Plasma cell differentiation is coupled to division-dependent DNA hypomethylation and gene regulation

Abstract

The epigenetic processes that regulate antibody secreting plasma cells are not well understood. Here, analysis of plasma cell differentiation revealed DNA hypomethylation of 10% of CpG loci that were overrepresented at enhancers. Inhibition of DNA methylation enhanced plasma cell commitment in a cell division-dependent manner. Examination of in vivo differentiating B cells stratified by cell division revealed a 5-fold increase in mRNA transcription coupled to DNA hypomethylation. Demethylation occurred first at binding motifs of NF-κB and AP-1 and later at those for IRF and Oct-2, and were coincident with activation and differentiation gene expression programs. These data provide mechanistic insight into the cell-division coupled transcriptional and epigenetic reprogramming and suggest DNA hypomethylation reflects the cis-regulatory history of plasma cell differentiation.

Resting naïve B cells rarely undergo mitosis¹, do not secrete immunoglobulins (Ig) and express only basal levels of transcripts². Upon activation through the B cell receptor or Toll-like receptors, B cells rapidly divide³ and differentiate into mitotically cycling plasmablasts, post-mitotic terminally differentiated plasma cells or memory B cells^4,5. Plasmablasts and plasma cells actively secrete Ig whereas memory B cells do not, but have the potential to rapidly differentiate upon subsequent antigen exposure. Despite the extensive study of B cell and plasma cell transcriptional programming^3,6, many mechanisms that govern differentiation remain unknown. While B cell differentiation requires cell division^4,5, the number of divisions does not solely determine plasma cell fate^5,7. This has led to a stochastic model of differentiation that is highly variable for individual B cells but leads to balanced progeny fates at a population level^5,7,8. One mechanism that could contribute to such cellular heterogeneity is epigenetic variability. Epigenetic marks, such as DNA methylation or histone modification, can enhance or repress gene transcription and are mitotically heritable^9,10. DNA methylation is necessary for hematopoietic stem cell renewal, restricts myeloid differentiation and allows for B cell commitment¹¹. During a B cell immune response, DNA methylation was remodeled in germinal center and memory B cells and plasma cells^12–14. However, the breadth, timing and function of these epigenetic changes in response to an in vivo stimulus are incompletely understood.

To gain insight into the epigenetic mechanisms that govern B cell differentiation, we used in vivo models to determine the direct relationships between DNA methylation, gene expression and cell division. We found that B cell differentiation was associated with targeted DNA hypomethylation and increased gene expression. Cell division was accompanied by a hierarchy of DNA hypomethylation events at cis-regulatory elements that corresponded with division-specific expression. Our results define a step-wise process of division-coupled epigenomic remodeling that allows B cells to adopt a new transcriptional program and cell fate.

Results

B cell differentiation is coupled to unique transcriptional states

We used an inducible in vivo model of B cell differentiation to examine the molecular events that could be traced to a defined stimulus. C57BL/6J mice challenged with the mitogen lipopolysaccharide (LPS) i.v. exhibited splenomegaly and a three-fold expansion of splenic B220⁺ B cells, while activated B220⁺GL7⁺ B cells increased from 2% to 35% of splenocytes three days post-challenge as compared to naïve mice (Supplementary Fig. 1a-c). Extrapolation of these data indicated that there were approximately 120 million new B cells in the splenic compartment (Supplementary Fig. 1d-f). Analysis of CD138⁺ differentiating B cells showed an admixture of cells with intermediate to low expression of B220 (Fig. 1a). B220 expression on CD138⁺ plasma cells is a marker of rapid cellular turnover in the spleen¹⁵ and bone marrow¹⁶, whereas B220^loCD138⁺ plasma cells represent a post-mitotic population¹⁵. Both B220^int and B220^lo CD138⁺ plasma cells were strongly induced three days post-challenge with LPS (Fig. 1a), and are herein referred to as plasmablasts (PB) and plasma cells (PC), respectively.

Figure 1.

B-cell differentiation is coupled to unique transcriptional states. (a) Flow cytometry showing splenic B220 and CD138 expression in naïve and LPS-challenged mice on day 3 (left). Quantitation of B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells (right). (b) Post sort purity of B cells, plasmablasts (PB) and plasma cells (PC). (c) Hierarchical clustering of expression data at 16,181 genes in the populations shown above. (d) Principle components analysis of expression data shown in c. (e) Scatterplot of expression changes in B220^intCD138⁺ plasmablasts (PB) and B220^loCD138⁺ plasma cells (PC) as compared to B cells from naïve mice. Differentially expressed genes (Supplementary Table 1) are shown in burgundy (plasmablasts), gold (plasma cells), or black (both). Dashed gray lines indicate expression changes of twofold. (f) Gene set enrichment analysis of expression changes in plasmablasts and plasma cells for genes regulated in human plasma cells¹⁷ (left, FDR <0.05) and the Reactome pathway ‘Mitotic M-M/G1 phases’ (FDR <0.01, plasmablasts only). Enrichment score is shown on top for both plasmablasts and plasma cells. Below is the overlap of genes from each gene set with the ordered expression changes of plasmablasts and plasma cells relative to B cells. *P <0.001 (two-tailed t-test). Data are representative from two experiments and 15 mice (a, mean and s.d.) or one experiment with three mice where two B cells, three plasmablasts, and two plasma cells were analyzed (c-f).

To characterize transcriptional stages in B cell differentiation, we isolated RNA from splenic B220⁺ B cells from naïve C57BL/6J mice and B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells from mice three days post-LPS challenge (Fig. 1b) and analyzed it using Illumina MouseRef-8 BeadArrays. Hierarchical clustering of all expression data stratified samples by cell type (Fig. 1c), indicating each cell type was transcriptionally unique. This observation was supported by principle components analysis (PCA), which stratified each biological replicate by cell type (Fig. 1d). Compared to naïve B220⁺ B cells, B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells up-regulated 937 and 567 genes, and down-regulated 1,016 and 501 genes, respectively (Supplementary Table 1). Despite the distinct gene expression programs of each cell type, 544 genes exhibited similar regulation in plasmablasts and plasma cells as compared to naïve B220⁺ B cells (Fig. 1e). Gene ontology annotation of genes commonly downregulated in both B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells encoded products involved in hematopoiesis (e.g. Hhex), immune system development (Bcl2, Irf8) and antigen presentation (Ciita, H2-Aa, H2-Ab1, H2-Eb1). Genes upregulated uniquely in B220^intCD138⁺ plasmablasts were associated with mitosis and cellular division (Supplementary Table 2). These observations were supported by gene set enrichment analysis (GSEA), which indicated B220^intCD138⁺ plasmablast and B220^loCD138⁺ plasma cell gene expression changes were similar to those previously reported in humans¹⁷ and identified mitotic pathways as selectively enriched in B220^intCD138⁺ plasmablasts (Fig. 1f, Supplementary Table 3). These data show LPS induces a robust B cell differentiation characterized by transcriptionally distinct B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells.

Plasmablasts and plasma cells undergo targeted DNA hypomethylation

To directly measure epigenetic changes during B cell differentiation, we extracted DNA from naïve splenic B220⁺ B cells, and day 3 LPS-induced B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells and the DNA methylation state was determined using reduced representation bisulfite sequencing (RRBS)¹⁸. In total, 911,004 CpGs were analyzed (Supplementary Table 4). Hierarchical clustering of the DNA methylation data separated B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells from naïve B220⁺ B cells (Fig. 2a), indicating that DNA methylation may play a significant role during B cell differentiation. PCA of DNA methylation data indicated B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells were distinct from B220⁺ B cells (Fig. 2b), suggesting a common epigenetic change in plasmablasts and plasma cells. The distribution of highly methylated loci in naïve B220⁺ B cells showed a substantial shift to more intermediate methylation in B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells (Fig. 2c). The average DNA methylation level was decreased in B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells as compared to naïve B220⁺ B cells (Fig. 2d), suggesting that B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells undergo global DNA hypomethylation.

Figure 2.

B220^intCD138⁺ plasmablasts (PBs) and B220^loCD138⁺ plasma cells (PCs) undergo DNA hypomethylation. (a) Hierarchically clustered heatmap of DNA methylation (DNAme) data for naïve splenic B220⁺ B cells and LPS-induced B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells. (b) Principle components analysis of DNAme data. (c) Probability distribution of DNA methylation. (d) Average DNA methylation for B cells, plasmablasts and plasma cells showing a genome-wide decrease. (e) Barplot of differentially methylated loci (DML). Methylated loci (hyper) are shown on a positive scale and demethylated loci (hypo) are shown as negative values. (f) Scatter plot of DNA methylation changes in plasmablasts and plasma cells. Differentially methylated loci are colored in burgundy (PB), gold (PC) and black (both). (g) Venn diagram of methylated (left; mDML) and demethylated (right; dDML) showing overlap of plasmablast and plasma cell differentially methylated loci. (h) Bar plot showing the percent of differentially methylated loci clustered into contiguous regions for plasmablasts and plasma cells. The shade denotes the number of CpGs in the contiguous region, where contiguous is defined relative to assay coverage. The expected (e) percentiles are shown to the left of the observed (o). (k) Genome plots of Irf4 and Arid3a showing demethylated regions. Coverage is indicated by black bars and plasmablast and plasma cell differentially methylated loci are shown. Average DNA methylation for B cells, PBs, and PCs are shown below. *P <0.05, **P <0.01, ***P <0.001 (Welch’s t-test (d), or permutation testing (h)). Data represent three biological replicates (d; mean and s.d.).

Differentially methylated loci (DML) were identified using a statistical model that accounts for sequencing depth and biological variation¹⁹. This identified 17,628 and 99,265 differentially methylated loci in B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells, respectively, as compared to naïve B220⁺ B cells (Supplementary Table 5). In both B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells, more than 99.7% of the DNA methylation changes were demethylated (Fig. 2e). RRBS data were validated at CpG loci near Irf4, Il10, Tnfrsf13b, Tnfrsf13c and Egr3 – genes implicated in plasma cell biology^20–24 – using combined bisulfite restriction analysis²⁵ in splenic naïve B220⁺ B cells and LPS-induced B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells (Supplementary Fig. 2). Moreover, DNA methylation in bone marrow B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells was also reduced at CpG loci near Irf4, Prdm1, Il10 and Cd86 genes as compared to splenic B220⁺ B cells (Supplementary Fig. 3), suggesting that these DNA methylation changes represent a common epigenetic program of plasma cells.

Although a large number of CpGs coordinately lost DNA methylation in both LPS-induced splenic B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells as compared to naïve splenic B220⁺ B cells, 90% of the CpGs interrogated did not significantly change their methylation state (Fig 2f), suggesting these changes were specific. In addition, there was a substantial overlap of differentially methylated loci between B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells (Fig. 2g). Among these, demethylated loci included CpGs around key genes expressed in plasma cells, such as Irf4, Prdm1 and Xbp1, while methylated loci included important regulators of B cell differentiation such as Irf8 and Egr3 (Fig. 2g). Finally, differentially methylated loci were clustered into contiguous regions; that is, more than 30% of plasmablast and 50% of plasma cell demethylated loci were adjacent to at least one other demethylated loci relative to assay coverage (Fig. 2h). Examples of such contiguous regions occurred at Irf4, a transcription factor necessary for B cell differentiation^26,27, and Arid3a, a gene known to regulate Ig heavy chain transcription²⁸ (Fig. 2i). Other examples included regions around Prdm1, which encodes Blimp-1, a master regulator of plasma cell fate²⁹, and Cflar, a gene involved in the regulation of caspase-induced apoptosis (data not shown). Together, these data identified a targeted DNA hypomethylation that occurs during B cell differentiation, impacting ~10% of CpG loci assayed.

Demethylated regions occur at B cell enhancers

To test whether demethylated regions in B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells were preferentially occurring at enhancer regions, the overlap of demethylated loci with active (H3K4^me1+H3K27^ac+H3K4^me3−) and poised enhancer regions (H3K4^me1+H3K27^ac−H3K4^me3−)³⁰ were assessed using an odds ratio. Here enhancers were identified in splenic B220⁺ B cells and splenocytes, the CH12 lymphoma cell line and thymus, testis and brain whole tissue homogenates from C57B/6J mice using data generated by the ENCODE consortium and others^31,32. We observed that demethylated loci were generally enriched at enhancers and this was most pronounced in active enhancers of B cells, splenocytes and CH12 cells (Fig. 3a), suggesting that these regulatory regions are utilized during plasma cell differentiation.

Figure 3.

Differentially methylated loci (DML) preferentially occur at B cell enhancers and near motifs of transcription factors required for B cell differentiation. (a) Overlap of B220^intCD138⁺ plasmablast (PB) and B220^loCD138⁺ plasma cell (PC) differentially methylated loci with active and poised enhancers. (b) Clustering of transcription factor motifs enriched near (≤50 bp) differentially methylated loci in plasmablasts and plasma cells. Motifs are clustered on the similarity of their consensus binding motif. Transcription factor families are denoted in black boxes (IRF: Inteferon Regulatory Factor; RHD: Rel Homology Domain; NFAT: Nuclear Factor of Activated T-cells; POU: Pituitary Octamer Unc-86 transcription factor; bZIP: basic Leucine Zipper Domain; bHLH: basic Helix-Loop-Helix). The HOMER³³ motif name is labelled on the right. Significance (FDR ≤0.05) was determined from HOMER³³ software. Enhancers were determined from publically available data^31,32 where active enhancers were defined as H3K4^me1+H3K27^ac+H3K4^me3− and poised enhacers as H3K4^me1+H3K27^ac−H3K4^me3-30. Data represent three biological replicates.

To determine which transcription factors may be active during B cell differentiation, we searched regions within 50 bp of demethylated loci for enriched transcription factor motifs relative to assay coverage using HOMER³³. Transcription factor motifs enriched in either B220^intCD138⁺ plasmablast and B220^loCD138⁺ plasma cell demethylated loci were clustered by the similarity of their consensus binding motif (Fig. 3b), showing that interferon regulatory factors (IRF), POU homeobox domain (POU), and basic leucine zipper domain (bZIP) motifs were overrepresented in both B220^intCD138⁺ plasmablast and B220^loCD138⁺ plasma cell demethylated regions, whereas Rel homology domain (RHD), NFATc1 and E2A motifs were only found significant in B220^loCD138⁺ plasma cell demethylated regions. This observation is in agreement with the fact that transcription factors that bind these motifs are critical determinants of B cell fate^26,34–36. These data suggest that DNA demethylation in B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells reflect transcription factor activity and enhancer usage.

DNA demethylation and differentiation is coupled to cell division

To address the role of DNA methylation during plasma cell differentiation, splenic B220⁺ B cells were labeled with carboxyfluorescein succinimidyl ester (CFSE) or cell trace violet (CTV) to track cellular division and cultured for three days in the presence of LPS, IL-2 and IL-5³⁷ and the DNA methylation inhibitor 5-azacytidine (5-azaC). Flow cytometry analysis showed splenocytes cultured with 5-azaC had a higher frequency of differentiation into CD138⁺ plasma cells compared to untreated cultures, and the effect of 5-azaC was dose-dependent (Fig. 4a,b). Moreover, 5-azaC induced plasma cell differentiation after fewer cell divisions compared to untreated cultures (Fig. 4c).

Figure 4.

Inhibition of DNA methylation facilitates plasma cell differentiation. (a) Flow cytometry analysis of CFSE staining, CD138 expression and viability exclusion dye on B cells differentiated for three days ex vivo with LPS, IL-2, IL-5 and treated with increasing amounts of 5-azacytidine (5-azaC). (b) Frequency of CD138⁺ cells by 5-azaC treatment. (c) Frequency of CD138⁺ cells by cell division and 5-azaC treatment. *P <0.001 (two-tailed t-test (b) or ANOVA (c)). Data are from three independent experiments with 1-5 mice per experiment pooled and performed in triplicate (b, c, mean and s.d.).

To determine how DNA methylation and gene expression changes occur in relation to cell division during B cell differentiation in vivo, we used an adoptive transfer model. Splenic B220⁺ B cells isolated from CD45.1 C57BL/6J mice were labeled with CFSE or CTV and transferred i.v. into CD45.2 μMT mice, which express defective membrane-truncated IgM heavy chains and as such are B cell-deficient³⁸. Hosts were challenged with LPS one day after adoptive transfer and three days post-challenge we found CD45.1⁺ donor B cells in the spleens of recipient mice (Supplementary Fig. 4a,b). CFSE dilution indicated that B cells transferred into LPS-challenged mice underwent 0 to >8 rounds of division, whereas B cells transferred into mice that did not receive LPS rarely divided (Supplementary Fig. 4c,d). Three days post-LPS challenge donor CD45.1⁺B220^intCD138⁺ plasmablasts and CD45.1⁺B220^intCD138⁺ plasma cells were only observed in the spleen of mice that received LPS (Fig. 5a). Donor CD45.1⁺B220^loCD138⁺ plasma cells had divided more than CD45.1⁺B220^intCD138⁺ plasmablasts‘ and no differentiation was observed prior to division six (Fig. 5b). B220 expression on CD45.1⁺ donor B cells was only lost after many rounds of division in LPS-challenged mice (Fig. 5c). These cells were viable and a large proportion of them expressed CD138 (Supplementary Fig. 4e-h). GL7 expression on donor CD45.1⁺ B cells was increased without division and progressively upregulated at later divisions in LPS-challenged mice (Fig. 5d). In agreement with previous ex vivo experiments^4,5, these data link plasma cell differentiation to cell division and suggest that the observed changes in gene expression and DNA hypomethylation may be related to the multiple rounds of mitotic division that plasmablasts and plasma cells undergo in vivo.

Figure 5.

Transcriptional amplification and DNA hypomethylation coincide with cellular division. (a) B220 and CD138 expression on transferred splenic naïve CD45.1⁺B220⁺ B cells showing LPS-induced CD138 expression (left). Quantification of B220^intCD138⁺ plasmablasts (PBs; center) and B220^loCD138⁺ plasma cells (PCs; right). (b) CFSE dilution on B220⁺ B cells, B220^intCD138⁺ plasmablasts, and B220^loCD138⁺ plasma cells (left). Quantification of B220^intCD138⁺ plasmablasts and B220^loCD138⁺ plasma cells by division (right). (c) B220 expression by division for control and LPS challenged mice (left). Quantitation of B220⁺ cells by division (right). (d) GL7 expression by division for control and LPS challenged mice (left). Quantitation of GL7⁺ cells by division (right). (e) Cell trace violet (CTV) staining and CD138 expression for CD45.1⁺ cells (top) and post sort purity (bottom) for the indicated populations. (f) Heatmaps of transcript expression (top) and DNA methylation (DNAme) (bottom) for populations shown in e. (g) Quantification of average mRNAs per cell (top) and DNAme (bottom) measured across 1,639,598 CpGs assessed. (h) Probability density for mRNA expression (top) and DNAme (bottom). The dashed gray line indicates the 90% detection level. (i) Principle component analysis for mRNA expression (top) and DNAme (bottom). *P <0.05, **P <0.01 and ***P <0.001 (two-tailed t-test (a, g)). Data are from two experiments with 4 and 6 mice per experiment (a-e) or one experiment performed in biological duplicate (f-i) (a-d, g; mean and s.d.).

To determine the relationship between DNA methylation and gene expression at successive stages of differentiation, we sorted cells based on CTV dilution representing divisions 0, 1, 3, 5, as well as cells that divided at least 8 times and were either CD138⁻ or CD138⁺ (referred to as ‘8⁻‘ and ‘8⁺’, respectively) (Fig. 5e) and RNA-seq was performed using external RNA controls consortium (ERCC) synthetic mRNA spike-in controls to quantitate the average number of mRNAs on a per cell basis (see Methods). In addition, dual restriction enzyme (MspI, TaqI) RRBS protocol was performed resulting in coverage at 1,639,598 CpGs (Supplementary Table 4). The resulting datasets were validated by direct comparison to data generated by RT-qPCR and a qPCR DNA methylation assay on independently isolated cells of the same division and CD138 expression (Supplementary Fig. 5). Heatmaps of mRNA expression and DNA methylation at distinct stages of B cell division indicated higher expression and lower DNA methylation at divisions 8⁻ and 8⁺ (Fig. 5f). Divisions 5, 8⁻ and 8⁺ had on average more mRNAs per cell as compared to division 0 and this corresponded with a significant decrease in average DNA methylation at division 8⁻ and 8⁺ (Fig. 5g). The distribution of mRNA in divisions 5, 8⁻, and 8⁺ identified a uniform increase across a wide range of expression levels as compared to undivided cells (Fig. 5h), suggesting that the increased mRNA expression was not solely attributable to a few highly expressed mRNAs, such as those encoding immunoglobulins. In addition, division 8⁻ and 8⁺ had reduced levels of highly methylated loci as compared to divisions 0, 1, 3, and 5 (Fig. 5h). PCA indicated that the greatest variation in gene expression was between division 8⁺ and all others (Fig. 5i). Gradual changes in total DNA methylation were observed between division 0 and divisions 1-5, but the largest variation in DNA methylation occurred after division 8 (Fig. 5i), suggesting that some DNA methylation changes in division 8⁻ may precede expression changes. Cumulatively, these data identify a division-dependent global amplification of mRNA that corresponds with DNA hypomethylation in antigen-induced differentiating B cells.

Next, pair-wise comparisons of gene expression between division 0 and divisions 1-8⁺ indicated that gene expression was mostly upregulated as cells divided, especially at later divisions (Fig. 6a). Differentially expressed genes were determined using criteria for both relative and absolute change in expression; thus, only genes upregulated twofold more than the average increase in expression for any division were determined significant. We also performed pairwise comparisons of DNA methylation, in which differentially methylated loci were determined relative to division 0. Few DNA methylation changes were found at early divisions, and most changes occurred between division 0 and divisions 5, 8⁻ and 8⁺ (Fig. 6b). Finally, we compared the correlations between gene expression changes and DNA methylation changes at each division. We observed that 98% of differentially methylated loci associated with differentially expressed genes lost DNA methylation and gained gene expression through division 5 as compared to division 0 (Fig. 6c). Differentially methylated loci inversely correlated with gene expression in divisions 8⁻ and 8⁺ (94% and 72%, respectively) (Fig. 6c). These data suggest that gene expression changes are highly correlated with local DNA methylation changes in antigen-induced differentiating B cells.

Figure 6.

Cell division coupled changes in gene expression and DNA methylation (DNAme). (a) Scatter plot of gene expression showing changes between undivided B cells (Div 0) and distinct divisions of differentiating B cells defined in Fig. 5e (bottom). Spike-in ERCC controls are shown as bright red triangles with a solid bright red regression line. The average regression between the two comparisons is shown as a black dashed line. Differentially expressed genes are shown in color denoting the division at which the gene was determined significant. (b) Scatterplot showing DNA methylation as in a. (c) Scatter plot of mRNA fold-change by change in DNA methylation showing correlation of gene expression and DNA methylation changes. Genes or CpG loci that are both differentially expressed and differentially methylated are colored as in a. Analysis was performed on data sets derived in Fig. 5e and represent the average of two biological replicates.

Quantitation revealed that 86% of division-specific differentially expressed genes were upregulated with increasing cellular division compared to undivided B cells and only 14% of genes were downregulated, with the majority of downregulated genes occurring between division 0 and 8⁺ (Fig. 7a). Gene ontology analysis indicated that genes upregulated in divisions 5, 8⁻ and 8⁺ as compared to division 0 were involved in mitosis and metabolism, and genes downregulated in division 8⁺ versus division 0 were cell activation genes (Fig. 7b, Supplementary Table 3). GSEA also indicated a large number of gene sets associated with genes upregulated in divisions 1 through 8⁺ compared to division 0 (Fig 7c). Genes sets upregulated in divisions 1 through 8⁺ as compared to division 0 were associated with cell division, Myc targets, the proteasome, endoplasmic reticulum (ER) stress and processing of proteins, while genes sets defined by NF-κB and type I interferon signaling were downregulated in divisions 8⁻ and 8⁺ as compared to division 0 (Fig. 7c, Supplementary Table 4). Genes that were dynamically regulated across cell division included downregulation of Cxcr5 and Icosl, transient upregulation of Cd83, Fas, Tlr9, Tbx21, Aicda and upregulation of Cxcl10, Schmt2, Txndc5, Rexo2, Selk, Cxcr4, and Xbp1 (Fig. 7d). As such, gene expression was globally upregulated with increasing division, yet annotation of genes that were preferentially increased or decreased through division stages indicated that pathways important for B cell and plasma cell biology were selectively regulated.

Figure 7.

Dynamic gene expression changes correspond with a hierarchy of DNA hypomethylation. (a) Barplot of differentially expressed genes (DEGs) relative to division 0. (b) Heatmaps of gene ontology results for up and downregulated genes. Rows represent the top ontology for division-specific differentially expressed genes. Columns depict the relationship to other division differentially expressed genes. (c) Barplot of positively and negatively correlated GSEA results (left). GSEA showing upregulation of the proteosome and Myc signaling and downregulation of TNF and IFN-α pathways (right). Plots are as in Fig. 1f. (d) RNA-seq analyses of differentially expressed genes (units are average mRNA per cell). (e) Barplot of differentially methylated loci relative to division 0 with gains (hyper) in DNA methylation (DNAme) plotted above and loses (hypo) below. (f) Plot of DNA methylation at division specific differtenially methylated loci. (g) Barplot representing the fraction of differentially methylated loci that fall into contiguous blocks. The expected (e) number is shown next to the observed (o). (h) Odds ratio of overlap for division-specific demethylated loci with tissue-specific enhancers. (i) Transcription factor motifs enriched in division-specific demethylated regions as described in Fig. 3b. (j) DNA methylation differences for differentially methylated loci near transcription factor motifs in i. (l) Expression for genes that contain transcription factor motifs in i with (+) or without (-) a demethylated loci. *P <0.001 (Fisher’s exact test (b, h), permutation testing (g), ANOVA (j), and Wilcoxon-rank sum test (k)). Analysis was performed on data sets derived in Fig. 5e and represent the average of two biological replicates.

When division-specific DNA methylation changes were compared to division 0 most changes were hypomethylation events at later divisions (Fig. 7e). Demethylated CpGs continued to lose DNA methylation at later divisions (Fig. 7f), and these were organized into contiguous regions, identifying these changes as focal epigenetic events (Fig. 7g). Division-specific DNA methylation changes had a substantial overlap with enhancers in B cells, splenocytes and CH12 cells (Fig. 7h). This was most pronounced at earlier divisions, and considerably less for enhancers in thymus, testis and brain tissue (Fig. 7h). We used HOMER³³ to overlap differentially methylated loci with known transcription factor motifs, and observed that demethylated loci preferentially occurred near (≤50 bp) motifs for bZIP, IRF, MADS, POU, and RHD (Fig. 7i). Demethylated loci occurred near NF-κB (RHD) and AP-1 (bZIP) motifs starting at division 3; whereas IRF and POU family members were only enriched in demethylated loci specific to divisions 8⁻ and 8⁺ (Fig. 7i), suggesting a hierarchy of transcription factor utilization. Demethylated loci proximal to the above motifs, lost more DNA methylation than those not associated with a motif (Fig. 7j). Genes associated with these motifs that contained demethylated loci were expressed at higher levels in each division than genes that contained the motif but were not demethylated (Fig. 7k). These data suggest that transcriptional changes are being driven by an ordered demethylation of key cis-regulatory elements.

Global gene expression and average DNA methylation were inversely correlated with each other (Fig. 8a). To understand individual correlations of gene expression with DNA methylation at each division, demethylated loci were annotated to the closest gene and distinct patterns of DNA methylation and gene expression were compared using a normalized Euclidean distance metric, such that similar patterns had smaller distances. This metric was used to categorize gene expression and DNA methylation patterns into four groups using K-means clustering (Fig. 8b). Genes in clusters 1 and 4 were negatively correlated with respect to expression, DNA methylation and division – with cluster 4 being the largest and most homogeneous group (Fig. 8b). In contrast, clusters 2 and 3 displayed a progressive loss in DNA methylation with transient gene expression in divisions 1 through 8⁻ (Fig. 8b). Gene ontology annotation revealed distinct functional categories for each cluster with ER stress enriched in clusters 1 and 4 and metabolic processes exclusive to cluster 4 (Fig. 8c). Cluster 2 was associated with anatomical morphogenesis and cell adhesion, and cluster 3 with genes involved in leukocyte activation (Fig 8c, Supplementary Table 2). DNA methylation and gene expression changes characteristic for clusters 1 and 4 were found at regions near Prdm1 and Il10, important regulators of plasma cell biology^20,29, as well as C1qbp and Rexo2, genes important for mitochondrial metabolism^39,40 (Fig. 8d). Several loci near the B cell activation genes Cd80, Cd83, Cd86, and Aicda, which encodes AID, a protein required for class-switch recombination and somatic hypermutation, and Ablim1, a molecule involved in cell adhesion and recently implicated in B cell differentiation²³ fell in clusters 2 and 3 (Fig. 8d and data not shown). As such, these analyses revealed several regions where DNA methylation was lost with transient gene expression occurring between divisions 1 and 8⁻, prior to silencing at division 8⁺. Together, these results suggest that DNA methylation reflects a historical footprint of gene regulation by cell division in B cell differentiation.

Figure 8.

Gene expression correlates with cell division and DNA methylation (DNAme). (a) Scatterplot of average DNA methylation level by the sum of mRNA per cell showing a global correlation between expression and DNA methylation. (b) Heatmap of gene expresssion-DNA methylation correspondance. 11,456 Differentially methylated loci found near 2,431 differentially expressed genes were measured using a normalized Euclidean metric and organized using K-means clustering (see Methods). Average gene expression and DNA methylation for the four K-means groups (right). (c) Heatmap of the top gene ontology result for each of the four K-means groups in b. Each row represents the top ontology for a K-means group and how that ontology is enriched in other groups is depicted by the columns. (d) Gene plots for differentially expressed genes with differentially methylated loci are shown (top). Specific gene expression, DNA methylation and cell division correlations are shown for CpGs denoted in black upward facing arrowheads. Plots correspond to the left and right most CpG assessed, respectively. P-values are determined using an ANOVA (a) or Fisher’s exact test (c). Spearman’s correlation coefficient (ρ) is shown in a. Analysis was performed on data sets derived in Fig. 5e and represent two biological replicates (b, d; mean and s.d.).

Discussion

Here we provide insight into the dynamic gene expression and epigenetic changes that occur during B cell differentiation in vivo. Plasmablasts and plasma cells had distinct gene expression programs, and both underwent focal DNA methylation changes at up to 10% of their DNA methylome. More than 99% of DNA methylation changes were demethylation events. Differentiating B cells increased their global mRNA expression by more than 5-fold in a division-dependent manner. This increase was attributable to transcriptional amplification of thousands of mRNAs, and not just immunoglobulins and may be required to maintain the B cell fate program, while the cells are undergoing massive proliferation and differentiation. Pairing gene expression analysis with DNA methylation data from the same division-specific, differentiating B cells provided critical insight into the functional categories of gene expression and DNA methylation changes, and indicated that demethylation occurred at transiently expressed activation genes and highly expressed plasma cell genes. Transcriptional changes were highly correlated with cellular division and DNA demethylation of transcription factor motifs, providing a mechanism by which DNA methylation changes controls enhancer accessibility and programmatic fate of differentiating B cells.

DNA methylation studies of human steady-state plasma cells showed 60-90% DNA methylation differences were hypomethylation events compared to B cells^13,14. This difference in observed DNA hypomethylation may reflect the kinetics of the process, as the plasma cells analyzed here were less than three days old, whereas the half-life of steady-state plasma cells can range from several days to more than 100 days depending on the subpopulation^16,41,42. If this is indeed the case, then our data suggest that an abrupt and targeted loss of DNA methylation occurs during the initial stages of differentiation, but this may be followed by gradual gains in DNA methylation as plasma cells age.

Demethylated loci in differentiating B cells were overrepresented at B cell enhancers and binding motifs of transcription factors required for B cell differentiation^{21,22,26,34,35,43}. This suggests that certain B cell enhancers are utilized upon activation, and as the cells divide and differentiate they make use of novel regulatory elements. This is supported by the co-localization of transcription factor binding motifs and demethylated regions, such that motifs for NF-κB and AP-1 – which are directly induced through LPS-TLR4 signaling – were enriched starting at division 3, whereas Oct-2 and IRF motifs were only enriched in cells that divided 8 times or more. Thus, the data indicate that the DNA hypomethylation observed represents a hierarchy of cis-regulatory events during B cell differentiation. Currently, known mechanisms of DNA demethylation include enzymatic-dependent active processes^44,45 and replication-coupled passive processes. Given the rapid replication of differentiating B cells, it is likely that a passive process partially accounts for the DNA demethylation observed here. This is supported by the cell division-specific DNA methylation analyses that found the majority of changes occurred in cells that had undergone many rounds of division. Yet, DNA methylation changes were focal at regulatory regions, suggesting a targeted process. One possibility is that DNA demethylation is facilitated by the binding of transcription factors that block replication coupled DNA methyltransferase activity⁴⁶. The observed demethylation could also result from 5-hydroxymethylation of cytosines (5-hmC) in B cells, which does not have a known mechanism of maintenance through mitosis. Although there is no evidence for active demethylation during B cell differentiation, such a process cannot be ruled out.

Treatment of cells ex vivo with 5-azaC resulted in augmented plasma cell differentiation in a division-dependent manner, and suggested that loss of DNA methylation is limiting for plasma cell differentiation. This may be in contrast to observations that inhibition of DNA methylation reduces germinal center formation¹². It is possible that DNA methylation functions to extend B cell activation at the cost of plasma cell differentiation. This is consistent with greater DNA hypomethylation in plasmablasts that divided at least 8 times as compared to B cells that also divided at least 8 times. These division-coupled epigenetic changes may also help explain why the number of divisions a B cell has undergone directly corresponds with the potential of that B cell to differentiate⁵. Such epigenetic mechanisms may contribute to both the variability of the B cell response at the population level, as well as the high correlation of cell fates between sibling B cells⁷. These data do not preclude other models of differentiation, such as the asymmetric division reported in T cells⁴⁷ or a transcription factor centric model, in which transcription factors drive or inhibit plasma cell formation⁶. Rather these data provide insight into an additional and critical layer of complexity that regulates plasma cell formation. The results support a model of DNA methylation remodeling that is coincident with gene expression and reflects the cis-regulatory history of plasma cells and the epigenetic reprogramming events through cell division and differentiation.

Online Methods

Mice and LPS challenge

C57/BL6J mice between 8-12 weeks of age were used for experiments except where otherwise specified. All animals were housed by the Emory Division of Animal Resources and all protocols were approved by the Emory Institutional Animal Care and Use Committee (IACUC). For animal studies a power analysis was performed and included in our approved IACUC. Investigators were not blinded and no animals were excluded from analysis. Experiments were balanced such that similar numbers of mice were included in each group. LPS challenge was administrated intravenously using 50 μg of Salmonella LPS (Enzo Life Sciences, ALX-581-008) and mice were analyzed 3 days post-challenge. LPS-induced B cell differentiation was analyzed in two experiments with 7 and 8 mice composed of 11 females and 4 males (Fig. 1). Microarray analysis and RRBS was performed on one experiment with 3 female mice. CD45.1 mice were of strain B6.SJL-Ptprc^a Pepc^b/BoyJ from Jackson Laboratories (#002014). B cell deficient mice (μMT) were previously described³⁸ and were also obtained from Jackson Laboratories (#002288, strain B6.129S2-Ighm^tm1Cgn/J). For cell division assays, 20 × 10⁶ CTV- or CFSE-labeled splenic CD45.1⁺B220⁺ B cells were adoptively transferred into μMT hosts and allowed to rest for 18-24 hours prior to LPS challenge. Here, transferred splenic CD45.1⁺B220⁺ B cells were gender-matched to hosts and this was performed in three experiments that included 3, 6 (Fig. 5a-d) and 4 mice (Fig. 5e-i) composed of 3 males and 10 females.

Cell Isolation and flow cytometric analysis

Splenic cell suspensions were made by mechanically forcing spleens through a 40 μm filter and lysing red blood cells with ACK lysis buffer (0.15 M NH₄Cl, 10 mM KHCO₃, 0.1 mM EDTA) for 30 seconds prior to quenching the reaction with 4 volumes RPMI 1640 media (Corning Cellgro) supplemented with 10% heat-inactivated fetal bovine serum (Sigma-Aldrich), 1% MEM - non essential amino acids, 100 μM Na pyruvate (Sigma), 10 mM HEPES pH 7.3 (Sigma), 0.0035% β-mercaptoethanol (Sigma-Aldrich). Cells were washed and resuspended at 10⁷ cells / ml in PBS with 1% BSA and 2 mM EDTA. Validation data for all antibodies used are available on the manufacturer’s website. Prior to staining, cells were blocked with anti-Fc (anti-CD16/CD32) (Tonbo Biosciences, 2.4G2) at a final concentration of 0.25 μg / 10⁶ cells for 15 minutes on ice. Staining panels included anti-CD11b (Tonbo Biosciences M1/70) and anti-CD11c (Tonbo Biosciences N418) conjugated to FITC or PerCP-Cy5.5 each at a concentration of 0.25 μg / 10⁶ cells to remove autofluorescent macrophages. The following stains, antibodies-fluorophore combinations were used to assess cellular phenotype: anti-B220-PerCP-Cy5.5 or -PE-Cy7 (Tonbo Biosciences, RA3-6B2) at 0.05 μg / 10⁶ cells; anti-CD43-FITC (BD #553270) at 0.125 μg / 10⁶ cells; anti-CD138-PE, -BV421, or -BV711 (BD, 281-2) at 0.025 μg / 10⁶ cells; anti-GL7-eFluor660 (eBioscience GL-7) at 0.025 μg / 10⁶ cells; anti-CD45.1-APC-Cy7 (Tonbo Biosciences A20); Viability Violet Stain (Life Technologies L34955), CFSE (Tonbo #13-0850) and CTV (Life Technologies #C34557) both at 10 μM / 10⁷ cells / ml. Cells were stained for 30 minutes and fixed using 1% paraformaldehyde prior to analysis. Staining panels included fluorescence minus one (FMO) controls to ensure that correct compensation was applied, as well as isotype controls to assess non-specific staining. Flow cytometric analysis was collected on a Becton Dickinson (BD) LSRII and FCS files were exported using FACSDiva (v6.2). Analysis of flow cytometric data was conducted in R/Bioconductor (v.3.2.2) using the ‘flowCore’ (v.1.36.9) package⁴⁸ or FlowJo software (v9.7.6). Code is available upon request.

Naïve B cells were isolated using immunomagnetic negative selection for CD43, CD4, and Ter-119 (Miltenyi #130-090-862) following the manufacturer’s protocol. Purity was confirmed by flow cytometric analysis. Plasmablasts and plasma cells were isolated by first enriching the CD138 positive fraction of splenocytes and/or bone marrow using a positive immunomagnetic enrichment on CD138 (Miltenyi #130-098-257) and then FACS using a BD FACS Aria II at the Emory Flow Cytometry Core Laboratory. Adoptively transferred cells were isolated by labeling CD45.1⁺ cells with CD45.1-APC-Cy7 (Tonbo Biosciences A20) and immunomagnetic enrichment using anti-APC beads (Miltenyi # 130-090-855) followed by FACS using a BD FACS Aria II at the Emory Flow Cytometry Core Laboratory.

Ex Vivo differentiation and 5-aza-cytidine treatment

B cells were differentiated ex vivo as previously described³⁷, but with the incorporation of division tracking dye. B cells were isolated using immunomagnetic separation as above, and stained with CTV or CFSE at a concentration of 20⁶ cells / mL in PBS. Cells were differentiated at an initial concentration of 0.5 × 10⁶ cells per ml with LPS (20 μg / ml; Sigma #L2630), IL-2 (20 ng / ml; eBioscience #14-8021), and IL-5 (5 ng / ml; eBioscience #14-8051). Half doses of LPS and cytokines were given on subsequent days. 5-azacytidine (Sigma Aldrich #A2385) was added to cultures every day at concentrations ranging from 50 nM to 500 nM as indicated.

Quantitative real-time PCR (qRT-PCR) analysis

Gene expression was validated using 20,000 cells sorted into RLT lysis buffer and total RNA was purified using the Quick-RNA MicroPrep Kit (Zymo Research). The entire RNA yield was reverse transcribed using SuperScript II reverse transcriptase (Invitrogen), diluted prior to quantitative real-time PCR (qRT-PCR) on a CFX96 instrument (BioRad) using SYBR Green incorporation. Expression levels were expressed relative to 18S ribosomal RNA levels for each gene analyzed. A full list of qRT-PCR primers can be found in Supplementary Table 7.

Code availability

All code is available upon request.

Microarray analysis

Total RNA from each cell type was extracted using the RNeasy mini prep kit (Qiagen) and used for microarray analysis on the MouseRef-8v2 BeadChip (Illumina). Gene expression data were quantile normalized using GenomeStudio v.2011.1 (Illumina) and exported for analysis. Quality control (QC) steps included mapping all probes to the mouse reference genome (mm9) using Bowtie⁴⁹ (v.1.0.0) and removing probes that had multiple alignments, or did not align to a UCSC Known Gene exon⁵⁰. The UCSC known gene database was obtained via the R/Bioconductor package ‘TxDb.Mmusculus.UCSC.mm9.knownGene’⁵¹ (v.3.2.2). This resulted in 22,907 of 25,697 probes passing quality control and coverage of 16,181 genes. Differential expression was identified using linear regression implemented in R/Bioconductor (v3.1.3)⁵¹. Multiple hypothesis testing was applied to probes with signal that was detected in at least one sample (N=10,322) and with a 2-fold change in expression or greater using a Benjamini-Hochburg false discovery rate (FDR)⁵². Those genes with a FDR less than 0.01 were considered significant (Supplementary Table 1).

RNA-seq analysis

For RNA-seq analysis, 40,000 cells were sorted into RLT buffer (Qiagen) with 1% 2-mercaptoethanol (Sigma), vortexed and snap frozen. Prior to extraction 5 μl of 1:2000 dilution of ERCC synthetic RNAs (ThermoFisher) were added to each sample. RNA was extract using RNeasy Mini Kit (Qiagen) following the manufacturer’s protocol. DNA was removed with RQ1 DNase (Promega) at 37°C for 30 minutes and RNA was purified using 3x AMPPure XP clean up (Beckman Coulter). Stranded mRNA-seq libraries were constructed using KAPA Biosystems Stranded mRNA-Seq Kit following the manufacturer’s protocol (KAPA Biosystems). Equal molar amounts of mRNA-seq libraries were amplified for 11 PCR cycles and purified using a 1.5x ratio of AMPPure XP beads (Beckman Coulter). Libraries were pooled in equal molar ratios based on the KAPA Library Quantification Kit (Kapa Biosystems) and sequenced on an Illumina HiSeq2500 with 50 bp paired-end reads.

Mapping and quantification of RNA-seq data

RNA-seq data was mapped back to the UCSC mouse genome mm9 using Tophat2⁵³ (v.2.0.13) with the following parameters “–p 8 –N2 –max-multihits 1 –read-gap-length 1” and the UCSC Known Genes⁵⁰ mm9 transcript file as a guide. The 92 ERCC sequences were added to the mm9 genome as artificial chromosomes. PCR duplicates were determined using Picard (http://broadinstitute.github.io/picard/) and removed from subsequent analyses. Reads that uniquely overlapped mm9 exons were determined in R (v.3.2.2) using the ‘summarizeOverlaps’ function in mode ‘IntersectionNotEmpty’ of the ‘GenomicAlignments’ package⁵⁴ (v.1.6.3). Reads per million (RPM) were calculated for each gene based on the number reads in all potential exons for a given gene and the total number of uniquely mappable reads per sample. Fragments per kilobase per million (FPKM) were calculated based on RPM and the total size of non-overlapping exons for a gene. The number of mRNA molecules per cell were calculated with the following equation:

$$\frac{{\mathrm{mRNA}}_{greneA}}{\mathit{cell}}=\frac{{\mathrm{FPKM}}_{geneA}}{\#\mathit{cells}}x\frac{\Sigma {\mathrm{Molecules}}_{ERCC}}{\Sigma {\mathrm{FPKM}}_{ERCC}}$$

Differential analysis of RNA-seq data

Differentially expressed genes (DEGs) were determined using EdgeR⁵⁵ (v.3.12.0) based on both relative and absolute changes in expression. Gene counts were calculated using all reads mapping to exons of unique UCSC mm9 Known Genes⁵⁰, determined as described above. For relative differences, normalization factors were determined using the EdgeR function “calcNormFactors”. For absolute differences, the normalization factors were determined as the sum of ERCC FPKM divided by the average ERCC FPKM across all samples. A minimum 2-fold change was imposed upon criteria for both relative and absolute differences. P-values calculated by EdgeR were corrected for multiple hypothesis testing using Benjamini-Hochberg FDR correction⁵². In total, DEGs had an FDR ≤0.01 with a fold-change ≥2 using both relative and absolute criteria to be considered significant.

Bioinformatic analyses of expression data

Heatmaps and hierarchical clustering of gene expression data used an ‘average’ or unweighted pair group method with arithmetic mean agglomeration method applied to the Z-score normalized probe signal (microarray) or average number of mRNAs / cell (RNA-seq) using the R/Bioconductor functions ’hclust’ and ‘image’ in a method very similar to that employed by the ‘heatmap’ function⁵¹ (R code available upon request). Principle components analysis was done using the R/Bioconductor function ‘prcomp’ also applied to Z-score normalized expression data.

Gene ontology analysis was conducted on differentially expressed genes (DEGs) using the R/Bioconductor package GOstats (v2.32.0)⁵⁶. For microarray data, all genes with probes that passed QC were used as background. For RNA-seq data, all UCSC mm9 Known Genes⁵⁰ were used as background. Gene Set Enrichment Analysis (GSEA v2.1.0) was performed using the pre-ranked list option. For microarray data the rank was determined by the average t-statistic for all probes mapping to a given gene. Only probes that were detected in at least one sample were used. For RNA-seq data the rank was determined by the −log₁₀(FDR) x sign(fold-change). Here, the FDR is the average FDR determined using both absolute and criteria described above.

DNA methylation assays

DNA isolation was obtained from cells digested overnight with Proteinase K and RNase at 67°C and extracted using phenol-chloroform-isoamylalcohol and ethanol precipitation. RRBS libraries were made from 10 to 500 ng of DNA and were digested overnight with 20 U MspI (New England Biolabs) following the manufacturer’s protocol. RRBS libraries for B cell divisions were also digested with TaqI in separate reactions (New England Biolabs). Digested DNA was purified using a 1.8X Solid Phase Reversible Immobilization (SPRI) clean-up with Agencourt AMPure XP beads (Beckman Coulter). Illumina compatible sequencing adaptors were used and contained fully methylated cytosine residues and were either NEXTflex Bisulfite-Seq Barcodes (BIOO Scientific) or were designed similar to those previously described⁵⁷ and synthesized by Integrated DNA Technologies. DNA was end-repaired and A-tailed and sequencing adaptors were ligated using the Hyper Prep Kit (KAPA Biosystems) following the manufacturer’s protocol. Adaptor-ligated DNA was bisulfite treated using the EpiTect Bisulfite Kit (Qiagen), modifying the manufacturer’s protocol by extending the denaturation thermocycler step from 5 to 10 minutes at 99°C. Adaptor-ligated bisulfite treated libraries were amplified 10-15 times using HiFi Uracil+ Polymerase (KAPA Biosystems) and library concentration was estimated using the KAPA quantification kit (KAPA Biosystems). Size was estimated using a high sensitivity DNA chip (Agilent Technologies). Libraries were sequenced using 50 bp single-end or paired-end reads on a HiSeq2500 by the Genome Technology Center at New York University (NYU).

Combined Bisulfite Restriction Analysis (COBRA) was performed similar to that previously described²⁵. Briefly, high molecular weight DNA was bisulfite treated (see above) and 1 ng of bisulfite converted DNA was amplified 35-40 times using JumpStart Taq polymerase (Sigma) and bisulfite primers (Supplementary Table 6). Half of the amplified product was digested with TaqI (NEB) or BstUI (NEB) and the other half was mock digested as a control, prior to visualization on 1.5-2% agarose gels. DNA methylation was also quantitated by a qPCR approach where genomic DNA was aliquoted into three equal portions where one was mock digested to quantitate the total amount of DNA, one was digested with the methyl-sensitive restriction enzyme HpaII to quantitate unmethylated DNA, and the final aliquot was digested with the methyl-insensitive isoschizomer MspI as a negative control. Equal portions of each aliquot were subjected to qPCR and DNA methylation levels were quantitated as the ratio of HpaII-digested material to mock-digested material. Quantitation was based upon a standard curve of genomic DNA and all primers (Supplementary Table 7) were between 90% and 100% efficient.

DNA methylation bioinformatic analyses

RRBS data were aligned to the in silico bisulfite converted genome (mm9) using Bismark⁵⁸ (v.0.13.1) with the Bowtie2⁵⁹ option. Binary alignment map (BAM) files were parsed to derive DNA methylation calls that were collapsed to the CpG level using custom R scripts that made use of the ‘Rsamtools’ (v.1.22.0) and ‘data.table’ (v.1.9.6) packages (code available upon request). Data were compiled into datasets that included a sample specific coverage (minimum 10X coverage per sample) and a group specific coverage (minimum 10X coverage per group). Hierarchical clustering and principle component analyses were performed on sample specific data sets in a manner analogous to that described for the gene expression analysis except no normalization was performed and heatmaps were ordered by increasing DNA methylation from top to bottom. The distribution of DNA methylation values were assessed using the ‘density’ function in R/Bioconductor and represents the probability density function across sample specific coverage. Average methylation was determined based on sample coverage and differences were determined by Welch’s t-test. B cell, plasmablast and plasma cell differentially methylated CpG loci (DML) were identified using Dispersion Shrinkage for Sequencing (DSS)¹⁹ and division specific DML were determined using the general experimental design version of DSS⁶⁰. DSS was applied to group level data and CpG loci that had an FDR ≤0.01 with a minimum change of 20% in DNA methylation were considered significant. Contiguous DML were defined as two or more DML that were located adjacent to each other on the genome relative to assay group coverage. Significance of DML contiguity was assessed by permutation analysis. This involved randomly permuting DML 1,000 times and calculating the percent of permuted DML that occurred in contiguous regions for each permutation. The P-value was determined by the number of times that the permuted value was equal to or greater than the actual value.

Overlap of DML with enhancer elements (methods described in Meta-analysis) was assessed using Fisher’s exact test⁶¹ implemented in R/Bioconductor. Transcription factor motifs enriched within 50 bp of DML were determined with HOMER software³³ (v.4.7.2) relative to RRBS assay coverage. Results with an FDR ≤ 0.05 were considered significant. Motif position weight matrices were clustered using the “PWMSimilarity” function of the “TFBSTools” package⁶² (v.1.8.2) with a minimum overlap of 6 nucleotides. Data were clustered using hierarchical clustering as described above. Code is available upon request.

Correlation of gene expression and DNA methylation

To analyze DNA methylation and gene expression correlation, CpG loci were annotated to the closest UCSC mm9 Known Gene⁵⁰ transcript using custom R / Bioconductor code (available upon request). CpG loci that were within 100 kb of a transcript were assigned to the closest gene. Subsequently, the expression fold-change for each gene was plotted by the change in absolute DNA methylation. Significance of inverse correlation was assessed using Fisher’s exact test⁶¹ (implemented in R) to determine if more DML-DEG correlations were negatively associated than expected by chance.

Meta-analysis

Analysis of histone modifications utilized previously described chromatin immunoprecipitation sequencing (ChIP-seq) experiments performed on primary B cells^{31–33,63,64}. ChIP-seq data were obtained from Gene Expression Omnibus experiments GSE30859³³, GSE38046⁶⁴, GSE51336³¹, GSE42706⁶³, and GSE51011³²; specific data sets are listed in Supplementary Table 4. Data were uniformly aligned and processed to the mouse genome (mm9) using Bowtie2⁵⁹ (v.2.1.0). ChIP-seq fragment size for each data set was calculated using the “chip-seq” (v.1.20.0) package in R/Bioconductor based on the SISSER method previously described⁶⁵Enriched regions for those published by the ENCODE project³¹ were downloaded from the UCSC genome browser (http://genome.ucsc.edu/cgi-bin/hgFileSearch?db=mm9). Enriched regions for other studies were determined using MACS software (v1.4)⁶⁶. Enhancers were determined for spleen, CH12 cells, thymus, and whole brain H3K4me1, H3K27ac, and H3K4me3 data by taking overlapping H3K4me1 and H3K27ac regions that did not overlap a region enriched for H3K4me3. Odds ratio and significance of overlap with DML were determined using Fisher’s exact test.