Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Nature Biotechnology (2024 )Cite this article crispr cas9 vector
Monoclonal antibodies are increasingly used to prevent and treat viral infections and are pivotal in pandemic response efforts. Antibody-secreting cells (ASCs; plasma cells and plasmablasts) are an excellent source of high-affinity antibodies with therapeutic potential. Current methods to study antigen-specific ASCs either have low throughput, require expensive and labor-intensive screening or are technically demanding and therefore not widely accessible. Here we present a straightforward technology for the rapid discovery of monoclonal antibodies from ASCs. Our approach combines microfluidic encapsulation of single cells into an antibody capture hydrogel with antigen bait sorting by conventional flow cytometry. With our technology, we screened millions of mouse and human ASCs and obtained monoclonal antibodies against severe acute respiratory syndrome coronavirus 2 with high affinity (<1 pM) and neutralizing capacity (<100 ng ml−1) in 2 weeks with a high hit rate (>85% of characterized antibodies bound the target). By facilitating access to the underexplored ASC compartment, the approach enables efficient antibody discovery and immunological studies into the generation of protective antibodies.
In contrast, antibody-secreting cells (ASCs; plasma cells and plasmablasts) have been undervalued as a rich source of highly specific antibodies because of technological limitations and their limited accessibility in peripheral blood3. Antibodies isolated from ASCs are thought to have on average higher affinity than those derived from memory B cells7,8,9. Additionally, ASCs are responsible for the active humoral immune response, that is, they secrete the circulating antibodies that confer protection against invading pathogens10. Interrogating this compartment therefore brings us closer to understanding the secreted antibody repertoire and can help bridge the gap between proteomic profiling of plasma antibodies and bulk sequencing of the B cell repertoire11.
However, studying the specificity of single ASCs is difficult because they secrete their antibodies and express few or no Igs on the surface12. For this reason, ASCs cannot be interrogated by conventional FACS in an antigen-specific manner with high throughput. Traditionally, the ASC compartment has therefore been interrogated by either unbiased plasmablast sorting13,14, which does not give information about the antigen specificity of single cells, or enzyme-linked immunospot (ELISpot) assays, which enumerate the frequency of antigen-specific cells but are not able to recover their genotype15.
Because the percentage of suitable antigen binders within the plasmablast population can be low, unbiased sorting can require substantial time and resource investment to screen antibodies from all cells for validation of binding. Recently, approaches based on compartmentalization of single cells in water-in-oil emulsion droplets have been applied in antibody discovery16,17,18,19,20 and sequencing21,22,23. Alternatively, antibody discovery from ASCs can be performed with commercial optofluidic systems24. However, functional assays in droplets are technically demanding, and standalone machines are prohibitively expensive, therefore both approaches are not accessible to the wider research community.
We address the limitations of current ASC screening methods by enabling the high-throughput interrogation of antigen-specific ASCs by conventional FACS. In our workflow, we first use droplet microfluidics to encapsulate single cells into an antibody capture hydrogel at 107 cells per h, creating a stable capture matrix around the cell that enables the concentration of secreted antibodies and simple addition and removal of detection reagents. We then use the multiplexed detection and high-throughput sorting capabilities of FACS to isolate antigen-specific ASCs for single-cell sequencing and recombinant antibody expression. The modular nature of the method enables its extension to other secreted molecules by simple replacement of capture and detection reagents.
We demonstrate the utility of this approach by screening millions of primary immune cells to isolate mAbs against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) from mouse and human ASCs. Tapping into these repertoires allowed us to collect a diverse pool of sequences of secreted antibodies. Of a representative subset of human antibodies, 95% bound the respective antigen, many with subnanomolar affinities and high neutralizing capacities (<100 ng ml−1), highlighting the benefits of interrogating the locus of the active humoral response. Our technology can generate pathogen-specific antibodies within 2 weeks, both democratizing and fast-tracking the development of antibody drug candidates.
We have addressed this challenge by compartmentalizing single ASCs into an antibody capture hydrogel by automated droplet microfluidics (at a rate of up to 107 cells per h), followed by the selection of the secreted antibody specificity with fluorescently labeled antigens by FACS at a similar speed (Fig. 1a). This combination of microfluidics and FACS enables the high throughput that is crucial for the success of antibody discovery campaigns.
a, Overview of the workflow—B cells (or enriched ASCs) are isolated from mice (bone marrow or spleen) or human PBMCs. Cells are mixed with liquid BG-agarose at 37 °C and encapsulated into picolitre water-in-oil emulsion droplets using a flow-focusing junction. Droplets are collected on ice for agarose gelation and demulsified, creating stable hydrogel beads around each cell. The BG-agarose is converted into an antibody capture matrix by the addition of recombinant capture reagents that are fused to the SNAP-tag, an enzyme that reacts with BG moieties. During incubation, antibodies secreted by a single cell are captured in the hydrogel surrounding the cell. Cells that have secreted antigen-specific antibodies are identified with fluorescently labeled detection reagents (antigens, secondary antibodies and antibodies against cell-surface markers), sorted using flow cytometry and sequenced. Antibody sequences can be obtained within 4 days, and recombinant antibodies for testing are generated in 2 weeks. b, Agarose-based antibody capture matrix. Agarose is chemically modified to contain BG moieties that react covalently with the SNAP-tag. Single-domain antibodies (VHHs) against the constant region of antibody light chains are expressed as SNAP-tag fusions and immobilized in the BG-agarose hydrogel, creating the capture matrix. c, Antibody capture by BG-agarose hydrogel beads functionalized with VHH–SNAP. Antibody capture (anti-streptavidin mouse IgG) and antigen binding (streptavidin–GFP) were analyzed by flow cytometry. The plot shows at least 220 events per condition at a 5% contour level. d, Antibody secretion by single OVA-specific mouse bone marrow plasma cells. Representative confocal microscopy image from two independent experiments showing a single cell encapsulated in VHH-functionalized BG-agarose stained with fluorescently labeled OVA (AF555), anti-CD138 antibodies (AF647) and anti-mouse IgG antibodies (AF405). e, Sorting of OVA-specific mouse bone marrow plasma cells. Hydrogel beads containing plasma cells that secreted OVA-specific IgG were sorted by FACS (gated as live/FLAG+/CD138+/IgM−/IgG+). The plots show 10,168 (CD138+) and 411 (IgG+) events at 2% contour level. f, Characteristics of mouse anti-OVA antibodies—variable domain genes (V and J), third complementarity-determining region amino acid sequences (CDR3s) and equilibrium dissociation constants (KD).
Specifically, B cells are mixed with liquid benzylguanine (BG)-agarose (at 37 °C) and encapsulated into monodisperse water-in-oil emulsion droplets of 25 µm diameter at kilohertz rates. The droplets are collected on ice (to solidify the agarose) and demulsified, creating a stable agarose microcompartment around each cell. BG-modification creates covalent attachment sites for the SNAP-tag, a 20 kDa engineered human O6-alkylguanine-DNA alkyltransferase that reacts specifically, quickly and irreversibly with different BG-containing substrates25. We synthesized BG-agarose by chemical modification of commercial low-melting-point agarose in a simple, two-step procedure (Extended Data Fig. 1a). To create an antibody capture matrix, we functionalized the BG sites with recombinant SNAP-tag fusion proteins of single-domain antibodies (VHHs) that bind to the constant region of antibody light chains (Fig. 1b; see Extended Data Fig. 2a for detailed considerations regarding the design of the capture matrix)26,27. This capture approach is highly modular, as any protein fused to the SNAP-tag can be immobilized in the agarose matrix. Through covalent functionalization with antibody capture reagents, antibodies secreted by encapsulated cells are immobilized around the cell, physically linking the secreted antibody and the cell of origin. With two different VHHs, recognizing either κ or λ light chains, any secreted antibody (for example, IgG, IgM and IgA) from any ASC can be captured efficiently. Light-chain-mediated capture enables multivalent binding in the hydrogel (one IgG molecule has two constant light chains). This capture mechanism leads to substantial avidity effects that significantly decrease the apparent rate constant for the dissociation (koff; Extended Data Fig. 2b–d) and increase the half-life of the VHH–antibody complex. Consequently, we expect minimal dissociation of captured antibodies from the VHHs during the course of the experiment. We have chosen VHHs as antibody capture reagents due to their small size, high affinity, stability and straightforward expression as SNAP-tag-fusions in Escherichia coli. Additionally, the chosen VHHs do not exhibit binding to antibodies from other species, which ensures their compatibility with antibody-based flow cytometry stainings26,27.
In the hydrogel, antibodies and cells are accessible for staining with fluorescently labeled detection reagents and antigens due to the large pore size of the agarose (>100 nm28,29), while unbound reagents can be removed by washing. Additionally, the small size (25 µm diameter) and monodispersity of the hydrogel beads enable sorting by conventional FACS devices. Hydrogel beads can therefore essentially be stained, washed and sorted like single cells, but they also provide additional information about the secreted antibody (secreted amount, isotype and antigen specificity). Finally, single-sorted cells in hydrogel beads are sequenced to enable recombinant expression and characterization of antigen-specific antibodies. FACS allows the detection of several phenotypic parameters or different antigens in one experiment and the correlation of the sequence of single ASCs with phenotypic antibody information through index sorting.
We estimated the capture capacity of the BG-agarose by reacting BG-agarose beads (25 µm diameter, 1.5% (wt/vol) agarose) with different quantities of a GFP-SNAP-tag fusion protein (GFP-SNAP), followed by the analysis of the GFP fluorescence of the beads by flow cytometry (Extended Data Fig. 1b). We observed saturation of the GFP signal at over 109 immobilized GFP molecules per bead. In the literature, average secretion rates of single ASCs range from 103 to 105 antibodies per s30,31, indicating that a completely functionalized capture matrix would not become saturated even after several hours of antibody secretion. Notably, in our system, the number of antibody capture sites is a function of bead size and the amount of added VHH–SNAP and is therefore well-defined and controllable. In contrast to capture on the cell surface32,33, antibody capture capacity does not depend on cell-intrinsic properties such as cell-surface molecules and should therefore be relatively uniform across the ASC population.
Next, we showed that light-chain-mediated capture enables both the interrogation of antigen binding and the detection of immobilized antibodies by flow cytometry (Fig. 1c). We captured a commercial anti-streptavidin antibody (IgG) in VHH-functionalized BG-agarose beads and stained the beads with both streptavidin–biotin–GFP and anti-IgG antibodies. Analysis by flow cytometry confirmed that captured antibodies can simultaneously bind to the functionalized agarose, the antigen, and the detection antibodies. Consistent with these findings, the epitopes of VHHs, anti-IgG antibodies and antigens do not overlap (Extended Data Fig. 2e,f).
With a capture system for secreted proteins at hand, we established an antibody discovery workflow with primary immune cells. To determine cell survival in the hydrogel and optimal duration of antibody secretion, we isolated mouse bone marrow plasma cells (CD138+) by positive magnetic selection and encapsulated them into BG-agarose. We found that the encapsulation process did not affect the viability of CD138+ cells and that cells remained highly viable (92%) in the hydrogel even 4 h after encapsulation (Extended Data Fig. 3a). To establish the time window for secretion, we functionalized the BG-agarose hydrogel with anti-mouse κ VHH–SNAP and monitored the immunoglobulin G1 (IgG1) secretion signal over time by flow cytometry (Extended Data Fig. 3c,d). We found that the percentage of IgG-secreting plasma cells increased over time, while the background IgG signal of beads that did not contain a plasma cell (CD138−) increased only considerably after 1 h and 45 min of incubation. Non-encapsulated CD138+ cells were IgG1-negative because mouse bone marrow plasma cells do not express surface IgG34. We measured IgG levels in the supernatant by ELISA and found that the antibody levels generated by encapsulated cells slightly increased over time compared to background but remained outside of the quantifiable range of the assay and were considerably lower than in the supernatant of not encapsulated cells (Extended Data Fig. 3b). Based on the secretion time course data, we believe that the method is robust for a time window of 30 min to 1 h and 45 min as it is possible to identify a distinct IgG+ population in these samples while still maintaining a low level of background. For antibody selections, we chose 1 h and 45 min as this longer secretion time enables the detection of a larger proportion of ASCs while still being short enough to fit the antibody selection experiment into a single day.
We then immunized mice with hen egg ovalbumin (OVA) to obtain OVA-specific B cells (Extended Data Fig. 4). To visualize secretion and capture of OVA-specific antibodies, we stained cells compartmentalized into VHH-functionalized hydrogels with fluorescently labeled antibodies against CD138, secondary antibodies against IgG and fluorescently labeled OVA. By confocal fluorescence microscopy, we observed colocalization of IgG and OVA signal in the hydrogel around the cell, indicating secretion of OVA-specific IgG (Fig. 1d). The IgG and OVA signal distribution in the bead and the lack of IgG signal on empty beads suggest no substantial cross-contamination (Extended Data Fig. 4a,b and Supplementary Fig. 4).
To obtain OVA-specific antibody sequences for recombinant expression, we encapsulated bone marrow-derived plasma cells for sorting by FACS (Extended Data Fig. 4c). After incubation, hydrogel beads were stained with fluorescently labeled monomeric OVA, antibodies against B cell and ASC surface markers (B220 and CD138), an anti-FLAG antibody for labeling of VHH-functionalized hydrogels, as well as anti-IgG and IgM secondary antibodies. We hypothesized that in the absence of VHH, secreted antibodies are removed during washing steps due to the low background binding and large pore size of the agarose. Only samples to which VHH–SNAP had been added contained an IgG+ population, confirming VHH-specific retention of antibodies and excluding the staining of cell-surface Igs (Extended Data Fig. 4d).
Plasma cells secreting OVA-specific antibodies (Fig. 1e and representative gating strategy in Extended Data Fig. 4e) were sorted into single wells of 96-well plates, and antibody variable genes were obtained by RT–PCR followed by Sanger sequencing. We recombinantly expressed two antibodies and found that they both bound OVA with high affinity using biolayer interferometry (BLI) (KD values of 2.36 nM and 0.68 nM, respectively; Fig. 1f and Extended Data Fig. 4f), suggesting that high-affinity binders can be identified by our workflow. In addition to bone marrow-derived plasma cells, we also analyzed OVA-specific spleen-derived ASCs (Extended Data Fig. 5). With this setup, we can therefore screen millions of cells in a matter of hours and generate high-affinity mAbs from primary mouse immune cells.
We then assessed our ability to isolate high-affinity antibodies against a relevant pathogen antigen. We immunized mice with the RBD of the wild-type (WT) SARS-CoV-2 spike protein, enriched plasma cells from bone marrow and used them in our workflow (Fig. 2a). Plasma cells secreting RBD-specific IgG antibodies were sorted with fluorescently labeled RBD-streptavidin tetramers into single wells of 96-well plates for Sanger sequencing (Fig. 2b and gating strategy in Extended Data Fig. 6b). Secreted antibodies could be identified by comparing the IgG signal of samples with and without VHH addition (Extended Data Fig. 6a).
a, Mouse immunization and analysis scheme. Bone marrow plasma cells (CD138+) were magnetically enriched and then used in our workflow. RBD-specific plasma cells were sorted with fluorescently labeled RBD-streptavidin tetramers. b, Sorting of RBD-specific mouse plasma cells. Cells were gated as live/CD138+, and IgG-secreting RBD-specific plasma cells were sorted by FACS. The plots show 79,629 (CD138+) and 1,623 (IgG+) events at 2% contour level. c, Overview of antibody sequences of sorted plasma cells. In total, 54 paired heavy- and light-chain sequences were obtained. The pie chart shows the 21 observed HV and LV gene combinations (HV–LV). HV–LV pairings are colored by the HV gene. Combinations that were characterized are shown in darker shades, while combinations that were not expressed are shown in lighter shades. The three most expanded expressed HV–LV combinations are highlighted with their CDRH3 amino acid sequence and frequency. d, Summary of anti-RBD ELISA. The plot shows the EC50 with an antibody concentration range of 0.0002–400 nM. Antibodies that did not bind RBD at 400 nM are shown at an arbitrary EC50 of 1,000 nM (gray diamonds). e, Characteristics of mouse anti-SARS-CoV-2 RBD antibodies with neutralizing capacity—variable domain sequences (V and J genes), third complementarity-determining region amino acid sequences (CDR3), equilibrium dissociation constants (KD) and IC50 against WT SARS-CoV-2. f, In-tandem epitope binning experiment with mRBD1 and mRBD2. g, Crystal structure of mRBD2 with SARS-CoV-2 RBD (PDB: 8BE1). Top left, the RBD (green) in complex with mRBD2 Fab fragment (purple and pink for light and heavy chains) is superimposed with RBD complexed with ACE2 (gray; PDB: 6M0J), showing how the Fab fragment overlaps significantly with ACE2. The main figure shows details of the RBD loop (green carbon atoms) binding to the CDRs of the mRBD2 Fab fragment.
To confirm SARS-CoV-2 RBD binding and perform further functional profiling, we recombinantly expressed 26 representative antibodies as full-length mouse IgG1 (Supplementary Table 1 and full-length sequences in Supplementary Table 2). In an indirect ELISA against RBD, we found that 22 antibodies (84.6%) bound RBD at the tested concentrations (up to 400 nM antibody; Fig. 2d and Extended Data Fig. 6c), including the representatives of all five of the most expanded clones.
To determine whether the RBD-specific binders were also able to neutralize SARS-CoV-2, we used luminescent reporter cells to perform live virus neutralization assays37,38,39. Despite the high apparent affinity of our plasma cell clones, only two antibodies (mRBD1 and mRBD2 (belonging to the fifth most expanded clone)) were able to neutralize WT SARS-CoV-2 with an IC50 of 3.17 µg ml−1 and 4.61 µg ml−1, respectively (Fig. 2e and Extended Data Fig. 6d), highlighting that selection of an antibody in an immune response does not guarantee neutralizing capacity.
To further characterize mRBD1 and mRBD2, we determined their binding affinities to RBD (Extended Data Fig. 6e) and performed in-tandem epitope binning assays by BLI (Fig. 2f). The binning assay indicated that the two antibodies do not share the same binding site on RBD, as an increase in BLI signal was observed when antibodies were added sequentially to immobilized RBD.
To provide further insights into the binding mechanism of mouse ASC-derived RBD-binding antibodies36,40, we obtained a co-crystal structure of the Fab fragment of mRBD2 with RBD. The structure shows that mRBD2 binds to a long disulfide-bonded loop (residues 474–489) on the RBD that is part of the angiotensin-converting enzyme 2 (ACE2)-binding interface (Fig. 2g). In the antibody, the binding site is formed of CDR2 and CDR3 from both light and heavy chains. Phe487 and Tyr489 of the RBD interact with CDR2 of the heavy chain, and Asn487 is hydrogen bonded to both heavy and light chains. Ser477 hydrogen bonds to heavy-chain CDR3, while Thr478 hydrogen bonds to light-chain CDR3. The conformation of the RBD loop itself is almost identical to that seen in the ACE2:RBD complex while moving through a hinge-like motion by a few Ångström relative to the rest of the RBD (Supplementary Fig. 5). Comparison of mRBD2 with the ACE2-blocking mouse plasmablast-derived antibody 2B04 (ref. 40) showed that their binding sites on RBD partially overlap, hinting at similar neutralization mechanisms.
To demonstrate that our technology has utility in pandemic response efforts, we adapted our protocol to isolate antibodies directly from humans. mAbs derived from humans after infection or vaccination have been shown to be efficacious and well-tolerated in the human body, representing a fast track to therapeutic development3. Furthermore, studying antigen-specific plasmablasts gives unique insights into the postinfection and vaccine responses of different individuals41,42.
To adjust the workflow for the capture of human antibodies, we exploited the modular setup of the capture system and exchanged the anti-mouse VHH with anti-human κ and λ VHHs27. First, we activated normal peripheral blood mononuclear cells (PBMCs) in vitro to stimulate antibody secretion43 and were able to detect antibody secretion and capture from single cells (Extended Data Fig. 7). The compatibility of the workflow with ex vivo-activated cells highlights the versatility of the technology and expands the scope of the method to the analysis of activated memory B cells.
Next, we identified SARS-CoV-2 RBD-specific antibodies directly from human antibody-secreting plasmablasts. We isolated PBMCs 7–9 days after the second dose of the mRNA vaccine BNT162b2 and enriched for B cells by negative selection (Fig. 3a). After encapsulation into BG-agarose and functionalization with anti-human κ and λ VHH–SNAP, we observed a distinct IgG-secreting, RBD-positive ASC population (Fig. 3b and Extended Data Fig. 8a,c). We then compared the antigen-specific signal of all plasmablasts in the sample with added VHH (enabling analysis of secreted antibodies) with the same sample incubated and stained in the same way but without the capture construct (−VHH, only RBD-specific antibodies on the cell surface can be investigated; Extended Data Fig. 8b). We did not observe an antigen-specific plasmablast population (RBD+) in the sample without capture reagent, providing evidence that the antibodies we discovered could not have been identified by FACS surface staining alone under these experimental conditions. Our results do not exclude the possibility that cells with surface Ig that specifically bind to RBD occur in the population, but our approach specifically targets ASCs and would thus not retain them a priori.
We sorted IgG-secreting RBD-positive cells from fresh or cryopreserved samples into single wells of 96-well plates for sequencing. Compared to freshly isolated PBMCs, ASC recovery and IgG secretion from cryopreserved samples were decreased, but no difference in the frequency of RBD-specific ASCs was observed (Extended Data Fig. 8c). Reduced ASC recovery from cryopreserved samples compared to fresh samples has been reported in the literature44.
From just one individual, we obtained 185 plasmablast sequences, belonging to a broad spread of HV–LV gene pairs (Fig. 3c). The 185 antibodies spanned 156 different VH sequences and 111 different VH clonotypes (Methods). As in the mouse study, we observed a population of extremely expanded VH clonotypes. Fifteen clonotypes had occupancy of ≥3 members, including an IGHV4-34 clone (10 members, 5.4% of total, representative CDRH3 ARAHLIGDCGGGSCYSGPDPSNWFDP), an IGHV4-39 clone (8 members, 4.3% of total, representative CDRH3 ARRRAGSYFKDLFDY) and an IGHV7-4-1 clone (8 members, 4.3% of total, representative CDRH3 ARVGRAAIAALDDAFDI).
Clonal clustering against the thousands of human SARS-CoV-2 response antibodies in CoV-AbDab showed that 17 (9%) of our plasmablast sequences belonged to the same VH clonotype as antibodies already characterized to be a SARS-CoV-2 RBD binder35. Several of these belonged to immunodominant VH clonotypes based on their identification across numerous independent studies on alternative B cell compartments from SARS-CoV-2 convalescent patients (such as IGHV3-53 with representative CDRH3 ARDLGEAGGMDV—four sequences in our study and six sequences from six other studies as of February 2022 (refs. 45,46,47,48,49,50)). This clustering demonstrates the presence of RBD-reactive bulk B cell and memory B cell clones among a postvaccination ASC population, confirming their differentiation/expansion as part of the active immune response against SARS-CoV-2.
Additionally, we performed Fv region structural clustering of our sequences alongside CoV-AbDab using the Structural Profiling of Antibodies to Cluster by Epitope (SPACE) protocol51, which clusters anti-coronavirus antibodies by predicted structure and can group them into sets that bind the same antigenic site. This enables an orthogonal consideration of immunodominant epitopes because antibodies that have the same topology but belong to a different VH clonotype can recognize the antigen with the same binding mode. SPACE yielded an additional 13 shared clusters for CoV-AbDab, highlighting cases where the same epitope was likely to be targeted, although the sequences did not belong to the same clonotype.
We then expressed 16 antibodies that either belonged to the most prevalent clonotypes not yet present in the CoV-AbDab (hRBD1-8, belonging to three different clonotypes) or were implicated in RBD binding based on SPACE structural clustering with antibodies found in the CoV-AbDab (hRBD9-16, belonging to eight different clonotypes) as full-length human IgG1 (Supplementary Table 3 and full-length sequences in Supplementary Table 4). Of these antibodies, 94% (15/16) bound RBD at the tested concentrations with affinities in the low picomolar to nanomolar range (Fig. 3d). Of the RBD binders, 80% (12/15) were also able to neutralize WT SARS-CoV-2. Among these, ten antibodies showed high neutralizing capacity, with IC50 values comparable to the REGEN-COV (Ronapreve) mAb cocktail (range approximately 10–100 ng ml−1; Fig. 3e). As observed for many other antibodies induced by WT SARS-CoV-2 (refs. 52,53), neutralizing capacity against Omicron BA.1 SARS-CoV-2 was generally reduced. Nonetheless, it was still quantifiable for five of our antibodies (Fig. 3e), and one antibody (hRBD16) retained most of its neutralizing capacity (IC50 of 68 ng ml−1 against WT compared to 398 ng ml−1 against Omicron BA.1). hRBD16 is highly homologous to CAB-B37, a memory B cell-derived antibody that can neutralize a wide range of SARS-CoV-2 variants54, suggesting that hRBD16 might also be broadly neutralizing. We used epitope binning by BLI to further characterize seven strongly neutralizing antibodies and determined that they bound to two distinct competition regions on WT RBD (hRBD3/4/9/12/16 and hRBD5/8 bound to the same regions; Fig. 3f). Conversely, the differences in neutralizing capacity against Omicron BA.1 (hRBD3/4/16 neutralized Omicron BA.1, while hRBD9/12 did not) indicate that although these antibodies bind to the same region on the RBD, they interact with different residues.