The antigenic anatomy of SARS-CoV-2 receptor binding domain

Abstract

Antibodies are crucial to immune protection against SARS-CoV-2, with some in emergency use as therapeutics. Here, we identify 377 human monoclonal antibodies (mAbs) recognizing the virus spike and focus mainly on 80 that bind the receptor binding domain (RBD). We devise a competition data-driven method to map RBD binding sites. We find that although antibody binding sites are widely dispersed, neutralizing antibody binding is focused, with nearly all highly inhibitory mAbs (IC₅₀ < 0.1 μg/mL) blocking receptor interaction, except for one that binds a unique epitope in the N-terminal domain. Many of these neutralizing mAbs use public V-genes and are close to germline. We dissect the structural basis of recognition for this large panel of antibodies through X-ray crystallography and cryoelectron microscopy of 19 Fab-antigen structures. We find novel binding modes for some potently inhibitory antibodies and demonstrate that strongly neutralizing mAbs protect, prophylactically or therapeutically, in animal models.

Keywords: coronavirus, SARS-CoV-2, anti-RBD antibody, receptor binding domain, antibody, immune responses, virus structure.

Graphical abstract

Figure S1

SARS-CoV-2 elicits binding and neutralizing antibodies against trimeric spike, RBD, and NP proteins, related to Figure 1 (A) Plasma from donors with confirmed SARS-CoV-2 infection were collected at 1-2 months after onset of symptoms and tested for binding to SARS-CoV-2 spike, RBD and N proteins by capture ELISA. (B) Neutralizing titers to authentic live virus. Data are representative of one experiment with 42 samples and presented as means ± s.e.m. (C) Comparison of the frequency of spike-reactive IgG expressing B cells in mild cases and severe cases measured by FACS. Small horizontal lines indicate the median. Data are representative of one experiment with 16 samples. The Mann–Whitney U test was used for the analysis and two-tailed P values were calculated (in B and C).

Figure S2

SARS-CoV-2 antibody isolation strategies, related to Figure 1 Human monoclonal antibodies from memory B cells were generated using two different strategies. (A) IgG expressing B cells were isolated and cultured with IL-2, IL-21 and 3T3-msCD40L cells for 13-14 days. Supernatants were harvested and tested for reactivity to spike protein by ELISA. (B) Antigen-specific single B cells were isolated using labeled recombinant spike or RBD proteins as baits. The IgG heavy and light chain variable genes from both strategies were amplified by nested PCR and cloned into expression vectors to produce full-length IgG1 antibodies.

Figure S3

Specificity and sequence analysis of 377 human antibodies, related to Figures 1, 2, and 5 (A) Epitope mapping of SARS-CoV-2 -specific antibodies against the RBD, S1 subunit (aa 16–685) and S2 subunit (aa 686-1213) were evaluated by ELISA, and the NTD-binders were identified by cell-based fluorescent immunoassay. Antibodies interacting with none of the subdomains were defined as trimeric spike. The number in the centers indicate the total number of tested antibodies. (B) Frequency of amino acid substitutions from germline in SARS-CoV2-specific heavy and light chains (n = 377). (C) Repertoire analysis of antibody heavy and light chains of anti-S (Non-RBD) and anti-RBD antibodies. At the center is the number of antibodies. Each slice represents a distinct clone and is proportional to the clone size. (D) Frequency of amino acid substitutions from germline in heavy and light chains of antibodies cross-reacting between SARS-CoV-2 and the 4 seasonal coronaviruses (n = 20).

Figure 1

Characterization of SARS-CoV-2-specific mAbs (A) Cross-reactivity of 299 anti-spike (non-RBD) and 78 anti-RBD antibodies to trimeric spike of human alpha- and beta-coronaviruses by capture ELISA. (B) Comparison of neutralization potencies (IC₅₀) between anti-spike (non-RBD) and anti-RBD antibodies against authentic SARS-CoV-2 using focus reduction neutralization test (FRNT). The Mann-Whitney U test was used for the analysis and two-tailed p values were calculated. (C) Correlation between SARS-CoV-2 neutralization and RBD:ACE2 blocking by anti-RBD antibodies. Antibodies with IC₅₀ <0.1 μg/mL, 0.1–1 μg/mL, and 1–10 μg/mL are highlighted in red, blue, and orange, respectively. (D) Plasma was depleted of RBD-specific antibodies using Ni-NTA beads coated with or without RBD, then evaluated for SARS-CoV-2 neutralizing activity by FRNT assay (n = 8). Results are expressed as percent neutralization of control without plasma. The percentage of depletion of neutralizing antibodies for each sample tested is indicated at the top of each panel. See also Figures S1, S2, and S3.

Figure 2

RBD anatomy and epitope definition based on mapping results (A) Pale gray RBD surface with cartoon depiction of one monomer rainbow colored from blue (N terminus) to red (C terminus) alongside gray surface depiction of RBD labeled to correspond to the adjacent torso (Torso Gaddi, Wikipedia, CC BY-SA 3.0, modified in Adobe Photoshop) used by analogy to enable definition of epitopes. (B) Cluster maps showing the output of the mapping algorithm with each spot corresponding to a “located” antibody and color-coded according to epitope. (C) BLI antibody data competition matrix (calculated values) output from cluster analysis showing the clustering into 5 epitopes. (D) RBD (gray)-ACE2 (purple) complex (PDB: 6M0J, (Lan et al., 2020)). RBD residues contacting ACE2 are shown in green. (E) Located antibodies mapped onto the RBD shown as a gray surface with the ACE2-binding site in green. The individual antibodies are depicted as spheres and color coded as in (B), those central to this paper are labeled. (F) As for (E), but antibodies are color-coded according to their ability to neutralize. See inset scale: red, strongest neutralizers; blue, weakest neutralizers. See also Figure S3.

Figure 3

RBD complexes The Fab-RBD complexes reported in this paper as determined by a combination of X-ray crystallography with the exception of Fab 40 for which the Fab-RBD has been excised from a cryo-EM structure of Fab 40 bound to the S protein. (A) Shows the front view and (B) shows the back view with the RBD surface shown in gray and Fabs drawn as cartoons with the heavy chain in red and the light chain in blue. The ACE2 footprint on the RBD is colored in green. See also Figure S5.

Figure 4

Spike morphology and Fab binding (A) Orthogonal views of the trimeric spike as a pale gray surface with one monomer depicted as a cartoon and rainbow colored from the N to the C terminus (blue to red). (B) Surface depiction of the electron potential map for the Spike-mAb 159 complex determined by cryo-EM to 4.1 Å resolution. The Spike is shown tilted forward and colored in teal apart from the RBDs (gray), and the fragment of mAb 159 that can be visualized is shown in orange. (C) Gray surface depiction of the RBD with a blue sphere denoting the location of Fab 45 as predicted using the mapping algorithm reported here. (D) Gray surface depiction of the RBD of the X-ray crystallographic structure of the observed RBD-Fab 45 complex. Fab 45 binds close to the predicted position but is slightly translated. The S309 Fab (the closest structure in the competition matrix on which the mapping algorithm was based) is shown superimposed. Both Fabs are depicted as a cartoon with the heavy chain in magenta and light chain in blue. (E) Orthogonal gray surface depictions of the RBD with Fab 384 bound and Fab CV07-270 superimposed onto the complex. These Fabs use the same heavy-chain V-gene but bind differently. They are drawn as cartoons with the heavy and light chains for Fab 384 in magenta and blue and those for CV07-270 in pale pink and light blue, respectively. See also Figures S4 and S5.

Figure S4

Crystal structures of the ternary complexes and overrepresentation of binding modes, related to Figures 4, 5, 6, and 7 (A) RBD-88-45, (B) RBD-253-75, (C) RBD-253H55L-75 and (D) RBD-384-S309 complexes. (E) Sequence alignment for HC CDR3s using public V-region 3-53, antibodies are represented by a number (from this study) or by PBD code and a name. (F) Comparison of binding modes of 150 (orange), 158 (cyan), 269 (magenta). (G) Superimposition of RBD-Fab complexes available in PDB (up to 21^st Oct. 2020). RBD is shown as gray surface, Febs as Cα traces with heavy chains in warm color and light chains in cool color. (H) The bound Fabs can be divided into four major clusters, neck (B38(7bZ5), CB6(7C01), CV30(6XE1), CC12.3(6XC4), CC12.1(6XC3), COV2-04(7JMO), BD629(7CHC), BD604(7CH4), BD236(7CHB)), left shoulder (p2b-2f6(7BWJ), BD368(7CHC), C07-270(6XKP)), left flank (EY6A(6ZCZ), CR3022(6YLA), S304(7JX3), COVA1-16(7JMW)) and right flank (S309 (7JX3)), according to their binding modes on RBD. (I) Outliers that include right shoulder binders (REGN10987 (6XDG), COVA2-39 (7JMP), CV07-250 (6XKQ), S2H14 (7JX3)). One Fab in the neck cluster is drawn as red and blue surface to show the relative position of the outliers.

Figure 5

Determinants of binding, CDR length (A) Fab 384 interaction: left panel overview of the interacting CDRs from the heavy chain (magenta) and light chain (cyan) with the RBD (gray surface). The interactions of the H3, H2, and L1 and L3 loops are shown in the adjacent panels. (B) The distribution of IGHV, IGKV, and IGLV gene usage of anti-RBD antibodies. Antibodies are grouped and colored according to their neutralization IC₅₀ values. (C) Left panel overview of the CDR interactions for Fabs 150 (magenta), 158 (cyan), and 269 (orange). Adjacent panels (top) show a close-up of the H3 loop interactions for each of these antibodies retaining the same color coding, and the bottom panel shows the interactions of the L3 loop and also the sequence alignment for the loops. (D) Back and side views of the complex of Fab 40 and RBD (gray surface) with the Fab drawn as a cartoon with the heavy chain in magenta and the light chain in blue. Fab 158 (gray cartoon) is superimposed. Note despite Fab 40 using the IGVH3-66 public V-gene, whereas 158 uses IGVH3-53 they bind almost identically. (E) Fab 75-RBD complex with the RBD drawn as a cartoon in magenta and the Fab similarly depicted with the heavy chain in orange and the light chain in gray. This antibody uses IGHV3-30 and is not a potent neutralizer. It can be seen that the only heavy-chain contact is via the extended H3 loop. See also Figures S3 and S4.

Figure 6

Determinants of binding, light-chain swapping, and glycosylation (A) Table of sequences of mAbs 253, 55, and 165. (B) Neutralization activity of authentic SARS-CoV-2 by the original mAb253, chimeric mAb253H55L, and chimeric 253H165L (presented as IC₅₀ values). Immunoglobulin heavy- and light-chain gene alleles are presented in the table. Data are from 3 independent experiments, each with duplicate wells and the data are shown as mean ± SEM. (C) The chimeric Fab 253H55L (mAb 253 [IGVH1-58] heavy chain combined with the light chain of mAb 55 [IGVK3-20]) in complex with the RBD here shown as a hydrophobic surface. The Fab is drawn as a ribbon with the heavy chain in magenta and the light chain in blue. This 10-fold increase in neutralization titer of this Fab compared to 253 appears to come from the single substitution of a tryptophan for a tyrosine making a stabilizing hydrophobic interaction. (D) CDRs with sugar bound in the RBD complexes with Fabs 88 (top panel) sugar bound to N35 in the H1 loop, 316 (middle panel) sugar bound to N59 in the H2 loop, and 253 (bottom panel) sugar bound to N102 in the H3 loop. Note that Phe 486 is marked by a diamond to enable the various orientations to be related. See also Figures S4 and S6.

Figure S6

Importance of antibody glycosylation, related to Figure 6 (A-C) Effect of mutation of the Asn residue glycosylated in the heavy chains of antibodies 88, 253 and 316 respectively. (D-F) |2Fo-Fc| electron density maps contoured at 1.2 σ showing the glycans at glycosylation sites at N35 of 88 (D), N59 of 316 (E) and N102 of 253 (F). (G) Relative binding position and orientation of CDR-H3 and glycans between 316 (green) and 88 (orange), and (H) between 316 and 253 (cyan). RBD is shown as a gray surface.

Figure 7

Determinants of binding, valency of interaction, and in vivo studies (A) Cryo-EM Spike-Fab complexes showing different RBD conformations. The density for the Spike is shown in teal, the RBD in gray, and Fab in orange. Left: “all RBDs down” conformation with Fab 316 bound. Middle: “one RBD up” conformation with one Fab 158 bound. Right: “all RBDs up” conformation with 3 Fab 88s bound. (B) Left: potently neutralizing Fab 159 (cartoon representation with red heavy chain and blue light chain) in complex with the NTD (gray transparent surface). Right: 159 is depicted with another NTD binding Fab (4A8) superimposed as a gray ribbon, the binding sites are separated by ~15 Å. (C) Fab 159 (magenta, HC; blue, LC) is drawn as a cartoon in its binding location on top of the NTD of the Spike that is drawn as a gray surface and viewed from the top (a full IgG is modeled onto one monomer showing that it cannot reach across to bind bivalently). (D) ELISA binding (blue) and FRNT neutralization (red) curves of ten full-length antibodies (solid lines) and corresponding Fab molecule (dash lines) against SARS-CoV-2. Data are from 2 independent experiments (mean ± SEM). (E–K) Seven- to 8-week-old male and female K18-hACE2 transgenic mice were inoculated by an intranasal route with 10³ PFU of SARS-CoV-2. At 1 dpi, mice were given a single 250 μg (10 mg/kg) dose of the indicated mAb by intraperitoneal injection. (E) Weight change (mean ± SEM; n = 5–10, two independent experiments: two-way ANOVA with Sidak’s post-test: ns, not significant, ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗∗p < 0.0001; comparison to the isotype control mAb treated group). (F–K) At 7 dpi tissues were harvested and viral burden was determined in the lung (F and G), heart (H), spleen (I), nasal washes (J), and brain (K) by plaque (F) or qRT-PCR (G–K) assay (n = 7–11 mice per group; Kruskal-Wallis test with Dunn’s post-test: ns, not significant, ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, ^∗∗∗∗p < 0.0001). Dotted lines indicate the limit of detection. See also Figures S4, S5, and S7.

Figure S5

Cryo-EM data, related to Figures 3, 4, and 7 Resolution and map quality at the RBD-Fab/IgG interface. (A-K) [left] Gold-standard FSC curve (FSC = 0.143 marked) generated by cryoSPARC for fab (or IgG in the case of 159)-spike structures, [right] showing map quality at the antigen/antibody interface with 40, 88, 150, 158, 316, 384, 253H55L RBD up, 253H55L RBD down, 253H165L, 159 RBD down, 159 RBD up, respectively. Classification of Cryo-EM datasets shows Spike heterogeneity for 384 and 159. (L) Gaussian filtered reconstructed volume (transparent gray) with refined spike (from two clusters of 384 following local variability analysis using cryoSPARC). At very low contour levels, and with Gaussian filtering, we are able to see slight evidence of one (right), or two (left) additional bound fabs. (M) Reconstructed volume for 159 in the RBD up (left) and down (right) positions, colored by spike chain (blue, green, purple) and IgG (orange). The RBD in the up position is indicated by a red arrow.

Figure S7

Prophylaxis with mAbs 40 and 88 protects against weight loss and decreases viral burden, related to Figure 7 (A-G) Seven to eight-week-old male and female K18-hACE2 transgenic mice were given a single 250 μ dose of the indicated mAbs by intraperitoneal injection. One day later mice were inoculated by intranasal route with 10³ PFU of SARS-CoV-2. (A) Weight change (mean ± s.e.m; n = 6, two independent experiments: two-way ANOVA with Sidak’s post test: ns, not significant, ^∗p < 0.05, ^∗∗∗∗p < 0.0001; comparison is to the isotype control mAb treated group). B-G. At 7dpi tissues were harvested and viral burden was determined in the lung (B-C), heart (D), spleen (E), nasal washes (F) and brain (G) by plaque assay (B) or RT-qPCR (C-G) assay (n = 6 mice per group. Kruskal-Wallis test with Dunn’s post-test: ns, not significant, ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001). Dotted lines indicate the limit of detection.