3.5.1. Dysregulation of Epigenetic Writer-Eraser Equilibria Diminish Plasticity of B Cells during Maturation
In
Figure 4 we discussed different scenarios of oncogenic perturbations in terms of a simplified scheme of epigenetic regulation. The systematic analysis of transcriptional activities of epigenetic modifiers presented in the previous subsections now enables us to compare the expected with the observed changes (
Figure 10a). Compared with B cells, the equilibria of histone methylation reactions shift in direction of methylated H3K9 and H3K27 and demethylated H3K4 if one uses the expression data as a proxy for enzyme activities. These shifts suggest the increase of repressed and the decrease of active promoters in lymphoma accompanied by DNA hypermethylation,
i.e., similar alterations as expected for EZH2 and MLL2 mutations (compare with the scenario in
Figure 4b). Hence, the latter mutations and the expression changes of the enzymes suggest similar effects on DNA methylation and gene activities.
Figure 9.
TCA-cycle related epigenetic compounds: (a) overview map and profiles of histone JmjC- and DNA TET-demethylases; (b) gene expression profiles of TCA-related enzymes and (c) gene set enrichment profiles of GO-gene sets related to TCA; (d) PSF-profiles of metabolic products along the TCA-cycle. Fumarate and succinate can act as antagonists of αKG to inhibit αKG-dependent dioxygenases. Note that the PSF profiles of all metabolites are very similar and virtually are anti-correlated with the expression of the demethylases shown in part a.
Figure 9.
TCA-cycle related epigenetic compounds: (a) overview map and profiles of histone JmjC- and DNA TET-demethylases; (b) gene expression profiles of TCA-related enzymes and (c) gene set enrichment profiles of GO-gene sets related to TCA; (d) PSF-profiles of metabolic products along the TCA-cycle. Fumarate and succinate can act as antagonists of αKG to inhibit αKG-dependent dioxygenases. Note that the PSF profiles of all metabolites are very similar and virtually are anti-correlated with the expression of the demethylases shown in part a.
The diversification of lymphoma data into different subtypes and healthy controls enables a refined view, for example, on the changes of enzyme expression between different stages of B cell development in the germinal center. For some of the enzymes (e.g., KMT6/EZH2 and KDM6B/JMJD3) one finds similar expression levels in GCB cells and in part of the lymphoma subtypes: For example, EZH2 is silenced in resting B cells but massively up-regulated in GCB cells, which undergo rapid proliferation and immunoglobulin affinity maturation. JMJD3 shows nearly the opposite trend being highly active in B cells but nearly inactive in GCB cells and lymphoma. A similar, although less pronounced trend is found for KMD6A/UTX, another relevant K27DM [
57].
These results reflect alterations of cellular programs during lymphocyte development, which are accompanied or even governed by epigenetic mechanisms. PRC2-mediated H3K27 trimethylation in healthy B cells is required on a selective core of PcG targets whose repression enables TF-dependent cell reprogramming [
90], e.g., to transform naïve B cells into highly proliferative GCB cells. This reprogramming potentially also includes H3K4 methylation, which in concert with H3K27 methylation form bivalent promoter domains. These combined histone marks are required to poise genes for activation or deactivation in response to developmental and differentiation cues [
91]. The resolution of the bivalent domains is mediated by the H3K4 and H3K27 demethylases. Mutations of MLL2 and EZH2 genes may both perturb this equilibrium. In consequence associated regulations go awry and lymphomas can ensue. Perturbations in the fine balance of GCB cell proliferation, differentiation and antigen exposure are assumed to lock GCB cells in an immature and proliferative state, which in collaboration with other lesions induce lymphoma.
Comparison of the transcriptional activities of the enzymes between the lymphoma subtypes reveals subtle differences, which suggest different types and degrees of disturbed equilibria. In all example profiles shown in
Figure 10a one sees a monotonous increase of the mean enzyme expression from DLBCL over IntL to BL suggesting a continuous shift of the histone methylation equilibria. We recently presented indications for pronounced chromatin remodeling between BL and DLBCL affecting first of all transformations between repressed, poised and active promoter states [
6]. These changes of promoter states potentially ensure alternative activation of proliferative (in BL and partly IntL) and inflammatory and developmental (in DLBCL and partly FL and IntL) expression programs and they are accompanied by aberrant DNA methylation in the promoter regions of the affected genes. Note that H3K4 and especially H3K27 methylation can tune not only developmental and “stemness” genes but also inflammatory processes needed to respond to external stimuli [
58]. These different types of genes have in common that their function requires a high degree of plasticity for cell fate decisions. These decisions should induce different kinds of functional differentiation including maturation stages of the cells, their proliferative and metabolic activity and also the ability for adequate immune response.
3.5.2. Activation of Gene Expression and of TCA Metabolism in Lymphoma Associates with Epigenetics
Bivalent and repressed promoters are prerequisites for the plasticity of the B cells required during their maturation in the GC. These genes can serve as hubs in TF networks that switch whole cascades of downstream genes either as suppressors, activators and/or enhancers of their transcriptional activity. In consequence, suppression of anti-proliferative programs and/or of activation of inflammatory processes is assumed to govern molecular mechanisms in lymphoma with respect to these functionalities. Increased proliferation requires up-regulation of the molecular machineries required for transcription and translation. In addition it needs activation of the metabolism delivering the energy needed for these processes as indeed observed in BL and IntL [
10].
Figure 10.
(
a) The expression profiles of the methyltransferases and demethylases of H3K9, H3K4, H3K27 and of DNA-CpG’s suggest a shift of expression of the affected genes towards repressed and CpG-methylated promoter states. The scheme is redrawn from
Figure 4 and supplemented by selected expression profiles of the respective enzymes determined from the lymphoma cohort studied. The triangles indicate the shift of the methylation-demethylation reactions in lymphoma compared with B cells deduced from the expression profiles from the respective enzymes. (
b) Expression profiles of MYC and of the mean total expression averaged over all SOM-metagenes of each samples. Total expression is consistently activated in lymphoma except MM compared with B and GCB cells, whereas MYC is on high level in BL and IntL carrying genetic activating MYC defects.
Figure 10.
(
a) The expression profiles of the methyltransferases and demethylases of H3K9, H3K4, H3K27 and of DNA-CpG’s suggest a shift of expression of the affected genes towards repressed and CpG-methylated promoter states. The scheme is redrawn from
Figure 4 and supplemented by selected expression profiles of the respective enzymes determined from the lymphoma cohort studied. The triangles indicate the shift of the methylation-demethylation reactions in lymphoma compared with B cells deduced from the expression profiles from the respective enzymes. (
b) Expression profiles of MYC and of the mean total expression averaged over all SOM-metagenes of each samples. Total expression is consistently activated in lymphoma except MM compared with B and GCB cells, whereas MYC is on high level in BL and IntL carrying genetic activating MYC defects.
To judge this overall balance we calculated the mean total expression level of each sample using the metagene expression data obtained in our SOM analysis. The results clearly reveal a bimodal distribution with high expression levels in GC-derived lymphoma on one hand and with low expression levels in healthy B and GCB cells and in MM sharing similar expression signatures with B cells (
Figure 10b). This result clearly supports the view that malignant transformations from B and/or GCB cells into GC-derived lymphoma are paralleled by the massive upregulation of the transcriptional activity in the cells. Interestingly, the profile of total expression partly resembles the PSF profiles of TCA metabolites (compare with
Figure 9c) but anti-correlates with the expression profiles of genes located in spot I and particularly with that of KDM4C shown in
Figure 7d. These results suggest that total gene expression in lymphoma and B cells is related to the TCA-energy metabolic activity, which, in turn, couples with the expression of epigenetic modifiers and particularly with KDM4C demethylating H3K9me3. Its low level in lymphoma (except MM) promotes trimethylation of H3K9 and recruitment of DNMTs which are on high level in lymphoma (see DNMT1,
Figure 7b). In final consequence, one expects increased CpG-methylation in agreement with the scheme in
Figure 10.
On the other hand, our data indicate subtle differences of the expression of a series of genes between B- and GCB-cells. Particularly, GCB-cells show higher total expression (
Figure 10b) and higher activity of KEGG-TCA and NADPH related genes compared with B-cells (
Figure 9c). This difference is possibly governed by a shift of the H3K9-methylation equilibrium, which suggests also changes in DNA promoter methylation (
Figure 10a) of genes affecting the energy metabolism. We indeed identified differential methylation patterns between B- and GCB-cells, where increased methylation is found for PRC2-targets and repressed bivalent chromatin states in GCB-cells [
6]. This result suggests that chromatin remodeling in the GC switches the state of metabolic activity between GCB- and B-cells.
We considered also another possible mechanism of global activation of transcription. Particularly, MYC can act as a universal amplifier of gene expression by hyper-activating still active genes via transcriptional pause release [
92,
93]. In consequence genes once activated by other mechanisms can be expected to become hyper-activated via aberrant MYC-overexpression. The expression profile of MYC (
Figure 7b) however considerably differs from that of global expression. MYC is on high levels in BL and to a less degree also in IntL compared with DLBCL and FL, mainly due to genetic defects amplifying the MYC gene in BL and part of IntL. TCA metabolic activity better associates with the global transcriptional level in GC-derived lymphoma, suggesting mutual relations and possible consequences for epigenetics as discussed above.
3.5.3. Asymmetric Activation of Methyl-Writers and -Erasers
Our study clearly shows that the expression of nearly all enzymes considered alters markedly between the lymphoma subtypes. For a holistic view we make use of the fact that SOM cartography maps the genes in an organized way. The structure of the map provides information about the underlying regulatory net because the arrangement of spots reflects their mutual co-variance structure (
Figure 2c). We assigned the location of the epigenetic modifiers in the map to the respective spots (see
Table 1), and with a more coarse resolution to the quadrants Q1 to Q4. Interestingly, we found strong depletion of epigenetic modifiers in Q4 opposed by their enrichment in Q2 and particularly also in Q1 and Q3 (
Figure 6). In the next step we rearranged the network of expression modules to better resolve its covariance structure (
Figure 11). It clearly reveals a “backbone” of mutually correlated modules, which sequentially connects spots from Q1 to Q3. A second backbone is formed by correlated spots mostly located in Q4 and partly in Z (spot MM) and Q3 (IM). It forms an almost separated entity connected via anti-correlated edges (in red) from the first, main backbone. The spots and thus also the respective quadrants contain co-regulated genes specifically up-regulated in different subtypes as indicated in
Figure 11. Importantly, almost each of the regulatory modes also affects a group of epigenetic modifiers. In other words, (de)regulation of epigenetics covers the whole transcriptional landscape of lymphoma. Moreover, the epigenetic modifiers enrich within the spot clusters when compared with the total number of genes in the spots (Fishers exact test:
p = 3.7 × 10
−6). Hence, epigenetic modifiers are affected by (de-)regulatory effects with higher probability than expected by chance.
The network can be decomposed into a subnet, which mainly refers to genes that antagonistically switch between BL on one hand and DLBCL/FL on the other hand. Interestingly, this subnet accumulates methyltransferases in Q3 that tend to repress gene expression of their target genes leading to antagonistic expression profiles in Q4 (the detailed assignment of enzymes to each of the spots is given in
Supplementary Figure S3). The imbalance between the gene expression of methyltransferases and demethylases between Q3 and Q4 can be rationalized partly by the requirement of maintenance methylation of DNA-CpG and histone methylation marks after cell division and DNA replication, which requires high activities of methyltransferases. The question whether upregulation of KMT and DNDMT expression in the highly proliferative subtypes BL and partly IntL ensures maintenance of DNA and histone methylation patterns or whether it leads to progressive loss of methylation requires further studies.
Figure 11b illustrates this antagonism between BL and DLBCL using a triangular scheme of lymphocyte development and lymphoma heterogeneity. Particularly the K27MTs in Q3 are expected to inhibit PRC2 targets, which indeed accumulate in the anticorrelated region Q4. These genes were subsumed as group 1 genes in [
6], being hypermethylated and overexpressed in lymphoma compared with the controls. In addition, compounds of the PRC1 and SWI/SWF complexes co-regulate with these MTs and their targets. This parallel suggests that the stabilization of repressed promoters and the opening/closing of chromatin are mechanisms that change the expression patterns between BL and DLBCL.
Another subnet in
Figure 11a contains genes that switch between the MM and IntL. It accumulates methyltransferases and demethylases in Q1 that repress or activate expression. Most of the demethylases are JmjC-family enzymes, which are repressed by TCA products such as fumarate and succinate as illustrated in the right panel of
Figure 11b. These demethylases together with the KMTs in Q1 then either repress or activate expression of their targets giving rise to group 3 and group 4 genes as genes that antagonistically change their expression between Q1 and Q4 [
6]. These gene groups are enriched in developmental regulators, genes related to immune response, PRC2 targets and also CIMP/GCIMP genes hypermethylated in colon and brain cancer, respectively.
Both subnets overlap in Q2 collecting most of the epigenetic modifiers including activating and repressing ones without clear preference (
Figure 11a). This overlap region contains modules that switch expression between (GC)B cells and lymphoma. We hypothesize that the underlying modes regulate transcriptional programs differentiating between healthy B and GCB cells. Other enzymes localize near spots H and MM referring to early and late stages of B cell maturation, respectively [
12].
Figure 11.
(
a) Network of expression modules governing lymphoma heterogeneity. The nodes refer to the spot clusters extracted from the SOM analysis (for functional assignments see [
6]). Green and red edges indicate positive and negative correlations with |w| > 0.3 using the weighted topological overlap correlation measure (see also
Figure 2). A “backbone” of correlated modules can be assigned to the Q1–Q3 quadrants of the SOM used to map enzyme activities. Accumulation of different types of modifiers and complexes in Q1–Q3 is shown in the right part. Q4 is almost depleted from modifiers. A detailed list of the modifiers found in each spot is given in
Figure S3. The network can be roughly divided into two subnets as described in the text. (
b) The DLBCL/BL and MM/IntL subnets explain the expression changes of three different groups of genes identified in [
6] (see text).
Figure 11.
(
a) Network of expression modules governing lymphoma heterogeneity. The nodes refer to the spot clusters extracted from the SOM analysis (for functional assignments see [
6]). Green and red edges indicate positive and negative correlations with |w| > 0.3 using the weighted topological overlap correlation measure (see also
Figure 2). A “backbone” of correlated modules can be assigned to the Q1–Q3 quadrants of the SOM used to map enzyme activities. Accumulation of different types of modifiers and complexes in Q1–Q3 is shown in the right part. Q4 is almost depleted from modifiers. A detailed list of the modifiers found in each spot is given in
Figure S3. The network can be roughly divided into two subnets as described in the text. (
b) The DLBCL/BL and MM/IntL subnets explain the expression changes of three different groups of genes identified in [
6] (see text).
In summary, network analysis of the epigenetic modifiers identifies two subnets related to differential expression between BL and DLBCL, and between MM and IntL subtypes, respectively. The former subnet is governed by methyltransferases upregulated in BL and repressing transcription of their target genes. The latter one contains demethylases which are presumably under metabolic control and which can activate and/or repress their targets.