Complex subsets but redundant clonality after B cells egress from spontaneous germinal centers

Affinity matured self-reactive antibodies are found in autoimmune diseases like systemic lupus erythematous. Here, we used fate-mapping reporter mice and single-cell transcriptomics coupled to antibody repertoire analysis to characterize the post-germinal center (GC) B cell compartment in a new mouse model of autoimmunity. Antibody-secreting cells (ASCs) and memory B cells (MemBs) from spontaneous GCs grouped into multiple subclusters. ASCs matured into two terminal clusters, with distinct secretion, antibody repertoire and metabolic profiles. MemBs contained FCRL5+ and CD23+ subsets, with different in vivo localization in the spleen. GC-derived FCRL5+ MemBs share transcriptomic and repertoire properties with atypical B cells found in aging and infection and localize to the marginal zone, suggesting a similar contribution to recall responses. While transcriptomically diverse, ASC and MemB subsets maintained an underlying clonal redundancy. Therefore, self-reactive clones could escape subset-targeting therapy by perpetuation of self-reactivity in distinct subsets.


Introduction
In autoimmune diseases like systemic lupus erythematosus (SLE), pathogenesis is caused by selfreactive antibody-secreting cells (ASCs) and sustained by memory B cells (MemBs). Numerous subsets of these B cells have been described in models of health and disease, but it is unknown whether and how these subsets change and relate to each other in autoimmune disease. Among MemBs, the major known subsets are conventional germinal center (GC)-and non-GC-derived MemBs (Taylor et al., 2012;Viant et al., 2021). In addition, atypical MemBs (AtMemBs) are present in models of acute and chronic infection in mice (Kim et al., 2019;Song et al., 2022), as well as in parasitic infection in humans (Portugal et al., 2017) and are similar to the age-associated B cells first described in ageing and autoimmune mouse models (Phalke and Marrack, 2018). DN2 (IgD−CD27−) cells resemble AtMemB cells and are an expanded subset of B cells in human SLE. They likely have a naive

Fate-mapped B cells are distributed into four main clusters
To track WT B cells as they develop in an autoimmune environment, we used a mixed bone marrow (BM) chimera model (Akama-Garren et al., 2021;Degn et al., 2017;van der Poel et al., 2019). Irradiated host mice received a combination of BM donor cells from WT and 564 Igi mice (Berland et al., 2006), which bear transgenic BCRs with anti-RNP specificities. In this model, WT donor B cells expand, are selected into spontaneous GCs, mature into ASCs, and contribute to the circulation of self-reactive antibodies. To track activated B cells and their derived populations, Aicda-CreERT2-EYFP fate-mapping reporter mice (Dogan et al., 2009) were used as donors for the WT B cell repertoire ( Figure 1A). Fate-mapped WT B cells were purified, sorted, and processed for droplet-based singlecell RNA-seq. Moreover, we compared the response to that of foreign antigen WT chimeras immunized with a primary and secondary dose of haptenated protein.
Using paired single-cell BCR sequencing and repertoire analysis, we observed clones with clear expansion in both autoimmune and immunized chimeras (Figure 1-figure supplement 1B). Indeed, in both conditions, we were able to find clones that had members belonging to all four clusters, exemplified by the phylogenetics of the most expanded clone in one autoimmune chimeric mouse ( Figure 1E). The GC mutational level was in a similar range between conditions, although autoimmune chimeras had more DZ and LZ replacement mutations. MemBs in autoimmune and immunized chimeras had a similar number of nucleotide replacement mutations. However, ASCs accumulated more mutations in immunized chimeras ( Figure 1F). Nevertheless, ASCs in both conditions reached similar maximum levels of mutations (20 nt in autoimmune and 19 nt in immunized chimeras).
Cells from autoimmune chimeras showed more isotype diversification in all compartments ( Figure 1G). Importantly, both preferential and shared clonal usage of heavy chain genes were observed
To investigate the potential targets, we produced six monoclonal antibodies using the VDJ sequences from two autoimmune chimeric mice. The VDJ sequences originated from members of four different clones, including the most expanded clone from one mouse. We tested the reactivity of these antibodies using a native protein HuProt array containing more than 20,000 recombinant proteins. We combined the monoclonal antibodies into two pools (Mab1 and Mab2), each containing two distinct clones. We identified multiple potential binders, as exemplified by the top 30 rank based on fluorescence Z-score for each pool ( Figure 1I). The potential targets include the housekeeping genes ARIH1, ARHGEF39, and DNAJC7, as well as the poorly described C6orf165 (CFAP206). We also observed reactivity for the known autoantigens SCL-70 and Sm/RNP, and the single-stranded binding protein SSBP1.
Thus, using single-cell RNA-seq, we confirmed that in the 564Igi mixed BM chimeric autoimmune model, B cells with a WT BCR repertoire break tolerance, expand in spontaneous GCs and develop into MemBs and ASCs in a seemingly unrestricted manner, much like they do in response to foreign antigens.
The two terminal clusters differed in key aspects. Xbp1 (X-box binding protein 1), a key modulator of the endoplasmic reticulum (ER) stress response and the unfolded protein response and a major antibody-secreting cell (ASC), DZ, LZ, and memory B cell (MemB) clusters among all cells sequenced from each chimera. Each dot represents a mouse. (E) Phylogenetic tree of the most expanded clone in one of the autoimmune chimeras. Clonal members colored by assigned cluster. (F) Accumulated replacement mutations in each cluster by condition. Violin plot with embedded Tukey boxplot. (G) Isotype usage per cluster for each condition. (H) Clonal usage of selected V genes by condition. Each dot represents a mouse in a Tukey boxplot. IGHV 1-75 was only found in two immunized chimeras. (I) Z-score ranking of fluorescent detection for monoclonal pools Mab1 (left) and Mab2 (right) after HuProt microarray binding assay. The monoclonals originate from two autoimmune chimeras. Statistical values correspond to two-tailed Mann-Whitney test (F) and unpaired t-tests (H).
The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Data for Figure 1D.
Source data 2. Data for Figure 1F.
Source data 3. Data for Figure 1H.    regulator of the transition from B cell to ASC, had slightly more expression in ASC_Late_2 than in ASC_Late_1, whereas ASC_Late_1 had Atf6 and Hspa5 as markers ( Figure 2C). These two genes encode Activation transcription factor 6 and Heat Shock Protein Family A (HSP70) Member 5, which are also involved in the regulation of ER stress. These results indicate that terminally differentiated ASCs might use different mechanisms to handle ER stress. Moreover, ASC_Late_2 had more reads mapped to Ig genes than ASC_Late_1 ( Figure 2D). Similar observations regarding differential usage of ER stress modulation pathways have been made using LPS-stimulated B cells (Scharer et al., 2020). However, we are not aware of other reports identifying in vivo genetic and repertoire differences of these terminally differentiated ASCs in spontaneous GCs. Although few surface markers were unique to the two subclusters, higher expression of Ptprc (CD45, B220), Itga4 (CD49d), and Itgb2 (CD18) were observed in ASC_Late_1 cells and Cd28 in ASC_Late_2 ( Figure 2E). Further characterization of the subclusters using gene set activity analysis with AUCell (Aibar et al., 2017) identified similar levels of gene expression for the Myc pathway, as expected, with the exception of ASC_early_2, which was characterized by proliferation markers. Importantly, a higher score signature of oxidative phosphorylation (OXPHOS) in ASC_Late_2 relative to ASC_Late_1 was observed, suggesting specific metabolic requirements ( Figure 2F).
We analyzed the antibody repertoire for the two terminal clusters, taking advantage of the paired Ig heavy and light chain sequence results. In our dataset, ASC_Late_2 consisted mainly of IgM ASCs, regardless of the immune status ( Figure 3A), whereas ASC_Late_1 cells had a highly diverse isotype usage, with IgG2c being the most common in autoimmune mice. ASC_Late_1 cells accumulated more replacement mutations than ASC_Late_2 cells, and more replacement mutations were observed in immunized than in autoimmune chimeras ( Figure 3B). Comparison of the clonal composition of the ASC clusters revealed extensive clonal overlap ( Figure 3C) in both conditions. This indicates that although the terminal states have transcriptomic and repertoire differences, clones have members with a potential of expansion in any cluster. Moreover, analysis of clonal usage of V genes identified multiple instances of redundancy in the ASC_Late clusters, for example IGHV 1-26, 1-69, and 14-4, among others ( Figure 3D).
In summary, we identified multiple ASC clusters as they differentiated in both autoimmune and immunized environments, with pseudotime analysis suggesting two terminal states with divergent transcriptomic and VDJ maturation profiles, as well as potentially distinct capacities for antibody secretion but that nonetheless share or contain similar clonal members.

MemBs are composed of diverse clusters with differences in their transcriptome and repertoire
An important factor in the pathology of autoimmune disease is the self-sustaining chronicity of autoreactive B cells, which suggests the presence of MemBs. However, it is unclear whether the MemB compartment is similar in autoimmune and immunized responses, and little is known about its internal complexity. To address this question, we reclustered B cells identified as MemBs ( Figure 1B), identifying four subclusters ( Figure 4A). Notably, all MemB subclusters were observed for both autoimmune and immunized mice ( Figure 4B). Remarkably, clusters MemB_1 and MemB_3 accumulated more replacement mutations than clusters MemB_2 and MemB_4, though the latter two still contained cells with similar maximum mutations as MemB_1 and MemB_3 ( Figure 4C). Examination of Ig isotype revealed that Cluster MemB_2 was mostly composed of IgM cells in both autoimmune and immunized chimeras. A hallmark of the inflammatory interferon-driven response common in lupus and viral infections is expression of IgG2c. While this isotype was nearly absent among the four subclusters of (E) Normalized gene expression level of surface markers Ptprc, Itgb2, Itga4, and Cd28 in ASC_Late_1 and ASC_Late_2. (F) Single-cell level scoring for MYC and OXPHOS signature profiling of ASC subclusters with AUCell, split by condition. Statistical values correspond to one-way analysis of variance (ANOVA) with Tukey correction for multiple comparisons (F).
The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Data for Figure 2F, antibody-secreting cell (ASC) OXPHOS autoimmune graph.

Figure 2 continued
Autoimmune Immunized Vgene Usage in 2 clones or more in autoimmune mice  MemB in the immunized mice, all four subclusters of MemB in autoimmune chimeras included the inflammatory isotype ( Figure 4D).
Overall, although distinct, the clusters displayed extensive overlap in gene expression (Figure 4figure supplement 1A). Comparing the transcriptomic profiles of the four clusters revealed distinct markers for each cluster ( Figure 4E, Figure 4-figure supplement 1B). MemB_1 cells had the highest expression of Cd83, a common B cell activation marker typically observed in LZ B cells, and of Il4i1, plus they had higher expression of Apex1, Eif5a, Eif4a1, Slc25a5, C1qbp, and Mif. MemB_2 cells had the highest expression of Fcrl5 and Cd72, similar to DN2 cells in humans (Jenks et al., 2018) and atypical memory B cells in mice and humans (Kim et al., 2019), along with higher expression of Zeb2, Apoe, Cd38, Cd81, Itgb1, and Syk, among others. Markers of MemB_3 were S100a10, Vim, Lgals1, Ass1, Itgb7, Sec61b, Anxa2, and Stk38. Interestingly, Vimentin (Vim) is a cytoskeleton component important for the filament reorganization following BCR stimulation (Tsui et al., 2018). Although unique markers for MemB_4 were scarce, Fcer2a (encoding for CD23) and Icosl showed the highest expression. Other MemB_4 markers include Cd55, Il2rg, Ets1, Ltb, Lmo2, and Zfp36.
Altogether, our interpretation is that MemB_1 likely represents recent GC-derived B cells, as it shows the highest single-cell score for the MYC pathway ( Figure 4-figure supplement 1C), MemB_2 corresponds to the AtMemBs described in chronic antigen exposure and aging and is analogous to DN2 cells in humans, MemB_3 corresponds to MemBs in an activated state, as it had the highest score for a signature of genes downregulated by Pten, a major regulator of B cell activation (Figure 4figure supplement 1D), and MemB_4 corresponds to a memory compartment dependent on interacting with T cells for reactivation.
To further understand the relationship between MemB subclusters, we investigated the clones present in the four MemB clusters. Remarkably, we found that in both autoimmune and immunized mice ( Figure 5A), all MemB clusters share clones with one another. This suggests that, regardless of their repertoire origin and specificities, MemBs can acquire any of the different cluster characteristics and that circulation among them might exist. Moreover, as observed in ASCs, regardless of their transcriptomic and repertoire differences, subclusters can have similar V gene usage for IGHV 1-15, 1-42, 1-53, 1-64, 1-75, 3-6, and 9-3, among others ( Figure 5B).
To evaluate the relationship between potential reservoir (memory) and active response (secretion), we looked for clonal intersection between ASCs and MemBs. We found that clones across all the MemB clusters can seed all the different ASC clusters, regardless of the autoimmune or immunized context ( Figure 5C). Cluster MemB_4 showed higher clonal contribution in immunized chimeras than in autoimmune chimeras, and in both conditions, most clones were shared between MemB_3 and all the ASC subclusters.

Validation of the GC origin of the MemB compartment
We generated a new set of autoimmune chimeras (n = 5, Figure 6A), which were maintained on a tamoxifen diet for a period of 8 weeks before removing tamoxifen for 4 weeks before analysis. Among the MemBs (AID-EYFP+ GL7− CD138−), we found that MemB_2 (FCRL5+) and MemB_4 (CD23+) cells were in similar proportions in all mice ( Figure 6B, Figure 6-figure supplement 1A).
Recent studies show that Aicda (AID) expression can precede the commitment and formation of GC B cells at least in immunized mice (Roco et al., 2019). This could suggest that the observations with AID reporter mice could also include non-GC-derived MemBs. Indeed, formation of extrafollicular MemBs cannot be ruled out from the AID-based model (Lee et al., 2011;Toyama et al., 2002).
As an alternative solution for GC fate mapping, we used S1pr2, which is an established fate marker for GC-derived B cells (Shinnakasu et al., 2016). To confirm the observations made with our single-cell dataset and Aicda-CreERT2-EYFP reporter chimeras, a new set of chimeras using the S1pr2-CreERT2-tdTomato reporter system in combination with 564Igi BM were constructed (n = 3, Figure 6C). The autoimmune chimeras, per mouse. Only V genes used in two clones or more are displayed. Statistical values correspond to two-tailed Mann-Whitney test (B).
The online version of this article includes the following source data for figure 3: Source data 1. Data for Figure 3B. donor S1pr2 reporter mice were also crossed with Prdm1-EYFP reporter mice to distinguish GC-derived ASCs.
To gain insight into the spatial distribution of Memory S1pr2tdTomato+ B cells, cryosections of splenic tissue were characterized by fluorescent scanning confocal microscopy. As expected, S1pr2t-dTomato+ cells localized mostly to GCs in the B cell follicles. Interestingly, a substantial number of GC-derived FCRL5+ MemBs localized close to the MZ and particularly near the bridging channels ( Figure 6E), delineated by CD169+ macrophages.
To confirm the MemB FCRL5+ localization, we injected anti-CD45 antibodies intravenously (i.v.) into autoimmune chimeras (S1pr2-CreERT2-tdTomato:564Igi) 5 min before sacrifice ( Figure 6F). This labels cells exposed to blood circulation in the spleen, as those residing in the MZ would be (Cinamon et al., 2008;Song et al., 2022). We observed a significant difference in the i.v. labeling for MemB FCRL5+ in contrast to MemB CD23+ cells, confirming a preferential MZ localization ( Figure 6G, Figure 6-figure supplement 1C). We confirmed the efficiency of MZ preferential labeling using a traditional gating strategy for MZ and follicular B cells in the same mice ( Figure 6-figure supplement 1D). This privileged location would allow MemB FCRL5+ cells to quickly reactivate upon exposure to foreign antigen but also perpetuate a detrimental response when reacting to self-antigens.
Altogether, we found an underlying complexity of MemBs that, in autoimmune and immunized mice, can be further subdivided into groups that share clonality with ASCs and have transcriptomic and VDJ repertoire signatures suggesting distinct roles in the immune response. Indeed, Memb_2 cells are evidence of memory contribution to the MZ and that the GC-derived MemB compartment can contain DN2-like cells and have the capacity to contribute to pathology through the preservation of harmful self-reactive specificities.

Discussion
The study of antibody-driven autoimmunity often relies on the use of mice with knock-in BCRs specific for a self-antigen. This approach has been important in establishing many of the current concepts of B cell autoimmunity such as receptor editing, follicular exclusion, and clonal deletion. However, introducing a BCR with a certain pre-determined affinity can potentially bias the observations of the rules governing the activation of those cells. Moreover, conclusions drawn from monogenic BCR mice are unlikely to accurately predict the response to self-antigen when a full repertoire of cells are competing for self-antigen and T cell help. Self-reactivity exists as a spectrum of affinities controlled by sensitive mechanisms that might have different regulatory properties (Smith et al., 2019;Tan et al., 2019).
The mixed BM chimera model used in our study allows for the pre-defined self-reactive cells to kick-start autoimmunity. In this environment, naive self-reactive WT B cells that normally undergo negative selection can spontaneously become activated and enter self-reactive GCs. Although this approach still depends on a BCR knock-in transgene, it provides a model to study the development of WT B cells as they mature in an autoimmune environment (Degn et al., 2017).
We found that self-reactive WT B cells mature and reach pausi-clonality through GC selection and eventually enter all the major mature B cell compartments, similar to that of the immunized controls. Both differential and overlapping clonal usage of V genes were observed in the autoimmune and Source data 1. Data for Figure 4B.
Source data 2. Data for Figure 4C, by cluster.
Source data 3. Data for Figure 4C, within cluster.   immunized chimeras. For example, among the most used V genes in both conditions were IGHV 1-53. This VH family member appears in many scenarios such as MZ B cells reacting to HIV gp120 immunization (Pujanauski et al., 2013) and self-reactive clones in FcgRIIB-deficient mice (Tiller et al., 2010;van der Poel et al., 2019). Intriguingly, 1-53 is among the most commonly used genes in the mouse response to the hapten group NP (Xue et al., 2019), most likely due to its similarity to IGHV 1-72. Thus, IGHV 1-53 can be considered part of the collection of baseline repertoire that gives rise to 'polyreactive' antibodies. Although polyreactive antibodies may be detrimental in autoimmunity, they are likely retained in the repertoire for use when a quick response to a foreign antigen is needed. Notably, mouse IGHV 1-53 is the mouse ortholog to human IGHV 1-69 (Ighv1-53, 2023). IGHV 1-69 is commonly observed in broadly neutralizing antibodies in the anti-viral response, but together with IGHV 4-34, it is also a feature of antibodies in autoimmune disease (Bashford-Rogers et al., 2018;Watson and Breden, 2012). Additionally, the ortholog to IGHV 4-34 in mice is IGHV 3-6 (Ighv3-6, 2023), which was the most clonally used gene in our autoimmune chimeras. This suggests that selfreactivity can be driven by predominant V genes that might initially be polyreactive but are further tailored in GCs. Altogether, these data suggest that there might be different ways to generate selfreactive pathogenic clones, starting with common V genes but leading to the usage of more specific ones. Distinguishing the contributions of originally polyreactive vs non-polyreactive clones will allow for better understanding of the mechanisms that lead to epitope spreading.

ASC and MemB intersect in immunized mice
Many characteristics have been attributed to ASCs: short-vs long-lived, plasmablasts vs plasma cells, and EF-vs GC-derived. We observed two terminal clusters of ASC development with clear distinctions in the number of Ig transcripts, metabolic requirements and ER-stress-related gene expression. ASC_ Late_1 had lower Ig counts and score signature for oxidative phosphorylation, together suggesting a reduced rate of antibody secretion (Lam et al., 2018a). Intriguingly, ASC_Late_1 had more replacement mutations than ASC_Late_2 and a more diversified usage of isotype. Given that MemBs have accumulated mutations over-time in the GC and that our sequencing dataset was processed ~5 weeks after tamoxifen treatment while the lifespan of short-lived ASCs is estimated at ~2 weeks, reactivated MemBs are the likely origin of most ASC_Late_1 cells. Although pseudotime analysis leads us to classify ASC_Late_2 as one of the two terminal clusters, the immediately preceding cluster ASC_Mid_2 also possesses characteristics of mature ASCs as well, including prominent expression of Cxcr4 and Slpi, which is a marker for BM and long-lived ASCs (Lam and Bhattacharya, 2018b), suggesting it could represent another terminal state.
We identified four MemB subclusters. MemB_1 has characteristics of GC B cells and likely represents recently egressed cells. MemB_2 has similar markers as atMemB and DN2 cells. MemB_3 shares similar characteristics to recently activated B cells. MemB_4 shows patterns of interaction with T cells (and resembles DN4 in humans). We suspect that the different properties of MemB_2 and MemB_4 exist to allow the B cell compartment to respond adequately in a diversity of contexts. Indeed, we observed distribution of MemB_2 GC-derived cells in the MZ and in bridging channels -convenient locations for immediate antigen surveillance and response. MemB_4 might represent T-dependent bona fide long-lived MemBs, maintaining a record of previous exposure and requiring involvement of T cells for retriggering.
Finally, we found that, to varying degrees, all subclusters of ASCs and MemBs shared members of the same clones. MemB_3 showed striking predominance in clonal overlap with ASCs of all subclusters, indicating this cluster is an activated version of all the other MemB subpopulations. The clonal overlap between all these subclusters was surprising, although not completely unexpected after finding overlapping clonality among ASCs and MemBs when analyzed separately. This overlap likely chimeras by flow cytometry (n = 3 chimeras). (E) Splenic localization of FCRL5+ S1PR2(Tomato)+ MemB cells by confocal microscopy. Overview of a spleen from autoimmune chimeras ( Figure 3C), left, and selected area, center, delineates a bridging channel for a close-up examination of FCRL5+ S1pr2tomato+, right. Arrow points to FCRL5+ MemBs. Scale bars represent 100 µm (left), 50 µm (center), and 20 µm (right). Arrow heads point to FCRL5+ S1PR2(Tomato)+ cells. (F) Experimental design for in vivo marginal zone labeling. S1pr2-CreERT2-tdTomato:564Igi chimeras intravenously (i.v.) injected with 5 µg of anti-CD45-APC for 5 min before organ extraction. (G) Flow cytometry gate strategy to evaluate i.v. CD45 labeling between MemB FCRL5+ and MemB CD23+ cells. Statistical values correspond to one-tailed unpaired Student t-test (G).
The online version of this article includes the following figure supplement(s) for figure 6:  provides a wider safety-net, an insurance of long-lived memory and a readiness to reactivate to deal with prospective pathogens. On the other hand, this implicates that the efforts to target any one cellular compartment for the sake of therapeutically ablating autoimmunity may overlook the presence of shared self-antigen specificity across different cellular niches.
Overall, using a WT BCR repertoire mouse model that maps spontaneous GC-derived cells, we observed a diversity of subgroups, reflecting the inner complexities of the ASC and MemB compartments, with specific transcriptomic and repertoire characteristics, but with underlying redundant clonality.

Study design
The purpose of this study was to characterize the post-GC populations of antibody-secreting and MemBs in the context of autoimmunity. We used GC fate-mapping and single-cell transcriptomics coupled to BCR repertoire analysis. The BM chimeric mouse model we used allows to track WT B cells as they develop and exit from spontaneous GCs. We contrasted the autoimmune chimeras with NP-OVA immunized chimeras.

Tissue processing
Mice were euthanized by cervical dislocation under isoflurane induced anesthesia. Spleens were extracted, dissected, and processed for immunofluorescence microscopy or flow cytometry analysis and cell sorting.

BM chimeras
BM chimeras were prepared using marrow from 564Igi mice and BCR WT donor as previously described (Degn et al., 2017). Mice were lethally irradiated at 1100 rad and kept on antibiotics (sulfamethoxazole/trimethoprim) through drinking water for 7 days after irradiation. Femurs and tibia from donor mice were cleaned from muscle tissue and subsequently rinsed with cell transfer buffer (HBSS supplemented with 10 mM N-2-hydroxyethylpiperazine-N-2-ethane sulfonic acid (HEPES), 1 mM Ethylenediaminetetraacetic acid (EDTA), and 2% heat inactivated fetal bovine serum). Marrow was extracted from bones by crushing them using mortar and pestle and the detached cells were resuspended in cell transfer buffer and passed through a 70-mM sterile filter. Cells were counted by using an erythrocyte lysed aliquot. All autoimmune chimeras were prepared at 2:1 ratio for 564Igi:WTreporter, with WT-reporter and irradiated hosts genotype as specified for each experiment. BM recipients received 15-20 × 10 6 cells i.v. in 100 ml cell transfer buffer by retroorbital i.v. injection approximately 8 hr post irradiation.

Fate-mapping tamoxifen induction
Mice were exposed to tamoxifen in two different ways, as specified in each experimental design figure. Mice were gavaged with 10 mg of tamoxifen (Sigma) dissolved in Corn Oil at 50 mg/ml twice, at days 4 and 7 post primary immunization in the case of immunized chimeras and at the same time for autoimmune chimeras. For validation experiments, BM chimeric mice were maintained in a tamoxifen enriched diet for the specified timeframes (Envigo) (Song et al., 2022).
Cell sorting for single-cell sequencing Spleens from autoimmune and immunized chimeras were processed following viability recommendations for droplet-based Chromium single-cell RNA-seq gene expression (10×). The full organs were dissociated in MACS buffer (PBS 1×, 0.5% BSA, 2 nM EDTA) using syringe plungers and 70 µm cell strainers, spun down at 1000 rpm for 5 min, resuspended in RBC lysis buffer and incubated on ice for 5 min. Samples were washed with MACS buffer and spun down 1000 rpm for min. Cells were enriched by negative selection with a Pan II B cell enrichment kit (Miltenyi) according to the provider specifications. During enrichment incubation cells were also stained for flow sorting. The antibodies used for staining were anti-GL7-PacBlue, ant-CD138-PE, anti-B220-PerCP/Cy5.5, anti-CD38-PECy7, and anti-CD45.1-APC from Biolegend. eFluor780 from Thermo Fisher was used for viability. After enrichment and antibody staining cells were resuspended in resuspension buffer (PBS 1×, 0.04% BSA). Reporter cells were sorted based on EYFP expression (Efluor 780-EYFP+ CD45.1+) at two-way purity sort with a FACSARIA II Special Order system (BD Biosciences) with 355, 405, 488, 640, and 592 nm lasers. Sorted cells were spun down and resuspended in resuspension buffer prior to singlecell encapsulation.

Droplet-based single-cell sequencing and data processing
The sorted cells were encapsulated with barcoded hydrogels using the Chromium system for Single Cell Immune Profiling, that allows coupling of transcriptomic and VDJ information per cell. cDNA libraries were prepared according to the manufacturer's recommendations. Library quality control and sequencing (NextSeq 500, Illumina) were performed by the HMS Biopolymers Facility. Cellranger (5.0.1) was used to generate the count matrices for gene expression and the VDJ contigs per cell using the multi function for each mouse. Reads were aligned to a custom mm10 reference genome incorporating the transcript sequences for EYFP and Cre Recombinase. R (4.1.2) was used for further processing the Cellranger gene expression count matrix output using the OSCA workflow (https:// bioconductor.org/books/release/OSCA/) as template and using the Single Cell Experiment (SCE) format. Scater and Scran were used for data QC. Cells were subset for less than 5% mitochondrial and 40% ribosomal content. Ig genes were excluded from clustering and posterior expression analysis. Correction was done with fastMNN function from batchelor. Clustering was performed using the Leiden algorithm. Markers for each cluster were identified using the scoreMarkers function from Scran. Pre-defined genesets were used for Oxidative Phosphorylation (GO:0006119), Myc upregulated (Wang et al., 2020;Chen et al., 2021) and PTEN_DN.V1_DN (GSEA) were used for signature analysis with AUCell. Data visualization was done with dittoSeq.

BCR repertoire analysis
The VDJ output from Cellranger multifunction was further processed following the Immcantation (https://immcantation.readthedocs.io/en/stable/) recommendations for 10× derived single-cell data. Genes were assigned using IgBlast and the IMGT reference sequence database. Clones were set using the DefineClones function from Change-O, with a 0.1691791 threshold defined using the findt-Threshold function from Shazam. countClones (Alakazam) and observedMutations (Shazam) functions were used for quantification.
Phylogenetic trees were generated with GCTree (DeWitt et al., 2018) using all VDJ sequences from a given clone and rooted in the germline obtained with CreateGermlines function from Change-O.
For target detection, the monoclonal antibodies were pooled in two different samples and processed for reactivity by CDI Labs with a HuProt v4.0 array, containing 21,000 human proteins. Monoclonal antibodies at 1 µg/ml were diluted in a final volume of 3 ml and were probed with the arrays for 2 hr at room temperature (RT). The arrays were washed according to the company protocol and were probed with the secondary antibody (Alexa-647-goat-anti-mouse IgG gamma-specific) under conditions optimized by CDI Labs for signal detection with GenePix software. Data were processed with CDI's proprietary data analysis software (Z-score analysis).

Immunofluorescence and confocal microscopy
Following 4% Paraformaldehyde (PFA) fixation on ice for 3 hr, spleens were embedded in OCT (Fisher Healthcare), frozen in dry ice and stored at −80°C. Spleens were cut into 10 µm sections, blocked for 1 hr at RT with Blocking buffer (PBS, 0.01% Tween20, 2% BSA, and 5% Fetal Bovine Serum (FBS)) and stained O.N. at 4°C with antibodies diluted in Blocking buffer. Anti-CD169-BV510 (Biolegend) and anti-FCRL5-APC (biotechne) were used for staining. Slides were washed three times in PBS, 0.01% Tween20 and mounted with FluoroGel (Electron Microscopy Sciences) prior to image acquisition. Images were acquired with an OLYMPUS FV3000R resonant scanning confocal microscope equipped with four laser lines (405, 488, 514, and 633 nm), hybrid galvo and fast resonant scanning capabilities, ultra-sensitive GaAsP detectors with full spectral imaging and motorized XYZ stage for tiling. The images were processed in Fiji.

Statistical analysis
Two-tailed Mann-Whitney, one-way analysis of variance with Tukey correction and one-tailed unpaired Student t-tests were performed with Prism 9 (GraphPad).
with the guidelines of the Laboratory Animal Center of National Institutes of Health. The Institutional Animal Care and Use Committee of Harvard Medical School approved all animal protocols (protocol number IS111).

Additional files
Supplementary files • Supplementary file 1. Clonal overlap between and among antibody-secreting cell (ASC) and memory B cell (MemB) subsets. This file contains the raw values for the circos plot graphs showing clonal intersects between subsets for Figures 3C and 5A, C.

Data availability
The sequencing data presented in this study have been submitted to the Gene Expression Omnibus under accession number GSE203132. Code is available on GitHub (copy archived at Castrillon, 2023).
The following dataset was generated: