BCR Selection and Affinity Maturation in Peyer’s Patches Germinal Centers

SUMMARY

Antigen-binding B cell receptor (BCR)/antibody variable regions are encoded by exons assembled in developing B cells by V(D)J recombination¹. Immensely diverse primary BCR repertoires derive from mechanisms that diversify V(D)J gene segment junctions that contribute to the antigen contact complementarity-determining region 3 (CDR3)¹. Primary B cells undergo antigen-driven BCR affinity maturation via somatic hyper-mutation (SHM) and cellular selection in germinal centers (GCs)^2,3. While most GCs are transient³, gut microbiota-dependent intestinal Peyer’s patch (PP) GCs are chronic⁴, with little known about their BCR repertoires or SHM patterns. To elucidate physiological PP GC BCR repertoires, we developed a high-throughput repertoire/SHM assay. PP GCs from different mice expand public clonotypes that often have canonical IgH CDR3s that appear far more frequently in naïve B cell repertoires than predicted, due to junctional biases during V(D)J recombination. Some public clonotypes are gut microbiota-dependent and encode antibodies reactive to bacterial glycans, while others are not. SPF fecal transfer to germ-free (GF) mice restored germ-dependent clonotypes, directly implicating BCR selection. Indeed, we identified recurrently selected SHMs in such public clonotypes, implicating affinity maturation in mouse PP GCs under homeostasis conditions. Thus, persistent gut antigens select recurrent BCR clonotypes to seed chronic PP GC responses.

Each newly generated (naïve) B lymphocyte expresses a BCR that harbors, respectively, one IgH and IgL variable region¹. However, collectively, B cells express huge repertoire of primary BCRs with different CDR3s, due to mechanisms that diversify variable (V), diversity (D) and joining (J) gene segment junctions during V(D)J recombination¹. Junctional diversification mechanisms (deletions, P element formation, and non-templated nucleotide (N) region additions by terminal deoxynucleotidyl transferase (TdT)¹) are estimated to generate 10¹¹ or more distinct BCRs^5,6, greatly exceeding the approximately 10⁸ steady-state primary mouse B cells⁷. Not all junctions are randomly diversified. For example, in the absence of TdT during lymphocyte development in the fetal liver, V(D)J junctions frequently use short micro-homologies within two recombining gene segments to generate canonical CDR3s^8–13.

Primary B cells undergo antigen-driven BCR affinity maturation via activation-induced cytidine deaminase (AID)-initiated SHM and cellular selection in “conventional” peripheral lymphoid GC structures found in lymph nodes or spleen^2,3. PPs are the major lymphoid tissues involved in gut adaptive immune responses, with continuous GCs in exposure to gut microbiome and dietary contents⁴. Compared to conventional GCs, PP GCs are unique in two major aspects⁴. First, they arise in the context of a homeostatic response instead of acute response. Second, the potential antigens PPs face are of enormous complexity. While chronic PP GCs are T cell dependent^14,15, affinity maturation to gut antigens at steady state has not been demonstrated^16,17. Mice in which BCR-deficient B cell development is driven by a knock-in EBV LMP2A gene form chronic PP GCs but not antigen-induced splenic GCs¹⁸. Moreover, identical pre-rearranged knock-in productive and passenger IgH VDJ exons have essentially identical, intrinsic SHM patterns in mouse PP GC B cells¹⁹, raising the question of whether PP GCs are sites of BCR diversification in an antigen non-specific manner, as occurs in chicken, sheep and rabbits^20–22. On the other hand, oral immunization of mice can induce conventional, acute PP GC responses that generate oligoclonal, affinity-matured antibodies²³, suggesting chronic PP GC responses are likely shaped by gut micro-environment². In depth elucidation of physiological BCR repertoires of chronic PP GCs from nontransgenic mice requires application of an appropriate high-throughput assay.

To elucidate PP GC B cell repertoires in specific-pathogen-free (SPF) C57BL/6 WT mice, we upgraded our HTGTS-Rep-Seq assay^24–26 to cover all V(D)J segment usage and also SHM profiles across the full length of IgH and IgL variable region exons. This “Rep-SHM-Seq” method (Extended Data Fig. 1a,b,c; Methods) uses optimized bait primers designed against a degenerate region at the 3’ end of all J_H or J_L segments to provide an unbiased assay for determining full IgH and IgL variable region exon repertoires. We further generated a downstream bioinformatic pipeline that incorporated clonotype and SHM analyses (Extended Data Fig. 1d; Methods). Clonotype is conventionally defined as identical V and J segments with more than 90% CDR3 nucleotide sequence identity. Rep-SHM-Seq uses genomic DNA as template and, critical to our experiments, detects both productive and non-productive V(D)J rearrangements. Thus, by assaying SHM patterns of non-productive V_HDJ_H sequences from many GC samples, we generated an intrinsic SHM pattern database for most mouse V_Hs. With this database, we employ a hierarchical Bayesian model to statistically compare SHM rates of each V_H nucleotide for given non-productive (intrinsic) patterns and productive patterns in GC B cells to follow affinity maturation (See Methods). Such experiments cannot be done by existing RNA-based repertoire sequencing methods, since out-of-frame non-productive mRNAs cannot be reliably measured²⁷.

To evaluate ability of Rep-SHM-Seq to follow a well-characterized immune response, we assayed splenic naïve and GC B cell repertoires of NP-CGG intraperitoneally (IP) immunized C57BL/6 mice. We detected significant increases in V_H1–72 and V_λ1 utilization above naïve B cell levels (Extended Data Fig. 2a, b), in accord with their known dominance in the C57BL/6 NP response^28–30. V_H1–72 GC rearrangements were associated with two major clonotypes and V_λ1 rearrangements were associated with one major clonotype in all 5 mice examined. Both V_H1–72 clonotypes utilized D1–1 and J_H2, (Extended Data Fig. 2c), and their associated V_H1–72 had G to T point mutation at residue 98 of CDR1 (33W->L, Extended Data Fig. 2d), all of which are known features of C57BL/6 NP response^29,31.

To understand the nature of chronic PP GCs, a key question is whether PPs are seeded by B cells harboring specific BCRs that then undergo clonal expansion⁴. To address this question, we analyzed naïve and GC B cell repertoires in PPs of SPF C57BL/6 mice. V_H, D, and J_H, as well as V_L and J_L, utilization patterns were essentially identical between PP and splenic naïve B cells (Extended Data Fig. 3), demonstrating a BCR repertoire with stable composition of variable region gene segment utilization across different lymphoid tissues of SPF C57BL/6 mice. However, PP GC B cell repertoires had significantly enriched utilization of 10 V_Hs compared to naïve V_H repertoires (Fig. 1a). Notably, we identified public clonotypes that recurred in multiple mice for 9 of these V_Hs, most of which are the largest clonotype among all the clonotypes using the same V_H (Fig. 1b). From the 18 different mice analyzed, these recurrent IgH clonotypes were present at the following frequency: V_H1–12RC (11/18), V_H1–11RC (8/18), V_H1–47RC (7/18), V_H9–4RC(6/18), V_H6–3RC (5/18), V_H6–6RC (5/18), V_H11–2RC (4–18),V_H2–9RC (4/18), V_H4–1RC (3/18) (Fig. 1b, Extended Data Fig. 4). To further assess the significance of recurrent PP GC clonotypes enriched in SPF mice, we assayed for this phenomenon in 10 germ-free (GF) mouse pools (3 mice each) to obtain sufficient GC B cell numbers (Extended Data Fig. 5a–c). Notably, the V_H1–12RC (shown in green) and V_H1–47RC (shown in red) were highly recurrent in GF mice (each found in 10/10 GF mice, Fig. 2a, Extended Data Fig. 5d), suggesting they are bacteria-independent. GF mice PP GCs had additional recurrent clonotypes not found in SPF mice, some of which use the same V_H and are closely related with only 1–2 amino acids difference in the CDR3 (Fig. 1, 2a, Extended Data Fig. 5e, Extended Data Table 1a).

Fig. 1. |. V_H and clonotype enrichment recurs in PP GCs.

a, V_H repertoire of productive V(D)J junctions in PP GC on the “+” y-axis vs naïve B cells on the “-” y-axis from 18 SPF mice. The ratio of each functional V_H segment is shown as % of total productive V(D)J rearrangements. V_H segments are ordered linearly with respect to relative proximity to the Ds based on their chromosome coordinates. Data is plotted as mean ± SEM. P-values for V usage are calculated by two-sided Mann-Whitney U test with FDR correction (*: p < 0.1, **: p < 0.05, ***: p < 0.01). See Supplementary Table 4 for exact p-values. b, GC productive clonotype distribution of each indicated V_H is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_H and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. Colors indicate clonotypes found among different mouse categories as compared with Fig. 2.

Fig. 2 |. Microbial and non-microbial factors underlie recurrent PP GC clonotypes.

a-c, V_H repertoire of productive V(D)J junctions in PP GC vs naïve B cells plotted as mean ± SEM from 10 samples of GF mice with 3 mice in each sample as noted by asterisk (a), 5 GF mice colonized with SFB-mono feces (b) and 10 GF mice colonized with SPF feces (c) for 3 weeks. GC productive clonotype distribution of each indicated V_H is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_H and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. Colors indicate clonotypes found among different gnotobiotic mouse categories, including SPF mice in Fig. 1b. P-values for V usage are calculated by Mann-Whitney U test with FDR correction (*: p < 0.1, **: p < 0.05, ***: p < 0.01). See Supplementary Table 4 for exact p-values. The recovery of germ-dependent clonotypes V_H1–64RC and V_H6–3RC in GF+SPF mice is significant compared to GF mice. (p = 0.0023 for V_H1–64RC and p = 0.0001 for V_H6–3RC; Fisher’s exact test, two-sided).

We extended these studies to 5 GF mice colonized with segmented filamentous bacteria (SFB)-mono-colonized (SFB-mono) mouse feces^32,33 and 10 GF mice transferred with SPF mouse feces. SFB grow primarily in the terminal ileum where they induce Th17 cell responses^32,33 and stimulate potent PP GC reactions^34,35. We orally-gavaged GF mice with SFB-mono or SPF mouse feces for 3 weeks and found, as expected³⁵, increased PP GC B cell numbers compared to those in GF mice (Extended Data Fig. 5a–c). The V_H1–12RC and V_H1–47RC clonotypes recurred in GF mice following either SFB colonization or SPF mouse feces treatment (Fig. 2b, c, Extended Data Fig. 5d). However, SFB colonization in GF mice further enriched a V_H1–64RC (3/5 mice, purple) not observed in SPF mice (Fig. 1, 2b, Extended Data Fig. 5f, Extended Data Table 1b). Given that our SPF mice harbor SFB among an enormous diversity of microbiota, detection of this clonotype might be masked by domination of their PP GC responses by other microbes. In this regard, oral gavage of GF mice with SPF feces also elicited V_H1–64RC (4/10 mice), and a few additional recurrent clonotypes not found in SPF mice (Fig. 1, 2c, Extended Data Fig. 5g, Extended Data Table 1c), perhaps due to a more-limited pool of competing bacteria following fecal transfer³⁶. The V_H6–3RC, observed in SPF mice, was also recovered as a recurrent PP GC clonotype (6/10 mice, blue) following SPF fecal transfer (Fig. 1, 2c, Extended Data Fig. 5h).

The recurrence and recovery of different PP GC clonotypes under different gnotobiotic conditions indicates that these clonotypes were selected by gut microbiota- or non-bacteria-derived antigens. In support of this, we identified 6 positively selected unique SHMs among 4 of the recurrent PP GC clonotypes (Fig. 3a, b, Extended Data Fig. 6a, b). Consistent with affinity maturation, all selected SHMs render amino acid changes and occur within a CDR region. Specifically, the V_H9–4RC had enrichment of 59K->I (Fig. 3a); the V_H6–3RC had enrichment of 61H->Y (Fig. 3b); the V_H1–12RC had enrichment of 50A->V and 55N->D (Extended Data Fig. 6a); the V_H1–64RC had enrichment of 58T->P and 59N->I (Extended Data Fig. 6b). For other recurrent clonotypes where we did not find significantly elevated mutations above the intrinsic pattern (Extended Data Fig. 7), highly complex gut antigens, conceivably could select for various mutations in different cells, which are diluted in the population and mask selected mutations. Alternatively, some recurrent clonotypes may already reach high antigen affinity with germline V_H sequences and/or intrinsic SHMs and, thus, not require further affinity maturation.

Fig. 3 |. Affinity maturation of recurrent PP GC clonotypes.

a-b, V_H SHM profile of indicated PP GC clonotypes and intrinsic pattern, plotted as mutation rate at each nucleotide. Sequences were stratified by overall V_H mutation rate¹⁹ (see Methods). Significance is determined by hierarchical Bayesian modeling with PEP_0.1 (see Methods; *: PEP_0.1<0.05, **: PEP_0.1<0.01, ***: PEP_0.1<0.005). PEP_0.1 value and amino acid changes are denoted for significantly enriched SHMs. Data is presented as mean ± SEM from mice containing indicated clonotype: 6 SPF mice (a), 5 SPF mice and 6 GF+SPF mice (b). c-g, Microbial glycan microarray reactivities of indicated mAbs. See full antibody sequences in Supplementary Table 5. Data is shown as average fold enrichment over negative-control mAb (anti-human PD1) of 4 technical repeats. Major peaks are annotated with glycan number and glycan names are indicated on the right panel.

We assayed the IgL repertoires of the PP GC samples from 9 SPF mice and found three significantly enriched V_Ls (Vκ14–111, Vκ4–68 and Vκ6–15) and three that were dominantly utilized (Vκ1–110, Vκ4–72, Vκ5–43 albeit usage did not reach significance), each with several recurrent IgL clonotypes (Extended Data Fig. 8a, Supplemental Table 1). To assess whether these recurrent IgL clonotypes were associated with the PP GC IgH recurrent clonotypes to form antibodies, we inferred IgH/IgL pairing based on correlated frequencies in the 9 SPF mice (see Methods) and identified potentially paired IgL clonotypes for 5 recurrent IgH clonotypes. All 5 IgH/IgL pairs recurred in multiple mice. Specifically, V_H1–12RC/Vκ5–43Jκ2, V_H1–47RC/Vκ1–110Jκ1, V_H9–4RC/Vκ1–110Jκ5, V_H112RC/Vκ4–72Jκ5 and V_H6–6RC/Vκ4–68Jκ2 recurred in 3 to 4 out of 9 SPF mice (Extended Data Fig. 8b). To confirm IgH/IgL pairing, we performed 10x Genomics single cell immune profiling RNA sequencing (scRNA-seq) of 4 SPF mice, and found three exact pairs (V_H1–12RC/Vκ5–43Jκ2, V_H1–47RC/Vκ1–110Jκ1 and V_H11–2RC/Vκ4–72Jκ5) (Extended Data Fig. 8b), indicating these IgH/IgL pairs make antibodies in vivo. We also detected related IgL clonotypes with a different Jκ for V_H1–12RC and V_H1–47RC by scRNA-seq (Extended Data Fig. 8b), suggesting choice of Jκ is flexible³⁷ for these antibodies. In this regard, scRNA-seq identified a very similar antibody to the inferred V_H9–4RC/Vκ1–110Jκ5 in which the paired IgL clonotype used Jκ4 (Extended Data Fig. 8b). We did not find V_H6–6RC by scRNA-seq due to more limited data set obtained by this method. For several other IgH recurrent clonotypes in SPF mice where we could not confidently infer recurrently paired IgLs by Rep-SHM-Seq (See Methods), V_H6–3RC was of particular interest since it was recovered in SPF fecal transferred GF mice (Fig. 2c). In this regard, we found a Vκ2–109Jκ2 clonotype paired with V_H6–3RC by scRNA-seq (Extended Data Fig. 8b), which allowed us to express an antibody with the V_H6–3RC in vitro for further characterization.

To further characterize recurrent PP antibodies, we cloned the V_H1–12RC/Vκ5–43Jκ2, V_H94RC/Vκ1–110Jκ4, V_H6–3RC/Vκ2–109Jκ2 pairs from real sequencing reads that contained mutations in both IgH and IgL, including the selected V_H mutations, expressed these mAbs and screened them on a microbial glycan microarray³⁸. Consistent with presence in GF mice, the V_H1–12RC mAb did not react with any microbial glycans in the microarray (Extended Data Fig. 6c). However, the V_H9–4RC mAb showed substantial binding to surface glycans of P. stuartii, P. aeruginosa and P. vulgaris (Fig. 3c), and the V_H6–3RC mAb showed substantial binding to lipopolysaccharides of S. marcescens (Fig. 3d), consistent with being selected by gut bacteria in the SPF mice. Since the glycan microarray used is from human bacterial pathogen species that do not normally reside in SPF mice³⁸, the V_H9–4RC and V_H6–3RC mAbs were likely induced by bacterial species from the mouse gut microbiome bearing similar antigenic glycan structures³⁹. Furthermore, reverting all the IgH and IgL mutations of the V_H9–4RC and V_H6–3RC mAbs to germline sequences greatly reduced their reactivity to these glycans (Fig. 3e, f), implicating affinity maturation of these antibodies in mouse PP GCs. Finally, replacing the IgH CDR3 of the mutation-reverted V_H6–3RC antibody with a same length CDR3 that utilizes the same V_H and J_H, but has a 50% difference in amino acid sequences (TGPRGFAY vs TDPAWFPY) completely abolished its glycan binding activity, highlighting the importance of its CDR3 sequence per se (Fig. 3g). For this antibody it is possible that, as for many antibodies, the IgL sequence per se is not a key factor for binding^40,41 or that the IgL isolated in the context of scRNA-seq actually fulfills this function.

AID-deficient mice, which do not undergo variable region exon SHM or BCR affinity maturation, never-the-less form spontaneous splenic GCs, which are thought to expand B cells with primary BCRs due to induction of aberrantly expanded system-wide gut microbiota^42,43. To assess whether appearance of recurrent clonotypes in response to the complex gut microbiome could occur in non-gut lymphoid tissues, we assayed spontaneous splenic GCs in AID-deficient mice and, indeed, found public clonotypes (Extended Data Fig. 9a). One such clonotype (V_H1–81RC, brown) is shared between splenic and PP GCs in the AID-deficient mice (Extended Data Fig. 9a, b), indicating potential selection by the same antigen. Thus, selection of public clonotypes within induced splenic GCs also applies to complex gut antigens as well as simple hapten antigens (e.g. NP-CGG, Extended Data Fig. 2). Note that the clonotypes found enriched in the hyper expanded PP GCs^42,44 (Extended Data Fig. 5a–c) of AID-deficient SPF mice were distinct from those of WT SPF mice (Fig. 1, Extended Data Fig. 9b, Supplementary Table 2), which, among other possibilities, might result from hyper-expanded gut flora with altered bacteria compositions^42,44.

Sequence analyses revealed that 8 of the 9 recurrent PP GC clonotypes found in SPF mice, and a subset of those in GF mice with or without fecal transfer from SFB-mono or SPF mice, were each associated with a canonical CDR3 in multiple mice (Fig. 4a, Extended Data Fig. 4, 5d, f, h, Extended Data Table 1). In this regard, the CDR3 of V_H1–12RC was found in 32 of 43 total mice from all categories analyzed and that of V_H1–47RC was found in 14 of 43 (Fig. 4a). Other canonical CDR3s occur in the mouse backgrounds where the clonotype existed (Fig. 4a). The remaining small portion of non-canonical CDR3s within each clonotype represents junctions with the same V_HD or DJ_H junction and occasional mismatches, many of which do not involve “N” or “P” regions (Supplementary Table 3) and could be explained by SHMs in the canonical CDR3 sequences. Given that the theoretical number of mouse CDR3s that can be generated by V(D)J junctional diversification exceeds mouse naïve B cell numbers by several orders of magnitude^5–7, the question arises as to how the same CDR3 nucleotide sequences occurred in PP GCs in many different mice. To address this question, we used IGoR⁴⁵ to estimate the intrinsic generation probability (P_gen) of the recurrent CDR3s. The P_gens of the 8 recurrent PP GC CDR3s are significantly higher than the average P_gen of the total CDR3 repertoire (Fig. 4b), and each is also among the highest of all possible CDR3s encoding the same amino acid sequence (Fig. 4c). These findings suggest that the recurrent PP GC CDR3s belong to a subset of CDR3s over-represented in primary repertoires due to biases in junctional diversification^8–13,46, which, in theory, could reflect evolution of the mouse’s V(D)J recombination machinery to generate antibodies recognizing common gut antigens. Thus, the recurrence of PP GC CDR3s likely results from combined influences of strong antigen-dependent BCR selection on set of V(D)J exon sequences that, due to junctional diversification bias, occur in primary repertoires at higher than predicted frequencies.

Fig. 4 |. Intrinsic junctional biases contribute to PP GC clonotype recurrence.

a, Table showing the observed recurrence of each PP GC clonotype and its canonical CDR3 in the mouse categories found with each clonotype. b, Violin plot comparing P_gen distribution of the 8 recurrent PP GC CDR3s with that of 578,654 CDR3s from sequenced naïve B cells. P-value is calculated by two-sided Mann-Whitney U test. c, Violin plot showing P_gen distribution of all possible CDR3 sequences encoding the same amino acids (AA) sequence as the canonical CDR3 for each PP GC recurrent clonotype. The number of CDR3 sequences for each plotted clonotype is: 3072 (V_H1–12RC), 147456 (V_H1–47RC), 6144 (V_H9–4RC), 3072 (V_H11–2RC), 2048 (V_H6–3RC), 1536 (V_H6–6RC), 9216 (V_H1–11RC), 221184 (V_H2–9RC). Red dot indicates the location of the canonical CDR3 among each P_gen distribution. Box plots are presented with median, upper and lower quartiles and whiskers showing 1.5x interquartile range.

The nature of homeostatic PP GC Ig repertoires, whether BCR selection is involved in their formation, and whether they undergo affinity maturation have been intriguing questions⁴ that we have now addressed (Extended Data Fig. 9c). Chronic PP GCs of C57BL/6 mice recurrently express a set of clonotype-specific antibodies, mostly with canonical CDR3 sequences, generated more frequently than expected in the naïve B cell repertoire. These antibodies are selected by gut microbial or non-microbial antigens to seed PP GCs, where they undergo affinity maturation. The highly selected PP GC antibodies could contribute to the host-microbiome symbiosis by targeting specific bacterial populations and confer protective capacity against pathogens via cross-reactivity on the surface glycans of microbiota.

Methods

Mice, immunization and cell lines

All mice used in this study are C57BL/6 background except the human V_H1–2 mouse model, which was of mixed 129/Sv and C57BL/6 genetic background. WT specific pathogen free (SPF) mice were purchased from Charles River Laboratories International and maintained in SPF conditions. AID^−/− mice were generated in the Alt lab from previous studies and maintained in SPF conditions in the animal facility of Boston Children’s Hospital. All SPF mice contained gut SFB as verified by 16S rRNA gene quantitative PCR analysis of fecal bacterial genomic DNA as described³². The human V_H1–2 mouse model was described previously⁵. Germ-free (GF) mice were maintained in germ-free isolators in the Littman lab in the animal facility of the Skirball Institute of the NYU School of Medicine. For SFB-mono or SPF fecal transfer, 5-week old GF mice were orally gavaged with SFB-mono or SPF mouse feces and kept in gnotobiotic iso-cages for 3 weeks. All animal experiments were performed under protocols approved by the Institutional Animal Care and Use Committee of Boston Children’s Hospital and New York University School of Medicine, with compliance to all relevant ethical regulations. For NP-CGG immunization, WT mice aged 8–12 weeks were immunized intraperitoneally with 100 μg of NP-CGG (N-5055A, Biosearch Technologies) in 100 μl PBS mixed with 100 μl of Imject® Alum (Thermo Scientific). Mice were sacrificed at day 10 after immunization with NP-CGG. The mouse Cer/Sis-deleted v-Abl pro-B cell line was made by CRISPR/Cas9-mediated targeting of the Cer/Sis elements that lie between V_κ and J_κ intervening region within an ATM-deficient v-Abl kinase transformed pro-B cell lines that contains an Eμ-Bcl2 transgene as described previously⁴⁷. Cer/Sis-deletion in the v-Abl pro-B cell line was authenticated by southern blot for the genomic modification. The Cer/Sis-deleted v-Abl pro-B cells were cultured in RPMI medium containing 15% (v/v) FBS, and were treated with 3 μM STI-571 for 4 days to induce V(D)J recombination prior to genomic DNA preparation²⁶. The C57BL/6 ES cell line was derived from a wild-type C57BL/6 mouse. Cell lines were not tested for mycoplasma contamination.

B cell isolation from mouse spleen and PPs

Spleen and PPs were dissected out from 8–12 weeks old mice, prepared into single cell suspensions and purified by EasySep® Negative Selection B cell Enrichment Kit (Stem Cell Techonologies) according to the manufacturer’s protocol. Purified B cells were stained with anti-B220-PE (1:2000) (eBiosciences), anti-CD38-APC (1:200) (eBiosciences) and anti-GL7-FITC (1:200) (Biolegend). GC (B220⁺GL7⁺CD38⁻) and nonGC (B220⁺GL7⁻CD38⁺) B cells were sorted from the same sample by fluorescence-activated cell sorting (FACS) in the Department of hematology/oncology flow cytometry research facility at Boston Children’s Hospital. Genomic DNA from sorted cells was prepared using a DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturer’s protocol. 4 to 18 independent mice were analyzed for each category as indicated in the respective figures.

Rep-SHM-Seq methodology

Rep-SHM-Seq was derived from HTGTS-Rep-seq²⁴ with several modification. To capture full-length V(D)J sequences in recovered junctions for SHM analysis, we designed bait primers close to the coding ends of J_Hs and J_Ls and used Illumina MiSeq 2 × 300-bp paired-end sequencing. Mixed J_H or J_L primers were designed from a highly degenerative region, so that we minimize the amplification biases caused by different primers. In this regard, we employed a mouse model with human V_H1–2 (hV_H1–2) replacing the mouse V_H81X with IGCR1 deletion so that hV_H1–2 accounts for half of the V_H usage in mature splenic B cell repertoire²⁵. From splenic B cells of these V_H1–2 replaced mice, we made a library from a hV_H1–2-specific bait to determine the ratio of each J_H in hV_H1–2DJ junctions, compared these rations with those of libraries made from mixed J_H1–4 baits and found a good correlation (r = 0.96). For Igκ repertoires, we employed a v-Abl kinase transformed pro-B cell line with Cer and Sis element deleted so that proximal V_κ3 family genes are highly utilized⁴⁸. We made libraries using a V_κ3 family degenerate bait (V_κ3d, annealing to V_κ3–2,7,10,12) and from mixed J_κ1,2,4,5 baits. The J_κ ratios in V_κ3–2,7,10,12J_κ junctions from the two set of baits gave a strong correlation (r = 0.91). For Igλ repertoires, we used V_λ1,2 degenerate bait and mixed J_λ1,2,3 baits to compare J_λ ratios in V_λ1,2J_λ junctions, which gave a strong correlation (r = 0.88). Mixed J_{κ, λ} baits were further optimized using WT mouse splenic naïve mature B cells to give a κ:λ ratio of 94%:6%, which is consistent with prior findings^49,50. These bait primers are listed in Supplementary Table 6.

The bioinformatics pipeline for Rep-SHM-Seq was constructed with three modules: run, mut and clonal. The “run” module was used to preprocess MiSeq reads and assign V, D, J segments, similar to the HTGTS-Rep-seq²⁴ pipeline, but with more stringent filters: only reads with more than 98% of the sequences having a sequencing Quality Score ≥20 were kept; only joined reads (R1 and R2 had an overlap region ≥10 bp and mismatch rate ≤8%) were kept. The qualified reads were assigned to V, D, J segments using IgBLAST software⁵¹ with default parameters and determined as productive configuration when the V(D)J rearrangement is in-frame and contains no stop codon; out of frame rearrangements or those with stop codons are considered as in a non-productive configuration. Germline V, D, J gene sequences were obtained from IMGT database⁵², manually curated, and used to generate IgBLAST sequence databases. We applied various stringencies to filter reads that can align to V, D, and J segments (IgBLAST score >150, total alignment length >100, overall mismatch ratio <0.1). The “mut” module was used to identify V segment mutations and make SHM profile. Only reads with more than 50% V length coverage were kept for SHM profile analysis. Mutations were identified by parsing the reads with the inferred germline sequence collected for each V segment to calculate the mutation frequency at each nucleotide position to profile SHM through the whole V exon. SHM profiles for productive and non-productive sequences were generated separately. The “clonal” module was used to cluster clonotypes based on CDR3 sequences using a previously described method⁵³. The reads with the same assigned V and J segments and CDR3 junction length were grouped. Within each group, reads were hierarchically clustered using single linkage with Hamming distance measured by their junction sequences (distance threshold = 0.1, corresponding to >90% CDR3 sequence identity). The number of clonotypes and input cells for each GC sample is summarized in Supplementary Table 7.

Library preparation for Rep-SHM-Seq and data analysis

All libraries were made using 400 ng genomic DNA as starting materials as described²⁴, so that the level of PCR amplification was kept the same among different samples. In cases when less than 400 ng genomic DNA was obtained from GC or non-GC B cell samples, genomic DNA from JM8a3, a C57BL/6 ES cell line, was added to make the starting DNA template amount consistent among samples. The library preparation procedure was mostly done as previously described^24,54 with a few modifications in the linear amplification-mediated PCR step, where we used one reaction of 50 μl instead of 8 reactions and performed 100 cycles instead of 80 cycles.

To get purer populations of GC and naïve B cells for analysis and minimize potential cross-contamination during FACS sorting, we further filtered sequencing reads by keeping mutated reads for B220⁺GL7⁺CD38⁻ samples as GC B cells and non-mutated reads for B220⁺GL7⁻CD38⁺ samples as naïve B cells. Duplicate junctions were included in the analyses as previously described^1,24.

V segment usage analysis

For each sample, the ratio of each functional V segment was calculated as % usage among productive V(D)J junctions. For GC vs naïve repertoire comparison, V segment usages from multiple samples were plotted in the same chart above and below x-axis along chromosome coordinates. A complete list of functional V_H and V_L segments in the order plotted along x-axis is shown in Supplementary Table 8. To minimize potential variance differences caused by data size differences, the libraries were each normalized by random sampling to the smallest library within each set (Supplementary Table 9). Significantly enriched V segments in GCs were identified by two-sided Mann-Whitney U test and unpaired Student’s t-test with multiple test correction by FDR (p-values shown in Supplementary Table 4). Similar patterns and degree of significance were obtained when compared without normalization.

Recurrent clonotype and CDR3 analysis

To identify recurrent GC clonotypes across different samples, we pooled the GC productive reads from normalized libraries together to carry out clonal clustering^53,55. In our study, a clonotype is considered as “enriched” in one sample only if it makes up more than 0.3% of all productive junctions, and as “recurrent” only if it is enriched in more than one mouse. The consensus CDR3 DNA sequence plot for each clonotype was generated by WebLogo as previously described⁵⁶. To determine the frequency of identified CDR3s in naïve repertoire, we used IGoR as a background model⁴⁵. We trained IGoR on the VDJ sequences of naïve B cells, with log likelihood converged after six expectation-maximization (EM) iterations. Then, we predicted P_gen for each VDJ sequence of naïve B cells, as well as each recurrent CDR3 by inputting its VDJ sequence. We compared the P_gen distribution of the 8 recurrent PP GC CDR3s with that of all VDJ sequences of naïve B cells by two-sided Mann-Whitney U test, and visualized them by violin plot generated by ggplot2⁵⁷. We also generated all possible ‘back-translated’ nucleotide sequences that could be translated to the same amino acid sequence of each recurrent CDR3, and predicted their P_gens after inserting them back to VDJ sequence. We examined the location of P_gen of observed recurrent ones in the P_gen distribution of ‘back-translated’ sequences.

SHM analysis with hierarchical Bayesian model

To compare SHM mutation rate between productive sequences in a clonotype and background intrinsic non-productive allele, at each nucleotide site of the V gene, we developed a hierarchical Bayesian model to deal with the variable read depth among mouse samples.

For a nucleotide site, denote the mutation rate is θ_i in a mouse individual (i = 1~m for m biological replicates in one group). The observation x_i mutant reads in n_i total reads covering the site is modeled as sampling from a binomial distribution with parameter θ_i. The mutation rate θ_i of m biological replicates is modeled as sampling from a shared prior beta distribution with parameters α and β. Beta(α, β) distribution can be re-parameterized as η=α + β and μ=α / (α + β), where μ is the mean of the distribution, representing the average mutation rate for the whole group of biological replicates. Here is the detailed model specification.

$$\begin{array}{l}\text{μ~ Uniform (0,1)}\\ \text{η~Gamma(1,1)}\\ \text{α=μ}\cdot \text{η; β=(1-μ)}\cdot \text{η}\\ {\text{θ}}_{\text{i}}\text{~Beta(α,β)}\\ {\text{x}}_{\text{i}}\text{~Binomial}\left({\text{n}}_{\text{i}}\text{;}{\text{θ}}_{\text{i}}\right)\end{array}$$

Given data, we applied JAGS⁵⁸ to generate Markov Chain Monte Carlo sample points to depict the posterior distribution of the parameters. We ran two parallel chains, each ran 100,000 iterations after 60,000 burn-in iterations. To compare SHM profile between a recurrent productive clonotype and intrinsic non-productive allele, we generated posterior sample points of μ and μ_intrinsic, calculated μ_diff = μ - μ_intrinsic, and the posterior probability P(μ_diff > Δ), where the biological effect size Δ is set to 0.1. We reported sites with posterior probability P(μ_diff > 0.1) > 0.95, or correspondingly posterior error probability(PEP)(μ_diff > 0.1) = 1 - P(μ_diff > 0.1) < 0.05 as significant sites. To deal with variation of SHM level in different mice, sequences from all samples were stratified by overall V_H mutation rate¹⁹ and those with mutation rate lower than 2.5% were used for the analysis.

Heavy-light chain pair inferrences

To obtain possible pairing between recurrent heavy-chain and light-chain clonotypes, we examined the usage percentage of each clonotype in each sample, and ranked light-chain clonotypes for each heavy-chain clonotype by a similarity measure of usage percentage among m samples defined as following. Denote the usage percentage of a heavy-chain clonotype as a m-dimensional vector X = [x₁, x₂, …, x_m], with its average as $\overline{x}=\frac{1}{m}{\sum }_{i=1}^m{x}_i$, and fluctuation vector $\text{Δ}X=X-\overline{x}$. Similarly, denote the usage vector of a light-chain clonotype as $Y=\left[y_1,y_2,\dots ,y_m\right]$, and also with its average $\overline{y}=\frac{1}{m}{\sum }_{i=1}^m{y}_i$ and fluctuation vector $\text{Δ}Y=Y-\overline{y}$. The similarity measure we used is defined as ${\cos }_{-}\mathrm{similarity}(\text{Δ}X,\text{Δ}Y)\cdot |X|\cdot |Y|$ , where the cosine similarity measured the similarity in two vector’s direction is calculated as ${\cos }_{-}\mathrm{similarity}(\text{Δ}X,\text{Δ}Y)=\cdot (\text{Δ}X\cdot \text{Δ}Y)/(|\text{Δ}X|\cdot |\text{Δ}Y|)$, and vector length $|X|=\sqrt{{\sum }_{i=1}^m{x}_i^2}$. We multiplied the cosine similarity by the usage percentage vector length, because the pairing between clonotypes with higher usage is more reliable. In order to keep confident pairing results, we only reported pairings with the similarity measure larger than 10⁻³.

Single cell RNA sequencing and data analysis

Droplet-based scRNA-seq datasets were produced using a 10x Genomics Chromium system. B cells isolated from PP GCs were sorted into 80% methanol/PBS and 15,000 cells were loaded onto the Chromium™ Controller instrument following the manufacturer’s recommendations. Cells were partitioned into Gel Beads in Emulsion in the Chromium™ Controller instrument where cell lysis and barcoded reverse transcription of RNA occurred. Libraries were prepared using 10x Genomics Library Kits and sequenced on an Illumina NextSeq500 according the manufacturer’s recommendations. Raw sequencing files were aligned to the mouse V(D)J sequence using Cell Ranger: V(D)J Pipelines (10x Genomics) and the V usage and clonotype profiles were generated and visualized by Loupe V(D)J Browser. Note that due to low viability of GC B cells after isolation in vitro, which is critical for cell recovery according to 10x genomics manufacture’s protocol, we were able to recover only 600~1200 cells for each sample. The low cell recovery rate and low yield of scRNA-seq for GC B cells may introduce some variability in profiling the GC BCR repertoire.

Production of monoclonal antibodies (mAbs)

Heavy chain sequences and their paired light chain sequences of the recurrent PP GC antibodies were determined by Rep-SHM-Seq and confirmed by scRNA-seq. The gblock gene fragments containing the heavy chain and the light chain sequences with or without mutations (Supplementary Table 5), with human constant region sequences (IgG1, Igκ) and a 6xHis tag at the C terminus of heavy chain were synthesized and cloned into pcDNA3.1+ plasmid. The antibodies were generated using the Expi293 expression system (ThermoFisher Scientific, Waltham, MA) according to the product manual. Basically, cell culture supernatants were harvested 5 days after transfection, cleared of cells by centrifugation at 3000 g for 15 min, and subsequently purified by HPLC coupled with HisTrap HP histidine-tagged protein purification columns (GE Healthcare Life Science, Chicago, IL, USA). mAbs were dialyzed by SnakeSkin Dialysis tubing, 10k MWCO, 22mm and stored at 4°C in PBS. The concentrations of the mAbs purified were determined by ELISA with goat anti-human IgG (Southern Biotech).

Microbial glycan microarrays

mAbs were diluted to 5 μg/ml and assayed against the microbial glycan microarray by the Consortium for Functional Glycomics at the Protein-Glycan Interaction Core (CFG) and National Center for Functional Glycomics at Beth Israel Deaconess Medical Center in Boston, MA. Raw data is publicly available under accession numbers CFG_3624, and CFG_3645 (www.functionalglycomics.org). A human anti-PD1 mAb was used as negative control mAb for this assay, and signals for all samples were quantified by calculating the fold change over the anti-PD1 mAb.

Other Statistical analysis

No statistical methods were used to predetermine sample size. All samples were randomly selected and researchers were blinded whenever possible. Fisher’s exact test and Spearman’s correlation coefficient were used as indicated in the figure legends. Statistical tests with appropriate underlying assumptions on data distribution and variance characteristics were used. Statistical analysis was performed by R 3.5.1 and Prism (version 6, GraphPad Software).

Data availability

The next-generation sequencing data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database under the accession number GSE140795. All figures have associated raw data deposited.

Code availability

The computational pipeline of Rep-SHM-Seq and code for statistical analyses tools used in this study are available at https://github.com/Yyx2626/HTGTSrep.

Extended Data

Extended Data Fig. 1 |. Overview of Rep-SHM-Seq.

a, Schematic showing the experimental procedure of the Rep-SHM-Seq method. P, productive; NP, non-productive; MID, multiple identifier. b, Diagram of IgH locus in the hV_H1–2 mouse model. The proportions of each J_H in hV_H1–2DJ junctions are compared between libraries made from splenic B cells with hV_H1–2 bait vs mixed J_H1–4 bait primers. c, Diagram of IgL locus of the v-Abl kinase transformed pro-B cell line. The J_κ proportions in V_κ3–2,7,10,12J_κ junctions are compared between libraries made from V_κ3d bait and mixed J_κ1,2,4,5 bait primers. The J_λ proportions in V_λ1,2J_λ junctions are compared between libraries made from V_λd1,2 bait and mixed J_λ1,2,3 bait primers. Positions of bait primers are indicated. Three biological repeats are included for each set of primers. r, Spearman’s correlation coefficient. P-values are determined by two-sided Spearman’s correlation test. d, Schematic showing the bioinformatic pipeline of Rep-SHM-Seq. PE, paired-end; QC, quality control.

Extended Data Fig. 2 |. Rep-SHM-Seq detects NP-specific GC selection.

a-b, V_H and V_L repertoire of productive V(D)J junctions in splenic GC vs naïve B cells plotted as mean ± SEM from 5 mice with NP-CGG IP immunization for 10 days. V_H and V_L segments are each ordered linearly based on their chromosome coordinates, with respect to relative proximity to the Ds or J_Ls. V_κs are plotted on the left of V_κs. GC productive clonotype distribution of each indicated V_H or V_L is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_H or V_L and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. P-values for V usage are calculated by two-sided Mann-Whitney U test with FDR correction (*: p < 0.1, **: p < 0.05, ***: p < 0.01). See Supplementary Table 4 for exact p-values. Enrichment of V_H14–3, the key V_H in BALB/c mouse NP response^59,60, is detected here by Rep-SHM-Seq with two recurrent clonotypes (Supplementary Table 2). c, Representative junctional structure of indicated common IgH clonotypes aligned with germline V_H (red), D (blue) and J_H (orange), with resected sequences shown in grey. Underline indicates microhomology. All possible segment assignments by IGoR with probabilities are shown in Supplementary Table 10. Canonical CDR3 sequences are shown in black, amino acid sequences in purple, and the clonotype consensus in sequence logo pictures. The number of mice containing the canonical CDR3 sequence or the clonotype consensus is indicated in parentheses following each sequence. d, V_H SHM profile of indicated clonotypes and intrinsic pattern, plotted as mutation rate at each nucleotide. Sequences were stratified by overall V_H mutation rate¹⁹ (see Methods). Data is presented as mean ± SEM from 5 mice. Significance is determined by hierarchical Bayesian modeling with PEP_0.1 (see Methods; *: PEP_0.1<0.05, **: PEP_0.1<0.01, ***: PEP_0.1<0.005). PEP_0.1 value and amino acid changes are denoted for significantly enriched SHMs.

Extended Data Fig. 3 |. Stable composition of variable region exon segments usage in naïve B cells across tissues and mice.

V_H (a), D, J_H (b), V_L (c) and J_L (d) repertoires of productive V(D)J junctions in splenic vs PP naïve B cells plotted as mean ± SEM from the 5 NP-CGG IP immunized mice. r, Spearman’s correlation coefficient. P-values are determined by two-sided Spearman’s correlation test.

Extended Data Fig. 4 |. Representative junctional structure of indicated recurrent PP GC IgH clonotypes in SPF mice.

Junctional structures are aligned with germline V_H (red), D (blue) and J_H (orange), with resected sequences shown in grey. Underline indicates microhomology. All possible segment assignments by IGoR with probabilities are shown in Supplementary Table 10. For certain clonotypes, the D could not be accurately annotated because of the short, aligned D sequence. Canonical CDR3 sequences are shown in black, amino acid sequences in purple, and the clonotype consensus (CDR3 nucleotide sequences with same V and J segments, same CDR3 length, and more than 90% similarity) in sequence logo pictures. The number of mice containing the canonical CDR3 sequence or the clonotype consensus is indicated in parentheses following each sequence.

Extended Data Fig. 5 |. FACS results of PP GC B cells and recurrent PP GC IgH CDR3s in different gnotobiotic mouse categories.

a, FACS plots showing the representative proportion of GC (GL7+CD38-) cells among B220+ cells in PPs. The FACS analysis was performed for 9 SPF mice, 10×3 GF mice, 5 GF+SFB mice, 10 GF+SPF mice, 15 AID−/− SPF mice, and representative results are shown. b, Plot comparing %GC of PP B220+ cells from different mouse categories. c, Plot comparing the number of sorted PP GC B cells in each mouse from different mouse categories. Data is presented as mean ± SEM from 9 SPF mice, 10×3 GF mice, 5 GF+SFB mice, 10 GF+SPF mice, 15 AID−/−SPF mice. P-values are calculated by two-sided Mann-Whitney U test. d, f, h, Representative junctional structure of common IgH clonotypes aligned with germline V_H, D and J_H for V_H1–47RC and V_H1–12RC in mice of GF, GF+SFB and GF+SPF conditions (d), V_H1–64RC in mice of GF+SFB condition (f) and V_H6–3RC in mice of GF+SPF condition (h). All possible segment assignments by IGoR with probabilities are shown in Supplementary Table 10. For certain clonotypes, the D could not be accurately annotated because of the short, aligned D sequence. Canonical CDR3 sequences are shown in black, amino acid sequences in purple, and the clonotype consensus in sequence logo pictures. The number of mice containing the canonical CDR3 sequence or the clonotype consensus is indicated in the parentheses following each sequence. e and g, GC productive clonotype distribution of the indicated V_H enriched in PP GC B cells vs naïve B cells in GF mice (e) and GF+SPF mice (g), is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_H and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. Note that the V_H5–17 recurrent clonotype in GF mice is different from the one in GF+SPF mice.

Extended Data Fig. 6 |. SHM selection in recurrent PP GC clonotypes.

a-b, V_H SHM profile of indicated PP GC clonotypes and intrinsic pattern, plotted as mutation rate at each nucleotide. Sequences were stratified by overall V_H mutation rate¹⁹ (see Methods). Significance is determined by hierarchical Bayesian modeling with PEP_0.1 (see Methods; *: PEP_0.1<0.05, **: PEP_0.1<0.01, ***: PEP_0.1<0.005). PEP_0.1 value and amino acid changes are denoted for significantly enriched SHMs. Data is presented as mean ± SEM from mice containing indicated clonotype: 11 SPF mice, 9×3 GF mice, 4GF+SFB mice and 8 GF+SPF mice (a), 3 GF+SFB mice and 4 GF+SPF mice (b). c. Microbial glycan microarray reactivities of mutated V_H1–12RC mAb. See full antibody sequences in Supplementary Table 5. Data shown as average fold enrichment over negative-control mAb (anti-human PD1) of 4 technical repeats.

Extended Data Fig. 7 |. Some recurrent PP GC clonotypes do not show selected SHMs.

V_H SHM profile of indicated PP GC clonotypes and intrinsic pattern, plotted as mutation rate at each nucleotide. Sequences were stratified by overall V_H mutation rate¹⁹ (see Methods). Significance is determined by hierarchical Bayesian modeling with PEP_0.1 (see Methods; ns: PEP_0.1>=0.05). Data is presented as mean ± SEM from mice containing indicated clonotype: 7 SPF mice, 10×3 GF mice, 4 GF+SFB mice and 6 GF+SPF mice (a), 4 SPF mice (b), 9×3 GF mice (c).

Extended Data Fig. 8 |. Recurrent PP GC IgH clonotypes are often associated with recurrent IgL clonotypes.

a, V_L repertoire of productive VJ junctions in PP GC vs naïve B cells plotted as mean ± SEM from 9 SPF mice related to Fig. 1. GC productive clonotype distribution of each indicated V_L is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_L and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. P-values for V usage are calculated by two-sided Mann-Whitney U test with FDR correction (*: p < 0.1, **: p < 0.05, ***: p < 0.01). See Supplementary Table 4 for exact p-values. b, Table showing paired IgL clonotype of recurrent PP GC IgH clonotypes, as inferred by Rep-SHM-Seq (See methods) from 9 SPF mice and/or detected by scRNA-seq from 4 SPF mice.

Extended Data Fig. 9 |. Recurrent clonotypes in chronic splenic and PP GCs in AID−/− mice.

a, V_H repertoire of productive V(D)J junctions in splenic GC vs naïve B cells plotted as mean ± SEM from 9 AID−/− SPF mice. b, V_H repertoire of productive V(D)J junctions in PP GC vs naïve B cells plotted as mean ± SEM from 15 AID−/− SPF mice. GC productive clonotype distribution of each indicated V_H is shown in long-tail distribution plot, with the y-axis representing fraction among all productive clonotypes for this V_H and x-axis representing the top clonotypes (rank ordered). The dominant recurrent clonotypes are marked with the number of mice containing this clonotype. The brown color indicates the clonotype found common in the splenic and PP GCs of AID−/− mice. The green color indicates the clonotype found in WT mice of different gnotobiotic categories, related to Fig. 1. P-values for V usage are calculated by two-sided Mann-Whitney U test with FDR correction (*: p < 0.1, **: p < 0.05, ***: p < 0.01). See Supplementary Table 4 for exact p-values. c, Schematic summarizing the main findings of the paper. Junctional biases during V(D)J recombination generates a diverse CDR3 repertoire for naïve B cells in PPs, with a set of CDR3s occurring at higher frequency, from which gut microbial or non-microbial antigens select recurrent IgH clonotypes in multiple mice, mostly with canonical CDR3 sequence and recurrent pairing of IgL. These recurrently selected antibodies contain selected SHMs, suggesting affinity maturation. The asterisk on V_H6–3RC/Vκ2–109 indicates this pairing was picked up from scRNA-seq data to make the antibody in vitro, not confirmed as recurrent pairing by Rep-SHM-Seq. The frequency of the BCRs/SHMs represented in this schematic does not correspond to their real frequency in the naïve or GC repertoire of PPs.

Extended Data Table. 1 |. Table summarizing recurrent PP GC clonotypes found in GF (a), GF+SFB (b) and GF+SPF mice (c), related to Fig. 2.

The J segment used, CDR3 length, canonical CDR3 sequence in nucleotides or amino acids and the number of mice containing the clonotype or canonical CDR3 nucleotide sequence are indicated.

a

clonotype	JH	CDR3 length (nts)	canonical CDR3 nt	canonical CDR3 aa	clonotype in # mice	canonical CDR3 in # mice
VH1-12RC	3	21	GCAAGAGAGGGGTTTGCTTAC	AREGFAY	10	9
VH1-12RC_GF	2	21	GCAAGAGAGGGCTTAGACTAC	AREGLDY	5	\
VH1-47RC	4	36	GCAAGGGGGAGTAACTACGACTATGCTATGGACTAC	ARGSNYDYAMDY	10	2
VH1-47RC_GF1	4	33	GCAAGGGGGAGTAACTACTATGCTATGGACTAC	ARGSNYYAMDY	6	\
VH1-47RC_GF2	2	36	GCAAGGGGGGGTAACTACGTGAACTACTTTGACTAC	ARGGNYVNYFDY	6	\
VH1-72RC_GF1	2	30	GCAAGATCGGACTATGGTAACTTTGACTAC	ARSDYGNFDY	9	\
VH1-72RC_GF2	2	27	GCAAGATCCGACTACTACTTTGACTAC	ARSDYYFDY	4	\
VH5-17RC_GF	3	24	GCAAAACTGGCCTGGTTTGCTTAC	AKLAWFAY	5	a
VH2-9RC_GF1	4	39	GCCAAACATGATGGTAACTACGACTATGCTATGGACTAC	AKHDGNYDYAMDY	3	\
VH2-9RC_GF2	1	42	GCCAAACATGAAGGTAACTTCGGGGACTGGTACTTCGATGTC	AKHEGNFGDWYFDV	3	3

b

clonotype	JH	CDR3 length (nts)	canonical CDR3 nt	canonical CDR3 aa	clonotype in # mice	canonical CDR3 in # mice
VH1 -12RC	3	21	GCAAGAGAGGGGTTTGCTTAC	AREGFAY	4	4
VH1-47RC	4	a6	GCAAGGGGGAGTAACTACGACTATGCTATGGACTAC	ARGSNYDYAMDY	4	3
VH1-64RC	2	36	GCAAGATGTACTACGGTAGTAGCCCCCTTTGACTAC	ARCTTVVAPFDY	3	2

c

clonotype	JH	CDR3 length (nts)	canonical CDR3 nt	canonical CDR3 aa	clonotype in # mice	canonical CDR3 in # mice
VH1-12RC	3	21	GCAAGAGAGGGGTTTGCTTAC	AREGFAY	8	8
VH1-12RC_GF	2	21	GCAAGAGAGGGGTTTGACTAC	AREGFDY	6	4
VH1-47RC	4	a6	GCAAGGGGGAGTAACTACGACTATGCTATGGACTAC	ARGSNYDYAMDY	6	6
VH6-aRC	3	24	ACAGACCCGGCCTGGTTTCCTTAC	TDPAWFPY	6	4
VH1-26RC_GF+SPF	1	24	GCAAGATCTCGGTACTTCGATGTC	ARSRYFDV	5	2
VH5-17RC_GF+SPF	4	30	GCAAGGCCGTATTACTATGCTATGGACTAC	ARPYYYAMDY	4	2
VH1-64RC	2	36	GCAAGGCCGTATTACTATGCTATGGACTAC	ARTTTVVAPFDY	4	\
VH1-31RC_GF+SPF	4	48	GCAAGGGCATATTACTACGGTAGTAGCTACAACTATGCTATGGACTAC	ARAYYYGSSYNYAMDY	2	2