Antibodies with potent and broad neutralizing activity against antigenically diverse and highly transmissible SARS-CoV-2 variants

The emergence of highly transmissible SARS-CoV-2 variants of concern (VOC) that are resistant to therapeutic antibodies highlights the need for continuing discovery of broadly reactive antibodies. We identify four receptor-binding domain targeting antibodies from three early-outbreak convalescent donors with potent neutralizing activity against 12 variants including the B.1.1.7 and B.1.351 VOCs. Two of them are ultrapotent, with sub-nanomolar neutralization titers (IC50 <0.0006 to 0.0102 μg/mL; IC80 < 0.0006 to 0.0251 μg/mL). We define the structural and functional determinants of binding for all four VOC-targeting antibodies, and show that combinations of two antibodies decrease the in vitro generation of escape mutants, suggesting potential means to mitigate resistance development. These results define the basis of therapeutic cocktails against VOCs and suggest that targeted boosting of existing immunity may increase vaccine breadth against VOCs.

resistance to therapeutic monoclonal antibodies, have increased transmissibility and to potentially increase pathogenicity (10)(11)(12)(13)(14). Additionally, vaccines designed based on the original WA-1 outbreak strain sequence elicit antibody responses that show decreased in vitro neutralizing activity against variants (14)(15)(16). In this study, we investigated antibodies isolated from convalescent subjects who were infected by the WA-1 strain during the first few months of the 5 outbreak, determined their reactivity against variants of concern (VOCs) and defined the structural features of their binding to spike.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 early ancestral SARS-CoV-2 viruses, these antibodies have high-potency against B.1.1.7 and B.1.351 VOCs.
The two most potent antibodies, A23-58.1 and B1-182.1, shared highly similar gene family usage in their heavy and light chains, despite being from different donors (Table S1). Both use IGHV1-58 heavy chains and IGKV3-20/IGKJ1 light chains. This antibody gene family 5 combination has previously been noted to be present in other COVID-19 convalescent subjects and has been proposed as a public clonotype (29)(30)(31)(32). To gain structural insights on the interaction between this class of antibodies and the SARS-CoV-2 spike, we mixed A23-58.1 Fab and spike at a molar ratio of 3.6:1 and purified the complex by size-exclusion chromatography. We collected single particle cryo-EM data on a Titian Krios and determined the structure of the complex at 3.39 10 Å resolution ( Figure 3A, Figure S4 and Table S2) and revealed that the antibody bound to spike with all RBDs in the up position, confirming the negative stain results ( Figure 3A, Figure 1G).
However, the cryo-EM reconstruction density of the RBD and A23-58.1 interface was poor due to conformational variation.
To resolve the antibody-antigen interface, we performed local refinement and improved 15 the local resolution to 3.89 Å which enabled detailed analysis of the mode of antibody recognition ( Figure S4). Antibody A23-58.1 bound to an epitope on the RBD that faces the 3-fold axis of the spike and is accessible only in the RBD-up conformation ( Figure 3A). The interaction buried a total of 601 Å 2 surface area from the antibody and 607 Å 2 from the spike (Table S3). The A23-58.1 paratope constituted all six complementarity-determining regions (CDR) with both heavy 20 chain and light chain contributing 73% and 27% of binding surface area, respectively (Figure 3B, Figure 3C and Table S3). The 14-residue-long CDR H3, which provided 48% of the heavy chain paratope, kinked at Pro95 and Phe100F (Kabat numbering scheme for antibody residues) to form 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/2021.02.25.432969 doi: bioRxiv preprint a foot-like loop that is stabilized by an intra-loop disulfide bond between Cys97 and Cys100B at the arch. A glycan was observed to attach the CDR H3 Asn96 ( Figure S4F). The CDRs formed an interfacial crater with a depth of ~10 Å and a diameter of ~20 Å at the opening. Paratope residues inside the crater were primarily aromatic or hydrophobic. With CDR H3 Pro95 and Phe100F paving the bottom, CDR H1 Ala33, CDR H2 Trp50 and Val52, and CDR H3 Val100A lined the 5 heavy chain side of the crater ( Figure 3B/C). On the light chain side, CDR L1 Tyr32 and CDR L3 residues Tyr91 and Trp96 provided 80% of the light chain binding surface ( Figure 3B,C). In contrast, paratope residues at the rim of the crater are mainly hydrophilic, for example, Asp100D formed hydrogen bonds with Ser477 and Asn487 of the RBD ( Figure 3B-C, Table S3).
The A23-58.1 epitope is composed of residues between b5 and b6 at the tip of RBD ( Figure  10 3B and 3D). With the protruding Phe486 dipping into the crater formed by the CDRs, these residues formed a hook-like motif that is stabilized by an intra-loop disulfide bond between Cys480 and Cys488. Aromatic epitope residues, including Phe456, Tyr473, Phe486 and Tyr489, provided 38% of the binding surface (237.5 Å 2 ) ( Figure 3B and 3D, Table S3). Lys417 and Glu484, which are located at the outer edge of the epitope, contributed only 3.7% of the binding surface 15 ( Figure 3B and Table S3). Overall, the cryo-EM analysis provided structural basis for the potent neutralization of the E484K mutant by A23-58.1. The binding mode of A23-58.1 is very similar to that of a previously reported IGHV1-58/IGKV3-20-derived antibody, S2E12 (29) confirming that they are members of the same structural class ( Figure 3C and 3D). In addition, sequence analysis indicates that B1-182.1 is likely also a member of this class -and thus shares the same 20 mode of recognition. In fact, B1-182.1 share a nearly identical IGHD2-15-derived CDR H3 sequence with S2E12 ( Figure 3C). The lack of impact of emerging resistance mutations on B1-182.1 can be explained by the same mechanism whereby A23-58.1 antibody also is not impacted 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 by these variants. Interestingly, while A23-58.1 and B1-182.1 both have 10-residue-long CDR L3s, S2E12 includes the insertion of Val91A ( Figure 3C) and the longer CDR L3 may therefore lead to differences in binding/neutralization capacity for viruses with mutations at the Glu484 location where CDR L3 contacts.
Both REGN10933 and CB6 bind to the same side of the RBD that A23-58.1 contacted ( Figure 3D and 3E). However, their binding surfaces were all shifted towards the saddle of the open RBD and 10 encircled residues Lys417, Tyr453, Glu484 and Asn501 within the epitope ( Figure 3D and 3E), mutations K417N and Y453F potently abolished key interactions and led to the loss of neutralization for both REGN10933 and CB6 ( Figure 2). In contrast, LY-CoV555 approached the RBD from a different angle with its epitope centered around Glu484 ( Figure 3D and 3F). Modeling indicated that mutation E484K may abolish key interactions with Arg50 and Arg96 of LY- CoV555 15 and cause a clash with CDR H3 of LY-CoV555. These structural data suggest that the unique binding modes of A23-58.1 and potentially B1-182.1 derived from the same germline enabled their high effectiveness against the new SARS-CoV-2 variants.
We next used the structural analysis to investigate the relative contribution of predicted contact residues on binding and neutralization ( Figure 3D). Cell surface expressed spike binding 20 by A23-58.1 and B1-182.1 were knocked out by F486R, N487R, and Y489R ( Figure 4A, Figure   S5), resulting in a lack of neutralization for viruses pseudotyped with spikes containing these mutations ( Figure 4B). In contrast, binding and neutralization of A19-46.1 and A19-61.1 were 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 minimally impacted by these changes ( Figure 4A,B, Figure S5). CB6, LY-CoV555 and REGN10933 binding and neutralization were also impacted by the three mutations, consistent with the structural analysis that these residues are commonly shared contact(s) among the impacted antibodies. Taken together, the shared binding and neutralization defect imposed by these mutations on A23-58.1 and B1-182.1 suggests that the hook-like motif and CDR crater are critical 5 for the binding of antibodies within the VH1-58 public class.
Next, based on structural modeling of the negative stain EM density ( Figure 1G), we chose several mutants to investigate the determinants of binding for A19-46.1 and A19-61.1. Under conditions where A23-58.1 and B1-182.1 were not impacted, we found that L452R, F490R and S494R knocked out binding for A19-46.1 and S494R knocked out binding for A19-61.1 ( Figure  10 4A, Figure S5). In addition, the partial overlap of susceptibility to the selected mutations between A19-46.1 and A19-61.1 is in agreement with the antibody competition data showing similar, but distinct profiles ( Figure 1F) and indicates that these antibodies represent distinct antibody classes.
To explore resistance mechanisms that might be generated during the course of infection, we applied antibody selection pressure to replication competent vesicular stomatitis virus (rcVSV) 15 expressing the WA-1 SARS-CoV-2 spike (rcVSV-SARS2) (33) to identify spike mutations that confer in vitro resistance against A23-58.1, B1-182.1, A19-46.1 or A19-61.1 ( Figure S6). rcVSV-SARS2 was incubated with increasing concentrations of antibody, and cultures from the highest concentration of antibody with >20% cytopathic effect (CPE) were carried forward into a second round of selection to drive resistance (27) (Figure S6). A shift to higher antibody concentrations 20 required for neutralization indicated the presence of resistant viruses. To define and determine the relative frequency of mutations that accumulated at positions within the spike of resistant viruses, we performed Illumina-based shotgun sequencing ( Figure S6). Variants present at a frequency of 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 greater than 10% and increasing from round 1 to round 2 were considered to be positively selected resistant viruses. For A19-46.1, three selection mutations were generated: Y449S (freq. 15%), N450S (freq. 16%) and F490V (freq. 58%) ( Figure 4C, Figure S7). The most dominant, F490V, was found in 58% of sequences and is consistent with the previous finding that F490R knocked out binding and neutralization of A19-46.1 ( Figure 4A,C). These residues are clustered near one 5 another on RBD and would be expected to be accessible when RBD is in the up or down position ( Figure S7). Several of these contacts are shared by Class II RBD antibodies (22,34) and REGN10933 (26,35). Despite these shared contacts, A19-46.1 is able to neutralize variants that REGN10933 cannot (i.e., Var10, Var11 and B.1.351) ( Figure 2A-C), indicating that A19-46.1 makes critical contacts that are able to overcome resistance mutations that affect REGN10933. 10 Three residues were positively selected in the presence of A19-61.1: K444E (freq. 57%), G446V (freq. 24%) and G593R (freq. 19%) ( Figure 4A) and did not overlap with those selected by A19-46.1. G593R is located outside the RBD domain and the others are clustered nearby S494R, identified previously ( Figure 4A, Figure S7). The highest frequency change was at K444 and represented 57% of the sequences. This residue has been shown to be critical for the binding 15 of Class III RBD antibodies such as REGN10987 (22,26,27,35). Taken together with the ability of A19-61.1 to block ACE2 binding ( Figure 1F, Figure S2) and differential neutralization between the Class III antibody REGN10987 and A19-61.1 against Var3 (N439K/D614G) (i.e., significantly reduced neutralization with REGN10987) ( Figure 2C), A19-61.1 likely targets an epitope distinct epitope from REGN10987 and other Class III RBD antibodies. 20 Finally, a single mutation, F486S (freq. 91%) was positively selected for when virus was incubated in the presence of A23-58.1. This is in agreement with our structural analysis ( Figure   3B) that showed that F486 is located at the tip of RBD "hook" and contributes to the binding interface in the antibody "crater". This is dominated by aromatic side chains that form the "hook" and "crater" interface ( Figure 3A,B). Therefore, one possible explanation for the loss in activity is through the replacement of a hydrophobic aromatic residue (i.e., phenylalanine) with a small polar side chain (i.e., serine) ( Figure 3C).
To probe the relevance of in vitro derived resistance variants to the potential for clinical 5 resistance we next examined spikes derived from circulating virus variants. We investigated the relative frequency of variants containing the mutations present in the GISAID sequence database using the COVID-19 Viral Genome Analysis Pipeline (cov.lanl.gov) (23) in which, as of February 11, 2021, there were 417,702 entries. Out of these, the original WA-1 residues critical for A19-46.1 activity (i.e., Y449, N450, L452, F490 and S494) were present in 99.81-99.99% of sequences 10 available. Of the residues identified in our experiments to mediate resistance to A19-46.1, Y449S, N450S, L452R, F490L/V and S494R, only F490V has been noted in the database (5 sequences, 0.001%) ( Figure 4, Figure S7). For A19-61.1, the ancestral WA-1 residues, K444, G446, S494 and G593, were present in 99.81-100% of entries. Of the resistance-inducing residues identified, i.e., K444E, G446V, S494R, and G593R, only G446V has been noted in the database (106 sequences, 15 0.03%) ( Figure 4, Figure S7). Finally, for A23-58.1 and B1-182.1 the ancestral WA-1 residues F486, N487 and Y489 were present in >99.99% of sequences and none of the binding/resistance mutations identified in our experiments were noted in the database. The relative lack of resistance mutations found in circulating viruses suggests that the in vitro derived mutations exact a fitness cost on the virus and are not tolerated during infection but could also reflect either under-sampling 20 or the absence of other sources of selection pressure.
Viral genome sequencing has suggested the possibility that in addition to spread via transmission, convergent selection of de novo mutations may be occurring (6-9, 13, 23, 36). 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 Therefore, effective therapeutic antibody approaches might require new antibodies or combinations of antibodies to mitigate the impact of mutations. Based on their complementary modes of spike recognition and breadth of neutralizing activity, we hypothesized that combination of B1-182.1 with either A19-46.1 or A19-61.1 would decrease the rate of in vitro resistance acquisition compared to each antibody alone. As a first test, we used negative stain EM 3D 5 reconstructions to determine whether these combinations are able to bind simultaneously to spike protein. Consistent with the competition data ( Figure 1F), we found that the Fabs in both combinations were able to engage spike simultaneously with RBD in the up position ( Figure 4D).
Furthermore, we noted that binding was in a 3:1 Fab:spike ratio in most of the observed particles ( Figure 4D), revealing that the epitopes of A19-46.1 and A19-61.1 on the spike are accessible in 10 both RBD up and down positions ( Figure 1G and Figure 4D). This suggests that the combination allows alternative preferential mode of RBD engagement (i.e., RBD up vs. RBD down) by A19-46.1 and A19-61.1 that is not seen in the absence of B1-182.1 or A23-58.1.
Next, we determined the rates of in vitro resistance acquisition of combined treatments compared to individual antibodies using an rcVSV SARS-CoV-2 resistance generation approach. 15 We evaluated the capacity of individual antibodies or combinations to prevent the appearance of rcVSV SARS -CoV-2-induced cytopathic effect (CPE) throughout multiple rounds of passaging in the presence of increasing concentrations of antibodies. In each round, the well with the highest concentration of antibody with at least 20% CPE was carried forward into the next round. We found that wells with A19-61.1 or A19-46.1 single antibody treatment reached the 20% CPE 20 threshold in their 50 µg/mL well after 3 rounds of selection ( Figure 4E). Similarly, B1-182.1 single antibody treatment reached >20% CPE in the 50 µg/mL wells after 4 rounds ( Figure 4E).
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 CPE threshold was reached only at a concentration of 0.08 µg/mL and did not progress to higher concentrations despite 5 rounds of passaging ( Figure 4E). While further data are required, these results suggest that such combinations may lower the risk that a natural variant will lead to the complete loss of neutralizing activity and suggests a path forward for these antibodies as combination therapies. 5 Worldwide genomic sequencing has revealed the occurrence of SARS-CoV-2 variants that increase transmissibility and reduce potency of vaccine-induced and therapeutic antibodies (10)(11)(12)(13)(14)(15)(16). Recently, there has been a significant concern that antibody responses to natural infection and vaccinations using ancestral spike sequences may have focused responses that are overcome by mutations present in more recent isolates (e.g., E484K in B.1.351) (12-16). As a first step to 10 address the risk of reduced antibody potency against new variants, we isolated and defined new antibodies with neutralization breadth covering newly emerging SARS-CoV-2 variants. Increased potency and breadth were mediated by binding to regions of the RBD tip that are offset from E484K, which is a major determinant of resistance in VOCs (12-16). Our results show that highly potent neutralizing antibodies with activity against these variants was present in at least 3 of four 15 convalescent subjects who had been infected with ancestral variants of SARS-CoV-2. Furthermore, two antibodies from different subjects used VH genes associated with previously described public clonotypes (29,30). Overall, these data establish the rationale for a vaccine boosting regimen that may be used to selectively induce immune responses that increase the breadth and potency of antibodies targeting the RBD region of the spike glycoprotein. 20 Furthermore, since both variant sequence analysis and in vitro time to escape experiments suggest that combinations of these antibodies may have a lower risk for loss of neutralizing activity, these 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 antibodies represent a potential means to achieve both breadth against current VOCs and to mitigate risk against those that may develop in the future. 105 and is also made available for use under a CC0 license.

References and Notes
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 virus neutralization by protective human antibodies. Science. 351, 1343-6 (2016).

Acknowledgments:
We would like to thank the staff of the Clinical Trials Program of the Vaccine Research Center and the volunteers that made this research possible. We also appreciate 5 the assistance of Dr. Ruth Hunegnaw for assistance with figure preparation. We are grateful to Tara L. Fox of NCEF for collecting cryo-EM data and for technical assistance with cryo-EM data processing. We would like to thank Avan Antia, Rachel L. Davis and Farida Laboune for technical assistance with sequencing.  Tables S1-S3 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 References (36-52)
(B) Gross binding epitope distribution was determined using an MSD-based ELISA testing against RBD, NTD, S1, S-2P or HexaPro. S2 binding was inferred by S-2P or HexaPro binding without binding to other antigens. Indeterminant epitopes showed a mixed binding profile. Total 10 number of antibodies (i.e., 200) and absolute number of antibodies within each group is shown. (F) Biolayer interferometry-based epitope binning experiment. Competitor antibody (y-axis) is bound to S-2P prior to incubation with the analyte antibody or ACE2 protein (x-axis) as indicated and percent competition range bins are shown as red (>=75%), orange (60-75%) or 20 white <60%) (n=2). mAb114 is an anti-Ebola glycoprotein antibody and is included as a negative control (37) 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021    (which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/2021.02.25.432969 doi: bioRxiv preprint A23-58.1 Fab bound to the RBD is shown in orange and blue. Structure of the RBD and A23-58.1 after local focused refinement was shown to the right. The heavy chain CDRs are colored brown, salmon and orange for CDR H1, CDR H2 and CDR H3, respectively. The light chain CDRs are colored marine blue, light blue and purple blue for CDR L1, CDR L2 and CDR L3, respectively. The contour level of Cryo-EM map is 5.7s. 5 (B) Interaction between A23-58.1 and RBD. All CDRs were involved in binding of RBD.
Epitope of A23-58.1 is shown in bright green surface with a yellow border (left, viewing from antibody to RBD). RBD mutations in current circulating SARS-CoV-2 variants are colored red.
Lys417 and Glu484 are located at the edge of the epitope. The tip of the RBD binds to a cavity formed by the CDRs (right, viewing down to the cavity). Interactions between 10 aromatic/hydrophobic residues are prominent at the lower part of the cavity. Hydrogen bonds at the rim of the cavity are marked with dashed lines. RBD residues were labeled with italicized font.
(D) Epitope of A23-58.1 on RBD. Epitope residues for different RDB-targeting antibodies are marked with * under the RBD sequence.
(E) Comparison of binding modes of A23-58.1 and REGN10933. One Fab is shown to bind to the RBD on the spike. The shift of the binding site to the saddle of RBD encircled Lys417, 20 Glu484 and Tyr453 inside the REGN10933 epitope (violet), explaining its sensitivity to the K417N, Y453F and E484K mutations.
105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 (F) Comparison of binding modes of A23-58.1 and LY-CoV555. One Fab is shown to bind to the RBD on the spike. Glu484 is located in the middle of LY-CoV555 epitope (light orange), explaining its sensitivity to the E484K mutation. 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 (D) Negative stain 3D reconstruction of the ternary complex of spike with Fab B1-182.1 and A19-46.1 (left) or A19-61.1 (right).
(E) rcVSV SARS-CoV-2 was incubated with increasing concentrations (1.3e-4 to 50 µg/mL) of either single antibodies (A19-46.1, A19-61.1 and B1-182.1) and combinations of antibodies (B1-182.1/A19-46.1 and B1-182.1/A19-61.1). Every 3 days, wells were assessed for CPE and the 5 highest concentration well with the >20% CPE was passaged forward onto fresh cells and antibody containing media. Shown is the maximum concentration with >20% CPE for each of the test conditions in each round of selection. Once 50 µg/mL has been reached, virus was no longer passaged forward and a dashed line is used to indicate maximum antibody concentration was reached in subsequent rounds. 10 105 and is also made available for use under a CC0 license.
(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC The copyright holder for this preprint this version posted March 1, 2021. ; https://doi.org /10.1101/10. /2021 (Subjects 1-4). Shown is the staining for RBD-SD1 BV421, S1 BV786 and S-2P APC or Ax647. Cells were sorted using indicated sorting gate (pink) and percent positive cells that were either RBD-SD1, S1 or S-2P positive is shown for each subject. (B) Gross binding epitope distribution was determined using an MSD-based ELISA testing against RBD, NTD, S1, S-2P or HexaPro. S2 binding was inferred by S-2P or HexaPro binding without binding to other antigens. Indeterminant epitopes showed a mixed binding profile. Total number of antibodies (i.e., 200) and absolute number of antibodies within each group is shown. (F) Biolayer interferometry-based epitope binning experiment. Competitor antibody (y-axis) is bound to S-2P prior to incubation with the analyte antibody or ACE2 protein (x-axis) as indicated and percent competition range bins are shown as red (>=75%), orange (60-75%) or white <60%) (n=2). mAb114 is an anti-Ebola glycoprotein antibody and is included as a negative control (37) (G) Negative stain 3D reconstructions of SARS-CoV-2 spike and Fab complexes. A19-46.1 and A19-61.1 bind to RBD in the down position while A23-58.1 and B1-182.1 bind to RBD in the up position. Representative classes were shown with 2 Fabs bound, though stoichiometry at 1 to 3 were observed.     Y144  A222V  A222V  R246I  R246I  K417N  K417N  K417N  K417N  K417N  N439K  N439K  Y453F  Y453F  E484K  E484K  E484K  E484K  E484K  N501Y  N501Y  N501Y  N501Y  N501Y  N501Y  N501Y  A570D  A570D  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  D614G  P681H  P681H  T716I/S982A/D1118H  S2 x 3  A701V  A701V  A701V RBD S1-C-term  Fig. 3. Structural basis of A23-58.1 binding and neutralization. (A) Cryo-EM structure of A23-58.1 Fab in complex with SARS-CoV-2 HexaPro spike. Overall density map is shown to the left with protomers colored light green, gray and cyan. One of the A23-58.1 Fab bound to the RBD is shown in orange and blue. Structure of the RBD and A23-58.1 after local focused refinement was shown to the right. The heavy chain CDRs are colored brown, salmon and orange for CDR H1, CDR H2 and CDR H3, respectively. The light chain CDRs are colored marine blue, light blue and purple blue for CDR L1, CDR L2 and CDR L3, respectively. The contour level of Cryo-EM map is 5.7s. (B) Interaction between A23-58.1 and RBD. All CDRs were involved in binding of RBD. Epitope of A23-58.1 is shown in bright green surface with a yellow border (left, viewing from antibody to RBD). RBD mutations in current circulating SARS-CoV-2 variants are colored red. Lys417 and Glu484 are located at the edge of the epitope. The tip of the RBD binds to a cavity formed by the CDRs (right, viewing down to the cavity). Interactions between aromatic/hydrophobic residues are prominent at the lower part of the cavity. Hydrogen bonds at the rim of the cavity are marked with dashed lines. RBD residues were labeled with italicized font. (C) Paratope of A23-58.1. Sequences of B1-182.1 and S2E12 were aligned with variant residues underlined. Paratope residues for A23-58.1 and S2E12 were highlighted in green and light brown, respectively. (D) Epitope of A23-58.1 on RBD. Epitope residues for different RDB-targeting antibodies are marked with * under the RBD sequence. (E) Comparison of binding modes of A23-58.1 and REGN10933. One Fab is shown to bind to the RBD on the spike. The shift of the binding site to the saddle of RBD encircled Lys417, Glu484 and Tyr453 inside the REGN10933 epitope (violet), explaining its sensitivity to the K417N, Y453F and E484K mutations. (F) Comparison of binding modes of A23-58.1 and LY-CoV555. One Fab is shown to bind to the RBD on the spike. Glu484 is located in the middle of LY-CoV555 epitope (light orange), explaining its sensitivity to the E484K mutation.   Fig. 4. Critical binding residue determination and mitigation of escape risk using dual antibody combinations (A) The indicated Spike protein mutations predicted by structural analysis were expressed on the surface of HEK293T cells and binding to the indicated antibody was measured using flow cytometry. Data is shown as Mean Fluorescence Intensity (MFI) normalized to the MFI for the same antibody against the WA-1 parental binding. Percent change is indicated by a color gradient from red (increased binding, Max 250%) to white (no change, 100%) to blue (no binding, 0%).
(C) Replication competent vesicular stomatitis virus (rcVSV) whose genome expressed SARS-CoV-2 WA-1 was incubated with serial dilutions of the indicated antibodies and wells with cytopathic effect (CPE) were passaged forward into subsequent rounds ( Figure S6) after 48-72 hours. Total supernatant RNA was harvested and viral genomes shotgun sequenced to determine the frequency of amino acid changes. Shown are the spike protein amino acid/position change and frequency as a logo plot.
(E) rcVSV SARS-CoV-2 was incubated with increasing concentrations (1.3e-4 to 50 μg/mL) of either single antibodies (A19-46.1, A19-61.1 and B1-182.1) and combinations of antibodies (B1-182.1/A19-46.1 and B1-182.1/A19-61.1). Every 3 days, wells were assessed for CPE and the highest concentration well with the >20% CPE was passaged forward onto fresh cells and antibody containing media. Shown is the maximum concentration with >20% CPE for each of the test conditions in each round of selection. Once 50 μg/mL has been reached, virus was no longer passaged forward and a dashed line is used to indicate maximum antibody concentration was reached in subsequent rounds.