key: cord-0960324-miz0jhkc authors: Rapp, Micah; Guo, Yicheng; Reddem, Eswar R.; Yu, Jian; Liu, Lihong; Wang, Pengfei; Cerutti, Gabriele; Katsamba, Phinikoula; Bimela, Jude S.; Bahna, Fabiana A.; Mannepalli, Seetha M.; Zhang, Baoshan; Kwong, Peter D.; Huang, Yaoxing; Ho, David D.; Shapiro, Lawrence; Sheng, Zizhang title: Modular basis for potent SARS-CoV-2 neutralization by a prevalent VH1-2-derived antibody class date: 2021-03-19 journal: Cell Rep DOI: 10.1016/j.celrep.2021.108950 sha: d32fd07534353b40946e281b7e8851c5f33547cf doc_id: 960324 cord_uid: miz0jhkc Antibodies with heavy chains that derive from the VH1-2 gene constitute some of the most potent SARS-CoV-2-neutralizing antibodies yet identified. To provide insight into whether these genetic similarities inform common modes of recognition, we determined structures of the SARS-CoV-2 spike in complex with three VH1-2-derived antibodies: 2-15, 2-43, and H4. All three utilize VH1-2-encoded motifs to recognize the receptor-binding domain (RBD), with heavy chain N53I enhancing binding and light chain tyrosines recognizing F486RBD. Despite these similarities, class members bind both RBD-up and -down conformations of the spike, with a subset of antibodies utilizing elongated CDRH3s to recognize glycan N343 on a neighboring RBD – a quaternary interaction accommodated by an increase in RBD separation of up to 12 Å. The VH1-2-antibody class thus utilizes modular recognition encoded by modular genetic elements to effect potent neutralization, with VH-gene component specifying recognition of RBD and CDRH3 component specifying quaternary interactions. Studies on human antibody responses to viral pathogens including HIV-1, influenza, EBOLA, and malaria have revealed prominent classes of similar neutralizing antibodies (nAbs), which arise commonly in numerous individuals in response to infection or vaccination (Ehrhardt et al., 2019; Imkeller et al., 2018; Joyce et al., 2016; Kallewaard et al., 2016; Pappas et al., 2014; Zhou et al., 2015; Zhou et al., 2013) . Such multi-donor antibody classes are defined based on similar V(D)J gene recombination and similar mode of structural recognition -with this combination indicative of a common evolutionary process in antibody development (Kwong and Mascola, 2012) . Multi-donor antibody classes are thought to arise based on effective function, combined with genetic accessibility due to class requirements for V(D)J recombination and somatic hypermutations of sufficient frequency as to be present in the antibody human repertoire. Some multi-donor antibody classes are potent, broadly neutralizing, and frequent in human antibody repertoire (Imkeller et al., 2018; Joyce et al., 2016; Zhou et al., 2013) . A prominent mode of rational vaccine design -"lineage-based vaccine design" -endeavors to elicit such antibody classes by vaccination (Haynes et al., 2012; Jardine et al., 2013; Kwong and Mascola, 2018) and this approach to human vaccination has recently entered clinical assessment (Diemert and McElrath, 2018) . Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), the causative agent of the ongoing Coronavirus disease 2019 (COVID-19) pandemic, has infected over 80 million people and has claimed over one million deaths since the outbreak began in late 2019 (Dong et al., 2020; Zhou et al., 2020; Zhu et al., 2020) . Therapies and vaccines are urgently needed to end the pandemic. Many nAbs have now been isolated from COVID-19 convalescent donors with the most potent nAbs showing promise as prophylactic or therapeutic agents (Brouwer et al., 2020; J o u r n a l P r e -p r o o f 4 al., 2020; Kreer et al., 2020; Liu et al., 2020; Robbiani et al., 2020; Rogers et al., 2020; Tortorici et al., 2020; Zost et al., 2020) . This growing set of nAbs provides an opportunity to identify effective human antibody responses to SARS-CoV-2 common in the population, which will inform therapeutic strategies and help to interpret vaccine readouts. SARS-CoV-2 nAbs predominantly target the viral spike glycoprotein, which interacts with angiotensin-converting enzyme 2 (ACE2) receptors on the host-cell surface to mediate virus entry (Hoffmann et al., 2020; Wang et al., 2020) . The ectodomain of the prefusion spike comprises three copies of both S1 and S2 subunits (Wrapp et al., 2020) . The S1 subunit is responsible for ACE2 binding, and the S2 subunit mediates fusion with host cell membrane. Each S1 subunit comprises an N-terminal domain (NTD) and a receptor binding domain (RBD). The RBDs are very flexible, adopting either an 'up' conformation (open state) or a 'down' conformation (closed state), with only the 'up' RBDs capable of interacting with ACE2 . Currently, many nAbs have been characterized which bind to epitopes on either the 'up' and/or 'down' RBDs (Barnes et al., 2020a; Liu et al., 2020; Tortorici et al., 2020) . These RBD-targeting nAbs neutralize SARS-CoV-2 through mechanisms that include competition with ACE2 for RBD binding and locking the RBDs in the 'down' conformation. Human SARS-CoV-2 nAbs develop with few somatic hypermutations and strong avidity effects (Barnes et al., 2020b; Kreer et al., 2020; Liu et al., 2020; Robbiani et al., 2020) . Characterization of genetic features of SARS-CoV-2 nAbs identified to date show enrichment of antibody variable genes including VH3-53, VH1-2, VH1-69, VH3-66, VH1-58, and VH3-30 Robbiani et al., 2020; Yuan et al., 2020) . So far, structural characterization of multiple nAbs have revealed two separate RBD-targeted classes derived from the similar VH3-53 and VH3-66 genes. One of these is characterized by heavy chain complementarity-determining region 3 J o u r n a l P r e -p r o o f 5 (CDRH3) of 15 amino acids or shorter -and recognizes RBD ridge in the 'up' position on SARS-CoV-2 spike (type I) (Yuan et al., 2020) . The second VH3-53/-66 class -with longer CDRH3srecognizes a similar region of RBD, but adopts an approach angle with the heavy and light chain orientation rotated 180 degrees (type II), suggesting CDRH3 to be critical for the classification of VH3-53/-66 antibodies (Barnes et al., 2020a; Wu et al., 2020a) . Different VH3-30 originated antibodies can recognize at least three different regions of RBDs (Barnes et al., 2020a; Hansen et al., 2020) , demonstrating that they do not form a single gene-restricted antibody class. Nevertheless, whether nAbs derived from VH1-2 and other germline genes form gene-restricted classes that represent shared effective antibody responses remains unaddressed. Currently, three VH1-2 potent nAbs (2-4, S2M11, and C121) show similar modes of RBD recognition but differences in quaternary epitope recognition (Barnes et al., 2020a; Liu et al., 2020; Tortorici et al., 2020) . It is thus still unclear whether RBD-targeting VH1-2 antibodies form a gene-restricted class. If so, what are the key genetic and structural signatures and determinants of neutralization potency? Here we present structures of three VH1-2-derived nAbs (2-15, 2-43, and H4), revealing that they recognize a SARS-CoV-2 RBD epitope with similar modes of RBD recognition and similar angles of approach. Overall, recognition is modular with RBD predominantly recognized by a VH1-2 gene encoded module. The second recognition module is represented by the diverse CDRH3s of the VH1-2 antibodies, which mediate quaternary recognition of N343 glycan from an adjacent RBD for a subset of class members. Thus, we define a multi-donor VH1-2 antibody class, members of which can achieve very high neutralization potency, which is prevalent in human responses to SARS-CoV-2. The shared genetic and structural signatures inform strategies to improve members of the VH1-2 antibody class. J o u r n a l P r e -p r o o f 6 To identify common features of the human antibody response to SARS-CoV-2, 158 spikespecific antibodies with characterized neutralization potencies were collected from 10 studies (Table S1 ) (Brouwer et al., 2020; Hansen et al., 2020; Ju et al., 2020; Kreer et al., 2020; Liu et al., 2020; Robbiani et al., 2020; Rogers et al., 2020; Tortorici et al., 2020; Wu et al., 2020b; Zost et al., 2020) . V(D)J gene usage analysis showed that VH1-2 was the second most frequently utilized germline gene ( Figure 1A , 25 in total). Comparison of neutralization potencies revealed that 11 of 25 (44%) of the VH1-2 antibodies are potent (half maximal inhibitory concentration (IC 50 ) <0.1µg/ml, Figure 1B ). Within four studies Liu et al., 2020; Tortorici et al., 2020; Zost et al., 2020) , VH1-2 antibodies ranked the most potent of gene-delimited sets of antibodies (IC 50 ranges from 0.015 to 0.0007µg/ml). All 25 of the VH1-2-derived antibodies have been reported to bind to the SARS-CoV-2 RBD (Brouwer et al., 2020; Hansen et al., 2020; Ju et al., 2020; Kreer et al., 2020; Liu et al., 2020; Robbiani et al., 2020; Rogers et al., 2020; Tortorici et al., 2020; Wu et al., 2020b; Zost et al., 2020) . Sequence alignment of the VH1-2 nAbs showed the heavy chains to carry few somatic hypermutations, with each having a unique CDRH3 with length varying from 11 to 21 amino acids ( Figure 1C and S1A, Kabat definition), and using different D genes ( Figure S1B ). The VH1-2 antibodies used both kappa and lambda genes with enrichment of the IGLV2-14 gene ( Figure 1C ). As described below, we determined structures of three nAbs: 2-15, 2-43, and H4, with the highly potent antibodies 2-15 and 2-43 isolated from donor '2' of our previous study while H4 was from a different donor (Wu et al., 2020b) . 2-15, 2-43, and H4 neutralize J o u r n a l P r e -p r o o f 7 authentic SARS-CoV-2 "live" virus with IC 50 of 0.0007, 0.003, and 0.896 µg/ml, respectively. These three antibodies derived from different heavy and light chain gene recombinations, and hence three different B cell lineages ( Figure S1B ). Both 2-43 and 2-15 utilize the DH2-15*01 gene and have a long CDRH3 of 20 amino acids ( Figure S1C ), but they have different HJ gene origins (JH6*03 and JH3*02 respectively). The light chains of 2-43 and 2-15 were derived from recombination of a novel allele of the VL2-14 gene with JL3*02 and JL1*01 respectively ( Figure S1D ). H4 used DH2-2*01 and JH2*01 genes and had a CDRH3 of 17 amino acids (Wu et al., 2020b) . The H4 light chain was derived from VK2-40*01 and JK4*01. To understand SARS-CoV-2 spike recognition by the VH1-2-derived antibodies, we used single-particle cryo-electron microscopy (cryo-EM) to produce high-resolution 3D-reconstructions of antigen-binding fragments (Fabs) from 2-15, 2-43, and H4 in complex with the SARS-CoV-2 spike (Table S2 ). The reconstruction of the 2-43 complex with spike, refined to a global resolution of 3.60Å ( Figure S2A -S2E), is significantly improved than in our previous study (5.8Å resolution) . The reconstruction revealed a predominant class with 3 Fab molecules bound per spike trimer ( Figure 2A ). Each 2-43 Fab used heavy chain variable domain to bind one primary RBD with an orientation similar to the previously published antibody 2-4 , with all RBDs in the 'down' conformation ( Figure 2A ). 2-43 heavy and light chains also recognize a quaternary epitope from the adjacent RBD. 3D classification revealed less populated states with 1 and 2 Fabs bound, but in every case the Fab was bound to a 'down' RBD. This dependence upon the 'down' conformation is likely due to extensive interactions between the Fab CDRH3 loop and the N-linked glycosylation at residue N343 on the adjacent RBD ( Figure S2E ). This suggests a J o u r n a l P r e -p r o o f 8 neutralization mechanism whereby the Fab simultaneously occludes ACE2 binding and locks the RBDs in the 'down' conformation. For the complex of 2-15 with spike ( Figure 2B ), approximately 56% of particles were bound to an RBD in the 'up' conformation, and 44% bound to RBD 'down' ( Figure S2F -S2I), differing from antibody 2-43, which bound only to "down" RBDs. Due to increased conformational heterogeneity of the RBD 'up' class, the RBD 'down' class was the focus of our structural analysis ( Figure 2B ). In our initial attempt, a 9-fold molar excess of Fab was incubated with the S trimer. However, this resulted in spike disassembly ( Figure S2J ). To overcome this issue for structure determination, we found that a 1:1 molar ratio left spike-complexes intact, though some spike disassembly was still observed ( Figure S2K ). Although we were unable to resolve the 2-15:spike interaction at atomic resolution, the cryo-EM analysis did reveal the overall orientation of the Fab along with the position of the peptide backbone for several of the CDR loops. To understand the binding mode of 2-15 at atomic resolution, we determined the crystal structure of 2-15 in complex with isolated RBD. The structure was determined by molecular replacement at 3.18 Å resolution to a final crystallographic R work /R free of 18.6%/23.8% with good overall geometry ( Table S3) . The electron density for RBD (corresponding to S1-subunit residues 333-527) was well defined for residues 344-518, with missing internal stretches of four (362) (363) (364) (365) (366) and seven (383-390) residues, with the C-terminal eighteen residues (519-537) disordered ( Figure 2C ). For Fab 2-15 all residues were visualized in density, with the exception of heavy chain residues 142-145 in the Fc region. We then docked the crystal structure to the 2-15-spike complex cryo-EM density map by superposing RBD from the crystal structure on RBD from the cryo-EM map, with subsequent rigid body fitting to density. J o u r n a l P r e -p r o o f 9 Overall, the conformation of 2-15 in the crystal structure agrees with the density observed in the cryo-EM map of the complex with spike -particularly in the V-gene-encoded CDR regions. However, the conformation of the long CDRH3 loop differs, and is far more elongated in the crystal structure ( Figure S2L ). In the cryo-EM map, it appears that the CDRH3 loop is pushed away from the N343 glycan and the helix encompassing residues 364-371 from the adjacent 'down' RBD. The two light chains exhibit more significant conformational differences. Save for CDRL3, which bends slightly away from RBD in the crystal structure, the conformations of the loops are quite similar, but the chain is shifted down closer to the RBD. Finally, a reconstruction of Fab H4 revealed spike complexes with a single Fab bound to RBD in the 'up' conformation ( Figure 2D and S2M to S2P). Interestingly, no Fab was seen bound to a 'down' RBD. The conformational flexibility of the RBD while in the 'up' conformation made a high-resolution cryo-EM reconstruction unattainable, and the map was refined to an overall resolution of 5.08Å. Similar to Fab 2-15, the peptide backbone for many CDR loops are observed, and a homology model of the Fab variable domain was docked into the density. The superimposition of modeled H4 to the 2-43/spike complex showed that H4 adopts an RBD approaching angle similar to 2-43 ( Figure S2Q ). However, the light chain of H4 rotates toward the interface between RBDs such that the long CDRL1 of H4 clashes with N343 glycan from an adjacent 'down' RBD ( Figure S2R ), which suggests a possible explanation for the apparent lack of H4 binding to 'down' RBDs. To understand the similarity in RBD recognition, we characterized the epitope and paratope interactions of 2-43 and 2-15 and compared to the three published VH1-2 antibodies 2-4, S2M11, J o u r n a l P r e -p r o o f 10 and C121 (Barnes et al., 2020a; Liu et al., 2020; Tortorici et al., 2020) . Overall, 2-43 interacts predominantly with the receptor binding motif (RBM, residues 438-508) on one 'down' RBD protomer while simultaneously binding the N343 glycan from an adjacent 'down' RBD protomer ( Figure 3A ). We describe the N343 glycan interaction below in the last section focused on quaternary interactions. The interaction between 2-43 and the 'primary' RBD bound through its RBM buries a total of 756Å 2 paratope surface area (BSA), 83% of which is contributed by heavy chain ( Figure 3A and S3A). Heavy chain framework 1 (FWH1), CDRH1, CDRH2, and CDRL3 of 2-43 form hydrogen bond networks with three RBD regions. In the first region, residues in FWH1 and the DE-loop in FWH3 form a groove to hold Y449 RBD . Hydrogen bonds are observed between FWH1 residues and Y449 RBD and Q498 RBD ( Figure 3A , second panel). In the second region, involving the 'flat' region of RBM, T30 HC forms a hydrogen bond with S494 RBD . In the third region or the RBD ridge region, Y33 HC , N52 HC , and S54 HC form a hydrogen bond network with E484 RBD . Light chain residues Y91 LC and S95 LC further hydrogen bond with G485 RBD and F486 RBD . In addition, Y30 LC , Y32 LC , Y91 LC , and Y100j HC also form a hydrophobic pocket to hold F486 RBD ( Figure S3B ), with π-π stacking observed between Y30 LC and F486 RBD (Figure 3A showed similar RBM approach angle and binding mode ( Figure 3C , 3D, S3C, and S3D). Typically, interactions between the VH1-2 antibodies and the RBD are mediated predominantly by heavy J o u r n a l P r e -p r o o f 11 chain (RBD A column in Figure S3C ). For all antibodies, the VH1-2 gene encoded residues form a module for RBM recognition. Despite these antibodies having different light chain gene origins, convergent tyrosine residues in CDRL1 and CDRL3 pack against F486 RBD ( Figure S3B ), which constitutes another module for RBD recognition. Because each antibody has a unique CDRH3, no conserved polar or hydrophobic interaction is observed (Figure 3 and S3B-S3D), albeit CDRH3 contributes the most BSA among the CDRs except in 2-15 ( Figure S3A ). Altogether, the germline gene residues from the VH1-2 gene as well as light chain V genes anchor the antibodies to the RBM in a similar mode with heavy chain V region playing a dominant role in determining the mode of recognition. To gain an overall understanding of similarity in binding orientations of the VH1-2 antibodies, we superposed RBD and antibody complexes and calculated pairwise root mean square deviation (RMSD) to compare relative binding orientations of the six VH1-2 antibodies (2-43, 2-15, H4, 2-4, C121, and S2M11) as well as 31 additional RBD-targeting antibodies originated from other VH genes. Overall, the VH1-2 antibodies have similar binding orientations (Figures 4A, 4B, and S4A). Their epitopes overlap with ACE2 binding site ( Figure S4B and S4C). The similarity in genetic origin, details of their interactions with RBD (unavailable for H4), and angle of approach, suggests that these six antibodies form a VH1-2 antibody class. The structural, recombination, and somatic hypermutation analyses (described below in the next section) presented here revealed critical residues that determine the specificity and binding affinity of the VH1-2 class antibodies. We define residues forming conserved side chain polar and/or hydrophobic interactions as genetic signatures. The heavy chain signatures include a T30-x-x-Y33 J o u r n a l P r e -p r o o f 12 motif in CDRH1 and a [NS] 52-x-[NIV]-S54 motif (x represents any amino acid) in CDRH2 ( Figure 4C ), with no conserved motif observed in CDRH3 ( Figure S1A ). Searches of germline gene databases using this signature showed only alleles of the VH1-2 gene to match both motifs ( Figure 4E ), suggesting that the antibody class is likely to be VH1-2-restricted. For light chain, residues Y32 LC and Y91 LC are signature residues that interact with both the 'primary' RBD ( Figure 3 and S3B) and N343 glycan from adjacent RBDs (described below in the quaternary recognition section). However, because the light chain interaction-signatures can be found in many germline genes ( Figure 4D and S4D), the VH1-2 antibody class may not be restricted with respect to light chain origin gene. To understand the conservation of the VH1-2 antibody epitope, we calculated the conservation score for each RBD residue and observed that the flat region is highly conserved in the natural SARS-CoV-2 reservoir. The four RBD residues critical for VH1-2 antibody recognition, Y449 RBD , E484 RBD , F486 RBD , and Q493 RBD , showed low mutation frequencies (approximately 0, 3, 7, and 7 per 10,000 sequences, respectively). Nonetheless, SARS-CoV-2 strains with mutations at the four positions are resistant to C121 and S2M11 of the VH1-2 antibody class (Barnes et al., 2020a; Tortorici et al., 2020) . In addition, the VH1-2 antibodies show high similarity in approach angle and epitope to antibody C144 ( Figure 4C and S4A), a type II VH3-53/-66 antibody. However, C144 uses different CDRH1 and CDRH2 motifs for interacting with the conserved residues in the flat region and ridge of RBD (e.g. the N32-Y33 and [T/S]54-G-G-[T/S]57 motifs in the VH3-53/-66 antibodies) ( Figure 4C ) (Barnes et al., 2020a; Wu et al., 2020a) . The CDRH2 of C144 shifts towards the RBD ridge, perhaps because the CDRH2s of the VH3-53/-66 genes are one residue shorter than the VH1-2 gene. J o u r n a l P r e -p r o o f 13 Nonetheless, both antibody classes use similar light chain genes including VL2-14 and VL2-23, which provide key Tyr residues that interact with F486 RBD . To understand the effects of somatic hypermutation (SHM) on binding affinity and neutralization potency, we reverted SHMs in the paratope regions of 2-43, 2-15, and 2-4 to their germline residues individually and in combination. 2-43 only has one SHM in the paratope region, S76T HC ( Figure S1A ), the reversion of which reduces the IgG apparent binding affinity and pseudovirus neutralization potency by about 6-fold and 5-fold respectively ( Figure 5A , 5B, and S4E). For 2-15, three paratope SHMs were observed: N53I HC , G55D HC , and Y32F LC . The reversion of N53I HC and Y32F LC individually reduced binding affinity by 56-fold and 12-fold respectively ( Figure 5A ) as well as neutralization potency by 217-fold and 42-fold respectively ( Figure 5B ). The results suggested that the N53I HC and F32Y HC mutations are critical for the affinity maturation of 2-15. Structural analysis showed that the N53I HC mutation allows the Ile side chain to fit a hydrophobic pocket on RBD ( Figure 3B and S5A). A convergent mutation is also observed in S2M11 (N53I HC ) and C121 (N53V HC ). We then introduced the N53I HC mutation to 2-43 and 2-4 and observed significant improvements of both binding affinity and neutralization potency (both IC 50 and maximum potency) ( Figure 5 ). These results suggested that mutation to a hydrophobic residue at 53 HC can be used to improve members of the VH1-2 antibody class. The crystal structure further showed that 2-15 32 LC does not interact with RBD ( Figure S3B ), the Y32F LC SHM may alter the interaction with RBD indirectly. For 2-4, the SHM A60T HC generates an N-glycosylation site at position N58 HC . The N58 HC glycan interacts with the RBD ridge ( Figure 3C ). The reversion of this SHM reduces binding affinity and neutralization by about 6-fold and 9-J o u r n a l P r e -p r o o f 14 fold respectively ( Figure 5 ). In summary, somatic hypermutation analysis revealed that the precursors of these antibodies could bind antigens with nanomolar apparent binding affinity, suggesting that the precursors of these antibodies can be activated efficiently by the SARS-CoV-2 spike protein. Nonetheless, SHMs significantly improve both the binding affinity and the neutralization potency of these antibodies. Because the observed SHMs are frequently generated by the somatic hypermutation machinery ( Figure S1A ), we anticipate that requirements for somatic hypermutations are unlikely to present a significant barrier for affinity maturation of this antibody class. In addition, previous studies showed that bivalent binding (avidity) is critical for certain RBD-directed antibodies to achieve high potency (Barnes et al., 2020b; . We therefore examined whether avidity contributes to the observed high potency of the VH1-2 antibody class. Comparison of neutralization potency between IgGs and Fabs revealed that the Fabs of 2-43, 2-15, and 2-4 have potencies of about 140-, 95-, and 14-fold lower than their IgG forms respectively ( Figure S5C ), suggesting that avidity effects are critical for achieving high neutralization potency by the VH1-2 class antibodies. Our previous study with a low resolution cryo-EM structure showed that 2-43 recognizes a quaternary epitope . However, details on the quaternary interactions and their functional relevance have not yet been characterized. Here, high resolution cryo-EM reconstructions revealed atomic-level interactions between 2-43 and the quaternary epitope. Overall, 2-43 interacts comprehensively with N343 glycan as well as helix 364-371 from an adjacent 'down' RBD protomer (RBD B ) ( Figure 6A and S6A), burying a total of 999Å BSA J o u r n a l P r e -p r o o f 15 ( Figure S3A ). The quaternary interaction is predominantly mediated by the long CDRH3, which is held by the two branches of the N343 glycan, the structure of which is highly flexible and has not been observed in previous studies. The tip of one N343 glycan branch is further stabilized by hydrogen bonding with 2-43 CDRL1 and the ridge of RBD A ( Figure S6A ). Altogether, 2-43 forms strong quaternary interactions with the adjacent 'down' RBD, which is critical for locking RBDs in the all 'down' conformation. In contrast to 2-43, the cryo-EM density maps of the 2-15 and H4 spike complexes revealed no quaternary contacts ( Figure 2D , S2L, and 6B). Comparison of quaternary interactions among the five VH1-2-derived antibodies revealed two structural groups. Group 1 includes 2-43, 2-4, and S2M11, which bind predominantly 'down' RBDs. Group 2 includes 2-15, H4, and C121, which bind both 'up' and 'down' or only 'up' RBDs. For group 1, the quaternary recognition is mediated mostly by CDRH3, followed by CDRL1 and CDRL2 ( Figure 6C and S3A). Differences between each antibody-specific CDRH3 determines that each antibody has a unique quaternary epitope ( Figure 6A , 6B, 6C, S3B, and S6B). Different from other class members which bind both spike and isolated RBD Robbiani et al., 2020; Tortorici et al., 2020; Wu et al., 2020b) , the quaternary interaction is indispensable for 2-43 . The quaternary epitope of 2-4, only part of which appears to be visible in the cryo-EM map comprises only N343 glycan from the adjacent 'down' RBD protomer ( Figure 6C ), likely because 2-4 has a short CDRH3. The quaternary epitope of S2M11 includes N343 glycan and helices 339-343 and 364-371 of the adjacent RBD ( Figure S6B ). Altogether, accommodation of N343 glycan from the adjacent 'down' RBD is a common feature of quaternary recognition by group 1 antibodies. The diverse CDRH3s of the group 1 antibodies form a module for quaternary recognition. In contrast, group 2 antibodies 2-15 and C121 show J o u r n a l P r e -p r o o f 16 minor interactions with neighboring 'down' RBDs ( Figure 2C , 6B, S2R, and S3A), which may explain why they do not lock the trimeric RBDs in an all 'down' conformation. To understand whether antibody binding induces conformation change in the SARS-CoV-2 spike protein, we compared the antibody-bound and ligand-free spike structures in the closed (all 'down') prefusion conformation. These comparisons showed that the binding of the VH1-2 antibodies (except S2M11), as well as other RBD-directed antibodies leads to significantly larger distances and less contacts between the trimeric 'down' RBDs ( Figure 6D and 6E). Each of these antibodies plays a role in bridging the interactions between RBDs as well as reorienting the conformation of N343 glycan. However, antibodies like 2-15, which does not form strong interactions with the adjacent RBD, may disassemble the spike when 3 Fabs bind to the RBDs ( Figure S2J and S2K). In addition, superposition of 2-15 on the 'down' RBD in ligand-free spike revealed the light chain to significantly clash with an adjacent 'up' RBD ( Figure S6D ), suggesting members of the VH1-2 antibody class to either bind only 'down' RBDs when a neighboring 'down' RBD is available or push the neighboring 'up' RBD away to bind 'down' RBDs. Consistent with the first mechanism, we did not observe a neighboring 'up' RBD adjacent to 2-15 and 2-4 bound 'down' RBDs in the cryo-EM data. In contrast, C121 adopts the second mechanism through light chain interaction with the adjacent 'up' RBD ( Figure S6D ) (Barnes et al., 2020a) . In summary, the quaternary interaction and angle of approach determine that the VH1-2 antibody class modulates SARS-CoV-2 spike conformation when binding to a 'down' RBD. In this study, we determined structures of three nAbs: 2-43, 2-15, and H4, which revealed a VH1-2 antibody class with a common RBD-binding mode and similar angle of approach. The VH1-J o u r n a l P r e -p r o o f 17 2 antibody class uses two modules for spike recognition with the VH1-2 gene encoded module for recognition of RBD and CDRH3 module for quaternary recognition. The VH1-2 antibody class has little or no restraint on CDRH3 length. The prevalence of the VH1-2 antibody class in numerous donors (Brouwer et al., 2020; Hansen et al., 2020; Ju et al., 2020; Kreer et al., 2020; Liu et al., 2020; Robbiani et al., 2020; Rogers et al., 2020; Tortorici et al., 2020; Wu et al., 2020b; Zost et al., 2020) suggests it to be a common component of the effective antibody response to SARS-CoV-2 that can include highly potent neutralizing antibodies. For some members of the VH1-2 class, recognition of a quaternary epitope can lock RBDs in the 'down' conformation, providing an additional mechanism to achieve high neutralization potency. The quaternary epitopes of the VH1-2 class include the N343 glycan and helix 364-371 of an adjacent RBD. A similar quaternary epitope is recognized by the type II VH3-53/-66 class antibody C144 (Barnes et al., 2020a) . Thus, this quaternary epitope appears to represent a supersite at which antibodies lock the RBDs in the all 'down' conformation. We also observe that the quaternary interaction can induce distinct conformational changes of the trimeric 'down' RBDs. The quaternary interaction observed in S2M11 does not significantly alter the distance between spike protomers ( Figure 4E ) (Tortorici et al., 2020) . In contrast, the quaternary interactions of 2-43 and 2-4, mediated predominantly by CDRH3, increase the distance of the trimeric RBDs ( Figure 4 ). For 2-15, the cryo-EM structure of one Fab bound spike revealed a CDRH3 conformation moving away from the adjacent 'down' RBD to avoid steric clash. Despite this, 2-15 also increases the distance between RBDs, which may be the result of mobile quaternary contacts that cannot be observed in the low resolution cryo-EM reconstruction. We anticipate that the interactions between RBDs could be further weakened when all three protomers are bound by 2-15, which may be the cause of the observed spike disassembly by 2-15 ( Figure S2J and S2K). The disassembly of spike J o u r n a l P r e -p r o o f 18 may be another mechanism for 2-15 to achieve ultrapotent neutralization. In addition, the crystal structure of 2-15 in complex with isolated RBD showed 2-15 to have another stable binding mode in the absence of a quaternary contact. We hypothesize that this binding mode may be observed when 2-15 binds to 'up' RBDs and cannot form quaternary interactions. The structural information we present also provides clues for further optimization of the VH1-2 antibody class. Despite members of the class achieving high potency with germline gene mediated interactions, we observe that somatic hypermutations can further improve this class significantly. The conserved mode of RBD recognition implies that members of the antibody class will be sensitive to similar viral escape mutations. In particular, K417N RBD , L452R RBD , E484K RBD , and N501Y RBD mutations are observed in emerging SARS-CoV-2 variants (e.g., B.1.351, B.1.1.7, and B.1.429) with transmission rates dramatically higher than preexisting strains (Davies et al., 2020; Tegally et al., 2020) . We found that positions 452 and 484 are within the VH1-2 antibody epitopes ( Figure S3E ). Mutation E484K RBD but not K417N RBD and N501Y RBD impairs both binding affinity and neutralization potency of the VH1-2 antibody class (Barnes et al., 2020a; Tortorici et al., 2020; Wang et al., 2021) . We anticipate that L452R RBD will also impair the recognition by the VH1-2 antibody class, but further investigation is required to examine the hypothesis. Cryo-EM data collection was performed at the National Center for Cryo-EM Access and Training (A) Gene usage of SARS-CoV-2 neutralizing antibodies. SARS-CoV-2 spike-specific VH1-2 antibodies are frequently induced in infected humans. VH1-2 antibodies are significantly enriched in the antigen-specific antibody repertoire than that of heathy donors. Antibody repertoires from 17 healthy donors were used for the analysis. (B) Many SARS-CoV-2 neutralizing VH1-2 antibodies (red) isolated from human donors achieve high potency, comparable to the most frequent IGHV3-53/-66 antibodies (blue). The half maximal inhibitory concentration (IC 50 ) of neutralization is shown except IC 100 for antibodies from Kreer et al., (2020) . Live virus neutralization potency is shown for antibodies from nine studies except the Hansen study. IC 50 values greater than 10µg/ml are set to 10µg/ml. Neutralizing antibodies targeting both RBD and non-RBD epitopes are included. (C) SARS-CoV-2 neutralizing VH1-2 antibodies use diverse light chain genes. (D) Sequence alignment of the heavy chain of six VH1-2 antibodies. Antibodies with structures reported in this study are highlighted in red. Residues identical to germline gene are shown as dots. See also Figure S1 and Table S1. H4 is colored cyan. See also Figure S2 , Table S2, and Table S3 . (A) Overview of the 2-43 epitope (left panel) and close-up view of the hydrogen bond networks between 2-43 and the SARS-CoV-2 RBD (right three panels). The RBD epitope recognized by 2-43 heavy (forest) and light (limegreen) chains are colored wheat and orange respectively (RBD A , light gray). Epitope residues interacting with both heavy and light chains are colored lemon. 2-43 also binds to N343 glycan (magenta) from a neighboring RBD (RBD B , dark gray). Hydrogen bonds and π-π stacking are shown as black dashed lines. (B) Overview of the epitope of 2-15 (left panel) and close-up view of the hydrogen bond networks and π-π stackings between 2-15 and the SARS-CoV-2 RBD (right three panels). 2-15 heavy and J o u r n a l P r e -p r o o f 21 light chains are colored marine and blue respectively. Epitope residues interacting with both heavy and light chains are colored lemon. (C) Overview of the epitope of 2-4 (left panel) and close-up view of the hydrogen bond networks between 2-4 and the SARS-CoV-2 RBD (right three panels). The heavy and light chains of 2-4 are colored chocolate and salmon respectively. Epitope residues interacting with both heavy and light chains are colored lemon. See also Figure S3 . (D) Distance between antibody free and antibody bound trimeric RBDs. Antibody binding induces significantly larger expansion of the trimeric RBDs. The three RBDs were colored gray. Residues at RBD interfaces were colored cyan, light blue, and blue for protomers RBD A , RBD B , and RBD C respectively. The distances between RBD protomers were measured using C⍺ of position 503. Note: due to low resolution, the interface residues between RBDs were not shown for 2-15 bound spike. 2-15 (dashed circle) binds to protomer RBD C . Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Zizhang Sheng (zs2248@cumc.columbia.edu). Expression plasmids generated in this study for expressing SARS-CoV-2 proteins and antibody mutants will be shared upon request. and Vero E6 cells (cat# CRL-1586™) were from ATCC. The mammalian expression vector that encodes the ectodomain of the SARS-CoV-2 S trimer was kindly provided by Jason McLellan (Wrapp et al., 2020) . SARS-CoV-2 S trimer expression vector was transiently transfected into Expi293 TM cells using 1 mg/mL of polyethylenimine IgGs were tested using a three-fold dilution series of IgGs with concentrations ranging from 33.3 nM to 1.2 nM. The association and dissociation rates were each monitored for 55s and 300s respectively, at 50µL/min. The bound spike/IgG complex was regenerated from the anti-his antibody surface using 10 mM Glycine pH 1.5. Blank buffer cycles were performed by injecting running buffer instead of IgG to remove systematic noise from the binding signal. The resulting data was processed and fit to a 1:1 binding model using Biacore Evaluation Software. Recombinant Indiana VSV (rVSV) expressing SARS-CoV-2 spikes were generated as previously described (Han et al., 2020; Liu et al., 2020) . HEK293T cells were grown to 80% confluency before transfection with pCMV3-SARS-CoV-2-spike (kindly provided by Dr. Peihui Wang, Shandong University, China) using FuGENE 6 (Promega). Cells were cultured overnight at 37°C with 5% CO 2 . The next day, medium was removed and VSV-G pseudo-typed ∆G-luciferase (G*∆G-luciferase, Kerafast) was used to infect the cells in DMEM at an MOI of 3 for 1 hr before washing the cells with 1X DPBS three times. DMEM supplemented with anti-VSV-G antibody (I1, mouse hybridoma supernatant from CRL-2700; ATCC) was added to the infected cells and they were cultured overnight as described above. The next day, the supernatant was harvested and clarified by centrifugation at 300g for 10 min and aliquots stored at −80°C. The 158 SARS-COV-2 neutralizing antibodies were collected from ten publications. We annotated these antibodies using IgBLAST-1.16.0 with the default parameters (Ye et al., 2013) . For antibodies which have cDNA sequences deposited, the V and J genes were assigned using SONAR version 2.0 (https://github.com/scharch/sonar/) with germline gene database from IMGT (Lefranc et al., 2009; Schramm et al., 2016) . For each antibody, the N-addition, D gene, and Paddition regions were annotated by IMGT V-QUEST (Brochet et al., 2008) . To identify somatic hypermutations, each antibody sequence was aligned to the assigned germline gene using MUSCLE v3.8.31 (Edgar, 2004) . Somatic hypermutations were identified from the alignment. In addition, the analysis of single cell antibody repertoire sequencing data of SARS-CoV-2 patient 2 from , from which 2-15 and 2-43 were isolated, showed that 29 of the 38 unique transcripts assigned to IGLV2-14*01 share nucleotide mutations G156T and T165G. These mutations lead to amino acid mutations E50D and N53K. Both nucleotide mutations are also observed in 82 of 90 unique IGLV2-14 transcripts from patient 1 of the same study. Because these transcripts having different VJ recombination and paired with different heavy chain genes, the chances that the two convergent mutations are the results of somatic hypermutation are very low. Thus, we suspect that both donors contain a new IGVL2-14 gene allele (IGVL2-14*0X), which was deposited to European Nucleotide Archive (ENA) with project accession numbers: PRJEB31020. 2-43 and 2-15 were assigned to the IGLV2-14*0X allele. For mAb 2-43, SARS-CoV-2 spike protein at a concentration of 2 mg/ml was incubated with six-fold molar excess per spike trimer of the antibody Fab fragments for 30 minutes in 10 mM sodium acetate pH 5.5, 150 mM NaCl, and 0.005% n-dodecyl-β-D-maltoside (DDM). 2µL of the J o u r n a l P r e -p r o o f 33 sample was incubated on C-flat 1.2/1.3 carbon grids for 30 seconds and vitrified using a Leica EM GP. Data were collected on a Titan Krios electron microscope operating at 300 kV, equipped with a Gatan K3 direct electron detector and energy filter, using the Leginon software package (Suloway et al., 2005) . A total electron fluence of 51.69 e-/Å2 was fractionated over 40 frames, with a total exposure time of 2.0 seconds. A magnification of 81,000x resulted in a pixel size of 1.058 Å, and a defocus range of -0.4 to -3.5 µm was used. For mAb 2-15, SARS-CoV-2 spike protein at a concentration of 1 mg/ml was incubated with a molar ratio of 1:1 Fab fragments to spike trimer for 30 minutes in 10 mM sodium acetate pH 5.5, 150 mM NaCl, and 0.005% % DDM. 2µL of the sample was incubated on C-flat 1.2/1.3 carbon grids for 30 seconds and vitrified using a Leica EM GP. Data were collected on a Titan Krios electron microscope operating at 300 kV, equipped with a Gatan K3 direct electron detector and energy filter, using the Leginon software package (Suloway et al., 2005) . All processing was done using cryoSPARC v2.15.0 (Punjani et al., 2017) . Raw movies were aligned and dose-weighted using patch motion correction, and the CTF was estimated using patch CTF estimation. A small subset of approximately 200 micrographs were picked using blob picker, followed by 2D classification and manual curation of particle picks, and used to train a Topaz neural network. This network was then used to pick particles from the remaining micrographs, which were extracted with a box size of 384 pixels. For the mAb 2-43 dataset, 2D classification followed by ab initio modelling and 3D heterogeneous refinement revealed 61,434 particles with three Fabs bound, one to each RBD. A reconstruction of these particles with imposed C3 symmetry resulted in a 3.78 Å map, as determined by gold standard Fourier shell correlation (FSC). Symmetry expansion followed by masked local refinement was used to obtain a 3.81 Å map of the Fab and RBD. The remainder of the S trimer was subjected to local refinement to obtain a 3.61 Å map. These two separate local refinements were aligned and combined using the vop maximum function in UCSF Chimera (Pettersen et al., 2004) . This was repeated for the half maps, which were used along with the refinement mask from the global non-uniform refinement to calculate the 3D FSC (ref) and obtain an estimated resolution of 3.60 Å. All maps have been submitted to the EMDB with the ID EMD-#####. For the mAb 2-15 dataset, 2D classification followed by ab initio modelling and 3D heterogeneous refinement revealed 16,590 particles with one Fab bound to an RBD in the 'down' conformation and 21,456 particles with one Fab bound to an RBD in the 'up' conformation. The particles with Fab bound to RBD down were refined using Non-uniform refinement and C1 J o u r n a l P r e -p r o o f 35 symmetry to a global resolution of 5.73 Å as determined by gold standard FSC. The RBD and Fab were masked and subjected to local refinement to obtain a map at 6.21 Å. The remainder of the trimer was also refined locally to 5.63 Å. A consensus map was obtained as described for mAb 2-43, with a resolution of 5.87 Å. All maps have been submitted to the EMDB with the ID EMD-#####. For the mAb H4 dataset, 2D classification followed by ab initio modelling and 3D heterogeneous refinement revealed 102,290 particles with one Fab bound to an RBD in the 'up' conformation. No classes with Fab bound to 'down' RBD were identified. 3D Variability Analysis was used to visualize the significant conformational heterogeneity of the 'up' RBD. Using the first and last frames of the reaction coordinate as reference maps, representing the extremes of the orientations adopted by the RBD, heterogeneous refinement was repeated to separate the Fabbound spikes into more homogeneous classes. 56,080 particles adopted a more stable conformation and were refined to 4.78 Å using homogeneous refinement and C1 symmetry. Like the previously described datasets, the Fab and RBD were refined locally to 5.03 Å, with the remainder of the S trimer being refined to 4.32 Å. The final consensus map was estimated to have a resolution of 5.07 Å. All maps have been submitted to the EMDB with the ID EMD-#####. An initial homology model of the 2-43 Fab was built using Schrödinger Release 2020-2:BioLuminate (Zhu et al., 2014) and of H4 using ABodyBuilder (Leem et al., 2016) . For mAb 2-15, the crystal structure determined here (PDB ID ####) was used as a starting model for the Fab variable domain and the associated RBD. For all models, the S trimer was modeled using the coordinates from PDB ID 6XEY. These models were docked into the consensus map using Chimera. The model was then fitted interactively using ISOLDE 1.0.1 (Croll, 2018) and COOT J o u r n a l P r e -p r o o f 36 0.8.9.2 (Emsley and Cowtan, 2004) and using real space refinement in Phenix 1.18 (Adams et al., 2004) . For Fab 2-43, in cases were side chains were not visible in the experimental data, they were truncated to alanine except for residues very close to the RBD:Fab interface. Both H4 and 2-15 were built as poly-alanine models due to the low resolution of the experimental data. Validation was performed using Molprobity (Davis et al., 2004) and EMRinger (Barad et al., 2015) . Models were submitted to the PDB with the following IDs: mAb 2-43 is 7L56, mAb H4 is 7L58, and 2-15 is 7L57. Figures were prepared using UCSF ChimeraX (Goddard et al., 2018) . For determination of the complex of with RBD, the two proteins were mixed at 1:1 molar ratio and incubated at 4.0°C for 60 min. RBD:2-15 complex was then isolated by gel filtration on Superdex-200 (GE Healthcare). Fractions containing complexes were pooled and concentrated to 6.5 mg/ml in SEC buffer. Screening for initial crystallization conditions was carried out in 96-well sitting drop plates using the vapour-diffusion method with a Mosquito crystallization robot (TTP LabTech) using various commercially available crystallization screens. Diffracting crystals were obtained from 0.1 M Hepes pH 7.5 and 70% MPD. Crystals were directly frozen and X-ray diffraction data was collected to 3.18 Å resolution at 100 K from a single flash-cooled crystal on beamline ID-C at the Advance Photon Source (APS) at Argonne National Laboratory. Diffraction data were processed with XDS (Kabsch, 2010) and scaled using AIMLESS (Evans, 2006) and Fab (PDB code, 4HK0) as search models. Manual rebuilding of the structure using COOT (Emsley et al., 2010) was alternated with refinement using Phenix refine (Afonine et al., 2012) and PDB-REDO (Joosten et al., 2014) . The Molprobity server was used for structure validation and J o u r n a l P r e -p r o o f 37 PyMOL (version 2.1, Schrödinger, LLC) for structure visualization (Chen et al., 2010) . A summary of the X-ray data collection and refinement statistics are shown in Table S2 . PISA was used to identify paratope-epitope interfaces and to calculate buried surface area (Krissinel and Henrick, 2007) . Hydrogen bonds were identified with a cutoff of 3.8Å. 2-15 model was submitted to the PDB with the following IDs: 7L5B. To measure the RBD approaching angle of antibodies, we first identify shared epitope residues among the five members of the VH1-2 antibody class. PyMOL was used to determine the centre of mass of the shared epitope residues. We then determined the center of mass for heavy (the centre of mass of the conserved Cys at 22 and 92) and light chains (the centre of mass of the conserved Cys at 23 and 88). The heavy and light chain approaching angle was determined by linking the chain centre of mass to the centre of mass of the shared epitope. PyMOL was used to make structure figures. For antibody B38, the epitope center of mass was determined using the epitope residues of the antibody. The statistical analyses for the pseudovirus neutralization assessments were performed using GraphPad Prism. The SPR data were fitted using Biacore Evaluation Software. Cryo-EM data were processed and analyzed using CryoSparc and Relion. Cryo-EM structural statistics were analyzed with Phenix and Molprobity. Statistical details of experiments are described in Method Details or figure legends. J o u r n a l P r e -p r o o f Figure S1 . Alignment of heavy and light chain sequences of IGHV1-2 antibodies. Related to Figures 1, 3, 4, 5. (A) Heavy chain sequence alignment of the VH1-2-derived antibodies with paratope residues of the VH1-2 class highlighted with underscore. Residues forming backbone hydrogen bonds, side chain hydrogen bond, hydrophobic contact with the SARS-CoV-2 RBD are colored magenta, red, green respectively. Residues contacting quaternary epitopes are colored cyan. Residues interacting with N343 glycan is shown with italic font. Antibody positions and CDR definitions are assigned using the Kabat scheme. The gene-specific substitution profile of VH1-2 gene shows the frequencies of somatic hypermutation to be generated by the somatic hypermutation machinery. (B) Heavy chain gene recombination of the six VH1-2 antibodies. For H4 and S2M11, no N-or P-addition sites were determined because the nucleotide sequences are unavailable. H4 paratope residues were not identified due to low resolution of the structure. (C) Light chain sequence alignment of the VH1-2 antibody class members with paratope residues highlighted with underscore. Side chains are shown for residues with buried accessible surface area more than 20Å 2 . (C) Overview of the S2M11 epitope (left panel) and close-up view of the hydrogen bond networks between S2M11 and the SARS-CoV-2 RBD (right three panels). The RBD epitope recognized by S2M11 heavy and light chains are colored wheat and orange respectively (RBD A , light gray). Epitope residues interacting with both heavy and light chains were colored lemon. S2M11 also binds to N343 glycan (magenta) from a neighboring RBD (RBD B , dark gray). Hydrogen bonds are shown as black dashed lines. PDB structure 7K43 is used. (D) Overview of the C121 epitope (left panel) and close-up view of the hydrogen bond networks between C121 and the SARS-CoV-2 RBD (right three panels) (PDB ID 7K8X). The RBD epitope recognized by C121 heavy (palecyan) and light (lightblue) chains are colored wheat and orange respectively (RBD A , light gray). Epitope residues interacting with both heavy and light chains are colored lemon. J o u r n a l P r e -p r o o f Figure S4 . Similarities in binding orientation and epitopes of RBD-targeting antibodies. Related to figure 4. (A) Clustering of SARS-CoV-2 RBD-targeting antibodies. For each pair of antibodies, the RBDs were superimposed and the RMSD of Ca atom between antibody variable domains were then calculated. Antibodies were then clustered using the calculated pairwise RMSD values. (B) Epitope residues of SARS-CoV-2 RBD-targeting antibodies. For each antibody, epitope residues are shown and non-epitope residues are masked with dots. ACE2 binding site residues are highlighted with cyan background. (C) The VH1-2 class antibodies have RBD epitopes that overlap with the ACE2 binding site. (D) Alignment of the light chain germline genes used by the SARS-CoV-2 RBD-targeting VH1-2 antibodies in Figure 1C as well as other germline genes having residues Y32 LC and Y91 LC . (D) Superimposition of C121 in complex with spike to antibody-free SARS-CoV-2 spike. The distance between the two RBD protomers increases in C121 bound spike, suggesting that the interaction of C121 light chain with the adjacent 'up' RBD induces conformation change in RBDs. PDB ID: 6XDG Cryo-EM structure: C105 Barnes et Recent developments in the PHENIX software for automated crystallographic structure determination Towards automated crystallographic structure refinement with phenix.refine EMRinger: side chain-directed model and map validation for 3D cryo-electron microscopy SARS-CoV-2 neutralizing antibody structures inform therapeutic strategies Structures of Human Antibodies Bound to SARS-CoV-2 Spike Reveal Common Epitopes and Recurrent Features of Antibodies IMGT/V-QUEST: the highly customized and integrated system for IG and TR standardized V-J and V-D-J sequence analysis Potent neutralizing antibodies from COVID-19 patients define multiple targets of vulnerability SARS-CoV-2 Neutralizing Antibody LY-CoV555 in Outpatients with Covid-19 MolProbity: all-atom structure validation for macromolecular crystallography The CCP4 suite: programs for protein crystallography ISOLDE: a physically realistic environment for model building into lowresolution electron-density maps Estimated transmissibility and severity of novel SARS-CoV-2 MOLPROBITY: structure validation and all-atom contact analysis for nucleic acids and their complexes A Phase I Trial to Evaluate the Safety and Immunogenicity of eOD-GT8 60mer Vaccine An interactive web-based dashboard to track COVID-19 in real time Polyclonal and convergent antibody response to Ebola virus vaccine rVSV-ZEBOV Coot: model-building tools for molecular graphics Features and development of Coot Scaling and assessment of data quality UCSF ChimeraX: Meeting modern challenges in visualization and analysis Identification of SARS-CoV-2 inhibitors using lung and colonic organoids Studies in humanized mice and convalescent humans yield a SARS-CoV-2 antibody cocktail B-cell-lineage immunogen design in vaccine development with HIV-1 as a case study SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor Antihomotypic affinity maturation improves human B cell responses against a repetitive epitope Rational HIV immunogen design to target specific germline B cell receptors The PDB_REDO server for macromolecular structure model optimization Vaccine-Induced Antibodies that Neutralize Group 1 and Group 2 Influenza A Viruses Human neutralizing antibodies elicited by SARS-CoV-2 infection Integration, scaling, space-group assignment and post-refinement Structure and Function Analysis of an Antibody Recognizing All Influenza A Subtypes Longitudinal Isolation of Potent Near-Germline SARS-CoV-2-Neutralizing Antibodies from COVID-19 Patients Inference of macromolecular assemblies from crystalline state Human Antibodies that Neutralize HIV-1: Identification, Structures, and B Cell Ontogenies HIV-1 Vaccines Based on Antibody Identification, B Cell Ontogeny, and Epitope Structure ABodyBuilder: Automated antibody structure prediction with data-driven accuracy estimation IMGT, the international ImMunoGeneTics information system Cross-Neutralization of a SARS-CoV-2 Antibody to a Functionally Conserved Site Is Mediated by Avidity Potent neutralizing antibodies against multiple epitopes on SARS-CoV-2 spike Phaser crystallographic software Rapid development of broadly influenza neutralizing antibodies through redundant mutations UCSF Chimera--a visualization system for exploratory research and analysis cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination Convergent antibody responses to SARS-CoV-2 in convalescent individuals Isolation of potent SARS-CoV-2 neutralizing antibodies and protection from disease in a small animal model SONAR: A High-Throughput Pipeline for Inferring Antibody Ontogenies from Longitudinal Sequencing of B Cell Transcripts Automated molecular microscopy: the new Leginon system Emergence and rapid spread of a new severe acute respiratory syndrome-related coronavirus 2 (SARS-CoV-2) lineage with multiple spike mutations in South Africa. medRxiv Ultrapotent human antibodies protect against SARS-CoV-2 challenge via multiple mechanisms Increased Resistance of SARS-CoV-2 Variants B.1.351 and B.1.1.7 to Antibody Neutralization. bioRxiv Structural and Functional Basis of SARS-CoV-2 Entry by Using Human ACE2 Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation An Alternative Binding Mode Antibodies to the SARS-CoV-2 Receptor Binding Domain A noncompeting pair of human neutralizing antibodies block COVID-19 virus binding to its receptor ACE2 IgBLAST: an immunoglobulin variable domain sequence analysis tool Structural basis of a shared antibody response to SARS-CoV-2 A pneumonia outbreak associated with a new coronavirus of probable bat origin Structural Repertoire of HIV-1-Neutralizing Antibodies Targeting the CD4 Supersite in 14 Donors Multidonor analysis reveals structural elements, genetic determinants, and maturation pathway for HIV-1 neutralization by VRC01-class antibodies Antibody structure determination using a combination of homology modeling, energy-based refinement, and loop prediction A Novel Coronavirus from Patients with Pneumonia in China Potently neutralizing and protective human antibodies against SARS-CoV-2