Generation of a single-cell B cell atlas of antibody repertoires and transcriptomes to identify signatures associated with antigen specificity

Summary Although new genomics-based pipelines have potential to augment antibody discovery, these methods remain in their infancy due to an incomplete understanding of the selection process that governs B cell clonal selection, expansion, and antigen specificity. Furthermore, it remains unknown how factors such as aging and reduction of tolerance influence B cell selection. Here we perform single-cell sequencing of antibody repertoires and transcriptomes of murine B cells following immunizations with a model therapeutic antigen target. We determine the relationship between antibody repertoires, gene expression signatures, and antigen specificity across 100,000 B cells. Recombinant expression and characterization of 227 monoclonal antibodies revealed the existence of clonally expanded and class-switched antigen-specific B cells that were more frequent in young mice. Although integrating multiple repertoire features such as germline gene usage and transcriptional signatures failed to distinguish antigen-specific from nonspecific B cells, other features such as immunoglobulin G (IgG) subtype and sequence composition correlated with antigen specificity.


INTRODUCTION
The importance of antibodies in drug development has promoted interest in developing more rapid and efficient methods for monoclonal antibody discovery. 1,2 Traditionally, antibody discovery has relied heavily on experimental screening approaches such as hybridomas, phage, or yeast display and more recently B cell cloning 3,4 ; however, the emergence of genomics methods and in particular deep sequencing and bioinformatics are also contributing to antibody discovery. 5 For example, deep sequencing of antibody repertoires from immunized mice has been used to identify clonally expanded plasma cells that are associated with antigen specificity. 6,7 In addition, the emergence of single-cell sequencing (scSeq) technologies has made it possible to identify endogenously paired variable heavy (VH) and variable light (VL) chain regions from single B cells at high throughput, which can be used to subsequently reconstruct antibodies for experimental testing. 8,9 For example, our group recently performed scSeq of plasma cells following immunization or viral infection in mice, and subsequent expression and screening of antibodies from highly expanded clones were shown to correlate with antigen specificity. 10,11 A similar approach was applied to plasma cells isolated from convalescent COVID-19 individuals and led to the discovery of a panel of antibodies, including potent neutralizing antibodies against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). 12 An innovative approach based on B cell selection with DNA-barcoded antigens followed by scSeq (LIBRA-seq) has been effective for the rapid isolation of antibodies against multiple antigens. 13,14 Further development of scSeq methods has made it possible to obtain both transcriptome and antibody (or B cell receptor, BCR) repertoire information from single B cells at high throughput, [15][16][17][18] thereby enabling an integrated analysis of B cell transcriptional phenotypes within the context of antibody clonal populations. 19 This integrated scSeq approach was recently used to identify a large and diverse population of antigen-specific B cells from convalescent COVID-19 individuals, leading to the discovery of highly potent neutralizing antibodies. 20 Although this work demonstrates the promise of integrated scSeq for viral antigens, it remains unknown how antibody repertoire and transcriptome features can be used to iScience Article identify antigen-specific B cells, especially within the context of challenging therapeutic target antigens, such as self-antigens that are often targets for autoimmune disease or cancer. 21,22 The discovery of antigen-specific B cells to antigens with high homology to self-proteins may be potentially enhanced through immunization in aged individuals as previous studies have demonstrated that the redistribution of B cell subsets that occurs during aging may favor the formation of antibody responses toward self-antigens. 23,24 Furthermore, it remains unknown whether the potential loss of negative selection occurring with immunosenescence can improve the ability to identify diverse populations of antigen-specific B cells. [25][26][27] Here, we perform integrated scSeq of antibody repertoires and transcriptomes of B cells isolated from mice immunized with a model therapeutic protein antigen (human tumor necrosis factor receptor-2 [TNFR2]) followed by antibody-antigen specificity profiling at high throughput. We generate a singlecell atlas of 100,000 B cells and experimentally screen 227 antibodies, thereby uncovering signatures associated with antigen-specific B cells. Importantly, only in young mice were clonally expanded B cells found to be strongly associated with antigen-specific antibodies, with very few observed antigen-specific B cells discovered in the older cohort. In addition to supporting the discovery of a diverse panel of antigenspecific B cells against a therapeutically relevant target, scSeq can also relate functional properties of antibodies to repertoire and transcriptional features and gain greater insight on B cell selection.

RESULTS
Clonal expansion is detected in young and old mice following immunization To profile B cell selection following serial protein immunizations, we immunized a cohort of 3-month-old (3m) male C57BL6 mice (n = 5) with five successive injections of 10 mg of the extracellular domain of human TNFR2 mixed with 20 mg of the adjuvant monophosphoryl lipid A (MPLA). Our previous findings demonstrated that immunization of young mice with an immunogenic protein antigen (ovalbumin [OVA]) results in highly expanded plasma cells within the bone marrow. 11 For this study, we selected an antigen that has potentially reduced immunogenicity given its high sequence similarity (61.3%) to the murine homolog of TNFR2. In contrast to more immunogenic antigens, we expected a weaker immune response against TNFR2 given its high similarity to self-proteins. For this reason, we utilized serial protein immunizations in combination with a strong adjuvant to trigger a stronger response, as has previously been performed. 10 As aging may be linked to a more antigen-experienced immune repertoire (encounter of higher number of antigens throughout life) together with an increased production of autoantibodies, 28,29 we hypothesized that immunization with TNFR2 in aged mice could potentially trigger stronger memory B cell responses against self-antigens with high sequence similarity, such as human TNFR2. Furthermore, it has been observed that antibody-forming B cells in aged mice exhibit higher absolute numbers in certain organs following cognate antigen interactions. 30 Hence, we additionally performed the same immunization schemes with additional cohorts of 12-month-old (12m) (n = 3) and 18-month-old (18m) (n = 3) male C57BL6 mice. With the exception of one mouse in the young cohort, all mice exhibited high antibody titers against TNFR2 ( Figure S1).
We isolated bone marrow plasma cells (BM PCs) (CD138 hi , TACI hi ) and spleen B cells (CD19 hi , IgM low , IgD low ) by fluorescence-activated cell sorting (FACS) and performed single-cell antibody repertoire and transcriptome sequencing using the 10x genomics 5 0 immune profiling pipeline ( Figure 1A). Following iScience Article library preparation, deep sequencing and alignment to reference germline genes, we recovered a total of 65,000 cells containing exactly one heavy chain (HC) and one light chain (LC), with an average of 3,200 cells per mouse. The majority of cells were of the immunoglobulin M (IgM) isotype for both spleen and bone marrow ( Figure 1B). Cells were grouped together based on sharing identical heavy and light chain complementarity determining region 3 (CDRH3+CDRL3) amino acid sequences (hereafter referred to as clone), resulting in a few hundreds of unique clones for each mouse or organ repertoire. Further investigation revealed that the majority of the BM PC repertoires demonstrated extensive clonal expansion, with approximately 75% of clones being expanded (clones supported by more than one cell) ( Figures 1D and  S2). On the other hand, we observed that the majority of splenic B cell clones were not clonally expanded as they possessed only a single-cell barcode, which was comparable across all three age groups ( Figures 1D and S2). While the majority of expanded clones for BM PCs were IgM, spleen repertoires had an extensive presence of IgG subtypes ( Figures 1E, S3 and S4). Nevertheless, clonally expanded cells of IgM, IgG, and IgA isotypes were still present in both organs ( Figures S5, S6 andS7). Previous single-cell antibody repertoire sequencing studies suggest a strong correlation between class-switched clones and the number of amino acid sequence variants existing within an individual clone. 11, 31 Here, we observed that the number of cells for the most expanded IgG and IgA clones demonstrated a minor correlation with the number of amino acid variants, whereas some IgM clones contained a considerable number of amino acid variants ( Figure 1F). Finally, we did not observe any differences in clonal expansion, isotype distribution, or the number of amino acid variants between the different age groups.

Single-cell sequencing reveals organ and isotype transcriptional heterogeneity
Next, we leveraged the capabilities of scSeq to integrate transcriptome information with antibody repertoire features. To this end, we performed unsupervised clustering and uniform manifold approximation projection (UMAP) on the combined dataset (all cells across all mice and cohorts), which gave rise to 14 distinct cell clusters based on global gene expression (Figures 2A, S8A and S9). Examination of the transcriptional space clearly distinguishes between clusters populated principally by cells derived from either bone marrow or spleen or clusters with cells originating from both organs at similar ratios (clusters 8, 9, and 12) (Figures 2A and S8B); a high degree of reproducibility was observed across samples ( Figure S8C). Differential gene expression analysis was used to define specific cell populations ( Figures 2B and S10). 32 Not surprisingly, this demonstrated that cell clusters associated with bone marrow were characterized almost completely by plasma cell markers, which was in contrast to the splenic transcriptional space where multiple distinct B cell phenotypes were present ( Figure 2A). Overlaying isotype information onto the UMAP revealed separation of IgG-, IgM-, IgA-, and IgD-expressing cells ( Figure S11). Differential gene expression analysis revealed exclusive genes defining IgG (ApoE, Gimap4), IgA (Ccr10, Glpr1), IgD (Sell, Fcer2a), and IgM (Ggh, Slc3a2) isotypes, some of which are consistent with previous findings 10 ( Figure 2C). Moreover, as we removed Ig genes before gene expression analysis, these isotype-specific signatures suggest a strong influence of isotype fate to the expression profile of B cells and appear robust throughout different infection or immunization conditions. 10 Such differences might be used in the future to select PCs of a certain isotope given their lack of BCR surface expression.
We next determined if there are age-associated transcriptional changes between young and old mice. We restricted our analysis to the 3m and 18m cohorts and quantified differential gene expression separately for each organ ( Figure 2D). Interestingly, we observed for both bone marrow and spleen a substantial downregulation in young mice of AC133103.1, an uncharacterized gene marker, and substantial upregulation of Eif3f which has been implicated in protein translation and cell growth with its expression significantly decreased in many human cancers 33,34 ( Figure 2D). We determined if expanded B cell clones were transcriptionally distinct compared to non-expanded clones (clones supported by only one single cell). Owing to the high heterogeneity of the splenic B cell population, we focused our analysis on BM PCs ( Figure 2E). Identification of cluster-defining genes in the expanded BM PC pool was not obvious; however, we did observe a downregulation of Vpreb3, which is implicated in B cell maturation 35,36 ( Figure 2E).

Clonal expansion correlates with antigen specificity in young but not old mice
We next determined the extent to which clonal expansion is correlated with antigen-specific binding. We focused on the most expanded clones that possessed cells mostly with an IgG subtype as these have been previously shown to be preferentially associated with antigen specificity. 11 More specifically, we selected for the most expanded IgG clones of each group. We additionally included 34 expanded IgG clones that were found in both organs of individual mice. Lastly, we included four clones that were found across ll OPEN ACCESS iScience 26, 106055, March 17, 2023 iScience Article multiple mice ( Figure 3A and Table S1). We used a previously established system for rapid antibody cloning and expression in mammalian cells by CRISPR-Cas9 genome editing. 37,38 The resulting antibodies, representing 204 unique IgG clones, were then screened by ELISA for binding to the TNFR2 antigen ( Figure 3A and Table S1). Unexpectedly, we observed that in young mice, there was a considerably higher fraction of  Table S1). This was most apparent in BM PC repertoires, where approximately 31.3% of clones tested (31 out of 99) from young mice were antigen specific, in contrast to only 7.1% (4 out of 56) in older mice ( Figure 3B). In the spleen, we also detected 21.4% (6 out of 28) and 14.2% (3 out of 21) antigen-specific clones for young and old mice, respectively ( Figure 3C). Furthermore, independent of age and organ, we observed that a higher degree of clonal expansion was not correlated with antigen specificity as antigen-specific and nonspecific clones were evenly distributed throughout the most expanded clones of each repertoire ( Figures 3B and 3C).
Having discovered TNFR2-specific antibody sequences among clonally expanded cells, we next determined whether clones present in multiple organs of the same mouse would exhibit antigen specificity. These organ-overlapping clones showed higher levels of clonal expansion when compared to clones observed in only one organ ( Figure 3D). We expressed and determined the antigen specificity of several of these organ-overlapping clones, some of which coincided with previously expressed expanded clones (marked with a star [*] in Figures 3B and 3C). This revealed that approximately 33% (12/37) of organoverlapping clones were also antigen specific. ( Figure 3E and Table S1). Furthermore, antigen specificity and clonal expansion were not correlated across organ-overlapping clones ( Figure 2E). In contrast to the correlation observed between age and antigen specificity in the clonally expanded cells, we did not observe any age-associated biases to antigen specificity within the organ-overlapping antibody pool.

Investigating immune repertoire and transcriptome features of antigen-specific BM PCs
We next determined whether repertoire or transcriptome features could differentiate between recently activated TNFR2-specific clones and nonspecific ones. We focused our computational analysis on the BM PCs due to the higher number of experimentally validated antibodies with specificity to TNFR2. We initially quantified the mean number of somatic hypermutations (SHMs) per clone in the full-length VH and VL regions, which revealed that, on average, the TNFR2-specific fraction of clones exhibited lower levels of SHM ( Figure 4A). Next, we visualized the distribution of IgG subtypes (IgG1, IgG1B, IgG2C, and IgG3) which showed a higher percentage of the IgG1 subtype among the antigen-specific fraction of clones ( Figure 4B), an observation that was also reflected on the cellular level ( Figure S12A). We next sought to relate TNFR2 specificity to germline gene usage. Visualizing the distribution of HC and LC V gene usage across TNFR2-specific and nonspecific sequences did not suggest any TNFR2-associated biases ( Figure S12B). We further determined if certain VH-VL germline combinations were enriched in the antigen-specific or nonspecific fraction. Circos plots of the experimentally verified antigen-specific and nonspecific clones did not show enrichment for certain V gene combinations in either of the two groups ( Figure 4C). The same observation was made when looking at other germline features, such as the J gene usage ( Figure S12C). To investigate whether sequence convergence could be detected within the binding fraction of TNFR2-specific clones, we initially visualized the amino acid sequences for the most frequent CDRH3 and CDRL3 lengths. This initially revealed little indication of amino acid or biochemical bias between antigen-specific and nonspecific sequences within the CDR3 region ( Figures 4D and S12D). Next, we constructed sequence similarity networks based on the edit distance of CDRs, 39 which demonstrated that antigen-specific clones demonstrated clustering across a range of edit distances (Figures 4E and  S12E). Some of these antigen-specific clusters contained clones utilizing different germline genes despite similar CDR3s, potentially suggesting convergent sequence motifs ( Figures 4F and S12F).
We next investigated whether distinct gene expression profiles could be detected between TNFR2-specific and nonspecific clones as these signatures could correspond to genes involved in recent iScience Article recruitment and migration compared to long-lived BM PCs present prior to immunization. Following differential gene expression analysis, we observed significantly expressed markers, both when including all TNFR2-specific sequences ( Figure 4G) and when restricting our analysis to the different age cohorts (Figures S13A and S13B). Among these differentially expressed genes, we saw an upregulation in the non-binder compartment of Slpi, a gene found to inhibit class switching, 40 and Xbp1, a gene associated with PC differentiation and survival. 41,42 The expression and identification of such genes might indicate a relevant signature of B cells with binding capability against the immunized protein. iScience Article DISCUSSION Here, we used single-cell antibody repertoire and transcriptome sequencing to investigate the extent that repertoire features can be leveraged to discover antigen-specific antibodies from immunized mice. scSeq allowed us to relate individual transcriptomes to the antibody repertoire for nearly one hundred thousand B cells, thereby providing large-scale insight into the relationship between gene expression, clonal selection, and antigen specificity. We observed comparable levels of clonal expansion in repertoires from both young and aged mice following serial immunizations with human TNFR2. In addition, we observed class-switched antibodies among the most expanded clones in both spleen and bone marrow repertoires.
The emergence of scSeq workflows has made it possible to comprehensively test and reconstruct the specificity of antibodies based on their immune repertoire profiles. Using antibody expression and screening, we were thus able to demonstrate that a fraction of expanded cells in the repertoires of young mice produced antibodies with specificity to TNFR2 following immunization. This fraction was comparable to our previous results in the context of immunization with OVA, where approximately 45% of the most expanded clones were found to be antigen specific. 11 Moreover, our results demonstrate an increased proportion of clonally expanded B cells that were not antigen specific in aged mice, suggesting either the occurrence of bystander expansion of B cells following vaccination as previously described 16,43 or that immunization-induced clonal expansion fails to exceed levels present in naive repertoires. This decrease of antigen specificity in aged mice may be linked to previous reports that naive B cell repertoires have restricted clonal diversity in aged individuals or that IgM+ B cells accumulate in the bone marrow during aging, thereby reducing available space for newly recruited B cells. [44][45][46] Other parameters linking immune senescence to a decreased number of antigen-specific plasma cells could involve alterations of pro-B/pre-B proliferative capacities, 47,48 a decline of number and size of germinal centers 49 and deficiencies in class switching. 50 Further measuring of affinities and, in the case of antiviral immunizations, neutralization potential would be beneficial to assess how aging impacts the quality of the antibody response. In addition, we observed that antigen specificity was largely stochastic and could not be predicted based on the clonal rank of expanded cells in both young and old mice, which is in accordance with previous results. 11 Having discovered TNFR2-specific sequences using clonal expansion and clonal overlap, we integrated single-cell transcriptomes with antibody repertoires in an attempt to investigate whether other factors could be incorporated into the selection criteria such as Ig isotype, SHM, or transcriptional signatures. We focused on the BM PC repertoires due to the higher number of experimentally validated antibodies with specificity to TNFR2. These repertoires remain particularly interesting since plasmablasts, short-, and long-lived PCs significantly contribute to circulating antibodies present in serum. 51 This comparison revealed that TNFR2-specific PCs had minor differences in immune repertoire features and similar transcriptional profiles to nonspecific PCs. More specifically, our data suggested that antigen-specific clones were less mutated in the V-and J-regions and preferentially expressed the IgG1 subtype. We expected that immunization would lead to clear transcriptional differences relating to recent selection between TNFR2-specific and nonspecific PCs. However, we observed minor transcriptional differences between these two populations of cells when performing unsupervised clustering and differential gene expression analysis. Transcriptional differences were however observed in IgM-, IgD-, IgA-, and IgG-expressing B cells, which was consistent with previous results, 31 as well as expansion-specific transcriptional clustering. 52,53 It should be noted that in this study we have explicitly selected for expanded IgG-expressing clones, in an attempt to increase the percentage of antigen-specific sequences as to our experience expansion is correlated with antigen specificity. 10,11 For other immunological questions, a more stochastic selection of clones might be more appropriate.
Single-cell antibody repertoire sequencing provides a detailed molecular quantification of clonal selection. Thus far, the majority of repertoire studies conducted have examined lymphocytes following infection or immunization using model proteins. 20,[54][55][56][57][58][59] In contrast, there is limited information regarding the behavior of humoral immunity triggered by antigens with high host-sequence similarity, which are common therapeutic targets for indications such as cancer and autoimmune disease. Therefore, our results have implications for the discovery of monoclonal antibodies targeting such therapeutic antigens, such as TNFR2, which is implicated in pro-and anti-inflammatory conditions. 21,22 It should be noted that in this study we utilized a hyper-immunization scheme (5 injections iScience Article of adaptive immunity and influence its quality. 60 Other aspects that have not been assessed in this study that could deeply affect the immune response, and thus antibody discovery, include the antigen dose, the choice of the adjuvant and delivery system, the immunization route, and the time intervals between priming and boosting. [61][62][63][64][65][66][67] For example, it has been shown that longer time intervals between priming and boosting can lead to a stronger B-cell response, eliciting a higher number of germinal center (GC)-B cells and ASCs. 68,69 Another interesting avenue that could also be employed in the discovery of monoclonal antibodies targeting such therapeutic antigens is multi-dose priming which has shown potential in mounting protection upon vaccination. 70 Finally, inclusion of both male and female samples should be considered in future studies, as sex-based differences in immune function and response might influence the outcome. 71,72 The response to vaccination in older individuals is characterized by lower titers of vaccine-specific antibodies, increase in autoantibodies, and shorter duration of antibody responses. 73,74 Despite these observations, it has been considered that antibody avidity or neutralization capacity is unaffected with age, suggesting that changes in vaccine-induced antibody responses during aging are a result of reduced antibody titer rather than differences in antibody functionality. [75][76][77] Here, our data strongly suggest that the functionality of ASCs is impaired with age and thus likely contributes to poor responses observed in older individuals. This is in line with observations concerning defects in the GC response in older individuals and poor formation of antigen-specific memory B cells. 78 Experiments of adoptive cell transfer of B cells from aged mice to young recipients showed that these B cells retained their ability for activation by antigen stimulation and their ability for GC reaction entry. 79 Future experiments investigating the functionality of adoptively transferred B cells from aged individuals to younger hosts would be of interest for a better understanding of how antibody function is affected with age. In addition, future studies focusing on the interplay of B cells and other components in the aged GC microenvironment will shed light on the functional impairment of ASCs with age. A promising example has been the transient blockade of IL-10 receptor signaling to improve Tfh-dependent GC response and the use of toll-like receptor 7 (TLR7) agonist to overcome age-related defects in cDC2 priming of T cells. 80,81 Understanding these age related differences in the GC microenvironment and ways to overcome them will be pivotal for vaccine efficacy in the elderly.

Limitations of the study
The results presented in this study require caution when interpreting and extrapolating findings relating to fundamental PC biology. One major concern relates to the low number of mice included in the study. Experimental variability between individual animals may influence the differences observed on both repertoire features and gene expression profiles. Similarly, the study considered mice of the same strain and sex, C57BL/6 male (Janvier), unquestionably influencing the results at hand. Moreover, we acknowledge that our transcriptional findings and repertoire feature comparisons of antigen-specific and nonspecific sequences are based on a low number of cells and clones, which may also contribute to the observed profiles. Future studies, including larger cohorts and a higher number of antigen-specific PCs, are required to corroborate our presented hypotheses. Finally, profiling repertoire features and antigen specificity of PC repertoires in naive mice would help elucidate mechanistic insights regarding PC antigen-driven selection.

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

ACKNOWLEDGMENTS
We acknowledge and thank Dr. Christian Beisel, Elodie Burcklen, and Ina Nissen at the ETH Zurich D-BSSE Genomics Facility Basel for support and assistance. We also thank the D-BSSE FACS facility for experimental support. Funding: This work was supported by the European Research Council Starting Grant 679403 (to STR), ETH Zurich Research Grants (to STR and AO), the Swiss National Science Foundation (grant 310030_166078 and 310030B_185374 to AO), and an ETH Seed Grant (AY).