• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of neuroncolAboutAuthor GuidelinesEditorial BoardNeuro-Oncology
Neuro Oncol. Jan 2011; 13(1): 84–98.
Published online Oct 5, 2010. doi:  10.1093/neuonc/noq110
PMCID: PMC3018904

Methylation profiling identifies 2 groups of gliomas according to their tumorigenesis

Abstract

Extensive genomic and gene expression studies have been performed in gliomas, but the epigenetic alterations that characterize different subtypes of gliomas remain largely unknown. Here, we analyzed the methylation patterns of 807 genes (1536 CpGs) in a series of 33 low-grade gliomas (LGGs), 36 glioblastomas (GBMs), 8 paired initial and recurrent gliomas, and 9 controls. This analysis was performed with Illumina's Golden Gate Bead methylation arrays and was correlated with clinical, histological, genomic, gene expression, and genotyping data, including IDH1 mutations. Unsupervised hierarchical clustering resulted in 2 groups of gliomas: a group corresponding to de novo GBMs and a group consisting of LGGs, recurrent anaplastic gliomas, and secondary GBMs. When compared with de novo GBMs and controls, this latter group was characterized by a very high frequency of IDH1 mutations and by a hypermethylated profile similar to the recently described glioma CpG island methylator phenotype. MGMT methylation was more frequent in this group. Among the LGG cluster, 1p19q codeleted LGG displayed a distinct methylation profile. A study of paired initial and recurrent gliomas demonstrated that methylation profiles were remarkably stable across glioma evolution, even during anaplastic transformation, suggesting that epigenetic alterations occur early during gliomagenesis. Using the Cancer Genome Atlas data set, we demonstrated that GBM samples that had an LGG-like hypermethylated profile had a high rate of IDH1 mutations and a better outcome. Finally, we identified several hypermethylated and downregulated genes that may be associated with LGG and GBM oncogenesis, LGG oncogenesis, 1p19q codeleted LGG oncogenesis, and GBM oncogenesis.

Keywords: glioblastoma, low-grade glioma, methylation profile

Gliomas, the most frequent primary brain tumors, are a heterogeneous group.1 Low-grade gliomas (LGGs; World Health Organization [WHO] grade II) progress slowly over many years, but glioblastomas (GBMs; WHO grade IV) are usually fatal within 12–18 months. Most gliomas are incurable. Therefore, better molecular characterization of these tumors is required to improve both their classification and their treatment. Extensive genomic and transcriptomic studies have already been performed on gliomas, but their epigenetic alterations remain to be characterized.25 DNA methylation is a highly regulated process in normal cells and becomes drastically modified in cancer cells.6 Hypermethylation of tumor suppressor gene promoters is a common mechanism for gene silencing.7 It is often associated with global hypomethylation of pericentrometric DNA sequences, leading to tumorigenesis.8 Several studies have reported aberrantly methylated genes in gliomas including TP53 and p14ARF,9 PTEN,10,11 p16/CDKN2A,12 PCDH-g-A11,13 PEG3,14 DIRAS3,15 EMP3,16 and CASP8.17 Among them, MGMT has been the most studied.18 Specific changes in methylation associated with different subtypes of gliomas also have been reported.1822 However, these studies were limited to only a few samples or a few genes,1922 or to only 1 subtype of glioma.18 More recently, the study of the methylation profile of the GBM samples included in the TCGA (The Cancer Genome Atlas) has enabled the identification of a glioma CpG island methylator phenotype (G-CIMP) that has been shown to be associated with IDH1 mutations and improved outcome.4

In the present study, we used an array-based method (Illumina's high-throughput single-nucleotide polymorphism [SNP] genotyping system)23 to analyze the methylation pattern of 1536 CpGs corresponding to 807 genes in a large set of LGGs and GBMs. The first aim was to investigate whether these 2 very different types of gliomas display different methylation profiles and whether the subgroups of GBMs and LGGs can be identified according to their methylation profile. The second aim was to identify new genes associated with LGG and GBM oncogenesis. For this purpose, we correlated methylation and expression levels to identify genes that are trascriptionally regulated by epigenetic alterations.

Materials and Methods

Patients

In total, 36 GBMs and 33 LGGs were selected based on tumor material as well as clinical and molecular data availability. Among GBMs, the median age was 54 years (range: 25–85), median survival was 13.9 months (95% CI: 12.5–19 months), and median follow-up was 13 months (range: 0.5–57). LGGs included 15 grade II oligodendrogliomas, 13 grade II oligoastrocytomas, and 5 grade II astrocytomas according to the WHO 2007 classification. Among LGGs, the median age was 36 years (range: 16–59). Median survival was not reached after a median follow-up of 32 months (range: 2–60). Nine samples from nontumor brains were also included as controls (normal brain [n = 2], amyotrophic lateral sclerosis [n = 2], and epileptic surgery [n = 5]). A second set of 8 paired initial and recurrent gliomas was also studied. This set consisted of 4 LGGs that progressed to an anaplastic glioma, 1 LGG and 2 anaplastic gliomas progressing to a secondary GBMs, and 1 initial and recurrent de novo GBM. All samples were studied on methylation arrays and comparative genomic hybridization (CGH) arrays. A subset of samples (17 LGGs, 29 GBMs, and 3 controls) was also studied on Affymetrix gene expression arrays.

DNA Methylation Profiling

DNA was extracted from frozen samples using a standard protocol (Qiagen, QIAmp DNA minikit). A total of 500 ng of DNA were bisulfite converted using the Gold DNA methylation kit (Zymo Research) according to the manufacturer's instructions. Methylation was detected for 1505 CpG sites using the Illumina GoldenGate Assay with BeadArray technology as described previously.23 This method quantifies the methylation level of CpGs mostly located between −500 and +500 bp from the transcription start site of 807 genes selected for their relevance in cancer studies. After bisulfite treatment, the assay was carried out according to the standard procedures of the GoldenGate genotyping assay24 using Illumina-supplied reagents and conditions. The array hybridization was conducted under a temperature gradient program, and the array was imaged using a BeadArray Reader.25 Image processing and intensity data extraction were performed and the data used as described previously.26 Each methylation data point represents the fluorescent signals from the M (methylated) and U (unmethylated) alleles. Background intensity, computed from a set of negative controls, was subtracted from each analytical data point. The ratio of fluorescent signals was then computed from the 2 alleles to reflect the fractional methylation level at each CpG site (β-value). Bead arrays provide a β-value between 0 and 1 as the proportion of methylation for a given CpG site.27,28 To avoid any gender-specific bias, X-linked genes were excluded from the analysis. Probes with a detection P-value higher than .05 in more than 5 samples and those containing an SNP within 10 bp of the targeted CpG were also excluded.

Methylight Assay

MGMT, RASSF1, and RBP1 methylation were studied on 25 out of the 33 LGGs and on 34 out of the 36 GBMs for which enough DNA was available. After sodium bisulfite conversion, genomic DNA was amplified by fluorescence-based, real-time quantitative PCR as described previously.29 Briefly, bisulfite-converted genomic DNA was amplified using locus-specific PCR primers flanking an oligonucleotide probe with a 5′ fluorescent reporter dye (6FAM) and a 3′ quencher dye (TAMRA). After crossing a fluorescence detection threshold, the PCR amplification yields a fluorescent signal proportional to the amount of PCR product generated. Control samples were included on each plate. PCR was performed using a 96-well tray and caps. The final reaction mixture had a volume of 25 µL and consisted of 600 nM of each primer, 200 nM probe, 200 µM deoxynucleotide triphosphate, and 12.5 µL of 2× buffer (Absolute QPCR ROX Mix, Abgene) containing a reference dye and 1 µL of bisulfite-converted DNA. The following conditions were used: 50°C for 2 minutes, 95°C for 10 minutes, followed by 45 cycles at 95°C for 15 seconds, and 60°C for 1 minute. Two sets of primers and probes, specifically designed for bisulfite-converted DNA, were used: a methylated set for each tested gene and a reference set, β-actin (ACTB), to normalize the input DNA. The specificity of the reactions for methylated DNA was confirmed separately using DNA extracted from lymphocytes of healthy individuals as a negative control and DNA extracted from the SNB-19 cell line (100% methylated) as a positive control. The percentage of methylated molecules at the target locus was calculated by dividing the target/ACTB ratio of a tumor by the target/ACTB ratio of SNB-19 DNA. The primers and the probes used are listed in Table 1 and correspond to those published previously.30

Table 1.
Short list of genes differentially methylated in both LGGs and GBMs in comparison to controls

CGH Array, Gene Expression, and Genotyping

CGH array was performed as described previously31 on arrays (Integragen) containing about 4500 sequence-validated bacterial artificial chromosomes (spotted in quadruplicate on the array) that cover the genome with a mean resolution of 0.7 Mb. Gene expression analysis was performed on a pangenomic Genechip Human Genome U133 Plus 2.0 expression array (54.000 probe sets; Affymetrix) as described previously.32 TP53, PTEN, and IDH1 mutations were assessed as reported previously.33

Statistical Analysis

All statistical analyses were carried out using the R-system software v.2.10.0 (http://www.r-project.org/) with Bioconductor packages (http://www.bioconductor.org), Center for Infomation Technology (CIT) packages, Comprehensive R Archive Network (CRAN) packages (http://cran.r-project.org/), and original R code (J.L.). Methylation data were managed and normalized with functions provided by the ‘Methylumi’ Bioconductor package. Distributions of β-values were plotted using the R function ‘density’. The χ2 and the Kolmogorov–Smirnov tests were applied to compare the β-value distributions between samples. Hierarchical clustering was based on Pearson's correlation and Ward's distance. Fisher's exact test was used to compare the annotations of the different clusters. Principal component analysis (PCA) was performed using the ‘ade4’ package, and methylation variability was measured with the standard variation. Methylation difference levels were assessed for each CpG by the Wilcoxon test (α = 5%). As an additional filter, we considered only CpGs that had a difference in median methylation levels higher than 0.17 and larger than twice the standard deviation in tumor samples. Genes differentially expressed were assessed by the unpaired t-test (α = 5%). In order to classify our samples according to Verhaak et al.'s system,5 we used their published centroid-based classifier. Survival curves were calculated according to the Kaplan–Meier method, and differences between curves were assessed using the log-rank test. TCGA GBM data set (http://cancergenome.nih.gov/) was used as an independent data set to assess DNA methylation array reproducibility, to perform a hierarchical clustering with the samples of the present study, and to assess the genes found to be regulated by methylation in the GBM samples of the present study.

Results

The methylation profiles of 33 LGG tumors, 36 GBMs, 8 paired initial and recurrent gliomas, and 9 nontumor brain tissues were obtained using the Illumina GoldenGate bead array.

DNA Methylation Array Reliability and Reproducibility

To assess the reliability of Illumina's bead arrays, we compared the results obtained with this technology with results obtained using the Methylight method for the methylation status of 3 genes (MGMT, RASSF1, and RBP1). Bead arrays provide a β-value between 0 and 1 that reflects the proportion of methylation for a given CpG.23,27 Methylight provides a methylated or unmethylated status for each gene. As shown in Fig. 1A–C, there was a good correlation between both techniques, as the median β-value was significantly higher in methylated samples than in unmethylated samples as defined by Methylight (MGMT [Δβ = 0.18, Wilcoxon test, P < 10−3], RASSF1 [Δβ = 0.72, P < 10−6], and RBP1 [Δβ = 0.82, P < 10−7]). Moreover, when considering a CpG with a β-value higher than 0.2 as methylated, both methods were in agreement for 70%, 81%, and 92%, respectively. However, as shown in Fig. 1A–C, Illumina's bead array sensitivity and specificity were variable according to the gene studied. In comparison with Methylight, Illumina's bead arrays had a high specificity but a low sensitivity for MGMT, a high sensitivity but a medium specificity for RASSF1, and a high specificity and a high sensitivity for only RBP1. As shown in Fig. 1D–F, Illumina's bead arrays underestimated MGMT methylation in both LGG and GBM, whereas it overestimated RASSF1 methylation in LGG. These differences might be explained by the fact that Methylight assesses the methylation status of a DNA segment potentially containing several CpGs, whereas Illumina's probes target only 1 specific CpG.

Fig. 1.
Comparison with Methylight. (A–C) Methylation levels (β-value) of MGMT, RASSF1, and RBP1 were compared with the methylation status obtained using Methylight (U for unmethylated and M for methylated). For each gene, the β-values ...

To validate the reproducibility of the bead arrays, we compared the methylation profiles in our GBM data set with publicly available data from TCGA (http://cancergenome.nih.gov/), which used the same bead arrays to study the methylation profiles of a series of 220 GBMs (http://cancergenome.nih.gov/). For this purpose, the median β-value for each probe in each data set was calculated and compared. We found that 95% of CpGs presented an absolute difference in their median β-value ≤0.17, which is the minimum difference that can be significantly observed on this array (Supplementary Material, Fig. S1).21 The global Pearson's correlation between the 2 vectors of median β-values was 0.95 (Supplementary Material, Fig. S1), showing that Illumina's bead array is a highly reproducible technology.

LGGs Display a Hypermethylated Profile

After filtering the data, the analysis was conducted on the methylation level of 1243 CpGs corresponding to 731 unique genes in 36 GBMs, 33 LGGs, and 9 nontumoral brain control samples. Among the 1243 CpGs, 869 (70%) were located in CpG islands that were often within gene promoter regions and known to regulate gene expression, and 374 (30%) were located in “shores,” up to 2 kb from islands, which have also been implicated in cancer gene expression regulation.6,34 As shown in Fig. 2, regardless of the sample type, CpGs within (CpGi) and outside (CpGo) CpG islands displayed 2 opposite methylation profiles. In agreement with the methylation status of CpGs in the human genome,6 about 65% of CpGi were totally unmethylated (β ≤ 0.20) and 65% of CpGo were completely methylated (β ≥ 0.8). CpGi were hypermethylated in both LGG and GBM compared with control samples (P < 2 × 10−16). Consistent with what has been reported for many cancers including gliomas, CpGo were hypomethylated in GBM compared with controls (P < 2 × 10−16).35 In contrast, CpGo were hypermethylated in LGG when compared with both control samples (P < 4 × 10−11) and GBMs (P < 2 × 10−16; Fig. 2). This finding was unexpected and suggests that LGGs have a specific hypermethylated profile.

Fig. 2.
Distribution of β-values in LGG, GBM, and control samples. The distribution of the β-values is given as a density plot for CpG within (A) and outside (B) CpG Islands according to sample types. The β-value, between 0 and 1, indicates ...

Identification of 2 Groups of Gliomas Characterized by Distinct Pathways of Tumorigenesis

To further study the differences between LGG and GBM methylation profiles, we performed an unsupervised hierarchical clustering of the 33 LGGs, 36 GBMs, and 9 controls using the methylation level of the 1243 CpGs (Fig. 3). Eight paired initial and recurrent gliomas were also included in this analysis (4 LGGs progressing to grade III gliomas, 1 LGG and 2 grade III gliomas progressing to a secondary GBM, and 1 recurrent de novo GBM).

Fig. 3.
Unsupervised hierarchical clustering of the 33 LGGs, 36 GBMs, 8 paired initial and recurrent gliomas, and 9 controls according to their methylation profile. Genes (lines) and samples (columns) are ordered according to a hierarchical clustering. β-values ...

As shown in Fig. 3, the samples were classified into 3 main clusters: a cluster of GBMs, a cluster of controls, and a cluster of LGGs (P < .001). Samples in the LGG cluster displayed a hypermethylated profile in comparison with both GBMs and controls (Fig. 3). The controls displayed a very homogeneous profile despite their different origins (normal brain, epilepsy, and amyotrophic lateral sclerosis). Four LGG samples clustered with the controls probably because they were contaminated with a non-neoplastic tissue. Beside histology, the main difference between the LGG and the GBM clusters was the frequency of IDH1 mutation. Indeed, there was no IDH1 mutation in the GBM group (0 of 35), whereas 93% (42 of 45) of samples in the LGG group had an IDH1 mutation (P < .001). Strikingly, the 5 GBMs with an IDH1 mutation (including 3 secondary GBMs) and the 6 IDH1 mutated anaplastic gliomas clustered into the LGG group, suggesting that the methylation profile varies not only according to histology but also according to glioma genetic characteristics. It also suggested that primary and secondary GBMs might be distinguished according to their methylation profile.

Other differences between the LGG and GBM clusters included a higher frequency of MGMT methylation as assessed by the Methylight technique (but not by the methylation array) in the LGG group (82% vs 48%; Fisher's test, P = .005), whereas EGFR amplification was observed exclusively in the GBM group (58% vs 0%, P < .001). When considering the samples for which the gene expression profile was available, according to the Verhaak et al. classification, there was a higher frequency of samples classified as proneural (13/20 vs 3/29, P < .001) and as neural (5/20 vs 1/29, P = .03) in the LGG group, whereas the GBM group was enriched in samples classified as mesenchymal (15/29 vs 2/20, P = .003) and as classical (10/29 vs 0/20, P = .003). Within the LGG group, we could not identify any differences among histological subtypes (oligodendroglial, astrocytic, and oligodendroglial) but we observed that 12 of 13 samples with a 1p19q codeletion formed a distinct subcluster (Fig. 3). Though heatmap inspection did not show consistent differences between 1p19q and non-1p19q codeleted LGGs, PCA of the methylation profile did distinguish these 2 groups of gliomas, suggesting that these 2 groups of LGGs really display a distinct methylation profile (Fig. 4). This result also suggests a strong relationship between the methylation profile and particular genetic characteristics of gliomas. Among GBMs, we could not identify an association between a particular subcluster and clinical or molecular characteristics including survival, PTEN and TP53 mutations, EGFR amplification, CDKN2A deletions, or MGMT methylation status. However, when considering samples for whom gene expression profiles were available, 1 subcluster was enriched in GBM classified as classical (9/17 vs 1/12, P = .02), whereas the other cluster was enriched in GBM classified as mesenchymal (11/12 vs 4/13, P < .001).

Fig. 4.
PCA of LGG samples. PCA distinguishes LGG samples with and without 1p19q codeletion.

Interestingly, when considering initial and recurrent tumors, we noticed that they had very similar methylation profiles. Indeed, except for 1 case, all recurrent tumors clustered together with the corresponding initial tumor, suggesting that the methylation profile of gliomas remains remarkably stable across their evolution, even at the time of anaplastic transformation.

Thus, at least with this set of genes, methylation profiling identified 2 groups of gliomas characterized by 2 distinct pathways of tumorigenesis: a group corresponding to de novo GBM and a group corresponding to LGG, recurrent anaplastic gliomas, and secondary GBM characterized by a hypermethylated profile and a high rate of IDH1 mutations.

GBMs with a Hypermethylated LGG-Like Profile Have a High Rate of IDH1 Mutations and a Better Prognosis

In order to further study the characteristics of the GBM that displayed an LGG-like hypermethylated profile, we used TCGA GBM GoldenGate data set to perform a new hierarchical clustering with these GBM samples and our data set. As shown in Fig. 5, we observed 2 main clusters: one large GBM cluster consisting of our de novo GBM samples and of most of TCGA GBM samples (201 of 220) and a second cluster consisting of our LGG samples, our IDH1 mutated GBM and grade III glioma samples, and 19 TCGA GBM samples. Interestingly, consistent with our findings, the 19 TCGA GBM samples that clustered with our LGG samples were more frequently classified as secondary GBM (Fisher's test, P = .006), had a much higher rate of IDH1 mutation (Fisher's test, P < 10−6), and were more frequently classified as proneural according to the Verhaak et al. classes (Fisher's test, P < 10−6) than TCGA GBM samples that clustered with our de novo GBM samples. These GBM samples had also a longer overall survival than TCGA GBM samples which clustered into the de novo GBM cluster (47.8 vs 10.9 months, P = 2 × 10−5, Fig. 6).

Fig. 5.
Unsupervised hierarchical analysis with the 220 TCGA GBM samples. Genes (lines) and samples (columns) are ordered according to a hierarchical clustering. β-values increase from green (unmethylated) to yellow to red (methylated). The following ...
Fig. 6.
Overall survival of GBM samples with an LGG-like methylation profile in TCGA data set. Kaplan–Meier survival curve of TCGA GBM samples demonstrating that TCGA GBM samples with a hypermethylated LGG-like profile (gray dashed curve) had a longer ...

Identification of Candidate Genes Implicated in LGG and GBM Oncogenesis

Finally, to identify genes differentially methylated among LGG, GBM, and controls, a Wilcoxon test was performed, and an additional filter was applied to select only genes with substantially increased or decreased methylation. To study the impact of methylation on the expression level of differentially methylated genes, the gene expression profiles of a subset of samples (17 LGGs, 29 GBMs, and 3 controls) were studied on Affymetrix U133v2 microarrays.

Twenty-three genes were differentially methylated in more than 20% of LGG and GBM samples when compared with controls (Table 2, Supplementary Material, Table S1). TRIP6, GFAP, and LEFTY2 were hypomethylated and significantly overexpressed in both LGG and GBM. The list of hypermethylated genes included several potential tumor suppressor genes, genes frequently silenced in cancers, genes known to inhibit tumor progression, and genes implicated in apoptosis. Five genes were downregulated in both LGG and GBM, including the potential tumor suppressor gene NBL1 as well as FZD9, which have been demonstrated to be silenced in several cancers (Table 2).

Table 2.
Short list of genes differentially methylated in both LGGs and GBMs compared with controls

Overall, 76 genes were completely unmethylated (n = 3) or methylated (n = 73) in more than 20% of LGG samples in comparison with controls. Two genes, PDGFRA and DDR1, were hypomethylated and overexpressed (Table 3, Supplementary Material, Table S1). The list of hypermethylated genes consisted of several potential tumor suppressor genes, genes reported to be silenced in cancers, and genes implicated in apoptosis or known to inhibit tumor progression. Most of these genes have not been implicated in gliomagenesis. Interestingly, several oncogenes were also found to be hypermethylated (eg, FGFR3, KIT, JAK3, MET, and TGFB2). Eighteen genes were hypermethylated and downregulated, including the potential tumor suppressors CAV2, FABP3, and TMEFF1 (Table 3). Additionally, 12 genes were differentially methylated between LGGs with and without 1p19q codeletions, with 2 genes being hypermethylated and downregulated in 1p19q codeleted LGGs (FGFR2 and TNFRS1B; Supplementary Material, Table S1). These genes might be implicated in alternative mechanisms of gliomagenesis in 1p19q codeleted LGGs or might be the markers of a different histogenesis.

Table 3.
Short list of genes differentially methylated in LGGs in comparison to controls

In total, 28 were completely unmethylated (n = 15) or methylated (n = 13) in more than 20% of GBM samples compared with controls (Table 4, Supplementary Material, Table S1). Eight genes were hypomethylated and upregulated, including several genes implicated in tumor progression (eg, MMP2 and MMP9; Table 4). The list of hypermethylated genes included the candidate tumor suppressors TUSC3 and GATA6 as well as several genes previously reported to be silenced in cancers. Six genes were hypermethylated and downregulated, including the candidate tumor suppressor TUSC3.

Table 4.
Short list of genes differentially methylated in GBMs in comparison to controls

In order to validate in an independent series the list of genes that seemed to be regulated by methylation in GBM vs normal brain, we studied the expression and methylation of these genes in TCGA GBM data set. This analysis demonstrated that except for 1 gene, SEMA3C, 20 out of the 21 genes that were found to be hypomethylated and overexpressed or hypermethylated and downregulated in our series of GBM had a similar pattern of expression and methylation in TCGA GBM samples (Supplementary Material, Table S2).

Discussion

The present study shows that the Illumina GoldenGate bead array is a powerful method for classification purposes, allowing the identification of specific methylation profiles related to different ways of tumorigenesis. However, due to the heterogeneity of methylation, the sensibility and specificity of methylation arrays can be variable among the genes. Therefore, as demonstrated for the assessment of MGMT methylation, results obtained on individual genes need to be validated using another technique.

Our first objective was to investigate whether LGG and GBM displayed different methylation profiles and whether the subgroups of GBM and LGG could be identified according to their methylation profiles. Interestingly, this has led us to conclusions very similar to those reached recently by Noushmehr et al.4 These authors performed an unsupervised hierarchical clustering of the GBM included in TCGA and found that a subgroup was characterized by a concerted hypermethylation at a large number of loci, reminiscent of the CpG island methylator phenotype described in colorectal cancers. They showed that in GBM, this G-CIMP was tightly associated with IDH1 mutations, with a proneural gene expression profile, and with prolonged survival. Most secondary GBMs were G-CIMP positive. Using the Methylight technique, they further studied the methylation profile of 8 genes characteristic of the G-CIMP in an independent series of gliomas and demonstrated that the G-CIMP was much more frequent in grade II/III gliomas than in GBM and was also associated with IDH1 mutations and a better outcome in these tumors. They also showed that the G-CIMP was stable across recurrence. In the present study, the initial strategy was to study the methylation profiles of GBM and LGG samples. We observed that they were dramatically different from each other and also clearly different from nontumor brain samples. LGG displayed a characteristic hypermethylated profile.4 Unsupervised analysis led us to the observation that the methylation profile of gliomas did vary not only according to the histology but also according to the genetic characteristics. Indeed, the glioma samples were classified into 2 clusters: a GBM cluster without IDH1 mutations consisting of only de novo GBM and an LGG cluster with 90% IDH1 mutated samples in which the GBM, including secondary GBM, and anaplastic gliomas with IDH1 mutations were classified. We also observed that the methylation profile of gliomas was remarkably stable across evolution, even at the time of anaplastic transformation. Finally, using TCGA data set, we found that TCGA GBM samples with an LGG-like hypermethylated profile that clustered with our LGG samples were characterized by a high rate of IDH1 mutations and had dramatic longer survival than the GBM samples that clustered into the de novo GBM cluster. Thus, our findings are consistent with those by Noushmehr et al.,4 and actually what we call here a hypermethylated profile does correspond to the G-CIMP described by these authors. It shows that epigenetic alterations occur early in gliomagenesis and characterize at least 2 distinct pathways for tumorigenesis: 1 involving LGG, anaplastic gliomas, and secondary GBM evolving from prior LGG (mostly IDH1 mutated), which are characterized by a G-CIMP/hypermethylated profile, and another encompassing de novo GBM. It now remains to be seen whether methylation profiling might identify other subgroups of gliomas. In the present study, LGGs with and without 1p19q codeletion seemed to have distinct methylation profiles. Concerning GBM, as previously noticed, classical and mesenchymal GBM also seemed to have different methylation profiles.4 Therefore, it is possible that the use of larger methylation arrays will allow the identification of other subgroups of gliomas. A large-scale methylation analysis also remains to be performed in anaplastic gliomas. In the present study, the set of paired initial and recurrent tumors included 6 anaplastic gliomas, all with an IDH1 mutation that clustered into the LGG cluster.

The second aim of this study was to identify new genes associated with gliomagenesis. For this purpose, we correlated the methylation status with the gene expression profile to identify hypomethylated/overexpressed genes and hypermethylated/downregulated genes. Among the differentially methylated genes, 27% displayed concordant under- or overexpression. We found that 20 of the 21 genes that seemed to be regulated by methylation in our GBM samples had a similar pattern of expression and methylation in TCGA GBM samples. However, the list of genes identified here will have to be validated by other techniques and their role in gliomagenesis will have to be assessed.

Despite their different methylation patterns, we identified a set of differentially methylated genes in both LGG and GBM. TRIP6, which regulates cell migration and has been demonstrated to promote cancer progression in colon cancer, was both hypomethylated and overexpressed in LGG and GBM.34 Among the hypermethylated genes, 5 (TES, MEST, CD81, FZD9, and TNRFS10A) were recently identified to be the most frequently hypermethylated genes in GBM but were not known to be hypermethylated in LGG.18 One of these genes, FZD9, was also downregulated in our series. FZD9 is involved in Wnt signaling and has been demonstrated to be silenced in most hematologic malignancies.35 Among the other hypermethylated and downregulated genes in LGG and GBM, NEBL1, AATK, and MAPK10 displayed interesting characteristics. NEBL1 is a potential tumor suppressor gene in neuroblastomas.36 AATK37 and MAPK1038 have not been implicated in cancer but play critical roles in neuronal apoptosis.

A large set of genes was differentially methylated in LGG compared with controls, including several potential tumor suppressor genes. Several oncogenes were also hypomethylated and overexpressed (namely PDGFRA and DDR1)39,40 in addition to several oncogenes that were hypermethylated and downregulated, such as the oncogenes FGFR3, KIT, MET, and TGFB2 (Table 2). The silencing of these oncogenes might play a role in the slower evolution of LGG when compared with GBM. Among the list of hypermethylated and downregulated genes in LGG, only MATK, THBS1, and TMEF have already been reported to be methylated or involved in gliomagenesis. MATK has been demonstrated to inhibit tumor progression through the inhibition of Src signaling41 and is silenced in gliomas.42 Furthermore, its expression inhibits the growth of astrocytoma cell lines.42 THBS1 has an antiangiogenic role and is hypermethylated in LGG.19 TMEFF is involved in growth factor signaling and has been suggested to be a potential tumor suppressor gene in brain tumors where it is downregulated.43 However, downregulation had not been linked with hypermethylation until now. Concerning the other hypermethylated and downregulated genes, CAV2 and FABP3 are potential tumor suppressor genes in prostate cancer as well as breast cancer and are silenced in other types of cancers.4448 CCNA1 binds to important cell cycle regulators and is silenced in several cancers, including cancers of the head and neck in which CCNA1 hypermethylation is inversely correlated with TP53 mutation status.49 RBP1 is frequently hypermethylated in lymphoma and gastro-intestinal cancers.50

We also identified 2 genes, FGFR2 and TNFRFS1B, specifically hypermethylated and downregulated in LGG with 1p19q codeletion in comparison with LGG without 1p19q codeletion. FGFR2 acts as an oncogene in several cancers, including breast cancer,51 but its silencing is oncogenic in thyroid cancers.52 Additionally, FGFR2 is silenced in pituitary adenomas.53 TNFRSF1B is implicated in neuronal and oligodendroglial apoptosis.54

In GBM, we identified 5 genes that were hypomethylated, overexpressed, and known to promote GBM proliferation and/or invasiveness: MMP2,55 MMP14,55 MMP9,55 SERPINE1,56 and SPP1.57 MMP9 has been previously demonstrated to be hypomethylated in GBM,18 and hypomethylation of MMP14 and MMP2 has been shown to induce their overexpression in GBM cells.58 All hypermethylated and downregulated genes, FLT3, HTR1B, NEFL, TUSC3, and ZNF215, have been recently reported to be hypermethylated in GBM.18 HTR1B is silenced in lung cancer and is deleted as well as downregulated in renal cancer and lymphoma.5961 TUSC3 is a candidate tumor suppressor thought to encode a subunit of the endoplasmic reticulum–bound oligosaccharyltransferase complex that catalyzes a pivotal step in the protein N-glycosylation process and is implicated in Mg2+ uptake.62 TUSC3 has been shown to be deleted or silenced in several cancers.63

Conclusion

This study is the first genome-wide study assessing the methylation profiles of different subtypes of gliomas. It identifies 2 groups of gliomas according to their methylation profile: a group of de novo GBMs and a group consisting of LGG as well as high-grade gliomas evolving from a prior LGG. Moreover, we found that the methylation profile of gliomas is remarkably stable across their evolution. By correlating the methylation profile and gene expression profile, we identified several genes that may play an important role in gliomagenesis.

Funding

J.L. was supported by the Ligue Nationale Contre le Cancer.

Acknowledgments

This work is part of the Carte d'Identité des Tumeurs (CIT) program (http://cit.liguecancer.net/index.php/en) from the Ligue Nationale Contre le Cancer. The results published here are in part based on data generated by TCGA pilot project established by the NIH and National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at their website (http://cancergenome.nih.gov).

Conflict of interest statement. None declared.

References

1. Louis DN, Ohgaki H, Wiestler OD, et al. The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol. 2007;114:97–109. doi:10.1007/s00401-007-0243-4. [PMC free article] [PubMed]
2. Cancer Genome Atlas Research Network. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008;455:1061–1068. doi:10.1038/nature07385. [PMC free article] [PubMed]
3. Parsons DW, Jones S, Zhang X, et al. An integrated genomic analysis of human glioblastoma multiforme. Science. 2008;321:1807–1812. doi:10.1126/science.1164382. [PMC free article] [PubMed]
4. Noushmehr H, Weisenberger DJ, Diefes K, et al. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma. Cancer Cell. 2010;17:510–522. doi:10.1016/j.ccr.2010.03.017. [PMC free article] [PubMed]
5. Verhaak RG, Hoadley KA, Purdom E, et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell. 2010;17:98–110. doi:10.1016/j.ccr.2009.12.020. [PMC free article] [PubMed]
6. Mueller WC, von Deimling A. Gene regulation by methylation. Recent Results Cancer Res. 2009;171:217–239. doi:10.1007/978-3-540-31206-2_13. [PubMed]
7. Cadieux B, Ching TT, VandenBerg SR, Costello JF. Genome-wide hypomethylation in human glioblastomas associated with specific copy number alteration, methylenetetrahydrofolate reductase allele status, and increased proliferation. Cancer Res. 2006;66:8469–8476. doi:10.1158/0008-5472.CAN-06-1547. [PubMed]
8. Amatya VJ, Naumann U, Weller M, Ohgaki H. TP53 promoter methylation in human gliomas. Acta Neuropathol. 2005;110:178–184. doi:10.1007/s00401-005-1041-5. [PubMed]
9. Baeza N, Weller M, Yonekawa Y, Kleihues P, Ohgaki H. PTEN methylation and expression in glioblastomas. Acta Neuropathol. 2003;106:479–485. doi:10.1007/s00401-003-0748-4. [PubMed]
10. Wiencke JK, Zheng S, Jelluma N, et al. Methylation of the PTEN promoter defines low-grade gliomas and secondary glioblastoma. Neurooncology. 2007;9:271–279. doi:10.1215/15228517-2007-003. [PMC free article] [PubMed]
11. Costello JF, Berger MS, Huang HS, Cavenee WK. Silencing of p16/CDKN2 expression in human gliomas by methylation and chromatin condensation. Cancer Res. 1996;56:2405–2410. [PubMed]
12. Waha A, Guntner S, Huang TH, et al. Epigenetic silencing of the protocadherin family member PCDH-gamma-A11 in astrocytomas. Neoplasia. 2005;7:193–199. doi:10.1593/neo.04490. [PMC free article] [PubMed]
13. Otsuka S, Maegawa S, Takamura A, et al. Aberrant promoter methylation and expression of the imprinted PEG3 gene in glioma. Proc Jpn Acad Ser B Phys Biol Sci. 2009;85:157–165. doi:10.2183/pjab.85.157. [PMC free article] [PubMed]
14. Riemenschneider MJ, Reifenberger J, Reifenberger G. Frequent biallelic inactivation and transcriptional silencing of the DIRAS3 gene at 1p31 in oligodendroglial tumors with 1p loss. Int J Cancer. 2008;122:2503–2510. doi:10.1002/ijc.23409. [PubMed]
15. Alaminos M, Davalos V, Ropero S, et al. EMP3, a myelin-related gene located in the critical 19q13.3 region, is epigenetically silenced and exhibits features of a candidate tumor suppressor in glioma and neuroblastoma. Cancer Res. 2005;65:2565–2571. doi:10.1158/0008-5472.CAN-04-4283. [PubMed]
16. Martinez R, Setien F, Voelter C, et al. CpG island promoter hypermethylation of the pro-apoptotic gene caspase-8 is a common hallmark of relapsed glioblastoma multiforme. Carcinogenesis. 2007;28:1264–1268. doi:10.1093/carcin/bgm014. [PubMed]
17. Hegi ME, Diserens AC, Gorlia T, et al. MGMT gene silencing and benefit from temozolomide in glioblastoma. N Engl J Med. 2005;352:997–1003. doi:10.1056/NEJMoa043331. [PubMed]
18. Martinez R, Martin-Subero JI, Rohde V, et al. A microarray-based DNA methylation study of glioblastoma multiforme. Epigenetics. 2009;4:255–264. [PubMed]
19. Gonzalez-Gomez P, et al. Promoter hypermethylation of multiple genes in astrocytic gliomas. Int J Oncol. 2003;22:601–608. [PubMed]
20. Martinez R, Schackert G, Esteller M. Hypermethylation of the proapoptotic gene TMS1/ASC: prognostic importance in glioblastoma multiforme. J Neurooncol. 2007;82:133–139. doi:10.1007/s11060-006-9264-4. [PubMed]
21. Uhlmann K, Rohde K, Zeller C, et al. Distinct methylation profiles of glioma subtypes. Int J Cancer. 2003;106:52–59. doi:10.1002/ijc.11175. [PubMed]
22. Yu J, Zhang H, Gu J, et al. Methylation profiles of thirty four promoter-CpG islands and concordant methylation behaviours of sixteen genes that may contribute to carcinogenesis of astrocytoma. BMC Cancer. 2004;4:65. doi:10.1186/1471-2407-4-65. [PMC free article] [PubMed]
23. Bibikova M, Lin Z, Zhou L, et al. High-throughput DNA methylation profiling using universal bead arrays. Genome Res. 2006;16:383–393. doi:10.1101/gr.4410706. [PMC free article] [PubMed]
24. Fan JB, Chee MS, Gunderson KL. Highly parallel genomic assays. Nat Rev Genet. 2006;7:632–644. doi:10.1038/nrg1901. [PubMed]
25. Barker DL, Theriault D, Che D, Dickinson T, Shen R, Kain R. Self-assembled random arrays: high-performance imaging and genomics applications on a high-density microarray platform. Proc SPIE. 2003;4966:1–11. doi:10.1117/12.483434.
26. Galinsky VL. Automatic registration of microarray images. II. Hexagonal grid. Bioinformatics. 2003;19:1832–1836. doi:10.1093/bioinformatics/btg260. [PubMed]
27. Bibikova M, Fan JB. GoldenGate assay for DNA methylation profiling. Methods Mol Biol. 2009;507:149–163. doi:10.1007/978-1-59745-522-0_12. [PubMed]
28. Lynch AG, Dunning MJ, Iddawela M, Barbosa-Morais NL, Ritchie ME. Considerations for the processing and analysis of GoldenGate-based two-colour Illumina platforms. Stat Methods Med Res. 2009;18:437–452. doi:10.1177/0962280208099451. [PubMed]
29. Eads CA, Danenberg KD, Kawakami K, et al. MethyLight: a high-throughput assay to measure DNA methylation. Nucleic Acids Res. 2000;28:E32. doi:10.1093/nar/28.8.e32. [PMC free article] [PubMed]
30. Ehrlich M, Woods CB, Yu MC, et al. Quantitative analysis of associations between DNA hypermethylation, hypomethylation, and DNMT RNA levels in ovarian tumors. Oncogene. 2006;25:2636–2645. doi:10.1038/sj.onc.1209145. [PMC free article] [PubMed]
31. Idbaih A, Marie Y, Lucchesi C, et al. BAC array CGH distinguishes mutually exclusive alterations that define clinicogenetic subtypes of gliomas. Int J Cancer. 2008;122:1778–1786. doi:10.1002/ijc.23270. [PubMed]
32. Ducray F, Idbaih A, de Reynies A, et al. Anaplastic oligodendrogliomas with 1p19q codeletion have a proneural gene expression profile. Mol Cancer. 2008;7:41. doi:10.1186/1476-4598-7-41. [PMC free article] [PubMed]
33. Sanson M, Marie Y, Paris S, et al. Isocitrate dehydrogenase 1 codon 132 mutation is an important prognostic biomarker in gliomas. J Clin Oncol. 2009;27:4150–4154. doi:10.1200/JCO.2009.21.9832. [PubMed]
34. Chastre E, Abdessamad M, Kruglov A, et al. TRIP6, a novel molecular partner of the MAGI-1 scaffolding molecule, promotes invasiveness. FASEB J. 2009;23:916–928. doi:10.1096/fj.08-106344. [PubMed]
35. Martin-Subero JI, Ammerpohl O, Bibikova M, et al. A comprehensive microarray-based DNA methylation study of 367 hematological neoplasms. PLoS One. 2009;4:e6986. doi:10.1371/journal.pone.0006986. [PMC free article] [PubMed]
36. Enomoto H, Ozaki T, Takahashi E, et al. Identification of human DAN gene, mapping to the putative neuroblastoma tumor suppressor locus. Oncogene. 1994;9:2785–2791. [PubMed]
37. Tomomura M, Morita N, Yoshikawa F, et al. Structural and functional analysis of the apoptosis-associated tyrosine kinase (AATYK) family. Neuroscience. 2007;148:510–521. doi:10.1016/j.neuroscience.2007.05.048. [PubMed]
38. Kuan CY, Whitmarsh AJ, Yang DD, et al. A critical role of neural-specific JNK3 for ischemic apoptosis. Proc Natl Acad Sci USA. 2003;100:15184–15189. doi:10.1073/pnas.2336254100. [PMC free article] [PubMed]
39. Martinho O, Longatto-Filho A, Lambros MB, et al. Expression, mutation and copy number analysis of platelet-derived growth factor receptor A (PDGFRA) and its ligand PDGFA in gliomas. Br J Cancer. 2009;101(6):973–982. doi:10.1038/sj.bjc.6605225. [PMC free article] [PubMed]
40. Ram R, Lorente G, Nikolich K, et al. Discoidin domain receptor-1a (DDR1a) promotes glioma cell invasion and adhesion in association with matrix metalloproteinase-2. J Neurooncol. 2006;76:239–248. doi:10.1007/s11060-005-6874-1. [PubMed]
41. Zhu S, Bjorge JD, Cheng HC, Fujita DJ. Decreased CHK protein levels are associated with Src activation in colon cancer cells. Oncogene. 2008;27:2027–2034. doi:10.1038/sj.onc.1210838. [PubMed]
42. Kim SO, Avraham S, Jiang S, Zagozdzon R, Fu Y, Avraham HK. Differential expression of Csk homologous kinase (CHK) in normal brain and brain tumors. Cancer. 2004;101:1018–1027. doi:10.1002/cncr.20442. [PubMed]
43. Gery S, Yin D, Xie D, Black KL, Koeffler HP. TMEFF1 and brain tumors. Oncogene. 2003;22:2723–2727. doi:10.1038/sj.onc.1206351. [PubMed]
44. Fra AM, Pasqualetto E, Mancini M, Sitia R. Genomic organization and transcriptional analysis of the human genes coding for caveolin-1 and caveolin-2. Gene. 2000;243:75–83. doi:10.1016/S0378-1119(99)00559-4. [PubMed]
45. Fukasawa M, Kimura M, Morita S, et al. Microarray analysis of promoter methylation in lung cancers. J Hum Genet. 2006;51:368–374. doi:10.1007/s10038-005-0355-4. [PubMed]
46. Goyal J, Smith KM, Cowan JM, Wazer DE, Lee SW, Band V. The role for NES1 serine protease as a novel tumor suppressor. Cancer Res. 1998;58:4782–4786. [PubMed]
47. Lu CY, Hsieh SY, Lu YJ, et al. Aberrant DNA methylation profile and frequent methylation of KLK10 and OXGR1 genes in hepatocellular carcinoma. Genes Chromosomes Cancer. 2009;48:1057–1068. doi:10.1002/gcc.20708. [PubMed]
48. Shi YE, Ni J, Xiao G, et al. Antitumor activity of the novel human breast cancer growth inhibitor, mammary-derived growth inhibitor-related gene, MRG. Cancer Res. 1997;57:3084–3091. [PubMed]
49. Tokumaru Y, Yamashita K, Osada M, et al. Inverse correlation between cyclin A1 hypermethylation and p53 mutation in head and neck cancer identified by reversal of epigenetic silencing. Cancer Res. 2004;64:5982–5987. doi:10.1158/0008-5472.CAN-04-0993. [PubMed]
50. Esteller M, Guo M, Moreno V, et al. Hypermethylation-associated inactivation of the cellular retinol-binding-protein 1 gene in human cancer. Cancer Res. 2002;62:5902–5905. [PubMed]
51. Katoh M. Cancer genomics and genetics of FGFR2 (Review) Int J Oncol. 2008;33:233–237. [PubMed]
52. Kondo T, Zheng L, Liu W, Kurebayashi J, Asa SL, Ezzat S. Epigenetically controlled fibroblast growth factor receptor 2 signaling imposes on the RAS/BRAF/mitogen-activated protein kinase pathway to modulate thyroid cancer progression. Cancer Res. 2007;67:5461–5470. doi:10.1158/0008-5472.CAN-06-4477. [PubMed]
53. Zhu X, Lee K, Asa SL, Ezzat S. Epigenetic silencing through DNA and histone methylation of fibroblast growth factor receptor 2 in neoplastic pituitary cells. Am J Pathol. 2007;170:1618–1628. doi:10.2353/ajpath.2007.061111. [PMC free article] [PubMed]
54. Inukai T, Uchida K, Nakajima H, et al. Tumor necrosis factor-alpha and its receptors contribute to apoptosis of oligodendrocytes in the spinal cord of spinal hyperostotic mouse (twy/twy) sustaining chronic mechanical compression. Spine. 2009;34:2848–2857. doi:10.1097/BRS.0b013e3181b0d078. [PubMed]
55. Levicar N, Nuttall RK, Lah TT. Proteases in brain tumour progression. Acta Neurochir. 2003;145:825–838. doi:10.1007/s00701-003-0097-z. [PubMed]
56. Colin C, Voutsinos-Porche B, Nanni I, et al. High expression of cathepsin B and plasminogen activator inhibitor type-1 are strong predictors of survival in glioblastomas. Acta Neuropathol. 2009;118:745–754. doi:10.1007/s00401-009-0592-2. [PubMed]
57. Ding Q, Stewart J, Jr, Prince CW, et al. Promotion of malignant astrocytoma cell migration by osteopontin expressed in the normal brain: differences in integrin signaling during cell adhesion to osteopontin versus vitronectin. Cancer Res. 2002;62:5336–5343. [PubMed]
58. Chernov AV, Sounni NE, Remacle AG, Strongin AY. Epigenetic control of the invasion-promoting MT1-MMP/MMP-2/TIMP-2 axis in cancer cells. J Biol Chem. 2009;284:12727–12734. doi:10.1074/jbc.M900273200. [PMC free article] [PubMed]
59. Takai D, Yagi Y, Wakazono K, et al. Silencing of HTR1B and reduced expression of EDN1 in human lung cancers, revealed by methylation-sensitive representational difference analysis. Oncogene. 2001;20:7505–7513. doi:10.1038/sj.onc.1204940. [PubMed]
60. Strefford JC, Stasevich I, Lane TM, Lu YJ, Oliver T, Young BD. A combination of molecular cytogenetic analyses reveals complex genetic alterations in conventional renal cell carcinoma. Cancer Genet Cytogenet. 2005;159:1–9. doi:10.1016/j.cancergencyto.2004.09.020. [PubMed]
61. Li CY, Chow C, Chan WY. Allelic imbalance mapped to 6q14.1 is associated with loss of expression of 5-HT receptor 1B in non-Hodgkin lymphomas. Diagn Mol Pathol. 2004;13:69–74. doi:10.1097/00019606-200406000-00002. [PubMed]
62. Zhou H, Clapham DE. Mammalian MagT1 and TUSC3 are required for cellular magnesium uptake and vertebrate embryonic development. Proc Natl Acad Sci USA. 2009;106:15750–15755. doi:10.1073/pnas.0908332106. [PMC free article] [PubMed]
63. Bashyam MD, Bair R, Kim YH, et al. Array-based comparative genomic hybridization identifies localized DNA amplifications and homozygous deletions in pancreatic cancer. Neoplasia. 2005;7:556–562. doi:10.1593/neo.04586. [PMC free article] [PubMed]
64. Möllemann M, Wolter M, Felsberg J, et al. Frequent promoter hypermethylation and low expression of the MGMT gene in oligodendroglial tumors. Int J Cancer. 2005;113:379–385. doi:10.1002/ijc.20575. [PubMed]
65. Hesson L, Bièche I, Krex D, et al. Frequent epigenetic inactivation of RASSF1A and BLU genes located within the critical 3p21.3 region in gliomas. Oncogene. 2004;23:2408–2419. doi:10.1038/sj.onc.1207407. [PubMed]

Articles from Neuro-Oncology are provided here courtesy of Society for Neuro-Oncology and Oxford University Press
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • MedGen
    MedGen
    Related information in MedGen
  • PubMed
    PubMed
    PubMed citations for these articles
  • Substance
    Substance
    PubChem Substance links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...