- Top of page
- Material and Method
- Supporting Information
The downregulation of specific genes through DNA hypermethylation is a major hallmark of cancer, although the extent and genomic distribution of hypermethylation occurring within cancer genomes is poorly understood. We report on the first genome-wide analysis of DNA methylation alterations in different neuroblastic tumor subtypes and cell lines, revealing higher order organization and clinically relevant alterations of the epigenome. The methylation status of 33,485 discrete loci representing all annotated CpG islands and RefSeq gene promoters was assessed in primary neuroblastic tumors and cell lines. A comparison of genes that were hypermethylated exclusively in the clinically favorable ganglioneuroma/ganglioneuroblastoma tumors revealed that nine genes were associated with poor clinical outcome when overexpressed in the unfavorable neuroblastoma (NB) tumors. Moreover, an integrated DNA methylation and copy number analysis identified 80 genes that were recurrently concomitantly deleted and hypermethylated in NB, with 37 reactivated by 5-aza-deoxycytidine. Lower expression of four of these genes was correlated with poor clinical outcome, further implicating their inactivation in aggressive disease pathogenesis. Analysis of genome-wide hypermethylation patterns revealed 70 recurrent large-scale blocks of contiguously hypermethylated promoters/CpG islands, up to 590 kb in length, with a distribution bias toward telomeric regions. Genome-wide hypermethylation events in neuroblastic tumors are extensive and frequently occur in large-scale blocks with a significant bias toward telomeric regions, indicating that some methylation alterations have occurred in a coordinated manner. Our results indicate that methylation contributes toward the clinicopathological features of neuroblastic tumors, revealing numerous genes associated with poor patient survival in NB.
Neuroblastoma (NB) is a childhood neuroblastic tumor arising from precursor cells of the sympathetic nervous system. These tumors are composed largely of immature neuroblasts and display a broad clinical spectrum ranging from rapid advancement and death due to disease to spontaneous regression.1 The diverse clinical behavior of NB is mirrored by extensive heterogeneity in the genetic abnormalities exhibited in the tumors, with MYCN amplification or deletion of chromosome 11q material representing distinct tumor subtypes with generally unfavorable clinical outcomes. In contrast, tumors that are characterized primarily by hyperdiploidy and few structural chromosome abnormalities have a more clinically favorable outcome. Other neuroblastic tumors include benign ganglioneuroma, which are composed primarily of Schwannian stromal and mature ganglion cells, and an intermediate tumor, ganglioneuroblastoma (GNB), which is stromal rich but also possess immature neuroblasts. GNB tumors are clinically less aggressive than NB.
Aberrant hypermethylation of cytidine bases within gene promoter regions is a well-known mechanism for the transcriptional silencing of tumor suppressor genes in many forms of cancer.2, 3 Studies based on a limited number of candidate genes indicate that aberrant hypermethylation is a clinically relevant event in NB tumors. For example, CASP8, a key regulator of apoptosis, and RASSF1A, a well-documented tumor suppressor gene, have been noted to undergo frequent hypermethylation in clinically unfavorable subtypes of NB.4–10 Most significantly, the methylation status of a limited set of genes indicated that patterns of methylation might prove useful for discriminating clinical risk groups of NB.4, 11, 12 There have also been reports indicating that some genes might become methylated in a coordinated manner, a “methylator phenotype,”11, 13 and that genes with methylated promoter regions display some level of clustering.7 A genome-wide assessment of DNA hypermethylation in either NB or the clinically more favorable neuroblastic tumors, however, has never been carried out.
Here, using genome-wide DNA methylation profiling, we identify DNA methylation differences found at gene promoter regions that might contribute to the clinicopathological differences between NB, GN and GNB tumors. Recently, it has also become apparent that recurrent large-scale blocks of contiguously methylated CpG islands occur in several types of cancer, including colorectal,14, 15 breast,16 astrocytoma17 and Wilm's tumors.18 In this report, we demonstrate for the first time that these contiguously hypermethylated regions also occur in neuroblastic tumors and that they have a highly biased distribution toward the terminal regions of chromosomes.
- Top of page
- Material and Method
- Supporting Information
To date, studies assessing methylation changes in NB have concentrated on a select number of candidate genes,5–7, 9, 10, 28, 29 but a genome-wide analysis of promoter region and/or CpG island DNA methylation has been lacking. We provide the first genome-wide assessment of DNA methylation patterns in NB, GN and GNB neuroblastic tumors. Our analysis indicates that recurrent large-scale blocks of contiguously hypermethylated promoter/CpG island sites (n = 70) occur in the tumors, and that there is a significant bias for these sites toward telomeric ends (31% of the blocks occurring <2 Mb from telomere). Chromosome 19 displayed the highest number of methylation blocks (eight blocks in total), which may be due to the greater gene density of this chromosome. The blocks mapping to imprinted regions (n = 3) or chromosome X (n = 4) are possibly related to normal developmental processes, which leaves a total of 63 methylation blocks that are potentially disease related. In contrast, DNA methylation analysis of three human chromosomes in normal tissue identified a significant correlation for methylation of regions only over distances ≤1,000 bp,30 suggesting that larger methylated regions may be disease specific.
A number of groups have also noted similar clusters of hypermethylated sites in other forms of cancer,14–18 but to the best of our knowledge, our report is the first demonstrating significant overrepresentation of methylated blocks toward the telomeres. In some cancer cells, particularly those that are telomerase active, subtelomeric repeats exhibit significant levels of methylation,31 and it is tempting to speculate that the mechanism leading to hypermethylation of these repeats can extend methylation further proximal. Interestingly, one of the hypermethylated blocks in NB included the HOXD family on chromosome 2, which was also identified as a region of large-scale DNA methylation in breast cancer16 and in brain tumors,17 indicating that at least one of these blocks occurs in multiple forms of cancer. As the characterization of these large regions of methylation has only been carried out in a very limited number of cancers, future studies should focus on determining the level of similarity of our 63 regions across different cancer types to elucidate further their relevance to cancer development.
NB, GN and GNB tumors differ greatly in their histopathological characteristics and in their clinical aggressiveness. Here, we demonstrate that large differences in the DNA methylation status of numerous loci also exist between the clinically unfavorable NB and more favorable GN/GNB tumor groups, allowing the stratification of both groups on the basis of DNA methylation profiling. The number of genes that are consistently hypermethylated in the GN/GNB group relative to NB is far greater (227 genes) than the opposite comparison (11 genes) even though there were approximately the same average number of hypermethylated sites per tumor in both categories. The reason for this bias is likely due to the extensive heterogeneity of the NB tumor group, consisting of at least three distinct genetic subtypes, MYCN amplified, 11q- and hyperdiploid. An analysis of a larger tumor cohort would likely identify additional hypermethylated sites specific to NB subtypes.
Of the genes that were hypermethylated in NB tumor group and not in the GN/GNB, the lower expression of one gene PCID2 correlated with poor survival in the 88 tumor set (p = 0.011).
This gene codes for a 399 amino acid protein containing a PCI domain and has no assigned specific function (Swiss-Prot:Q5JVF3), making it an interesting novel NB candidate gene for further follow-up. ATXN2 was also predominantly methylated in the NB tumors compared to the GN/GNB group. Wiedemeyer et al.32 have previously reported that the ectopic expression of wild-type ataxin-2 in SHEP Tet21N NB cells has a profound effect on the susceptibility of the cells to undergo apoptosis. Our analysis suggests that ATXN2 undergoes methylation in the more aggressive primary NB tumors, which possibly leads to an increased resistance to apoptosis.
Overexpression of NME2, a nucleoside diphosphate kinase that has been previously associated with an aggressive disease course in NB,33 was one of the genes identified as being hypermethylated in the GN/GNB tumors but not in NB. Overexpression of this gene was significantly associated with poor survival in our NB tumor cohort. Other genes that were selectively methylated in the GN/GNB group included EIF4G1 and TRIM32. High levels of EIF4G1 protein have been associated with the formation of tumor cell emboli, which promote invasion in breast cancer,34 and it has also been shown to be highly overexpressed in advanced squamous cell carcinoma.35 Interestingly, high expression of EIF4G1 in our tumor cohort was also associated with poor survival. TRIM32 expression is elevated in human head and neck cell carcinoma, and it has also been shown that overexpression of this gene leads to enhanced cell growth and transforming ability by promoting degradation of ABI2.36
It can be hypothesized that genes that are hypermethylated exclusively in GN/GNB contribute to the more benign phenotype of this group. An important caveat to this hypothesis is that the GN/GNB tumors have much larger amounts of Schwann cells, thus the methylation profiles could be representative of a cell type that might be of nonmalignant origin,37 although Mora et al.38 have suggested that these cells are of tumor origin. Nevertheless, the stromal cell component seems to contribute to a less aggressive phenotype,39 and the hypermethylation of specific genes within these cells could be contributing to this phenomenon. Further studies involving DNA methylation profiling of isolated cell types will have to be carried out to further address this issue.
DAC, commonly known as decitabine, causes reactivation of epigenetically silenced genes and is a widely reported method to validate potential gene targets to establish their silencing as a consequence of DNA methylation.40 Genome-wide screens for epigenetically silenced genes using expression arrays have been used previously to identify novel target genes in various cancer types.41 In our study, we observe a substantial number of genes that are reexpressed after DAC treatment; however, only 30% of these genes exhibited substantial promoter region methylation before DAC treatment. In addition, DAC treatment also led to greater than 5-fold downregulation of similar numbers of genes. Exposing cells to DAC clearly causes a rampant cascade of secondary events, resulting from activation of transcription factors or miRNAs, which makes interpretation of expression microarrays exceedingly complicated. In agreement with our results, most DAC studies based on small numbers of genes have also reported only partial overlap of genes that are hypermethylated and that are reexpressed in response to DAC.42, 43In vitro studies involving DAC are further complicated by the fact that there were significantly more peaks of hypermethylation detected in cell lines when compared to primary tumors, consistent with the observation of Smiraglia et al.44 on other types of cancer. Our results indicate that the DNA methylation status for specific genes in cell lines is not necessarily indicative of the situation in primary tumors. Nevertheless, a more precise understanding of the effects of DAC (decitabine) on gene expression is certainly warranted given that it is an FDA-approved cancer therapeutic.45
Our integrated analysis of genome-wide DNA methylation, aCGH and mRNA expression profiles has identified numerous new candidate loci for potential functional studies, and the value of a similar approach was also demonstrated in the analysis of osteosarcoma cell lines by Sadikovic et al.46 An interesting candidate identified from our study is RBP7 (CRBP IV), a cellular retinol-binding protein gene on chromosome 1p36.22, which was concomitantly deleted and hypermethylated in ∼50% of MYCN-amplified tumors and reexpressed in cell lines treated with DAC. This locus was also shown to be hypermethylated and reactivated by DAC treatment in nasopharyngeal carcinoma cell lines and might confer resistance to retinoic responsiveness.47 In addition, we show that promoter region hypermethylation of CDC42 is an alternative method of silencing this gene, which was previously shown to be downregulated by 1p deletion and repressed by MYCN binding.48 Ectopic upregulation of this locus causes NB cells to undergo differentiation.
In conclusion, through the integration of genomic, epigenetic and expression data, we have identified differentially methylated genes related to more aggressive neuroblastic tumor phenotypes and large-scale blocks of contiguously hypermethylated sites with an enrichment at telomeric regions. Consistent and widespread differences in methylation patterns between neuroblastic tumor subtypes indicate that epigenetic differences contribute to the phenotypic characteristics of these tumors. The identification of concurrent hypermethylation/deletion inactivation events in genes that are related to poor survival in NB warrants functional follow-up to assess their exact involvement in disease mechanism.