|Home | About | Journals | Submit | Contact Us | Français|
Candidate gene investigations have indicated a significant role for epigenetic events in the pathogenesis of medulloblastoma, the most common malignant brain tumor of childhood. To assess the medulloblastoma epigenome more comprehensively, we undertook a genomewide investigation to identify genes that display evidence of methylation-dependent regulation. Expression microarray analysis of medulloblastoma cell lines following treatment with a DNA methyltransferase inhibitor revealed deregulation of multiple transcripts (3%–6% of probes per cell line). Eighteen independent genes demonstrated >3-fold reactivation in all cell lines tested. Bisulfite sequence analysis revealed dense CpG island methylation associated with transcriptional silencing for 12 of these genes. Extension of this analysis to primary tumors and the normal cerebellum revealed three major classes of epigenetically regulated genes: (1) normally methylated genes (DAZL, ZNF157, ASN) whose methylation reflects somatic patterns observed in the cerebellum, (2) X-linked genes (MSN, POU3F4, HTR2C) that show disruption of their sex-specific methylation patterns in tumors, and (3) tumor-specific methylated genes (COL1A2, S100A10, S100A6, HTATIP2, CDH1, LXN) that display enhanced methylation levels in tumors compared with the cerebellum. Detailed analysis of COL1A2 supports a key role in medulloblastoma tumorigenesis; dense biallelic methylation associated with transcriptional silencing was observed in 46 of 60 cases (77%). Moreover, COL1A2 status distinguished infant medulloblastomas of the desmoplastic histopathological subtype, indicating that distinct molecular pathogenesis may underlie these tumors and their more favorable prognosis. These data reveal a more diverse and expansive medulloblastoma epigenome than previously understood and provide strong evidence that the methylation status of specific genes may contribute to the biological subclassification of medulloblastoma.
Medulloblastoma is an invasive embryonal tumor of the cerebellum and the most common malignant brain tumor of childhood. Current risk stratification of medulloblastoma patients is based on clinical features, with high-risk patients comprising children <3 years of age at diagnosis, patients that present with metastatic disease, and patients with residual tumor following surgery. However, the clinical utility of this risk stratification is limited, and outcome is variable within current high- and standard-risk groups.1,2 Recent molecular studies have provided critical insights into the biological basis of medulloblastoma development. These investigations offer promise for a more robust subclassification of this heterogeneous group of tumors and have identified biological markers with potential clinical utility.3–7
The study of epigenetic changes in medulloblastoma development offers significant potential for an improved understanding of its molecular pathogenesis, alongside the identification of molecular stratification markers and therapeutic targets. Restriction landmark genome scanning studies have suggested that multiple loci undergo changes in their methylation status during medulloblastoma development,8 but relatively few gene-specific events have been identified to date, and most studies in medulloblastoma have been limited to the analysis of individual genes in limited cohorts of tumors. Moreover, relationships between medulloblastoma methylation patterns and normal patterns of somatic methylation observed in the normal cerebellum have not been widely assessed.9 Genomewide approaches have been successfully used to identify epigenetically regulated genes in such tumor types as colorectal, esophageal, and breast carcinomas.10–12 Approaches include the use of expression arrays to identify epigenetically silenced candidate tumor suppressor genes, which are upregulated following inhibition of methyltransferases or histone deacetyllases, and CpG island arrays, which allow the investigation of changes in the methylation status of multiple CpG islands in a single experiment.10–12 The predictive value of epigenomic data has also been demonstrated in recent studies, such as in ovarian cancer, where array-based analysis was used to create a 112-gene “methylation signature” able to predict progression-free survival with 95% accuracy.13 Recent studies have begun to apply such array-based techniques to medulloblastoma epigenetics; however, subsequent detailed follow-up of methylated genes identified by these techniques has not assessed their diagnostic or prognostic potential.14,15
In this article, we report investigations aimed at developing a more comprehensive understanding of the nature of the medulloblastoma epigenome and its clinical significance. Using microarray-based analysis of gene expression changes induced by inhibition of DNA methylation, we show that a significant proportion of genes is regulated by methylation-dependent mechanisms in medulloblastoma cell lines. Our subsequent detailed investigation of a series of these genes in primary medulloblastomas and the normal cerebellum has revealed distinct classes of epigenetically regulated genes in medulloblastoma and identified novel molecular events in its pathogenesis. Importantly, we identify COL1A2 as a frequently methylated and epigenetically silenced gene in medulloblastoma. COL1A2 methylation is strongly associated with all clinicopathological disease subgroups except infant tumors with desmoplastic/nodular histology, providing evidence that a distinct molecular pathogenesis underlies these tumors and their more favorable clinical outcome. Based on these data, methylation signatures appear to show significant potential as biological markers for the discrimination of specific disease subgroups in medulloblastoma.
Demethylating treatment and analyses of gene expression and methylation status of multiple genes were performed on three independent medulloblastoma cell lines (D283Med, MEB-MED8A, and D425Med). Additional cell lines DAOY, UW228-3, MHH-MED1, D341Med, D384Med, and D556Med were analyzed for COL1A2 methylation status. All cells were grown under recommended culture conditions. Cell line DNA was extracted using the Qiagen DNeasy kit (Qiagen Ltd, Crawley, UK).
Normal cerebellar specimens consisted of snap-frozen postmortem material from three fetuses (18, 19, and 22 weeks’ gestational age: two males, one female), one female infant (newborn), and one male adult (67 years) who had died of nonneoplastic conditions. Three additional cerebella from male patients (prenatal, 25 months, and 60 years) were analyzed for COL1A2 methylation status.
The methylation status of candidate genes identified was analyzed in an initial cohort of 16 primary medulloblastomas, selected to represent all the major disease subclassifications. The cohort consisted of seven classic tumors, three large-cell/anaplastic tumors, and six tumors of the desmoplastic/nodular histopathological subtype:2 7 male and 9 female cases, 4 of whom were infants (<3 years), 11 children (3–15 years), and 1 adult (≥16 years). COL1A2 methylation status was analyzed in a further 44 primary tumors. The total cohort of 60 tumors analyzed for COL1A2 methylation consisted of 33 classic, 9 large-cell/anaplastic, and 18 desmoplastic/nodular tumors: 7 metastatic-stage (M-stage) 3, 1 M-stage 2, 45 M-stage 0/1, and 7 data unavailable;2 34 males and 26 females, of whom 14 were infants, 41 children, and 5 adults. (For clinical and pathological data for individual cases, see Figs. 2 and and33 below.) DNA was extracted from snap-frozen tissues using standard methods and from formalin-fixed, paraffin-embedded tissue using a Qiagen DNeasy kit (Qiagen). Loss of heterozygosity analysis (LOH) was performed on 32 tumor DNA samples for which constitutional DNA was available. DNA from peripheral blood samples was prepared as previously described.16 Newcastle and North Tyneside Research Ethics Committee approval was obtained for the collection, storage, and biological study of all material.
Cell lines were grown in the presence or absence of the demethylating agent 5′-aza-2′-deoxycytidine (5-azaCdR) as previously described.17
Total RNA was extracted as previously described17 from cell lines D283Med, D425Med, and MEB-MED8A, which had been grown in parallel cultures with or without 5-azaCdR treatment. Microarray expression analysis was performed at the Newcastle University microarray facility; RNA was converted to biotin-labeled cRNA and hybridized to the Human Genome U133A array according to manufacturer protocols (Affymetrix, Santa Clara, CA, USA). MAS5 software (Affymetrix) was used for data processing, normalization, and calculation of signal intensities. Signal intensities were compared between treatment conditions using Microsoft Excel software.
CpG islands were predicted by analyzing 20,000 bp of sequence surrounding the transcriptional start site of each gene using transcript information obtained from the Ensembl genome browser (www.ensembl.org) and the program CpG Plot (www.ebi.ac.uk/emboss/cpgplot). Parameters selected were regions of DNA longer than 100 bp with an observed:expected ratio of the frequency of the dinucleotide CpG of >0.6 and an overall G/C content of >50%. Previously published primers were used to analyze the methylation status of CDH1 and S100A6.18,19 For all other genes, primers suitable for bisulfite sequencing were designed using the program Methprimer (www.urogene.org/methprimer). Full details of the Ensembl transcript information, position of predicted CpG islands, PCR products, and primer sequences are provided in a Supplementary Data Table available (online only, with this article at neuro-oncology.dukejournals.org).
Bisulfite treatment of DNA was carried out using the EpiTect Bisulfite Kit (Qiagen). We used 30 ng of bisulfite-treated DNA per PCR reaction, which was carried out using standard conditions. POU3F4 and CDH1 PCR reactions required the addition of betaine (Sigma-Aldrich, Gillingham, UK) to the PCR mix (final concentration, 1 M). Annealing temperatures for each primer pair are provided in the supplementary data. PCR products were purified using the PureLink PCR purification kit (Invitrogen, Paisley, UK). Purified products were directly sequenced with a CEQ DTCS kit (Beckman Coulter, High Wycombe, UK) using the antisense primer to obtain the reverse sequence. Sequenced products were analyzed on a CEQ 2000XL DNA analysis system (Beckman Coulter), and the methylation status at each CpG residue was determined by assessing the relative peak intensities, as previously described.20
RNA extraction, cDNA synthesis, and analysis of PCR products were performed as described previously.17 Reverse transcription (RT)-PCR primers were designed using transcript information obtained from the Ensembl genome browser. Primer sequences, exon locations, and annealing temperatures are provided in the supplementary data. RT-PCR of ACTB (encoding β-actin, a house-keeping gene) was used as a control for RNA concentration, as described previously.17
cDNA was used as a template in real-time PCR reactions specific to the COL1A2 transcript, and expression was assessed relative to expression of a control gene (TBP) using the comparative Ct method for relative quantification. Real-time PCR was performed on the ABI PRISM 7900HT Detection System using TaqMan reagents and primers designed to amplify regions encompassing exons 6 and 7 of the COL1A2 transcript and exons 5 and 6 of the TBP transcript (Applied Biosystems, Foster City, CA, USA) according to manufacturer instructions. Analyses were performed in triplicate.
Whole-cell protein extracts were extracted from subconfluent medulloblastoma cells as previously described.21 We subjected 90 μg of protein lysate to polyacrylamide gel electrophoresis and Western blot analysis using standard methods. Membranes were probed with a primary rabbit polyclonal antibody specific to collagen type 1 (Abcam, Cambridge, UK) and then reprobed without stripping with a mouse monoclonal antibody specific to α-tubulin (Sigma-Aldrich) as a control. Appropriate secondary antibodies with horseradish peroxidase conjugates (Dakocymation Ltd., Glostrup, Denmark) were bound to the primary antibodies, followed by incubation with enhanced chemiluminescence substrate (Pierce, Rockford, IL, USA). Protein bands were visualized by autoradiography.
Analysis was performed using published Affymetrix U133Av2 expression array data available for a cohort of 46 medulloblastomas4 (available at www.stjuderesearch.org/data/medulloblastoma/). Twelve tumors from this cohort were assessed in the present study and were used to compare COL1A2 expression in methylated and unmethylated tumors. The complete array data set was used to analyze COL1A2 expression in different clinicopathological subtypes. Expression data were subjected to log10 transformation prior to statistical analysis.
Two polymorphic microsatellite markers were used to assess the heterozygosity status of the COL1A2 region: D7S527 and D7S1820 (www.ncbi.nlm.nih.gov/genome/sts). The forward primer from each set was end-labeled with a fluorescent Beckman dye (Sigma-Aldrich). Size markers (Beckman Coulter) were added to a 1:10 dilution of the PCR products, which were then run and analyzed on the CEQ 2000XL DNA analysis system. LOH was assessed as previously described.7
To identify genes showing evidence of epigenetic regulation in medulloblastoma, three medulloblastoma cell lines (D425Med, D283Med, and MEB-MED8A) were cultured in the presence or absence of the demethylating agent 5-azaCdR. The optimal 5-azaCdR concentration and treatment time were initially determined by analyzing effects on reexpression of the RASSF1A gene (methylated in all three cell lines)16 by real-time quantitative PCR (data not shown). The lowest 5-azaCdR concentration and shortest exposure at which maximal RASSF1A reexpression was achieved were selected for use in subsequent experiments, in order to minimize any toxicity-related and/or secondary gene expression responses in treated cells. Following treatment with 5-azaCdR under optimized conditions (5 μM, 72 h), resultant expression changes were assessed using the Affymetrix Human Genome U133A array, which contains >22,000 probe sets recognizing 18,400 independent transcripts. Changes in probe signal intensity following 5-azaCdR treatment were calculated, and probes detecting an expression increase >3-fold were classed as being upregulated and thus showing strong evidence of methylation-dependent transcriptional regulation.
Analysis of individual cell lines indicated an expansive number of epigenetically regulated genes in medulloblastoma; between 3% and 6% of probe sets examined showed evidence of methylation-dependent upregulation (632–1,387 of ~22,000; Fig. 1A). Data sets generated for each cell line were compared in order to identify probes detecting increased expression in multiple cell lines. This analysis identified 19 probes upregulated in all three cell lines (Fig. 1A; listed in Table 1), which were prioritized for further analysis to characterize in detail the nature of epigenetic events observed in medulloblastoma. Probe sequence identity and specificity were verified using the BLAST sequence alignment tool (www.ncbi.nlm.nih.gov/BLAST). Expression changes observed by array analysis were independently validated by RT-PCR for six selected genes. Increases in expression observed by microarray were confirmed in all cases (Fig. 1B). In addition, upregulation of S100A6 and S100A10 has also been validated previously.17
The 19 probes mapped to genes involved in a diversity of cellular pathways related to tumorigenesis (including signal transduction, cell division, cell adhesion, and extracellular matrix assembly). While six of the genes have previously been reported to be methylated in other cancers (CDH1, PLK2, COL1A2, S100A6, S100A10, LXN), only one gene (CDH1) has previously been identified as methylated in medulloblastoma.10,18,22–26 Another gene, ID2B, is a pseudogene (accession NM_001039082; www.ncbi.nlm.nih.gov) and was not analyzed further.
CpG islands usually occur 5′ to the transcriptional start site of a gene and are often associated with the promoter region. However, methylation of flanking DNA sequences can also mediate epigenetic silencing.27 Thus, in order to identify CpG islands potentially involved in regulating the expression of the genes of interest, the presence of CpG islands within a 20,000-bp region surrounding the transcriptional start site of each gene was predicted in silico using CpG Plot software. Predicted CpG islands were identified for 16 of the 18 genes, which were taken forward for further analysis. CpG island lengths and positions, relative to the transcriptional and translational start sites of each gene, are shown in the supplementary data. CpG islands were not identified in CGAT1 and HNMT. Although this finding does not exclude a role for DNA methylation in their regulation, these two genes were not investigated further.
To determine whether the transcriptional silencing and upregulation following 5-azaCdR treatment observed for the 16 candidate genes harboring 5′ CpG islands were associated with DNA methylation of these regions, the methylation status of the CpG islands of all 16 genes was assessed by bisulfite sequence analysis in medulloblastoma cell lines (Fig. 2). Seventy-five percent (12 of 16) of genes showed methylation of their CpG islands in all three cell lines (COL1A2, LXN, MSN, CDH1, S100A6, S100A10, HTATIP2, ZNF157, POU3F4, ASPN, HTR2C, and DAZL), consistent with their epigenetic transcriptional regulation, while the remaining four genes were consistently unmethylated (PLK2, ABL3, AP4E1, and EPOR).
We next examined the relationship between methylation patterns detected in the medulloblastoma cell lines and those observed in the normal cerebellum. Based on these patterns, methylated genes can be categorized into three major groups (Fig. 2A, B). Group 1 genes show evidence of enhanced methylation in all three medulloblastoma cell lines compared with the normal cerebellum. Group 1A genes (COL1A2, HTATIP2, S100A6, S100A10, and CDH1) are methylated in all three cell lines but unmethylated in the normal cerebellum, while group 1B genes (LXN) show evidence of variable levels and distribution of partial methylation in the normal cerebellum, in contrast to complete methylation at all CpG sites examined in the three cell lines. Group 2 genes are X-linked (POU3F4, MSN, and HTR2C) and exhibit their predicted methylation patterns in the normal cerebellum (i.e., partially methylated in female cerebella as a consequence of X inactivation by hemimethylation, and unmethylated in male cerebella) but display aberrant patterns of complete methylation in cell lines (all of which are derived from male patients). Group 3 genes (DAZL, ZNF157, and ASPN) exhibit equivalent patterns of methylation in normal cerebella and medulloblastoma cell lines, indicating that these patterns reflect normal somatic methylation. Because group 3 gene methylation is not tumor specific, these genes were not analyzed further. These comparisons demonstrate that somatic DNA methylation is a significant feature of the normal cerebellum and indicate a complex and gene-specific interrelationship between methylation patterns observed in the normal cerebellum and their disruption in medulloblastoma cell lines.
The nine genes showing evidence of contrasting methylation patterns between normal cerebella and cell lines were analyzed in a cohort of 16 primary medulloblastomas, representative of all major disease subtypes (Fig. 2). Of the five group 1A genes showing complete methylation in all three cell lines and no methylation in normal cerebellum samples, COL1A2 was completely methylated in a large proportion of primary tumors (75%; 12 of 16), and S100A10 was methylated partially in 25% (4 of 16) of cases, while the remaining genes (HTATIP2, S100A6, and CDH1) were unmethylated in the tumor samples studied. The group 1B gene LXN, which exhibits partial methylation in the normal cerebellum, showed complete methylation in 94% (15 of 16) of primary tumors. The normal sex-specific methylation patterns of the three X-linked group 2 genes (POU3F4, MSN, and HTR2C) were disrupted in tumors, with 9 of 16 (56%) of tumors exhibiting aberrant methylation of one or more of these genes. Hypermethylation of POU3F4 and HTR2C was observed in three of seven and four of seven tumors from male patients, respectively. Where partial methylation was seen for these genes in male patients, it differed from the normal female methylation pattern (50% methylation at all sites), with variation in the amount of methylation (0%–100%) across individual sites. Conversely, all three genes showed hypomethylation in tumors from female patients; three such tumors showed concordant hypomethylation of all three genes and may result from loss of one complete copy of the X chromosome (see “Discussion”), whereas POU3F4 and HTR2C were hypomethylated in a further one and two cases, respectively.
In summary, six of the nine genes investigated in primary tumors showed evidence of disrupted methylation patterns in a significant proportion (>15%) of cases compared with the normal cerebellum. Of these genes, COL1A2 showed most frequent evidence of complete and tumor-specific methylation and was therefore selected for further investigation.
The epigenetic silencing of COL1A2 was next verified in extended studies of medulloblastoma cell lines and primary tumors, using three independent experimental approaches. First, COL1A2 methylation status was assessed in an extended panel of medulloblastoma cell lines. Two examples of unmethylated cell lines were identified (DAOY, UW228-3), while all remaining cell lines were completely methylated (78%; 7 of 9 lines total; Fig. 3). Real-time PCR analysis was used to assess COL1A2 mRNA levels in unmethylated (DAOY, UW228-3) and selected methylated (D283Med, D425Med) cell lines, alongside normal cerebellar samples (Fig. 4A). COL1A2 expression was reduced in COL1A2 methylated cell lines compared with COL1A2 unmethylated cell lines and normal cerebellum. Following 5-azaCdR treatment, expression of COL1A2 increased in the two methylated cell lines, but not in the two unmethylated cell lines, strongly supporting its epigenetic regulation by DNA methylation (Fig. 4A). Second, collagen type 1 protein levels were assessed in the selected cell lines by Western blot analysis (Fig. 4B). Collagen type 1 expression was markedly lower in methylated cell lines, indicating that the transcriptional silencing observed for COL1A2 has consequent effects on protein expression levels. Third, COL1A2 methylation status was assessed in a cohort of 12 primary tumors for which expression data, generated using the Affymetrix U133Av2 expression microarray, were available from a previously published study (Fig. 4C).4 Methylation analysis revealed that 4 of 12 tumors were unmethylated for COL1A2, and 8 of 12 were methylated (Fig. 3). Unmethylated tumors typically showed higher levels of COL1A2 expression than did methylated cases, and the mean expression level was approximately twice as high in the unmethylated tumor group as in the methylated tumors. These differences showed a trend toward, but did not reach, statistical significance in this limited cohort (p = 0.142, Student’s t-test). The unavailability of matched DNA/RNA samples from further samples precluded a wider analysis in the present study, but this limited data set is consistent with a relationship between COL1A2 methylation status and expression in primary tumors. Together, data from these three independent approaches strongly support the epigenetic transcriptional silencing of COL1A2 by DNA methylation in both medulloblastoma cell lines and primary tumors.
COL1A2 methylation patterns were assessed in detail in extended cohorts of primary tumors (n = 60) and normal cerebella (n = 8) (Fig. 3), which included samples shown in Figs. 2 and and4.4. In total, 77% (46 of 60) of the primary tumor cohort was methylated for COL1A2. Where present, COL1A2 methylation typically involved dense methylation across the whole CpG island affecting all sites analyzed. No evidence of methylation was observed at any CpG residue in unmethylated tumors and cell lines. All eight normal cerebella were similarly unmethylated, further confirming the specificity of aberrant COL1A2 methylation patterns observed in cell lines and primary tumors.
To determine whether genetic mechanisms contribute to COL1A2 inactivation in medulloblastoma, the polymorphic microsatellite markers D7S527 and D7S1820, which flank the COL1A2 locus on 7q22.1 (1.6 Mb telomeric and 0.7 Mb centromeric to the COL1A2 transcription start site, respectively), were used to assess whether LOH in this region is a feature of medulloblastoma. Thirty-two paired constitutional blood and tumor DNA samples were assessed, of which 28 were informative for at least one marker. No evidence of LOH at the COL1A2 locus was found (data not shown). Although we cannot rule out a role for smaller intragenic mutations or deletions, the presence of two COL1A2 alleles, coupled with dense patterns of complete COL1A2 methylation in tumors, indicates that (1) deletion of the COL1A2 locus is not a common feature of medulloblastoma and (2) biallelic epigenetic inactivation by DNA methylation represents the predominant mechanism of COL1A2 inactivation in these tumors.
An assessment of associations between COL1A2 methylation patterns and clinical and pathological medulloblastoma disease features (i.e., sex, age at diagnosis, metastatic stage, and pathological subtype) was performed in our extended cohort of 60 primary tumors (methylation patterns and clinical data are shown in Fig. 3). COL1A2 methylation was significantly associated with the nondesmoplastic histopathological morphophenotype; 86% (36 of 42) of tumors with nondesmoplastic pathology were methylated, compared with 56% (10 of 18) of desmoplastic/nodular tumors (Fisher’s exact p = 0.019). In addition, COL1A2 methylation was significantly associated with patient age at diagnosis (87% [40 of 46] of patients ≥3 years vs. 43% [6 of 14] of patients <3 years; Fisher’s exact p = 0.002). Most notably, these observed relationships appear to be founded on results for the infant desmoplastic medulloblastoma subgroup. Infant desmoplastic medulloblastomas appear to represent a distinct medulloblastoma subgroup based on their COL1A2 methylation status; COL1A2 is rarely methylated in this tumor group (13% [1 of 8] of cases), compared with 90% (9 of 10) of desmoplastic cases ≥3 years at diagnosis (p = 0.002) and 87% (45 of 52) of cases outside the infant desmoplastic subgroup (p < 0.0001, both Fisher’s exact test). No further statistically significant relationships were observed.
If COL1A2 expression is regulated by its methylation and infant desmoplastic tumors are rarely methylated, we would predict that COL1A2 expression would be higher in this group of tumors than in other medulloblastomas. In order to test this hypothesis, we examined Affymetrix U133Av2 expression microarray data available from a cohort of 46 medulloblastomas (comprising 7 infant desmoplastic tumors, 7 childhood [≥3 years at diagnosis] desmoplastic tumors, 4 infant nondesmoplastic tumors, and 28 childhood nondesmoplastic tumors).4 Consistent with the methylation-dependent regulation of COL1A2 expression in primary medulloblastomas, infant desmoplastic tumors showed significantly higher COL1A2 mRNA expression than other medulloblastomas (Fig. 5; p = 0.004, Student’s t-test) in this cohort. Similarly, within the desmoplastic histopathological subtype, infant desmoplastic tumors had significantly higher COL1A2 expression than did childhood desmoplastic cases (p = 0.04, Student’s t-test), which had a mean expression level comparable to that of nondesmoplastic tumors (2,866 vs. 2,613; p = 0.69, Student’s t-test).
The application of array-based methodologies to effect a global analysis of DNA methylation patterns and epigenetically regulated genes has started to provide insights into the nature and extent of the epigenome.10–15 In this study, we used a pharmacological expression reactivation approach to identify genes upregulated following demethylating treatment in medulloblastoma cell lines prior to their detailed analysis to facilitate characterization of epigenetic events contributing to the medulloblastoma epigenome and their clinical and biological significance in medulloblastoma development.
Our data suggest a wide-ranging role for epigenetic events in medulloblastoma cells; between 3% and 6% of transcripts analyzed (632–1,387 of ~22,000) showed evidence of >3-fold upregulation in each cell line following demethylation treatment. This figure represents only an approximation of the number of genes methylated, because some methylated genes may show <3-fold upregulation, while, conversely, some genes will be upregulated by methylation-independent mechanisms. Nevertheless, the data indicate an extensive epigenome and are corroborated by previous restriction landmark genome scanning studies, which have indicated that approximately 6% of CpG islands are methylated in medulloblastoma cell lines.8
Eighteen independent genes were upregulated following demethylation treatment in all three cell lines tested and selected for further analysis. Unlike conventional candidate gene-based studies, these genes were selected in an unbiased manner without previous knowledge of their function or methylation status in other cancers and should therefore represent a truer picture of the classes of epigenetically regulated genes in the medulloblastoma genome. Sixteen of these genes harbored predicted CpG islands in their 5′ regions, and CpG island methylation was confirmed for 12 (75%) of these in medulloblastoma cell lines. Although culture-related de novo methylation events can be common in cell line models,28 methylation of 9 of the 12 genes was consequently demonstrated in either primary medulloblastomas or normal cerebellar tissues, underlining the utility of this pharmacological reactivation approach for the identification of epigenetically regulated genes relevant in development and disease.
Detailed analysis of the methylated genes identified has provided significant insights into the nature of the normal and malignant cerebellar epigenomes. Somatic methylation is a significant feature of normal cerebellar tissues, illustrated by group 3 genes (DAZL, ZNF157, and ASPN), which show consistent patterns of methylation in normal cerebella derived from individuals across a wide range of ages (postnatal to adult). These patterns are reflected in medulloblastoma cell lines and are thus unlikely to play any direct role in tumorigenesis. These genes add to the growing numbers of genes that have been detected as methylated in the normal cerebellum17,20 and suggest that the role of DNA methylation in the transcriptional silencing of genes in normal somatic tissues has perhaps been underestimated.
Group 2 genes indicate the potential involvement of X-linked genes in medulloblastoma pathogenesis. These genes show the expected sex-specific patterns of methylation in the normal cerebellum, which become disrupted in a proportion of tumors. Inactivation of one copy of the X chromosome in female mammals is associated with methylation of promoter CpG islands on this chromosome.29 Therefore, these genes will appear hemi-methylated in normal tissue from females, whereas in male tissues they will be unmethylated. Based on the three X-linked genes identified and analyzed (POU3F4, MSN, and HTR2C), this pattern was disrupted in the majority of primary medulloblastomas (9 of 16) and cell lines (3 of 3) by both the loss of methylation in tumors from female patients and the gain of methylation in tumors from male patients. The loss of methylation in female tumors generally occurred concordantly in all three genes analyzed and may reflect loss of one copy of the X chromosome, which is a frequent occurrence in primary medulloblastomas.30,31 Conversely, POU3F4 and HTR2C were hypermethylated in a proportion of tumors from male patients, suggesting that hypermethylation of X-linked genes may contribute to medulloblastoma tumorigenesis in males. Roles for POU3F4 or HTR2C in tumorigenesis have not been investigated previously, but HTR2C encodes 5-hydroxytryptamine (serotonin) receptor 2C and is involved in neuronal migration, cell division, and differentiation.32 Together, these findings suggest pleiotropic and sex-specific roles for the epigenetic regulation of X-linked genes in medulloblastoma. Their further investigation, particularly in the context of the male predominance observed for medulloblastoma,2 will be of interest.
Finally, the six autosomal group 1 genes identified and analyzed (group 1A, COL1A2, S100A10, S100A6, HTATIP2, and CDH1; group 1B, LXN) were methylated in a tumor-specific manner in all cell lines and thus represent candidate tumor suppressor genes in medulloblastoma. Group 1A genes are unmethylated in the normal cerebellum but display aberrant hypermethylation in primary tumors. S100A10 was methylated in 4 of 16 tumors; we have previously investigated S100A6 and S100A10 methylation in a large cohort of primary tumors (not including the 16 tumors analyzed here) and found both genes to be methylated in a proportion of cases (5 of 40 and 4 of 35, respectively),17 suggesting that transcriptional silencing of these genes is important in a subset of primary tumors. Our analysis of COL1A2 in an extended tumor and cell line cohort showed frequent methylation in primary tumors (46 of 60, 77%), similar to that seen in cell lines (7 of 9, 78%). This suggests that methylation of COL1A2 in cell lines accurately reflects the methylation status of primary tumors. HTA-TIP2 and CDH1 were not methylated in the cohort of 16 primary tumors investigated in this study, suggesting they are unlikely to be frequently methylated in medulloblastoma. Findings for CDH1 are consistent with a previous study (methylation in 0 of 38 cases).33 Group 1B genes, exemplified by LXN, show complete methylation in tumors against a background of partial and variable methylation in the normal cerebellum; this enhanced methylation status suggests a possible tumor suppressor role for LXN, which has recently been identified as a negative regulator of stem cell population size in mice.34 Alternatively, because LXN expression has been shown to be restricted to certain cell populations in the rat brain,35 it may show cell-type-specific methylation-dependent silencing within the cerebellum,36 with the methylation seen in medulloblastomas reflecting their cell type(s) of origin.
We have identified tumor-specific (predominately biallelic) methylation across the COL1A2 CpG island associated with transcriptional silencing in the majority (77%) of primary medulloblastomas. COL1A2 methylation status provides important insights into the molecular pathogenesis of specific medulloblastoma disease subgroups and has clear utility as a medulloblastoma biomarker. Current histopathological classification divides medulloblastoma tumors into two broad categories: nondesmoplastic (classic and large-cell anaplastic, 85% of all tumors) and desmoplastic (15% of tumors).2 Clinically, medulloblastoma patients younger than 3 years of age are treated as a distinct group and have a significantly worse prognosis than do older children.37 The reduced radiotherapy doses given to these children are a contributing factor, but it is unclear whether the biology of infant medulloblastomas differs from that of those arising in older patients. Desmoplastic/nodular tumors are more frequent in patients younger than 3 years of age (~55% of tumors, compared with ~5% of tumors from children 3–16 years of age) and are associated with a better prognosis than are nondesmoplastic tumors in this infant group.6,38 However, any biological basis for these clinicopathological associations remains to be identified.
COL1A2 methylation is a frequent event in nondesmoplastic tumors from both infant (<3 years; 83%, 5 of 6) and childhood (≥3 years; 86%, 31 of 36) cases. In contrast, COL1A2 methylation in desmoplastic/nodular tumors is clearly associated with patient age: COL1A2 methylation is rare in infant desmoplastic/nodular cases (13%, 1 of 8) compared with its frequent occurrence in childhood desmoplastic/nodular tumors (90%, 9 of 10). These findings are further corroborated by expression microarray data derived from 46 medulloblastomas, in which COL1A2 expression is significantly higher in infant desmoplastic/nodular tumors than in other tumor subtypes.4 Together, these data demonstrate that the epigenetic inactivation of COL1A2 is strongly associated with medulloblastoma development but is not a significant feature of infant desmoplastic/nodular tumors. The clinicopathological associations observed for COL1A2 appear to be distinctive among gene-specific epigenetic events assessed in medulloblastoma to date; other frequently methylated genes in medulloblastoma, such as RASSF1A, do not show any significant clinicopathological associations and are frequently methylated in infant desmoplastic/nodular tumors.20
Although further work is now required to examine the molecular profiles of these subgroups of tumors in more detail, COL1A2 is the first biomarker identified to suggest that, despite similar histologies, desmoplastic/nodular tumors arising in infants are molecularly distinct from their counterparts in older children and may represent a unique disease subgroup at the molecular level. Differential COL1A2 methylation between nondesmoplastic and desmoplastic/nodular cases within the infant subgroup provides a molecular correlate for the more favorable clinical outcome reported for desmoplastic/nodular tumors in infants and offers a strong biological basis for the development of desmoplastic/nodular histology as a therapeutic stratification marker in this group of tumors in future groupwide clinical trials.
Any functional role for COL1A2 inactivation in medulloblastoma development now requires further investigation. COL1A2, together with COL1A1, encodes collagen type 1, a major component of the extracellular matrix.39 A possible tumor suppressor role for COL1A2 in medulloblastoma is supported by a number of independent lines of evidence, including the following: (1) nondesmoplastic infant medulloblastomas (associated with COL1A2 inactivation) have a worse outcome than do desmoplastic/nodular tumors in this age group;6,38 (2) disruption of the integrity of the extracellular matrix is a frequent event in promoting tumor growth and progression;39 (3) COL1A2 methylation leading to its epigenetic inactivation has previously been reported in other cancers, including colorectal cancer and hepatoma;22,40 and (4) direct evidence for a tumor suppressor role has been demonstrated in studies showing that COL1A2 is a downstream target of the epidermal growth factor (EGF)/EGF receptor signal transduction pathway and can suppress cell transformation by Ras and other oncogenes.41,42 In contrast, patients with osteogenesis imperfecta and Ehlers-Danlos syndrome, which arise due to mutations in the COL1A2 gene, do not show increased predisposition to medulloblastoma, although osteogenesis imperfecta type II is associated with abnormalities in CNS development.43 Finally, independent of any role in tumorigenesis, the epigenetic status of COL1A2 may reflect the developmental biology of different medulloblastoma histological and molecular subtypes, and our data raise the hypothesis that infant desmoplastic/nodular tumors may have distinct developmental origins.
J.A.A. and J.C.L. contributed equally to this work. This work was supported by grants from the Katie Trust, the Samantha Dickson Brain Tumor Trust, Charlie’s Challenge, and the SDBTT Kieron Clark Fund. Medulloblastomas investigated in this study include samples provided by the U.K. Children’s Cancer and Leukaemia Group (CCLG) tumor bank, as part of CCLG-approved biological study BS-2007-04. Cell lines D384Med, D556Med, and D425Med were gifts from Dr. D. Bigner (Duke University, Durham, NC, USA). MHH-MED1 and MEB-MED8A were gifts from Prof. T. Pietsch (University of Bonn Medical Centre, Bonn, Germany). UW228-3 was a gift from Dr. J. Silber (University of Washington, Seattle, WA, USA). The remaining cell lines were obtained from the American Type Culture Collection (Manassas, VA, USA). Four normal cerebellar DNAs were a gift from Dr. M. Fruhwald (University of Munster, Munster, Germany).