Array-based DNA methylation profiling of primary lymphomas of the central nervous system

Background Although primary lymphomas of the central nervous system (PCNSL) and extracerebral diffuse large B-cell lymphoma (DLBCL) cannot be distinguished histologically, it is still a matter of debate whether PCNSL differ from systemic DLBCL with respect to their molecular features and pathogenesis. Analysis of the DNA methylation pattern might provide further data distinguishing these entities at a molecular level. Methods Using an array-based technology we have assessed the DNA methylation status of 1,505 individual CpG loci in five PCNSL and compared the results to DNA methylation profiles of 49 DLBCL and ten hematopoietic controls. Results We identified 194 genes differentially methylated between PCNSL and normal controls. Interestingly, Polycomb target genes and genes with promoters showing a high CpG content were significantly enriched in the group of genes hypermethylated in PCNSL. However, PCNSL and systemic DLBCL did not differ in their methylation pattern. Conclusions Based on the data presented here, PCNSL and DLBCL do not differ in their DNA methylation pattern. Thus, DNA methylation analysis does not support a separation of PCNSL and DLBCL into individual entities. However, PCNSL and DLBCL differ in their DNA methylation pattern from non- malignant controls.


Background
Primary lymphomas of the central nervous system (PCNSL) are highly malignant B-cell lymphomas confined to the central nervous system (CNS) with a poor prognosis [1]. They are considered as separate entity within the updated WHO classification [1], although they cannot be distinguished histologically and immunophenotypically from extracerebral diffuse large B-cell lym-phoma (DLBCL). However, based on the remarkably worse clinical course and prognosis it is still a matter of debate whether PCNSL differ from systemic DLBCL with respect to their molecular features and pathogenesis. Interphase cytogenetic and molecular genetic studies have shown that PCNSL share a variety of features with systemic DLBCL. These include rearranged immunoglobulin (IG) gene segments with evidence for ongoing somatic hypermutation, aberrant somatic hypermutation of non-IG genes, translocations affecting the IG and BCL6 genes, gains in chromosome band 18q21, and mutations of the PRDM1 gene [2]. With respect to their gene expression profile PCNSL segregate along the spectrum of systemic DLBCL including the ABC-and GCB-types of DLBCL [3,4].
Epigenetic silencing of functionally important genes by DNA methylation may also contribute to PCNSL development [5][6][7][8][9][10]. By studying DNA methylation of 14 tumor suppressor genes in 25 PCNSL using methylation-specific PCR, Chu et al. [5] demonstrated that all PCNSL had methylated at least two of the genes studied. Although these findings suggest DNA methylation of tumor suppressor genes to be a common event in PCNSL, previous studies on DNA methylation have been limited to a rather small number of a few selected genes. The recent availability of array-based techniques offers the opportunity for a more comprehensive DNA methylation profiling [11,12]. Here, a series of PCNSL was studied for DNA methylation of 1,505 CpG sites from 807 selected genes including a significant number of genes relevant for tumorigenesis. By comparing DNA methylation profiles of PCNSL to those recently obtained for normal hematopoietic controls and 49 systemic DLBCL [12], we identified 194 genes to be differentially methylated between PCNSL and normal controls. Four genes were putatively differentially methylated between PCNSL and systemic DLBCL. Based on the DNA methylation pattern these lymphoma entities did not segregate suggesting DNA methylation pattern of PCNSL and systemic DLBCL to be highly comparable.

DNA extraction and samples
DNA samples from five tumors diagnosed as PCNSL (two of ABC-and three of GCB-subtype, all from female patients) as described recently [3] and classified according to the WHO classification 2008 [1] were subjected to array-based DNA methylation profiling. Systemic lymphoma manifestation was excluded by extensive staging. All studies were approved by local ethics committees. Informed consent was provided according to the Declaration of Helsinki. DNA extraction was performed as described previously [13]. The DNA samples of 49 systemic DLBCL and of 10 normal controls have been described in detail recently [12,14]. The tumor cell con-tent of DLBCL samples was >70% as verified by a panel of experienced pathologists.

DNA methylation profiling using universal BeadArrays
DNA methylation analyses were performed using the GoldenGate Methylation Cancer Panel I (Illumina Inc., San Diego, CA) as described previously [12]. The array allows assaying 1,505 CpG sites from 807 selected genes, which include numerous genes relevant for tumorigenesis including oncogenes, tumor suppressor genes, and genes involved in metastasis, differentiation, cell cycle control, and apoptosis. The complete dataset is provided as supplementary information (Additional file 1). The reproducibility and accuracy of the GoldenGate Cancer Panel I based DNA methylation analysis has been demonstrated extensively [11,12].

Analyses of the DNA methylation data obtained by the GoldenGate Methylation Cancer Panel I
DNA methylation profiling data from 10 hematopoietic controls (analysed in replicates) and 49 systemic DLBCL obtained recently with the same platform served for comparison and have been reanalyzed in this study [12]. As detailed in Additional file 2, normal controls contained two samples of tonsillar germinal center B-cells, two normal peripheral blood samples, and six lymphoblastoid cell lines. Systemic DLBCL were classified as non molecular Burkitt lymphoma (non-mBL) by gene expression profiling and included 29 ABC and 20 GCB-type DLBCL [12,14]. Identification of imprinted CpGs and gender-specifically methylated CpGs on chromosome X has been performed as described recently [12]. Using BeadStudio software (ver.3, Illumina Inc., San Diego, CA) agglomerative hierarchical clustering was performed in PCNSL, systemic DLBCL, and normal hematopoietic controls excluding CpG loci located in imprinted genes and Xchromosomal genes with gender-specific methylation (1,284 CpGs corresponding to 716 genes were included for further analysis). The statistical analyses used to define the different DNA methylation subgroups have been detailed recently [12]. The data obtained from the arrays were not normalized. In addition, differential methylation analysis (DMA) has also been performed using the BeadStudio Software. CpG loci with calculated DiffScores below -30 or above 30 (corresponding to a p-value of p > 0.001, based on a t-statistic) simultaneously showing an absolute DeltaBeta value above 0.3 (corresponding to a difference of 30% in the DNA methylation level) were considered as differentially methylated. The global DNA methylation data per case shows a bimodal distribution in which beta values < 0.25 defines the unmethylated CpGs and beta values> 0.75 the methylated CpGs [12]. Thus, beta values below 0.25 and above 0.75 were selected as threshold-values to define unmethylated and methylated CpG loci for further analysis.

Enrichment for PcG-marks and promoter classes in different DNA methylation groups
Identification of Polycomb target genes and promoter classification into promoters with high (HCP), intermediate (ICP), and low (LCP) CpG content respectively, are based on recent publications [15][16][17]. Annotation lists of Polycomb group (PcG) marks and promoter classes were compared with the genes analyzed for methylation via gene symbol or locuslink ID. Proportions of genes in the different methylation groups were compared with respect to PcG-marks and promoter classes as described previously [12].

Principal component analysis (PCA)
PCA has been performed using the Omics Explorer, Version 2.0 Beta (Qlucore AB, Lund, Sweden).

Gene ontology
The "Gene Annotation Tool to Help Explain Relationships" (GATHER; http://gather.genome.duke.edu) has been used to determine the enrichment of individual ontology terms in the group of genes differentially methylated in PCNSL and normal controls (with respect to the overall composition of the GoldenGate array). Ontology terms with an at least twofold increase in numbers of genes between the tested group and the array containing at least five individual genes, were tested for significance of enrichment by performing Fisher's exact test.

Statistics
Fisher's exact test and Mann-Whitney test were performed using GraphPad Prism version 4.02 for Windows, Graph-Pad Software, San Diego, CA.

Array-based DNA methylation profiling of PCNSL
The BeadArray technology [11,18] was applied to perform quantitative DNA methylation analyses of 1,505 individual CpGs corresponding to 807 genes relevant for tumorigenesis in five PCNSL. Results were compared to those recently obtained for ten hematopoietic controls and 49 systemic DLBCL [12]. Genes known to be imprinted and X-chromosomal genes with gender-specific methylation were excluded from the analysis, as they are partially methylated under physiological conditions and might represent a confounding variable to classify cases according to their DNA methylation profile. Thus, a total of 1,284 CpGs corresponding to 716 genes entered the analysis.

Identification of genes differentially methylated between PCNSL and hematopoietic controls
PCA based analysis applying most stringent conditions differentiates normal controls from both PCNSL and systemic DLBCL samples in a highly significant fashion (p = 1.2 × 10 -12 , q = 3.9157 × 10 -11 ; Fig. 1A). 33 CpG loci (cor-responding to 30 genes) differentiating between PCNSL and controls performed in replicates were significantly hypermethylated while three CpG loci showed hypomethylation in PCNSL compared to hematopoietic controls (Fig. 1B).
Supervised cluster analysis of the 1,284 CpGs comparing the five PCNSL to ten hematopoietic controls resulted in 296 differentially methylated CpGs corresponding to 194 genes (Additional file 3). Of these 194 genes, 153 genes were hypermethylated while 41 genes were hypomethylated in PCNSL. Unsupervised cluster analysis of the five PCNSL and the 10 hematopoietic controls along with the 49 systemic DLBCL samples for all 1,284 CpGs separated all lymphoma samples from normal controls (Fig. 1C). These findings suggest that the methylation profile of PCNSL strongly differs from that of normal hematopoietic tissues including tonsillar germinal center B-cells and lymphoblastoid cells.

CpG loci differentially methylated in PCNSL and systemic DLBCL
The methylation pattern of both PCNSL and systemic DLBCL were very heterogeneous with respect to total DNA methylation (data not shown). To address the question whether the DNA methylation pattern of PCNSL differs from that of systemic DLBCL, we compared DNA methylation values of 1,284 CpG loci from the five PCNSL with those of the 49 systemic DLBCL. Of the 194 genes differentially methylated between normal controls and PCNSL, using the same analysis approach, 157 genes were among the 174 genes differentially methylated between controls and DLBCL (Additional file 4). Neither supervised nor unsupervised cluster analysis separated PCNSL from systemic DLBCL (Fig. 1C, D, and data not shown). A differential methylation analysis (DMA) of PCNSL and systemic DLBCL cases yielded only four CpGs differentially methylated between PCNSL and systemic DLBCL (ESR1, EFNA1, MATK, and PDE1B). However, hierarchical cluster analysis performed on these four CpG loci failed to distinguish PCNSL from systemic DLBCL (Additional file 5). Thus, methylation analysis of these CpG loci does not allow a proper assignment of unknown samples to either PCNSL or systemic DLBCL. Furthermore, PCA failed to differentiate between PCNSL and systemic DLBCL (Fig. 1A, and data not shown). In conclusion, PCNSL did not exhibit a specific DNA methylation signature as compared to systemic DLBCL including more than 700 genes which had entered this study.

Classification of genes based on their DNA methylation pattern in PCNSL as compared to non-neoplastic hematopoietic tissues
To get further insight into DNA hypermethylation in PCNSL, we compared the class of genes methylated in PCNSL but not in controls (meP/umC, n = 138) to the classes of genes unmethylated or methylated in both PCNSL and controls (umP/umC, n = 348 genes and meP/ meC, n = 91 genes), respectively. These groups provided the basis for further analysis (Additional file 6).

The group of genes methylated in PCNSL but unmethylated in non-neoplastic hematopoietic controls is enriched for polycomb targets in embryonic stem cells and promoters with high CpG content
The group of genes unmethylated in the hematopoietic controls but methylated in PCNSL was highly significantly enriched for genes repressed by the polycomb (PcG) repressing complexes (PRC2) in embryonic stem cells compared to the genes present on the array ( Fig. 2A; RR = 2.55; p < 0.0001, Fisher's exact test). This group was also enriched for the simultaneous presence of all three PRC2 marks as described previously [12,15,16]: EED (RR = 2.62), SUZ12 (RR = 2.68) and 3 mK27-H3 (RR = 2.54) (Fig. 2B). However, the targets of a specific PRC2 mark were not preferentially enriched in any group (Fig. 2C). While the percentage of PcG target genes in the group of genes unmethylated (um) in both controls (C) and PCNSL (P) (umP/umC) did not differ significantly from the one of the GoldenGate array (RR = 0.81; p = 0.1675), the percentage of PcG target genes was significantly reduced (RR = 0,30; p = 0.0010) in the group of genes methylated (me) in both PCNSL and normal controls (meP/meC) ( Fig. 2A, B).

(1%)
Since it was reported previously that particular promoters with high CpG content and CpG islands become methylated in tumors [19][20][21][22], we have sorted the genes in the groups defined above in genes having promoters with high (HCP), intermediate (ICP), or low (LCP) CpG content [12], respectively. While the group of genes methylated in both PCNSL and controls (meP/meC) was enriched for genes with low CpG content promoters (LCP; RR = 2.21; p < 0.0001), the remaining two groups were enriched by genes having promoters with high CpG content (umP/umC: RR = 1.17; p = 0.0023; meP/umC: RR = 1.35; p < 0.0001; Fig. 2D). The CpG loci belonging to genes methylated in PCNSL and controls were furthermore significantly enriched for loci located outside of CpG Islands (RR = 2.24; p < 0.0001).

The group of genes differentially methylated in PCNSL is enriched for genes involved in neurological processes and cellular signaling pathways
As genes studied with the methylation-specific BeadArray were selected for their involvement in tumorigenesis, they will "by definition" be deregulated in various tumors. Even considering this bias, we tested whether genes differentially methylated in PCNSL and normal controls were Enrichment of PcG marks and HCP promotors in the group of genes methylated in PCNSL but unmethylated in non-neoplastic hematopoietic tissues Genes de novo methylated in PCNSL (meP/umC) predominantly had promoters with high CpG content (p < 0.0001). In contrast, genes having promoters with low CpG content were enriched in the group showing high methylation in PCNSL and normal controls (meL/meC; p < 0.0001).

Genes with PcG marks [%]
enriched for specific Gene Ontology (GO) terms. Surprisingly, we observed that terms particularly involved in neurophysiological and perception processes were significantly enriched (Additional file 7). However, the same GO terms were also enriched in systemic DLBCL. Thus, a PCNSL specific enrichment of individual GO terms differing from DLBCL was not identified.

Discussion
In the present study, we determined DNA methylation patterns in tumor samples of five patients suffering from PCNSL and compared them to the methylation pattern of 49 systemic DLBCL and ten normal hematopoietic controls, which have been characterized in detail before [12,14]. The GoldenGate Cancer Panel I used in this study offers the possibility to analyze 1,505 single CpG loci corresponding to 807 genes in parallel. To prevent any bias because of gender specific DNA methylation and imprinting effects, X-chromosomal genes and genes known to be imprinted, the methylation extent of which varies under normal physiological conditions, have been excluded from further analyses. The reproducibility and the accuracy of this array based approach have been extensively demonstrated before [12,18].
Previous studies on DNA methylation in PCNSL were limited by their restriction to a low number of selected genes. Our data are in line with Chu et al. and Gonzales-Gomez et al. [5,8] [27] and analysis of the CDKN2A locus might be affected by recurrent deletions of this gene in PCNSL [28].
A further analysis of genes which were unmethylated in hematopoietic controls and methylated in PCNSL (meP/ umC) showed a significant enrichment of genes which are repressed by components of the PRC2 in embryonic stem cells. Components of the PRC2 complex are essential for the maintenance of the undifferentiated state of embryonic stem cells by suppressing genes called PcG target genes, the activation of which leads to cellular differentiation [16,29]. Indeed, hypermethylation of PcG target genes has been described for several tumor entities [30] including systemic mature aggressive B-cell lymphomas [12]. This enrichment of genes controlled by PRC2 during stem cell maintenance could lead to the hypothesis that the tumor cells of PCNSL cells derive from a stem cell like progenitor cell. Alternatively, dedifferentiation and epigenetic reprogramming of an already differentiated precursor cell could also explain the DNA methylation pattern [12]. The latter model is supported by previous data lending support for the hypothesis that the tumor cells derived from mature GC exit B-cells [3]. Finally, genes anyhow silenced in the hematopoietic lineage could become switched off by DNA methylation in tumor cells [12].
Interestingly, genes with three PcG-marks in the promoter region were significantly enriched, while there was no enrichment of a specific mark (neither EED, SUZ12 or 3 meK27-H3) in PCNSL. This is in line with results obtained for systemic mature B-cell lymphomas [12]. Furthermore, also similar to the scenario described for other B-cell lymphomas, in particular genes with high CpG content promoters become methylated in PCNSL. This correlates well with the known phenomenon that CpG islands in promoter regions become methylated during tumorigenesis [31][32][33].
In a recent study addressing genomic imbalances in PCNSL [34] several genes hypermethylated in PCNSL, including e.g. ERBB3, PDE1B, ASCL1, and BCAM were located in regions with recurrent genomic gains. However, they were not expressed in the tumor [3]. Since DNA methylation has been associated with gene repression in numerous studies [35][36][37], our data on DNA methylation might offer a putative explanation why increased gene dosage does not lead to increased gene expression and even transcriptional silencing in PCNSL. However, since a number of recent studies have shown little correlation between tumor specific hypermethylation and changes in gene expression [12,38,39], this must be addressed by future studies.

Conclusions
Comparing the DNA methylation state of 1,284 CpG loci in PCNSL, systemic DLBCL, and hematopoietic controls, a DNA methylation pattern exclusively specific for PCNSL was not identified. The methylation data do not allow dis-tinguishing PCNSL from systemic DLBCL on the basis of DNA methylation levels. However, the present study does not exclude the possibility that analysis of a larger number of CpG loci of the genome in a larger series of PCNSL might identify PCNSL specific features.