Chronic viral infections, such as human immunodeficiency virus and hepatitis C, are long-standing global health burdens that have evaded control by an antibody-mediated vaccine. The inability to clear virus from the body can cause irreparable and potentially fatal organ damage if left untreated. SARS-CoV-2 is the newest virus with the potential to cause a persisting infection1, with severe disease correlating to dysregulation in antibody responses2. Efficiently treating patients who battle to clear the infection has proven problematic. Thus, it is critical to understand why the immune system fails to produce effective adaptive responses to chronic virus. While progress in this area has focused on T cell dysfunction, the mechanistic changes that lead to sustained B cell dysfunction have remained elusive.

Antibody production is a critical component of effective antiviral immune responses. During a T cell-dependent response, antigen-activated B cells form an early wave of antibody-secreting cells (ASCs), followed by a germinal center (GC) reaction within which B cells undergo affinity maturation. The generation of a high-affinity repertoire is a finely regulated process driven by transcription factors. Among them, c-Myc is transiently expressed in the GC light zone (LZ) and is responsible for selection of antigen-specific clones for further expansion in the dark zone (DZ)3,4,5, followed by differentiation into high-affinity ASCs and memory B cells. Antibodies help clear infections through several key mechanisms: direct neutralization, non-neutralizing functions and mediating affinity maturation of GC B cells. However, antibodies can be a double-edged sword during chronic infection. Ineffective humoral responses are underpinned by three major alterations: polyclonal B cell expansion6,7,8, altered B cell memory9 and changes to the quantity and quality of antibody. Excessive IgG in the serum leads to an overload of immune complexes6 (ICs) that suppress antibody-mediated innate effector functions and counterintuitively limits viral clearance10,11. IC deposition during chronic infection can cause destructive tissue pathology in both patients and in mice12,13. However, quantity is not the sole consideration for understanding how antibodies contribute to protection versus pathogenicity, as it does not necessarily correlate with antibody efficacy10,14. Efficacy arises from neutralizing capacity and Fc-mediated functions, both critical for clearing viral infection15,16,17,18. Despite the evidence of humoral dysregulation during chronic infection, few therapeutic options exist for modulating humoral responses to recover functionality and drive viral clearance in chronic infection.

Epigenetic modifiers, such as chromatin-remodeling histone modifiers and DNA methyltransferases, are vital regulators underpinning B cell fate19,20. During chronic viral infection, histone modifiers regulate functional differences between exhausted CD8+ T cells and effector memory subsets21,22. Epigenetic regulators have been shown to dysregulate B cells in lymphomas23,24 and in autoimmune conditions25. Yet, the B cell-intrinsic roles of epigenetic regulators in humoral dysfunction during chronic infection are undetermined. We hypothesized that the chronic inflammatory environment generated by persistent infections could perturb the expression and activity of epigenetic modifiers in the GC during the production of high-affinity ASCs. Using a combination of deep-sequencing, genetic tools and small molecule inhibitors, we demonstrate that a dysfunctional B cell fate in viral infection is dictated by the histone modifier BMI-1.


BMI-1 is increased in B cells during chronic viral infection

To elucidate how chronic viral infection disrupts B cell differentiation and identify key intrinsic determinants of dysfunction, we used the comparative lymphocytic choriomeningitis virus (LCMV) model. C57BL/6 mice were infected with either LCMV-WE or LCMV-Docile, two closely related strains of LCMV, to induce either an acute (WE) or chronic (Docile) infection. Using this model, we investigated whether chronic infection resulted in changes to the epigenomic landscape within responding B cells. GC B cells (CD19+IgDloCD95+CD38loCD138) were isolated 14 d after infection and both assay for transposase-accessible chromatin using sequencing (ATAC-seq) and RNA-seq was undertaken (Fig. 1a). ATAC-seq was used to determine regions of chromatin accessibility within the genome that may be altered during chronic infection and correlated to changes in gene expression as determined by RNA-seq. Principal-component analysis (PCA) showed that samples within the same group (WE or Docile) clustered together. Overall, 889 differentially accessible regions (DARs) were identified between acute and chronic GC B cells, with an overall increase (561 DARs) in accessibility in chronic infection (Fig. 1b,c and Extended Data Fig. 1a–c). The frequency of DARs in the promoter-transcription start site region was 11%, with the majority of DARs within the intergenic and intron regions (Extended Data Fig. 1b). In accordance with increased chromatin accessibility, RNA-seq identified 545 differentially expressed genes (DEGs), with 505 genes upregulated in chronic GC B cells compared to acute GC B cells (Fig. 1d and Extended Data Fig. 1d–h; select ASCs and interferon (IFN)-regulated genes indicated). To determine which of these DEGs may be directly regulated by changes to chromatin accessibility at their loci, overlap analysis of the datasets was performed. A total of 39 DEGs in GC B cells responding to LCMV-WE or LCMV-Docile had corresponding DARs identified by ATAC-seq (Extended Data Fig. 1i,j; several different DARs were identified for individual genes). Ingenuity Pathway Analysis (IPA) was then utilized to gain further insight into the gene networks altered in chronic infection. In accordance with an increase of ASCs in chronic infection, IPA revealed differentiation and quantity of plasma cells, as well as GC B cells, in the top ten disease and biological functions altered across the two conditions (Fig. 1e,f). This was despite excluding CD138+ cells from the sort-purified GC B cells used for ATAC-seq and RNA-seq analyses.

Fig. 1: Gene expression and chromatin accessibility regions in GC B cells during chronic viral infection are distinct from acute viral infection.
figure 1

a, Schematic of experimental setup. GC B cells were sort-purified from wild-type mice infected with either LCMV-WE or LCMV-Docile 14 d previously. b, PCA in two dimensions of DARs in acute (salmon) and chronic (green) GC B cells. c, Heat map of ATAC-seq signals for unique and shared peak groups; n = 4 mice per group. d, Heat map of gene expression by RNA-seq for differential expressed genes in the two clusters; n = 2 mice per group. e, Bar charts of significantly enriched top ten list of IPA-based diseases and bio function in chronic GC B cells. Each bar represents a high-level functional category as calculated from the Fisher’s exact test (P ≤ 0.05). f, IPA analysis: upregulated or downregulated genes predicted to affect the differentiation of plasma cells. g, Accessibility changes in PRC1 targets within DARs (ATAC-seq; reads per million (RPM); top) and PRC1 targets within the DEG dataset (RNA-seq; counts per million (CPM); bottom); ***P = 0.0001, ****P < 0.0001 (top); *P = 0.0312, ****P < 0.0001 (Wilcoxon matched-pairs signed-rank test, two-tailed P value) (bottom). h, Plots of normalized counts for canonical PRC1 genes (RNA-seq); n = 2 mice per group, data represent mean ± s.e.m. i, Schematic of infection and tamoxifen administration of Bmi1CreERT2Rosa26EYFP reporter mice. Wild-type (WT) and Bmi1CreERT2Rosa26EYFP reporter mice were infected with either LCMV-WE or LCMV-Docile. Tamoxifen was administered every 24 h from d9–d11 and EYFP+ cells were assessed in B cell subsets at d14. j, Flow cytometric representative histograms of EYFP+ cells within the lymphocyte population. k, EYFP+ lymphocytes (frequency of live cells); n = 3 WT (WE or Docile), n = 9 reporter (WE) and n = 7 reporter (Docile), combined from three independent experiments. l, Frequency of EYFP+ GC B cells; EYFP+ B220loCD138hi ASC; n = 3 WT mice (WE or Docile), n = 9 reporter mice (WE) and n = 7 reporter mice (Docile), combined from three independent experiments. m, Uninfected reporter mice received three doses of tamoxifen, 24 h apart and assessed 48 h after the final dose. Frequency of B220loCD138hi ASCs in mesenteric lymph nodes (mLNs), spleen and bone marrow (BM); n = 8, combined from two individual experiments. Data represent mean ± s.e.m. (km); *P < 0.05, **P < 0.01, ***P < 0.001 (Mann–Whitney U-test, two-tailed P value).

Source data

The increase in chromatin accessibility regions led us to hypothesize that epigenetic modifiers that repress gene expression may be dysregulated during chronic viral infection. Two important complexes that are known regulators of B cell fate are the Polycomb repressive complexes (PRCs). PRC1 and PRC2 consist of several subunits with the ability to support and deposit H2AK119Ub1 and H3K27me3, respectively, regulating expression of target genes required for formation and maintenance of GCs23,24,26. We observed 49 PRC1 targets (Fig. 1g) and 80 PRC2 targets (Extended Data Fig. 1k) with altered chromatin accessibility in chronic GC B cells. In our DEG dataset, 63 PRC1 targets (Fig. 1g) and 74 PRC2 targets were differentially expressed (Extended Data Fig. 1k). Among the overlapped targets from the two datasets, the analysis identified seven altered PRC1 targets (Gfi1, Gnaz, Ifi27, Oas3, Sclo4a1, Stat1 and Tmem184b) and eight altered PRC2 targets (Evc, Gnaz, Max, Mx1, Oas3, Plk2, Stat1 and Xaf1).

We next assessed whether expression of PRC components themselves was altered in chronic GC B cells. Expression of components of the canonical PRC1 (Fig. 1h), non-canonical PRC1, canonical PRC2 or PRC2 co-factors (Extended Data Fig. 1l) were mostly unchanged in chronic infection, with Ring1b, Suz12, Rbbp4 and Bcor decreased, whereas Bmi1 was increased compared to acute GC B cells. The increase in Bmi1 was a surprising finding. Previous reports have not observed BMI-1 in murine GC B cells26, although it has been detected in the LZ of GC in human tonsils27 and is expressed in human ASCs28. BMI-1 is a Polycomb group protein that activates the catalytic component RING1B of PRC1 (ref. 29), which in turn deposits monoubiquitin at lysine 119 of histone H2A30. Functionally, BMI-1 allows stable silencing of gene expression, regulating lineage commitment, self-renewal, proliferation and survival31,32,33,34,35,36. BMI-1 was originally identified as an oncogenic collaborator of c-Myc37,38 that is upregulated in Myc-induced B cell malignancies39 and myeloma34,40; however, its role during humoral responses to infection has not been determined.

We proposed that Bmi1 upregulation in chronic GC B cells may be connected to the increase in plasma cell gene networks. To understand murine Bmi1 expression dynamics, we used a tamoxifen-inducible Bmi1CreERT2Rosa26EYFP reporter mouse to temporally examine Bmi1 expression on a per-cell basis. We first assessed which B cell subsets express BMI-1 in a tractable, non-infectious model, the hapten carrier 4-hydroxy-3-nitrophenylacetyl-keyhole limpet hemocyanin (NP-KLH). Bmi1CreERT2Rosa26EYFP mice were immunized with NP-KLH precipitated in the adjuvant alum, administered with tamoxifen at d3–5 and assessed at d7 to detect the yellow fluorescent protein (YFP) expressed under the control of Rosa26 regulatory sequences in BMI-1+ cells (Extended Data Fig. 2a). Flow cytometric analyses revealed that B220loCD138hi ASCs upregulated Bmi1 (Extended Data Fig. 2b). We undertook further analysis to determine whether BMI-1-expressing cells were enriched in a particular subset of ASCs. Enhanced YFP (EYFP) expression was modulated in the IgA+ and IgG1+ populations in secondary lymphoid tissues (Extended Data Fig. 2c) and EYFP+ was significantly enriched in plasma cells compared to plasmablasts (Extended Data Fig. 2d).

Bmi1CreERT2Rosa26EYFP reporter mice were infected with LCMV-WE or LCMV-Docile and tamoxifen was administered every 24 h from d9–11 (Fig. 1i). Infection with LCMV-Docile induced a significant increase in the frequency of EYFP+ lymphocytes, compared to LCMV-WE (Fig. 1j,k). In both GC B cells and in B220loCD138hi ASCs, BMI-1 was increased ~threefold in chronically infected mice, compared to acutely infected mice, with a higher proportion within the ASC population (Fig. 1l). Within the GC population, the increase in EYFP+ cells in LCMV-Docile-infected mice was localized to the LZ population, whereas DZ cells had an equivalent frequency of EYFP+ cells following acute or chronic infection (Extended Data Fig. 2e). Furthermore, when steady-state ASCs were assessed for EYFP expression, EYFP+ cells were significantly increased in the bone marrow compared to the spleen and mesenteric lymph nodes (Fig. 1m). Thus, Bmi1 expression was enriched in the ASC population compared to GC B cells, correlating with an increase in tissues with higher frequencies of long-lived plasma cells. The increase in Bmi1 transcript in EYFP+ subsets, as well as increased Bmi1 transcript and BMI-1 protein in B cell subsets responding to LCMV-Docile compared to LCMV-WE from wild-type mice, was confirmed by quantitative PCR with reverse transcription (RT–qPCR) and immunoblot analyses (Extended Data Fig. 2f–k). Combined, BMI-1 is conventionally enriched in ASCs and differentially expressed in GC B cells responding to either acute or chronic strains of LCMV.

Bmi1 deletion in B cells accelerates LCMV-Docile clearance

We next asked whether modulation of BMI-1 in B cells in chronic infection determined productive outcomes to viral infection. To delete Bmi1 in mature B cells, we generated Bmi1f/fCd23Cre/+ mice by crossing mice that carry floxed alleles of Bmi1 with mice expressing a Cre recombinase under the control of the B cell-specific Cd23 promoter. Bmi1f/fCd23Cre/+ and control Cd23Cre/+ mice were infected with LCMV-Docile and assessed throughout infection. Notably, Bmi1f/fCd23Cre/+ mice had significantly reduced viral titers (fivefold change) compared to controls at d28 after infection (Fig. 2a). Time-course analyses demonstrated that viral titers were similar between control and BMI-1-deficient mice at d7; however, a 2.4-fold decrease was detected by d14 (Fig. 2b), concomitant with a significant increase in weight gain compared to Cd23Cre/+ mice (Extended Data Fig. 3a). At d14 after infection, spleens from Bmi1f/fCd23Cre/+ mice were noticeably smaller than control mice (Fig. 2c) and splenocyte numbers were reduced (Fig. 2d). Frequencies of innate, CD4+ and CD8+ T cell populations were comparable between Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice (Extended Data Fig. 3b–f). Functionally, there was no difference in killing capacity of gp33+CD8+ T cells (Extended Data Fig. 3g) or production of IFN-γ on a per-cell basis (Extended Data Fig. 3h). Infection with LCMV disrupts splenic architecture41,42,43, particularly B cell follicles and T cell zones (Fig. 2e; Cd23Cre/+ mice; spleens from uninfected mice are shown in Extended Data Fig. 3i). In contrast, spleens from Bmi1f/fCd23Cre/+ mice displayed intact follicular structures and T cell areas (Fig. 2e; Bmi1f/fCd23Cre/+). Thus, depletion of BMI-1 in mature B cells resulted in normalized splenic size and architecture and accelerated viral clearance.

Fig. 2: Conditional deletion of Bmi1 in B cells results in accelerated clearance of chronic viral infection.
figure 2

a, Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice were infected with LCMV-Docile and livers were assessed for viral plaque-forming units (p.f.u.) at d28, n = 10 Cd23Cre/+ and n = 11 Bmi1f/fCd23Cre/+mice, combined from three independent experiments. Data represent mean ± s.e.m. ***P = 0.000264 (Mann–Whitney U-test, two-tailed P value). b, Time course of viral p.f.u. of the indicated time points after infection (d7, n = 3; d14 and d28, n = 10 Cd23Cre/+ and n = 11 Bmi1f/fCd23Cre/+ mice, combined from 1–3 experiments per time point). Data represent mean ± s.e.m. d7, P = 0.234; d14, **P = 0.0095; d28, ***P = 0.000264 (Mann–Whitney U-test, two-tailed P value). c, Representative spleens from Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice infected with LCMV-Docile at d14 after infection. d, Splenocyte number at the indicated time points after infection; d7, n = 9 mice per group; d14, n = 15 Cd23Cre/+ and n = 18 Bmi1f/fCd23Cre/+ mice; d28, n = 11 Cd23Cre/+ and n = 12 Bmi1f/fCd23Cre/+ mice, combined from at least two individual experiments per time point. Data represent mean ± s.e.m. d7, *P = 0.0244; d14, ***P = 0.001; d28, *P = 0.0214 (Mann–Whitney U-test, two-tailed P value). e, Histological analysis of spleens from Cd23Cre/+ and Bmi1f/fCd23Cre/+ at d7 and d21 after infection; B220 (cyan), CD3 (magenta). Scale bar, 100 μm. d7, n = 3 Cd23Cre/+ spleens and n = 4 Bmi1f/fCd23Cre/+ spleens; d21, n = 3 spleens per group. One experiment per time point.

Source data

Increase in antibody function in the absence of BMI-1

Changes to antibody quality and quantity contribute to ineffective humoral responses during chronic viral infection. Given that B cell-specific deletion of BMI-1 led to accelerated viral clearance, we hypothesized that the efficacy of antibody, either through neutralizing or non-neutralizing effector function, was increased in Bmi1f/fCd23Cre/+ mice. To test this, LCMV-Docile-infected Bmi1f/fCd23Cre/+ and Cd23Cre/+ mice were assessed for neutralizing antibody capacity. Neutralizing antibodies formed to chronic LCMV emerge late in the response, increasing approximately 2 months after infection44. By contrast, antibodies from Bmi1f/fCd23Cre/+ mice had an increased capacity to reduce plaque formation in vitro at d14, compared to antibodies from control mice (Fig. 3a).

Fig. 3: Pleiotropic function of BMI-1 in the humoral response to chronic viral infection.
figure 3

a, Cd23Cre/+ or Bmi1f/fCd23Cre/+ mice were infected with LCMV-Docile. Assessment of the ability of serum at d14 post-infection to reduce viral infectivity in vitro (no serum, no s.); n = 6 Cd23Cre/+ and n = 9 Bmi1f/fCd23Cre/+ mice, combined from three independent experiments. Data represent mean ± s.e.m. *P < 0.05 (Mann–Whitney U-test, two-tailed P value). b, Purified IgG from infected Cd23Cre/+ or Bmi1f/fCd23Cre/+ mice was assessed for N-glycan structures by capillary electrophoresis. Relative abundance of specific types of N-glycan structures of total IgG at d14 after infection (G2, digalactosylated; F, fucosylated; Z1, sialylated); n = 5 mice per group, data combined from two independent experiments. Data represent mean ± s.e.m. **P = 0.0079 (Mann–Whitney U-test, two-tailed P value). c, ADCC assay used to determine frequency of dead target MC57G cells after incubation with WT NK cells and sera from d14 post-LCMV-Docile-infected mice, assessed by flow cytometry; n = 7 mice per group, data combined from two independent experiments. Data represent mean ± s.e.m. **P = 0.0023 (Mann–Whitney U-test, two-tailed P value). d, Representative flow cytometric plot of B220loCD138hi ASC in LCMV-Docile-infected mice. e,f, Frequency and number of ASCs in mice infected with either LCMV-WE or LCMV-Docile at d14 (WE, n = 5 Cd23Cre/+ and n = 6 Bmi1f/fCd23Cre/+ mice; Docile, n = 14 Cd23Cre/+ and n = 15 Bmi1f/fCd23Cre/+ mice) (e) and in LCMV-Docile-infected mice at indicated time points (d7, n = 6 mice per group; d14, n = 14 Cd23Cre/+ and n = 15 Bmi1f/fCd23Cre/+ mice; d21, n = 6 mice per group; d28, n = 8 Cd23Cre/+ and n = 9 Bmi1f/fCd23Cre/+ mice, combined from at least two experiments per time point) (f). Data represent mean ± s.e.m. Frequency of live cells: d7, **P = 0.0022; d14, *P = 0.0102; d21, *P = 0.0411; d28, P = 0.8884 (Mann–Whitney U-test, two-tailed P value); cell number: d7, P = 0.8182; d14, **P = 0.0012; d21, *P = 0.026; d28, P = 0.4907 (Mann–Whitney U-test, two-tailed P value). g, Frequency and number of CD95hiCD38loIgDlo GC B cells at d14. WE, n = 5 Cd23Cre/+ and n = 6 Bmi1f/fCd23Cre/+ mice; Docile: n = 14 Cd23Cre/+ and n = 15 Bmi1f/fCd23Cre/+ mice. Data were combined from at least two experiments and represent mean ± s.e.m. Frequency of live cells: *P = 0.0194, **P = 0.0015, ***P = 0.0005; cell number: *P = 0.0418, **P = 0.0024 (Mann–Whitney U-test, two-tailed P value). h, Assessment of total IgG2c immune complexes at d14 after infection; n = 2 mice per group, data are representative of three independent experiments and represent mean ± s.e.m. (Mann–Whitney U-test, two-tailed P value). i,j, BM of Ly5.1 and Bmi1f/fCd23Cre/+ were mixed in a 1:1 ratio and used to reconstitute irradiated recipients; mice were infected and assessed for ASCs as per schematic; n = 9 per group, data were combined from two independent experiments and represent mean ± s.e.m. ***P = 0.002 (Mann–Whitney U-test, two-tailed P value).

Source data

The tuning of antibody quality occurs by glycosylation of a conserved N-linked glycan of the IgG constant region (Fc), which can be dynamically modulated by the pathogen-influenced microenvironment during an immune response45,46. Shifts in glycosylation can have antiviral or anti-inflammatory effects, regulate complement activation and enhance the effectiveness of neutralizing antibodies16,47. Therefore, to elucidate the efficacy of antibody produced in BMI-1-deficient mice, we assessed IgG Fc N-linked glycan profiles. While most glycosylation patterns remained unchanged in the absence of BMI-1 (Extended Data Fig. 4a), there was a significant decrease in G2Z1F, whereas the corresponding G2Z1 remained unchanged (Fig. 3b). Afucosylated forms of IgG can increase antibody efficacy48, enhance antibody-dependent cellular cytotoxicity (ADCC)49 and induce effective innate cell function during chronic viral infection50. Specifically, the G2Z1F glycoform has been shown to be enriched in the IgG2c subset and is thought to act as a dampener of this pro-inflammatory isotype51. To determine whether antibody efficacy was increased by reduction of this glycoform, we compared the ability of antibody from Bmi1f/fCd23Cre/+ and Cd23Cre/+ mice to induce ADCC. Using natural killer (NK) cells from wild-type mice, ADCC was significantly enhanced using sera collected from LCMV-Docile-infected Bmi1f/fCd23Cre/+ mice, compared to sera from Cd23Cre/+ mice (Fig. 3c). Taken together, these data reveal BMI-1 as a key regulator of antibody efficacy in chronic viral infection.

We next investigated the kinetics of ASC production. The frequency of CD19+IgD+ cells (enriched in naive B cells) was comparable between Bmi1f/fCd23Cre/+ mice and controls, as was the ability of CD19+IgD to form GC B cells (Extended Data Fig. 4b–e). At d14, the total number of CD19+IgD+ and GC B cells was significantly reduced. BMI-1-deficient GC B cells were able to isotype switch to IgG2c and to maintain detectable virus-specific IgG2c (Extended Data Fig. 4f,g). In contrast, B cell-specific deletion of BMI-1 altered the dynamics of ASCs. In LCMV-Docile-infected Cd23Cre/+ mice, ASC formation peaked at d14 before declining to low levels by d21 (Fig. 3d–f). In comparison, ASC frequency, but not total number, was increased in Bmi1f/fCd23Cre/+ mice at d7, before a significant decrease in both frequency and number at d14 compared to controls (Fig. 3f). Notably, there was not a complete absence of ASCs; in fact, the reduction of d14 ASC frequency and number (Fig. 3e) resembled that observed in LCMV-WE (acute)-infected mice. Similarly, GC B cell frequency and number (Fig. 3g) and circulating ICs (Fig. 3h) in BMI-1-deficient mice infected with LCMV-Docile, resembled that observed in control mice infected with LCMV-WE. To confirm that changes to the ASC population were B cell-intrinsic, 1:1 mixed bone-marrow chimeras were established with Ly5.1 and Bmi1f/fCd23Cre/+ bone marrow (Fig. 3i). Infected chimeras showed a significant reduction in BMI-1-deficient ASCs compared to Ly5.1 (Fig. 3j). Thus, induction of neutralizing antibodies, augmentation of antibody-dependent effector function and reduction in circulating ICs are consistent with more potent immune responses in LCMV-Docile-infected Bmi1f/fCd23Cre/+ mice.

Relationship of c-Myc and BMI-1 in chronic viral infection

Affinity maturation is highly regulated to avoid the survival and expansion of potentially autoreactive or cancerous clones. A critical regulator of the selection process is c-Myc3,4. Given the role of BMI-1 in Myc-induced B cell malignancies39 and the heterogeneity of both Bmi1 and Myc3,4 expression in the GC, we hypothesized that BMI-1 and c-Myc may collaborate to regulate the efficacy of the antibody response to infection. EYFP+ and EYFP GC B cells were isolated from BMI-1 reporter mice infected with either LCMV-WE or LCMV-Docile (Fig. 4a). Of note, Myc expression was selectively expressed in the EYFP+ GC B cell subset, with a 21-fold increase compared to EYFP GC B cells (Fig. 4b). The fold change of Myc expression in EYFP+ GC B cells responding to chronic infection was considerably higher compared to the acute WE-infected cohort, in which there was a twofold, nonsignificant increase in EYFP+ over EYFP (Fig. 4b). We next investigated whether Bmi1 deletion affected Myc expression. Total GC B cells from LCMV-Docile-infected Cd23Cre/+ mice had increased expression of Myc compared to LCMV-WE (Fig. 4c). B cell-specific BMI-1 deletion selectively reduced Myc expression in LCMV-Docile-infected mice to the level observed in LCMV-WE infection (Fig. 4c). In contrast, there was no significant change in gene expression of other reported targets of BMI-1, Bcl2l11 (encoding Bim)34 and Pmaip1 (encoding Noxa)33 in acute or chronically infected Bmi1f/fCd23Cre/+ mice (Extended Data Fig. 4h).

Fig. 4: Regulation of c-Myc in BMI-1-deficient mice.
figure 4

a, Schematic of infection and tamoxifen administration of Bmi1CreERT2Rosa26EYFP reporter mice. b, RT–qPCR analyses of Myc expression in EYFP and EYFP+ GC B cells from Bmi1CreERT2Rosa26EYFP reporter mice infected with LCMV-Docile (d7 after infection; 2-𝜟𝜟Ct method relative to EYFP; reference gene is Ku70). WE, n = 7 EYFP and n = 5 EYFP+; Docile, n = 10 EYFP and n = 8 EYFP+; combined from two (WE) and four (Docile) experiments. Data represent mean ± s.e.m. **P = 0.0016, ****P < 0.0001 (Mann–Whitney U-test, two-tailed P value). c, Myc expression in GC B cells isolated from Cd23Cre/+ or Bmi1f/fCd23Cre/+ mice, d14 after infection, assessed by RT–qPCR (2-𝜟𝜟Ct method relative to Cd23Cre/+ WE-infected mice), n = 4 mice per group. Data represent mean ± s.e.m. *P = 0.0286 (Mann–Whitney U-test, two-tailed P value). d, Histological analysis of spleens from Cd23Cre/+ and Bmi1f/fCd23Cre/+ at d7 after infection; IgD (cyan), CD3 (red), Myc (yellow), PNA (magenta). Scale bar, 100 μm. Orange arrows indicate c-Myc-expressing cells. e, Quantification of Myc+PNA+ cells per IgDlo regions within a follicle; data are combined from ten individual regions from n = 4 mice per group, combined from two independent experiments and represent mean ± s.e.m. ****P < 0.0001 (Mann–Whitney U-test, two-tailed P value). fh, B cells were stimulated in vitro with CD40L, IL-4, IL-5 and with increasing concentrations of IL-21 as indicated and assessed for Bmi1 (f), Myc (g) and plasmablast frequency (h); n = 4–5 mice per group, combined from two independent experiments. Data represent mean ± s.e.m. (fh). *P < 0.05 (Mann–Whitney U-test, two-tailed P value).

Source data

Within the GC, c-Myc is localized to the LZ in immunization models3,4. Given the increase in Myc in chronic infection, we hypothesized that c-Myc would no longer be tightly regulated within the GC. Indeed, histological analyses of Cd23Cre/+ mice infected with LCMV-Docile revealed c-Myc expression throughout the GC (Fig. 4d and Extended Data Fig. 5a). In contrast, GCs in Bmi1f/fCd23Cre/+ mice were mostly absent of c-Myc expression, with only a few cells evidently expressing c-Myc (Fig. 4d,e and Extended Data Fig. 5b). Therefore, conditional deletion of BMI-1 restored regulated expression of c-Myc within the GC.

Given the changes in expression and distribution of c-Myc, we next examined whether somatic hypermutation and/or affinity maturation of GC B cells was altered at d14 post-infection (Extended Data Fig. 5c–f). No clear differences were observed in mutation levels (Extended Data Fig. 5c). To examine affinity maturation in BMI-1-sufficient versus BMI-1-deficient GC B cells, the ratio of non-synonymous to synonymous mutations (N/S) (Extended Data Fig. 5d) and selection pressure (Extended Data Fig. 5e,f) was assessed. The ratio of N/S in the complementarity-determining regions (CDRs) was decreased in GC B cells isolated from Cd23Cre/+ mice responding to chronic infection, compared to acute infection, suggestive of reduced affinity maturation. In contrast, GC B cells isolated from LCMV-Docile-infected Bmi1f/fCd23Cre/+ mice had a N/S ratio comparable to that of Cd23Cre/+ mice infected with LCMV-WE (Extended Data Fig. 5d). We then quantified the link to selection pressure more precisely with BASELINe (Extended Data Fig. 5e,f). B cell receptor (BCR) sequences from either Cd23Cre/+ or Bmi1f/fCd23Cre/+ mice infected with LCMV-WE had more positive selection in the CDR region than those infected with LCMV-Docile; however, the trend toward positive selection returned in Bmi1f/fCd23Cre/+ mice responding to chronic infection, compared to Cd23Cre/+ mice, although it was not completely restored to that of acute infection. In contrast to the CDR, there was little change in the negative selection detected in the framework regions (Extended Data Fig. 5f). Taken together, affinity maturation seemed to be improved in Bmi1f/fCd23Cre/+ mice during chronic infection.

The elevation in BMI-1 and c-Myc expression in chronic GC B cells prompted us to ask whether there was a particular pathogen-influenced microenvironmental change that may induce these differences in B cell transcriptional programs. A key candidate was IL-21, a cytokine that regulates B cell behavior52,53 and can induce context-specific Myc expression54. Notably, IL-21 production by helper T cells in chronic viral infection is substantially higher than in acute viral infection55. We therefore tested whether there was a dose-dependent induction of Bmi1 and Myc in in vitro-stimulated B cells, co-cultured with increasing concentrations of IL-21 (Fig. 4f–h). While B cells cultured with low IL-21 concentrations did not increase Bmi1 (Fig. 4f) or Myc (Fig. 4g) over the baseline stimuli (CD40L, IL-4 and IL-5), significant upregulation was evident at high IL-21 levels (Fig. 4f,g). The observed increase was not simply due to a change in plasmablast frequency within the cultures, as there was no significant change in plasmablast frequency between 25–50 ng ml−1 IL-21 (Fig. 4h). Together, these data reveal that the disruption in c-Myc regulation during chronic infection correlated with increased IL-21 and that conditional deletion of BMI-1 restores expression and localization of c-Myc-expression in GC B cells.

H2AK119Ub1 is decreased at key loci in chronic infection

The non-canonical form of PRC1 (which excludes BMI-1) and PRC2 are required for GC formation26. The ability of BMI-1-deficient B cells to form GCs and upregulation of BMI-1 in ASCs, suggest a switch between the non-canonical and the canonical forms of PRC1 to regulate B cell fate. This switch seemed to be disrupted in chronic GC B cells, with upregulation of Bmi1, Myc and ASC-related genes (Fig. 5a). Accordingly, c-Myc-target genes were significantly enriched in chronic GC B cells, compared to the acute gene expression profile (Fig. 5b,c), concomitant with identification of several c-Myc targets within DARs (Fig. 5c).

Fig. 5: H2AK119Ub1 is decreased at key loci in B cells responding to chronic viral infection.
figure 5

a, Heat map of selected genes in GC B cells isolated from WT mice infected with either LCMV-WE or LCMV-Docile, assessed by RNA sequencing. b, Gene set enrichment analysis (GSEA) (canonical pathway Hallmark gene set) plot of Myc targets using the RNA-seq dataset shown in Fig. 1. c, Accessibility changes in Myc targets within DARs (ATAC-seq dataset shown in Fig. 1; top) and Myc targets within the DEG dataset (RNA-seq dataset shown in Fig. 1; bottom). ****P < 0.0001 (Wilcoxon matched-pairs signed-rank test, two-tailed P value). d,e, Chromatin immunoprecipitation (ChIP)–qPCR analyses of H2AK119Ub1 (d) or H3K27me3 (e) in CD19+IgD (Sw B cells) and CD19+IgD+ (UnSw) B cells isolated 7 d after infection from Cd23Cre/+ (LCMV-WE or LCMV-Docile infected) or Bmi1f/fCd23Cre/+ (LCMV-Docile infected) mice; n = 3. Data represent mean ± s.e.m. *P < 0.05, **P < 0.01 (Unpaired Student’s t-test, two-tailed P value). f, RNA-seq analysis of GC B cells isolated 14 d after infection from Cd23Cre/+ (LCMV-WE or LCMV-Docile infected) or Bmi1f/fCd23Cre/+ (LCMV-Docile infected) mice. Heat map (left) showing DEGs that were detected in both (Figs. 1 and 5) RNA-seq datasets (FDR < 0.05, logFC > 0.58). Heat maps of clusters C1, C3 and C4 (right). Red denotes genes that were also identified in the ATAC-seq DARs dataset in Fig. 1.

Source data

Given that the non-canonical PRC1 deposits H2AK119Ub1 at key loci in GC B cells, it was unclear how changes in BMI-1 expression affected PRC1 function, if at all. We first asked whether changes in H2AK119Ub1 and H3K27me3, the histone marks regulated by PRC1 and PRC2, respectively, were altered at key loci in acute versus chronic GC B cells. There was a clear change in H2AK119Ub1 in control GC B cells between LCMV-WE- and LCMV-Docile-infected mice (Fig. 5d). B cells from LCMV-Docile mice had a significant decrease in H2AK119Ub1 at Myc and Prdm1 loci (Fig. 5d), whereas H2K27me3 was unchanged (Fig. 5e). B cell-specific BMI-1 deletion in chronically infected mice elevated H2AK119Ub1 at Myc and Prdm1 loci in unswitched B cells, with a less prominent increase at Irf4 in IgD B cells. While there was a trend increase in H3K27me3 in the absence of BMI-1, in most cases this change was not significant (Fig. 5e). Thus, conditional deletion of BMI-1 restored monoubiquitination at H2AK119 in chronic B cells to similar levels as control B cells responding to acute viral infection.

To determine whether the BMI-1-dependent change in H2AK119Ub1 regulated gene expression, we performed RNA-seq analysis of acute versus chronic GCB from Cd23Cre/+ or Bmi1 Cd23Cre/+ mice (Extended Data Fig. 6a). Using the DEGs identified in both Fig. 1 and Extended Data Fig. 6 datasets to explore whether BMI-1-deficiency restored chronic GC B cell gene expression to resemble that of acute GCB, we identified five distinct signatures (Fig. 5f). BMI-1 deletion partially restored gene profiles to that of control GCB responding to acute infection. Among these five clusters (C1-5), genes listed in C1 and C4 showed similar expression patterns between BMI-1-sufficient acute GC B cells and BMI-1-deficient chronic GC B cells (magnified heat maps; Fig. 5f). Taken together, these results underscore the crucial role of BMI-1 in regulating gene expression in B cells responding to chronic infection.

BMI-1 regulates ASCs in response to NP-KLH immunization

The differences observed in BMI-1 expression and function in B cells responding to acute versus chronic viral infection prompted us to investigate the specific role of BMI-1 during B cell differentiation in a tractable, non-infectious model. Thus, Bmi1f/fCd23Cre/+ and Cd23Cre/+ mice were immunized with NP-KLH in alum and humoral responses were assessed. Histological analyses revealed the loss of splenic IgG1bright cells at d7 after immunization (Fig. 6a). Correspondingly, there was a significant decrease in the total frequency of BMI-1-deficient ASCs (Fig. 6b), antigen-specific IgG1+ ASCs (Fig. 6c) and antibody at d14 and d28 (Fig. 6d,e), although d7 IgG1 and IgM antibody levels were comparable between BMI-1-deficient mice and controls (Extended Data Fig. 6b). As ASCs can arise from either GC-independent or GC-dependent pathways, we assessed whether the defect in ASCs was due to a loss of GC B cells. Bmi1f/fCd23Cre/+ B cells formed GCs, isotype switched and produced appropriate DZ and LZ ratios, in a manner comparable to Cd23Cre/+ mice (Extended Data Fig. 6c–e). The ability of B cells to form GCs in Bmi1f/fCd23Cre/+ mice is consistent with a previous study demonstrating that NP+IgDlo B cell could expand after immunization in Bmi1 germline knockouts56. While only a small fraction of GC B cells was EYFP+ after immunization, Myc was ~sixfold upregulated in EYFP+ compared to EYFP GC B cells (Extended Data Fig. 6f). To assess whether modulating BMI-1 expression affected affinity maturation, Vh186.2 in GC B cells was sequenced (Fig. 6f,g and Extended Data Fig. 6g). The overall number of mutations (Fig. 6f) and CDR3 length (Extended Data Fig. 6g) was comparable between Bmi1f/fCd23Cre/+ and Cd23Cre/+ mice; however, there was an increase in mutations in CDR1 (P = 0.0594) and a significant decrease in mutations in CDR3 (P = 0.0448) in the absence of BMI-1 (Fig. 6f). Next, the affinity-enhancing replacement of the codon encoding for amino acid tryptophan by that encoding leucine at position 33 (W33L) of Vh186.2 was measured. The percentage of the W33L mutation was increased in GCB from Bmi1f/fCd23Cre/+ compared to Cd23Cre/+ mice (71% compared to 55%, respectively) (Fig. 6g). Therefore, while GC B cells could form in Bmi1f/fCd23Cre/+ mice, B cell-specific BMI-1 deletion led to an increase of high-affinity variants within GC B cells, but a decrease in ASC output.

Fig. 6: BMI-1 regulates plasma cell survival.
figure 6

a, Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice were immunized with NP-KLH in alum. Histological analyses of spleens at 7 d after immunization, stained with B220 (red) and IgG1 (blue), representative of two individual experiments. Scale bar, 100 μm. b, Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice were immunized with NP-KLH in alum and the frequency of B220loCD138hi cells was assessed by flow cytometry; d7, n = 8 Cd23Cre/+ and n = 7 Bmi1f/fCd23Cre/+ mice; d14, n = 5 mice per group; d28, n = 9 Cd23Cre/+ and n = 10 Bmi1f/fCd23Cre/+ mice; combined from 2–3 individual experiments per time point. Data represent mean ± s.e.m. d7, *P = 0.0427; d14, **P = 0.0079; d28, *P = 0.0211 (Mann–Whitney U-test, two-tailed P value). c, ELISpot analyses of NP+IgG1+ ASCs at indicated time points after immunization; d7, n = 8 Cd23Cre/+ and n = 7 Bmi1f/fCd23Cre/+ mice; d14, n = 7 mice per group; d28, n = 9 Cd23Cre/+ and n = 10 Bmi1f/fCd23Cre/+ mice; combined from 2–3 individual experiments per time point. Data represent mean ± s.e.m. d7, ***P = 0.0003; d14, **P = 0.0076; d28, **P = 0.0020 (Mann–Whitney U-test, two-tailed P value). d,e, NP+IgG1+ serum antibody at d14 (d) and d28 (e) after immunization; d14, n = 3 Cd23Cre/+ and n = 4 Bmi1f/fCd23Cre/+ mice; d28, n = 4 mice per group; representative of two individual experiments per time point. Data represent mean ± s.e.m. *P < 0.05 (Mann–Whitney U-test, two-tailed P value). f, Single NP+IgG1+CD95hiCD38loCD138 GC B cells were sort-purified from Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice immunized with NP-KLH in alum (three mice pooled per group) at d21 after immunization. The Vh186.2 gene was sequenced; shown is the number of mutations per sequence across CDR and FW regions. g, Trp-to-Leu mutations at position 33 of Vh186.2 in GC B cells sort-purified from Cd23Cre/+ (42 sequences) and Bmi1f/fCd23Cre/+ (44 sequences). h, Uninfected Bmi1f/fPrdm1Rosa26CreERT2 mice and controls were administered with tamoxifen and assessed for ASC frequency. i, Frequency of ASCs assessed by flow cytometric analyses at d7 and d14 after tamoxifen analyses; n = 5 mice per group, combined from two individual experiments per time point. Data represent mean ± s.e.m. *P < 0.05, **P < 0.01 (Mann–Whitney U-test, two-tailed P value).

Source data

The data presented thus far suggested that BMI-1 was important for GC to ASC differentiation. We next asked whether BMI-1 directly regulated ASC survival independent of its role in differentiation. To specifically determine whether steady-state ASCs relied on BMI-1 for their survival, we generated Bmi1f/fPrdm1Rosa26CreERT2 mice in which BMI-1 is inducibly deleted in Prdm1 (the gene encoding Blimp-1)-expressing cells. Uninfected Bmi1f/fPrdm1Rosa26CreERT2 and control mice were administered with tamoxifen and ASCs were assessed 7 and 14 d later (Fig. 6h). Tamoxifen administration significantly reduced the frequency of ASCs at d14, and ASC number at both time points (Fig. 6i), but not GC B cells (Extended Data Fig. 6h–j) demonstrating that BMI-1 is an important regulator of ASC survival.

Small molecule inhibition of BMI-1 depletes ASCs

Targeting detrimental antibodies and ICs is a therapeutic goal for chronic infectious and inflammatory disorders. Given that BMI-1 regulated ASC survival, we assessed small molecule inhibition of BMI-1 to modulate ASC and IC formation. First, naive B cells were stimulated in vitro to induce plasmablast differentiation, which was significantly reduced by PTC-209 (ref. 57) in a dose-dependent fashion (Fig. 7a and Extended Data Fig. 7a). While plasmablasts were reduced, the ability of B cells to undergo division was unaffected across doses (Extended Data Fig. 7b), with the exception of 1.25 μM (not shown). Next, C57BL/6 mice were infected with LCMV-Docile and after 2 d to allow initial activation of the immune response, were given daily administrations of PTC-028, an inhibitor that targets and degrades BMI-1 through hyperphosphorylation58 (Fig. 7b). LCMV-Docile-infected mice that received the vehicle control lost weight before stabilization at 85% of the original body weight at d8 after infection (Fig. 7c). In contrast, LCMV-Docile-infected mice administered with PTC-028 lost <5% body weight up to d3 after infection, but then recovered by d5 (Fig. 7c). Similar to Bmi1f/fCd23Cre/+ mice, ASC frequency and number were significantly reduced after PTC-028 treatment, compared to controls (3.3-fold and 1.8-fold, respectively; Fig. 7d). Concomitantly, total (Fig. 7e) and LCMV-specific IgG2c IC (Fig. 7f) were significantly decreased in the PTC-028 group compared to vehicle controls. While liver viral titers were not significantly reduced with this dosage regimen (not shown), splenic architecture was restored in mice administered PTC-028, similar to that observed in Bmi1f/fCd23Cre/+ mice (Extended Data Fig. 7c). Collectively, these data reveal BMI-1 as a viable therapeutic target to deplete ASCs and reduce IC formation in immune disorders.

Fig. 7: BMI-1 can be targeted to deplete plasma cells in vitro and in vivo.
figure 7

a, B cells were stimulated in vitro with LPS and IL-4 and in the presence or PTC-209 or dimethylsulfoxide (DMSO) at the indicated concentrations. B220loCD138hi ASCs were assessed after 4 d by flow cytometry. Data are representative of three individual experiments; *P = 0.0117, ***P = 0.0002 compared to DMSO control (one-way analysis of variance). b, Schematic of PTC-028 administration in vivo. c, Cd23Cre/+ and Bmi1f/fCd23Cre/+ mice were infected with LCMV-Docile and administered with PTC-028 or vehicle control. c, mouse weights assessed at the indicated time points after infection. d, Frequency and number of ASCs assessed by flow cytometry. e, Assessment of total IgG2c+ ICs. f, Assessment of LCMV-specific IgG2c+ ICs; n = 4 mice per group, data are representative of two individual experiments and represent mean ± s.e.m. (cf); *P < 0.05 (Mann–Whitney U-test, two-tailed P value).

Source data


Antibodies are a key modulator of both innate and adaptive arms of an immune response. However, in chronic infections, the quality and quantity of antibodies are dysregulated and antibodies can themselves become pathogenic. Defining the mechanistic steps that shift humoral responses from effective to ineffective when viral infections persist has been elusive. Here, we identify BMI-1 as a critical regulator of B cell differentiation, which can be targeted therapeutically. Deletion of BMI-1 in B cells led to pleiotropic effects on the immune response: accelerated viral clearance, reduced splenomegaly coupled with improved lymphoid architecture, increased neutralizing capability and effector function of antibody. Expression of Myc, the transcription factor that is required for B cell selection in the GC3,4, was concentrated in Bmi1-expressing cells and regulated by BMI-1 in chronic viral infection. Further, the preservation of lymphoid architecture, reduction in excessive IC or changes to the N-glycan profile of antibody within ICs may work collectively to support affinity maturation in the GC11,59,60. These results highlight the critical direct and indirect roles of B cells controlling effective versus ineffective immune responses to persistent viral infection.

c-Myc is expressed in proliferating cells but is known to be tightly regulated in cells undergoing selection in the LZ3,4. Similarly, in acute infection BMI-1 was expressed in only a small fraction of GC B cells but upregulated upon ASC differentiation. Chronic viral infection deregulated expression of both BMI-1 and c-Myc, concomitant with an increase in GC B cells and ASCs. Increased c-Myc may result in the disruption of GC B cell division and selection, skewing cells toward premature differentiation5. BMI-1 and c-Myc cooperate in lymphoid transformation37,38 and can directly and indirectly regulate each other in a context-dependent manner61,62,63. Although a number of pathogen-influenced changes in the microenvironment have been recorded in chronic infection41,64,65, we reasoned that cytokine differences within the GC may mediate sustained induction of BMI-1 and c-Myc. One key cytokine that is notably increased in chronic LCMV infection, compared to acute, is IL-21 (ref. 55). Accordingly, high concentrations of IL-21, but not low levels, induced BMI-1 and c-Myc in stimulated B cells. Persisting GCs have been proposed to be beneficial in inducing broadly neutralizing antibodies66. However, we highlight the importance of appropriate regulation of this process under sustained inflammatory pressure. In particular, BMI-1 plays an essential role in dictating whether B cells recruited into a response against infection will contribute to a successful humoral response.

The specific role of BMI-1 in ASCs reveals the importance in the composition of the heterogeneous PRC1 during B cell differentiation, changing from non-canonical26 to canonical components during an immune response to regulate GC and ASC biology, respectively. Here, H2AK119Ub1 was decreased at key loci in GC B cells in chronic infection, compared to acute, suggesting a deregulation in PRC function. BMI-1 deletion increased H2AK119Ub1, in a seemingly paradoxical manner as BMI-1 is known to support the ubiquitin ligase activity of RING1B29. This suggests that the aberrant recruitment of BMI-1 into the PRC1 in chronic GCs disrupts the non-canonical function of the PRC1, concomitant with downregulation of BCOR, a regulator of GC biology26. It also suggests that this dynamic process is regulated by microenvironmental signals within the GC. Understanding the relationship between cytokine signaling and resultant biochemical changes in the composition of the PRCs may implicate other chronic inflammatory mediators that lead to detrimental B cell differentiation.

Small molecule inhibitors of epigenetic regulators are rapidly emerging as important therapeutic tools and as such are in clinical trials for a number of disorders67,68. Excessive IC deposition induces substantial tissue damage in chronic infections and other antibody-mediated diseases in which depleting pre-existing ASCs is a therapeutic goal69,70. We propose that BMI-1 could be targeted with a small molecule inhibitor to deplete pre-existing detrimental plasma cells. This strategy may also be used to tackle pathogenic antibody-dependent enhancement, in which antibody produced from initial viral exposure leads to multi-organ damage upon re-exposure. In identifying BMI-1 as a critical determinant of productive antibody responses to chronic viral infection, this study has important therapeutic implications for immune disorders in which pathogenic antibodies play a role in morbidity.



All animal experimentation was performed following the Australian National Health and Medical Research Council Code of Practice for the Care and Use of Animals for Scientific Purposes guidelines and in accordance with institutional regulations. The Monash Animal Ethics Committee approved all procedures. Bmi1f/fCd23cre mice were generated by crossing Cd23cre mice provided by M. Busslinger (Institute of Molecular Pathology)71 with Bmi1f/f mice provided by S. Morrison (University of Texas Southwestern)72. Bmi1CreER mice (010531; Jackson Laboratories)73 were crossed with Rosa26EYFP mice74. Bmi1f/fPrdm1Rosa26CreERT2 mice were generated by crossing Bmi1f/f with Prdm1Rosa26CreERT2 mice provided by D. Tarlinton (Monash University). Ly5.1 mice were kindly provided by C. Zaph (Monash University). All mice were on a C57BL/6 background, backcrossed and maintained at the Monash Animal Research Platform. Both male and female mice were used in the experimental research project, between 42 d and 4 months for intact mice; up to 6 months for bone-marrow chimera experiments.

For immunization, mice were injected intraperitoneally (i.p.) with 100 µg 100 µl−1 per mouse of NP13-KLH precipitated in adjuvant alum. For LCMV infection, 100 μl of LCMV-WE (3 × 103 p.f.u.) or LCMV-Docile (2 × 106 p.f.u.) was given intravenously (i.v.) per mouse tail vain. For tamoxifen administration, Bmi1CreERT2RosaEYFP and Bmi1f/fPrdm1Rosa26CreERT2 mice were treated with tamoxifen (Sigma) dissolved in peanut oil and ethanol at 5 mg ml−1. Then, 90 µl of tamoxifen or vehicle control (peanut oil and ethanol) was administered to each mouse for 3 d consecutively via oral gavage at indicated time points.

Flow cytometry, cell sorting and antibodies

Single-cell suspensions were resuspended in PBS 2% FCS and stained for flow cytometric analysis. Live or fixed cells were analyzed on the Fortessa or Canto (BD); cytometry data were acquired with FACSDiva software (BD) and analyzed with FlowJo (TreeStar). FcγRII/III (2.4G2; supernatant) and Normal Rat Serum (Sigma-Aldrich) were used to block nonspecific binding. For sort purification, cells were stained with fluorochrome-labeled antibodies (Supplementary Table 1) and purified by FACSAria or Influx (BD). Gating strategies used are shown in Extended Data Fig. 8.



Splenic portions were fixed using periodate–lysine–paraformaldehyde fixative for 6–8 h at 4 °C. Tissues were transferred to a 30% sucrose solution and incubated overnight at 4 °C. Spleens were then embedded in OCT (Tissue-Tek) and kept frozen at −80 °C until staining. Seven-micrometer tissue sections were cut via microtome (Leica) and mounted on Superfrost Plus slides. Sections were fixed with cold acetone (Sigma) for 10 min, blocked with PBS–5% FCS and stained using the indicated antibodies (Supplementary Table 2). Slides were observed under ×4 and ×10 magnification using an Olympus IX71 inverted microscope. Images were captured with a mounted high-sensitivity cooled SPOT RT KE CCD camera; white balance and exposure were set using SPOT 5.1 imaging software. Confocal microscopy images were acquired on a Leica SP8 laser scanning confocal microscope on a DMi8 inverted stand. Overview images were acquired with a multi-immersion objective HC PL APO ×20/0.75 IMM CS2 objective with oil immersion. For higher resolution images an HC PL APO ×63/1.40 CS2 oil immersion objective was used. Image resolution was 1,024 × 1,024. Quantitation of PNA+Myc+ cells within IgDlo regions, as well as 100 µm scale bars and pseudo-colors, were assigned to raw images using ImageJ software. Brightness and contrast were applied equally to each image. Threshold and despeckle tools were used for quantification analysis.


Splenic portions were frozen in OCT (Tissue-Tek). Seven-micrometer sections were cut using a microtome (Leica) at −20 °C and fixed with acetone. Staining was performed using described antibodies (Supplementary Table 3). Slides were viewed under ×4 and ×10 magnification using an Olympus CKX41 microscope and images were captured with a mounted Nikon DP-12 camera. Raw images were taken using the Nikon NIS-Element software platform and ImageJ was used to add 100-μm scale bars to each image.

Deletion analysis

Sort-purified subsets were processed using the NucleoSpin Tissue kit (Macherey-Nagel, 740952.250). Genomic DNA was amplified using the following primers: Bmi1 LoxP forward (5ʹ-GCTAGCATTCCTGGTTTTGC-3ʹ), Bmi1 LoxP reverse (5ʹ-GGCACAGTGATGAGGTGTTG′3ʹ) and Bmi1 excised (5ʹ-CACGAGGTGCTTCTTTCCTC-3ʹ).

RNA purification and quantification

Sort-purified cell subsets were centrifuged, lysed in RLT buffer (QIAGEN) and passed through a gDNA eliminator spin column (QIAGEN) to remove gDNA. Total RNA was extracted using the RNeasy Plus Micro kit (QIAGEN, cat. no. 74034) according to manufacturer’s instructions. First-strand complementary DNA synthesis was performed with a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems, cat. no. 4368814). RT–qPCR was performed with Rotor-Gene Q using the QuantiNovaSYBR Green PCR kit (QIAGEN, cat. no. 208054). The forward and reverse primers for Bmi1 (ref. 75) were 5ʹ-AATTAGTCCCAGGGCTTTTCAA-3ʹ and 3ʹ-TCTTCTCCTCATCTGCAACTTCTC-5ʹ; Myc: 5ʹ-GCTGTTTGAAGGCTGGATTTC-3ʹ and 3ʹ-GATGAAATAGGGCTGTACGGAG-5ʹ; Bcl2l11: 5ʹ-CGACAGTCTCAGGAGGAACC-3ʹ and 3ʹ-CCTTCTCCATACCAGACGGA-5ʹ; and Pmaip1: 5ʹ-GAACGCGCCAGTGAACCCAA-3ʹ and 3ʹ-CTTTGTCTCCAATCCTCCGG-5ʹ. The housekeeping Ku70 gene was used as an internal control for normalization. To assess Ku70 expression, the following set of primers was used: 5ʹ-TCGCCTTTACTGAGAAGGTGAC-3ʹ and 3ʹ-TGCTGCAGGACTGGATTCTC-5ʹ. The 2-𝜟𝜟Ct method was used to determine relative changes in gene expression.


Sort-purified subsets were processed as published76. Briefly, sorted cells were incubated with 0.6% formaldehyde for 10 min. Crosslinking was stopped by addition of glycine to a final concentration of 0.25 M and cross-linked chromatin was sonicated to reduce DNA size to 200–500 base pairs (bp). Sonicated chromatin was then used for ChIP analysis. Immunoprecipitation was undertaken with mouse antibodies against H3K27me3 (Millipore, 07-449) and H2AK119Ub1 (Cell Signaling Technology, 8240). RT–qPCR was performed as described above. Primers used to amplify a fragment including the transcription start site of indicated genes were: Prdm1 5ʹ-GAGTAGTCAGTCGCTCGCTC-3ʹ and 3ʹ- GGTCCCTCCTTTTCTACGGC-5ʹ; Irf4 5ʹ-GAGCTGAAGAAAGCCAGGGT-3ʹ and 3ʹ-CTTTATAGAGCCGGAGGCGG; and Myc 5ʹ-GCGCGAGCAAGAGAAAATGG-3ʹ and 3ʹ- GCTCCGGGGTGTAAACAGTA-5ʹ. Data were normalized by dividing the total copies of no-antibody control, H3K23me3 or H2AK119Ub1 to copies of the total input. Each value was then multiplied by a factor of 100 to obtain the frequency (%) of input.


For ELISA, LCMV-WE and Docile viral lysates were used to coat 96-well high-binding plates (Sarstedt) overnight at 4 °C. Plates were then blocked with PBS 1% BSA for 1 h, washed and loaded with serially diluted serum for a 4 h incubation at 37 °C. Plates were washed again and incubated for 1 h at 37 °C with IgG2c conjugated to horseradish peroxidase (HRP; Southern Biotech), followed by washing and development with OPD-substrate solution (Sigma-Aldrich). For ELISPOT and ELISA of NP-KLH samples, multiscreen HA plates (Millipore) were coated overnight with NP12BSA or NP1BSA for quantification of NP-specific IgG1+ ASCs. Plates were blocked, washed and loaded with samples prepared in RPMI 5% FCS, 50 μM 2-mercaptoethanol and 2 mM glutamine before an overnight incubation at 37 °C. Plates were washed and subsequently incubated for 1 h at 37 °C with secondary antibody (IgG1) conjugated to alkaline phosphatase (Southern Biotech) for 1 h. Plates were washed and developed with the BCiP/NBT reaction (Sigma-Aldrich). For ELISA analysis, high-binding plates were coated with NP12BSA, serum samples followed by IgG1-HRP incubated as described above, before detection with OPD-substrate solution. A non-linear fit is shown (GraphPad Prism).

IC precipitation

ICs were precipitated from mouse sera11,77. Briefly, serum was incubated with a 1:1 ratio of PBS 5% PEG6000 solution overnight at 4 °C. The serum–PEG samples were mixed with RPMI 2.5% PEG6000 in a 1:3 ratio and spun down at 2,000g for 30 min at 4 °C. PEG precipitates were resuspended to the initially used serum volume in cold PBS and then quantified by ELISA.

T cell cytotoxicity assay and IFN-γ production

Cytotoxicity capacity of CD8+ T cells were assessed as previously described78. Briefly, CD8+ T cells were negatively enriched using the CD8α+ T cell (mouse) isolation kit (Miltenyi Biotec). Splenocytes from Ly5.1 mice were used as targets, labeled with cell trace violet (CTV; Molecular Probes) and loaded with gp33–41 peptide at 0.1 µM. Enriched CD8+ T cells were plated with peptide-pulsed (CTV high) or unpulsed (CTV low) splenocyte target cells from Ly5.1 mice in either a 1:1 ratio of 2.5:1 ratio and incubated for 5 h at 37 °C, stained for surface markers, propidium iodide (Molecular Probes), anti-CD45.1:APC-Cy7 (A20; BioLegend) and anti-CD45.2:PE (104; BioLegend) and assessed by flow cytometry. For cytokine production, approximately 1 × 106 splenocytes were stimulated in vitro for 5 h with IL-2 10 U ml−1 (Peprotech) and brefeldin A (BD Cytofix/Cytoperm) in the presence or absence of 1 µM H2-Db gp33–41 peptide. Cells were surface stained with anti-CD8:BUV395 (53-6.7, BD Biosciences), then fixed and permeabilized for intracellular expression of IFN-γ:FITC (XMG1.2, BD Biosciences) and analyzed by flow cytometry.

Bone-marrow chimeras

Lethally irradiated Ly5.1 mice (2 × 4.5 Gy) were reconstituted with 50% Ly5.1 and 50% Ly5.2-Cd23Cre/+ or 50% Ly5.2-Bmi1f/fCd23Cre/+ bone marrow.

Plaque-forming unit assay and infectivity reduction assay

Livers were weighed and the calculated volume of DMEM (Gibco) was added to obtain 500 mg ml−1 of suspension. One stainless steel bead (QIAGEN) was added per sample and liver tissues were homogenized using the TissueLyser LT (QIAGEN) for 5 min at 50 oscillations s−1. Beads were removed and samples were spun for 10 min at 15,000 r.p.m. (4 °C) to pellet cell debris. For the p.f.u. assay, homogenized liver suspensions were serially diluted and added onto MC57G cells. An overlay mixture of 2% methylcellulose solution and 2× DMEM (Gibco) was added after virus adsorption and plates incubated (37 °C for 48 h). Plates were then stained with primary VL-4 antibody (previously described79) and the secondary HRP goat anti-rat IgG (Jackson ImmunoResearch). Plaques were developed (HRP Substrate kit; Vector) and counted to quantify viral load. For infectivity reduction assays, sera from LCMV-infected experimental mice were inactivated at 56 °C for 30 min to eliminate pre-existing virus. Sera were then diluted and incubated for 90 min at 37 °C with a known quantity of LCMV (30 p.f.u.). The antibody-virus mixtures were added onto MC57G cells and plaques were calculated. The neutralization capacity of the serum was expressed in percentage of plaque reduction, referring to the no-serum control sample.


Target cells MC57G were cultured in complete DMEM (DMEM, 10%, NaPyr, L-Glu, NEAa, 2-ME and HEPES, pH 7.2–7.5), labeled with CTV and infected with LCMV-Docile at a multiplicity of infection of 1, 24 h before the cytotoxicity assay. Splenic NK cells were isolated from naive C57BL/6 mice using nylon wool filters (Polysciences) and seeded in complete RPMI-1640 (RPMI-1640, 5% FCS, 1% penicillin, streptomycin and 2 mM glutamine) supplemented with 250 ng ml−1 IL-2 for 5 d. ADCC was initiated by the addition of serum, previously collected from mice infected with LCMV-Docile, to the target cells. NK cells were combined in a 10:1 effector:target ratio. The plate was further incubated for 2 h at 37 °C. Cells were collected and stained with Fixable Viability Stain (BD Biosciences) to evaluate the frequency of dead cells by FACS.

IgG purification

IgG antibodies were purified from plasma according to manufacturer’s protocol (Melon Gel IgG Purification kit, Thermo Fisher Scientific). Purified IgG antibody samples were centrifuged through 100 kDa Amicon Ultra filters (Merck) at 14,000g for 10 min to remove excess serum proteins and buffer was exchanged into PBS. Purity was confirmed via SDS–PAGE (Bio-Rad) and IgG concentrations were measured using a nanodrop spectrophotometer (Bio-Rad).

IgG N-linked glycan profiling

IgG N-linked glycosylation patterns were measured80,81 according to ProfilerPro glycan profiling LabChip GXII Touch protocol on the LabChip GXII Touch instrument (PerkinElmer). Microchip capillary electrophoresis laser-induced fluorescence (CE-LIF) analysis of digested and labeled N-linked glycans was performed. The relative prevalence of major N-linked glycan profiles of IgG was analyzed using the LabChip GX Reviewer (PerkinElmer) software. Peaks were assigned based on the migration of known standards and glycan digests45. Peak area and relative prevalence of each glycan pattern were calculated.

In vitro B cell stimulation

B cells were isolated using a B cell negative enrichment kit (STEMCELL). Cells were resuspended in RPMI 5% FCS, 50 μM 2-mercaptoethanol and 2 mM glutamine. Then, 5 × 104 labeled B cells were cultured with CD40L (R&D Systems) in combination with IL-4 and IL-5 (R&D Systems) and in the presence of increasing IL-21 (Peprotech) at the indicated ng ml−1 concentrations.

Small molecule inhibitors

In vitro, 5 × 104 CTV-labeled B cells were cultured with LPS (Sigma) in combination with 50 ng ml−1 IL-4 (R&D Systems) in the presence of DMSO or PTC-209 (Sapphire Bioscience) at the indicated concentrations. In vivo, mice were infected with LCMV-Docile and administered with 15 mg kg−1 body weight per mouse of PTC-028 (Sapphire Bioscience) small molecule inhibitor or vehicle control (0.5% hydroxypropyl methylcellulose and 1% Tween 80) via oral gavage every 24 h.

ATAC sequencing

Briefly, 50,000 cells were pelleted and washed with PBS. One-step permeabilization and tagmentation method was used by resuspending the cell pellets in Digitonin (EZSolution), Tween-20 (Sigma) buffer. High-molecular-weight tagmented DNA was then removed by incubating samples with a 0.7× ratio of Agencourt AMPureXP beads (Invitrogen). The same process was applied using a 1.2× ratio of beads to exclude low-molecular-weight DNA. Purified DNA was then pre-amplified with indexed primers and HiFi HotStart Polymerase Ready Mix (KAPA Biosystems). After estimation of addition PCR cycles required82, the genomic library was fully amplified using the SYBR qPCR Master Mix with primers included in the Illumina Library Quantification kit for Bio-Rad iCycler (KAPA Biosystems). Libraries were sequenced on a HiSeq2500 using 50-bp paired-end Illumina chemistry. For data processing, reads were aligned to the NCBI37/mm10 build of the Mus musculus genome using Bowtie2 (ref. 83) and PCR duplicate reads were flagged using the Picard MarkDuplicates tool. Enriched accessible regions were determined using MACS2 (ref. 84) and reads overlapping peaks were annotated for each sample using GenomicRanges85 tool in R. DARs were determined using DESeq2 (ref. 86). PCA was performed using the vegan package in R.

RNA sequencing

Briefly, 250,000 cells were pelleted and washed with PBS. RNA purification was performed using the RNeasy Micro kit (QIAGEN) or Direct-zol RNA Microprep kit (Zymo Research), according to the manufacturer’s instructions. Total mRNA was quantified with the Qubit Fluorometer (Invitrogen) and RNA integrity number values were generated for quality assignment by using the Agilent Bioanalyzer. Quality control-assessed libraries were then sequenced at the Illumina NextSeq500. For data processing, reads were aligned to the NCBI37/mm10 build of the M.musculus genome using the Subread aligner. Library counts were obtained using featureCounts and log2fold change were processed applying the DeSeq2 function. A PCA plot was generated using the PCAplot tool included in the DeSeq2 library. Volcano plot and quality control analysis were acquired using the Degust tool87. K-means analysis and heat maps were carried out using the kmeans and heatmap.2 tools in RStudio. PRC and Myc target analyses plots were generated using information about targets from published datasets88,89,90,91,92,93.


QIAGEN’s IPA software was used for functional pathway analysis of DEGs identified in this study. Analysis was performed using expression cutoff of 0.58-fold and FDR of 0.05. GSEA was performed on unfiltered normalized counts to identify signaling pathway that are differentially activated in GC B cells between acute and chronic infection. Analysis was executed by using the MSigDB collection of the Hallmark gene set covering principal molecular functions.

Vh sequencing

Splenocytes from NP-KLH-immunized mice were pooled and single GC B cells were sort-purified into 96-well plates containing PBS on a BD FACSAria equipped with a plate sorter and prepared for sequencing using nested primers for Vh186.2 and Cγ1 as previously described94. DNA Sanger sequencing was performed by Macrogen, Singapore, using 3ʹ primers from the second round of PCR. Vh186.2 sequences were aligned to germline using IMGT V-Quest and sequences of poor quality, those not identified as Vh186.2, miscalls, incomplete sequences or those containing stop codons were excluded from further analysis. The number and frequency of mutations was identified by comparison to the germline sequence. The number of sequences with the position 33 mutation, tgg (Trp) -> ttg, ctt or tta (Leu), was counted.

LCMV BCR sequencing

Methods were adapted from previous work95. Sort-purified bulk GC B cells (B220+IgDloCD38-CD95+CD138) were processed with a NucleoSpin Tissue mini kit for DNA (Macherey-Nagel, 740952), according to manufacturer’s instructions. Genomic DNA was pre-amplified for mouse antibody heavy-chain library construction (using primers listed in Supplementary Table 4), amplicons were purified and Illumina NexteraXT adaptors were added to the Igh library95. Paired-end sequencing using Illumina MiSeq V2 chemistry and 251-bp paired-end reads was performed (Hudson Institute). For data processing, high-throughput BCR data of raw sequences from eight mice along with their metadata were stored, annotated and analyzed with ImmuneDB96. V and J gene assignment was performed using the Anchoring method97. For clone assignment, sequences in each mouse were divided into bins of common Vh and Jh gene assignment and the same CDR3 length. All of the sequences with 85% or more similarity in their CDR3 amino acids were grouped into the same clone. For mutation analysis, the number of unique mutations per clone were calculated in CDRs and framework regions, defined previously98. We considered both the overall levels of non-synonymous and synonymous mutations and the overall ratio of non-synonymous to synonymous mutations in each region of each mouse. For selection pressure analysis, the overall selection pressure in CDRs and framework regions was assessed across all clones of each mouse. Selection pressure analysis was performed with BASELINe (Bayesian Estimation of Antigen-driven Selection in Ig Sequences)99, part of the Shazam package100.


Sort-purified GC B cells were resuspended in RIPA buffer (Sigma, R0278) supplemented with Roche complete mini protease inhibitor cocktail (Roche, 11836170001). Protein lysates were prepared with 4× Laemmli Sample Buffer (Bio-Rad, 1610747) supplemented with β-mercaptoethanol. Samples were run on 12% SDS–PAGE gels (Bio-Rad, 1610174). Transfer blotting on nitrocellulose membranes was performed (40 min at 90 V). Membranes were blocked in 5% milk. Primary antibodies were: BMI-1 (D20B7; Cell Signaling Technology, 6964) and histone H3 (D1H2; Cell Signaling Technology, 4499). Secondary antibodies were peroxidase AffiniPure goat anti-rabbit IgG (H + L) (Jackson ImmunoResearch, 111-035-144). H3 controls of samples were run in parallel due to technical limitations of developing both H3 and BMI-1 on the same blot. Membranes were developed using Clarity Western ECL Substrate (Bio-Rad, 1705060) and images were taken at the ChemiTouch (Bio-Rad) at different exposure times. Intensity of bands was quantified using the ‘lane and bands’ tool in Image Lab software (Bio-Rad, v.6.1).

Statistical analysis

The Mann–Whitney nonparametric two-tailed test, unpaired Student’s t-test, Wilcoxon matched-pairs signed-rank test or a one-way analysis of variance was used to analyze the significance of data (GraphPad Prism). Statistical values are ****P < 0.0001, ***P < 0.001, **P < 0.01 and *P < 0.05. Data are represented as mean ± s.e.m. No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications20,101. Data distribution was assumed to be normal but this was not formally tested. There was no randomization to data collection or organization of experimental conditions. Animals were assigned to experimental groups based on genotype, sex and age matching. Mice assessed to be nonresponders to stimuli were excluded from analyses. Data collection and analysis were not performed blind to the conditions of the experiments.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.