Neutralizing monoclonal antibodies elicited by mosaic RBD nanoparticles bind conserved sarbecovirus epitopes

Increased immune evasion by SARS-CoV-2 variants of concern highlights the need for new therapeutic neutralizing antibodies. Immunization with nanoparticles co-displaying spike receptor-binding domains (RBDs) from eight sarbecoviruses (mosaic-8 RBD-nanoparticles) efficiently elicits cross-reactive polyclonal antibodies against conserved sarbecovirus RBD epitopes. Here, we identified monoclonal antibodies (mAbs) capable of cross-reactive binding and neutralization of animal sarbecoviruses and SARS-CoV-2 variants by screening single mouse B cells secreting IgGs that bind two or more sarbecovirus RBDs. Single-particle cryo-EM structures of antibody-spike complexes, including a Fab-Omicron complex, mapped neutralizing mAbs to conserved class 1/4 RBD epitopes. Structural analyses revealed neutralization mechanisms, potentials for intra-spike trimer cross-linking by IgGs, and induced changes in trimer upon Fab binding. In addition, we identified a mAb-resembling Bebtelovimab, an EUA-approved human class 3 anti-RBD mAb. These results support using mosaic RBD-nanoparticle vaccination to generate and identify therapeutic pan-sarbecovirus and pan-variant mAbs.


In brief
Sarbecovirus spike receptor-binding domains (RBDs) include conserved and variable epitopes, suggesting that antibodies against conserved regions would protect against future sarbecovirus spillovers and SARS-CoV-2 variants. Fan et al. structurally and functionally characterized monoclonal antibodies elicited by a mosaic-8 RBD-nanoparticle vaccine candidate, demonstrating crossreactive binding, neutralization, and targeting of desired epitopes.

INTRODUCTION
Spillover of animal SARS-like betacoronaviruses (sarbecoviruses) resulted in two human health emergencies in the past 20 years: the SARS-CoV epidemic in the early 2000s and the current COVID-19 pandemic caused by SARS-CoV-2. Large coronavirus reservoirs in bats are predictive of future cross-species transmission, 1-3 necessitating a vaccine that could protect against emerging coronaviruses. In addition, SARS-CoV-2 variants of concern (VOCs) have been discovered throughout the current pandemic, designated as such due to increased transmissibility and/or resistance to neutralizing antibodies. [4][5][6][7] In the case of Omicron VOCs, a large number of substitutions in the SARS-CoV-2 spike protein receptor-binding domain (RBD), and detectable cross-variant neutralization, 8 results in reduced efficacies of vaccines and therapeutic monoclonal antibodies (mAbs). 5,9 Comparison of the variability of RBDs across sarbecoviruses and within SARS-CoV-2 variants suggest that vaccines and mAbs targeting the more conserved neutralizing antibody epi-topes (class 4 and class 1/4; nomenclature from Barnes et al. 10 and Jette et al. 11 could protect against future zoonotic spillovers and SARS-CoV-2 VOCs. By contrast, antibodies targeting the less conserved class 1 and class 2 RBD epitopes that directly overlap with the binding footprint for human angiotensin-converting enzyme 2 (ACE2) receptor, the SARS-CoV-2 host receptor, recognize a portion of the RBD that exhibits sequence variability between sarbecoviruses, 10 which is also where VOC and variant of interest (VOI) substitutions accumulate. Class 3 RBD epitopes are more conserved than class 1 and class 2 epitopes but exhibit some variation across sarbecoviruses, suggesting the potential for continued variability among SARS-CoV-2 VOCs. 10 Here, we investigated the RBD epitopes of mAbs isolated from mosaic RBD-and homotypic RBD-immunized mice to characterize the antibody response to RBD nanoparticles. Binding and neutralization results, together with cryoelectron microscopy (cryo-EM) structures of antibody Fab-spike trimer complexes, suggested that the mosaic RBD-nanoparticle vaccine approach works as designed to target conserved epitopes and Figure 1. Utilizing antibody avidity effects suggests a strategy to target antibodies to conserved regions of sarbecovirus RBDs (A) Hypothesis for preferential stimulation of B cells with cross-reactive BCRs by mosaic (left) versus homotypic (right) RBD nanoparticles. Left: green crossreactive BCRs can crosslink between a conserved epitope (green circles) on adjacent RBDs in a mosaic RBD nanoparticle to enhance binding to a more occluded, but conserved, epitope through avidity effects. Middle: yellow BCRs recognizing an accessible strain-specific epitope (yellow triangle) can crosslink between adjacent RBDs on a homotypic nanoparticle to enhance binding through avidity effects. Right: yellow BCRs against a strain-specific orange epitope cannot crosslink between adjacent RBDs on a mosaic RBD nanoparticle that presents different versions of the epitope (colored triangles).
(legend continued on next page) ll OPEN ACCESS Article could be used both for more broadly protective vaccines and as a method to produce therapeutic neutralizing mAbs that would not be affected by Omicron or future SARS-CoV-2 VOC substitutions.

RESULTS
The majority of mosaic-8-elicited mouse mAbs identified as binding two or more RBDs are cross neutralizing The hypothesis behind enhanced elicitation of cross-reactive antibodies by mosaic RBD-nanoparticles is that B cell receptors (BCRs) recognizing conserved RBD epitopes are stimulated to produce cross-reactive Abs through bivalent binding of BCRs to adjacent RBDs, which would rarely occur when RBDs are arranged randomly on a nanoparticle ( Figure 1A). 12,13 By contrast, homotypic RBD-nanoparticles are predicted to stimulate BCRs against immunodominant strain-specific epitopes presented on all RBDs ( Figure 1A). The more conserved class 4 and class 1/4 epitopes ( Figure 1B) targeted by polyclonal antibodies in mosaic-8 RBD-nanoparticle antisera are unlikely to vary in SARS-CoV-2 VOCs ( Figure 1C; Data S1) because they contact other portions of the spike trimer, unlike class 1 and 2 RBD epitope regions targeted by homotypic SARS-CoV-2 RBDnanoparticle antisera that are not involved in contacts with non-RBD portions of spike ( Figure 1B). 12 We produced and characterized nanoparticles presenting randomly arranged RBDs from SARS-CoV-2 WA1 and seven animal sarbecoviruses (Pang17, RaTG13, WIV1, SHC014, Rs4081, RmYN02, and Rf1) (mosaic-8 RBD-mi3) and nanoparticles presenting only SARS-CoV-2 WA1 RBDs (homotypic SARS-CoV-2 RBD-mi3) 15 (Figures 1D and S1). Mice were primed and boosted with either mosaic-8 or homotypic SARS-CoV-2 RBD-nanoparticles in AddaVax adjuvant. We used a Berkeley Lights Beacon Optofluidic system to screen a subset of B cells for binding to one or more labeled RBDs (Data S1). B cells secreting IgGs binding at least one RBD were exported, and the variable domains of heavy-and light-chain genes were sequenced and subcloned into expression vectors containing genes encoding human IgG C H 1-C H 2-C H 3 domains, human C H 1, or human C L domains. From 39 exported cells, we isolated genes for 15 RBD-binding mAbs (Table S1) that were expressed as IgGs and Fabs. The 15 unique IgG sequences included 13 derived from mosaic-8 immunized mice and identified as binding to R2 (six mAbs) or to one (seven mAbs) labeled RBDs and two derived from homotypic RBD-nanoparticle immunized mice and identified as binding to R2 RBDs ( Figure 2A; Table S1). Two mAbs from mosaic-8 immunized mice were excluded from analyses after showing no detectable binding to purified RBDs (Table S1).
To identify RBD epitopes, we assessed potential competition with proteins that bind to known RBD epitopes, using the four human anti-RBD mAbs used as controls for ELISAs ( Figure 2A) plus other potential competitor or control mAbs: C022 (class 1/4), 11,17 CR3022 (class 4), 22 COVA1-16, 23 C135 (class 3), C110 (class 3), C105 (class 1), 17 and a soluble human ACE2-Fc construct. 11 The ELISA revealed the expected competition for the characterized human mAbs, validating its use for mapping RBD epitopes. Three of the five m8a mAbs (M8a-3, M8a-31, and M8a-34) mapped to class 1/4 or class 4 epitopes, M8a-28 mapped to the class 3 RBD region, and Ma-6 did not compete with any of the labeled anti-RBD IgGs ( Figure S2D). The identification of a class 3 RBD epitope for M8a-28 rationalized its potent neutralization of SARS-CoV-2 variants and limited neutralization of animal sarbecoviruses ( Figure 2B). The class 1/4 RBD epitope identification explained the lower neutralizing potency of M8a-3, M8a-31, and M8a-34, since this class of anti-RBD mAb tends to show less potent neutralization but broader sarbecovirus cross-reactivity, than other classes due to the more occluded nature of the class 1/4 epitope. 11,12,24 Of the two HSW mAbs, HSW-1 showed no detectable competition, and HSW-2 competed with CR3022, a class 4 anti-RBD mAb. These results demonstrated that most of the mAbs identified during Beacon screening mapped to the more conserved class 1/4, 4, and 3 RBD epitopes.
Cryo-EM structures of Fab-spike trimer complexes reveal cross-reactive recognition and rationalize neutralization results To deduce recognition and neutralization mechanisms, we used single-particle cryo-EM to solve structures of Fabs from the seven cross-reactive mAbs complexed with a SARS-CoV-2 6P spike trimer 25 (Figures 3, 4A, and 5; Table S2; Data S1). Each of the five M8a Fabs were bound to the SARS-CoV-2 WA1 spike, and the M8a-31 Fab was also complexed with the Omicron BA.1 spike (Figures 3A-3F; Table S2; Data S1). We observed one Fab bound to each of the three ''up'' RBDs, except for the M8a-28spike structure in which all three RBDs were ''down'' ( Figure 3C) and the M8a-6-spike structure, which showed only one wellresolved Fab per trimer.
A 3.1 Å resolution M8a-3 Fab-spike complex structure revealed Fab V H -V L interactions with ''up'' RBDs using all six CDRs along with residues within the light-chain framework region 2 and 3 (FWRL2 and FWRL3) ( Figures 3A, 4B, and S3A; Data S1). Consistent with the competition ELISA results (Figure S2D), comparison of the M8a-3 Fab-RBD interaction with previously characterized representative anti-RBD antibodies in different structural classes 10,11 showed overlap with the class 1 and class 4 RBD epitopes ( Figure S3A) and a binding footprint adjacent to that of ACE2 receptor ( Figures 3A and 4A). This was similar to the human mAb C118, a class 1/4 anti-RBD antibody that blocks ACE2 binding without substantially overlapping with the ACE2 receptor binding footprint 11 and competes with M8a-3 for RBD binding ( Figure S2D). The M8a-3-spike structure recognized a largely conserved region of the RBD ( Figure 4B), consistent with ELISA and neutralization results where M8a-3 neutralized and/or bound to most of the sarbecoviruses and the SARS-CoV-2 variants tested ( Figure 2). A 3.2 Å spike trimer structure complexed with the related, but mostly non-neutralizing, M8a-6 mAb showed three ''up'' RBDs but only one well-resolved Fab ( Figures 3B and S3B; Data S1). The M8a-6 Fab shared a similar RBD epitope and approach angle as M8a-3 ( Figures 3A and 4A; Figure S3B), interacting with the RBD using all six CDRs plus framework regions FWRH2, FWRL2, and FWRL3 ( Figure 4C). Furthermore, M8a-6 also recognized a similar epitope as C118 11 and M8a-3, involving mostly conserved RBD residues ( Figures 4C and  S3B). Despite sharing high sequence identity and similar binding epitopes on SARS-CoV-2 RBD with M8a-3, M8a-6 was ELISAs conducted with duplicate samples at least twice (for first seven mAbs) or once (for remaining mAbs). The same EC 50 values are presented for M8a-11 and M8a-26, which shared the same protein sequences. (B) Left: neutralization potencies (IC 50 values) of mAbs against SARS-CoV-2 variants and indicated sarbecoviruses. IC 50 s are reported from neutralization assays that were conducted using duplicate samples at least twice except for a single assay for M8a-28 against Omicron BA.1. Right: median IC 50 values. Significant differences calculated using Tukey's multiple comparison test between mAbs linked by horizontal lines are indicated by asterisks: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Medians are represented by black lines for IC 50 values of each mAb. See also Figure S2, Table S1, and Data S1. non-neutralizing against SARS-CoV-2 and only weakly neutralizing against BtKY72, whereas M8a-3 neutralized SARS-CoV-2 D614G with a 0.18 mg/mL IC 50 ( Figure 2B). These different neutralization profiles likely result from a weaker interaction of M8a-6 compared with M8a-3 with CoV spikes, as demonstrated by incomplete binding of Fabs in the M8a-6-spike complex cryo-EM structure ( Figure 3B) and the lack of competition of M8a-6 IgG with any of the IgGs with known epitopes ( Figure S2D). To investigate whether M8a-6 binds more weakly to its RBD epitope than M8a-3, we used surface plasmon resonance (SPR) to examine binding of M8a-3 and M8a-6 compared with C118 11 to a set of eight RBDs (Data S1). Visual inspection of sensorgrams and kinetic and equilibrium constants (when they could be derived by fitting data to a 1:1 binding model) showed weaker RBD binding by M8a-6 than by M8a-3 or C118. To map its epitope, we determined a cryo-EM structure of M8a-34 Fab bound to the WA1 spike trimer at 3.5 Å resolution ( Figure 3E; Data S1), revealing interactions of three Fabs with three ''up'' RBDs ( Figures 3E and S3E) that were modeled using an M8a-34 Fab-RBD crystal structure (Table S3). M8a-34 Fab interacted with the RBD through all three CDRHs as well as CDRL1 and CDRL3 ( Figures 4E and S3G). The M8a-34 epitope was similar to epitopes of other class 1/4 mAbs including M8a-3, M8a-6, and M8a-31, which overlapped with the binding epitopes of CR3022 (class 4) and C118 (class 1/4) ( Figures 4A and S3E), again consistent with its binding and neutralizing properties ( Figure 2) and competition ELISA results ( Figure S2D).
M8a-28, which showed the lowest degree of cross-reactive RBD binding (Figure 2A), mapped to the class 3 epitope instead of the more conserved class 1/4 and class 4 epitopes (Figure S2D), and except for M8a-6, it showed the lowest levels of cross-reactive sarbecovirus neutralization of the five mAbs isolated from mosaic-8 immunized mice ( Figure 2B). Single-particle cryo-EM structures of the M8a-28 Fab-spike complex were determined in two conformational states: a 2.8 Å structure with each of three Fabs binding to a ''down'' RBD ( Figure 3F) and a 3.1 Å structure with two Fabs bound to adjacent ''down'' RBDs and a third Fab at lower occupancy bound to a flexible ''up'' RBD (Data S1). The Fab-RBD interaction was mediated by all six CDRs, plus FWRH3 and FWRL1 ( Figures 4F and S3H).  Figure 2B), consistent with its epitope spanning more variable RBD residues than epitopes of class 4 and class 1/4 anti-RBD mAbs. 11 Despite broad recognition of sarbecovirus RBDs (Figure 2A), the HSW mAbs exhibited overall weaker neutralization potencies across the sarbecoviruses tested, with all IC 50 values >10 mg/mL ( Figure 2B). To compare recognition properties with the M8a Fabs, we determined a cryo-EM structure of HSW-1 bound to WA1 spike at 3.1 Å resolution, revealing a single well-ordered Fab bound to a trimer with two ''up'' RBDs and one ''down'' RBD ( Figures 5A and S4A; Data S1). The bound HSW-1 Fab interacted with two RBDs: one ''up'' RBD (1 RBD) and the adjacent ''down'' RBD (2 RBD) ( Figures 5A and S4A). Interactions between HSW-1 and the 1 RBD were mediated by FWRH1, CDRH1, CDRH3, CDRL1, CDRL2, CDRL3, and FWRL2 and only by the HSW-1 light chain for the 2 RBD (Figures 5A and 5B). Structural comparisons showed the epitope of HSW-1 overlapped somewhat with the binding epitopes of C118 (class 1/4) and CR3022 (class 4) and included mostly conserved residues ( Figure S4A).
We next used cryo-EM to investigate HSW-2-spike interactions, observing two main populations of particles: unliganded intact spike trimers and a Fab-spike S1 domain protomer complex (Data S1). From the latter, we obtained an EM reconstruction at 4.1 Å of HSW-2 Fab bound to the WA1 S1 domain ( Figures 5C and S4B) using a crystal structure of an HSW-2 Fab-RBD complex (Table S3) Figure S4, Tables S2 and S3, and Data S1. ll OPEN ACCESS ( Figure S2D), and although their binding poses differed, the HSW-2 and CR3022 epitopes overlapped ( Figure S4B). 22 S1 shedding resulting from mAb binding has been suggested as a possible neutralization mechanism for CR3022 and other class 4 anti-RBD mAbs; 22,26,27 however, HSW-2 was largely non-neutralizing ( Figure 2B). To determine accessibility of the HSW-2 epitope in an intact spike trimer, we aligned the RBD portion of the HSW-2 Fab-RBD structure to RBDs from spike structures with all ''up'' or all ''down'' RBDs, finding steric clashes in both cases ( Figures 5F and 5G). The inability of the HSW-2 Fab to access either ''up'' or ''down'' RBDs in an intact spike trimer is consistent with the observation that HSW-2 showed weak or no neutralization activity ( Figure 2B) despite binding almost all RBDs evaluated by ELISA (Figure 2A).
In summary, structural studies corroborated the competition assay mapping of the mouse mAb epitopes ( Figure S2D) and further revealed details of RBD recognition in the context of spike trimers.
To understand how substitutions in VOCs might affect binding of the mAbs for which we had Fab-spike structures, we mapped their binding epitopes compared with Omicron RBD substitutions ( Figures 4B-4F, 5B, and 5D). Most of the Omicron substitutions were in the more variable ACE2 receptor binding region ( Figures 1A and 4A; Data S1), with fewer substitutions in conserved regions ( Figures 1A, 4B-4F, 5B, and 5D; Data S1). Omicron substitutions were mainly at the peripheries of the RBD epitopes of the m8a mAbs isolated from mosaic-8-immunized mice (Figures 4B-4F), and there were no Omicron substitutions within the binding epitopes of the two HSW mAbs isolated from homotypic nanoparticle-immunized mice ( Figures 5B and 5D). Despite the Omicron substitutions not greatly affecting RBD binding by the seven mAbs (Figure 2A), some of the class 1/4 M8a mAbs showed somewhat reduced neutralization potencies ( Figure 2B).
Although RBD binding correlates with neutralization potencies for polyclonal antisera from RBD-nanoparticle immunized animals, 15 this is true for all mAbs, e.g., CR3022 binds to SARS-CoV-2 RBD but neutralizes only weakly or not at all. 30 One mechanism by which Omicron or other RBD substitutions could indirectly affect neutralization potencies of mAbs without affecting binding to isolated RBDs is by changing the dynamics of the conversion between ''up'' and ''down'' RBD conformations on spike trimers. Some classes of anti-RBD mAbs have a strong or absolute preference for binding an ''up'' versus a ''down'' RBD, e.g., most class 1 and class 4 anti-RBD mAbs only recognize ''up'' RBDs. 10 To assess whether the mAbs investigated here recognized ''up'' and/or ''down'' RBDs, we evaluated the accessibility of their epitopes on a spike by mapping each binding epitope onto an unliganded trimer structure with one ''up'' and two ''down'' RBDs (PDB: 6VYB) ( Figure S5) and a trimer with all ''up'' RBDs (PDB: 7RKV) ( Figure S6). The class 4 and 1/4 epitopes of M8a-3, M8a-6, M8a-31, M8a-34, and HSW-1 were buried when RBDs adopted the ''down'' conformation ( Figures S5A-S5D and S5F) but fully exposed in the ''up'' RBDs ( Figures S6A-S6D and S6F). Although the HSW-2 class 4 epitope was buried in ''down'' RBD conformation ( Figure S5G) and could be partially exposed in an ''up'' RBD conformation ( Figure S6G), structure alignments showed that HSW-2 cannot bind ''up'' or ''down'' RBDs in the context of a spike trimer ( Figures 5F and 5G). By contrast, the class 3 epitope of M8a-28 was exposed in both RBD conformations ( Figures S5E  and S6E). Likely related to these observations, only the M8a-28-bound trimer structure showed an inter-protomer RBD distance of 31 Å ( Figure 6I) equivalent to that of an unliganded trimer  (legend continued on next page) (28-40 Å ) ( Figure 6A). The other class 4 and 1/4 mAb Fab-bound trimer structures showed larger inter-protomer RBD distances (up to 70 Å ), corresponding to 11-34 Å more outward displacement of RBDs in comparison with unliganded or class 1-or ACE2-liganded spike trimer structures (Figures 6B-6H). 10 This outward displacement of RBDs could result in spike trimer destabilization, leading to S1 shedding. 11,18,22,26 Another property of antibodies that could affect their neutralization potencies relates to their ability to utilize bivalency. Since IgG antibodies have two identical Fab arms, they can increase their apparent affinities for binding to tethered antigens through avidity effects, which can occur through either inter-spike crosslinking (simultaneous binding of two neighboring spike trimers) or intra-spike cross-linking (simultaneous binding of two neighboring RBDs within the same spike trimer). To evaluate whether the M8a or HSW mAbs could enhance their binding through intra-spike crosslinking, we measured distances between neighboring Fabs in the Fab-spike structures to predict if simultaneous binding of both IgG Fabs to adjacent RBDs on a trimer would be possible. A distance of %65 Å between the C termini of the C H 1 domains of adjacent bound RBD-bound Fabs is required to allow the N-termini of the two chains of an IgG hinge to each of the C-termini of two bound Fabs. 10 Measured distances in spike trimers complexed with the M8a-3 (126, 130, and 159 Å ), M8a-34 (107, 110, and 150 Å ), or M8a-28 (144 Å ) Fabs were too large to permit intra-spike cross-linking ( Figures 6C, 6G, and 6I). Although we could not measure analogous distances in the M8a-6-spike structure because only one Fab was bound ( Figure 6D), the similar epitope and pose for M8a-3 and M8a-6 ( Figures 3A, 3B, 4B, and 4C) suggest that an IgG version of M8a-6 is unlikely to crosslink adjacent RBDs. Thus, the weak binding of M8a-6 to a spike trimer could not be improved by intra-spike crosslinking avidity effects, again rationalizing its lack of neutralizing activity ( Figure 2B). For spike trimers complexed with M8a-31 Fab ( Figures 6E and 6F), distances between the C termini of adjacent C H 1 domains were measured as 52 and 49 Å for M8a-31 Fab bound to the WA1 and Omicron BA.1 spikes, respectively, suggestive of potential intra-spike crosslinking. We could not evaluate potentials for intra-spike crosslinking for HSW-1 or HSW-2 because either only one Fab was bound per spike (HSW-1) ( Figure 5A) or the reconstructions showed Fab binding to dissociated S1 monomer (HSW-2) ( Figure 5C).
We also used modeling to assess how the RBD-nanoparticles used to elicit the mAbs investigated here might engage with bivalent BCRs. To address this issue, we asked whether the geometric arrangement of RBDs on mosaic-8 RBD-mi3 nanoparticles would permit bivalent engagement of neighboring RBDs by IgGs, here representing membrane-bound BCRs hypothesized to engage adjacent RBDs ( Figure 1D). We first constructed IgG models of each of the Fabs in the M8a and HSW Fab-spike structures (Figures 3 and 5). Next, we asked if it was sterically possible for both Fabs of an IgG to interact with the epitope identified from its cryo-EM structure on adjacent RBDs on a modeled RBD-mi3 nanoparticle. For each of the seven mAb epitopes, we found that the RBD-mi3 nanoparticle geometry was predicted to allow simultaneous recognition of adjacent RBDs by both Fabs of an IgG ( Figure S7), thus confirming that the geometric arrangement of RBD attachment sites on SpyCatcher-mi3 would allow BCR engagement through avidity effects. mAbs elicited by mosaic-8 RBD-nanoparticles resemble EUA-approved therapeutics or a potent cross-reactive human class 1/4 anti-RBD antibody Human mAbs that received emergency use authorization (EUA) for COVID-19 treatment include class 1 and class 2 anti-RBD mAbs that are no longer effective against SARS-CoV-2 variants and class 3 anti-RBD mAbs, two of which, Bebtelovimab and Cilgavimab, retain at least partial efficacy against Omicron variants ( Figures 7A and 7B). The epitope identified for M8a-28 ( Figure 7C) resembles epitopes of the class 3 anti-RBD therapeutic mAbs ( Figures 7D-7G), as evaluated by comparisons of common RBD epitope buried surface areas (BSAs) ( Figure 7B). Some of these mAbs, including M8a-28 ( Figure 2B), neutralize Omicron VOCs, but their epitope locations within a region that varies among sarbecoviruses suggests that future SARS-CoV-2 variants are likely to include substitutions that reduce or completely abrogate their efficacies (Figures 7C-7G). By contrast, the more occluded class 1/4 RBD epitope ( Figure 7A), to which bound mAbs can inhibit ACE2 receptor binding, 11,23,24 exhibits less variability across sarbecoviruses likely because substitutions that affect its contacts as a ''down'' RBD with other spike trimer regions limit its variability between SARS-CoV-2 VOCs and other sarbecoviruses. 12

DISCUSSION
Here, we characterized mouse mAbs elicited by mosaic (M8a mAbs) or homotypic (HSW mAbs) RBD-nanoparticles using both structural and functional analyses, showing that mosaic nanoparticles induce potently neutralizing antibodies that cross-react between animal sarbecoviruses and SARS-CoV-2 VOCs. Although we identified only five mAbs that bound to R2 RBDs from mosaic-8 immunized mice in these first experiments, one mAb (M8a-3) was both cross-reactive and strongly neutralizing and two others (M8a-31 and M8a-34) were less potently neutralizing but were cross-reactive against SARS-CoV-2 variants and animal sarbecoviruses. Another mAb (M8a-28) potently neutralized SARS-CoV-2 variants and resembled therapeutic antibodies in current use. Encouragingly, M8a-3, M8a-28, and M8a-31 neutralized all Omicron variants against which they were evaluated (BA.1, BA.2, BA.2.12.1, and BA.4/BA.5), although the Omicron lineage of SARS-CoV-2 had not emerged at the time these experiments were initiated. Structural studies showed that all five mAbs target the desired more conserved epitopes (class 3 and class 1/4) rather than the class 1 and class 2 RBD epitopes more commonly elicited by vaccination or infection. 26,31,32 By contrast, the only two mAbs isolated from homotypic SARS-CoV-2 nanoparticle-immunized mice that were identified as binding R2 RBDs during screening targeted different epitopes and were only weakly-or non-neutralizing.
Structural studies of Fab complexes with SARS-CoV-2 spike trimers, including one with Omicron BA.1, demonstrated that four of the five mAbs isolated from mosaic-8 immunized mice recognized conserved epitopes, as designed in the immunization approach and shown for polyclonal antisera raised in mice by mosaic-8 RBD-nanoparticle immunization. 12 By contrast, antibodies raised in homotypic RBD-nanoparticle immunized mice more commonly recognize variable class 1 and class 2 RBD epitopes, 12 likely explaining why it was more difficult in the current study to isolate single B cells from homotypic RBD-nanoparticle immunized mice secreting IgGs that bound R2 labeled RBDs. The two cross-RBD binding mAbs we were able to isolate from homotypic RBD-nanoparticle immunized mice showed binding to multiple sarbecovirus RBDs but were only weakly-or nonneutralizing. Corroborating this, the HSW-1-spike structure showed only one bound Fab per trimer compared with three bound Fabs per trimer in the structures of more potently neutralizing mAbs, and the HSW-2 Fab epitope was incompatible with binding to its RBD epitope on intact spike trimer, resulting in a trimer dissociation.
The fact that four of five mouse mAbs identified as binding to R2 different RBDs during B cell screening after mosaic-8 immunization target the class 1/4 epitope, in common with the potent, cross-reactive, and protective S2X259 human mAb, 24 supports the potential for using mosaic RBD-nanoparticles as immunogens to efficiently elicit cross-reactive and potent neutralizing mAbs against SARS-CoV-2 variants and animal sarbecoviruses that could spill over to infect humans. In addition, our finding that potent cross-reactive mAbs were identified from relatively few B cells suggest that high-throughput screening of larger samples from animals immunized with mosaic-8 RBD-mi3 could be used to identify many new therapeutic mAbs, which could then be used to prevent or treat infections of Omicron and future SARS-CoV-2 variants. Finally, together with previous challenge and serum epitope mapping studies, 12 these results further validate mosaic-8 RBD-nanoparticles as a broadly protective vaccine candidate.

Limitations of the study
The new mAbs characterized here were derived from immunizations of mice, raising concerns that they could differ from human antibodies elicited by the same immunogens. For example, mouse antibodies generally have shorter CDRH3s than human antibodies. 33 The CDRH3 lengths of the 7 mouse mAbs we characterized structurally ranged from 9 to 16 amino acids (IMGT definition); 34 hence, these mAbs included CDRH3s equivalent to the average length of their human counterparts (15.5 ± 3.2 amino acids). 33 In addition, the class 1/4 and class 4 antibodies primarily elicited by mosaic-8 RBD-nanoparticle immunization 12 tend to rely less on long CDRH3s than, e.g., class 2 anti-RBD antibodies 10 that are less commonly elicited by these immunogens. Another concern is that the murine repertoire might lack V H and V L gene segments that provide humans with public responses against SARS-CoV-2 RBDs, 35 of which V H 3-53/V H 3-63, 36,37 V H 3-30, 17 and V H 1-2 38 antibodies have been described. However, epitope mapping of the anti-RBD antibodies including V H domains encoded by these V gene segments shows that they mainly target more variable RBD epitopes. 35,38 Thus, our working model is that the mouse humoral response to our immunogens is likely to be qualitatively similar to human responses, although particular V gene segments may differ. Future analyses are necessary to  6VYB) showing that the class 3 epitope is exposed, whereas the class 1/4 epitope is partially occluded in the context of the spike trimer. The binding epitopes of representative class 3 (S309/Sotrovimab, PDB: 7JX3) and class 1/4 (C118, PDB: 7RKV) anti-RBD antibodies were identified by PDBePISA. 28 (B) Class 3 anti-RBD mAbs that currently or previously received emergency use authorization (EUA) approval for human administration by the US Food and Drug Administration (modified from Zhou et al. 29 ) compared with M8a-28 (this study). Of the human mAbs, only LY-CoV1404/Bebtelovimab retains full neutralization potency against Omicron BA.1, 29 Figure S1).

OPEN ACCESS
Article directly compare antibodies raised in mice versus humans against mosaic-8 RBD-nanoparticle immunogens.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:  generated from crystal structures of M8a-34-RBD and HSW-2-RBD complexes have been deposited at the PDB under accession codes PDB: 7UZC and 7UZD, respectively. Additional information required to analyze the data reported in this paper is available from the lead contact upon request.
This paper does not report original code.
Expi293T cells (Gibco) for protein expression were maintained at 37 C and 8% CO 2 in Expi293 expression medium (Gibco). Transfections were carried out with an Expi293 Expression System Kit (Gibco) and maintained under shaking at 130 rpm. All cell lines were derived from female donors and were not specially authenticated.
Bacteria E. coli DH5 Alpha cells (Zymo Research) used for expression plasmid productions were cultured in LB broth (Sigma-Aldrich) with shaking at 250 rpm at 37 C.
E. coli BL21-CodonPlus (DE3)-RIPL (Agilent Technology) used for producing SpyCatcher003-mi3 were cultured in 2xYT media 220 rpm at 37 C, IPTG was added at OD of 0.5 and induction lasted for 5 hours at 30 C. conjugated mi3 particles were determined using a Bio-Rad Protein Assay (Bio-Rad). Conjugated nanoparticles were characterized by SEC, SDS-PAGE, and electron microscopy imaging as shown in Figures S1C-S1E, and by electron microscopy, SEC and dynamic light scattering previously. 12 For negative-stain electron microscopy imaging of mosaic-8 and homotypic SARS-CoV-2 RBD-nanoparticles: ultrathin, holey carbon-coated, 400 mesh Cu grids (Ted Pella) were glow discharged (60 s at 15 mA), and a 3 mL aliquot of SEC-purified RBD-nanoparticles was diluted to 40-100 mg/mL and applied to grids for 60 s. Grids were negatively stained with 2% (w/v) uranyl acetate for 30 s, and images were collected with a 120 keV FEI Tecnai T12 transmission electron microscope at 42,000x magnification.

Immunizations
Immunizations were done using protocols, #19023, approved by the City of Hope IACUC committee. Experiments were conducted using 4-6-week-old female C57BL/6 mice (Charles River Laboratories). Immunizations were carried out as previously described 15 using intraperitoneal injections of 5 mg of conjugated RBD-mi3 nanoparticle (calculated as the mass of the RBD, assuming 100% efficiency of conjugation to SpyCatcher003-mi3) in 100 mL of 50% v/v AddaVax TM adjuvant (Invivogen). Animals were boosted 4 weeks after the prime with the same quantity of antigen in adjuvant. A final booster was administered intraperitoneally 3 days before mouse spleen harvest.

Beacon
Plasma B cells were isolated from immunized animals for characterization on a Berkeley Lights Beacon instrument. Spleens were isolated from two immunized mice per condition and prepared into single cell suspensions as described. 15 Plasma B cells were isolated by CD138 + cell enrichment (Miltenyi Biotec CD138 + plasma cell isolation kit). Enriched plasma B cell samples were loaded onto an OptoSelect 11k chip (Berkeley Lights) in BLI Mouse Plasma Cell Media (Berkeley Lights). Single cells were then isolated in individual nanoliter-volume compartments (Nanopens using light-based OptoElectro Positioning (OEP) manipulation with settings optimized for plasma B cells. From Mosaic-8 RBD-nanoparticle immunized animals, 9,695 cells were penned in one chip, of which 7,747 were single cell pens. For homotypic SARS-CoV-2 RBD-nanoparticle immunized animals, 9,130 cells were penned in a second chip, of which 7,699 were single cell pens (Data S1). On chip fluorescence assays were used to identify cells secreting antibodies specific to RBD antigens. Briefly, C-terminally Avi-tagged RBDs were modified with site-specific biotinylation (Avidity) according to the manufacturer's protocol and immobilized on streptavidin-coated beads (Berkeley Lights). Assays were conducted by mixing beads coupled with one of four RBDs used for screening with a fluorescently labeled goat anti-mouse secondary antibody Alexa568 at 1:2500 dilution and importing this assay mixture into the OptoSelect 11k chip. Assays were conducted post 30 minutes incubation after cell penning at 36 C. Images were acquired every 5 minutes for 9 cycles while the beads remained stationary in the main channel above the Nanopens of the OptoSelect chip. Antibodies specific for the immobilized RBD bound the antigen-coupled beads, which sequestered the fluorescent secondary antibody, creating a ''bloom'' of fluorescent signal immediately above Nanopens containing plasma B cells. Beads were washed out of the chip, and this assay was conducted for each of the four RBDs. After completion of all assays, RBD-specific cells of interest were exported using OEP from individual nanopen chambers to individual wells of a 96-well PCR plate containing lysis buffer.
After running assays and selecting positive blooms with single cells, we ran the OptoSeq BCR Export workflow, which performs reverse transcription overnight on the chip and exports cell lysates containing cDNA on capture beads onto a 96 well plate. cDNA amplification and chain-specific PCR were performed the following day and run on an agarose gel to confirm that bands of the correct size were present. PCR products were then purified using AMPure XP magnetic beads and submitted for Sanger sequencing at the City of Hope Sequencing Core.

Cloning
Sequences for V H and V L domains were codon optimized using GeneArt (Thermo Fisher Scientific) and gene blocks for each domain were purchased from Integrated DNA Technologies (IDT). Expression constructs were assembled using Gibson reactions. 58,59 The heavy chain for IgG expression was constructed by subcloning the V H gene into a p3BNC expression vector encoding the human IgG C H 1, C H 2, and C H 3 domains, and the heavy chain for Fab expression was constructed by assembling the V H gene into a p3BNC expression vector encoding a human C H 1 and a C-terminal 6x-His tag. The expression plasmid for the light chain was constructed by subcloning the V L gene into a p3BNC vector that also encoded kappa human C L . The numbering of V H and V L protein sequences and the identification of the V gene segments were determined using the ANARCI server. 60 IgG and spike trimer production and purification Proteins were expressed in Expi293 cells by transient transfection. IgGs and a previously described human ACE2-Fc construct 11 were purified from cell supernatants using MabSelect SURE columns (Cytiva), and His-tagged Fabs were isolated from cell supernatants using Ni-NTA columns (Qiagen). IgGs, ACE2-Fc, and Fabs were further purified by SEC using a HiLoad 16/600 Superdex 200 column (Cytiva). Purified proteins were concentrated using a 100 kDa and 30 kDa cutoff concentrator (EMD Millipore), respectively, to 10 to 15 mg/mL, and final concentrated proteins were stored at 4 C until use. 6P versions 25 of soluble SARS-CoV-2 WA1 and SARS-CoV-2 Omicron BA.1 spike trimers were isolated from cell supernatants using a pre-packed Ni-NTA column (Cytiva). Eluents from Ni-NTA purifications were subjected to SEC using a HiLoad Superdex 200 16/600 column followed by a Superose 6 10/300 (Cytiva) column. Peak fractions were pooled and concentrated to 6 mg/ml, flash frozen in 50 mL aliquots, and stored at -80 C until use.
phenix.refine in Phenix 51 and Coot, 52 respectively. The refined models were subsequently used as input models for docking into cryo-EM maps of Fab-spike complexes.
Cryo-EM sample preparation SARS-CoV-2 S-Fab complexes were formed by incubating purified spike trimer and Fabs at a 1.1x molar excess of Fab per spike protomer at room temperature for 30 minutes to a final concentration of 2 mg/mL. Fluorinated octylmaltoside solution (Anatrace) was added to the spike-Fab complex to a final concentration of 0.02% (w/v) prior to freezing, and 3 mL of the complex/detergent mixture was immediately applied to QuantiFoil 300 mesh 1.2/1.3 grids (Electron Microscopy Sciences) that had been freshly glow discharged with PELCO easiGLOW (Ted Pella) for 1 min at 20 mA. Grids were blotted for 3 to 4 seconds with 0 blot force using Whatman No.1 filter paper and 100% humidity at room temperature and vitrified in 100% liquid ethane using a Mark IV Vitrobot (Thermo Fisher Scientific).
Cryo-EM data collection and processing Single-particle cryo-EM datasets for complexes of SARS-CoV-2 WA1 spike 6P with M8a-3 Fab, M8a-6 Fab, M8a-28 Fab, M8a-31 Fab, M8a-34 Fab or HSW-1 Fab and SARS-CoV-2 Omicron BA.1 spike 6P with M8a-31 Fab were collected using SerialEM automated data collection software 43 on a 300 keV Titan Krios (Thermo Fisher Scientific) cryo-electron microscope equipped with a K3 direct electron detector camera (Gatan). For SARS-CoV-2 WA1 spike 6P complexed with HSW-2, a dataset was collected with SerialEM 43 on a 200 keV Talos Arctica cryo-electron microscope (Thermo Fisher Scientific) equipped with a K3 camera (Gatan). Movies were recorded with 40 frames, a defocus range of -1 to -3 mm, and a total dosage of 60 e -/Å 2 using a 3x3 beam image shift pattern with 3 exposures per hole in the superresolution mode with a pixel size of 0.416 Å for the collections on the Krios and a single exposure per hole in the superresolution mode with a pixel size of 0.4345 Å for the collection on the Talos Arctica. Detailed data processing workflows for each complex structure are outlined in Data S1. All datasets were motion corrected with patch motion correction using a bining factor of 2, and CTF parameters were estimated using Patch CTF in cryoSPARC v3.2. 44 Particle picking was done with blob picker in cryoSPARC using a particle diameter of 100 to 200 Å , and movies and picked particles were inspected before extraction. Particles were extracted and classified using 2D classification in cryoSPARC. 44 After discarding ice and junk particles, the remaining particles were used for ab initio modeling with 4 volumes, which were futher refined with heterogenerous refinement in cryoSPARC. 44 Subsequent homogeneous and non-uniform refinements were carried out for final reconstructions in cryoSPARC. 44 Because Fab interactions with 'up' RBDs are generally not well resolved in Fab-spike complex structures, 18 we used masks to locally refine and improve the interfaces of Fabs bound to 'up' RBDs when necessary. For local refinements, masks were generated using UCSF Chimera 45 and refinements were carried out in cryoSPARC. 44 Cryo-EM structure modeling and refinement An initial model of the M8a-3 Fab-spike trimer complex was generated by docking a single-particle cryo-EM Fab-SARS-CoV-2 spike 6P complex structure (PDB 7SC1) into the cryo-EM density using UCSF Chimera. 45 The model was refined using real space refinement in Phenix. 51 The Fab amino acid seqence was manually corrected in Coot. 52 The model of the M8a-3 Fab-spike complex was subsequently used for docking and model generation for remaining Fab-spike trimer complexes. For the Fab-spike complexes that we have RBD-Fab crystal structures for (M8a-6 Fab-RBD, M8a-34 Fab-RBD and HSW-2 Fab-RBD structures), we first docked the spike trimer (PDB 7SC1) in the EM density map, manually fitted the RBDs in Coot 52 and refined the spike trimer using phenix.real_-space_refine. 51 The RBD-Fab structures were then aligned to each of the RBDs in the corresponding Fab-spike complexes, and the RBD regions in the EM model were replaced by the RBDs from crystal structures upon structural alignments in Coot. 52 The final model containing the spike trimer and the Fabs were subsequently refined with phenix.real_space_refine. 51 Iterative real space refinement and model building were separately carried out in Phenix 51 and Coot. 52 Single-particle cryo-EM refinement statistics are reported in Table S2.

Structure analyses
Structure figures were made using UCSF ChimeraX. 47,48 Distances were measured using PyMOL v2.4.0. 55 Interacting residues between a Fab and RBD were analyzed by PDBePISA 28 using the following interaction definitions: potential H bonds were defined as a distance less than 3.9 Å between the donor and acceptor residues when H was present at the acceptor and there was an A-D-H angle between 90 and 270 ; potential salt bridges were defined between residues that were less than 4 Å . Sequence alignments were done using Geneious (https://www.geneious.com/). Buried surface areas (BSAs) were calculated by PDBePISA using a 1.4 Å diameter probe. 28 To evaluate the potential for intra-spike crosslinking by the two Fabs of a single IgG binding to adjacent RBDs within a single spike trimer, we measured the distances between the Ca atoms of the C-terminal residues of the C H 1 domains of adjacent RBD-binding Fabs in the structures of mAb-spike complexes as described previously. 10 A cut-off of no more than 65 Å was used to identify IgGs whose binding orientation could allow for both Fabs to bind simultaneously to adjacent RBDs in a single spike trimer. This cut-off was larger than the distance measured between comparable residues of C H 1 domains in intact IgG crystal structures (42Å , PDB: 1HZH; 48Å , PDB: 1IGY; 52Å , PDB: 1IGT) to account for potential influences of crystal packing, flexibilities in the elbow bend angle relating the V H -V L and C H 1-C L , and uncertainties in the placements of C H 1-C L domains in cryo-EM structures of the Fab-spike complexes. 10