduced as the first analyte. The second antibody is introduced as 14 the second analyte. As controls, buffer alone, an isotype IgG and 5A6 IgG were 15 included as the second analyte. See also Figure S1 , Table S1, and Table S2 . The IC 50 was calculated by a variable slope four parameter non-linear regression 2 model in Graphpad PRISM 7 Software. Pseudo-and live virus neutralization assays 3

were not performed for 6F8 Fab due to 6F8 IgG's low and similar potency to other 4 clones. 5

(C) Evaluation of antiviral activity of 5A6 in a model of reconstituted human airway 6 epithelia (HAE). Viral genome quantification was performed using RT-qPCR, and 7 results are expressed in relative viral production (intracellular or apical) compared to 8 control. Bars represent the mean ± SD in duplicates. ***P <0.001, ** P<0.01 and *P 9

<0.05 compared to the control (no Ab) by one-way ANOVA. The trans-epithelial 10 resistance (TEER in Ω/cm 2 ) was measures at 48hpi. interfaces, and appears to trap the pre-fusion Spike by locking closed one RBD. Table S3 and Movie S1-S3. and ACE2 itself, on the conformational cycle of the SARS-CoV-2 Spike trimer. On 30 the viral surface, the Spike is found predominantly in either the closed conformation, 31 or a "receptor seeking" conformation with one RBD open. When serially bound by 1 ACE2 or an orthosteric mimetic antibody like 2H4, the Spike trimer passes through a 2 series of conformations that eventually permit S1 shedding and the S2 post-fusion 3 transition that mediates membrane fusion. Alternatively, allosteric antibodies such as 4 3D11 can advance the trimer directly to the end of the opening process, potentiating 5 formation of syncytia through fusion of neighboring cells. Allosteric opening most 6 likely contributes to lower potency in a high-affinity receptor-blocking antibody, and 7 might even suggest the possibility antibody-dependent enhancement of infection. 8

Finally, the Spike might instead be recognized by 5A6, which inhibits membrane 9 fusion and syncytia formation by preventing S1 shedding and trapping the pre-fusion 10 trimer. By enjoining the exposure and conformational transition of the S2 subunit, the 11 5A6 complex represents an unproductive dead-end for the Spike trimer. See also 12 Figure S6 . BLI measurements are similar, however 3D11 IgG and ACE2-Fc bind significantly 1 more weakly to relatively unrestricted Fc-RBD than to Spike trimer (note that the 2 3D11 IgG binds >10x more tightly than 3D11 Fab in both sets of experiments, 3 indicating avid binding). 5A6 IgG binds somewhat more tightly to Spike trimer than to 4

Fc-RBD, perhaps indicating that RBDs within a Spike trimer have particularly 5 favorable geometries for binding. is more open than that bound to ACE2, yet the neighboring RBD trapped closed by 4 5A6 is more closed than in the singular ACE2 complex or the fully closed Spike. 

Further information and requests for resources and reagents should be 5 directed to and will be fulfilled by the Lead Contact, Dr. Cheng-I Wang 6 (Wang_ChengI@immunol.a-star.edu.sg). 7

All requests for resources and reagents should be directed to and will be 9 fulfilled by the Lead Contact. All antibodies are proprietary and can be obtained 10 through a Materials Transfer Agreement. Other materials will also be available from 11

the Lead Contact with a completed Materials Transfer Agreement. 12

Atomic coordinates and cryo-EM maps are deposited in EMDB and PDB as 14 follows. Spike:5A6 complex I has accession codes PDB: 7KQB and EMD-22993. 15

The focused refinement of Spike:5A6 using a mask including 5A6 and two RBDs has The Singapore strain of SARS-CoV-2 live virus used in this study was isolated 2 from a nasopharyngeal swab of a patient in Singapore (Young et al., 2020) . The 3

French strain of SARS-CoV-2 live virus used in this study was isolated from one of 4 the first COVID-19 cases confirmed in France: a 47-year old female patient 5 hospitalized in January 2020 in the Department of Infectious and Tropical Diseases, 6

Bichat Claude Bernard Hospital, Paris (Lescure et al., 2020) . The complete viral 7 genome sequence was obtained using Illumina MiSeq sequencing technology, was 8 then deposited after assembly on the GISAID EpiCoV platform (Accession ID 9

EPI_ISL_411218) under the name BetaCoV/France/IDF0571/2020. 10

Antibody discovery from phage display library 12

Anti-SARS-CoV-2 Spike RBD antibodies were isolated from an HX02 human 13 then purified from the culture supernatant using Protein G Agarose (Merck Millipore, 10

Cat#16-266) following the manufacturer's instructions. After elution, the purified 11 antibodies were dialyzed at 4°C for 4-20 hours against 1x PBS, for 3 times and 12 concentrated to 1-2 mg/ml using 10MWCO Vivaspin 20 (Sartorius, Cat#VS2001). 13

The tag-less Fab fragments were produced using the ExpiCHO transient with 1 M Sodium Chloride) with a sequential linear gradient of 0% to 5% in 5 min, 5% 23 to 15% in 30 min, and 15% to 100% in 20 min of Buffer B injection at a flow rate of 1 24 ml/min. The resulting purified Fab fragments were dialyzed at 4°C for 4-20 hours 1 against 4 liters of 20 mM Histidine, 150 mM NaCl, pH 6.6, for 3 times and 2 concentrated to 1-2 mg/ml using 10MWCO Vivaspin 6 (Sartorius, Cat#VS0601). 3

Anti-SARS-CoV-2 Spike RBD IgG antibodies were tested in an ELISA against 5 biotinylated recombinant SARS-CoV-2 Spike protein RBD-mFc (Sino Biological, 6

Cat#40592-V05H) to assess binding avidity for the target. In brief, NeutrAvidin 7 protein (Thermo Fisher Scientific, Cat#31000) was coated at 5 µg/ml onto 96-well 8 ELISA plates in coating buffer (8.4 g/L NaHCO 3 , 3.56 g/L Na 2 CO 3 , pH 9.5) overnight 9 at 4°C. After blocking with 1% Casein (Thermo Fisher Scientific, Cat#A37528) for 10 two hours, biotinylated antigen at 0.2 µg/ml was added to the plates and captured by 11

NeutrAvidin during one-hour incubation at room temperature. After washing with 12 0.05% PBST for 5 times, the IgG antibodies were added at different concentrations 13 

Anti-SARS-CoV-2 Spike RBD IgG or Fab antibodies were tested in an ELISA 21 against iodixanol-gradient-purified SARS-CoV-2 pseudoviruses with an isotype IgG 22

used as a negative control antibody. In brief, 1 µg/ml of pseudoviral particles were 23 coated in coating buffer onto 96-well ELISA plates overnight at 4°C. After blocking 24 with 1% Casein (Thermo Fisher Scientific, #A37528) for two hours, serially diluted 1 IgG or Fab antibodies starting from 20 nM (IgG) or 300 nM (Fab) with five-fold 2 dilutions were added to the plates and incubated for an hour at room temperature. 3

The wells were then washed again with 0.05% PBST, followed by addition of HRP 4 conjugated anti-human Fc antibody (JACKSON ImmunoResearch, Cat#109-036-5 098) or HRP conjugated anti-human Fab antibody (JACKSON ImmunoResearch, 6

Cat#109-036-097) for one-hour incubation before HRP activity was measured at 450 7 nm with addition of TMB substrate (Surmodics, BioFX®, Cat#TMBW-1000-01). 8

Anti-SARS-CoV-2 Spike RBD IgG antibodies were tested in a competition 10 ELISA to assess their ability to block the Spike protein RBD from binding to human 11 Spike RBD IgG antibodies were pre-incubated with 0.5 nM biotinylated Spike protein 15 RBD-mFc (Sino Biological, Cat#40592-V05H) for one hour at room temperature 16 before they were added to the ELISA plates coated with ACE2-Fc. After one-hour 17 incubation, the wells were washed with 0.05% PBST for five times and HRP 18 conjugated streptavidin (Biolegend, Cat#405210) was added at a dilution of 1:3000, 19

and incubated for another one hour before HRP activity was measured at 450 nm 20 with addition of TMB substrate (Surmodics, BioFX®, Cat#TMBW-1000-01). 21

Digestion reaction for each IgG was prepared using immobilized 23

FabALACTICA microspin columns (Genovis). 100 µL of IgG at 5 mg/mL 24 concentration in digestion buffer (150 mM sodium phosphate, pH 7.0) were 1 incubated overnight on each column. Digested sample was further purified using a 2 HiTrap Protein L column followed by size-exclusion chromatography (Superdex 75 3 10/300 GL) using an Äkta Pure FPLC (GE Healthcare). 4

The expression plasmid containing the prefusion S ectodomain as used in super resolution pixel size) at the sample. A dose rate of 8 e -/(pix · sec), or 11.5 e -3 /(Å 2 · sec), and a frame rate of 0.05 sec/frame was used with a total exposure time of 4 5.9 sec, for a total dose of 67.7 e -/Å 2 . Automated data collection was performed 5

using SerialEM (Mastronarde, 2005) . 6

Dose-weighted, motion-corrected sums down-sampled to the physical pixel 8 size were obtained from the super-resolution DED movies using UCSF Motioncor2 9 (Zheng et al., 2017) . For the Spike trimer, CTF estimation was performed in 10 cryoSPARC (Punjani et al., 2017) followed by blob-based particle picking, 2D 11 classification, ab initio modelling, 3D classification, and 3D refinement. For images of 12 antibody complexes, particles were instead picked using templates generated from 13 the apo trimer structure, and the apo trimer was likewise used as an initial model in 14 3D classification. The resolution of the interface between the Spike RBD and the 5A6 15 Fab was further improved using naïve focused refinements. Additional 3D 16 classification of the Spike:3D11 complex was performed in Relion 3.1 (Scheres, 17 2012) . Processing details are given in Table S3 and Figure S5 . RBDs and three copies of ACE2 bound (PDB: 7a98) was used with 3D11, and 1

Spike with two open RBDs and one copy of ACE2 bound (PDB: 7a95) was used with 2 2H4. Missing segments and side chains in the RBDs were built using Coot. 3

Interactive, density-restrained molecular dynamics simulations in ChimeraX 4 (Goddard et al., 2018) and ISOLDE (Croll, 2018) were used to finalize the models, 5

and atomic b-factors were calculated using PHENIX (Afonine et al., 2018) . Models 6 for the Spike:5A6 and Spike:3D11 complexes were first built into density maps from 7 whole-particle cryo-EM reconstructions, and then further refined using maps from 8 focused refinements of the Fab and Spike RBD. Maps of Spike:2H4 complexes were 9

of lower resolution and model building was terminated after the docking step 10 described above. Model statistics and density fit information are presented in Table  11 S3 and Figure S5 . harvested, centrifuged at 700 g for 10min to remove cell debris and filtered through a 22 0.45 µm filter unit (Sartorius, Cat#16555). Lenti-X p24 rapid titer kit (Takara Bio, 23

Cat#632200) was used to quantify the viral titres following the manufacturer 24 instructions. pTT5LnX-CoV-SP plasmid with D614G mutation was generated using 1

QuickChange Lightning Multi Site-Directed Mutagenesis Kit (Agilent, Cat#210513) 2

and was used to generate mutant pseudovirus expressing SARS-CoV-2 Spike 3 protein carrying D614G mutation. 4

To concentrate and purify the pseudovirus particles expressing the SARS-6

CoV-2 Spike glycoproteins, pre-cleared 40 mL viral supernatant was concentrated by 7 20% sucrose gradient centrifugation at 10,000 g for 4 hours at 4°C in an SW41 Ti 8 rotor with no brake. Upon removal of supernatant, 1mL of PBS was added to the 9 virus pellet and left at 4°C overnight. Concentrated virus was further purified by an 10

OptiPrep (60% [wt/vol] iodixanol, STEMCELL Technologies, Cat#07820) velocity 11 gradient. Iodixanol gradients were prepared in PBS in 1.2% increments ranging from Luminometer. The relative luciferase units (RLU) were converted to percent 5 neutralization and plotted with a non-linear regression curve fit using GraphPad 6 PRISM. 7

The potency of the IgG or Fab antibodies were determined in neutralizing live 9 SARS-CoV-2 virus assays. In brief, 25 µl of 100 TCID50 of SARS-CoV-2 live virus 10 (isolated from a nasopharyngeal swab of a patient in Singapore) was mixed with an 11 equal volume of serially diluted IgG or Fab antibodies and incubated at 37°C for one 12 hour before the mixture was added to 50 µl of Vero E6 C1008 cells in suspension. 13

The infected cells were incubated at 37°C incubator for four days and the cell 14 viability was determined using Viral ToxGloTM Assay (Promega, Cat#G8941). The 15 potency of 2H4, 3D11 and 5A6 IgG antibodies in neutralizing live SARS-CoV-2 virus 16 assays was also determined by measuring the viral genome copy number (GCN). 25 17 µl of 100 TCID50 of SARS-CoV-2 live virus (isolated from a nasopharyngeal swab of 18 a patient in Singapore) was mixed with an equal volume of serially diluted 2H4, 3D11 19 or 5A6 IgG antibodies and incubated at 37°C for one hour before the mixture was 20 added to 50 µl of 4x10 5 

The potency of 5A6 IgG was tested in neutralizing a live virus strain 7 Volt/Ohm Meter for TEER) and expressed as Ohm/cm2. 7

Vero E6 cells were transfected with S protein bearing furin recognition 9 mutation (R682RAR to A682AAR) with C-terminal GFP tag by Lipofectamin 2000 10 (Invitrogen) and were cultured on µ-Slide 8 well chamber slides (Ibidi, Cat#80826). 11

The transfection efficiency was monitored by percentage of GFP positive cells and 12 optimized within 15-30% to achieve the best signal-to-noise ratio in the following cell-13 cell fusion assay. After 48 hours, cells were treated with various antibodies diluted in 14 DMEM without FBS for 1 hour at 37°C. Cells were then treated with 15 µg/ml trypsin 15 and incubated at 37°C for another 2 hours. After trypsin treatment, cells were fixed 16 with 4% PFA at room temperature for 15 mins and the cell nuclei were stained with 17 DAPI. Images were taken by Olympus confocal microscope. 18

Binding affinity of purified Fab to RBD was measured on the Octet96Red 20 system (ForteBio). Anti-human IgG Fc (AHC) sensors were first loaded with 1 µg/ml 21 of Fc-RBD for 10 min, followed by kinetics buffer (phosphate-buffered saline buffer 22

supplemented with 0.1% Tween-20 and 0.1% BSA) for 5 min to establish a stable 23 baseline. The sensors were then dipped into different concentrations of each Fab 24 from 100 nM to 3.125 nM in two-fold dilutions for 6 min, and then in kinetics buffer 1 again for 10 min to measure association and dissociation. Assays were run at 25°C 2 and data was analysed on the Octet System Data Acquisition Software version 3 9.0.0.4. using the 1:1 Langmuir binding model. 4

Avidity of anti-SARS-CoV-2 Spike RBD IgG antibodies for RBD was 6 measured on the Octet96Red system. Anti-hIgG Fc capture (AHC) sensors were 7 used. The sensors were loaded with 1 µg/ml of Fc-RBD (made in-house) in assay 8 buffer (phosphate-buffered saline buffer supplemented with 0.1% Tween-20 and 9 0.1% BSA) for 10 min, quenched in 0.5 mg/ml of isotype IgG in assay buffer for 10 10 min, then dipped in assay buffer for 12 min for the system to stabilize. To measure 11 the association of 5A6, the sensors were dipped in a range of 5A6 IgG hydroxysuccinimide) solution and the 5A6 antibody was immobilized to the sensor 21 tips using a concentration of 7.5 µg/ml of 5A6 in 10 mM sodium acetate pH 6 buffer. 22

After quenching in 1M ethanolamine, the 5A6-immobilized sensor tips were dipped in 23 (ANOVA) was used to compare differences between groups. Differences were 1 considered statistically significant at confidence levels *P < 0.05 or **P < 0.01, 2 ***P < 0.001. First, the density map for Spike:5A6 complex I is displayed using a low iso-surface 3 threshold, highlighting the binding geometry. The Spike is shown with RBDs in coral, 4

NTDs in plum, the S2 core in rose brown, and 5A6 Fabs is shown in goldenrod. 5

Next, the map is shown in transparency overlaid on a colored copy of the map with a 6 higher density threshold, revealing high resolution details. The high threshold map is 7 replaced by a ribbon diagram of the complex, employing the same color scheme. 8

Finally, the ribbon diagram is shown embedded in the low resolution Spike:5A6 IgG 9 density. Post-fusion state PDB: 7a92 S1 shedding J o u r n a l P r e -p r o o f

Real-space refinement in PHENIX for cryo-EM and 3 crystallography

SARS-CoV-2 neutralizing antibody structures inform therapeutic strategies

Receptor binding and priming of the spike 10 protein of SARS-CoV-2 for membrane fusion

Potent neutralizing antibodies from COVID-19 patients define multiple targets of 14 vulnerability

A model for neutralization of 16 viruses based on antibody coating of the virion surface

Persistence of viral RNA, 20 pneumocyte syncytia and thrombosis are hallmarks of advanced COVID-19 21 pathology

Potent Neutralizing Antibodies against SARS-CoV-2 24 Identified by High-Throughput Single-Cell Sequencing of Convalescent Patients' B 25 Cells

A neutralizing human antibody binds to the N-terminal 28 domain of the Spike protein of SARS-CoV-2

ISOLDE: a physically realistic environment for model building into 30 low-resolution electron-density maps

Antibody-mediated neutralization of 33 flaviviruses: a reductionist view

Mechanisms of 35 Rabies Virus Neutralization

Designing Human Antibodies by Phage Display

UCSF ChimeraX: Meeting modern challenges in visualization 5 and analysis

A novel human anti-interleukin-1β neutralizing monoclonal antibody 8 showing in vivo efficacy

Studies in humanized mice and 11 convalescent humans yield a SARS-CoV-2 antibody cocktail

SARS-CoV-2 Cell 15 Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven 16

Syncytial Virus-Neutralizing Monoclonal Antibodies Motavizumab and Palivizumab 19 Inhibit Fusion

CoV-2 by Destruction of the Prefusion Spike

Human neutralizing antibodies elicited by SARS-CoV-2 infection. 25

Prognostic 28 Value of HIV-1 Syncytium-Inducing Phenotype for Rate of CD4+ Cell Depletion and 29 Progression to AIDS

Tracking 32 Changes in SARS-CoV-2 Spike: Evidence that D614G Increases Infectivity of the 33 COVID-19 Virus

Structure of the SARS-CoV-2 spike receptor-binding domain 36 bound to the ACE2 receptor

IMGT unique numbering for 39 immunoglobulin and T cell receptor variable domains and Ig superfamily V-like 1 domains

Clinical and virological data of the first cases of COVID-19 in Europe: a case series

Cross-Neutralization of a 8 SARS-CoV-2 Antibody to a Functionally Conserved Site Is Mediated by Avidity

Cross-neutralization antibodies against SARS-CoV-2 and RBD 12 mutations from convalescent patient antibody libraries

Development of therapeutic antibodies for the treatment of diseases

Automated electron microscope tomography using robust 17 prediction of specimen movements

The pathogenesis of respiratory syncytial 19 virus disease in childhood

The anti-HIV drug nelfinavir mesylate (Viracept) is a potent 22 inhibitor of cell fusion caused by the SARSCoV-2 spike (S) glycoprotein warranting 23 further evaluation as an antiviral against COVID-19 infections

Substitution at Aspartic Acid 1128 in the SARS Coronavirus Spike 27

Glycoprotein Mediates Escape from a S2 Domain-Targeting Neutralizing Monoclonal 28 Antibody

UCSF Chimera--a visualization system for exploratory 31 research and analysis

CoV-2 by a human monoclonal SARS-CoV antibody

Characterization and Treatment of SARS-CoV-2 in Nasal and Bronchial Human 38 Airway Epithelia

3D variability analysis: Resolving continuous 1 flexibility and discrete heterogeneity from single particle cryo-EM

cryoSPARC: 4 algorithms for rapid unsupervised cryo-EM structure determination

Isolation of potent SARS-CoV-2 neutralizing 8 antibodies and protection from disease in a small animal model

Systematic 12 analysis of monoclonal antibodies against Ebola virus GP defines features that 13 contribute to protection

A Bayesian View on Cryo-EM Structure Determination

SARS-CoV-2 direct 18 cardiac damage through spike-mediated cardiomyocyte fusion

A human neutralizing antibody targets the receptor-binding site of 22 SARS-CoV-2

Respiratory Syndrome Coronavirus Receptor and Activates Virus Entry

Antibody Elbow Angles 28 are Influenced by their Light Chain

HIV-induced T cell 30 syncytia are self-perpetuating and the primary cause of T cell death in culture

Characterization of the receptor-binding domain (RBD) of 2019 novel coronavirus: 34 implication for development of RBD protein as a viral attachment inhibitor and 35 vaccine

Chapter Four -Structural insights into 37 coronavirus entry

Human antibodies by 1 design

Exploring avidity: understanding the 3 potential gains in functional affinity and target residence time of bivalent and 4 heterobivalent ligands

Unexpected Receptor 7 Functional Mimicry Elucidates Activation of Coronavirus Fusion

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

Human-IgG-Neutralizing Monoclonal Antibodies Block the 14 SARS-CoV-2 Infection

A 17 human monoclonal antibody blocking SARS-CoV-2 infection

Comparative Protein Structure Modeling Using 20 MODELLER

Antibody-antigen interactions: new structures 22 and new conformational changes

Structural 25 Basis for Potent Neutralization of Betacoronaviruses by Single-Domain Camelid 26 Antibodies

Cryo-EM structure of the 2019-nCoV 29 spike in the prefusion conformation

A noncompeting pair of human neutralizing antibodies block COVID-19 32 virus binding to its receptor ACE2

Pathological findings of COVID-19 associated with acute 35 respiratory distress syndrome

Epidemiologic Features and

Clinical Course of Patients Infected With SARS-CoV-2 in Singapore

A highly conserved cryptic epitope in the receptor binding 4 domains of SARS-CoV-2 and SARS-CoV

Structural and 7 Functional Analysis of the D614G SARS-CoV-2 Spike Protein Variant

SARS-CoV-2 spike-11 protein D614G mutation increases virion spike density and infectivity

MotionCor2: anisotropic correction of beam-induced motion for improved 15 cryo-electron microscopy

Perspectives on therapeutic neutralizing antibodies 17 against the Novel Coronavirus SARS-CoV-2

A pneumonia outbreak associated with a new 20 coronavirus of probable bat origin

The assay was run at 25°C. Sensor tips were regenerated in 10 mM glycine at pH 1 2.7 and neutralized in PBS with 0.1% Tween-20 before another cycle of sandwich 2 assay was performed. Each sensor tip was used in a total of 3 cycles. Data analysis 3 was done in the Octet System Data Acquisition Software version 9.0.0.4. 4

StreptagII-tagged prefusion S ectodomain, diluted to 10 µg/mL in 10 mM (https://www.aatbio.com/tools/ic50-calculator). One-way analysis of variance 24