Rare, convergent antibodies targeting the stem helix broadly neutralize diverse betacoronaviruses

Summary Humanity has faced three recent outbreaks of novel betacoronaviruses, emphasizing the need to develop approaches that broadly target coronaviruses. Here, we identify 55 monoclonal antibodies from COVID-19 convalescent donors that bind diverse betacoronavirus spike proteins. Most antibodies targeted an S2 epitope that included the K814 residue and were non-neutralizing. However, 11 antibodies targeting the stem helix neutralized betacoronaviruses from different lineages. Eight antibodies in this group, including the six broadest and most potent neutralizers, were encoded by IGHV1-46 and IGKV3-20. Crystal structures of three antibodies of this class at 1.5–1.75-Å resolution revealed a conserved mode of binding. COV89-22 neutralized SARS-CoV-2 variants of concern including Omicron BA.4/5 and limited disease in Syrian hamsters. Collectively, these findings identify a class of IGHV1-46/IGKV3-20 antibodies that broadly neutralize betacoronaviruses by targeting the stem helix but indicate these antibodies constitute a small fraction of the broadly reactive antibody response to betacoronaviruses after SARS-CoV-2 infection.


In brief
The characteristics of antibodies that broadly neutralize coronaviruses are poorly understood. Here, Dacon et al. identify a class of stem helix-specific monoclonal antibodies from COVID-19 convalescent donors that neutralize diverse betacoronaviruses, use an IGHV1-46/IGKV3-20 gene signature, and bind in a conserved manner to the spike protein.

INTRODUCTION
Betacoronaviruses constitute one of four coronavirus genera and are a major cause of respiratory disease (V' Kovski et al., 2021). They can be divided into five subgenera, of which three currently contain members that are pathogenic to humans. HCoV-OC43 and HCoV-HKU1 are lineage A betacoronaviruses that cause mild upper respiratory disease, whereas MERS-CoV (lineage C), SARS-CoV, and SARS-CoV-2 (lineage B) are responsible for severe outbreaks that led to a large number of  (Fung and Liu, 2019). SARS-CoV-2, the causative agent of COVID-19, has claimed more than six million lives since the first cases emerged in late 2019 (Dong et al., 2020). The currently dominant SARS-CoV-2 Omicron subvariant BA.5 is resistant to most monoclonal antibody (mAb) therapeutics available in the clinic (Yamasoba et al., 2022;Takashita et al., 2022b). Furthermore, other betacoronaviruses infect a range of animal species that regularly come into contact with humans, increasing the possibility of future zoonotic spillover (Peck et al., 2015). Therefore, there is an urgent need to develop vaccines and therapeutic mAbs that broadly target betacoronaviruses.
The major immune target on the coronavirus surface is the spike protein, a homotrimeric type I viral fusion protein that is composed of two subunits, S1 and S2 (Li, 2016). The S1 subunit uses either its N-terminal domain (NTD) or C-terminal domain (CTD) as the receptor-binding domain (RBD) to engage host cell receptors. Following receptor engagement, the S2 subunit undergoes conformational rearrangements to bridge and fuse the virus and host cell membrane, allowing the release of virus genetic material into the host cell cytoplasm. The SARS-CoV-2 spike protein is the target of currently available COVID-19 vaccines and therapeutic mAbs (Edwards et al., 2022). Although these vaccines are predominantly based on whole-spike constructs, most of the neutralizing antibody response following immunization is thought to be directed against the RBD. Similarly, all therapeutic mAbs available for public use target this domain. However, given the poor conservation of the RBD across different betacoronavirus lineages (Li, 2015), these vaccines and therapies are unlikely to be effective against betacoronaviruses that are distantly related to SARS-CoV-2. Instead, more conserved regions of the spike protein may be more suitable for the design of vaccines that cover a wider range of betacoronaviruses.
Here, we performed an epitope-agnostic screen to identify mAbs that broadly neutralize betacoronaviruses, with the goal of studying the nature of these antibodies and the characteristics of their target epitopes. We found that the majority of broadly reactive mAbs were non-neutralizing and bound to an epitope that included the K814 residue. However, 11 mAbs targeted the conserved stem helix in the S2 subunit and cross-neutralized betacoronaviruses from different subgenera, highlighting the importance of this site as a target of neutralizing antibodies in conjunction with reports from previous studies (Li et al., 2022;Pinto et al., 2021;Sauer et al., 2021;Zhou et al., 2022b). Eight of these mAbs, isolated from multiple donors, used the same germline gene combination of IGHV1-46/ IGKV3-20. Crystal structures of three Fab-peptide complexes of antibodies COV89-22, COV30-14, and COV93-03 revealed that they all targeted the stem helix in a similar way. Two IGHV1-46/IGKV3-20 mAbs, COV89-22 and COV72-37, limited disease in the Syrian hamster model. In summary, these data suggest that the broadly reactive antibody response to betacoronaviruses after SARS-CoV-2 infection largely focuses on an immunodominant, weakly neutralizing site, but a minor part of this response consists of broadly neutralizing mAbs with shared gene usage that target the stem helix. Therefore, stem helix-specific vaccine constructs that elicit this antibody class may be an efficient way to generate protective antibody responses to betacoronaviruses, including all SARS-CoV-2 variants of concern.

Identification of mAbs that broadly neutralize betacoronaviruses
To isolate mAbs with broad reactivity, we selected 19 COVID-19 convalescent donors that had plasma reactivity to diverse betacoronaviruses from a previously described cohort (Cho et al., 2021). A total of 673,671 IgG + and 305,142 IgA + memory B cells (MBCs) from these donors were screened in a two-step workflow that utilized sequential oligoclonal and monoclonal B cell culture to downselect B cells of interest. Recombinant mAbs were screened for binding to spike protein from the betacoronaviruses SARS-CoV-2, SARS-CoV, MERS-CoV, HCoV-HKU1, and HCoV-OC43, as well as from the alphacoronaviruses HCoV-NL63 and HCoV-229E. We isolated six mAbs that targeted multiple coronavirus genera by binding the conserved fusion peptide, as recently described (Dacon et al., 2022). From this screen, we also obtained a panel of 54 IgG mAbs and one IgA mAb that were broadly reactive to betacoronaviruses but were mostly unreactive to alphacoronavirus spike proteins, with a few exceptions. All 55 mAbs bound to both SARS-CoV-2 and SARS-CoV spike, and the overwhelming majority of mAbs (53 of 55) also bound to HCoV-OC43 spike (Figure 1). Furthermore, 70.9% (n = 39) of the mAbs bound to the spike proteins of all five human-infecting betacoronaviruses. We next screened the 55 broadly reactive mAbs in neutralization assays against SARS-CoV-2, SARS-CoV, MERS-CoV, and HCoV-NL63 envelope pseudoviruses, as well as authentic HCoV-OC43, to assess the breadth and potency of their neutralization. Eighteen mAbs neutralized at least one virus, among which the mAbs COV89-22, COV30-14, COV72-37, COV44-26, and COV44-74 neutralized all four of the human betacoronaviruses tested ( Figure 1). However, the majority of mAbs were non-neutralizing, and, consistent with their spike binding being largely restricted to the betacoronavirus genera, none of the 55 mAbs neutralized the alphacoronavirus HCoV-NL63 even at the highest concentration tested (100 mg/mL).

Broadly neutralizing mAbs against betacoronaviruses target the stem helix
To identify the spike domain targeted by these broadly reactive mAbs, we tested the mAb panel for binding to SARS-CoV-2 RBD, NTD, S1, and S2. Flow cytometry analyses revealed the SARS-CoV-2 spike S2 subunit as the target of the majority of the mAbs ( Figure S1A). Subsequent surface plasmon resonance (SPR)-based epitope binning analysis demonstrated that the mAbs could be separated into two groups that were distinct from control mAbs targeting the fusion peptide ( Figure S1B) (Dacon et al., 2022). The mAbs sorted into Group A (n = 11) competed for epitope binding with a previously described stem helix-targeting mAb S2P6 (Pinto et al., 2021), whereas mAbs sorted into Group B (n = 40) bound to a separate epitope on the S2 subunit. To further investigate the specific binding sites of these mAbs, we performed peptide mapping using an array of overlapping 15-mer biotinylated peptides spanning the S2 subunit of SARS-CoV-2. Consistent with the epitope binning analysis, antibodies in Group A bound to peptides covering the The area under the curve (AUC) from titration of mAb binding to spike proteins from human betacoronaviruses SARS-CoV-2 Wuhan-Hu-1 (CoV-2), SARS-CoV (CoV-1), MERS-CoV, HCoV-HKU1, and HCoV-OC43, as well as alphacoronaviruses HCoV-NL63 and HCoV-229E, is shown on the panel of the left. Influenza H1 hemagglutinin (HA) was included as a control antigen and L9 IgG1 (malaria specific; Wang et al., 2020) was included as a negative control mAb for binding experiments. AUC values for each antigen are shown after subtraction with values for the negative control antigen CD4. The antibody titers at 50% neutralization (NT 50 ) against SARS-CoV-2 Wuhan-Hu-1, SARS-CoV, MERS-CoV, HCoV-NL63 envelope-pseudotyped virus, as well as authentic HCoV-OC43, are shown on the right. Neutralizing mAbs are ranked by their breadth of neutralization and the geometric mean of their NT 50 values. Cells highlighted in blue denote mAbs that did not show neutralizing potency at the highest concentration tested (100 mg/mL). Negative control mAbs for neutralization are DEN3 (dengue-specific; Rogers et al., 2020) for SARS-CoV-2, SARS-CoV, MERS-CoV and HCoV-NL63, and CV503 (SARS-CoV-2 RBD-specific; Cho et al., 2021) for HCoV-OC43. NT 50 values were calculated using the doseresponse-inhibition model with 5-parameter Hill slope equation in GraphPad Prism version 9.3.1.
Group B did not bind to any of the 15-mer peptides ( Figure S2A), suggesting that these antibodies recognize a conformational epitope within the S2 subunit. To identify this epitope, we utilized a shotgun mutagenesis approach, wherein S2 subunit residues were individually mutated to alanine in the context of the whole SARS-CoV-2 spike protein to generate a panel of spike mutants. We screened three mAbs in Group B,, against this panel and identified a single amino acid, K814, as critical for binding of all three mAbs ( Figure S2B). K814 is located at a poorly characterized site just N-terminal to the S2 0 cleavage site and fusion peptide region and is part of a loop that extends to the side of the spike protein ( Figure 2B). This residue has also been recently identified as a target of two other SARS-CoV-2 S2-specific mAbs, suggesting that this is a common recognition site (Chen et al., 2021). We named this site K814+, as the epitope recognized by the Group B mAbs most likely encompasses more than K814, but no surrounding amino acid was clearly identified as a target of all three mAbs from this group ( Figure S2B).
We compared the neutralization breadth of the Group A and B mAbs to determine the utility of each S2 site as a neutralizing epitope. Strikingly, 10 of 11 of the Group A (stem helix-specific) A B D B C Figure 2. Broadly neutralizing mAbs against betacoronaviruses target the stem helix (A) Heatmap of SARS-CoV-2 Wuhan-Hu-1 S2 peptide array. SPR was used to measure binding responses to 15-mer peptides (x axis, 3-aa offset) spanning the SARS-CoV-2 Wuhan-Hu-1 S2 subunit. Open triangle indicates the S1/S2 cleavage site, closed triangle indicates the S2 0 cleavage site; FP, fusion peptide: HR1, heptad repeat 1; C Helix, central helix; CD, connector domain; SH, stem helix; HR2, heptad repeat 2. (B) Sequence conservation of native SARS-CoV-2 spike protein (PDB: 7N1Q) using sequence alignment of 28 betacoronaviruses representing each of the 5 subgenera. Insets show K814 and surrounding residues, as well as the stem helix region. Generated using ChimeraX. (C) Number of betacoronavirus lineages neutralized by group A (stem helix) and group B (K814+) mAbs. (D) Alignment of stem helix region of betacoronavirus spike proteins using the MAFFT v7.0 software and L-INS-i algorithm. Percent identity of amino acid residues was calculated using only betacoronavirus isolates. See also Figures S1, S2 (legend continued on next page) mAbs were capable of neutralizing at least two different betacoronavirus lineages ( Figure 2C). Moreover, the broadest neutralizing mAbs isolated in this study  all belonged to Group A. In contrast, the majority of Group B mAbs did not neutralize a single betacoronavirus, and only a single mAb from this group cross-neutralized coronaviruses from two different lineages. These findings suggest that the majority of broadly reactive mAbs against betacoronaviruses target the K814+ site and are poorly neutralizing, whereas a minority target the stem helix and are capable of broadly neutralizing betacoronaviruses. Therefore, we decided to focus our efforts on further characterizing the Group A, stem helix-specific mAbs. We performed a sequence alignment of 28 isolates representing the five betacoronavirus subgenera to determine the degree of conservation of the stem helix sequence among betacoronaviruses (Figures 2B and 2D). In particular, amino acids F1148, E1151, K1157, and N1158 within the stem helix are highly conserved (>90%) within the betacoronavirus subgenera, which is consistent with the breadth observed in the stem helix-specific mAbs (Figure 2C). All SARS-CoV-2 variants of concern identified to date, including the Omicron subvariant BA.5, have identical sequences in this region. The human alphacoronaviruses HCoV-229E and HCoV-NL63 have divergent sequences at this location, which explains the lack of binding and neutralization of the Group A mAbs to these viruses (Figures 1 and 2D).
Stem helix-specific mAbs from multiple donors use an IGHV1-46/IGKV3-20 gene signature To investigate the genetic profile of the stem helix-specific mAbs, we examined their heavy and light chain V gene usage, as well as their complementarity-determining region 3 (CDR3) amino acid sequences. Interestingly, 10 of 11 mAbs targeting the stem helix used an IGHV1-46 heavy chain ( Figure 3A). In eight mAbs, this heavy chain was paired with an IGKV3-20 light chain. Of the remaining 44 broadly reactive mAbs, only one used IGHV1-46 and a different mAb used IGKV3-20. This V gene preference was not due to the expansion of a single B cell clone or V gene bias from a single donor, as the 10 mAbs were isolated from six different donors and only three (COV44-26, COV44-54, and COV44-74) were clonally related. The VH nucleotide mutation levels of the IGHV1-46 stem helix mAbs were between 4.8% and 12.2%, and the VH amino acid mutations were between 9.3% and 21.4%, indicative of prior experience in a germinal center ( Figure S3A). A comparison of the heavy chain CDR3 sequences of the IGHV1-46 mAbs revealed that this group of mAbs had divergent HCDR3 sequences, supporting a role for IGHV1-46-specific elements, such as HCDR1 and HCDR2, in binding to the stem helix ( Figure 3B). The light chain CDR3s of the IGHV1-46/IGKV3-20 mAbs were more similar ( Figure 3B), but this was unsurprising, given the large contribution of IGKV3-20 residues to this region.
Next, we compared the potency and breadth of the IGHV1-46/ IGKV3-20 mAbs to the other broadly reactive mAbs. Notably, the six most potent and broadly neutralizing mAbs in our panel used the IGHV1-46/IGKV3-20 combination, and all eight mAbs in this group neutralized at least two betacoronaviruses ( Figure 3C). These findings suggest that the ability to produce IGHV1-46/ IGKV3-20 mAbs is advantageous for immune defense against betacoronaviruses. To determine the frequency of B cells using these V genes, we used the iReceptor database (https:// gateway.ireceptor.org/) (Corrie et al., 2018) to screen published next-generation B cell receptor sequencing datasets from healthy individuals where at least 1 million rearranged sequences were obtained (Briney et al., 2019;DeKosky et al., 2015DeKosky et al., , 2016DeWitt et al., 2016;Tipton et al., 2015). At least 0.65% of B cells (average 2.1%) in each individual (n = 21) used IGHV1-46 (Figure 3D). Only six donors met our criteria for VL gene analysis, but at least 7.05% (average 11.2%) of B cells in each donor used IGKV3-20 ( Figure 3D). Furthermore, a separate study that performed deep sequencing of VK genes in four individuals (Jackson et al., 2012) found that IGKV3-20 was the most common kappa gene in all donors. Collectively, these findings suggest that IGHV1-46 and IGKV3-20 are commonly used individually by B cells in healthy individuals, although their combination would have a lower probability.
Of the eight IGHV1-46/IGKV3-20 stem helix-specific mAbs, COV44-26, COV44-54, and COV44-74 belonged to the same clonal lineage, allowing us to investigate the effects of affinity maturation on binding to the stem helix. We evaluated the binding of the putative unmutated common ancestor (UCA) and intermediates of this lineage to the SARS-CoV-2 spike protein and peptide 154 from the stem helix ( Figures S3B and S3C). The UCA bound well to the stem helix peptide and was able to bind to the SARS-CoV-2 spike protein, albeit more weakly than all the other members of the lineage, which bound similarly to the spike and stem helix. When comparing the sequences of these mAbs, the HCDR2 stood out as a region where the UCA was substantially different from the other members of the clonal lineage ( Figure S3D). We also produced VJ germline-reverted versions of the potent IGHV1-46/IGKV3-20 mAbs COV89-22, COV30-14, and COV72-37 and compared the characteristics of the germline and mature versions of these mAbs ( Figures  S3E and S3F). The germline mAbs were capable of binding the SARS-CoV-2 stem helix and betacoronavirus spikes (with the exception of the COV72-37 and COV89-22 germlines with the MERS-CoV spike) but were mostly non-neutralizing. In contrast, the mature forms of the mAbs were superior in both binding and neutralization. Taken together, these results suggest that naive B cells carrying IGHV1-46/IGKV3-20 are capable of engaging the stem helix, but somatic mutations increase both binding and neutralization potency of this class of mAbs.
We conducted an alanine scan on the stem helix peptide to determine whether the IGHV1-46/IGKV3-20 mAbs preferentially (F) Effect of mutations on COV89-22 binding in a shotgun mutagenesis assay. A residue was considered critical if mutation of this residue resulted in a reduction of binding signal for COV89-22 but not control mAb C, which targets a linear epitope not in this region (see Figure S3H). (G) Sequence logo plot of spike protein from 28 aligned betacoronavirus isolates representing each of the 5 subgenera. Amino acid residues are colored by hydrophobicity. Numbering is based on the SARS-CoV-2 Wuhan-Hu-1 sequence. Yellow stars indicate amino acids absent in SARS-CoV-2 Wuhan-Hu-1 Spike protein but present in at least one other sequence used for the alignment. Created using Weblogo3.0. See also Figure S3.
formed contacts with a distinct set of amino acids from the other mAbs targeting this site ( Figure 3E). We also used the spike shotgun mutagenesis assay to further examine the binding profile of COV89-22, a high-affinity binder and the most potent mAb in our panel ( Figures 3F, S3G, and S3H), since it was less susceptible to mutations in the context of the stem helix peptide. There was no clear difference between the binding profiles of COV44-74 and COV49-51, which use IGKV3-20, and COV44-37 and COV44-56, which use IGKV2-28, suggesting that the light chains were more permissive for specificity of these mAbs toward the stem helix. Overall, residues F1148, L1152, and F1156 were important for the majority of IGHV1-46 mAbs, whereas the sole non-IGHV1-46 mAb, COV77-09, only required F1156. F1148 is conserved in all betacoronavirus sequences examined, whereas L1152 and F1156 are conserved in >80% of the sequences (and with similar amino acid types as mutations) ( Figures 2D and 3G). When we examined SARS-CoV-2 spike sequences from the GISAID database (https://gisaid.org/) (Elbe and Buckland-Merrett, 2017) for mutation frequencies at these positions, we found that mutations at F1148, L1152, and F1156 were only present in 0.0002%, 0.0004%, and 0.002% of all sequences, respectively. To determine the effects of mutations at these positions on virus fitness, we produced pseudoviruses carrying single F1148A, L1152A, or F1156A mutations, or a triple F1148A/ L1152A/F1156A mutation, in parallel with wild-type (WT) pseudovirus produced at the same time using the same protocol. We compared the infectivity of each undiluted pseudovirus prepara-  Table S1. tion and observed a clear reduction in the infectivity of all mutants ( Figure S3I). To determine if this was due to a defect in pseudovirus production (e.g., due to spike misfolding) or the infective capacity of intact virions (or both), we quantified the p24 antigen concentration of each preparation. There was a clear reduction in p24 concentration for all mutants ( Figure S3J), and when the infectivity was normalized based on this count, only F1148A and the triple mutant showed substantially reduced function ( Figure S3K). Collectively, the data indicate that a mutation at each of the three positions (F1148, L1152, and F1156) impairs virus production, with F1148A further reducing the infectivity of the virions that are produced.
Crystal structure of three IGHV1-46 stem antibodies in complex with the stem helix peptide To decipher how COV89-22, COV30-14, and COV93-03 interact with the S2 stem helix in neutralizing SARS-CoV-2, the Fabs of these three antibodies were complexed with the 15-mer peptides 154 or 155, which cover the stem helix region (Figures 2A  and 4). The crystal structures of the COV89-22/COV30-14/ COV93-03-peptide complexes were determined at 1.6, 1.5, and 1.75-Å resolution, respectively (Figures 4 and S4; Table  S1). All residues of peptide 154 and thirteen of fifteen residues of peptide 155 were visible in the electron density maps (Figure S4A). Eleven residues of both peptides had a buried surface area (BSA) > 0 Å 2 in the interface with antibody ( Figure S5A). These antibodies share the same IGHV and IGKV germlines (IGHV1-46/IGKV3-20) as another anti-stem helix antibody S2P6 (Pinto et al., 2021). They contain 15/12/12 a.a. somatic mutations in the heavy chain and 8/11/5 a.a. in the light chain variable regions (VH/VL) of COV89-22/COV30-14/COV93-03, respectively ( Figures S6A and S6B). These three antibodies contact the stem helix peptide via CDR1, CDR2, and CDR3 in the heavy chain and CDR1 and CDR3 in the light chain ( Figure 4A). The BSA of each residue of the stem helix peptide exhibited a similar distribution among COV89-22, COV30-14, and COV93-03 ( Figure S5A). Furthermore, the molecular surface contact area reveal that they share very similar contact patterns for the main chain and side chain of the peptides among COV89-22/ COV30-14/COV93-03 ( Figure S5B). Among the three Fab-peptide complexes, F1148, L1152, Y1155, F1156, and H1159 of the peptide make hydrophobic interactions with a largely hydrophobic groove in the antibody composed of common residues from the heavy and light chains ( Figures 4B, 4C, and S5C). The aromatic residues, Y/H91-Y96 motif in LCDR3 and Y32 in LCDR1, create a hydrophobic cavity to accommodate F1148, L1152, and Y1155, consistent with the substantial loss of binding with Ala mutations in the spike protein ( Figures 3E and 3F). Furthermore, the RRNY residues (29-32) of LCDR1 along with S93 in LCDR3 in COV89-22 form a network of H-bonds and salt bridges with E1151 of the stem helix peptide ( Figure 5A) that accounts for a decrease in binding to E1151A in the spike protein ( Figure 3F). The equivalent TGRY and TSNY residues of LCDR1 in COV30-14 and COV93-03 contribute H-bonds but no direct salt bridges with stem helix peptide ( Figure 5A). In addition, D1153 of the stem helix peptide hydrogen bonds with Y33 and also forms backbone-backbone interactions between residues 1148 and 1149 with residue 97 in HCDR1 and HCDR3 in COV89-22 and COV30-14 ( Figure 5B). In COV93-03, the Y96 sidechain replaces the residue 97 interaction but here contrib-utes two H-bonds with D1153. HCDR3 Y96 also enhances hydrophobic and aromatic interactions among Y33, L1152, Y1155, and F1156. These findings suggest that all known IGHV1-46/IGKV3-20 antibodies mainly target the region F1148 to F1156 of the SARS-CoV-2 stem helix with a highly similar binding mode, and the key residues are consistent with those identified by the Ala scanning of the spike protein. Notably, IGHV1-46 encodes residues that contact the stem helix including the CDR2 I50, whereas IGKV3-20 (and the closely related IGKV3D-20) encodes the YGSSP motif, which includes key contact residues, consistent with the frequent use of these genes by stem helix-specific mAbs ( Figure S6C). In complex with the antibodies here, the monomeric stem helix peptide forms a helix as observed in both pre-fusion and post-fusion states of the SARS-CoV-2 spike ( Figures S4B and S4C). However, antibody binding to this region would clash with the three-helix bundle in the stem region in the pre-fusion state and in the post-fusion state, which suggests that binding to the spike requires a conformational change or increased dynamics from its pre-fusion form to a more open state, perhaps along a trajectory toward its post-fusion form.

Response to stem helix following vaccination and infection
To investigate antibody responses to the stem helix after COVID-19 vaccination or SARS-CoV-2 infection, we screened polyclonal IgG isolated from serum or plasma from the following donors for binding to peptide 154 from the stem helix region: individuals vaccinated with mRNA-1273 (Moderna)  Table S1.
( Figure S7A), unvaccinated donors recovering from a recent SARS-CoV-2 infection, and unvaccinated COVID-19-naive donors. COVID-19-naive individuals had negligible antibodies to this peptide, indicating a minimal contribution from previous infections by seasonal betacoronaviruses such as HCoV-HKU1 and HCoV-OC43 ( Figure S7B). There was an increase in the level of stem helix-specific antibodies after the second vaccination (p < 0.001), but this rapidly declined and was not restored by the booster dose ( Figure S7B). As a group, the convalescent individuals had higher responses than the naive donors (p = 0.0049) but did not have higher responses than the vaccinated individuals (p = 1). Overall, vaccination with mRNA-1273 and natural infection did not induce high levels of antibodies against the stem helix region.  Table S2.
Stem helix-specific mAbs neutralize SARS-CoV-2 variants of concern and inhibit fusion We tested three of the most potent stem helix-specific mAbs, COV89-22, COV30-14, and COV72-37, for their ability to neutralize SARS-CoV-2 variants of concern. The three mAbs neutralized all variants tested including Omicron BA.4/5 ( Figure 6A), which is consistent with the identical sequence of this region in all variants of concern ( Figure 2D). We also confirmed that these mAbs neutralized authentic SARS-CoV-2 and MERS-CoV, as well as a panel of betacoronaviruses in a second pseudovirus assay ( Figure 6B; Table S2). We then proceeded to test if these mAbs inhibit fusion of cells expressing SARS-CoV-2 spike protein and cells expressing ACE2, which is a potential mechanism of action of stem helix-specific mAbs (Li et al., 2022;Pinto et al., 2021;. We found that COV89-22, COV30-14, and COV72-37 inhibited fusion in both an imaging-based and quantitative assay, wherein fusion results in the release of an enzyme that cleaves a chromogenic substrate ( Figures 6C and 6D).

COV89-22 and COV72-37 limit disease in SARS-CoV-2infected Syrian hamsters
We tested COV89-22 and COV72-37 for the ability to limit disease in the Syrian hamster model of SARS-CoV-2 infection (Figure 7). To allow for optimal Fc function, we converted the Fc regions of the two mAbs to hamster IgG2. Each mAb was administered intraperitoneally at a 16-mg/kg dose, followed by intranasal infection with 10 5 plaque-forming units (PFU) of SARS-CoV-2 one day later. Disease progression in the hamsters (n = 12 per group) was monitored through daily assessment of changes in body weight, as well as histopathology measurements on days 3 and 7. As expected from previous studies, untreated hamsters lost around 10% of body weight through day 6 post-infection ( Figure 7A) (Cho et al., 2021). In contrast, hamsters treated with COV89-22 and COV72-37 maintained body weight similar to the uninfected controls throughout the study (p < 0.001 relative to untreated hamsters from days 2-7 for both mAbs). Most hamsters in the COV89-22 and COV72-37 groups showed only mild signs of interstitial pneumonia A B C Figure 7. COV72-37 and COV89-22 limit SARS-CoV-2-mediated disease in Syrian hamsters (A and B) Clinical outcomes of SARS-CoV-2 exposed Syrian hamsters after prophylaxis with stem helix mAbs. (A) Weight change was assessed using a mixedeffects repeated measures model with Dunnett's post-test multiple comparison (n = 12 animals from days 0-3 and n = 6 animals from days 4-7), and error bars represent mean ± SD. (B) Pneumonia pathology distribution scores on days 3 and 7 were analyzed by a Kruskal-Wallis test with Dunn's post-test multiple comparison (n = 6-12 animals per condition), between the mAb-treated and mock-treated groups on each day. Bars show median ± interquartile range. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001 and ns, not significant.
(C) Images of sagittal sections of the left lung lobe from untreated Syrian hamsters and those administered COV89-22 and COV72-37. Scale bars, 4 mm.
based on histopathological examination of lung tissue (median score 1 or less), consistent with the body weight data ( Figures  7B and 7C). Hamsters treated with COV89-22 showed a reduced pathology score from days 3 to 7, whereas the control group worsened during this period ( Figure 7B). Collectively, these findings suggest that COV89-22 and COV72-37 are effective in limiting disease in this model of SARS-CoV-2 infection.

DISCUSSION
In this study, we identify convergent IGHV1-46/IGKV3-20 mAbs from several individuals that target the S2 stem helix for broad neutralization of betacoronaviruses. Betacoronavirus-neutralizing mAbs targeting this region have previously been identified, including the IGHV1-46/IGKV3-20 mAb S2P6 (Pinto et al., 2021), affirming the importance of this region as a target site of neutralizing antibodies (Li et al., 2022;Pinto et al., 2021;Sauer et al., 2021;Zhou et al., 2022aZhou et al., , 2022b (Scheid et al., 2011;Wu et al., 2011), and SARS-CoV-2 RBD (IGHV3-53/IGHV3-66) ( Barnes et al., 2020;Yuan et al., 2020a). Here, the IGHV1-46/IGKV3-20 combination is relatively uncommon in the wider mAb panel as it is not used by any of the K814-+-specific mAbs, which constitute the majority of the mAbs described in this study. However, IGHV1-46/IGKV3-20 is used very frequently (72.7%) by mAbs targeting the stem helix, suggesting positive selection due to favorable binding to the stem helix (Zhou et al., 2022a). Accordingly, we found that VJ germline-reverted versions of these mAbs are capable of binding to the stem helix, suggesting that naive B cells using IGHV1-46/ IGKV3-20 already have the ability to target this epitope. Nevertheless, both binding and neutralization are improved with somatic mutations, highlighting the importance of the germinal center reaction in enhancing the antibody response to this site. This study provides information that could be useful for the design of next-generation coronavirus vaccines. The continuous emergence of new SARS-CoV-2 variants of concern that evade neutralizing antibody responses has provided strong motivation to develop vaccines that target more conserved regions of the spike protein. The S2 subunit, which is more conserved than S1, is currently being explored as a candidate for this purpose (Shah et al., 2021). However, these data suggest that the K814+ site in S2 is immunodominant and triggers broadly reactive but not broadly neutralizing antibodies. In contrast, the stem helix elicits fewer antibodies, perhaps due to limited accessibility in the pre-fusion spike but is a better target for eliciting neutralizing antibodies. Of a panel of 55 broadly reactive mAbs toward betacoronaviruses, the 11 mAbs that targeted the stem helix were also the 11 most potent mAbs based on the average NT 50 value against a panel of four betacoronaviruses. Therefore, a targeted construct that focuses the immune response on the stem helix and avoids the immunodominant K814+ site or an S2 construct that masks this site may be promising for design of a broad betacoronavirus vaccine. Moreover, the crystal struc-ture and mutagenesis data show the precise binding mode of potent stem helix-specific mAbs and identify key stem helix residues that must be included in the vaccine construct to elicit the desired antibody response. The conserved mode of interaction of the IGHV1-46/IGKV3-20 mAbs with the stem helix can serve as a template for the design of germline-targeting immunogens that aim to activate these B cell lineages.
The major drawback of mAbs targeting the stem helix is their lower in vitro neutralization potency relative to the RBD-specific mAbs that have been developed as clinical products (Dougan et al., 2021;Gupta et al., 2021;Weinreich et al., 2021;Takashita et al., 2022a). However, the stem helix-specific mAbs are more likely to retain function against new SARS-CoV-2 variants than mAbs targeting the RBD, which has shown the ability to accumulate diverse mutations without substantial or any loss of binding to ACE2 (Starr et al., 2020). Furthermore, in vitro potency does not always reflect efficacy in humans, as other factors such as Fc activity also contribute to protection (Bartsch et al., 2021). For instance, the therapeutic mAb sotrovimab has lower in vitro potency than most other therapeutic mAbs in the clinic but showed similar efficacy in preventing progression to severe COVID-19 disease in humans (Dougan et al., 2021;Gupta et al., 2021;Weinreich et al., 2021;Takashita et al., 2022a). The stem helix-specific mAbs described here, in particular COV89-22, were effective in preventing disease mediated by SARS-CoV-2 in a hamster model. Therefore, stem helix-specific mAbs and vaccine constructs should be further explored as countermeasures that could be immediately utilized for protection from future SARS-CoV-2 variants or novel betacoronaviruses.
Limitations of the study As mentioned above, the stem helix-specific mAbs described here have lower potency than potent RBD-specific mAbs, which have NT 50 values in the ng/mL range. These mAbs will have to be further characterized to determine whether they are potent enough to be used to prevent COVID-19 or reduce the risk of progression to severe disease in humans. Furthermore, we only evaluated the in vivo efficacy of these mAbs against SARS-CoV-2, and it is unclear if they also function in vivo against other betacoronaviruses, although we note that previously described mAbs with similar specificity showed in vivo function against MERS-CoV (Wang et al., 2021;Zhou et al., 2022a). Although these data are useful for vaccine design, we have not performed vaccination experiments in this study and thus cannot draw any definitive conclusions with regard to the efficacy of stem helix-based vaccines. Whether a stem helix-based vaccine can elicit sufficient antibody titers to neutralize betacoronaviruses in humans remains to be investigated. Although the stem helix is well conserved in betacoronaviruses and has an identical sequence in all SARS-CoV-2 variants of concern, it cannot be guaranteed that a new variant of concern with mutations in this region will not emerge in the future. This will have to be closely monitored.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We thank the blood sample donors at the New York Blood Center; Sandhya Bangaru, Gabriel Ozorowski, Alba Torrents de la Peñ a, and Andrew Ward for providing HCoV-OC43 and MERS-CoV spike; Henry Tien for technical support with automated crystalization; Jeanne Matteson for contributions to mammalian cell culture; Robyn Stanfield for technical assistance; Gavin Wright and Nicole Muller-Sienerth (Wright lab, University of York) for providing recombinant CD4; Melanie Cohen and Julie Laux for assistance with cell sorting; Walter Hardy for assistance in screening the Ala scan library; the staff of Stanford Synchrotron Radiation Laboratory (SSRL) Beamline 12-1 and the Advanced Photon Source (APS) Beamline 23-ID-B for assistance. We thank Nick Vaughan, Kurt Cooper, Becky Reeder, Marisa St Claire, Kyra Hadley, Amanda Hischak, Randy Hart, and Nejra Isic for assistance with hamster experiments. We thank Jiro Wada for assistance with graphics for histopathology lung sections. We gratefully acknowledge all data contributors, i.e., the authors and their Originating laboratories responsible for obtaining the specimens, and their Submitting laboratories for generating the genetic sequence and metadata and sharing via the GISAID Initiative. This work was supported by the Di-

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Joshua Tan (tanj4@nih.gov).

Materials availability
Antibodies described in this manuscript are available through a Materials Transfer Agreement (MTA) with the National Institute of Allergy and Infectious Diseases. Plasmids generated in this study have been deposited to Addgene.
Data and code availability d Crystal structures have been deposited into the Protein Data Bank (PDB: 8DTR, 8DTT, 8DTX for COV30-14, COV93-03 and COV89-22, respectively). Antibody sequences have been deposited in GenBank (accession numbers OP377774-OP377795). d This paper does not report original code. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Study cohort
Anonymized samples of whole blood and plasma from COVID-19 convalescent patients were obtained from a previously described cohort (Cho et al., 2021). Inclusion criteria included an age of 18 years or above and RT-PCR confirmation of SARS-CoV-2 infection. All samples were collected at least 2 weeks after resolution of symptoms, and all donors signaled consent by signing the standard New York Blood Center (NYBC) blood donor consent form. Participants met inclusion criteria and assented to provide samples. Of these, samples from 19 were selected for inclusion in this study following analysis of plasma IgG reactivity.
Whole blood, plasma and serum samples were collected from recipients of the SARS-CoV-2 mRNA-1273 vaccine (Moderna) at the NIH Clinical Research Center in Bethesda, MD under protocols approved by the NIH Institutional Review Board, ClinicalTrials.gov identifiers: NCT00001281 and NCT05078905. Inclusion criteria for the vaccine study were age (R 18 years), HIV status (negative), no known history of SARS-CoV-2 infection (verified by nucleocapsid antibody responses), and no previous doses of COVID-19 vaccines. 16 participants met inclusion criteria and provided written informed consent to have their blood products used for research purposes. Blood samples were collected serially at baseline (prior to receiving the initial vaccine dose), 30 d after administration of the second dose, pre-booster (third dose) baseline, and 30 d after administration of the booster. A further blood sample was also collected from 3 participants at 30 d after documented SARS-CoV-2 infection. Samples were not randomized or blinded, but were anonymized.
Cell culture Memory B cells (MBCs) were derived from cryopreserved peripheral blood mononuclear cells (PBMCs) by flow sort and cultured in IMDM (Gibco, 31980-030) supplemented with 10% HI-FBS (Gibco, 10438-026), 100 ng/mL IL21 (Gibco, PHC0211), 0.5 mg/mL R848 (Invivogen, tlrl-r848) and 13 Mycozap (Lonza, VZA-2021). Sf9 and High Five cells were cultured in HyClone insect cell culture medium (GE Healthcare, SH30280.03). Sf9 cells were seeded and incubated at 28 C in T25 and T175 flasks for bacmid transfection and generation of baculoviruses, respectively. High Five cells were incubated at 28 C with shaking at 110 rpm for 72 h for protein expression. Irradiated 3T3-CD40L cells were generated as previously described (Moir et al., 1999;Huang et al., 2013) and cryopreserved for use in MBC cultures. FreeStyle 293-F cells were cultured in Freestyle 293 Expression media (ThermoFisher Scientific, 12338018). HeLa cells were cultured in DMEM (Lonza, 08028) supplemented with 10% FBS, 13 penicillin/streptomycin and Glutamax. Rhabdomyosarcoma cells were cultured in DMEM (Gibco, 11966) supplemented with 10% HIFBS, 4500 mg/mL glucose, 1 mM sodium pyruvate, 1 mM HEPES and 50 mg/mL gentamicin (Quality Biological, 120-098-661) and cultured in a T225cm 2 flask at 37 C and 5% CO 2. HuH7.5 cells (provided by Dr. Deborah R. Taylor, US FDA), used to propagate MERS-CoV pseudovirus for use in neutralization assays, were cultured in DMEM with 10% BSA, 2 mM glutamine and 13 penicillin/streptomycin (D10). 293 flpin-TMPRSS2-ACE2 cells (provided by Dr. Adrian Creanga, VRC/NIH) were cultured in D10 with 100 mg/mL hygromycin. Hamster model Golden Syrian hamsters were sourced from Envigo (Indianapolis, IN USA). Animals were acclimated at IRF facility for 10 d and weighed 2 d prior to study commencement. Group selection was made at 5-6 weeks, assigning groups based on weight. Equal ll OPEN ACCESS Article numbers of males and females were assigned to each of eight groups of n = 12 according to weight. Animal research was conducted under an IACUC approved protocols at the IRF in compliance with the Animal Welfare Act and other federal statutes and regulations relating to animals and experiments involving animals. The facilities where this research was conducted are fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, International and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 2011.

METHOD DETAILS
Coronavirus spike proteins Expression of SARS-CoV-2 spike, SARS-CoV-2 RBD, SARS-CoV-2 NTD, SARS-CoV spike and SARS-CoV RBD has been described elsewhere (Cho et al., 2021;Lv et al., 2021;Yuan et al., 2020b). Briefly, the RBDs were cloned into an in-house pFast-Bac vector, fused with a gp67 signal peptide and an His 6 tag flanking the N-and C-terminus of the RBD. The bacmids were generated via Bac-to-Bac system (Life Technologies). The bacmid was then transfected into Sf9 cells. High (5 to 10) multiplicity of infection (MOI) of baculovirus-infected High Five cells (Life Technologies) was achieved according to the manufacturer's manual to produce RBD and spike proteins. The supernatant of the infected High Five cells was harvested around 72 h post-infection at 28 C with shaking at 110 rpm.
The cloning, expression and purification of the recombinant HA were performed as described in previous studies (Dreyfus et al., 2012;Ekiert et al., 2009). The hemagglutinin ectodomain (11-329 of HA1 and 1-174 to HA2, in H3 numbering) from A/Solomon Islands/03/2006 (H1N1)  was linked to an N-terminal gp67 signal peptide and to a C-terminal BirA biotinylation site, thrombin cleavage site, T4 trimerization domain and 6xHis-tag of a customized pFastBac vector. Recombinant bacmid DNA was generated using the Bac-to-Bac system (Life Technologies). The bacmid was transfected into Sf9 cells using FuGENE HD (Promega) to generate baculovirus. The baculovirus was subsequently used to infect High Five cells (Life Technologies) at the MOI of 5 to 10. High Five cells were then incubated at 28 C and shaking at 110 rpm for 72 h for HA expression. The recombinant HA was purified by Ni-NTA resin followed by size exclusion chromatography, buffer exchanged into 20 mM Tris, 150 mM NaCl, pH 8.0, and concentrated for the binding assay.

Sequence alignment of coronaviruses
To evaluate the conservation of the primary protein structure of spike, a multiple sequence alignment was performed using the following full-length sequences: . Sequences were aligned using the MAFFT v7 server using a BLOSUM62 scoring matrix and L-INS-I algorithm. The sequence alignment was used to generate a sequence logo plot using the Weblogo 3.0 server (Crooks et al., 2004;Schneider and Stephens, 1990) and to color conserved amino acid residues on a fulllength spike protein (PDB: 7N1Q) using Chimera X.
For analysis of mutation frequencies at positions F1148, L1152 and F1156 of the SARS-CoV-2 spike protein, 5,604,512 high-quality SARS-CoV-2 spike sequences from the GISAID database (https://gisaid.org/; Complete, High Coverage options selected) were retrieved on August 24, 2022. Sequences with multiple stop codons were excluded from the mutant count.

Optofluidic-based isolation of B cells
Flow-sorted MBCs (CD19 + IgA + /IgG + ) were mixed with irradiated feeder cells (irr-3T3-CD40L cells). 100 mL of this cell suspension was dispensed to each well of a 384-well plate (50 MBCs, 3000 feeders per well), and cultures incubated at 37 o C and 5% CO 2 for 10 d. On day 9, culture supernatants were collected and analyzed for reactivity against multiplexed CoV antigen beads by flow cytometry. From these data, culture wells of interest were specified. On day 10, MBCs from these wells of interest were pooled. These cells were washed in MACS buffer (0.5% w/v BSA in PBS with 2mM EDTA), and approximately 2.3 3 10 4 cells were loaded onto an OptoSelect 11k chip (Berkeley Lights). This chip was loaded into a Beacon analyzer and each individual B cell sorted into its own nanoliter-volume pen by the action of OEP light cages. 7 mm streptavidin beads (Spherotech, SVP-60-5) coated with 10 mg/mL of both MERS-CoV spike and OC43-CoV spike were incubated with 2.5 mg/mL goat anti-human IgG-Alexa Fluor 647 (Jackson Immunoresearch 109-606-170) and goat anti-human IgA-Cy3 (Jackson Immunoresearch 109-166-011). These beads were then immobilized in the channels of the OptoSelect 11k chip. Binding of secreted antibody from penned MBCs to the beads was detected in the CY5 channel (indicating IgG binding) or the TRED channel (indicating IgA binding); images from these channels were captured at 6 min intervals over a 30 min total time course. In a second step of this assay, MERS/OC43 beads were washed out of the chip and replaced with beads bound instead to SARS-CoV-2 spike that were otherwise prepared in the same manner. Antibody binding was again monitored by fluorescent image capture. This two-step procedure allowed for the identification of MBCs producing bona fide cross-reactive antibodies. These select MBCs were exported out of pens, again by the action of OEP light cages, and delivered directly into individual wells of a 96-well plate, where they were immediately lysed by Dynabeads mRNA DIRECT lysis buffer (Life Technologies, 61011). Plates were sealed, snap-frozen on dry ice, and placed in a -80 o C freezer until required. mAb sequence analysis and expression RT-PCR was performed on MBC lysates to amplify heavy and light chain sequences (Cho et al., 2021;Wang et al., 2020;Tiller et al., 2008) of cross-reactive antibodies (PCR primers from Wang et al., 2020). Sequences were then resolved by Sanger Sequencing (Eurofins and Genewiz). The software Geneious Prime (Version 2021.0.3, https://www.geneious.com) was then used for analysis of VH and Vl/Vk genes, CDR3 sequences, and percentage of somatic mutations, with reference to the International Immunogenetics Information System database (IMGT) (Lefranc, 2014). VJ-germline sequences were obtained by reverting the V and J genes to the closest germline based on the IMGT database. The chord diagram showing the relationship between antibody and light chains was generated using the circlize package in R (Gu et al., 2014). Pairs of VH and Vl/Vk sequences were matched and subsequently commercially cloned into plasmids containing an IgG1 or relevant light chain backbone, and expressed as recombinant antibody (Genscript). mAbs were also expressed in-house by transient transfection of Expi293 cells (ThermoFisher Scientific, A14527) using the ExpiFectamine 293 Transfection Kit (ThermoFisher Scientific, A14524) according to manufacturer's instructions. These recombinant antibodies were purified using HiTrap Protein A columns (Cytiva/GE Healthcare Life Sciences, 17040303). Sequence alignment of CDR3 heavy and light chains was performed using MAFFT v7 server using a G-INS-I algorithm. Amino acid residues were colored according to physicochemical properties. COV44-26, COV44-54 and COV44-74 were determined to be the same lineage based on the following criteria: same heavy chain and light chain V genes, >90% identity in CDR3 amino acid sequence. Lineage analysis, including inference of the unmutated common ancestor (UCA) and putative intermediates of the COV44-26, COV44-54 and COV44-74 clonal family, was performed using Cloanalyst (Kepler, 2013). For inferred lineage members containing the ambiguous nucleotide r, the nucleotide g was used (matching the germline) to allow translation and expression as recombinant antibodies. Lineage trees were visualized using the Interactive Tree of Life (iTOL) (Letunic and Bork, 2021).
HLA typing of donor cDNA HLA typing was necessary to identify the source of mAbs isolated from screens of donor-pooled B cells. Amplified cDNA from single cell isolates was subjected to an amplicon-based sequencing by synthesis approach using a commercially available ScisGoÒ-HLA-v6 kit (Scisco Genetics Inc., Seattle WA). This protocol uses a two-stage amplicon-based PCR for locus amplification and sample barcoding. Although this kit is designed for amplification from genomic DNA, a portion of kit amplicons was functional to amplify product from cDNA. Briefly, samples were sequentially subjected to two-stage PCR amplification following manufacturer's instructions, after which reactions were combined, purified, and applied to a MiSeq using Illumina Version 2 chemistry with 500-cycle, paired-end sequencing (Illumina, San Diego, CA). Data were assembled and analyzed using specially-adapted SciscloudÒ (Scisco Genetics Inc., Seattle WA) computational tools for the assembly of HLA genomic sequences derived from the ScisGoÒ-HLA-v6 kit. This software was made available as part of the kit. HLA class I and II genes could then be compared with typing data taken for each donor prior to sample processing, allowing for unambiguous identification of corresponding samples. mAb binding to coronavirus antigens Four-fold serial dilutions of recombinant mAbs in 0.5% BSA w/v in PBS, for a final dilution series of 47.7 pg/mL -200 mg/mL, were incubated with multiplexed CoV antigen beads at room temperature for 30 min. Beads were then washed and stained with 2.5 mg/mL goat anti-human IgG Alexa Fluor 647 (Jackson Immunoresearch, 109-606-170). Samples were acquired on the iQue Screener Plus (Intellicyt) and resulting data were analysed with FlowJo (Version 10.8.1). Titration curves and AUC analyses were performed on GraphPad Prism (Version 9.3.1); values were reported after subtraction of binding values to the negative control antigen CD4. For binding of mAbs to the SARS-CoV-2 spike and stem helix peptide (peptide 154), the L9 negative control curves are the same in Figures S3B and S3E.
V gene usage survey The iReceptor database (https://gateway.ireceptor.org/home) was surveyed to assess the frequency of circulating B cells expressing VH and VL genes of interest in healthy human donors. Study data were queried and downloaded from the AIRR Data Commons (Christley et al., 2020) using the iReceptor Gateway (Corrie et al., 2018). Only healthy donors with large datasets (R1310 6 sequences) were included, from a total of five studies (Briney et al., 2019;DeKosky et al., 2015;DeKosky et al., 2016;DeWitt et al., 2016;Tipton et al., 2015).

Epitope binning by SPR
For epitope binning, cross-reactive mAbs were coupled to a HC30M chip (Carterra) and analysed by the Carterra LSA. The running buffer used was 0.05% BSA w/v in HEPES-buffered saline with Tween-20 and EDTA (HBSTE). Chip conditioning involved successive injections of 50 mM NaOH, 500mM NaCl and 10mM glycine pH 2, before priming with MES supplemented with 0.05% Tween. The primed chip was then activated with a 1:1 mixture of 400 mM EDC and 100 mM NHS (ThermoFisher Scientific) immediately prior to direct coupling of 10 mg/mL of mAbs in pH 4.5 acetate buffer onto discrete spots on the chip. Excess chip binding sites were blocked with 1M ethanolamine, pH 8.5. 100 nM SARS-CoV-2 S2 subunit was pre-mixed in a 1:1 ratio with 2 mM of each sandwiching antibody and the mAb-spike complexes were then injected onto the array. After each sandwiching antibody injection, the chip was regenerated by three successive injections of 10 mM glycine pH 2.0. Binning data were analyzed with Epitope Software (Carterra).

SARS-CoV-2 S2 binding kinetics
Fab fragments were prepared using the Pierce Fab Preparation kit (Thermo Fisher Scientific, 44985) following the manufacturer's protocol with slight modifications. Briefly, 250-500 mg of each cross-reactive mAb was digested using immobilized papain for 3 h at 37 o C. The resulting digest was applied to Protein G Hi-Trap spin columns (Cytiva) to purify Fabs from Fc fragments and undigested mAbs. Residual reducing agent was removed using a Zeba Desalting Column 7K MWCO (ThermoScientific, 89882). Protein concentrations were determined using A280 measurements and Fab digests were confirmed using reducing and non-reducing SDS-PAGE.
For analysis of antibody binding kinetics, Fabs were coupled to an HC30M chip (0.56 mg/mL) and a three-fold dilution series of SARS-CoV-2 S2 subunit or spike protein was injected in ascending concentration without regeneration. A 10 min association and 30 min dissociation time were used. Association (k a ) and dissociation rates (k d ), as well as dissociation constants (K D ) were calculated using the Kinetics Software (Carterra).

SARS-CoV-2 S2 peptide mapping
Lyophilized 15-mer peptides that carried an N-terminal biotin tag with 12 amino acid overlap were synthesized (JPT Peptide Technologies) to span the SARS-CoV-2 S2 subunit (Ser686 -Lys1211, Accession #YP_009724390.1). Additionally, eight biotinylated peptides from H1 haemagglutinin protein were included as negative controls. 1 mg/mL peptide stocks were prepared in DMSO, then peptides were diluted to 0.1 mg/mL in 0.05% BSA w/v in HBSTE and captured onto SAD200M streptavidin-coated chips (Carterra). Cross-reactive mAbs were successively injected onto the peptide array at 10 mg/mL and regenerative binding was measured with 5 min association phase followed by a 1 min dissociation phase. Regeneration was achieved using three successive injections of 10 mM glycine pH 2.0 following each antibody injection. Data were analyzed using the Epitope Software (Carterra). To perform alanine scan experiments, the wild-type sequence 1142 QPELDSFKEELDKYFKNHTS 1161 and variants with alanine substitutions at each amino acid position were synthesized with modifications described above. Biotinylated peptides were captured to a streptavidin-coated chip (Carterra) and regenerative binding was measured as described above.

Fab expression for crystallization
The variable domains of heavy chain (VH) and light chain (VL) of COV89-22, COV72-37 and COV30-14 were codon optimized (Genscript) and fused with an N-terminal secreting signal peptide and a human Fab expressing vector. The Fab was expressed by co-transfection of heavy and light chain plasmids at a 2:1 ratio (in weight) in the ExpiCHO expression system (Life Technologies) according to the Max Titer protocol in the manufacturer's manual. Supernatants were harvested, centrifuged, and purified with