Isolation and characterization of RBD-specific NAbs from COVID-19 convalescent donors.
To elucidate the degree to which SARS-CoV-2 will adapt to evade NAbs, we first generated RBD-specific NAbs by screening a cohort of 36 convalescent patients infected with early-circulating SARS-CoV-2 strains during January-March 2020 and selected four donors with high titres of plasma RBD-binding antibodies and NAbs against a SARS-CoV-2 pseudovirus (Extended Data Fig. 1). We then conducted single-cell PCR experiments to generate human monoclonal antibodies (mAbs) from memory B cells using the SARS-CoV-2 RBD as the bait. Ninety-three RBD-specific antibodies were identified (Fig. 1a and Supplementary Table 1). More than 95% of the RBD-specific mAbs did not bind to the denatured form of RBD, indicating that the epitopes targeted by RBD-specific antibodies induced by natural infection are highly dependent on conformation. Furthermore, approximately 12-24% of isolated mAbs from a given donor showed cross-reactivity with SARS-CoV; these findings suggested that cross-reactive mAbs could be induced by natural infection, as the SARS-CoV-2 and SARS-CoV RBDs share 76% amino acid identity20.
Remarkably, 50-72% of the mAbs from each donor exhibited detectable pseudovirus SARS-CoV-2 neutralization (IC50 <5µg/ml), with 12-48% design-ated high-potency mAbs (IC50 ≤50ng/ml) (Fig. 1b and Supplementary Table 1). Expectedly, 25-64% of the mAbs from each donor had measurable receptor-blocking activity (IC50 <5 µg/ml), suggesting that most of the RBD-specific NAbs protect against virus infection via mechanisms that block attachment to the cellular receptor ACE2. More importantly, three antibodies potently neutralized the SARS-CoV-2 pseudovirus with an IC50 of 14-25 ng/ml and moderately neutralized the SARS-CoV pseudovirus with an IC50 of 0.9-5 μg/ml. They efficiently blocked binding of the SARS-CoV-2 and SARS-CoV RBDs to ACE2 (Extended Data Fig. 2a-c and Table 1), unlike a previously described broadly reactive NAb, S309, which was independent of receptor binding inhibition22. As shown in Fig. 1d and Supplementary Table 2, the NAbs were nearly unrestricted in the germline gene repertoire. Among these heavy chains, 73% originated from IGHV3 and IGHV4. We also observed that RBD-binding NAbs were strongly biased towards IGHV3-53/3-66, consistent with the findings of previous studies4,23,24, and suggest that they play an important role in the humoral immune response to SARS-CoV-2 infection25.
Finally, we verified that the 19 selected mAbs could efficiently neutralize authentic SARS-CoV-2 infection with IC50 values from 20 ng/ml to 1.8 μg/ml, and we found that several NAbs exhibited neutralizing activity comparable to that of CB6, which is in clinical26. The 19 NAbs were used as probes to search for the binding determinants of RBD-specific NAbs due to their high affinity for the SARS-CoV-2 RBD and potent neutralization of SARS-CoV-2 (Table 1 and Supplementary Fig. 1).
Mapping and characterization of protective antigenic sites and antibody binding hot spots on RBD.
To define the epitopes recognized by selected RBD-specific NAbs, we first performed competitive binding experiments. Three well-described mAbs targeting independent epitopes, CB6, CR3022 and S309, were used as controls22,26,27. Our panel of NAbs could be classified into 5 groups (Table 2). The group 1 and group 2 antibodies competed with CB6, while the group 4 antibodies competed with S309. The three cross-reactive antibodies in group 5 competed with CR3022. The antibodies in group 2 and group 3 may have larger footprints than those in the other groups because they competed with antibodies from two of the other groups. Interestingly, 4/5 of the group 1 antibodies utilized VH3-53/3-66 and had short CDRH3 lengths of 9-13 amino acids (Extended Data Table 1).
To obtain a comprehensive view of the antigenic sites on the SARS-CoV-2 RBD and the determinants of RBD NAb recognition, we performed global RBD alanine scanning mutagenesis (at nearly 190 RBD amino acid positions) with a panel of 17 mAbs derived from the five groups. Functional epitope mapping identified 33 binding determinants for NAbs and defined four major antigenic sites (1-4) targeted by RBD-specific NAbs based on their structural locations and epitope competition results. Site 1 and Site 2 overlap with ACE2 binding sites, while Site 3 and Site 4 are located outside the sites (Fig. 2a, b and Supplementary Table 3). Some antibodies bind to only one site, whereas others contact more.
The group 1 antibodies mainly bound to site 1. We demonstrated that some mutations, namely, N417A, F456A, N460A, A475V, F486A, and N487A, led to less binding for multiple group 1 antibodies. To verify whether these residues were crucial for the dominant VH3-53/3-66 germline antibodies, we selected eleven VH3-53/3-66 germline antibodies, including CB6 and B38 antibodies whose structures have previously been characterized26,28, as probes (Extended Data Fig. 3a). We demonstrated that most of the VH3-53/3-66 NAb prototypes were sensitive to mutations at F456 and N487 in different extent,while N417, N460, A475 and F486 were also involved in the contact of some VH3-53/3-66 NAbs (Fig. 2c). The heavy chains of CB6 and B38 use a similar structural mode for epitope recognition (Extended Data Fig. 3b). The conserved germline-encoded CDRH1 and CDRH2 together with the distinct CDRH3 contribute to tight contact with the core epitope on the RBD formed by N417, F456, N460, A475 and N487 residues within antigenic site 1, suggesting that mutations at these positions may give rise to resistance to VH3-53/3-66-prototype NAbs. With regard to the site 2 binding region, we observed that substitutions at shared positions (N450, L452, E484 and F490) reduced the binding of NAbs in groups 2 and 3 to the RBD. In addition, R346 within antigenic site 3 showed favourable interactions with three mAbs from group 3 and group 4. Antibodies in group 5 distinctly preferentially bound to the residues within site 4 (Fig. 2a). We further investigated whether the identified residues involved in antibody binding also influence ACE2 binding (Fig. 2a). As expected, substantial loss of ACE2 binding was caused by mutations in surface residues, namely, F456A, F486A, N487A, Y505A and Y449A, each of which is involved in ACE2 binding according to structural analysis5,29. Two core region mutations distant from the ACE2 binding surface, N343A and W436A, also resulted in loss of ACE2 binding.
We analyzed the relationship between neutralization potency and each antigenic site (Fig. 2d and Supplementary Table 4). Over 70% of the highly potent NAbs targeted antigenic sites 1 and 2; thus, antigenic sites 1 and 2 are the prime targets of SARS-CoV-2 neutralizing antibodies. Furthermore, SARS-CoV-2 and SARS-CoV cross-reactive mAbs mainly targeted antigenic sites 3 and 4 (Fig. 2e), indicating that these two sites are conserved exposed sites, consistent with the findings of previous studies22,27. Taken together, the results have implications for the design of SARS-CoV-2 vaccines, and the binding hot spots of RBD-specific NAbs identified here will support direct and intentional monitoring of immune escape mutants.
The residues essential for RBD folding and antigen conformation are evolutionarily conserved among sarbecoviruses.
Our landscape of mapping data also demonstrated that mutations at approximately 20% of the positions (38 of 190 positions) led to substantial loss of binding for nearly all NAbs as well as recombinant hACE2 (Extended Data Table 2). Most of these residues were buried within the RBD core structure, thus were likely to facilitate RBD folding. Our data also revealed the importance of disulfide bonds in the RBD. (Fig. 3a). Furthermore, these residues were highly conserved across clade 1, 2 and 3 sarbecoviruses30,31, including human and animal isolates (Fig. 3b). The fact that the unchanged residues are so evolutionarily conserved means that they play vital roles and that their preservation is necessary for virus survival. Additionally, point mutations can strongly affect protein stability, which may in turn affect protein function, as illustrated by studies on other viruses19,32. The top 17 destabilizing mutations predicted by the two structure-based methods MAESTRO33 and DUET34 showed high free energy change (ΔΔG) values and low average antibody binding percentages. Mutations resulting in improper RBD folding should be considered in determining functional epitope of antibodies by alanine scanning. Based on structural analysis of the RBD-CB6 complex and the RBD-B38 complex, the conserved Y421 residue is part of the epitope of these complexes because it forms hydrogen bonds with G54 in the CDRH2 of CB6 and B38 (Extended Data Fig. 3b). Collectively, the data suggest that some residues may have low mutational tolerance, so targeting these positions with antibodies could limit viral escape.
Natural substitutions of antibody binding hot spots confer resistance to multiple NAbs.
To investigate the residue polymorphism of each antigenic site, we computed the sequence variability at positions that were binding determinants for selected mAbs (Extended Data Fig. 4a, b). The data showed that site 1 and site 2 were more polymorphic than site 3 and site 4. Mutations were more frequently introduced in the positions with variable sequences between SARS-CoV-2 and SARS-CoV such that multiple sites were replaced by the same residues or by residues with similar biophysical properties at the corresponding positions of SARS-CoV RBD. Additionally, residues at positions 417, 475, 484, 452, 490 and 346, which are key recognition sites for multiple NAbs, were highly polymorphic. In contrast, some conserved residues that were proven to be critical for ACE2 binding by alanine scanning, such as N487, Y505, Y449, W436 and N343 (Fig. 2a), had limited variability, suggesting that these residues have a low inherent tolerance for mutations.
To explore the impacts of natural mutations on NAb binding, we constructed and expressed RBDs with single-amino acid substitutions that are present in circulating human isolates of SARS-CoV-2. As shown in Fig. 4a and Supplementary Table 5, for residues at which several alternate amino acids with different side chains were selected, the different substitutions did not contribute equally to NAb binding. For example, alanine scanning revealed that the F456A mutation caused loss of binding of VH3-53/3-66 NAbs, but the natural F456L variation did not result in resistance to VH3-53/3-66 NAbs. We also observed that L452R rather than L452M led to substantial loss of binding to 24-34L, 25-F8, 24-12K, and 28-15L. E484A and E484K rather than E484D resulted in resistance to 26-34L, 24-12K and 25-F8. F490V, F490S, and F490L each resulted in strong resistance to 26-34L, 25-F8 and 24-12K. P337R instead of P337S conferred resistance to S309, and R346S instead of R346T caused significant loss of binding of 24-12K, 28-15L and 25-C9. V382E rather than V382L reduced the binding activity of 28-26K and CR3022 in group 5. Collectively, the data suggest that different properties of amino acid substitutions, including hydrophobicity, polarity and charge, might determine resistance in terms of requirements for interactions with mAbs.
We further purified 17 mammalian cells expressed RBD single-point mutants and measured their binding activity with a panel of NAbs from the five groups (Fig. 4b). The fold changes in the EC50 values compared to those of the wild-type RBD were investigated, and the results were consistent with those of our preliminary mutational scan. Moreover, the molecular mechanisms of the effects of the mutations on three well-characterized mAbs, CB6, S309 and CR3022, were well explained by the structures (Fig. 4c-e). For example, replacement of K417 with Asn (N) greatly weakened CB6 binding affinity by disrupting a strong salt bridge between K417 in the SARS-CoV-2 RBD and the CB6 CDRH3. However, E340K and K378N disrupted the key hydrogen bonds with S309 and CR3022 respectively.
To further investigate whether the binding escape mutants exhibited NAb resistance, we constructed a panel of 17 SARS-CoV-2 pseudovirus variants, including 16 single RBD mutants, to examine their impacts on the neutralization potency of the 12 NAbs mentioned above (Fig. 4f). Since the dominant S sequence variant seen in clinical isolates is D614G, all the SARS-CoV-2 pseudovirus variants we constructed were coupled with the D614G variant35. As expected, in agreement with the EC50 value results (in which the RBD substitutions resulted in high EC50 values for NAbs), the pseudovirus variants correspondingly conferred resistance to NAbs with high IC50 values. Notably, the most frequent RBD variants seen in clinical isolates, N501Y and S477N, remained similarly sensitive to the majority of the selected NAbs; only 24-1L failed to neutralize N501Y. Substitutions responsible for major antigenic escape were in antigenic site 1 (K417N, F486L), antigenic site 2 (N450K, E484K, L452R, F490S) and antigenic site 3 (R346S). Using our immune escape mapping strategy, we identified a natural mutant, E340K, in the circulating virus that conferred resistance to a broadly reactive NAb, S309, and five mutants that resulted in resistance to CB6. These findings could inform the therapeutic use of these antibodies in clinical studies. Expectedly, the South African variant (RBD-K417N/E484K/N501Y) facilitated resistance to a somewhat wider range of NAbs than single mutations, which conferred complete resistance to five highly potent NAbs targeting two major antigenic clusters 1 and 2. However, the site 3-targeting antibodies 25-C9 and S309 and the site 4-targeting antibody 28-26K retained their ability to neutralize the B.1.351 pseudovirus.
Impacts of RBD mutations on ACE2 binding affinity.
To investigate how the antigenic escape residues identified by our study affect the RBD-ACE2 interaction, the binding affinities of eighteen mutated RBDs to ACE2 were analysed with BIAcore 8K (Supplementary Fig. 2). As shown in Fig. 5a, the majority of the mutants retained or even exhibited enhanced hACE2 binding. N501Y, L452R and S477N mutants exhibited high affinity for ACE2 and exhibited 9.24- to 14.66-fold higher binding affinity than wild-type RBD. These data provide a reasonable explanation for the high frequencies of the three mutations in clinical sequencing data (Extended data Fig. 4). Unlike N501Y, which induced tighter binding with ACE236, S477N and L452R occurred at sites that were likely not in the ACE2 contact region (Fig. 5b). It is possible that the mutations altered the charge within the flexible loop region of the RBM, creating a more favourable environment for binding. Notably, the key antibody escape mutations K417N, N450K, E484K, F490S and R346S had limited effects on ACE2 binding affinity with fold changes between 0.4 and 2.5, suggesting that they were not accompanied by loss of fitness. Despite of the strengthened binding to ACE2 caused by the substitution N501Y, the affinity of the South African B.1.351 variant for ACE2 was only 2.8-fold higher than that of the wild-type RBD, as it was balanced out by two additional mutations (K417N and E484K). Overall, our current data are likely to be useful for understanding the evolutionary mechanism that governs the emergence of viral escape mutants.
The key natural mutations were able to escape neutralization by COVID-19 convalescent donors.
To examine whether pseudoviruses with the key antigenic escape mutations conferred resistance to convalescent plasma from the first wave of SARS-CoV-2 infection in early 2020, nine out of 36 convalescent plasma samples (2, 6, 23, 24, 25, 26, 27, 28 and 32) were selected, and they exhibited different degrees of S-binding, receptor-blocking and neutralizing activity (Extended Data Table 3). The resistance profile of each human convalescent plasma sample was distinct, possibly because of the different repertoires of antigenic sites on the RBD targeted by polyclonal antibodies (Fig. 6a, b). Markedly, K417N, F486L, L452R, E484K and R346S resulted in resistance to at least six plasma samples, as the NAb titres were approximately 2-4 times lower than those for the wild type, indicating that NAbs targeting these key residues were enriched in human convalescent plasma. On the other hand, both F490S and N450K resulted in resistance to neutralization by plasma samples 6 and 24; in particular, the NAb titre of plasma sample 24 against F490S was reduced by 4.6 times. N501Y and S477N, two of the most abundant RBD mutations, provide resistance to a very limited number of human convalescent plasma, indicating that they may not pose a threat to human immune protection against natural infection, in agreement with the conclusions of previous studies10,37,38. Finally, we demonstrated that compared with the wild-type residues, the N501Y, K417N, and E484K mutations in the B.1.351-variant pseudovirus dramatically reduced the neutralizing ability of all the plasma samples, with 2.1- to 7.4 -fold reductions in NAb titres, which are consistent with previous studies15. In addition, this combination of mutations resulted in more resistance than single mutations. Overall, these data suggest that circulating viruses with single mutations at antigen binding hot spots could be resistant to neutralization by human convalescent plasma but that no single amino acid mutation can enable robust escape. The evolution of co-mutations at distinct major antigenic sites is worthy of considerable attention.