Plasmablasts derive from CD23^– activated B cells after the extinction of IL-4/STAT6 signaling and IRF4 induction

The production of antibody-secreting cells (ASCs) from B cells (that generate both cycling plasmablasts [PBs] and long-lived plasma cells [PCs]) is crucial for adaptive immunity and involves several complex molecular mechanisms that can be hijacked by tumoral processes that then give rise to germinal center (GC)–derived lymphomas.^1-3  Terminal B-cell differentiation is an autonomous, stochastic process^4,5  that occurs in the GC and outside the follicules.⁶  The collaboration of T cells and B cells has an impact on the rate of cell differentiation by modulating the levels of signals provided to the B cell.^7,8  It was initially acknowledged that T follicular helper (Tfh) cells secreted interleukin-21 (IL-21), but these cells were subsequently shown to produce IL-4. In mice and in humans, IL-4 and IL-21 were shown to have both redundant and respective functions in B-cell differentiation that were linked to specific times and sites of action.^9-13  Tfh cells are the dominant source of IL-4 in vivo; this cytokine drives B-cell expansion, maturation, and class switch recombination for murine and human antibodies. IL-4 activates STAT6 by phosphorylation, inducing expression of major histocompatibility complex class II, CD23 (encoded by FCER2), and the IL-4 receptor (IL-4R).¹⁴  This activation is critical for T-cell–dependent adaptive immune responses in vivo, as shown by the similar defects presented by Stat6 knockout (KO) and Il4ra KO mice.^14,15  Tfh-produced IL-4 is also essential in follicular lymphoma^16,17  in relation to activating STAT6 mutations¹⁸  and CD23 expression heterogeneity linked to differential prognosis.¹⁹  However, the mechanisms underlying the various actions of IL-4 have yet to be fully characterized, as well as whether these mechanisms are still active once B cells have committed to becoming PBs.

The cell identity distinction between B cells and PBs is maintained by mutually antagonistic interactions between regulatory factors²⁰ ; of these, interferon regulatory factor 4 (IRF4) is essential for both GC B cells and PCs.^21,22  In mice, GC B cells express Cbl/Cblb ubiquitin ligases that enable moderate B-cell receptor (BCR) and CD40L signals to promote B-cell clonal expansion and antibody affinity by mediating Irf4 protein ubiquitination and degradation. Once high affinity is achieved, strong BCR and CD40L signals induce a loss of Cbl/Cblb activities, which allows an increase in Irf4 protein levels and thus drives PB differentiation, GC egress, and high immunoglobulin production.²³  It is not known whether this mechanism also operates in humans.

Our controlled in vitro model of human B-cell differentiation already enabled us to characterize the final steps in B cells to PB commitment on the molecular level^8,24,25  and to demonstrate the role of 5-hydroxymethylation in this transition.²⁵  We used this model in our study to characterize the precise time at which the fate of the B cell is decided. An integrated genomic analysis revealed that both the IL-4/STAT6 and CBLB ubiquitin ligase pathways are extinguished during the emergence of human PBs.

Additional details regarding the methods are given in supplemental Methods, available on the Blood Web site.

Peripheral blood mononuclear cells were obtained from the French Blood Bank (Établissement Français du Sang de Bretagne, Rennes, France). Healthy volunteers were recruited after approval by the French Ministry of Higher Education and Research (AC-2014-2315), and they provided informed consent in line with the Declaration of Helsinki. Human naive B cells (NBCs) were purified (purity >99%) by negative selection using magnetic cell separation (Miltenyi Biotech).

Cell culture conditions, antibodies, flow cytometry, and fluorescence-activated cell sorting procedures have been described previously,^24,25  except for the use of Violet Proliferation Dye 450 (BD Biosciences) to enable sorting of resting or proliferating day 4 cells (day 4^hi/lo) and the use of the CaspGLOW Kit (eBioscience) to exclude active caspase-3–positive (apoptotic) cells. When indicated, cells were treated with the proteasome inhibitor MG132 (10 µM; Sigma-Aldrich) or cycloheximide (50 µg/mL; Sigma-Aldrich). Cytology analyses were done after May-Grünwald-Giemsa staining.

On day 4, after 2 hours of starvation in fetal calf serum–free medium, cells were stimulated (or not) with 20 ng/mL IL-4 for 10 minutes, subsequently fixed with 1.6% paraformaldehyde, permeabilized with cold methanol, and then stained with a STAT6(pY641)-Alexa 488 monoclonal antibody (mAb). For cells collected on day 6, the following antibodies were added: CD38-APC, CD23-PE-Cy7, and active caspase-3–PE (all from BD Biosciences). Data were acquired on a CytoFLEX flow cytometer (Beckman Coulter) and analyzed by using Flowlogic software (Miltenyi Biotec).

RNA was extracted and reversed transcribed into complementary DNA. Quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) analyses were performed with TaqMan Gene Expression Master Mix and Assays-on-Demand sets (Invitrogen). Gene expression levels were quantified using ABL1 as an endogenous control. The 2^–ΔΔCT method was used to determine the relative expression of each gene.

Cell pellets were lysed in radioimmunoprecipitation assay buffer completed with protease inhibitors. After concentration was determined, proteins were separated by using a 4% to 12% acrylamide gel and were transferred onto nitrocellulose membranes. Gels were revealed with horseradish peroxidase–conjugated secondary antibodies. The following antibodies were used for biochemical assays: anti-CBLB (D3C12, #9498), anti-CBL (#2747), anti-IRF4 (#4964), p21 Waf1/Cip1 (12D1, #2947), anti-STAT6 (#9362) (all from Cell Signaling) and anti-BLIMP1 (3H2-E8, #NB600-235; BioTechne), anti-IL-4R (AF-230, R&D Systems), and anti-β-actin (#AC-15; Sigma-Aldrich).

After protein extraction using an NP-40 lysis buffer, cell lysates were incubated overnight with 2 µg of rabbit mAb anti-IRF4 (D9P5H) or a rabbit mAb isotype control (DA1E; both from Cell Signaling). After incubation for 1 hour with Dynabeads Protein A (Thermo Fisher Scientific) and magnetic separation, immunoprecipitates were analyzed by western blotting.

For IRF4 ubiquitination analysis, day 4–activated B cells were cultured for 14 hours in the presence of cocktail 2 (IL-2 + IL-10 + IL-4), and 10 µM MG132 was added during the 4 last hours. Cell lysates were enriched for ubiquitinated proteins using the Signal-Seeker Ubiquitination Detection Kit (Cytoskeleton) and following the manufacturer’s instructions. After immunoprecipitation of ubiquitinated proteins during 2 hours with ubiquitination affinity beads or control beads, immunoprecipitates were analyzed by western blotting.

Three independent paired experiments were performed. RNA was extracted and sent for library preparation and sequencing by Helixio (Clermont-Ferrand, France). Reads were aligned against the hg19 human genome using TopHat2,²⁶  keeping only reads with a single alignment. Raw counts for each sample were obtained using HTSeq counts.²⁷  Differential expression of filtered genes (HTSfilter, 11 803 genes) was analyzed using DESeq2²⁸  in R (version 3.3.1; https://www.r-project.org). Differentially expressed genes were selected with an adjusted value of P < .05 and a fold-change >2 after Wald’s test and Benjamini-Hochberg correction. Principal component analysis (PCA) was performed on normalized read counts for the fold-change–filtered genes. Canonical pathways and biological functions were generated in an Ingenuity Pathways Analysis.

Assay for Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq) was performed on 1 set of day 4^lo, P2/CD23⁺, P2/CD23^–, and P1 cell populations, according to published methods.²⁹  After fluorescence-activated cell sorting, 50 000 cells were manually counted and were treated with Tn5 transposase. Libraries were amplified and sequenced (2 × 75 bp, paired-ends; Illumina HiSeq4000) by IntegraGen (Evry, France). Reads were aligned against the hg19 human genome, using Bowtie1.³⁰  Open regions were determined with Model-Based Analysis of ChIP-Seq2 (MACS2)³¹  and filtered with a value of q < 0.001. Read counts were determined by using HTSeq counts, and differentially opened (DO) regions were identified with DESeq2²⁸  and a likelihood ratio test P-value cutoff of .1. Read densities were calculated by creating a common set of regions, counting the number of reads overlapping with each feature, and normalizing by sequencing depth and region size.

ATAC-seq and RNA sequencing (RNA-seq) data were integrated by applying a cutoff distance between regions and transcription start sites of 500 kb. The best region and gene pairs were then determined by calculating the correlation between gene expression and ATAC-seq densities and keeping only pairs with Pearson’s r > 0.8. Graphical representations, data integration, and statistical analyses were generated with R and locally developed bash scripts. Unless otherwise indicated, groups were compared in a 2-tailed Wilcoxon’s test.

Human blood NBCs were differentiated in vitro into PBs.²⁴  On day 4, cell division tracking using violet proliferation dye 450 (VPD 450) discriminated between VPD^hi (day 4^hi) and VPD^lo (day 4^lo) subsets and on day 6, together with the CD38 marker, showed 3 cell populations characterized previously^24,25  and referred to as P1, P2, and P3 (Figure 1A). We determined that the nonapoptotic P2 cells contained CD23⁺ and CD23^– subsets, respectively, denoted as P2/CD23⁺ and P2/CD23^–. Only P2/CD23^– cells were able to generate PBs, as shown previously²⁵  (Figure 3D; supplemental Figure 4). To explain this divergent differentiation potential between P2/CD23⁺ and P2/CD23^– cells, we conducted an integrated RNA-seq and ATAC-seq analysis.

A PCA of paired RNA-seq data from day 4^lo, P2/CD23⁺, P2/CD23^–, and P1 cells showed that 78% of the variance was explained by the first 2 axes (rPC1 and rPC2). We traced 2 paths from day 4^lo cells (the common progenitor) to P2/CD23⁺ and P2/CD23^– cells (Figure 1B), which were also shown by a SLICER³²  trajectory (supplemental Figure 1A). After placing both samples and genes on the RNA-seq PCA plot, the positions of the cell populations and of B-cell and PC identity genes²⁵  led us to assign the rPC1 axis to PB differentiation (Figure 1C). Indeed, PC identity genes aggregated on the positive side of the rPC1 axis (corresponding to P2/CD23^– and P1 populations), and B-cell identity genes aggregated on the negative side (associated with day 4^lo and P2/CD23⁺ cells; Figure 1D). A gene set enrichment analysis (GSEA) of the ranked genes confirmed the enrichment of PC and NBC signatures at the positive and negative ends of the rPC1 axis, respectively (Figure 1E-F). A comparative GSEA of the 2 P2 populations showed that P2/CD23⁺ cells were enriched in gene signatures overexpressed in activated B cells from Irf4²¹  and Blimp1³³  KO mice (Figure 1G).

We next performed ATAC-seq on day 4^lo, P2/CD23⁺, P2/CD23^–, and P1 cell populations. As expected, a comparison with the Gm12878 lymphoblast cell line chromHMM track³⁴  showed that most ATAC-seq regions were active promoters and enhancers with a distribution that differed significantly from the genome as a whole (supplemental Figure 1A). These regions overlapped with open chromatin regions recently identified in mice³⁵  and with enhancers linked to promoters (FANTOM5 database³⁶ ). To identify potential regulatory regions, accessible regions were associated with genes (see “Methods”). The regions associated with PRDM1 and IRF4 (Figure 2A; supplemental Figure 1B) overlapped with binding sites for BACH2 and BCL6^37-40  and with enhancers and super-enhancers found in PBs²⁵  and in a multiple myeloma cell line.^41,42  The ATAC-seq read densities in our cell populations and unpublished data on NBCs confirmed that regions associated with genes at the positive end of the rPC1 axis opened up progressively during differentiation, whereas regions associated with genes at the negative end of the rPC1 axis tended to close (Figure 2B). In P1 cells, 5-hydroxymethylcytosine read densities increased for gene bodies at the positive end of the rPC1 axis and for the associated ATAC-seq regions.²⁵  This suggests that rPC1-associated ATAC-seq regions were enhancers activated during terminal B-cell differentiation (supplemental Figure 1C).

A DESeq2²⁸  analysis of the ATAC-seq data revealed 2658 DO regions. Unsupervised hierarchical clustering defined 6 clusters that were specific for 1 or 2 cell populations (C1-C6; Figure 2C; supplemental Figure 2A-C). A transcription factor binding site (TFBS) enrichment analysis using MEME-AME⁴³  and HOMER⁴⁴  analysis tools revealed that the potential regulators differed for each cluster consistently with the progression of B-cell differentiation (Figure 2D; supplemental Figure 2E). For example, C1 (associated with day 4^lo cells) was enriched in TFBSs for SPIB, PU.1, and NF-kB, in keeping with an activated B-cell state, whereas C6 was enriched in IRF4 binding sites, in line with the plasmablastic nature of P1 cells. We observed a weak enrichment in BACH2 binding sites (mainly in C2 and C3) and no significant enrichment in PAX5 binding sites, perhaps because we focused on late-stage differentiation.^8,45  IRF4 and PU.1:IRF8 motifs were strongly enriched in C5 and C6, fitting with the regulatory balance between IRF8 and IRF4.^46,47  PRDM1 TFBSs were significantly enriched in C6 only and at a lower level than IRF4 TFBSs, confirming that IRF4 is required earlier than PRDM1 in terminal B-cell differentiation.²¹  Interestingly, STAT6 binding sites were highly enriched in C2 and C3, associated respectively with the P2/CD23⁺ population and both P2 populations, whereas these sites were lost in C4, C5, and C6, corresponding to later differentiation states (Figure 2D).

Genes were associated with regions in each cluster, and annotation with GREAT⁴⁸  and enrichr⁴⁹  tools confirmed that each cluster corresponded to a different stage in differentiation. C6 was the largest cluster and was most strongly enriched in PC-expressed genes and genes upregulated during the differentiation of NBCs into PCs (Figure 2E; supplemental Figure 2D-E; supplemental Table 1).

To complete our investigation of the chromatin landscape and transcriptomic changes, a systematic evaluation of epigenetic factors referenced in an extensive database⁵⁰  revealed that among the 398 differentially expressed genes between P2/CD23⁺ and P2/CD23^– cells, we found only 7 of 95 histone genes (all overexpressed in P2/CD23^– cells) and 8 of 720 epigenetic factor genes (4 overexpressed in each cell population). Epigenetic factors overexpressed in P2/CD23^– cells were mostly linked to gene expression (PRDM1, GADD45A, ZMYND11), whereas those overexpressed in P2/CD23⁺ had either an ambiguous or repressing function (DEK, SATB1, ZNF516). A GSEA showed no enrichment in epigenetic processes (Gene Ontology [GO], Kyoto Encyclopedia of Genes and Genomes [KEGG], and Reactome databases) in P2/CD23⁺ cells and only limited enrichment in P2/CD23^– cells, mainly as a result of histone gene overexpression (data not shown).

Altogether, these findings confirmed that P2/CD23⁺ cells were unable to differentiate into PBs and stayed in an activated B-cell state, whereas P2/CD23^– cells committed toward the PC lineage. The fates of these 2 cell populations seemed regulated mainly by different transcription factors (TFs) (eg, STAT6 for P2/CD23⁺ cells and IRF4 for P2/CD23^– cells) and not by differential expression of epigenetic factors.

We noted above the transcriptomic similarity between P2/CD23⁺ cells and Irf4-KO activated B cells, which showed an enrichment in IRF4 TFBSs as differentiation progressed (Figures 1G and 2D). In mouse GC B cells, the quantity of Irf4 was shown to be regulated by Cbl and Cblb.²³  The expression of CBL in our in vitro model was quite stable, but that of CBLB diminished strongly in P2/CD23^– and P1 cells whereas IRF4 expression was induced (Figure 3A). The inability of P2/CD23⁺ cells to differentiate might therefore be a result of sustained expression of CBLB and a reduction in IRF4 protein levels. These differences between cell populations in gene expression, including BLIMP1/PRDM1, were confirmed by qRT-PCR (supplemental Figure 3A). In a western blot analysis of the cell populations (which agreed with the mRNA data), CBL expression was relatively stable throughout differentiation. We observed a negative correlation between IRF4 expression and CBLB expression in our cell populations, especially for cells collected beyond day 4 (Figure 3B). We first assessed IRF4 protein half-life and observed stable levels after several hours of cycloheximide treatment in primary cells (Figure 3C) and in the OCI-LY7 cell line (supplemental Figure 3B). IRF4 protein levels did not increase after proteasome inhibition (Figure 3D), and protein ubiquitination assays revealed very low levels of ubiquitinated forms of IRF4 in day 4 activated B cells 8 hours after adding the second stimulation (Figure 3E). This latter effect was observed at day 6 in P2/CD23^– and P1 cells (expressing very low levels of CBLB) rather than in P2/CD23⁺ cells (expressing low levels of IRF4 at this point; supplemental Figure 3C). No physical interaction of CBLB and IRF4²³  could be shown in our system after endogenous IRF4 immunoprecipitation (Figure 3F). After short interfering RNA downregulation of CBLB in day 4^lo cells, there was no increase in IRF4 protein levels in P2/CD23⁺ cells (data not shown) and only a slight increase in the proportion of P1 cells on day 6 (supplemental Figure 3D). The accessibility of an ATAC-seq region linked to the CBLB promoter³⁶  was correlated with CBLB expression. Furthermore, recent EZH2 and H3K27me3 profiling data⁵¹  from a similar in vitro model⁵²  highlighted the presence of the repressing histone mark H3K27me3 at the promoter and in the first intron of CBLB in pre-PBs and PBs (Figure 3G). Overall, these results suggest that although a downregulation of CBLB occurs at the last stage of human B-cell differentiation concomitant with an increase in IRF4 expression, CBL/CBLB-mediated ubiquitin degradation is not a major IRF4 regulation pathway in our system.

We hypothesized that the rPC2 axis could also explain differences and similarities between P2 cell populations. The ATAC-seq densities of regions associated with genes at the negative end of the rPC2 axis (associated with P2/CD23⁺ cells) showed a specific increase in this population (Figure 4A). Gene ranking and a GSEA of the negative end of the rPC2 axis revealed an enrichment in IL-4 response genes^53,54  (Figure 4B-C). IL-4R target genes (FCER2, IL4R, and STAT6) were expressed more in P2/CD23⁺ than in P2/CD23^– cells (Figure 1C; supplemental Figure 4A). Hence, the split in the fate of P2 cells may be the result of a differential response to IL-4 (added on day 4 in our in vitro system).

When positioned on the RNA-seq PCA plot, the genes from a B-cell–specific IL-4 response signature⁵⁴  were predominantly located in the P2/CD23⁺ quadrant, and few genes were close to the P2/CD23^– and P1 populations (Figure 4D, top). The positions of the genes were consistent with their expression levels in these cell populations (Figure 4D, middle) and with the accessibility of ATAC-seq regions associated with both sets of genes (Figure 4D, bottom). When day 4^lo, P2/CD23⁺, and P2/CD23^– populations were compared for IL-4 response signatures by GSEA, P2/CD23⁺ showed the greatest enrichment and day 4^lo cells showed the lowest enrichment (supplemental Figure 4). Interestingly, the IL-4R signaling in B lymphocytes signature containing signaling pathway members only, showed that the pathway was completely silenced in P2/CD23^– cells but not in day 4^lo cells (supplemental Figure 4B). Accordingly, we observed multimodal downregulation of the IL-4/STAT6 pathway in P2/CD23^– and P1 cells, with decreases in STAT6 and IL4R mRNA expression (Figure 5A), STAT6 and IL-4R protein levels (Figure 5B), and STAT6 phosphorylation (Figure 5C). These data are consistent with the changes in chromatin accessibility and the loss of STAT6 TFBS enrichment in C4-C6 clusters (Figure 2D). Overall, these results indicate that after an early response to IL-4 (see “D4lo⁺ 30 min IL-4” in Figure 5C), downregulation of the IL-4 signaling pathway is required in P2/CD23^– cells for the last step in PB commitment. Conversely, a sustained IL-4 response (Figure 5A-C) maintained the P2/CD23⁺ cells in an activated state and dampened their capacity to differentiate (Figure 5D). Indeed, cell-sorted day 5 P2/CD23⁺ cells were able to generate more P1 cells in the absence of IL-4 (Figure 5D, bottom left). By using a pan-caspase inhibitor treatment (QVD-OPH), we showed that this result was independent of the increase of apoptosis observed in the presence of IL-4 (Figure 5D, bottom right).

Interestingly, the ATAC-seq region linked to the CBLB promoter (Figure 3G) contained a potential STAT6 binding site. A meta-analysis of STAT6 chromatin immunoprecipitation sequencing data^55-57  could not confirm this; however, STAT6 binding sites were found in regulatory regions around the CBLB locus, including its promoter (supplemental Figure 5).

To assess differences between committed P2/CD23^– cells and P1 cells, we identified 183 differentially expressed genes (supplemental Table 2); 35 were upregulated in P1 cells (including the PC marker CD38), and 148 were upregulated in P2/CD23^– cells. Most of the latter were B-cell identity genes (eg, BACH2, CIITA, and CBLB), as shown by their position on the RNA-seq PCA plot (Figure 6A). A GSEA showed that P2/CD23^– cells were similar to the previously described⁵²  pre-PBs and had a specific molecular phenotype (supplemental Figure 6A).

This differential expression was sustained by differences in chromatin accessibility: DO regions associated with P1 upregulated genes (including CD38) were fully open in P1 cells only, and were all in clusters C5 and C6. Conversely, regions associated with genes upregulated in P2/CD23^– cells tended to close during differentiation (Figure 6B-C) and were found in C1, C2, and C3. A TFBS analysis of ATAC-seq regions associated with genes upregulated in either of the cell populations highlighted different families of TFs for P2/CD23^– and P1 cells (supplemental Figure 6B). The only common TFBS was PU.1:IRF8, an important regulator of terminal B-cell differentiation.^46,47  A differential TFBS analysis showed an enrichment of thyroid hormone receptor alpha motifs in P1 cells and of several motifs in P2/CD23^– cells, including 1 corresponding to TBX21, consistent with TBX21 overexpression in these cells (supplemental Table 2; supplemental Figure 6C).

One hallmark of PB differentiation is immunoglobulin production. We therefore looked at ATAC-seq regions within the immunoglobulin heavy chain coding locus. The region with the greatest fold-change in accessibility was located at the IGHM locus and belongs to the P1-specific cluster C6. Its accessibility differed significantly when comparing P1 and P2/CD23^– cells and was greater in P2/CD23^– cells than in P2/CD23⁺ cells (Figure 6D). The same dynamics were observed at the IGHA1 promoter and at the hs3 enhancer of the 3′ RR2 regulatory region⁵⁸  (Figure 6D-E). Overall, these data showed that P2/CD23^– cells were still imprinted by their former activated B-cell identity and were precursors of PBs.⁵² 

In late B-cell differentiation, a cell can become a post-activated B cell or produce 2 antibody-secreting cells. We showed previously²⁵  that B cells that had committed to becoming PBs downregulate expression of the CD23 cell surface marker. This finding enabled us to characterize the split in the cell fate by comparing P2/CD23⁺ and P2/CD23^– populations: CD23^–downregulation reflected the necessary inhibition of the IL-4/STAT6 signaling, and P2/CD23^– cells increased IRF4 mRNA and protein expression, which promoted their commitment to the PB lineage.^21-23 

Our chromatin accessibility analysis suggested that molecular modifications sustained the observed transcriptional changes and were not related to major changes in the expression of epigenetic factors. This is consistent with the fact that the activity of epigenetic factors important for B-cell differentiation, such as EZH2,⁵¹  is controlled by posttranslational modifications.⁵⁹  The TFBS analysis in the 6 clusters of DO regions indicated a coherent dynamic genomic landscape throughout B-cell differentiation. Interestingly, 2 key identity genes (BACH2 and PRDM1) were involved, both of which act on opposite sides of the same axis. This finding is in agreement with our previous results showing that BACH2 acts as a lock on the differentiation of B cells into PCs but is released early by IL-2–driven mechanisms.⁸  This release is stochastic, however, and depends on the output of signal integration. The P2/CD23⁺ cells may not have been stimulated above a certain threshold; hence, they maintained high levels of BACH2 expression and could not differentiate. In accordance with these findings, the transcriptome of the P2/CD23⁺ cells was similar to that seen in Irf4-KO or Prdm1-KO murine B cells, which are unable to differentiate.^21,33 

IL-4 has been described as a key cytokine in B-cell differentiation, with roles in the early extrafollicular production of short-lived PBs and in promotion of the GC B-cell identity.^10-13  We found that prolonged IL-4 signaling in P2/CD23⁺ cells impaired PC differentiation, in line with previous observations that IL-4 can both promote and inhibit B-cell differentiation.^60,61  In fact, IL-4 is able to induce both BCL6 and IRF4. Both factors promote the GC B-cell identity in a pre-GC context but later act antagonistically; BCL6 maintains the GC reaction, whereas IRF4 induces B-cell egress from the GC.^10-13,21  Therefore, just as a transient decrease in Tfh help is necessary to commit GC B-cell precursors to their GC fate,⁶²  a drop in IL-4 signaling in committed cells may be required for them to become a PB. Overall, the emergence of 2 cell populations with highly divergent fates results from the differential response to Tfh cytokines and signal integration linked to the heterogeneity of postactivated B cells, which is detectable very early in the differentiation process, as demonstrated with IL-2 priming.⁸ 

Our second major finding concerns the differential expression of the CBLB ubiquitin ligase in P2/CD23⁺ vs P2/CD23^– populations, and its impact on IRF4 protein levels. Li et al²³  showed in mice that Cbl/Cblb promotes Irf4 degradation in low-affinity GC B cells. Strong CD40 and BCR signals trigger the degradation of Cbl/Cblb, leading to elevated Irf4 levels and the egress of PBs from the GC. In human primary B cells, we observed that CBL mRNA and protein levels were stable, IRF4 and CBLB expression levels were inversely correlated, and IRF4 protein levels were quite stable. However, contrary to the mouse study, ubiquitinated forms of IRF4 and IRF4/CBLB interactions were very hard to prove in our system (Figure 3; supplemental Figure 3). Moreover, we could not detect any increase in IRF4 levels and only a slight effect on PB numbers after CBLB short interfering RNA inhibition. In our system, it is therefore likely that ubiquitination is not the primary regulating pathway of levels of IRF4 (a very stable protein; Figure 3C-D), which may be more related to an increase in IRF4 transcription levels between P2/CD23⁺ and P2/CD23^– cells. This discrepancy with the mouse GC B-cell model could be the result of either species differences in molecular mechanisms or the fact that our system does not faithfully model GC B cells. Given these differences, it may be interesting to test in humans the role of CBL/CBLB on antigen processing recently described in mice.⁶³  Interestingly, complete downregulation of CBLB was achieved only in P1 cells but not in P2/CD23^– cells. This was consistent with our findings indicating that P2/CD23^– cells are similar to pre-PB cells described in vivo and characterized by a specific molecular signature; hence, these cells are 1 step away from becoming PBs.^25,52 

Our comparative TFBS analysis of ATAC-seq regions in P2/CD23^– vs P1 populations raised 2 points warranting further investigation. First, the only differential TF motif identified in the P1 cells was related to the thyroid hormone receptor. It was shown that thyroid hormone regulates plasma cell differentiation both in vitro⁶⁴  and in vivo⁶⁵ ; however, the timing and the underlying molecular mechanisms were not investigated. Second, we identified an enrichment in TBX21 (also known as T-bet) motifs and TBX21 overexpression specifically in P2/CD23^– cells. Interestingly, CD23^–/T-bet⁺ B-cell subsets were reported in autoimmunity, chronic infections, and aging, both in mice and in humans.^66-72  These CD21^–/lo/T-bet⁺ atypical B cells were described to be or to give rise to pre-ASCs^69,70,72  in agreement with the pre-PB nature of the P2/CD23^– cells. Interestingly, several genes described in these T-bet⁺ B cells, such as TLR7 or FCRL5 were also upregulated in P2/CD23^– cells, whereas the most common co-marker of these B-cell subsets, CD11c (encoded by ITGAX), was not (supplemental Figure 6C). Hence, an investigation of the heterogeneity of the P2/CD23^– cell population is required, especially at the single-cell level. Our observation of IL-4/STAT6 pathway extinction in the final step of PB commitment is in line with IL-4 pathway extinction,⁶⁶  Antagonism by IL-4 of IL-21–driven T-bet upregulation⁶⁷  and the repressive effect of IL-4 on pre-ASC differentiation of T-bet⁺ B-cell subsets^69,72  suggest a common requirement of IL-4 signaling extinction for further ASC differentiation of these different B-cell subsets.

Our data shed new light on the mechanisms involved in B-cell fate (Figure 7). For CD23^– cells committed to differentiation into PCs, the concomitant extinction of IL-4/STAT6 signaling and increase in IRF4 expression is essential. Both pathways act in opposite manners on PC differentiation. These findings are relevant for follicular lymphomas, known to be IL-4 dependent and to present modifications of the IL-4/STAT6 pathway.

The authors thank David Fraser at Biotech Communication for editing services.

This work was funded by an internal grant from the Hematology Laboratory, Pôle de Biologie, Centre Hospitalier Universitaire de Rennes, Rennes, France, and by a grant from Association Pour la Recherche Contre le Cancer (PJA 20181207839). A.P. received fellowships from Région Bretagne and from Ligue Nationale de Lutte Contre le Cancer.

Cell sorting was performed at the Biosit Flow Cytometry and Cell Sorting Facility CytomeTRI (University of Rennes 1, Rennes, France).

Contribution: A.P. performed cell culture, molecular, and protein experiments, designed certain experiments, and wrote the first draft of the manuscript; F.C. prepared ATAC-seq libraries, analyzed and integrated omics data, created tools for analysis, and wrote and revised the manuscript; G.C. performed cell culture, cytometry, protein, and cell sorting experiments and revised the manuscript; M.H. helped with experiments needed for revision; F.D. helped with protein analyzes; T.F. designed the cell differentiation model, performed RNA-seq Ingenuity Pathways Analysis, supervised the project, and wrote and revised the manuscript; and all authors read and approved the revised manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Thierry Fest, INSERM U1236, Faculté de Médecine, University Rennes 1, 2 Avenue du Pr Léon Bernard, CS 34317, F-35043 Rennes Cedex, France; e-mail: thierry.fest@univ-rennes1.fr.