The isocitrate dehydrogenase genes IDH1
are mutated in >70%of lower-grade gliomas (grades II and III), in some glioblastomas4,5
, and in leukaemias and several other cancers6,7
. The most common IDH1
mutations in glioma (>95%) result in an amino acid substitution at arginine 132 (R132), which resides in the enzyme’s active site. Mutation of IDH imparts the ability to produce 2-hydroxyglutarate (2-HG), a potential oncometabolite8–10
. Alterations in the methylation landscape have been shown to have important roles during oncogenesis11
. CIMP has emerged as a distinct molecular subclass of tumours in a number of human malignancies, including glioblastoma1–3
. This phenotype is associated with extensive, coordinated hypermethylation at specific loci1,2,12,13
. In glioblastomas, G-CIMP is associated with the proneural subgroup of tumours and IDH mutation1
. Exactly how mutant IDH promotes tumorigenesis and causes G-CIMP—or CIMP in any type of human cancer—is unknown.
To determine whether IDH1
mutation directly causes G-CIMP, we used immortalized primary human astrocytes14
and constructed isogenic cells expressing either mutant IDH1 (R132H), wild-type IDH1, or neither. These astrocytes are well characterized14–17
. Introduction of wild-type IDH1 and the R132H IDH1 mutant resulted in equal expression of protein (modest threefold increase) (). Expression of mutant but not wild-type IDH1 in human astrocytes resulted in the production of 2-HG (). To determine whether mutant IDH1 altered the methylation landscape, we analysed genomic DNA from these cells using the Illumina InfiniumHumanMethylation450 platform. The platform provides genome-wide coverage and is both well validated and highly reproducible18,19
Introduction of mutant IDH1 into human astrocytes remodels the methylome
Previous data demonstrated that de novo
DNA methylation in in vitro
models occurs over extended periods, requiring time to ‘lock in’ epigenomic changes12,20
. We thus analysed the methylomes of astrocytes expressing mutant or wild-type IDH1 over successive passages (up to 50). Analysis using self-organizing maps demonstrated that mutant IDH1 progressively remodelled the glial methylome over time (), an effect that was not seen in control astrocytes. Expression of mutant IDH1 caused a marked increase in hypermethylation at a large number of genes, although there was a small group of hypomethylated genes as well ( and Supplementary Fig. 1a
and Supplementary Table 1
). Surprisingly, expression of wild-type IDH1 also reshaped the methylome but in a manner that differed from effects due to expression of mutant IDH1 (). Expression of wild-type IDH1 caused hypomethylation at specific loci, suggesting that both the production of 2-HG and the levels of α-ketoglutarate can affect the methylome. Unsupervised hierarchical clustering of the methylome data showed that the hypermethylated genes included both genes that underwent de novo
methylation as well as genes that originally possessed low levels of methylation but subsequently acquired high levels of methylation (). Control astrocytes did not undergo these methylome changes (). Mutant IDH1-induced remodelling of the methylome was progressive and reproducible, and resulted in significant changes in gene expression ( and Supplementary Fig. 1a
, Supplementary Tables 2
We sought to define the methylation targets of mutant IDH in astrocytes. Of the 44,334 CpG sites that were differentially methylated in mutant IDH-expressing cells, 30,988 sites were hypermethylated (3,141 unique genes with promoter CpG island methylation changes; Supplementary Table 1
). Transcriptional module mapping showed that the genes undergoing methylation changes were highly enriched for polycomb complex 2 (PRC2)-targeted loci (Supplementary Fig. 1b
and Supplementary Table 4)12,21
. These observations demonstrate that mutant IDH1 is sufficient to reshape the epigenome by altering the global methylation landscape.
Lower-grade gliomas (LGGs; World Health Organization grades II and III) and secondary glioblastomas are biologically distinct from primary or de novo
. Present knowledge of G-CIMP is based on the examination of primary glioblastomas in which IDH mutations are infrequent1,4,5
. To determine the impact of IDH mutation on the methylation landscape in primary LGGs, we generated a high-resolution, genome-wide set of LGG methylome data from patients with complete clinical follow-up using the same Infinium 450K platform as described earlier (72 WHO grade II and III gliomas; and Supplementary Table 5). We first performed consensus clustering ( and Supplementary Fig. 2a
) and unsupervised hierarchical clustering ( and Supplementary Fig. 2b
) to identify LGG subgroups. We identified two robust DNA methylation clusters, one encompassing tumours with markedly high methylation levels (cluster 2) and another without the hypermethylated loci (cluster 1). Cluster 2 tumours demonstrated a characteristic DNA methylation profile with high-coordinate cancer-specific methylation at a subset of loci, concordant with the G-CIMP phenotype defined in glioblastomas (Supplementary Fig. 2b
and Supplementary Table 6)1
. The composition of the G-CIMP group in these LGGs was confirmed by two independent clustering methods (K
-means consensus and two-dimensional hierarchical clustering) (). Probes defining CIMP in LGGs included those in CpG islands and shores (Supplementary Fig. 2c, d
) and were enriched for PRC2-target genes (Supplementary Table 7). Global expression profiles showed that G-CIMP+ tumours possessed markedly different transcriptional profiles than G-CIMP− tumours (Supplementary Tables 8 and 9). EpiTYPER (Sequenom) mass spectrometry was used to validate the methylation status of loci in both the astrocyte model and in the tumours (Supplementary Fig. 2e–g
Global epigenetic analysis of LGGs reveals dependence of G-CIMP on IDH mutation
To determine the mutational status of IDH1
, we sequenced the entire coding sequence of the two genes in all the samples above (). Ninety-eight per cent (49/50) of the G-CIMP+ tumours possessed either an IDH1
mutation or IDH2
mutation. Notably, none of the G-CIMP− tumours possessed mutant IDH (Supplementary Fig. 2h
). These genomic data show that G-CIMP is highly dependent on the presence of IDH mutation and, in LGGs that are CIMP−, IDH mutations do not occur (0%). Currently, the methylation status of O
-6-methylguanine DNA methyltransferase (MGMT) is a widely used molecular biomarker for glioblastoma prognosis and response to temozolomide24
. In LGGs, G-CIMP associated with markedly better clinical endpoints ( and Supplementary Figs 3– 6
, Supplementary Tables 10 and 11). Importantly, G-CIMP was significantly superior to MGMT methylation or MGMT
messenger RNA expression as a predictor of survival ().
We next sought to define the nature of the methylome differences between IDH mutant and wild-type tumours and characterize the effects of these differences on the LGG transcriptome. shows a principal component analysis (PCA) of methylome and expression data from our tumours. PCA shows that G-CIMP+ and G-CIMP− LGGs methylome subgroups correlate with marked transcriptome differences (). Of the 140,016 sites that were differentially methylated between IDH mutant and wild-type tumours, 121,660 were hypermethylated (Supplementary Table 6). There were 2,611 unique genes with alterations in promoter CpG islands represented in this group. Consistent with the results in , a volcano plot showing differentially methylated genes between G-CIMP+ and G-CIMP− tumours was highly asymmetric (). A starburst plot showing the relationship between DNA methylation and expression is shown in . Integration of the normalized gene expression and DNA methylation gene sets identified 429 genes with both significant hypermethylation and downregulation and 176 genes that were hypomethylated and upregulated in G-CIMP+ LGGs (Supplementary Table 12). Among these genes are those known to be involved in glioma initiation and outcome, including CDKN2C
(refs 25, 26
IDH1 mutation directly generates the methylation patterns present in G-CIMP tumours
As a critical experiment to prove causality between IDH1 mutation and G-CIMP, we performed an in-depth comparison of methylation marks and gene expression alterations between human astrocytes expressing mutant IDH1 and the LGGs with endogenous IDH1 mutation. We first focused on the comparison of methylation marks and found that both sets of methylome alterations targeted similar loci. Gene set enrichment analysis (GSEA) of the mutant IDH1-induced methylation changes in the isogenic astrocyte system () and the G-CIMP genes demonstrated very significant enrichment and concordance ( and Supplementary Table 13 and Supplementary Fig. 7
). Importantly, the genes that were methylated after mutant IDH1 expression correctly classified LGG tumours into CIMP+ or CIMP− groups with very high accuracy ( and Supplementary Table 14). To confirm the impact of these alterations on glioma pathobiology, we used the transcriptomic footprint of mutant IDH to generate an expression signature (mutant IDH repression signature) composed of the most significantly methylated and downregulated genes in both the isogenic astrocyte system and the G-CIMP gene set (17 genes; Supplementary Table 15). As expected, this signature classified an independent LGG cohort (Rembrandt) into two distinct subgroups ( and Supplementary Figs 8–10
and Supplementary Table 16). Together, our findings show that introduction of mutant IDH reprograms the epigenome and generates the foundations of G-CIMP.
IDH mutation is highly enriched in the CIMP+, proneural subgroup of glioblastomas. Using data from The Cancer Genome Atlas (TCGA), we applied the mutant IDH repression signature as a classifier to the transcriptomes of all four subgroups of glioblastomas27
. The signature segregated IDH mutant and wild-type proneural glioblastomas into two distinct subgroups associated with very different prognoses, but did not do so in other glioblastoma subgroups (Supplementary Fig. 11a, b
). These data demonstrate that mutant IDH-induced epigenomic alterations have profound biological implications within the proneural class of glioblastomas that are specific for this subclass. Comparison of gene expression programs that occur in astrocytes expressing mutant IDH1 to those in LGG tumours that harbour the IDH mutation showed remarkable similarity ( and Supplementary Fig. 12
). Moreover, introduction of mutant but not wild-type IDH1 into astrocytes resulted in the upregulation of nestin (and other genes associated with stem cell identity) at the time of DNA methylation increase and the adoption of a neurosphere/stem-like phenotype ( and Supplementary Fig. 13
. These data suggest that mutant IDH1 functions by interfering with differentiation state.
Functional implications of IDH1-mutation-induced alterations in the glioma epigenome
Our data show that IDH1 mutation is the mechanistic cause of G-CIMP. To gain further insight, we determined the effects of mutant IDH1 on histone alterations in our astrocyte system. (left) shows that expression of the IDH1 mutant increases levels of H3K9me2, H3K27me3 and H3K36me3, consistent with previous findings29
. Chromatin immunoprecipitation (ChIP) experiments examining representative genes that undergo hypermethylation show H3K9 and H3K27 methylation are both enriched in cells expressing mutant IDH1 (, right). As both of these marks can promote DNA methylation, alterations in histone marks may contribute to the accumulation of DNA methylation.
Next, we determined the effects the mutation had onTET2-dependent 5-hydroxymethylcytosine (5hmC) levels. We used a well-established assay9,29
and first confirmed that we were able to detect TET-dependent alterations in 5hmC (Supplementary Fig. 14
). We found that expression of the IDH1 mutant in astrocytes resulted in a significant decrease in 5hmC (, right). Expression of TET2 in the astrocytes produced 5hmC, which was inhibited by mutant but not wild-type IDH1 (, left). Because TET-mediated production of 5hmC is a primary mode of DNA demethylation30
, inhibition of this activity in the IDH1-mutant-expressing astrocytes may be a mechanistic basis for accumulation of DNA methylation, ultimately leading to a CIMP pattern.
IDH mutation and the CIMP phenotype are two very common features in cancer, the underlying mechanisms for which are obscure. The fundamental questions regarding these features are (1) how the IDH mutation contributes to oncogenesis, and (2) what the root cause of CIMP is. Our data address these important questions by demonstrating that IDH mutation is the cause of CIMP and leads to the CIMP phenotype by stably reshaping the epigenome. This remodelling involves modulating patterns of methylation on a genome-wide scale, changing transcriptional programs and altering the differentiation state. Our observations suggest that the activity of IDH may form the basis of an ‘epigenomic rheostat’, linking alterations in cellular metabolism to the epigenetic state. In summary, these data provide a mechanistic framework for how IDH mutation leads to oncogenesis and the molecular basis of CIMP in gliomas. We believe our observations have critical implications for the understanding of gliomas and the development of novel therapies for this disease.