De novo antibody identification in human blood from full-length single B cell transcriptomics and matching haplotype-resolved germline assemblies

A donor-specific IGH assembly is highly contiguous and fully phased

To assess the full repertoire of donor-specific germline immune alleles, we depleted B and T cells from whole blood and extracted DNA for whole genome nanopore sequencing (Methods). Approximately 45× coverage of nanopore sequencing reads was de novo assembled and phased, resulting in a highly contiguous personalized whole genome assembly (statistics in Supplemental Table S1). The IGH, IGL, and IGK loci were isolated from this assembly for in-depth annotation and study.

The 2 Mb IGH region was contained within a 79 Mb contig and separated into two phase blocks. The phase block gap did not contain heterozygous single-nucleotide polymorphisms (SNPs), but was manually resolvable using a heterozygous 172 bp deletion between IGHV3-20 and the pseudogene IGHV3-19, resulting in a fully phased IGH region (Fig. 2A). Additionally, structural variant calling supported the inclusion of a 37 kb alternate contig representing the known 5-10-1/3-64D structural variant (Lefranc 2001a), which was manually placed into haplotype 2. All IGHJ and constant gene segments were present in the expected order. A block of IGHD gene segments was missing from each haplotype; haplotype 1 was missing D3-3 through D2-8, and haplotype 2 was missing D6-6 through D5-1.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 2.

A donor-specific, haploid assembly was created using Flye and HapDup (Methods). Colored bars indicate the position on the contig of functional V, D, J, and C gene segments. (A) The fully phased IGH locus, with the dashed lines depicting alternative structures in sections where the assembly differs structurally between the two haplotypes, such as the IGHV3-9 and IGHV1-8 structural variant. The D gene cluster is expanded to show that a block of D gene segments is missing from each haplotype. (B) The IGL locus, where the IGL C V-CLUSTER was assembled into a separate contig from the remaining V-CLUSTERs and the J-C-CLUSTERs. All gene segments were shared between both haplotypes.

The IGL V clusters A and B, as well as the J-C cluster, were contained on a fully phased 26 Mb contig; V cluster C was present on a separate phased contig (Fig. 2B). The IGL region typically contains 7–11 highly similar J-C cassettes, ranging from 3.5 to 5.6 kb, each containing a J and a C gene segment (Lefranc 2001c). This IGL J-C cluster contains nine cassettes, including one novel IGLC3 allele that differed from the closest known alleles by a single SNV (Supplemental Fig. S1).

Although ∼660 kb of IGK sequence was assembled, it was relatively discontiguous (split into 7+ contigs). Furthermore, IGK could not be fully resolved due to two major challenges: (1) It consists of two highly similar ∼400 kb segmental duplications and (2) these segmental duplications are separated by 800 kb of highly repetitive sequence (Lefranc 2001b). Consequently, the assembler collapsed the homologous sequences between the two segmental duplications, and allele calling was not possible for collapsed genes.

Of the 295 total annotated IG gene segments, 288 (97.6%) matched perfectly with entries in the IMGT or OGRDB databases (Lees et al. 2020). We identified 2 IGLV alleles, 1 IGHV allele, and 1 IGLC allele (which appeared in both haplotypes) with novel polymorphisms that are not present in the IMGT or OGRDB databases. Each of these novel alleles in the assembly are supported by reads that align to each locus (Supplemental Fig. S1), as well as by expressed IG transcripts (Supplemental Table S2). Reads aligned to the remaining two gene segments indicate the likely presence of a 1 nt deletion error in the assembled sequence. Of the three bona fide novel V gene alleles, two are in framework region 1, while the third (IGLV2-14*03-like) involved a Glu (E) to Asp (D) change in residue 56 (Kabat numbering) in CDR2. The latter could potentially impact antibody reactivity or provide a different pathway for affinity maturation, as has been shown for other germline polymorphisms (Yuan et al. 2023).

We observed consistent 5-methylcytosine (5mC) hypomethylation at CpG sites downstream from IGHV genes, as well as around the D and J gene clusters (Supplemental Fig. S2). The significance of this observation is unclear, but it warrants further investigation in future studies.

Single-cell transcripts and isoforms are differentially expressed among B cell subpopulations and unambiguously annotated using nanopore sequencing

To understand the expression of immune genes in response to an immune challenge, our donor received a measles, mumps, and rubella (MMR) vaccination followed by a longitudinal series of blood draws. Daily PBMC samples were monitored for the appearance of CD27⁺⁺/CD38⁺⁺ expressing ASCs by fluorescence-activated cell sorting (FACS) (Supplemental Fig. S3). This putative population of ASCs appeared on day 6 postvaccination, at which point we collected 80 mL of whole blood for B cell enrichment and sorting. We used a multiparameter FACS strategy (Methods), involving doublet exclusion, viability gating, and a T/monocyte/NK cell exclusion step to ensure a highly pure population of CD19⁺/IgM⁻ B cells for sorting into CD38⁺⁺/CD27⁺⁺ ASCs in the P7 gate (referred to henceforth as the ASC gate) and CD38⁻/CD27⁺ MBCs in the P8 gate (MBC gate). Approximately 1000 and 30,000 cells were collected from the ASC and MBC gates, respectively, and were subsequently processed for single-cell transcriptome sequencing.

To characterize single-cell expression within the B cell subpopulations, the single cell, whole transcriptome libraries of both the ASC gate and MBC gate were split in half and sequenced by both short-read and nanopore sequencing platforms. Single-cell cDNA sequences from the short-read and nanopore libraries of both the ASC and MBC gates were compared for cell barcode overlap and unique molecular identifier (UMI) counts per cell. UMI comparisons revealed a high degree of concordance between the two sequencing technologies in both the ASC gate (Supplemental Fig. S4) and MBC gate (Supplemental Fig. S5).

Because nanopore sequencing generates reads covering full-length transcripts, expression was tabulated not only at the gene-level, but also at the transcript-level in order to quantify isoform expression of each gene. While the gene expression landscape of cells in the MBC gate can be assessed using short reads (Fig. 3A), the transcript-level expression provided by long-read sequencing provides an alternative view (Fig. 3B) with the potential for uncovering cell state differences that would be obscured at the level of gene expression. An analysis of differential transcript expression across cell clusters (Methods) revealed a total of 1785 transcripts whose expression in cells from one cluster was significantly higher or lower than in all cells in the remaining clusters (Fig. 3C; Supplemental Table S3).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 3.

Single-cell expression data from the MBC gate. (A) Uniform Manifold Approximation and Projection (UMAP) visualization of cells clustered by gene expression quantified by short-read sequencing. (B) UMAP visualization of cells clustered by isoform expression quantified by nanopore sequencing. (C) Volcano plot showing isoforms that are differentially expressed in one cluster relative to the entire population of cells, with Bonferroni-adjusted P-values. One differentially expressed isoform of the apoptosis regulating gene BCL2 is highlighted. (D) Cluster-specific expression of the BCL2 gene and the differential transcript usage of three BCL2 isoforms. BCL2–202 isoforms make up only 4% of BCL2 transcripts in cluster 3 but represent 32% of BCL2 transcripts in cluster 8.

The differentially expressed transcripts can highlight how different isoforms of the same gene can be preferentially expressed in some cells. For example, the apoptosis regulator BCL2 has multiple splice variants (Warren et al. 2019) and shows distinct isoform usage patterns between clusters. In cluster 3, BCL2–201 dominates, with 69% of total BCL2 expression, while BCL2–202 contributes only 4%. Cluster 8 shows lower overall BCL2 expression but increased BCL2–202 usage at 32% (Fig. 3D). While both isoforms encode the same protein and differ only in their 5′ UTRs, this dramatic shift in isoform preference suggests important biological processes that would be missed by gene-level analysis alone.

For the ASC gate, gene expression clustering of the cells based on nanopore sequencing revealed multiple cellular subpopulations, including CD38⁺⁺ plasmablasts and plasma cells (Fig. 4A). Due to dynamic gating during the sorting process, some CD27⁺/CD38⁺ activated MBCs were also collected in the ASC gate, but these are easily identifiable as a separate population due to their distinct expression profiles (Fig. 4A,B). Based on expression, cluster 2 appears to represent a population of nonproliferating plasma cells (MKI67-low, IRF4-high), while cells in cluster 4 are likely actively proliferating plasmablasts (MKI67-high, IRF4-low) (Fig 4B). Cells in both clusters 2 and 4 produce a high level of IGHC transcripts, consistent with their role in high production and secretion of antibodies. The remaining 374 CD38⁻ cells in clusters 0, 1, and 3 are most likely residual MBCs that were missorted into the ASC gate, as they have with low expression of IGHC, DERL3, IRF4, and MKI67 but high expression of MS4A1/CD20 and HLA-DR. Nonetheless, our expression data were clearly able to differentiate these cells from plasmablasts or plasma cells.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 4.

Single-cell expression data from ASC gate. (A) UMAP visualization of cells clustered by gene expression quantified by nanopore sequencing. (B) Expression of marker genes differentiating in each cluster. (C) Proportion of reads assigned to IGHG1 and IGHG3 within each cell from the ASC gate in both the short-read and nanopore sequencing data sets. (D) Schematic illustrating the genes involved in glycosylation of the conserved asparagine residue in the CH2 domain of the heavy chain. (E) Differential expression of two genes in the glycosylation pathway, B4GALT1 and FUT8, in certain antibody isotypes of plasmablasts and plasma cells from the ASC gate. Each point represents a single cell.

Antibody specificity derives from the sequence of the complementarity determining regions (CDRs) of the light and heavy chains. The isotype of an antibody, such as IgA, IgD, IgE, IgG, and IgM, is determined by the sequence of the constant domains in the heavy chain. IgG and IgA are further divided into subtypes based on the hinge region. While each cell is expected to produce antibodies of a single isotype and subtype (e.g., IGHG1), the short-read sequencing data revealed ambiguous isotype assignments. Most ASCs had a mix of IGHG1 and IGHG3 assignments from short reads, likely due to high sequence similarity in the limited portion of the constant domains that are captured in the fragmented short-read sequencing library. In contrast, the nanopore sequencing data showed no such ambiguous isotype and subtype assignments, with nearly all full-length transcript reads from each cell mapping definitively to either IGHG1 or IGHG3 (Fig. 4C).

Antibody glycosylation phenotypes

Differential glycosylation of antibodies affects their Fc effector functions, serum half-life, and isotype shaping (Irvine and Alter 2020). GWAS studies have also linked SNPs in glycosyltransferases, such as ST6GAL1, B4GALT1, FUT8, and MGAT3, to IgG N-glycosylation phenotypes and certain isotypes (Shen et al. 2017; Wahl et al. 2018). Several core glycosylation genes are known to have specific impact on antibody-dependent cellular cytotoxicity (ADCC), isotype shaping, and inflammation (Fig. 4D). However, it is unclear if these genes are preferentially expressed at the level of single B cell clones.

We examined the expression of core antibody glycosylation genes in both the ASC and MBC gates. In the majority of plasmablasts and plasma cells in the ASC gate, FUT8, a fucosyltransferase, was significantly overexpressed relative to B4GALT1 in cells producing IgA1 and IgA2 (Fig. 4E). The reverse was true in cells producing IgG1. Since increased fucosylation is negatively correlated with ADCC function, our data suggest attenuation of ADCC activity in the secreted IgAs and enhanced ADCC activity in the secreted IgGs.

In the MBC gate, glycosylation patterns were not broadly associated with specific isotypes. However, high expression of MGAT3, MGAT2, and SLC35C1 was restricted to a minority of MBCs (Supplemental Fig. S6). Furthermore, high MGAT3 and SLC35C1 expression in these MBC clones was mutually exclusive, consistent with their opposing roles in modulating ADCC activity.

IG transcript enrichment and germline assembly mapping enables high-confidence consensus sequences for antibody synthesis

To first quantify the relative proportions of secreted and membrane-bound antibodies in each cell, we searched for reads containing the IGHC M exons, which reside on the 3′ end of the IGH transcript and harbor the trans-membrane, connecting, and cytoplasmic tail regions. Although the majority of transcripts contained the secreted form of IGHC lacking the M exons, a subset of MBCs primarily expressed membrane-bound transcripts (Fig. 5A). In contrast, all of the ASCs primarily expressed the secreted forms.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 5.

Characterizing single-cell IG sequences and antibody synthesis. (A) The percentage of IGH reads in each cell possessing the M exons indicative of a membrane-bound antibody, stratified by cell type. (B) Proportions of each isotype identified from consensus IGH sequences, stratified by cell type. (C) CDR3 amino acid length distributions for all productive IGH consensus sequences, stratified by cell type. CDR3 lengths are enumerated between the junction anchor residues: Cys 104 and Try 118 (IMGT numbering) for each sequence. (D) Mutation frequency compared to the germline sequences in the FWR1–4 and CDR1–3 regions for each cell type and isotype. (E) Dandelion plot of sequences clustered within each nonsingleton clone by VDJ amino acid sequence similarity. (F) Putative positive recombinantly expressed antibodies sequenced from the ASC gate were screened by flow cytometry for binding to MeV surface expressed proteins (F or RBP/H) on virus infected Raji-DC-SIGN B cells (multiplicity of infection [MOI] 0.01). MMR-vaccinated, hyper-immune sera from the donor at the time of cell collection was used as a positive control. (G) Neutralization assay in Vero cells for MMR-82 (MeV+) and hyper-immune sera against a GFP-expressing MeV-Edmonston recombinant virus (1000 IU/96-well). Data are presented as percent infection (mean ± SD) compared to infection in media alone (set as 100%). Experiments were performed in triplicate. Anti-HNV monoclonal antibody (mAb) is an irrelevant IgG1 mAb against an unrelated virus and serves a negative control.

Next, we paired heavy and light chain IG transcripts from each cell so that functional antibodies could be generated directly from the consensus sequences. The ASCs generated a large number of IG transcript reads from the whole transcriptome sequencing of each cell: on average 5319 from the light chain and 357 from the heavy chain.

The MBCs, however, had a much more diverse transcriptome that resulted in fewer IG transcript reads from each cell. To increase the IG transcript coverage for these cells, a second sequencing library was generated using biotin probe enrichment to enrich for IGH, IGL, and IGK-containing transcripts (Methods). This resulted in a 130-fold enrichment in the percentage of IG transcripts compared to the matching whole transcriptome sequencing data set and made it possible to recreate high-accuracy heavy and light chain sequences from several thousand MBCs.

We created a custom pipeline (Supplemental Fig. S7) to produce heavy and light chain IG consensus sequences for each cell (Methods). This pipeline generated 761 and 7653 consensus sequences that were predicted to be productive from 616 and 5564 cells in the ASC and MBC gates, respectively. We determined the isotype of each consensus sequence using IgBLAST (Fig. 5B). The length distributions varied by isotype and were consistent with expectations based on the literature and reference database (IMGT–GENE-DB) (Supplemental Fig. S8; Lefranc et al. 2005). To assess the accuracy of these consensus sequences, we examined the CH1 regions from the constant genes because they are unaffected by somatic hypermutation. In the 4040 productive IGH consensus sequences, we found 96% of the CH1 regions to be identical to the germline CH1 alleles, with a mean accuracy of 99.984% (Supplemental Fig. S9).

Observed variation in the length of sequences from each isotype can be explained by the variable lengths of the CDR3 sequences for each subpopulation (Fig. 5C). A Gaussian-like distribution of CDR3 lengths in the memory population was consistent with average CDR3 lengths in peripheral MBC populations in healthy individuals, with a median of 14 amino acids (aa) (Tian et al. 2007; Duty et al. 2009; Dong et al. 2020). We observed that 0.2% (eight sequences) of the MBC contained CDR3 lengths of 30 aa or longer, including one CDR3 of 37 aa, that were not selected in the ASC populations. Nonsymmetrical distributions in the ASC populations, plasma cells, and plasmablasts show evidence of length biases, indicative of selection.

Nucleotide mutation frequencies observed in the V gene segment were within the standard range for postgerminal center, class-switched B cell populations with an average mutation frequency of 7.7% across all isotypes (Fig. 5D; Wrammert et al. 2008; Dong et al. 2020; Budeus et al. 2023). Analysis of mutation by isotype revealed rarer IgD MBC and plasma cell populations (1% and 0.03% abundance within the two populations, respectively), which did not contain IgM transcripts, had high mutation frequencies (>12%), and had increased JH6 usage (see below section on germline-specific repertoire analysis). This population is reminiscent of class-switched IgD B cell populations, mostly observed in germinal centers, that are reported to have high mutation rates (>10%) and preferential usage of JH6 segments (Zheng et al. 2004; Koelsch et al. 2007; Seifert and Küppers 2016; Budeus et al. 2023).

The Immcantation (Gupta et al. 2015) suite of tools was used to call clones based on the heavy chain junction length and sequence identity; subclones were called by light chains when available; 1219 clones had multiple members with cluster sizes ranging from 2 to 16 (Fig. 5E). Certain clones varied in their heterogeneity; for example, a group containing 14 IgD clones had more divergent sequences than a more tightly clustered group of 11 IgA2 clones.

Because the sequencing data were generated using Oxford Nanopore Technologies’ R9 flow cells, we assessed whether significant consensus accuracy improvements could be expected using the newer R10 flow cells. We sequenced leftover cDNA from the MBC gate using an R10 flow cell and compared the resulting consensus VDJ regions to the cell barcode- and IG locus-matched sequences from the R9 data. Overall, the proportion of productive consensus sequences increased from 78% to 87%. The jump is mainly driven by IGH productivity, which increased from 80% to 95%; 656 out of 1010 barcodes with unproductive IGH sequences in the R9 data had productive IGH sequences in the R10 data, with the vast majority becoming productive due to a G homopolymer correction. Still, among the filtered sequences used for the final R9 data set, the vast majority (94%) of sequences agreed with the R10 consensus. For future experiments, R10 would give a higher proportion of productive data, but R9 data are still useful with sufficient filtering.

Antibody synthesis and binding results

To investigate the function of novel antibody sequences identified in the ASCs, we generated 100 monoclonal antibodies based on consensus heavy and light chain sequences. The first batch of 50 antibodies was synthesized as full-length IgG1 IGs, and the subsequent 50 were engineered as single-chain variable fragment (scFv)-IgG1-Fc fusions. This approach allowed for both the assessment of native antibody binding characteristics (full-length IgG1) and the rapid screening of multiple antibody specificities simultaneously (see Methods).

Initially, the first batch of antibodies were screened for binding to MMR virus antigens via commercial ELISA kits. One of the full-length IgG1 antibodies, MMR-32 yielded positive binding results. MMR-32 displayed apparent polyreactivity as it bound to ELISA plates coated with both mumps and rubella virus antigens (Supplemental Fig. S10A–C). None of these antibodies exhibited neutralizing activity against MeV, MuV, or RubV (Supplemental Fig. S10D–F).

Subsequent screening of the 50 scFv-Fc fusions by FACS on ExpiCHO cells transiently transfected with MeV and MuV fusion (F) and receptor-binding protein (RBP) expression constructs identified several clones demonstrating binding to MeV-F proteins (Supplemental Fig. S11), notably MMR-82 and MMR-205. Further characterization confirmed MMR-82 binding to native MeV envelope proteins on infected Raji-B cells (Fig. 5F). Importantly, MMR-82 effectively neutralized MeV infection (Fig. 5G).

Since these ASCs were not antigen-baited, we tested whether any of the recovered antibodies were specific to other common infectious agents. Since up to 85% of adults >40 years are seropositive for human cytomegalovirus (HCMV), we screened 50 mAbs for reactivity against HCMV in an immunofluorescent assay and detected at least one positive clone (MMR-8) (Supplemental Fig. S12).

None of the antibodies, including the scFv-Fc fusions, demonstrated positive binding to the viral antigens in this ELISA format while hyper-immune polyclonal sera demonstrated the expected reactivity in the ELISA assays (Supplemental Fig. S10A–C). The lack of reactivity among the monoclonal antibodies highlights the possibility of false-negative results with ELISA-based screening methods since MMR-82 was obviously a positive binder to MeV-F.

Germline-specific repertoire analysis of antibody sequences confirm B cell class and allows haplotype distinction

Our pipeline enabled unambiguous assignment of IGH and IGL gene family usage (Fig. 6A; Supplemental Figs. S13, S14), revealing distinct gene segment repertoires in cells collected from the MBC and ASC gates. Similarly, IGK repertoires, assigned using the IMGT–GENE reference database, showed distinct VK segment usage across populations (Supplemental Fig. S14). The MBCs exhibited a diverse VH and JH segment repertoire, with preferential use of VH3 and VH4 gene families (Fig. 6A), and JH4 segments (Supplemental Fig. S14), typical of healthy adult MBCs (Tian et al. 2007). IgD-only MBCs expressed a preferential use of JH6 segments reminiscent of IgD class-switched B cells (63.89% JH6 usage vs. 19.1% in total memory cells vs. 20.9% in IgG1 memory cells) (Supplemental Fig. S15). In contrast, plasmablasts and plasma cells displayed a more restricted and biased VH repertoire, suggesting selective biases for antigen-specific ASCs (Fig. 6A).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 6.

Donor-specific V gene usage and mutation rates. (A) Overall IGHV gene usage frequency for all complete and productive heavy chain consensus sequences with an in-frame CDR3. Cells with a low V gene identity, high mutation rates (≥15%) are stratified (light blue), as they are sometimes filtered out by programs such as Hi-V-QUEST (IMGT). (B) Statistics for ambiguous IGH and IGL V gene calls when using IgBLAST with various databases. (C) V gene usage for cells expressing one of the heterozygous V alleles (left), or that expressed a homozygous V allele but were haplotype-discriminable by a heterozygous J allele (right). (D) Ratio of replacement to silent mutations (R/S) in framework regions (FWRs) versus CDRs. The dotted line represents theoretical R/S ratios for unselected CDRs at R/S = 2.9. The red lines are the mean R/S values. The elevated replacement mutations in the CDRs suggest cells that have undergone antigen selection.

A subset of sequences showed low V segment identity (<85%) against germline alleles, with mutation rates of 7%–9% for MBCs and plasmablasts, and 20% for plasma cells. Though 11% in MBCs and 13% in plasma cells were from the IgD-only population, most of these highly mutated sequences were on IgG and IgA bearing cells. Some highly mutated B cells are to be expected in healthy adults, as multiple exposures to the same immunizing antigens can recall B cells into further maturation rounds. We found population-specific VH usage biases in these highly mutated sequences (Fig. 6A, light blue bars). For example, VH1–2 segments were exclusive to the plasma cell population, suggesting that these patterns reflect B cell maturation rather than technical artifacts.

When using a deduplicated IMGT–GENE database with IgBLAST 14% of all functional V allele calls representing 57 different IGH and IGL genes are ambiguously assigned to multiple equally likely alleles. When using a personalized reference database, ambiguous calls are reduced by over 90%; only 1.5% of calls, representing 15 genes, remain ambiguous (Fig. 6B). For example, while IGLV10-54 allele assignments remain ambiguous even when using the personalized database, the allelic ambiguity for IGHV3-30 is completely resolved by using the personalized database (Supplemental Fig. S16). The unambiguous calls are also needed to distinguish transcripts from each haplotype.

Sequences can be traced back to a specific haplotype if they include a gene or allele that is only present on one haplotype. Fifty-five percent of IGH consensus sequences could be assigned a haplotype based on a heterozygous IGHV or IGHJ allele; 27% and 28% of consensus sequences came from haplotypes 1 and 2, respectively. Only 0.3% of the consensus sequences had a contradictory haplotype assignment between the V allele and the J allele, which supports the phasing accuracy of the assembly and is likely the result of hypermutation. Although the overall usage of both haplotypes is similar, certain alleles are used more often than others. For example, IGHV7-4-1*02 was used by 115 cells, versus a single cell using the *01 allele (Fig. 6C). The latter is known to be a poorly expressed VH allele in humans (Ohlin 2021). Two other poorly expressed human IGHV alleles (IGHV1-2*05 and IGHV4-4*01), known to be in linkage disequilibrium and typically found on the same haplotype (Ohlin 2021), were also present in our haplotype 2 and were found in 9 and 0 cells, respectively (Fig. 6C). Altogether, these data support the veracity of our analytic pipeline.

Mutational analysis of IGH shows an elevated ratio of replacement to silent mutations (R/S) in the CDR regions (mean = 3.7–3.9 depending on cell type) compared to the FWR and compared to an estimated theoretical R/S ratio previously calculated for unselected, random mutation (2.9) (Fig. 6D; Dörner et al. 1997, Wrammert et al. 2008; Cho et al. 2019). Increases in average R/S ratios of CDRs from multiple rounds of affinity maturation and selection in germinal center reactions is generally observed in memory and ASC B cell populations, and similarly, we see these increased averages for our selected populations. The R/S pattern holds true for the consensus sequences regardless of the level of IGHV nucleotide identity.