IDH1 mutation is sufficient to establish the glioma hypermethylator phenotype

Abstract

Both genome-wide genetic and epigenetic alterations are fundamentally important for the development of cancers, but the interdependence of these aberrations is poorly understood. Glioblastomas and other cancers with the CpG island methylator phenotype (CIMP) constitute a subset of tumours with extensive epigenomic aberrations and a distinct biology. Glioma CIMP (G-CIMP) is a powerful determinant of tumour pathogenicity, but the molecular basis of G-CIMP remains unresolved. Here we show that mutation of a single gene, isocitrate dehydrogenase 1 (IDH1), establishes G-CIMP by remodelling the methylome. This remodelling results in reorganization of the methylome and transcriptome. Examination of the epigenome of a large set of intermediate-grade gliomas demonstrates a distinct G-CIMP phenotype that is highly dependent on the presence of IDH mutation. Introduction of mutant IDH1 into primary human astrocytes alters specific histone marks, induces extensive DNA hypermethylation, and reshapes the methylome in a fashion that mirrors the changes observed in G-CIMP-positive lower-grade gliomas. Furthermore, the epigenomic alterations resulting from mutant IDH1 activate key gene expression programs, characterize G-CIMP-positive proneural glioblastomas but not other glioblastomas, and are predictive of improved survival. Our findings demonstrate that IDH mutation is the molecular basis of CIMP in gliomas, provide a framework for understanding oncogenesis in these gliomas, and highlight the interplay between genomic and epigenomic changes in human cancers.

Figure 1. Introduction of mutant IDH1 into human astrocytes remodels the methylome

a, Expression of wild-type and mutant IDH1 (R132H) in immortalized human astrocytes (passage 5). b, Overexpression of mutant IDH1 but not wild-type(WT) IDH1 in human astrocytes leads to production of 2-HG. Error bars show 1 standard deviation (s.d.) (n = 2). c, Self-organizing map (SOM) analysis of methylome data for wild-type IDH1-expressing, mutant IDH1-expressing (R132H), and parental (control) cell lines shows changes in the methylome in mutant IDH1-expressing and wild-type IDH1-expressing astrocytes, compared to parental cells. Mosaic patterns are pseudo-coloured SOMs from different time points (P indicates passage number). Tile colours indicate methylation level of centroids. d, Hierarchical clustering showing divergence of the methylome of IDH1-expressing astrocytes from that of parental astrocytes. MUT, mutant; PAR, parental. e, Heatmap showing the 10,678 most significant differentially methylated probes (ANOVA) in IDH1 mutant astrocytes and parental astrocytes (passages 2 and 40). Colour scale indicates β values. f, Kinetics of differential methylation in mutant and wild-type-expressing astrocytes. Error bars indicate inter-quartile range (n = 2).

Figure 2. Global epigenetic analysis of LGGs reveals dependence of G-CIMP on IDH mutation

a, Identification of G-CIMP by K-means consensus clustering of LGG samples. Unsupervised clustering was performed with the most variant probes (9,711 probes, top 2%). Tumours are listed in the same order along the x and y axes. G-CIMP status is indicated by the black and white bars. Consensus index values range from 0 to 1, with 0 being dissimilar (white) and 1 being similar (red). K = 2 is identified by the Lorenz curve. b, Two-dimensional (2D) hierarchical clustering of the same probes as in a identified the same two clusters. Each row represents a tumour and each column represents a probe. CIMP and IDH mutation status are indicated by the colour code. The level of DNA methylation (β value) for each probe is represented by colour scale (red, methylated; blue, non-methylated). Only cancer-specific events were used. c, Kaplan–Meier survival curve of Memorial Sloan-Kettering Cancer Center (MSKCC) patients (n = 72) with LGG (grade II and III). d, Receiver operating characteristic (ROC) curve comparing the sensitivity and specificity of G-CIMP status compared with MGMT methylation or MGMT expression status, in LGGs. Areas under the curve are noted in the inset. G-CIMP, MGMT methylation and MGMT expression were determined as described in Methods.

Figure 3. IDH1 mutation directly generates the methylation patterns present in G-CIMP tumours

a, The methylomes and transcriptomes of LGGs are distinct. PCA plot of LGG tumours for all methylation probes (left) and expression probes (right) (n = 52). PC, principal component. b, Starburst plot (left) for comparison of DNA methylation and gene expression. The log₁₀ (FDR-corrected P value) is plotted for β value forDNAmethylation (x axis) and gene expression (y axis) for each gene. Black dotted line shows the FDR-adjusted P value of 0.05. Red points indicate downregulated and hypermethylated genes in G-CIMP+ LGGs versus G-CIMP− LGGs. Blue points show hypomethylated and upregulated genes. Volcano (right) plot of all CpG loci analysed for G-CIMP association. The β-value difference in DNA methylation between G-CIMP+ and G-CIMP− tumours is plotted along the x axis. The P value between G-CIMP+ and G-CIMP− tumours is plotted on the y axis (−log₁₀ scale). Red indicates significantly different probes. c, Concordance between hypermethylated sites in mutant IDH1-expressing astrocytes and G-CIMP+ LGGs. GSEA shows significant enrichment between 730 hypermethylated unique CpG sites identified in IDH1 mutant astrocytes (ANOVA between passage 2 and 40) and those present in G-CIMP+ gliomas. GSEA correlation shown in colour scale. ES, enrichment score; FDR, false discovery rate; FEWR, familywise error rate; NES, normalized enrichment score; NOM, nominal P value. d, Differential methylation in IDH mutant astrocytes correctly classifies G-CIMP in the human LGGs. Two-dimensional-unsupervised hierarchical clustering of 81 human gliomas with top variant probes (n = 10,000) frommutant IDH1 astrocytes. Tumours are shown on the y axis, probes along the x axis. Methylation (β value) for each probe is represented with the colour scale. G-CIMP classification as determined by the astrocyte-derived data is denoted by the colour bars at the left. e, Kaplan–Meier survival curve of 115 patients with grade II or grade III gliomas in the Rembrandt Database grouped by CIMP status. P value calculated by log rank.

Figure 4. Functional implications of IDH1-mutation-induced alterations in the glioma epigenome

a, Concordance of transcriptional programs regulated by mutant IDH1 in astrocytes and G-CIMP in LGGs. P value for significance is shown along the x axis. Yellow lines indicate threshold of significance (P = 0.05). b, Mutant IDH1 results in the expression of markers of self-renewal and stem cell identity. Left, mutant IDH1 results in expression of nestin. P indicates passage number. Right, expression of mutant IDH1 promotes the adoption of a neurosphere phenotype. Astrocytes (passage 15) that express IDHR132H or IDH1 wild type were used in the neurosphere assay. Error bars indicate 1 s.d. **P < 0.01 (t-test). c, Alterations in histone marks in IDH1-mutant-expressing human astrocytes. Left, western blot results are shown using the indicated antibodies. Astrocytes are from passage 27. Right, ChIP of the indicated histone marks for representative hypermethylated genes. Error bars indicate 1 s.d. *P < 0.05. d, Mutant IDH1 inhibits the production of 5hmC in human astrocytes. Left, mutant IDH inhibits TET2-dependent 5hmC production in astrocytes. Parental astrocytes were infected with lentivirus directing the expression of TET2 catalytic domain and green fluorescent protein (GFP) ± mutant IDH1. FACS analyses are shown for 5hmC. Right, astrocytes expressing IDH R132H (passage 10) have less 5hmC than astrocytes that do not express the IDH mutant.