High-throughput mapping of B-cell receptor sequences to antigen specificity

Summary

B-cell receptor (BCR) sequencing is a powerful tool for interrogating immune responses to infection and vaccination, but it provides limited information about the antigen specificity of the sequenced BCRs. Here, we present LIBRA-seq (LInking B-cell Receptor to Antigen specificity through sequencing), a technology for high-throughput mapping of paired heavy-/light-chain BCR sequences to their cognate antigen specificities. B cells are mixed with a panel of DNA-barcoded antigens, such that both the antigen barcode(s) and BCR sequence are recovered via single-cell next-generation sequencing. Using LIBRA-seq, we mapped the antigen specificity of thousands of B cells from two HIV-infected subjects. The predicted specificities were confirmed for a number of HIV-and influenza-specific antibodies, including known and novel broadly neutralizing antibodies. LIBRA-seq will be an integral tool for antibody discovery and vaccine development efforts against a wide range of antigen targets.

In Brief

LIBRA-seq enables high-throughput mapping of BCR sequence to antigen specificity at the single-cell level.

Graphical Abstract

INTRODUCTION

The antibody repertoire – the collection of antibodies present in an individual – responds efficiently to invading pathogens due to its exceptional diversity and ability to fine-tune antigen specificity via somatic hypermutation (Briney et al., 2019; Rajewsky, 1996; Soto et al., 2019). This antibody repertoire is a rich source of potential therapeutics, but its size makes it difficult to examine more than a small cross-section of the total repertoire (Brekke and Sandlie, 2003; Georgiou et al., 2014; Wang et al., 2018; Wilson and Andrews, 2012). Historically, a variety of approaches have been developed to characterize antigen-specific B cells in human infection and vaccination samples. The methods most frequently used include single-cell sorting with fluorescent antigen baits (Scheid et al., 2009; Wu et al., 2010), screens of immortalized B cells (Buchacher et al., 1994; Stiegler et al., 2001), and B cell culture (Bonsignori et al., 2018; Huang et al., 2014; Walker et al., 2009, 2011). However, these methods to couple functional screens with variable heavy (V_H) and variable light (V_L) immunoglobulin gene sequences are low throughput; generally, individual B cells can only be screened against a few antigens simultaneously.

Recent advances in next-generation sequencing (NGS) enable high-throughput interrogation of antibody repertoires at the sequence level, including paired heavy and light chains (Busse et al., 2014; Dekosky et al., 2013; Tan et al., 2014). However, annotation of NGS antibody sequences for their cognate antigen partner(s) generally requires synthesis, production and characterization of individual recombinant monoclonal antibodies (DeFalco et al., 2018; Setliff et al., 2018). Recent efforts to develop new antibody screening technologies have sought to overcome throughput limitations while still uniting antibody sequence and functional information. For example, natively-paired human BCR heavy and light chain amplicons can be expressed and screened as Fab (Wang et al., 2018) or scFV (Adler et al., 2017b, 2017a) in a yeast display system. Although these various antibody discovery technologies have led to the identification of potently neutralizing antibodies, they remain limited by the number of antigens against which single cells can simultaneously be screened efficiently.

Inspired by previous methods combining surface protein marker detection with single-cell RNA sequencing (Peterson et al., 2017; Stoeckius et al., 2017), and T cell epitope determination with T cell receptor sequence (Zhang et al., 2018), we developed LIBRA-seq (LInking B-cell Receptor to Antigen specificity through sequencing) to simultaneously recover both antigen specificity and paired heavy and light chain BCR sequence. LIBRA-seq is a next-generation sequencing-based readout for BCR-antigen binding interactions that utilizes oligonucleotides (oligos) conjugated to recombinant antigens. Antigen barcodes are recovered during paired-chain BCR sequencing experiments and bioinformatically mapped to single cells. To demonstrate the utility of LIBRA-seq, we applied the method to peripheral blood mononuclear cell (PBMC) samples from two HIV-infected subjects, and from these, we successfully identified HIV- and influenza-specific antibodies, including both known and novel broadly neutralizing antibody (bNAb) lineages. LIBRA-seq is high-throughput, scalable, and applicable to many targets. This single, integrated assay enables the mapping of monoclonal antibody sequences to panels of diverse antigens theoretically unlimited in number, and facilitates the rapid identification of cross-reactive antibodies that may serve as therapeutics or vaccine templates.

Results

LIBRA-seq method and validation

LIBRA-seq transforms antibody-antigen interactions into sequencing-detectable events by conjugating barcoded DNA oligos to each antigen in a screening library. All antigens are labeled with the same fluorophore, which enables sorting of antigen-positive B cells by fluorescence-activated cell sorting (FACS) before encapsulation of single B cells via droplet microfluidics. Antigen barcodes and BCR transcripts are tagged with a common cell barcode from bead-delivered oligos, enabling direct mapping of BCR sequence to antigen specificity (Figure 1A).

Figure 1. LIBRA-seq assay schematic and validation.

(A.) Schematic of LIBRA-seq assay. (Top left) Fluorescently-labelled, DNA-barcoded antigens are used to (top right) sort antigen-positive B cells before (bottom) co-encapsulation of single B cells with bead-delivered oligos using droplet microfluidics. Bead-delivered oligos index both cellular BCR transcripts and antigen barcodes during reverse transcription, enabling direct mapping of BCR sequence to antigen specificity following sequencing. Note: elements of the depiction are not shown to scale, and the number and placement of oligonucleotides on each antigen can vary.

(B.) The assay was initially validated on Ramos B-cell lines expressing BCR sequences of known neutralizing antibodies VRC01 and Fe53 with a three-antigen screening library: BG505, CZA97 and H1 A/New Caledonia/20/99.

(C.) Between the minimum (y axis, top) and maximum (y axis, bottom) LIBRA-seq score for each antigen, different cutoffs were tested for their ability to classify each VRC01 cell and Fe53 cell as antigen-positive or -negative, where antigen-positive is defined as having a LIBRA-seq score greater than or equal to the cutoff being evaluated, and antigen-negative is defined as having a LIBRA-seq score below the cutoff. A series of 100 cutoff thresholds between the respective minimum and maximum antigen-specific LIBRA-seq scores were evaluated. At each cutoff, the percent of total VRC01 cells (left column of each antigen subpanel) and percent of total Fe53 cells (right columns) that were classified as positive for a given antigen is represented on a white (0%) to dark purple (100%) color scale.

(D.) For each B cell, the LIBRA-seq scores for each pair of antigens was plotted. Each axis represents a range of LIBRA-seq scores for each antigen. Density of total cells is shown, with purple to yellow indicating lowest to highest number of cells, respectively.

(E.) The LIBRA-seq score for BG505 (y-axis) and CZA97 (x-axis) for each VRC01 B cell was plotted. Each axis represents a range of LIBRA-seq scores for each antigen. Density of total cells is shown, with purple to yellow indicating lowest to highest number of cells, respectively.

See also Figure S1 and S6.

To test the ability of LIBRA-seq to accurately unite BCR sequence and antigen specificity, we devised a proof-of-principle mapping experiment using two Ramos B-cell lines with different BCR sequences and antigen specificities (Weaver et al., 2016). These engineered B-cell lines do not display endogenous BCR and instead express specific, user-defined surface IgM BCR sequences (Weaver et al., 2016). To that end, we chose two well-characterized BCRs: VRC01, a CD4-binding site-directed HIV-1 bNAb (Wu et al., 2010), and Fe53, a bNAb recognizing the stem of group 1 influenza hemagglutinins (HA) (Lingwood et al., 2012). We mixed these two populations of B-cell lines at a 1:1 ratio and incubated them with three unique DNA-barcoded antigens: two stabilized trimeric HIV-1 Env proteins (SOSIP) from strains BG505 and CZA97 (Georgiev et al., 2015; van Gils et al., 2013; Ringe et al., 2017), and trimeric hemagglutinin from strain H1 A/New Caledonia/20/1999 (Whittle et al., 2014) (Figure 1B; Supplemental Figure 1A–C).

We recovered 2321 cells with BCR sequence and antigen mapping information, highlighting the high throughput potential of LIBRA-seq (Supplemental Figure 1D). For each cell, the LIBRA-seq scores for each antigen in the screening library were computed as a function of the number of unique molecular identifiers (UMIs) for the respective antigen barcode (Methods). The LIBRA-seq scores of each individual antigen reliably categorized Ramos B cells by their specificity (Figure 1C). Overall, cells fell into two major populations based on their LIBRA-seq scores, and we did not observe cells that were cross-reactive for influenza HA and HIV-1 Env (Figure 1D). Further, VRC01 Ramos B cells bound both BG505 and CZA97 with a high correlation between the scores for these two antigens (Pearson’s r=0.84), demonstrating that LIBRA-seq readily identifies B cells that bind to multiple HIV-1 antigens (Figure 1E).

Isolation of Antibodies from a Known HIV bNAb Lineage from Donor NIAID45

We next used LIBRA-seq to analyze the antibody repertoire of donor NIAID45, who had been living with HIV-1 without antiretroviral therapy for approximately 17 years at the time of sample collection. This sample was selected as an appropriate target for LIBRA-seq analysis because a large lineage of HIV-1 bNAbs had been identified previously from this donor (Bonsignori et al., 2018; Wu et al., 2010, 2015). This lineage consists of the prototypical bNAb VRC01, as well as multiple clades of clonally related antibodies with diverse neutralization phenotypes (Wu et al., 2015). We used the same BG505, CZA97, and H1 A/New Caledonia/20/99 antigen screening library as in the Ramos B-cell line experiment, and recovered paired V_H:V_L antibody sequences with antigen mapping for 866 cells (Figure 2A; Supplemental Figures 1D, 2A). These B cells exhibited a variety of LIBRA-seq scores among the three antigens (Figure 2B), as can be expected from a polyclonal sample possessing a wide diversity of B cell specificities and antigen affinities. The cells displayed a few patterns based on their LIBRA-seq scores; generally, cells were either (1) HA^highEnv^low or (2) HA^lowEnv^high (Figure 2B). Additionally, we observed cells that were double positive for both HIV Env variants, BG505 and CZA97, suggesting HIV-1 strain cross-reactivity of these B cells (Figure 2B).

Figure 2. LIBRA-seq applied to a human B cell sample from HIV-infected donor NIAID45.

(A.) LIBRA-seq experiment setup consisted of three antigens in the screening library: BG505, CZA97, and H1 A/New Caledonia/20/99, and the cellular input was donor NIAID45 PBMCs.

(B.) After bioinformatic processing and filtering of cells recovered from single-cell sequencing, the LIBRA-seq score for each antigen was plotted (total, 866 cells). Each axis represents a range of LIBRA- seq scores for each antigen. Density of total cells is shown, with purple to yellow indicating lowest to highest number of cells, respectively.

(C.) 29 VRC01 lineage B cells were identified and examined for phylogenetic relatedness to known lineage members and for sequence features, with phylogenetic tree showing relatedness of previously identified VRC01 lineage members (black) and members newly identified using LIBRA-seq (red). Each row represents an antibody. Sequences were aligned using clustalW and a maximum likelihood tree was inferred using maximum likelihood inference. The resulting tree was visualized using an inferred VRC01 unmutated common ancestor (UCA) (accession MK032222) as the root. For each antibody isolated from LIBRA-seq, a heatmap of the LIBRA-seq scores for each HIV antigen (BG505 and CZA97) is shown; a scale of tan-white-purple represents LIBRA-seq scores from −2 to 0 to 2; in this heatmap, scores lower or higher than that range are shown as −2 and 2, respectively. Levels of somatic hypermutation (SHM) at the nucleotide level for the heavy and light chain variable genes as reported by IMGT are displayed as bars, with the numerical percentage value listed to the right of the bar; length of the bar corresponds to level of SHM. Amino acid sequences of the complementarity determining region 3 for the heavy chain (CDRH3) and the light chain (CDRL3) for each antibody are displayed. The tree was visualized and annotated using iTol (Letunic and Bork, 2019).

See also Figure S1, S2, and S6.

To further validate the utility of LIBRA-seq in monoclonal antibody isolation, we next sought to identify members of the VRC01 antibody lineage from the LIBRA-seq-identified antigen-specific B cells. We observed 29 BCRs that were clonally related to previously-identified members of the VRC01 lineage (Figure 2C). All LIBRA-seq-identified BCRs had high levels of somatic hypermutation and utilized IGHV1–2^*02 along with the characteristic five-residue CDRL3 paired with IGVK3–20 (Figure 2C). These B cells came from multiple known clades of the VRC01 lineage, with sequences with high identity and phylogenetic relatedness to lineage variants VRC01, VRC02, VRC03, NIH45–46, and others (Figure 2C). Of these, 25 (86%) had a high LIBRA-seq score for at least 1 HIV-1 antigen, three (10%) had mid-range scores (between 0 and 1) for at least 1 HIV-1 antigen, and only one of the VRC01 lineage B cells had negative scores for both HIV-1 antigens (Figure 2C, Supplemental Figure 2B). We recombinantly expressed three of the LIBRA-seq-identified lineage members, named 2723–3055, 2723–4186 and 2723–3131, to confirm the ability of these antibodies to bind the screening probes. Antibody 2723–3131 showed binding to CZA97 and BG505 by enzyme linked immunosorbent assay (ELISA) (Figure 3A), and neutralized two Tier 1 viruses but no viruses on a global panel of representative HIV-1 strains (deCamp et al., 2014) (Figure 3B). Both 2723–3055 and 2723–4186 bound to BG505 and CZA97, and potently neutralized 12/12 and 11/12 viruses on a global panel, respectively (Figure 3A–B). Together, the results from the donor NIAID45 analysis suggest that the LIBRA-seq platform can be successfully used to down-select cross-reactive bNAbs in prospective antibody discovery efforts.

Figure 3. Characterization of LIBRA-seq-identified antibodies from donor NIAID45.

(A.) Antigen specificity as predicted by LIBRA-seq was validated by ELISA for a subset of monoclonal antibodies belonging to the VRC01 lineage. Data are represented as mean ± SEM for one ELISA experiment. ELISA data are representative of at least two independent experiments.

(B.) Neutralization of Tier 1, Tier 2, and control viruses by VRC01 and newly identified VRC01 lineage members, 2723–3131, 2723–4186, and 2723–3055. IC50 values are shown from high potency (0.0001 μg/ml, red) to low potency (50 μg/ml, green). Lack of neutralization IC50 for concentrations tested is displayed as white.

(C.) Sequence characteristics and antigen specificity of newly identified antibodies from donor NIAID45. Percent identity is calculated at the nucleotide level, and CDRH3 and CDRL3 lengths and sequences are noted at the amino acid level. LIBRA-seq scores for each antigen are displayed as a heatmap with a LIBRA-seq score of −2 displayed as light yellow, 0 as white, and a LIBRA-seq score of 2 as purple; in this heatmap, scores lower or higher than that range are shown as −2 and 2, respectively. ELISA binding data against BG505, CZA97, and H1 A/New Caledonia/20/99 is displayed as a heatmap of the AUC analysis calculated from the data in Supplemental Figure 3A with AUC of 0 displayed as light yellow, 50% max as white, and maximum AUC as purple. ELISA data are representative from at least two independent experiments.

See also Figure S2 and S3.

Identification of Additional Anti-HIV and Anti-influenza Antibodies from Donor NIAID45

To further validate the ability of LIBRA-seq to accurately identify antigen-specific B cells, we produced a number of putative HIV-specific and influenza-specific monoclonal antibodies from donor NIAID45 that did not belong to the VRC01 lineage. In particular, we recombinantly produced seven additional anti-HIV antibodies, three of which were clonally related (2723–2121, 2723–422, and 2723–2304) (Figure 3C). We selected these seven antibodies because all had high LIBRA-seq scores for at least one HIV-1 antigen. All seven antibodies bound the antigens by ELISA as expected based on the respective LIBRA-seq scores, with high similarity between the patterns of LIBRA-seq scores and ELISA area under the curve (AUC) values (Figure 3C, Supplemental Figure 3A, Methods). We further characterized one of these antibodies, 2723–2121, and determined that it bound to a stabilized BG505 trimer (DS-SOSIP) (Kwon et al., 2015) by surface plasmon resonance (SPR) (Figures S3B and S3C). Antibody 2723–2121 competed for trimer binding with VRC01 (Supplemental Figure 3D), neutralized three Tier 1 pseudoviruses and 2/11 Tier 2 pseudoviruses from a global panel (Supplemental Figure 3E), and mediated trogocytosis and antibody-dependent cellular phagocytosis (Supplemental Figure 3F). In addition to the HIV-specific antibodies, we also characterized two antibodies predicted to have influenza specificity based on their LIBRA-seq scores for H1 A/New Caledonia/20/99 (Figure 3C). In agreement with the LIBRA-seq scores, antibodies 2723–2859 and 2723–3415 bound H1 A/New Caledonia/20/99 but not BG505 or CZA97 by ELISA, confirming the ability of LIBRA-seq to simultaneously isolate antibodies to multiple diverse antigens (Figure 3C, Supplemental Figure 3A).

Discovery of an HIV bNAb from Donor N90 Using a Nine-Antigen Screening Library

Having validated LIBRA-seq with three antigens on both Ramos B-cell lines and primary B cells from a patient sample, we sought to increase the number of antigens in the screening library. To that end, we screened the B-cell repertoire of NIAID donor N90 against nine antigens (Figure 4A). We selected this sample because a single broadly neutralizing antibody lineage (VRC38) targeting the V1/V2 epitope was isolated previously from this donor; however, the neutralization breadth of the VRC38 lineage could not account for the full serum neutralization breadth (Cale et al., 2017; Wu et al., 2012). This suggested that there could be additional bNAb lineages present in the B cell repertoire of N90, and we reasoned that utilizing multiple SOSIP probes could help accelerate identification of such antibodies. Thus, we sought to determine whether LIBRA-seq could accomplish two goals: (1) to recover antigen-specific B cells from the VRC38 lineage, and (2) to identify new bNAbs that could neutralize viruses that are resistant to neutralization by the VRC38 lineage.

Figure 4. LIBRA-seq applied to a sample from NIAID donor N90.

(A.) LIBRA-seq experiment setup consisted of nine antigens in the screening library: 5 HIV-1 Env (KNH1144, BG505, ZM197, ZM106.9, B41), and 4 influenza HA (H1 A/New Caledonia/20/99, H1 A/Michigan/45/2015, H5 Indonesia/5/2005, H7 Anhui/1/2013), and the cellular input was donor N90 PBMCs.

(B.) 18 VRC38 lineage B cells were identified and examined for phylogenetic relatedness to known lineage members as well as for sequence features, with phylogenetic tree showing relatedness of previously identified VRC38 lineage members (black) and members newly identified using LIBRA-seq (red). Each row represents an antibody. Sequences were aligned using clustalW and a maximum likelihood tree was inferred using maximum likelihood inference. The resulting tree was visualized, with a germline-reverted antibody from lineage VRC38 (Methods) as the root. For each antibody isolated from LIBRA-seq, a heat map of the LIBRA-seq scores for each HIV antigen is shown; tan-white-purple represents LIBRA-seq scores from −2 to 0 to 2; in this heatmap, scores lower or higher than that range are shown as −2 and 2, respectively. Levels of somatic hypermutation (SHM) at the nucleotide level for the heavy and light chain variable genes as reported by IMGT are displayed as bars, with the numerical percentage value listed to the right of the bar; length of the bar corresponds to level of SHM. Amino acid sequences of the complementarity determining region 3 for the heavy chain (CDRH3) and the light chain (CDRL3) for each antibody are displayed. The tree was visualized and annotated using iTol (Letunic and Bork, 2019).

(C-D) For each combination of (C.) influenza hemagglutinins or (D.) HIV SOSIPs, the number of B cells with high LIBRA-seq scores (>= 1) is displayed as a bar graph. The combinations of antigens are displayed by filled circles, showing which antigens are part of a given combination. Each combination is mutually exclusive. The total number of B cells with high LIBRA-seq scores for each antigen is indicated as a horizontal bar on the bottom left of each subpanel.

See also Figure S1, S4, S5, and S6.

To increase the number of antigens in our screening library, we utilized a panel that consisted of five HIV-1 Env trimers from a variety of clades, BG505 (clade A), B41 (clade B), ZM106.9 (clade C), ZM197 (clade C) and KNH1144 (clade A) (van Gils et al., 2013; Harris et al., 2011; Joyce et al., 2017; Julien et al., 2015; Pugach et al., 2015; Ringe et al., 2017), along with four diverse hemagglutinin trimers (H1 A/New Caledonia/20/99, H1 A/Michigan/45/2015, H5 A/Indonesia/5/2005, and H7 A/Anhui/1/2013) (Figure 4A, Supplemental Figure 1A). After applying LIBRA-seq to donor N90 PBMCs, we recovered paired V_H:V_L antibody sequences with antigen mapping for 1465 cells (Supplemental Figure 1D, 4A). Within this set of cells, we identified 18 B cells that were members of the VRC38 lineage (Figure 4B). Of these, 17 had high LIBRA-seq scores for at least one HIV antigen, and one had no high LIBRA-seq scores but had a mid-range score for two SOSIPs (Figure 4B).

We next focused our analysis on the B cells with the highest LIBRA-seq scores in the N90 sample, focusing on cells that had LIBRA-seq scores for any antigen above one (901 cells) (Figure 4C–D, Supplemental Figure 5A). We observed 32 cells that had high LIBRA-seq scores for three of the four influenza antigens (Figure 4C); we recombinantly produced one of these, 3602–1707, and confirmed broad influenza recognition, with high correlation between LIBRA-seq scores and ELISA AUC (Spearman correlation 0.77, p=0.015) (Figure 5A, Supplemental Figure 4B).

Figure 5. Characterization of LIBRA-seq-identified antibodies from donor NIAID N90.

(A.) Sequence characteristics and antigen specificity of newly identified antibodies from donor N90. Percent identity is calculated at the nucleotide level, and CDR length and sequences are noted at the amino acid level. LIBRA-seq scores for each antigen are displayed as a heatmap with a LIBRA-seq score of −2 displayed as light yellow, 0 as white, and a LIBRA-seq score of 2 as purple; in this heatmap, scores lower or higher than that range are shown as −2 and 2, respectively. ELISA binding data is displayed as a heatmap of the AUC analysis calculated from the data in Supplemental Figure 4B with AUC of 0 displayed as light yellow, 50% max as white, and maximum AUC as purple. ELISA data are representative from at least two independent experiments.

(B.) Neutralization of Tier 2 and control viruses by newly identified antibody 3602–870. IC50 values are shown from high potency (0.0001 μg/ml, red) to low potency (50 μg/ml, green). Lack of neutralization IC50 for concentrations tested is displayed as white.

(C.) Inhibition of BG505 DS-SOSIP/293F binding to 3602–870 IgG in presence of VRC34 Fab (diamond), PGT145 Fab (square) and VRC01 Fab (triangle).

See also Figure S4 and S5.

We also observed cells that had high LIBRA-seq scores for each of the different HIV-1 antigens, including 124 cells that had high scores for four or more SOSIPs (Figure 4D). We then down-selected SOSIP-high B cells based on having high LIBRA-seq scores to at least 3 SOSIP variants. In particular, we identified two members from the same antibody lineage that had high LIBRA-seq scores for BG505, KNH1144, ZM106.9 and ZM197. This lineage utilized the germline genes IGHV1–46 and IGK3–20 and was highly mutated in both the heavy- and light-chain V gene. We recombinantly expressed one of the lineage members, 3602–870, which was 28.5% mutated in its heavy chain V gene and 17.0% mutated in its light chain V gene and had a 19 amino acid CDRH3 and 9 amino acid CDRL3 (Figure 5A). 3602–870 bound all SOSIP probes by ELISA (Spearman correlation of 0.97, p<0.001 between LIBRA-seq scores and ELISA AUC) and neutralized 79% of tested Tier 2 viruses (11 of 14), including several viruses that were not neutralized by VRC38.01 (Cale et al., 2017) (Figure 5A–B, Supplemental Figure 4B). Of note, 3602–870 neutralized BG505 and ZM197, both of which were used as probes in the antigen screening library (Figure 5B). 3602–870 bound BG505 DS-SOSIP by SPR and competed for BG505 DS-SOSIP binding with VRC01 Fab (Figure 5C, Supplemental Figure 4C).

In summary, LIBRA-seq enabled the high-throughput, highly multiplexed screening of single B cells from an HIV-infected subject against a large antigen panel. This resulted in the identification of hundreds of antigen-specific monoclonal antibody leads from donor N90, with high-resolution antigen specificity mapping helping to facilitate rapid lead prioritization to identify a novel bNAb lineage.

DISCUSSION

Here, we developed a novel method to interrogate antibody-antigen interactions via a sequencing-based readout. After validating the approach on cell lines with known BCRs, we applied LIBRA-seq to prospective antibody discovery. We identified members of two known HIV-specific bNAb lineages from previously characterized human infection samples and a novel bNAb lineage. Additionally, we identified many other candidate broadly-reactive HIV-specific antibodies, and validated specificity for a subset of them. Within both HIV-1 infection samples, we also isolated influenza-specific antibodies using hemagglutinin screening probes, highlighting the utility of LIBRA-seq for simultaneously screening B cell repertoires against multiple, diverse antigen targets. In principle, the NGS-based coupling of antibody sequence and specificity enables screening of potentially millions of single B cells for reactivity to a larger repertoire of epitopes than purely fluorescence-based methods, since sequence space is not hindered by spectral overlap. Using LIBRA-seq may therefore help to maximize lead discovery per experiment, an important consideration when preserving limited sample.

Beyond LIBRA-seq’s utility in antibody discovery, the high-throughput coupling of antibody sequence and specificity can enable high-resolution immune profiling. For example, in donor N90 we observed an increase in IGHV gene somatic hypermutation between B cells that had a high LIBRA-seq score for a single HIV-1 antigen versus B cells that had high LIBRA-seq scores for multiple HIV-1 antigens (Figure 6A). We also observed the use of specific germline genes to be more frequent in B cells that exhibited broad, as opposed to strain-specific, HIV-1 antigen reactivity (Figure 6B, Supplemental Figure 5B). The elucidation of such relationships, enabled by the LIBRA-seq technology, may guide germline-targeting vaccine design efforts (Dosenovic et al., 2019; Jardine et al., 2013, 2016; Stamatatos et al., 2017) and can provide insights into the requirements for the acquisition of HIV-1 antigen cross-reactivity. The application of LIBRA-seq to antibody discovery and immune profiling should translate into rapid accumulation of new data, leading to novel insights into basic and applied immunology.

Figure 6. Sequence properties of the antigen-specific B cell repertoire.

(A.) IGHV gene identity (y-axis) is plotted for cells with high (>=1) LIBRA-seq scores for any combination of 1 through 5 HIV-1 SOSIP antigens (x-axis). Each distribution is displayed as a kernel density estimation, where wider sections of a given distribution represent a higher probability that B cells possess a given germline identity percentage. The median of each distribution is displayed as a white dot, the interquartile range is displayed as a thick bar, and a thin line extends to 1.5x the interquartile range. The violin ranges were limited to the observed data. Included are cells with IgG or IgA constant heavy genes as determined by Cell Ranger.

(B.) Each dot represents an IGHV germline gene, plotted based on the number of B cells reactive to only 1 HIV-1 SOSIP antigen (x axis) and the number of B cells reactive to 3 or more HIV-1 SOSIP antigens (y axis) that are assigned to that respective IGHV germline gene. Only B cells with high (>=1) LIBRA-seq scores for any HIV-1 antigen and with IgG or IgA constant heavy genes as determined by Cell Ranger are shown.

See also Figure S5.

STAR Methods

LEAD CONTACT AND MATERIALS AVAILABILITY

Further information and requests for resources and reagents should be directed to the Lead Contact, Ivelin Georgiev (Ivelin.Georgiev@Vanderbilt.edu). Antibody plasmids generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Human Subjects

Donor NIAID45:

Peripheral blood mononuclear cells were collected from donor NIAID45 on July 12, 2007. Donor NIAID45 was enrolled in investigational review board approved clinical protocols at the National Institute of Allergy and Infectious Diseases and had been living with HIV without antiretroviral treatment for approximately 17 years at the time of sample collection. Donor N90: Peripheral blood mononuclear cells were collected from donor N90 on May 29, 2008. Donor N90 was enrolled in investigational review board approved clinical protocols at the National Institute of Allergy and Infectious Diseases and had been living with HIV without antiretroviral treatment through the timepoint of sample collection since diagnosis in 1985 (Wu et al., 2012).

Cell Lines

Ramos B-cell lines were engineered from a clone of Ramos Burkitt’s lymphoma that do not display endogenous antibody, and they ectopically express specific surface IgM B cell receptor sequences. The B cell lines used expressed B cell receptor sequences for HIV-specific antibody VRC01 and influenza-specific antibody Fe53. The cells are cultured at 37°C with 5% CO ₂ saturation in complete RPMI, made up of RPMI supplemented with 15% fetal bovine serum, 1% L-Glutamine, and 1% Penicillin/Streptomycin. Although endogenous heavy chains are scrambled, endogenous light chain transcripts remain and are detectable by sequencing. We thus identified and classified single Ramos Burkitt’s B cells as either VRC01 or Fe53 based on their heavy chain sequences. These Ramos B- cell lines were validated for binding to our antigen probes by FACS (Supplemental Figure 1B).

METHOD DETAILS

Antigen expression and purification

For the different LIBRA-seq experiments, a total of six HIV-1 gp140 SOSIP variants from strains BG505 (clade A), CZA97 (clade C), B41 (clade B), ZM197 (clade C), ZM106.9 (clade C), KNH1144 (clade A) and four influenza hemagglutinin variants from strains A/New Caledonia/20/99 (H1N1) (GenBank ACF41878), A/Michigan/45/2015 (H1N1) (GenBank AMA11475), A/Indonesia/5/2005 (H5N1) (GenBank ABP51969), and A/Anhui/1/2013 (H7N9) (GISAID EPI439507) were expressed as recombinant soluble antigens.

The single-chain variants (Georgiev et al., 2015) of BG505, CZA97, B41, ZM197, ZM106.9, and KNH1144 each containing an AviTag, were expressed in FreeStyle 293F mammalian cells (ThermoFisher) using polyethylenimine (PEI) transfection reagent and cultured for 5–7 days. FreeStyle 293F were maintained in FreeStyle 293F medium or FreeStyle F17 expression medium supplemented with 1 % of 10% Pluronic F-68 and 20% of 200 mM L-Glutamine. These cells were cultured at 37°C with 8% CO ₂ saturation and shaking. After transfection and 5–7 days of culture, cultures were centrifuged at 6000 rpm for 20 minutes. Supernatant was filtered with Nalgene Rapid Flow Disposable Filter Units with PES membrane (0.45 μm), and then run slowly over an affinity column of agarose bound Galanthus nivalis lectin (Vector Laboratories cat no. AL-1243-5) at 4°C. The column was washed with PBS, and proteins were eluted with 30 mL of 1 M methyl-α-D-mannopyranoside. The protein elution was buffer exchanged 3X into PBS and concentrated using 30kDa Amicon Ultra centrifugal filter units. Concentrated protein was run on a Superose 6 Increase 10/300 GL or Superdex 200 Increase 10/300 GL sizing column on the AKTA FPLC system, and fractions were collected on an F9-R fraction collector. Fractions corresponding to correctly folded antigen were analyzed by SDS-PAGE, and antigenicity by ELISA was characterized with known monoclonal antibodies specific for that antigen.

Recombinant HA proteins all contained the HA ectodomain with a point mutation at the sialic acid-binding site (Y98F), T4 fibritin foldon trimerization domain, AviTag, and hexahistidine-tag, and were expressed in Expi 293F mammalian cells using Expifectamine 293 transfection reagent (Thermo Fisher Scientific) cultured for 4–5 days. Culture supernatant was harvested and cleared as above, and then adjusted pH and NaCl concentration by adding 1M Tris-HCl (pH 7.5) and 5M NaCl to 50 mM and 500 mM, respectively. Ni Sepharose excel resin (GE Healthcare) was added to the supernatant to capture hexahistidine tag. Resin was separated on a column by gravity and captured HA protein was eluted by a Tris-NaCl (pH 7.5) buffer containing 300 mM imidazole. The eluate was further purified by a size exclusion chromatography with a HiLoad 16/60 Superdex 200 column (GE Healthcare). Fractions containing HA were concentrated, analyzed by SDS-PAGE and tested for antigenicity by ELISA with known antibodies. Proteins were frozen in LN₂ and stored at −80C° until use.

All HIV gp140 SOSIP variant antigens and all influenza hemagglutinin variant antigens included an AviTag modification at the C-terminus of their sequence, and after purification, each AviTag labeled antigen was biotinylated using the BirA-500: BirA biotin-protein ligase standard reaction kit (Avidity LLC, cat no. BirA500).

Oligonucleotide barcodes

We used oligos that possess a 13–15 bp antigen barcode, a sequence capable of annealing to the template switch oligo that is part of the 10X bead-delivered oligos, and contain truncated TruSeq small RNA read 1 sequences in the following structure: 5’-CCTTGGCACCCGAGAATTCCANNNNNNNNNNNNNCCCATATAAGA*A*A-3’, where Ns represent the antigen barcode. For the cell line and NIAID45 experiments, we used the following antigen barcodes: CATGATTGGCTCA (BG505), TGTCCGGCAATAA (CZA97), GATCGTAATACCA (H1 A/New Caledonia/20/99). For the N90 experiment, we used longer antigen barcodes (15 bp), as follows: TCCTTTCCTGATAGG (ZM106.9), TAACTCAGGGCCTAT (KNH1144), GCTCCTTTACACGTA (ZM197), GCAGCGTATAAGTCA (B41), ATCGTCGAGAGCTAG (BG505), CAGGTCCCTTATTTC (A/Indonesia/5/2005), ACAATTTGTCTGCGA (A/Anhui/1/2013), TGACCTTCCTCTCCT (A/Michigan/45/2015), AATCACGGTCCTTGT (A/New Caledonia/20/99). Oligos were ordered from Sigma-Aldrich and IDT with a 5’ amino modification and HPLC purified.

Conjugation of oligonucleotide barcodes to antigens

For each antigen, a unique DNA barcode was directly conjugated to the antigen itself. In particular, 5’amino-oligonucleotides were conjugated directly to each antigen using the Solulink Protein-Oligonucleotide Conjugation Kit (TriLink cat no. S-9011) according to manufacturer’s instructions. Briefly, the oligo and protein were desalted, and then the amino-oligo was modified with the 4FB crosslinker, and the biotinylated antigen protein was modified with S-HyNic. Then, the 4FB-oligo and the HyNic-antigen were mixed together. This causes a stable bond to form between the protein and the oligonucleotide. The concentration of the antigen-oligo conjugates was determined by a BCA assay, and the HyNic molar substitution ratio of the antigen-oligo conjugates was analyzed using the NanoDrop according to the Solulink protocol guidelines. AKTA FPLC was used to remove excess oligonucleotide from the protein-oligo conjugates, which were also checked using SDS-PAGE with a silver stain.

Fluorescent labeling of antigens

After attaching DNA barcodes directly to a biotinylated antigen, the barcoded antigens were mixed with streptavidin labeled with fluorophore phycoerythrin (PE). The streptavidin-PE was mixed with biotinylated antigen at a 5X molar excess of antigen to streptavidin. 1/5 of the streptavidin-PE conjugate was added to the antigen every 20 minutes with constant rotation at 4°C.

Enrichment of antigen-specific B cells

For a given sample, cells were stained and mixed with fluorescently labeled DNA-barcoded antigens and other antibodies, and then sorted using fluorescence activated cell sorting (FACS). First, cells were counted and viability was assessed using Trypan Blue. Then, cells were washed with DPBS supplemented with 1 % Bovine serum albumin (BSA) through centrifugation at 300 g for 7 minutes. Cells were resuspended in PBS-BSA and stained with a variety of cell markers. For donor NIAID45 PBMCs, these markers included CD3-APCCy7, IgG-FITC, CD19-BV711, CD14-V500, and LiveDead-V500. Additionally, fluorescently labeled antigen-oligo conjugates (described above) were added to the stain. For donor N90 PBMCs, these markers included Ghost Red 780, CD14-APCCy7, CD3-FITC, CD19-BV711, and IgG-PECy5. Additionally, fluorescently labeled antigen-oligo conjugates were added to the stain. After staining in the dark for 30 minutes at room temperature, cells were washed 3 times with PBS-BSA at 300 g for 7 minutes. Then, cells were resuspended in DPBS and sorted on the cell sorter. Antigen positive cells were bulk sorted and then they were delivered to the Vanderbilt VANTAGE sequencing core at an appropriate target concentration for 10X Genomics library preparation and subsequent sequencing. FACS data were analyzed using Cytobank (Kotecha et al., 2010).

10X Genomics single cell processing and next generation sequencing

Single-cell suspensions were loaded onto the Chromium Controller microfluidics device (10X Genomics) and processed using the B-cell Single Cell V(D)J solution according to manufacturer’s suggestions for a target capture of 10,000 B cells per 1/8 10X cassette for B cell lines, 9,000 cells for B cells from donor NIAID45, and 4,000 for donor N90, with minor modifications in order to intercept, amplify and purify the antigen barcode libraries. The library preparation followed the CITE-seq protocol (available at https://cite-seq.com), with the exception of an increase in the number of PCR cycles of the antigen barcodes. Briefly, following cDNA amplification using an additive primer (5’-CCTTGGCACCCGAGAATT*C*C-3’) to increase the yield of antigen barcode libraries (Stoeckius et al., 2017), SPRI separation was used to size separate antigen barcode libraries from cellular mRNA libraries, PCR amplified for 10–12 cycles, and purified using 1.6X purification. Sample preparation for the cellular mRNA library continued according to 10X Genomics-suggested protocols, resulting in Illumina-ready libraries. Following library construction, we sequenced both BCR and antigen barcode libraries on a NovaSeq 6000 at the VANTAGE sequencing core, dedicating ~2.5% of a flow cell to each experiment, with a target 10% of this fraction dedicated to antigen barcode libraries. This resulted in ~334.5 million reads for the cell line V(D)J libraries (~96,500 reads/cell), ~376.3 million reads for donor NIAID45 V(D)J libraries (~79,300 reads/cell), and ~272.4 million reads for the N90 V(D)J libraries (~151,400 reads/cell). The N90 antigen barcode libraries were also sequenced a second time.

Processing of BCR sequence and antigen barcode reads

We developed a pipeline that takes paired-end FASTQ files of oligo libraries as input, processes and annotates reads for cell barcode, UMI, and antigen barcode, and generates a cell barcode - antigen barcode UMI count matrix. BCR sequence reads were processed using Cell Ranger (10X Genomics) using GRCh38 as reference. For the antigen barcode libraries, initial quality and length filtering was carried out by fastp (Chen et al., 2018) using default parameters for filtering (Supplemental Figure 6A). In a histogram of insert lengths, this resulted in a sharp peak of the expected insert size of 52–54 bp (Supplemental Figure 6B–D). Fastx_collapser was then used to group identical sequences and convert the output to deduplicated fasta files. We proceeded to process just the R2 sequences, as the entire insert is present in both R1 and R2. Each unique R2 sequence was processed one-by-one using the following steps: (1) The reverse complement of the R2 sequence was determined. (2) The sequence was screened for possessing an exact match to any of the valid 10X cell barcodes present in the filtered_contig.fasta file output by Cell Ranger during processing of BCR V(D)J FASTQ files. Sequences without a BCR-associated cell barcode were discarded. (3) The 10 bases immediate 3’ to the cell barcode were annotated as the read UMI. (4) The remainder of the sequence 3’ to the UMI was screened for a 13 or 15 bp sequence within a hamming distance of 2 to any of the antigen barcodes used in the screening library. Following this processing, only sequences with lengths of 51 to 58 were retained. After processing each sequence one-by-one, we screened for cell barcode - UMI - antigen barcode collisions. Any cell barcode - UMI combination that had multiple antigen barcodes associated with it was removed. We then constructed a cell barcode - antigen barcode UMI count matrix, which served as the basis of subsequent analysis. Additionally, we aligned the BCR contigs (filtered_contigs.fasta file output by Cell Ranger, version 2.2.0 CellRanger 10X Genomics) to IMGT reference genes using HighV-Quest (Alamyar et al., 2012). The output of HighV-Quest was parsed using ChangeO (Gupta et al., 2015), and merged with the UMI count matrix.

Determination of LIBRA-seq Score

Starting with the UMI count matrix, we set all counts of 1, 2, or 3 UMIs to 0, with the idea that these low counts could likely be attributed to noise. After this, the UMI count matrix was subset to contain only cells with a count of at least 4 UMIs for at least 1 antigen. We also removed cells that had only non-functional heavy chain sequences as well as cells with multiple functional heavy chain sequences using different IGHV genes, reasoning that these may be multiplets. We then calculated the centered-log ratios (CLR) of each antigen UMI count for each cell (Mimitou et al., 2019; Stoeckius et al., 2017). Because UMI counts were on different scales for each antigen, possibly due to differential oligo loading during oligo-antigen conjugation, we rescaled the CLR UMI counts using the StandardScaler method in scikit learn (Pedregosa and Varoquaux, 2011). Lastly, we performed a correction procedure to the scaled CLRs from UMI counts of 0, setting them to the minimum for each antigen for donor NIAID45 and N90 experiments, and to −1 for the Ramos B-cell line experiment. These CLR-transformed, scaled, corrected values served as the final LIBRA-seq scores. LIBRA-seq scores were visualized using Cytobank (Kotecha et al., 2010) and Matplotlib (Hunter, 2007). Cells with a LIBRA-seq score of 1 or greater for Donor N90 data were also visualized using UpSet plots (Lex et al., 2014) using the UpSetPlot package in Python. Donor NIAID45 and N90 data were subsetted to include only cells with a functional light chain.

Phylogenetic trees

Phylogenetic trees of antibody heavy chain sequences were constructed in order to assess the relatedness of antibodies within a given lineage. For the VRC01 lineage, the 29 sequences identified by LIBRA-seq and 53 sequences identified from the literature were aligned using clustal within Geneious. We then used the PhyML maximum likelihood (Guindon et al., 2010) plugin in Geneious (available at https://www.geneious.com/plugins/phyml-plugin/) to infer a phylogenetic tree. The resulting tree was then rooted to the inferred unmutated common ancestor (Bonsignori et al., 2018) (accession MK032222). Names for sequences and their accession include H01+07.F.1 (KP840594); H03+06.C.1 (KP840597); H03+06.E.1 (KP841560); H4.E.6 (KP841696); H4.E.5 (KP841700); H4.E.4 (KP841639); H4.E.3 (KP841608); H4.E.2 (KP841609); H4.E.1 (KP841701); H5.C.1 (KP840607); H5.F.1 (KP840608); H08.F.1 (KP840603); H08.H.1 (KP840835); VRC03b (KP840671); VRC03f (KP840674); VRC03g (KP840675); DH651.1 (MK032223); DH651.3 (MK032225); DH651.9 (MK032231); DH651.8 (MK032230); VRC06c (KP840678); VRC06d (KP840679); VRC06e (KP840680); VRC06f (KP840681); VRC06g (KP840682); VRC06h (KP840683); DH651.2 (MK032224); DH651.4 (MK032226); DH651.5 (MK032227); DH651.6 (MK032228); DH651.7 (MK032229); VRC06 (JX466923.1); VRC03 (GU980706.1); NIH45–46 (HE584543); VRC01 (GU980702); VRC01c (KP840658); VRC01d (KP840659); VRC01e (KP840660); VRC01f (KP840661); VRC01h (KP840663); VRC01i (KP840664); VRC01j (KP840665); VRC02 (GU980704); VRC07b (KP840666); VRC07c (KP840667); VRC07d (KP840668); VRC07e (KP840669); VRC07f (KP840670); VRC08c (KP840685); VRC08d (KP840686); VRC08e (KP840687); H03+06.A.0 (KP841501); VRC01UCA (MK032222). A similar process was used to build a phylogenetic tree for the VRC38 lineage, with one exception. Rather than using an inferred germline precursor, we germline-reverted framework 1, CDR1, framework 2, CDR2, framework 3, and framework 4 and used the junction nucleotide sequence of the lineage member with the least IGHV somatic mutation (VRC38.03). Trees were annotated and visualized in iTol (Letunic and Bork, 2019). While trees were constructed based on heavy chains, all VRC01 and VRC38 B cells had a correct light chain transcript, although sometimes additional light chain transcripts were also observed. One LIBRA-seq-identified VRC38 lineage member, 3602–1544, contained a single nucleotide deletion in the Cell Ranger-determined contig sequence in framework 2; this was manually corrected prior to inferring the phylogenetic tree.

Antibody expression and purification

For each antibody, variable genes were inserted into plasmids encoding the constant region for the heavy chain (pFUSEss-CHIg-hG1, Invivogen) and light chain (pFUSE2ss-CLIg-hl2, Invivogen and pFUSE2ss-CLIg-hk Invivogen) and synthesized from GenScript. mAbs were expressed in FreeStyle 293F or Expi293F mammalian cells (ThermoFisher) by co-transfecting heavy chain and light chain expressing plasmids using polyethylenimine (PEI) transfection reagent and cultured for 5–7 days. FreeStyle 293F (ThermoFisher) and Expi293F (ThermoFisher) cells were maintained in FreeStyle 293F medium or FreeStyle F17 expression medium supplemented with 1% of 10% Pluronic F-68 and 20% of 200 mM L-Glutamine. These cells were cultured at 37°C with 8% CO ₂ saturation and shaking. After transfection and 5–7 days of culture, cell cultures were centrifuged at 6000 rpm for 20 minutes. Supernatant was 0.45 μm filtered with Nalgene Rapid Flow Disposable Filter Units with PES membrane. Filtered supernatant was run over a column containing Protein A agarose resin that had been equilibrated with PBS. The column was washed with PBS, and then antibodies were eluted with 100 mM Glycine HCI at pH 2.7 directly into a 1:10 volume of 1 M Tris-HCI pH 8. Eluted antibodies were buffer exchanged into PBS 3 times using 10kDa Amicon Ultra centrifugal filter units.

Enzyme linked immunosorbent assay (ELISA)

For hemagglutinin ELISAs, soluble hemagglutinin protein was plated at 2 μg/ml overnight at 4°C. The next day, plates were washed three times with PBS supplemented with 0.05% Tween20 (PBS-T) and coated with 5% milk powder in PBS-T. Plates were incubated for one hour at room temperature and then washed three times with PBS-T. Primary antibodies were diluted in 1% milk in PBS-T, starting at 10 μg/ml with a serial 1:5 dilution and then added to the plate. The plates were incubated at room temperature for one hour and then washed three times in PBS-T. The secondary antibody, goat anti-human IgG conjugated to peroxidase, was added at 1:20,000 dilution in 1% milk in PBS-T to the plates, which were incubated for one hour at room temperature. Plates were washed three times with PBS-T and then developed by adding TMB substrate to each well. The plates were incubated at room temperature for ten minutes, and then 1 N sulfuric acid was added to stop the reaction. Plates were read at 450 nm.

For recombinant trimer capture for single-chain SOSIPs, 2 μg/ml of a mouse anti-AviTag antibody (GenScript) was coated overnight at 4°C in phosphate-buffered saline (PBS) (pH 7.5). The next day, plates were washed three times with PBS-T and blocked with 5% milk in PBS-T. After an hour incubation at room temperature and three washes with PBS-T, 2 μg/ml of recombinant trimer proteins diluted in 1% milk PBS-T were added to the plate and incubated for one hour at room temperature. Primary and secondary antibodies, along with substrate and sulfuric acid, were added as described above. Data are represented as mean ± SEM for one ELISA experiment. ELISAs were repeated 2 or more times. The area under the curve (AUC) was calculated using GraphPad Prism 8.0.0.

TZM-bl Neutralization Assays

Antibody neutralization was assessed using the TZM-bl assay as described (Sarzotti-Kelsoe et al., 2014). This standardized assay measures antibody-mediated inhibition of infection of JC53BL-13 cells (also known as TZM-bl cells) by molecularly cloned Env-pseudoviruses. Viruses that are highly sensitive to neutralization (Tier 1) and/or those representing circulating strains that are moderately sensitive (Tier 2) were included, plus additional viruses, including a subset of the antigens used for LIBRA-seq. Murine leukemia virus (MLV) was included as an HIV-specificity control and VRC01 was used as a positive control. Results are presented as the concentration of monoclonal antibody (in μg/ml) required to inhibit 50% of virus infection (IC₅₀).

Surface Plasmon Resonance and Fab competition

HIV-1 Env BG505 DS-SOSIP was produced either in GnT1- or 293F cells and purified as described previously (Do Kwon et al., 2015). The binding of antibodies 2723–2121 and 3602–870 to BG505 DS-SOSIP was assessed by surface plasmon resonance on Biacore T-200 (GE-Healthcare) at 25°C with HBS-EP+ (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, and 0.05% surfactant P-20) as the running buffer. Antibodies VRC01 and PGT145 were tested as positive control, and antibody 17b was tested as negative control to confirm that the trimer was in the closed conformation. Antibodies 2723–2121 and 3602–870 were captured on a flow cell of CM5 chip immobilized with ~9000 RU of anti-human Fc antibody, and binding was measured by flowing over a 200 nM solution BG505-DS SOSIP in running buffer. Similar runs were performed with VRC01, PGT145 and 17b IgGs. To determine their epitopes, antibodies 2723–2121 IgG and 3602–870 were captured on a single flow cell of CM5 chip immobilized with anti-human Fc antibody. Next 200 nM BG505 DS-SOSIP, either alone or with different concentrations of antigen binding fragments (Fab) of VRC01 or PGT145 or VRC34 was flowed over the captured 2723–2121 or 3602–870 flow cell for 60s at a rate of 10 μl/min. The surface was regenerated between injections by flowing over 3M MgCl₂ solution for 10 s with flow rate of 100 μl/min. Blank sensorgrams were obtained by injection of same volume of HBS-EP+ buffer in place of trimer with Fabs solutions. Sensorgrams of the concentration series were corrected with corresponding blank curves.

ADCP, ADCD, Trogocytosis, ADCC Assays

Antibody-dependent cellular phagocytosis (ADCP) was performed using gp120 ConC coated neutravidin beads as previously described (Ackerman et al., 2011). Phagocytosis score was determined as the percentage of cells that took up beads multiplied by the fluorescent intensity of the beads. Antibody-dependent complement deposition (ADCD) was performed as in (Richardson et al., 2018a) where CEM.NKR.CCR5 gp120 ConC coated target cells were opsonized with mAb and incubated with complement from a healthy donor. C3b deposition was then determined by flow cytometry with complement deposition score determined as the percentage of C3b positive cells multiplied by the fluorescence intensity. Antibody dependent cellular trogocytosis (ADCT) was measured as the percentage transfer of PKH26 dye of the surface of CEM.NKR.CCR5 target cells to CSFE stained monocytic cell line THP-1 cells in the presence of HIV specific mAbs as described elsewhere (Richardson et al., 2018b). Antibody-dependent cellular cytotoxicity (ADCC) was done using a GranToxiLux based assay (Pollara et al., 2011) with gp120 ConC coated CEM.NKR.CCR5 target cells and PBMCs from a healthy donor. The percentage of granzyme B present in target cells was measured by flow cytometry.

QUANTIFICATION AND STATISTICAL ANALYSIS

ELISA error bars (standard error of the mean) were calculated using GraphPad Prism version 8.0.0. The Pearson’s r value comparing BG505 and CZA97 LIBRA-seq scores for Ramos B-cell lines was calculated using Cytobank. Spearman correlations and associated p values were calculated using SciPy in Python.

DATA AND CODE AVAILABILITY

Data Availability Statement

Raw sequencing data used in this study are available on the Sequence Read Archive under BioProject accession number PRJNA578389.

Identified antibody sequences related to the VRC01 lineage have been deposited to GenBank under accession codes MN580550 – MN580578 (heavy chain) and MN580579 – MN580607 (light chain). Identified antibody sequences related to the VRC38 lineage have been deposited to GenBank under accession codes MN580608 – MN580625 (heavy chain) and MN580626 – MN580643 (light chain). Other sequences from antibodies identified and recombinantly produced as part of this study have been deposited to GenBank under accession codes MN580644 – MN580654 (heavy chain) and MN580655 – MN580665 (light chain).

Supplementary Material

1. Supplemental Figure 1. Purification of DNA-barcoded antigens and LIBRA-seq validation sorting schematic on Ramos B-cell lines, Related to Figures 1, 2, 4 and STAR Methods.

A. After barcoding each antigen with a unique oligonucleotide, antigen-oligo complexes are run on size exclusion chromatography to remove excess, unconjugated oligonucleotide from the reaction mixture. DNA-barcoded BG505 was run on the Superose 6 Increase 10/300 GL column and all other DNA-barcoded antigens were run on the Superdex 200 Increase 10/300 GL on the AKTA FPLC system. For size exclusion chromatography, dotted lines indicate DNA-barcoded antigens and fractions taken. The second peak indicates excess oligonucleotide from the conjugation reaction.

B. Binding of VRC01 or Fe53 Ramos B-cell lines to DNA-barcoded, fluorescently labeled antigens via flow cytometry. VRC01 cells bound to DNA-barcoded BG505-PE, DNA-barcoded CZA97-PE, and not DNA-barcoded H1 A/New Caledonia/20/99-PE. Fe53 cells bound to DNA-barcoded H1 A/New Caledonia/20/99-PE.

C. Gating scheme for fluorescence activated cell sorting of Ramos B-cell lines. VRC01 and Fe53 Ramos B cells were mixed in a 1:1 ratio and then stained with LiveDead-V500 and a DNA- barcoded antigen screening library consisting of BG505-PE, CZA97-PE, and H1 A/New Caledonia/20/99-PE. Gates as drawn are based on gates used during the sort, and percentages from the sort are listed.

D. For each experiment, the categorization of the number of Cell Ranger-identified (10X Genomics) cells after sequencing is shown. Each category (row) is a subset of cells of the previous category (row).

2. Supplemental Figure 2. Identification of antigen-specific B cells from donor NIAID45 PBMCs, Related to Figures 2 and 3.

A. Gating scheme for fluorescence activated cell sorting of donor NIAID45 PBMCs. Cells were stained with LiveDead-V500, CD14-V500, CD3-APCCy7, CD19-BV711, IgG-FITC, and a DNA- barcoded antigen screening library consisting of BG505-PE, CZA97-PE, and H1 A/New Caledonia/20/99-PE. Gates as drawn are based on gates used during the sort, and percentages from the sort are listed. These plots show a starting number of 50,187 total events. Due to the visualization parameters, 18 IgG-positive, antigen-positive cells are displayed, but 3400 IgG-positive, antigen-positive cells were sorted and supplemented with 13,000 antigen-positive B cells for single-cell sequencing. A small aliquot of donor NIAID45 PBMCs were used for fluorescence minus one (FMO) staining, and were stained with the same antibody panel as listed above with the exception of the HIV-1 and influenza antigens.

B. LIBRA-seq scores for BG505 (y-axis) and CZA97 (x-axis) are shown. Each axis represents the range of LIBRA-seq scores for each antigen. Density of total cells is shown. Overlaid on the density plot are the 29 VRC01 lineage members (dots) indicated in light green.

3. Supplemental Figure 3. Characterization of antibodies from donor NIAID45, Related to Figure 3.

A. Antigen specificity as predicted by LIBRA-seq was validated by ELISA for a variety of antibodies isolated from donor NIAID45. Antibodies were tested for binding to BG505, CZA97, and H1 A/New Caledonia/20/99. Data are represented as mean ± SEM for one ELISA experiment. ELISAs were repeated 2 or more times.

B. Binding of BG505 DS-SOSIP/GnT1- (resulting in Man5-enriched glycans) or BG505 DS-SOSIP/293F cells (complex glycans) to 2723–2121 IgG.

C. Binding of BG505 DS-SOSIP/GnT1- trimer to PGT145 IgG, VRC01 IgG, 17b IgG, and 2723–2121 IgG.

D. Inhibition of BG505 DS-SOSIP/GnT1- binding to 2723–2121 IgG in presence of VRC34 Fab (diamond), PGT145 Fab (square) and VRC01 Fab (triangle).

E. Neutralization of Tier 1, Tier 2, and control viruses by antibody 2723–2121 and VRC01. Results are shown as the concentration of antibody (in μg/ml) needed for 50% inhibition (IC50).

F. Levels of ADCP, ADCD, ADCT-PKH26, and ADCC displayed by antibody 2723–2121 compared to VRC01. HIVIG was used as a positive control and the anti-RSV mAb Palivizumab as a negative control.

4. Supplemental Figure 4. Identification of antigen-specific B cells from donor N90 PBMCs, Related to Figure 4 and 5.

A. Gating scheme for fluorescence activated cell sorting of donor N90 PBMCs. Cells were stained Ghost Red 780, CD14-APCCy7, CD3-FITC, CD19-BV711, and IgG-PECy5 along with a DNA-barcoded antigen screening library consisting of BG505-PE, KNH1144-PE, ZM197-PE, ZM106.9-PE, B41-PE, H1 A/New Caledonia/20/99-PE, H1 A/Michigan/45/2015-PE, H5 Indonesia/5/2005-PE, H7 Anhui/1/2013-PE. Gates as drawn are based on gates used during the sort, and percentages from the sort are listed. 5450 IgG-positive, antigen-positive cells were sorted and supplemented with 1480 IgG-negative, antigen-positive B cells for single-cell sequencing. A small aliquot of donor N90 PBMCs were used for fluorescence minus one (FMO) staining, and were stained with the same antibody panel as listed above without the antigen screening library.

B. Antigen specificity as predicted by LIBRA-seq was validated by ELISA for two antibodies isolated from donor N90. Antibodies were tested for binding to all antigens from the screening library: 5 HIV-1 SOSIP (BG505, KNH1144, ZM197, ZM106.9, B41), and 4 influenza HA (H1 A/New Caledonia/20/99, H1 A/Michigan/45/2015, H5 Indonesia/5/2005, H7 Anhui/1/2013). Data are represented as mean ± SEM for one ELISA experiment. ELISAs were repeated 2 or more times.

C. Binding of BG505 DS-SOSIP grown in GnT1- (resulting in Man5-enriched glycans) or 293F cells (complex glycans) to 3602–870 IgG.

6. Supplemental Figure 6. Sequencing preprocessing and quality statistics, Related to Figures 1, 2, 4 and STAR Methods.

A. Quality filtering of the antigen barcode FASTQ files. Fastp (Chen et al., 2018) was used to trim adapters and remove low-quality reads using default parameters. Shown are read and base statistics generated from the output html report from each of the Ramos B cell experiment (left), primary B cell experiment from donor NIAID45 (middle), and primary B cell experiment from donor N90 (right).

B. Shown is a distribution of insert sizes of the antigen barcode reads from the Ramos B-cell line experiment, as output from the fastp html report.

C. Shown is a distribution of insert sizes of the antigen barcode reads from the donor NIAID45 experiment, as output from the fastp html report.

D. Shown is a distribution of insert sizes of the antigen barcode reads from the donor NIH90 experiment, as output from the fastp html report.

Supplementary Figure 5. Supplemental Figure 5. Analysis of antigen reactivity for B cells from donor N90, Related to Figures 4–6.

A. Each graph shows the LIBRA-seq score for an HIV antigen (y-axes) vs. an influenza antigen (x-axes) in the screening library. The 901 cells that had a LIBRA-seq score above one for at least one antigen are displayed as individual dots. IgG cells (591 of 901) are colored orange and cells of all other isotypes are colored blue. Red lines on each axis indicate a LIBRA-seq score of one. Only 9 of the 591 IgG cells displayed high LIBRA-seq scores for at least one HIV-1 antigen and one influenza antigen, confirming the ability of the technology to successfully discriminate between diverse antigen specificities.

B. V gene usage of broadly HIV-reactive B cells. For each IGHV gene, the number of B cells with IgG or IgA constant heavy gene and high (>=1) LIBRA-seq scores for 3 or more HIV-1 SOSIP variants is displayed as a bar. The x-axis shows only IGHV genes used by at least 1 B cell with a high LIBRA-seq score for at least 1 HIV-1 antigen and an IgG or IgA C_H gene.

• LIBRA-seq: high-throughput mapping of BCR sequence to antigen specificity

• Identified HIV- and influenza-specific B cells in two HIV-infected subjects

• Predicted antigen reactivity for thousands of single B cells

• Identified a previously unknown broadly neutralizing HIV antibody