Transcriptional alterations in glioma result primarily from DNA methylation-independent mechanisms

Abstract

In cancer cells, aberrant DNA methylation is commonly associated with transcriptional alterations, including silencing of tumor suppressor genes. However, multiple epigenetic mechanisms, including polycomb repressive marks, contribute to gene deregulation in cancer. To dissect the relative contribution of DNA methylation-dependent and -independent mechanisms to transcriptional alterations at CpG island/promoter-associated genes in cancer, we studied 70 samples of adult glioma, a widespread type of brain tumor, classified according to their isocitrate dehydrogenase (IDH1) mutation status. We found that most transcriptional alterations in tumor samples were DNA methylation-independent. Instead, altered histone H3 trimethylation at lysine 27 (H3K27me3) was the predominant molecular defect at deregulated genes. Our results also suggest that the presence of a bivalent chromatin signature at CpG island promoters in stem cells predisposes not only to hypermethylation, as widely documented, but more generally to all types of transcriptional alterations in transformed cells. In addition, the gene expression strength in healthy brain cells influences the choice between DNA methylation- and H3K27me3-associated silencing in glioma. Highly expressed genes were more likely to be repressed by H3K27me3 than by DNA methylation. Our findings support a model in which altered H3K27me3 dynamics, more specifically defects in the interplay between polycomb protein complexes and the brain-specific transcriptional machinery, is the main cause of transcriptional alteration in glioma cells. Our study provides the first comprehensive description of epigenetic changes in glioma and their relative contribution to transcriptional changes. It may be useful for the design of drugs targeting cancer-related epigenetic defects.

Figure 1.

Aberrant methylation at CGI/promoters is not the main contributor to transcriptional alteration in glioma. (A) Classification of the 14,714 genes analyzed in this study. (B) DNA methylation level (mean β-values) of the 11,795 CGI/promoters (rows) analyzed in IDHmut and IDHwt glioma and control (normal brain tissue) samples (columns). Left columns show their hypermethylation or hypomethylation status in IDHmut and IDHwt glioma samples compared with controls. (C,D) Differential expression status of genes associated with hypermethylated (C) or hypomethylated (D) CGI/promoters in IDHwt (left) and IDHmut (right) glioma samples compared with controls.

Figure 2.

Extent of transcriptional alterations in IDHwt and IDHmut glioma samples. (A) Volcano plot analysis of differential gene expression in IDHwt (left) or IDHmut (right) glioma samples. Blue and red dots represent genes that were significantly down- or up-regulated, respectively, compared with healthy controls (n = 14,714 genes analyzed). (B) Circular karyotype showing the duplication (red) and deletion (blue) frequencies at the 14,714 analyzed genes in IDHwt (outer circles) and IDHmut (inner circles) samples. Genes showing a significant correlation between CNV and expression are symbolized by an orange (up-regulated) or green (down-regulated) line. (C) Correlation analysis between CNV and expression levels for the EGFR and HOXA13 genes in IDHwt (yellow dots, left) and IDHmut (blue dots, right) glioma samples. Black dots indicate value in healthy controls. EGFR overexpression correlated with increased copy number in IDHwt glioma samples. (D) Classification of the genes with expression alterations that did not correlate with CNV.

Figure 3.

Four expression defect classes. (A) Integrative analysis of differential gene expression and methylation in eight IDHwt glioma samples identified four main defect classes: gain of expression with gain of methylation (Meth+/Exp+), gain of expression with CGI/promoter remaining unmethylated (No Meth/Exp+), loss of expression with gain of methylation (Meth+/Exp−), and loss of expression with the CGI/promoter remaining unmethylated (No Meth/Exp−). (B) Differential DNA methylation analysis in all IDHwt glioma samples (n = 55) versus controls (n = 8) (delta of the mean β-value). Glioma samples were grouped in the four classes of expression defects defined in A. The methylated and methylable status of genes is indicated in the left column. (C) Integrative analysis of differential expression and methylation at selected Meth+/Exp+ (upper), No Meth/Exp+ (middle), and No Meth/Exp− (lower) genes in 42 IDHwt glioma samples compared with controls (n = 8). (D) Integrative analysis of differential gene expression and methylation in an independent cohort of 135 IDHwt glioma samples (validation cohort) also identified the four main defect classes. Odds ratio and significance of the overlap (Fisher's exact test) between the data of the validation cohort and our cohort, for each defect category, are shown on the right panel.

Figure 4.

Genes with bivalent chromatin signature in ES cells are more prone to be deregulated in IDHwt glioma. (A) Gene Ontology terms (biological processes) enriched in genes from the four defect categories. For each category, the four highest terms are shown. (B) Distribution of genes of each defect category according to their chromatin signature in human ES cells: (none) gray; (bivalent) black; (H3K4me3-only) blue; (H3K27me3-only) purple. As reference, the distribution of the 14,714 genes analyzed in this study according to their chromatin signatures in human ES cells is shown in the left panel. (C) Expression level and chromatin signatures of genes of the four defect categories in human ES cells, neural progenitor cells (NPCs), and healthy brain. For comparison, the same analysis is provided on the right panel for genes without expression defect (unaffected) in IDHwt glioma samples.

Figure 5.

Expression from genes with methylated CGI/promoter. (A) Data mining–derived ChIP-seq read density data for H3K27me3 (pink) and H3K4me3 (blue) in “Meth+/Exp+” genes in a ±2 kb window centered on their TSS, in healthy brain (left) and IDHwt-derived cell lines (right). The mean ChIP-seq signal values are shown on the lower panels for “Meth+/Exp+” genes (red line) and for the 14,714 analyzed genes (black line) that were used as normalized reference. (B) Heatmap showing CpG sites density and their mean methylation level in a ±2 kb window centered on the TSS of “Meth+/Exp+” genes and enriched (upper) or depleted (lower) for H3K4me3 in IDHwt glioma samples compared with healthy controls. The ChIP-seq read density obtained in IDHwt-derived cell lines is shown on the right panels. (C) Genome Browser view at the TWIST1 and FOXD3 loci to show H3K4me3 enrichment, differential DNA methylation, and the oriented RNA-seq signal. These two loci are representative of genes that initiate from an H3K4me3-marked TSS embedded in a methylated CGI/promoter in IDHwt samples. (D) HOXC11 is representative of genes in which expression initiates from an alternative TSS in IDHwt glioma samples.

Figure 6.

Transcription factor binding motifs in the promoters of genes overexpressed in glioma samples. (A) Data mining–derived ChIP-seq read density data for H3K27me3 (pink) and H3K4me3 (blue) at “No Meth/Exp+” genes in a ±2 kb window centered on their TSS in healthy brain (left) and IDHwt-derived cell lines (right). The mean ChIP-seq signal values are shown on the lower panel for all “No Meth/Exp+” genes (orange line) and for those that are (dotted green line) or not (dotted pink line) marked by H3K4me3-only in ES cells, NPC, and brain. The black line, used as normalized reference, shows the value for all analyzed genes. (B) Transcription factor motif enrichment in the CGI/promoter of “No Meth/Exp+” genes, calculated using i-cis Target and represented as a normalized enrichment score (NES). Enrichment is shown for genes that are (green squares) or are not (pink squares) marked by H3K4me3-only in ES cells, NPC, and brain. When a transcription factor possesses several binding motifs, data are presented as a box plot. (C) Expression status, assessed by RNA-seq, of the transcription factors identified in B. The middle column shows their expression status in healthy control (n = 5) (white, not expressed; gray, expressed: fpkm > 1) and the right column their expression in IDHwt glioma samples (n = 8). The left column shows the motif enrichment in all “No Meth/Exp+” genes (black), and those marked (green) and not marked (pink) by H3K4me3-only in ES cells, NPC, and brain. (D) Expression versus controls of selected overexpressed transcription factor identified in C assessed by RT-qPCR in 42 IDHwt glioma samples. Details for each sample are provided in the lower panel (P-value by Mann–Whitney U test).

Figure 7.

Gene repression is associated with H3K27me3 gain. (A) Data mining–derived ChIP-seq read density data for H3K27me3 (pink) and H3K4me3 (blue) at “Meth+/Exp−” and “No Meth/Exp−” genes in a ±2 kb window centered on their TSS, in healthy brain (left) and IDHwt-derived cell lines (right). The mean ChIP-seq signal values are shown in the lower panels for “Meth+/Exp−” (purple line) and “No Meth/Exp−” (blue line) genes. Genes in the “No Meth/Exp−” group were further subdivided in genes marked (dotted light blue line) and not marked (dotted dark blue line) by H3K4me3-only in ES cells, NPC, and brain. The black line used as normalized reference shows the value for all analyzed genes. (B) ChIP analysis of H3K9ac, H3K4me3, and H3K27me3 at selected genes in IDHwt (n = 7) and control (n = 5) samples. The precipitation level was normalized to that obtained at the TBP promoter (for H3K4me3 and H3K9ac) and at the SP6 promoter (for H3K27me3; P-values calculated with the Mann–Whitney U test). (C) Detail for each sample of the ChIP analysis at the PCSK6 locus. Heatmaps of the expression and methylation values are in the upper panel. (D) Expression level of “Meth+/Exp−” (purple column) and “No Meth/Exp−” (blue column) genes and of all analyzed genes (white column) in healthy controls. (E,F) Principal component analysis. (E) Two-dimensional scatter plot of the values of each “Meth+/Exp−” (purple dots) and “No Meth/Exp−” gene (blue dots) along the first (Dim 1) and second (Dim 2) principal component. For each class defect, the centroids are shown by colored squares. (F) H3K4me3 and expression levels in healthy brain are the variables that most contributed to and were significantly correlated with the first principal component.

Figure 8.

Working model. (A) In glioma, alterations in the control of the H3K27me3 signature could be one of the main contributors to the four types of transcriptional defects observed at CGI/promoter-controlled genes (upper). In this model, genome-wide hypomethylation induces H3K27me3 redistribution that could lead to ectopic expression of genes that are normally repressed by polycomb proteins, including some genes encoding transcription factors. These overexpressed transcription factors could then promote the aberrant expression of their target genes (dotted arrow). Similarly, alterations in the interplay between the polycomb complex and the transcriptional machinery could affect H3K27me3 fate during ES and/or neural stem cell differentiation. Specifically, this alteration could lead to the aberrant maintenance of bivalency and silencing at a subset of genes that are normally specifically expressed in brain. At genes that are normally poorly expressed in healthy brain, this process is associated with gain of DNA methylation in glioma. Beside defects in the H3K27me3 signature, we also identified a subset of genes that are apparently constitutively associated with H3K4me3-only, regardless of their expression status in brain and glioma (lower). The mechanisms underlying their transcriptional deregulation remain to be determined. (B) Percentage of unaffected and affected CGI/promoter-controlled genes for each of the four described defects in IDHwt and IDHmut glioma samples from our cohort.