DNA methylation dynamics during B cell maturation underlie a continuum of disease phenotypes in chronic lymphocytic leukemia

Abstract

Charting differences between tumors and normal tissue is a mainstay of cancer research. However, clonal tumor expansion from complex normal tissue architectures potentially obscures cancer-specific events, including divergent epigenetic patterns. Using whole-genome bisulfite sequencing of normal B cell subsets, we observed broad epigenetic programming of selective transcription factor binding sites coincident with the degree of B cell maturation. By comparing normal B cells to malignant B cells from 268 patients with chronic lymphocytic leukemia (CLL), we showed that tumors derive largely from a continuum of maturation states reflected in normal developmental stages. Epigenetic maturation in CLL was associated with an indolent gene expression pattern and increasingly favorable clinical outcomes. We further uncovered that most previously reported tumor-specific methylation events are normally present in non-malignant B cells. Instead, we identified a potential pathogenic role for transcription factor dysregulation in CLL, where excess programming by EGR and NFAT with reduced EBF and AP-1 programming imbalances the normal B cell epigenetic program.

Figure 1

Epigenetic programming during B cell maturation. (a) Top, FACS sorting markers used to isolate the analyzed B cell subsets after selection of CD19⁺ cells. Bottom, the frequency of IGHV3 mutations in each subpopulation. (b) Top, TWGBS summary comparing naive B cells and high-maturity memory B cells. Bottom, methylation heat maps for the top 5,000 most variable windows. (c) Enrichment of differentially methylated windows among chromatin states, defined using the 15-state ChromHMM model (hypermethylated, >20% change; hypomethylated, >40% change), in the comparison of naive B cells and high-maturity memory B cells. Fold enrichment was calculated independently in 19 lymphoblastoid B cell lines (error bars, s.e.m.). TSS, transcriptional start site. (d) Bubble scatterplot of 223 transcription factor motifs displaying the prevalence and fold enrichment of each motif in hypomethylated windows (>40% change). Bubble size corresponds to the P value. (e) Bubble scatterplot of 78 transcription factor ChIP-seq peaks (determined in GM12878 cells) in hypomethylated windows (>40% change). Bubbles are colored according to the cognate motif in d. (f) Examples of loci differentially methylated in normal B cells at single-CpG resolution. Regions of hypermethylation (pink) within the gene body of MSI2 (left) and hypomethylation (blue) upstream of NOTCH1 (right) are shown. Chromatin states (colors correspond to those used in c) and transcription factor binding site peaks from GM12878 cells are indicated. (g) Average methylation level of hypomethylated windows (>20% change) containing high-confidence transcription factor binding sites (TFBSs) and hypermethylated windows (>20% change) overlapping regions of transcriptional elongation in each B cell subtype. (h) Left, DNA methylation phylogenetic tree diagram of blood cell development of common hematopoietic cell types. Each branch tip represents a single sample. Right, diagram of B cell maturation constructed using high-confidence methylation (450K) at transcription factor binding sites. HSCs, hematopoietic stem cells; NK cells, natural killer cells; CC, centrocytes; CB, centroblasts. (i) Scatterplots displaying the change in methylation of 450K probes in naive B cells after in vitro stimulation and 5 d in culture (y axis) versus the difference in methylation between naive B cells and high-maturity memory B cells (x axis). Naive B cells were stimulated with CD40L (left) or with CD40L and antibody to IgM (αIgM; right). (j) Bubble scatterplot of transcription factor motifs overlapping hypomethylated CpGs following CD40L stimulation and 5 d in culture.

Figure 2

Evaluation of the maturation state of CLL using DNA methylation. (a) The average level of methylation at transcription factor binding sites for all six significantly programmed transcription factors in 267 CLL samples assessed by 450K array. Samples are ranked according to the average methylation level for all transcription factor binding sites. The average β value (global methylation) is also shown for each sample. (b) Heat map showing the methylation at all high-confidence AP-1, EBF1 and RUNX3 binding sites and in hypermethylated transcriptional elongation regions in 267 CLL samples. Consensus clustering of CLL samples identifies three methylation subtypes (LP-CLL, IP-CLL and HP-CLL). The level of IGHV homology (identity) for each CLL is indicated. (c) Scatterplot displaying the average methylation of AP-1, EBF1 and RUNX3 binding sites versus hypermethylated transcriptional elongation regions. (d) DNA methylation phylogenetic tree diagram of the DKFZ CLL sample cohort (n = 128) and normal B cell subtypes generated using AP-1, EBF1 and RUNX3 binding sites and hypermethylated transcriptional elongation regions. Each line represents a CLL sample; CpG methylation values for normal samples were averaged according to subtype. Colors indicate CLL subtypes as in c. (e) DNA methylation phylogenetic tree diagram of the CLL samples selected for RNA-seq analysis. Patient numbers are indicated for each sample; shaded areas represent the CLL subtypes. (f) Heat map of 459 genes differentially expressed in the LP-CLL and HP-CLL subtypes (FDR q < 0.05). Samples are ordered according to their position within the phylogenetic analysis in e. Genes that have been previously reported to be differentially expressed in the IGHV subtypes or to have a role in CLL pathogenesis are also indicated. (g) Scatterplot showing the correlation between the degree of methylation maturation (as assessed by the trunk distance from naive B cells in the phylogenetic analysis) and the acquisition of the HP-CLL expression state.

Figure 3

The impact of DNA methylation programming in patients with CLL. (a) DNA methylation heat map of 18 loci in an independent, clinically annotated set of 327 patients with CLL assessed by MassARRAY. Consensus clustering identifies three clusters of CLL cases (blue, LP-CLL; yellow, IP-CLL; brown, HP-CLL). (b) Kaplan-Meier plots showing time from diagnosis to treatment (top) and overall survival (bottom) for each CLL subtype (P < 0.0001, log-rank test). (c) The methylation maturation score of CLL samples separated by methylation subtype. This value depicts the degree of maturity by combining all available MassARRAY methylation data, taking into account the direction of programming (hyper- or hypomethylation) (LP-CLL, n = 163; IP-CLL, n = 61; HP-CLL, n = 103). Horizontal bars represent mean values. (d) Statistical summary using the methylation maturation score as a continuous variable within the CLL clusters. The hazard ratio (HR) and confidence interval (CI) of the effect of the score within each subtype per outcome variable were estimated from proportional hazards models. Significant associations (P < 0.05) are highlighted in bold.

Figure 4

Previously identified aberrant methylation in CLL is found in comparison of normal B cell subtypes. (a) Comparison of the methylation changes in CLL with those in high-maturity memory B cells using naive B cells as a reference. Data for 450K probes located within regions analyzed in previous studies were averaged for each cell type; CLL data from all three subtypes were combined. The gene list was obtained from Florean et al.. (b) DNA methylation profiles of representative promoters that were falsely identified as aberrantly methylated regions shown for six normal B cell subsets (left) and the CLL subtypes (right). The promoter region of ZAP70, a region reported to be differentially methylated between IGHV subtypes, is also shown. TWGBS data were used to generate the profiles for the normal B cell subtypes, and the CLLs (n = 12 samples averaged per subtype) were analyzed by MassARRAY. The position of the highly prognostic CpG at +223 in the ZAP70 locus is indicated. (c) Comparison of the difference in methylation for CLLs with wild-type IGHV and CLLs with mutated IGHV versus the difference in methylation from naive B cells to high-maturity memory B cells, displaying all 450K probes that differ by >10% between the IGHV subtypes. Probes that are hyper- and hypomethylated between naive B cells and high-maturity memory B cells are highlighted.

Figure 5

Deficiency in DNA methylation programming in LP-CLLs results from loss of expression of the EBF1 and FOS transcription factors. (a) Differences in methylation from naive B cells to high-maturity memory B cells and to LP-CLL using the 450K array. Data for CpGs were averaged for each subtype (naive B cells, n = 5; high-maturity memory B cells, n = 5; LP-CLLs, n = 107). CpGs were categorized as having failed (red) or aberrant (blue) hypomethylation versus successful hypomethylation (gray). Bottom, the total number of CpGs per category per CLL subtype. (b) Proportional Venn diagram showing the number of CpGs that exhibit failed hypomethylation in the CLL subtypes. (c) Left, bubble scatterplot of transcription factor motif enrichment in failed (versus successful) hypomethylation. Bubble size corresponds to the P value. Right, enrichment P values of 78 transcription factor ChIP-seq peaks in GM12878 cells in failed hypomethylation. (d) DNA methylation in normal B cells and CLLs covering the TNF locus. Mean DNA methylation (450K) levels for the naive B cell, high-maturity memory B cell, LP-CLL and HP-CLL subtypes are designated by colored lines; shaded areas indicate differences in methylation (relative to naive B cells) that exhibit successful (gray), failed (pink) or aberrant (blue) programming. Bottom, transcription factor binding in GM12878 cells. (e) EBF1 expression from RNA-seq data highlighting EBF1⁺ CLL cases. Right, frequency of EBF1⁺ cells in the CLL subtypes, combining RNA-seq, qPCR and microarray data. RPKM, reads per kilobase of transcript per million mapped reads. (f) Average methylation levels of EBF1 binding sites (versus other programmed transcription factor binding sites) in EBF1⁺ and EBF1⁻ CLLs. Differences were assessed by t test. (g) FOS expression time course after TPA induction as assessed by qPCR (error bars, s.e.m.). (h) Heat map showing genomic copy number profiling of chromosome 14q in 21 CLLs. CLL subtype is indicated. (i) Average methylation of AP-1 binding sites (versus other programmed transcription factor binding sites) in del14q versus disomic 14q CLLs. Cases with a subclonal deletion were considered separately. P values were assessed by t test.

Figure 6

Transcription factor binding sites enriched in CLL-specific hypomethylation. (a) Proportional Venn diagram showing the number of aberrantly hypomethylated CpGs in LP-CLL and HP-CLL (illustrated in Fig. 5a). (b) Bubble scatterplot of transcription factor motifs displaying the prevalence and fold enrichment of each motif in aberrant hypomethylation. Bubble size corresponds to the P value of the association. Right, enrichment P value of 78 transcription factor ChIP-seq peaks in GM12878 and K562 cells in LP-CLL–specific hypomethylated regions. P values were assessed by Fisher’s exact test. (c) Composite CpG methylation levels surrounding transcription factor motifs (range of ±1 kb) in aberrantly hypomethylated regions from 450K data. E2A and SPI1 motifs exemplify motifs equally aberrantly hypomethylated across the CLL subtypes, whereas NFAT and EGR motifs display additional hypomethylation in LP-CLLs. (d) Expression of EGR2 in each CLL subtype as determined by qPCR during a 5-h time window after TPA induction (error bars, s.e.m.). (e) Summary of the nonsynonymous EGR2 DNA-binding domain mutations found in 670 CLL cases showing the amino acid changes for each CLL subtype. WT, wild type. (f) Bubble scatterplot of transcription factor motif enrichment in regions specifically hypomethylated in EGR2-mutated CLLs. Bubble size corresponds to the P value of the association.

Figure 7

Summary of global and transcription factor binding site DNA methylation programming in normal B cells and CLL. Pie charts display the proportion of CpGs across the genome that are modified relative to naive B cells during normal DNA methylation programming and in CLL development. The cartoons present a summary of the dynamic DNA methylation programming in normal and CLL samples. Selective transcription factor binding sites and regions of transcriptional elongation are significantly enriched for DNA methylation changes. Many of the regions targeted for hypomethylation are marked with a histone enhancer chromatin signature, implying that hypomethylation of these regions is involved in the establishment of enhancers during maturation. Some binding sites, such as those for NF-κB, IRF and OCT transcription factors, are programmed earlier than others, such as binding sites for EBF, AP-1 and RUNX transcription factors. Overall, hypermethylation occurs more rarely but is enriched in regions of transcriptional elongation. CLL samples mostly retain the state of methylation programming at transcription factor binding sites corresponding to the stage from which the founder cell arose, with from ~70–100% of the program in normal memory cells achieved overall. In the more aggressive LP-CLL subtype, EBF and AP-1 binding sites fail to achieve mature levels of programming relative to other successfully programmed transcription factor binding sites. EGR (and NFAT) sites are aberrantly hypomethylated relative to normal cells.