BA.2.12.1, BA.4 and BA.5 escape antibodies elicited by Omicron infection

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) Omicron sublineages BA.2.12.1, BA.4 and BA.5 exhibit higher transmissibility than the BA.2 lineage1. The receptor binding and immune-evasion capability of these recently emerged variants require immediate investigation. Here, coupled with structural comparisons of the spike proteins, we show that BA.2.12.1, BA.4 and BA.5 (BA.4 and BA.5 are hereafter referred collectively to as BA.4/BA.5) exhibit similar binding affinities to BA.2 for the angiotensin-converting enzyme 2 (ACE2) receptor. Of note, BA.2.12.1 and BA.4/BA.5 display increased evasion of neutralizing antibodies compared with BA.2 against plasma from triple-vaccinated individuals or from individuals who developed a BA.1 infection after vaccination. To delineate the underlying antibody-evasion mechanism, we determined the escape mutation profiles2, epitope distribution3 and Omicron-neutralization efficiency of 1,640 neutralizing antibodies directed against the receptor-binding domain of the viral spike protein, including 614 antibodies isolated from people who had recovered from BA.1 infection. BA.1 infection after vaccination predominantly recalls humoral immune memory directed against ancestral (hereafter referred to as wild-type (WT)) SARS-CoV-2 spike protein. The resulting elicited antibodies could neutralize both WT SARS-CoV-2 and BA.1 and are enriched on epitopes on spike that do not bind ACE2. However, most of these cross-reactive neutralizing antibodies are evaded by spike mutants L452Q, L452R and F486V. BA.1 infection can also induce new clones of BA.1-specific antibodies that potently neutralize BA.1. Nevertheless, these neutralizing antibodies are largely evaded by BA.2 and BA.4/BA.5 owing to D405N and F486V mutations, and react weakly to pre-Omicron variants, exhibiting narrow neutralization breadths. The therapeutic neutralizing antibodies bebtelovimab4 and cilgavimab5 can effectively neutralize BA.2.12.1 and BA.4/BA.5, whereas the S371F, D405N and R408S mutations undermine most broadly sarbecovirus-neutralizing antibodies. Together, our results indicate that Omicron may evolve mutations to evade the humoral immunity elicited by BA.1 infection, suggesting that BA.1-derived vaccine boosters may not achieve broad-spectrum protection against new Omicron variants.


Article
is an urgent and immediate need to investigate the receptor binding and immune-evasion capabilities of these new Omicron variants.

Structural analyses of Omicron spike
We expressed and purified the prefusion-stabilized trimeric ectodomains of BA.1, BA.2, BA.3, BA.2.12.1, BA.2.13 and BA.4/BA.5 spike (S-trimer). All the S-trimers contain Gly-Ser-Ala-Ser (GSAS) and 6P mutations along with the T4 fibritin trimerization domain for increased stability 14,15 . We determined the cryo-electron microscopy (cryo-EM) structures of these S-trimers at overall resolutions of 3.1-3.5 Å. Together with the previously reported BA.1 structure 16 , this enabled us to compare the structural differences across Omicron sublineages ( Fig. 1a and Extended Data Fig. 1b). In contrast to the BA.1 S-trimer, which is stably maintained in an open conformation with one 'up' RBD and two 'down' RBDs 16 , BA.2 and BA.2.12.1 spike exhibits two conformational states corresponding to a closed form, with all three RBDs in the down configuration and an open form with one RBD in the up position. Of note, one RBD in BA.2.13 was clearly disordered, representing a stochastic movement, which, together with BA.2 and BA.2.12.1, suggests structural heterogeneity in the S-trimers of BA.2 sublineages. Most BA.3 and BA.4 S-trimers adopt closed or semi-closed forms (Fig. 1a). The differences in the RBD up or down conformation could be allosterically modulated by mutations and deletions in the N-terminal domain or near the furin cleavage site, but the detailed mechanism remains unclear. The BA.4/ BA.5 spike that we used in our experiments also contains the N658S mutation, which was present in early BA.4/BA.5 sequences but later disappeared owing to the lower transmissibility of this variant, and may correlate with the more closed RBD configurations of the BA.4/ BA.5 S-trimer. Of note, S-trimers from the BA.2 sublineage harbour relatively less compact architectures in the region formed by the three copies of S2 (Fig. 1b). By contrast, BA.1, BA.3 and BA.4/BA.5 spike possess relatively tight inter-subunit organization with more buried areas between S2 subunits (Fig. 1b). In line with structural observations, thermal stability assays also verified that S-trimers from BA.2 sublineages were the least stable among these variants, which might confer an enhanced fusion efficiency (Fig. 1c).
Next, we measured the binding affinity between human ACE2 (hACE2) and S-trimers of the Omicron variants by surface plasmon resonance (SPR) (Extended Data Fig. 1c). The BA.4/BA.5 S-trimer showed a decreased binding affinity with hACE2 compared with those of the other Omicron subvariants; however, this measurement could be misleading, owing to the additional N658S mutation. To exclude the potential influence of N658S, we also examined the binding affinities of the RBDs of the Omicron variants for hACE2 (Fig. 1d). The RBDs of Delta (B.1.617.2) and the circulating Omicron subvariants exhibited similar binding affinities for ACE2, except for the BA.3 RBD, which showed a lower affinity, similar to that of the ancestral WT strain. Additionally, the BA.2 subvariants displayed slightly higher binding affinities for hACE2 than the other Omicron variants. To further explore the molecular basis for the altered binding affinities of these variants to hACE2, we performed molecular dynamics simulations based on the cryo-EM structures and examined the effects of substitutions in the RBD on the interaction with hACE2 (Extended Data Fig. 1d). The results reveal that the lack of G496S in BA.2 sublineages meant that the hydrogen bond with hACE2 K353 was regained, increasing their binding capability, in line with experimental observations revealed by deep mutational scanning (DMS) assay 17 . However, a local conformational perturbation surrounding spike residues 444-448 disrupted the hydrophilic interaction between BA.3 spike (S446) with hACE2 Q42, presumably owing to the single mutation G446S rather than double mutations of G446S and G496S (Extended Data Fig. 1d)   The binding surface areas between S2 subunits of the variants are calculated in the table on the right. c, Thermoflour analysis for these Omicron variants. Analyses were performed as biological duplicates. d, Binding affinities of RBDs of Omicron variants for hACE2 measured by SPR. Analyses were performed as biological duplicates.
in BA.4/BA.5 spike decreases hACE2 binding activity owing to reduced hydrophobic interaction (Extended Data Fig. 1d). We also noted potential reductions in hydrophilic interactions owing to R493Q reversion. Notably, two reports claimed recently that BA.4/BA.5 RBD and spike (S2P) showed higher binding affinity to hACE2 compared with BA.1 and BA.2 spike, owing to L452R and R493Q reversion 18,19 . Despite this discrepancy, we conclude that BA.2 subvariants and BA.4/BA.5 are able to maintain high binding affinities for hACE2.  Table 1). Plasma samples were collected four weeks after the booster shot or four weeks after discharge from hospital following COVID-19 illness. In plasma from individuals who had received an inactivated virus (CoronaVac) or RBD protein (ZF2001) booster six months after two doses of CoronaVac, BA.1, BA.1.1 and BA.2 showed no significant difference in resistance to neutralization by plasma (Fig. 2a, (Fig. 2a,b). The decrease in neutralization was clearer in plasma obtained from individuals infected by BA.1 who had received three doses of CoronaVac before infection (Fig. 2c) = 38). c, Individuals who, after receiving 3 doses of CoronaVac, had been infected with BA.1 and recovered (n = 50). d, People who had recovered from SARS and received 2 doses of CoronaVac and 1 dose of ZF2001 (n = 28). P-values were calculated using two-tailed Wilcoxon signed-rank tests of paired samples. The geometric mean titre is shown above each group of points. e, Neutralizing activity against SARS-CoV-2 variants and sarbecoviruses by therapeutic NAbs. Green, half-maximal inhibitory concentration (IC 50 ) ≤ 30 ng ml −1 ; white, 30 ng ml −1 < IC 50 < 1,000 ng ml −1 ; red, IC 50 ≥ 1,000 ng ml −1 ; *, IC 50 ≥ 10,000 ng ml −1 . All neutralization assays were performed as biological duplicates. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant (P > 0.05).
To delineate the underlying antibody-evasion mechanism of BA.2.13, BA.2.12.1 and BA.4/BA.5, in particular the escape from the humoral immunity induced by recovery from BA.1 infection or recovery from SARS in vaccinated individuals, we began by isolating RBD-targeting NAbs from such individuals 27,31 (Extended Data Fig. 3a). We isolated antigen-specific memory B cells by fluorescence-activated cell sorting (FACS) of pooled peripheral blood mononuclear cells (PBMCs) using double WT RBD + selection for triple-vaccinated people, WT RBD + and SARS-CoV-1 RBD + selection for vaccinated individuals who had recovered from SARS and double BA.1 RBD + selection for individuals who had recovered from BA.1 infection (Extended Data Fig. 3b). We then performed single-cell V(D)J sequencing (scVDJ-seq) with BA.1 RBD and WT RBD feature barcodes on CD27 + IgM − antigen-specific memory B cells (Extended Data Fig. 3b). We also extracted the productive heavy-light chain paired V(D)J sequences and expressed the antibodies in vitro as human IgG1. We found that the majority of Omicron-reactive memory B cells from triple-CornonaVac-vaccinated individuals who had recovered from BA.1 infection could also bind to WT RBD (Fig. 3a). By contrast, only around a quarter of Omicron-reactive memory B cells isolated from unvaccinated individuals who had recovered from BA.1 infection could bind to WT RBD (Fig. 3a). The cross-reactive antigen-binding property was observed only in IgM − CD27 + memory B cells, but not in IgM + CD27 − naive B cells (Extended Data Fig. 2b). V(D)J sequence analysis revealed significantly higher heavy chain V-domain somatic hypermutation rates of BA.1 and WT cross-reactive B cell receptors (BCRs) than that of BA.1-specific BCRs (Fig. 3b), which indicates that the cross-reactive memory B cells were further affinity-matured compared with BA.1-specific memory B cells. Together, these data suggest that post-vaccination infection with Omicron BA.1 recalls mainly WT-induced memory B cells.
To further specify the epitope distribution of NAbs elicited by post-vaccination BA.1 infection, we applied high-throughput yeast display-based DMS assays 2,3 and determined the mutation profiles for escape for 1,640 RBD-binding antibodies. Among these antibodies, 602 were from individuals who had recovered from WT SARS-CoV-2 or triple vaccinees, 614 were from individuals who had recovered from post-vaccination BA.1 infection, and 410 SARS-WT SARS-CoV-2 cross-reactive antibodies were from vaccinated individuals who had recovered from SARS (Supplementary Table 2). We also included 14 antibodies with published DMS profiles in this analysis 2,32,33 . Of note, among the 614 antibodies from individuals who had recovered from post-vaccination BA.1 infection, 102 were BA.1-specific and did not bind to the WT RBD. The escaping mutation profiles of these 102 BA.1-specific NAbs were determined by DMS based on the BA.1 RBD. The remaining 1,538 WT RBD-reactive antibodies were sorted by unsupervised clustering into 12 epitope groups according to their WT-based mutational escape profiles using t-distributed stochastic neighbour embedding (t-SNE) (Fig. 3c), adding 6 additional epitope groups to our previous classification 3 .
Groups A-C recapitulate our previous taxonomy 3 in which the members target mainly the ACE2-binding motif [34][35][36][37][38] (Fig. 3h). Group D antibodies, such as REGN-10987, LY-CoV1404 and COV2-2130, bind to the linear epitope 440-449 on the RBD and are divided into D1 and D2 subgroups. Group D1 is more affected by mutations of R346 and L452, whereas D2 antibodies are not and interact more with P499 (Fig. 3h). Additionally, groups E and F are divided into E1-E3 and F1-F3, covering the front and back of the RBD and roughly corresponding to class 3 and class 4 antibodies, respectively 37 (Fig. 3h). Group E1 occupies the S309 binding site, whose epitope involves G339, T345 and R346. Group E2 antibodies bind to the front chest of RBD 35 . Group E2.1 binding is most affected by mutations of R346 and A348, whereas E2.2 is most affected by K356 and R357. Groups E3 (S2H97 site) and F1 (S304 site) bind to highly conserved regions on the bottom of the RBD, interacting mainly with K462/E516/L518 and S383/T385/K386, respectively. Antibodies in groups E1-E3 and F1 do not compete with ACE2 (Fig. 3f), whereas F2 and F3 antibodies compete with ACE2 and their binding is affected by T376, K378, D405, R408 and G504, corresponding to class 1/4 39 . We tested the pseuodovirus-neutralizing efficacy of antibodies in each group against SARS-CoV-1, SARS-CoV-2 D614G, Pangolin-GD and RaTG13, as well as their ability to bind to 22 sarbecovirus RBDs using ELISA (Supplementary Tables 2 and 3). Antibodies within the same cluster shared a common sarbecovirus neutralization potency and binding spectrum ( Fig. 3g and Extended Data Fig. 4). In total, we identified five clusters of antibodies exhibiting broad sarbecovirus-binding ability, namely groups E1, E3, F1, F2 and F3 (Extended Data Fig. 4), and antibodies in groups E1, F2 and F3 showed potent neutralizing activity against SARS-CoV-1 (Fig. 3g).
Of note, we found that plasma from individuals who had recovered from post-vaccination BA.1 infection displayed enrichment of group E2.1, E2.2 and F1 antibodies (Fig. 3d,e), which do not compete for binding with ACE2 (Fig. 3f). BA.1 does not harbour mutations on the epitopes of these NAb groups, which may explain why post-vaccination BA.1 infection is more likely to stimulate those NAbs. Although they are not enriched, the ACE2-competing group B and D1 antibodies remain highly abundant after infection. Since group E2, D1 and B antibodies are sensitive to mutation at residues 452 and 486 (Fig. 3h), it is highly probable that BA.2.12.1, BA.2.13 and BA.4/BA.5 can specifically target those antibodies, rationalizing the large decrease in NT 50 of plasma from individuals who have recovered from BA.1 infection against those variants (Fig. 2c).
To examine our hypothesis, we measured pseudovirus neutralization of these NAbs against BA.2.12.1, BA.2.13 and BA.4/BA.5, as well as the major Omicron variants BA.1, BA.1.1, BA.2 and BA.3 (Extended Data Fig. 5). NAbs from different epitope groups displayed distinct neutralizing activities against Omicron subvariants. BA.1-stimulated antibodies (following recovery from BA.1 infection) and WT-stimulated (following recovery from WT infection or vaccination, with or without previous SARS-CoV-1 infection) showed significantly higher potency and breadth in most epitope groups, confirming the increased affinity maturation of these antibodies (Extended Data Fig. 5).
Omicron subvariants evaded most WT-stimulated group A, B and C NAbs, although a subset of these antibodies showed broad effectiveness against Omicron (Extended Data Fig. 5). These broad NAbs were largely enriched by BA.1 stimulation, are generally encoded by similar heavy chain V genes compared with WT-stimulated antibodies and display higher convergence (Extended Data Fig. 6a,b). These broad ACE2-competing NAbs in groups A, B and C have been shown to be enriched in individuals who received a booster dose of mRNA vaccine 39 , which probably accounts for the high neutralizing activity against Omicron variants in plasma of individuals who had received three doses of mRNA vaccine. Nevertheless, BA.1-stimulated group B and C NAbs were significantly evaded by BA.4 owing to F486V and L452R RBD mutations, concordant with results from DMS (Extended Data Fig. 7a,b), which explains the strong humoral immune-evasion ability of BA.4/BA.5.

Omicron escapes broad sarbecovirus NAbs
In total, five clusters of antibodies were found to exhibit broad sarbecovirus-neutralizing ability with diverse specificity, namely groups E1, E3, F1, F2 and F3 (Extended Data Fig. 4). Whereas Group E3 and F1 antibodies demonstrated weak neutralizing activity against all variants owing to their highly conserved binding sites (Extended Data Fig. 8a-c), we found that E1, F2 and F3 NAbs-which are effective against BA.1, were rare in individuals after infection with WT SARS-CoV-2 or Omicron but enriched in vaccinated individuals who had recovered from SARS infection-displayed a systematic reduction in neutralization activity against BA.2 subvariants, BA.3 and BA.4/BA.5 (Figs. 2e and 5a-c). This observation explains the low NT 50 for Omicron subvariants other than BA.1 in plasma from individuals who had recovered from SARS infection (Fig. 2d). The mechanisms behind the loss of neutralization by these broad-specificity sarbecovirus antibodies WT *** *** *** *** *** *** ***    require investigation, as they may prove to be crucial for developing broad-spectrum sarbecovirus vaccines and antibody therapies.

KA ANKR P V A L T GGN L I V YRGP T KP A E KKPR A T S T GNYNK I PWDGT T KP A E KR PR A T S T GNYNK I PWGGT T T P A E KR PR A T QT GNYNK I PWDG I T T P A E KR PNSS T SGNYN L L I NPGT T S P A EMR PNS
To study how BA.2 subvariants, BA.3 and BA.4/BA.5 could systematically reduce the neutralization efficacy of group E1 antibodies, we solved the cryo-EM structures of two group E1 BA.1-neutralizing antibodies, BD55-3152 and BD55-5840, in complex with BA.1 spike proteins using cryo-EM ( Fig. 5d and Extended Data Fig. 9a,b). Similar to S309, the epitope of group E1 antibodies includes an N-linked glycan on N343 (Fig. 5d). Group E1 antibodies are also generally sensitive to mutation of G339, E340, T345 and especially R346, as indicated by their escaping mutation profiles (Fig. 5h). Notably, the newly acquired mutations of BA.2 do not overlap with the shared epitope of E1 antibodies, suggesting that the systematic reduction in neutralization is not caused by amino acid substitution and is potentially owing to structural alteration. To explore this hypothesis, we further determined the cryo-EM structure of the prefusion-stabilized BA.2 spike in complex with the BD55-5840 Fab (Fig. 5e). A structural comparison with the BA.1 RBD binding to BD55-5840 described above suggests that the 366-377 hairpin loop displays significant conformational differences due to S371F and T376A mutations ( Fig. 5e and Extended Data Fig. 9d). The overall positions of residues 375 and 376 are displaced by more than 3 Å, which probably further decreases the binding of group F2 and F3 NAbs in addition to the T376A side-chain substitution. As a result, the bulky phenylalanine resulting from the S371F mutation interferes with the positioning of the glycan moiety attached to N343, which in turn shifts the heavy chain of BD55-5840 upward. This may explain the decreased binding between BD55-5840 and S309, rationalizing their reduced neutralizing activity ( Fig. 5a and Extended Data Fig. 9e). The N343 glycan is critically recognized by almost all group E1 NAbs, including S309. Thus, this group of broad and potent NAbs is probably affected by the S371F mutation in a systematic manner through displacement of the N343 glycan.
The epitopes of group F2 and F3 antibodies cover a continuous surface on the back of the RBD and can only bind to RBDs in the up configuration (Fig. 2b). To probe how BA.2 escapes group F2 and F3 antibodies, we solved the cryo-EM structure of two representative BA.1-neutralizing antibodies-BD55-1239 from group F2 and BD55-3372 from group F3-in complex with BA.1 and Delta spike protein, respectively (Fig. 5f,g and Extended Data Fig. 9a). RBD mutations on T376, K378 and R408 can lead to escape from neutralization by group F2 antibodies (Fig. 5h). Indeed, these residues are centred on the core of the BD55-1239 epitope and are fairly conserved across sarbecoviruses (Fig. 5h). Notably, D405N and R408S, which are present in Omicron BA.2 sublineages, may alter the antigenic surface, disrupting the binding of F2 antibodies (Fig. 5f) and completely abolishing the neutralizing capacity of F2 antibodies (Fig. 5b). Similarly, the D405N and R408S mutations harboured by BA.2 subvariants could interrupt the heavy chain binding of F3 antibodies, causing large-scale escapes of BA.1-neutralizing group F3 NAbs (Fig. 5c). These observations were

BA.1-specific NAbs exhibit narrow breadths
In addition to the WT-BA.1 cross-reactive NAbs, we also investigated the epitope distribution of BA.1-specific NAbs that do not react with WT RBD. We built a yeast display variants library based on the BA.1 RBD, and determined the escape mutation maps of 102 BA.1-specific antibodies. By integrating the analysis of the entire dataset of 1,640 SARS-CoV-2 RBD antibodies, we derived the embedded features of the BA.1-specific NAbs and performed clustering and t-SNE analysis (Fig. 6a). The 102 NAbs were clustered into four BA.1-specific epitope groups, which we designated A Omi , B Omi , D Omi and F3 Omi , since these groups are closely related to the corresponding WT epitope groups (Fig. 6a,e). These antibodies all compete for binding with ACE2 and potently neutralize BA.1 but do not neutralize SARS-CoV-2 D614G or SARS-CoV-1 (Fig. 6b-d) because of the differences in the spike protein: N417K/Y501N/H505Y for A Omi , A484E/K478T for B Omi , K440N for D Omi and R498Q/Y501N for F3 Omi , as indicated by average escape maps of each group (Fig. 6e,f). Some of the previously circulating variants also harbour these same mutations-such as N501Y in Alpha (B.1.1.7), K417N/E484K/N501Y in Beta (B.1.351) and T478K in Delta-and only a small subset of the antibodies exhibit neutralizing activity against these variants (Fig. 6e). Moreover, nearly all of the BA.1-specific NAbs showed poor cross-reactivity against other Omicron subvariants (Fig. 6d) = 32). The geometric mean of the fold change in IC 50 relative to BA.1 is shown above each plot. Wilcoxon signed-rank test of paired samples, compared with IC 50 for BA.1. All neutralization assays were performed as biological duplicates. and antibodies in B Omi are evaded by BA.4 because of F486V. Binding of some group D Omi antibodies may be affected by S446G but were not detected by DMS; these antibodies were evaded by BA.2 subvariants and BA.4 (Fig. 6g). To further validate the results obtained by DMS, we constructed pseudoviruses based on BA.1 harbouring the reverting spike mutations N417K, K440N, S446G, K478T, A484E, R498Q, Y501N and H505Y, as well as BA.1 spike(D405N) and BA.1 spike(R408S). Virus expressing BA.1 spike(D405N) did not produce sufficient titres for further experiments despite multiple attempts. We therefore used BA.2 spike(N405D) instead. We found that the N417K, R498Q, Y501N and H505Y reversions indeed led to evasion of most A Omi and F3 Omi group antibodies, consistent with results from DMS (Fig. 6g). K484E and K478T are the major escaping mutants responsible for the poor breadth of B Omi NAbs (Fig. 6d). S446G caused a small subset of D Omi antibodies to lose neutralization potency, whereas R498Q and K440N resulted in the majority of D Omi NAbs not binding to WT RBD. Of note, expression of BA.1 spike(R408S) did not reduce neutralization by BA.1-specific NAbs, whereas BA.2 spike(N405D) restored the neutralization potency of A Omi and F3 Omi group antibodies against BA.2, indicating that D405N is the determinant of their poor cross-reactivity among BA.2, BA.3, BA.4 and BA.5 sublineages (Fig. 6d,g). These BA.1-specific NAbs displayed different heavy chain V gene usage compared to WT-reactive antibodies in the corresponding epitope group. Specifically, antibodies in A Omi and B Omi groups did not show significant convergence. IGHV3-53 and IGHV3-66 contributes only to a small subset of group A Omi antibodies. Instead, group D Omi antibodies were dominated by IGHV2-70 and IGHV5-51, whereas F Omi was dominated by IGHV4-59 (Extended Data Fig. 10). These three heavy chain V genes also appeared in WT-reactive antibodies, but were relatively rare and did not show significant epitope enrichment (Extended Data Fig. 6a,b).
Here we have shown that Omicron is continuously evolving under immune pressure, and rationalized the appearance of R346K (BA.1.1), L452 substitutions and the F486V mutation, which have enabled increased immune evasion. In contrast to when Omicron first appeared, new Omicron sublineages are able to target the humoral immunity induced by Omicron itself, such as by post-vaccination Omicron infection. Omicron breakthrough infections mainly recall WT-induced memory B cells 40,41 , which in turn narrow the diversity of elicited antibodies and may further drive the appearance of future mutants. These phenomena pose a challenge to the current herd immunity established through WT-based vaccination and infection by BA.1 and BA.2 variants, which is concordant with recent observations 42 . Similarly, these results also suggest that Omicron BA.1-based vaccine may not be the ideal antigen for inducing broad-spectrum protection against emerging Omicron sublineages.
By combining high-throughput single-cell sequencing and high-throughput yeast display-based DMS, we have demonstrated the ability to decipher the complicated humoral immune repertoire elicited by Omicron infection and the underlying immune-evasion mechanism of L452 and F486 mutations. The ability to dissect the entire humoral immunity into distinct antibody epitope groups greatly increases the resolution of antibody and mutational escape research. The antibodies in each epitope group show highly concordant attributes and features, which will facilitate the investigation of the immune-evasion mechanism of circulating variants. This work and the comprehensive data that we have generated here will inform the development of broad-spectrum sarbecovirus vaccines and therapeutic antibodies.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04980-y.

Plasma and PBMC isolation
Blood samples were obtained from 40 volunteers who had received 3 doses of CoronaVac, 39 individuals who had received 2 doses of Coro-naVac and 1 booster dose of ZF2001, 54 individuals who had recovered from BA.1 infection who had previously received 3 doses of Corona-Vac 43,44 , and 30 individuals who had recovered from SARS who had received 2 doses of CoronaVac and 1 dose of ZF2001. The volunteers' blood samples were obtained four weeks after the booster shot or four weeks after discharge from the hospital following BA.1 infection. COVID-19 disease severity was defined as asymptomatic, mild, moderate, severe or critical according to the WHO Living Guidance for Clinical Management of COVID-19 45 . Relevant experiments with plasma from SARS convalescents and SARS-CoV-2 vaccinees were approved by the Beijing Ditan Hospital Capital Medical University (ethics committee archiving no. LL-2021-024-02), the Tianjin Municipal Health Commission, and the ethics committee of Tianjin First Central Hospital (ethics committee archiving no. 2022N045KY). Written informed consent was obtained from each participant in accordance with the Declaration of Helsinki. All participants provided written informed consent for the collection of information, storage and usage of their clinical samples for research purpose, and publication of data generated from this study.
Whole blood samples were mixed and subjected to Ficoll (Cytiva, 17-1440-03) gradient centrifugation after 1:1 dilution in PBS+2% FBS to isolate plasma and PBMCs. After centrifugation, plasma was collected from upper layer and cells were collected at the interface. PBMCs were further prepared by centrifugation, red blood cells lysis (Invitrogen eBioscience 1X RBC Lysis Buffer, 00-4333-57) and washing steps. Samples were stored in FBS (Gibco) with 10% DMSO (Sigma) in liquid nitrogen if not used for downstream process immediately. Cryopreserved PBMCs were thawed in DPBS + 2% FBS (Stemcell, 07905).

B cell RNA and feature barcode data analysis
Using Cell Ranger (v6.1.1) pipeline, the mRNA fastq reads were processed and aligned to the human GRCh38 genome for gene expression profile. Genes expressed in fewer than 10 cells and cells expressing fewer than 100 genes or high-level mitochondria genes were removed to filter out low-quality data. Raw counts were normalized and scaled with Seurat 46 (v 4.0.3), while principal components analysis and uniform manifold approximation and projection were performed for cluster and visualization. Cell types were identified using SingleR 47 (v1.6.1) with Monaco human immune data 48 . Feature barcode reads were also counted by Cell Ranger (v6.1.1) as antibody capture library, and a cell was considered to bind the corresponding antigen of dominant feature barcodes (>25% in this cell).

Antibody sequence analysis
The antibody sequences obtained from 10X Genomics V(D)J sequencing were aligned to GRCh38 reference and assembled as immunoglobulin contigs by the Cell Ranger (v6.1.1) pipeline. Non-productive contigs and B cells that had multiple heavy chain or light chain contigs were filtered out of the analysis. V(D)J gene annotation was performed using NCBI IgBlast (v1.17.1) with the IMGT reference. Mutations on V(D)J nucleotide sequences were calculated by using the igpipeline, which compared the sequences to the closest germline genes and counted the number of different nucleotides. For antibodies from public sources whose original sequencing nucleotide sequences were not all accessible, the antibody amino acid sequences were annotated by IMGT/ DomainGapAlign 49 (v4.10.2) with default parameters. V-J pairs were visualized with the R package circlize (v0.4.10).

High-throughput antibody-escape mutation profiling
The magnetic-activated cell sorting-based antibody-escape mutation profiling system 3,17 was used to characterize mutation escape profile for NAbs. In brief, ACE2-binding mutants were induced overnight for RBD expression and washed followed with two rounds of Protein A antibody-based negative selection and MYC tag-based positive selection to enrich RBD-expressing cells. Protein A antibody-conjugated products were prepared following the protocol for Dynabeads Protein A (Thermo Fisher, 10008D) and incubated with induced yeast libraries at room temperature for 30 min with shaking. MYC tag-based positive selection was performed according to the manufacturer's instructions (Thermo Fisher, 88843).
After three rounds of sequential cell sorting, the obtained cells were recovered overnight. Plasmids were extracted from pre-and post-sort yeast populations by 96-Well Plate Yeast Plasmid Preps Kit (Coolaber, PE053). The extracted plasmids were then used to amplify N26 barcode sequences by PCR. The final PCR products were purified with 1X AMPure XP magnetic beads (Beckman Coulter, A63882) and submitted to 75bp single-end sequencing at Illumina Nextseq 500 platform.

Processing of DMS data
Single-end Illumina sequencing reads were processed as previously described. In brief, reads were trimmed to 16 or 26 bp and aligned to the reference barcode-variant dictionary with dms_variants package (v0.8.9). Escape scores of variants were calculated as F × (n X,ab /N ab )/ (n X,ref /N ref ), where n X,ab and n X,ref is the number of reads representing variant X, and N ab and N ref are the total number of valid reads in antibody-selected (ab) and reference (ref) library, respectively. F is a scale factor defined as the 99th percentiles of escape fraction ratios. Variants detected by less than six reads in the reference library were removed to avoid sampling noise. Variants containing mutations with ACE2 binding below −2.35 or RBD expression below −1 were removed as well, according to data previously reported. For BA.1 RBD-based libraries, due to the lack of corresponding ACE2-binding and RBD expression data, we used the RBD expression of Beta RBD-based DMS as filter instead 50 , and did not perform the ACE2-binding filter. Mutations on residues that use different amino acids in Beta and BA.1 were not filtered, except R493P, S496P, R498P, H505P and all mutations on F375, which were excluded in the analysis owing to low expression. Finally, global epistasis models were built using dms_variants package to estimate mutation escape scores. For most antibodies, at least two independent assays were conducted and single mutation escape scores were averaged across all experiments that pass quality control.

Antibody clustering and visualization
Site total escape scores, defined as the sum of escape scores of all mutations at a particular site on RBD, were used to evaluate the impact of mutations on each site for each antibody. Each of these scores is considered as a feature of a certain antibody and used to construct a feature matrix A N×M for downstream analysis, where N is the number of antibodies and M is the number of features (valid sites). Informative sites were selected using sklearn.feature_selection.VarianceThreshold of scikit-learn Python package (v0.24.2) with the variance threshold as 0.1. Then, the selected features were L2-normalized across antibodies using sklearn.preprocessing.normalize. The resulting matrix is referred as A′ N×M′ , where M′ is the number of selected features. The dissimilarity of two antibodies i, j is defined as 1 − Corr(A′ i ,A′ j ), where Corr(x,y) is the Pearson's correlation coefficient of vectors x and y.
We used sklearn.manifold.MDS to reduce the number of features from M′ to D = 20 with multidimensional scaling under the above metric. Antibodies are clustered into 12 epitope groups using sklearn.cluster. KMeans of scikit-learn in the resulting D-dimensional feature space. Finally, these D-dimensional representations of antibodies were further embedded into two-dimensional space for visualization with t-SNE using sklearn.manifold.TSNE of scikit-learn. For the 102 BA.1-specific antibodies that were assayed with BA.1 RBD-based yeast display library, the 20-dimensional embedding was generated using multidimensional scaling (MDS) with the DMS profile of all 1,640 antibodies, but clustering and t-SNE were conducted independently. To project these antibodies onto the t-SNE space of 1,538 antibodies assayed by WT RBD-based DMS, we calculated the pairwise Euclidean distance between 102 antibodies using BA.1 RBD-based DMS and 1,538 antibodies using WT RBD-based DMS in the 20-dimensional MDS space. The position of each BA.1-specific antibody in the original t-SNE space is defined as the average position of its ten nearest antibodies using WT RBD-based DMS. All t-SNE plots were generated by R package ggplot2 (v3.3.3).  (T19I, L24S, del25-27, del69-70, G142D, V213G, G339D, S371F,  S373P, S375F, T376A, D405N, R408S, K417N, N440K, G446S, L452R,  S477N, T478K, E484A, F486V, Q498R, N501Y, Y505H, D614G, H655Y,  N679K, P681H, N764K, D796Y, Q954H, N969K) plasmids were constructed using the pcDNA3.1 vector. G*ΔG-VSV virus (VSV G pseudotyped virus, Kerafast) was used to infect 293T cells (American Type Culture Collection (ATCC), CRL-3216), and spike protein-expressing plasmid was used for transfection at the same time. After culture, the supernatant containing pseudovirus was collected, filtered, aliquoted, and frozen at −80 °C for further use.

ELISA
To detect the broad-spectrum binding of the antibodies among Sarbecovirus, we used a panel of 20 synthesized sarbecovirus RBDs (Sino Biological Technology) (Supplementary Table 3). According to the sequence of 20 RBDs, a set of nested primers was designed. The coding sequences were obtained by the overlap PCR with a 6× His tag sequence to facilitate protein purification. The purified PCR products were ligated to the secretory expression vector pCMV3 with CMV promoter, and then transformed into Escherichia coli XL1-blue competent cells. Monoclones with correct transformation were cultured and expanded, and plasmids were extracted. Healthy HEK293F cells were passaged into a new cell culture and grown in suspension at 37 °C, 120 RPM, 8% CO2 to logarithmic growth phase and transfected with the recombinant constructs by using liposomal vesicles as DNA carrier. After transfection, the cell cultures were followed to assess the kinetics of cell growth and viability for 7 days. The cell expression supernatant was collected, and after centrifugation, passed through a Ni column for affinity purification. The molecular size and purity of eluted protein was confirmed by SDS-PAGE. Production lot numbers and concentration information of the 20 sarbecovirus proteins are shown in Supplementary  Table 3) in PBS was pre-coated onto ELISA plates (NEST, 514201) at 4 °C overnight. The plates were washed and blocked. Then 1 μg ml −1 purified antibodies or serially diluted antibodies were added and incubated at room temperature for 20 min. Next, Peroxidase-conjugated AffiniPure Goat Anti-Human IgG (H+L) ( JACKSON, 109-035-003) was applied and incubated at room temperature for 15 min. Tetramethylbenzidine (TMB) (Solarbio, 54827-17-7) was added onto the plates. The reaction was terminated with 2 M H 2 SO 4 after 10 min incubation. Absorbance was measured at 450 nm using Ensight Multimode Plate Reader (PerkinElmer, HH3400). ELISA A 450 measurements at different antibody concentrations for a particular antibody-antigen pair were fit to the model y = Ac n /(c n + E n ) using the R package mosaic (v1.8.3), where y is the A 450 value and c is the corresponding antibody concentration. A, E and n are parameters, where E is the desired EC 50 value for the specific antibody and antigen.

Antibody and ACE2 competition for RBD
Omicron RBD (Sino Biological, 40592-V08H121) protein in PBS was immobilized on the ELISA plates at 4 °C overnight. The coating solution was removed and washed 3 times with PBST and the plates were then blocked for 2 h. After blocking, the plates were washed 5 times, and the mixture of ACE2-biotin (Sino Biological, 10108-H27B-B) and serially diluted competitor antibodies was added followed by 30 min incubation at room temperature. Peroxidase-conjugated Streptavidin ( Jackson ImmunoResearch, 016-030-084) was added into each well for another 20 min incubation at room temperature. After washing the plates five times, TMB (Solarbio, 54827-17-7) was added into each well. After 10 min, the reaction was terminated with 2 M H 2 SO 4 . Absorbance was measured at 450 nm using Ensight Multimode Plate Reader (PerkinElmer, HH3400). The ACE2 competition coefficient was calculated as (B − A)/B, where B is the A 450 value with 0.3 μg ml −1 antibody and A is the A 450 value with 6 μg ml −1 antibody.

Biolayer interferometry
Biolayer interferometry assays were performed on Octet RED 384 Protein Analysis System (Fortebio) according to the manufacturer's instructions. To measure the binding affinities, monoclonal antibodies were immobilized onto Protein A biosensors (Fortebio) and the fourfold serial dilutions of Omicron S-trimer (BA.1 and BA.2) in PBS were used as analytes. Data were collected with Octet Acquisition 9.0 (Fortebio) and analysed by Octet Analysis 9.0 (Fortebio) and Octet Analysis Studio 12.2 (Fortebio).

S-trimer thermal stability assay
The thermal stability assay was performed to detect the exposed hydrophobic residues with an MX3005 qPCR instrument (Agilent) with SYPRO Red (Invitrogen) as fluorescent probes. We set up 25 μl reaction system (pH 8.0) containing 5 μg of target protein (S-trimer of Omicron lineage), 1000× SYPRO Red, and ramped up the temperature from 25 °C to 99 °C. Fluorescence was recorded in triplicate at an interval of 1 °C.

Surface plasmon resonance
Human ACE2 was immobilized onto CM5 sensor chips using a Biacore 8K (GE Healthcare). Serial dilutions of purified S-trimer or RBD of Omicron lineages were injected, ranging in concentrations from 100 to 6.25 nM. The response units were recorded at room temperature using BIAcore 8K Evaluation Software (v3.0.12.15655; GE Healthcare), and the resulting data were fitted to a 1:1 binding model using BIAcore 8K Evaluation Software (v3.0.12.15655; GE Healthcare).

Protein expression and purification for cryo-EM study
The S6P expression construct encoding the SARS-CoV-2 spike ectodomain (residues 1-1208) with six stabilizing Pro substitutions (F817P, A892P, A899P, A942P, K986P and V987P) and a GSAS substitution for the furin cleavage site (residues 682-685) was previously described 15 51 . For protein production, these expression plasmids, as well as the plasmids encoding the antigen-binding fragments (Fabs) of the antibodies described here, were transfected into the HEK293F cells using polyethylenimine (Polysciences). The conditioned media were collected and concentrated using a Hydrosart ultrafilter (Sartorius), and exchanged into the binding buffer (25 mM Tris, pH 8.0, and 200 mM NaCl). Protein purifications were performed using the Ni-NTA affinity method, followed by gel filtration chromatographies using either a Superose 6 increase column (for the spike proteins) or a Superose 200 increase column (for the Fabs). The final buffer used for all proteins is 20 mM HEPES, pH 7.2, and 150 mM NaCl.