|Home | About | Journals | Submit | Contact Us | Français|
Epigenetic remodeling is a hallmark of cancer, with the frequent acquisition of de novo DNA methylation in CpG islands. However, the functional relevance of de novo DNA methylation in cancer is less well-defined. To begin to address this issue in B-cells, we used BeadArray assays to survey the methylation status of 1,500 cancer-related CpG loci in two molecular subtypes of diffuse large B-cell lymphoma (ABC-DLBCL and GCB-DLBCL) and cognate normal B-cell populations. We identified 81 loci that showed frequent de novo DNA methylation in GCB-DLBCL and 67 loci that showed frequent de novo DNA methylation in ABC-DLBCL. These de novo methylated CpG loci included reported targets of polycomb repressive complexes (PRC) in stem cells. All candidate loci in GCB-DLBCL are proximal to genes that are poorly expressed or silent in purified normal germinal center (GC) B-cells. This is consistent with the hypothesis that de novo DNA methylation in cancer is more frequently involved in the maintenance rather than the initiation of gene silencing (de novo repression). This suggests that epigenetic switching occurs during tumorigenesis with de novo DNA methylation locking in gene silencing normally mediated by transcriptional repressors. Furthermore, we propose that similar to de novo genetic mutations, the majority of de novo DNA methylation events observed in tumors are passengers not causally involved in tumorigenesis.
Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous group of cancers arising from normal B-cell populations that have acquired characteristic groups of genetic and epigenetic changes [Coiffier, 2001; De Paepe and De Wolf-Peeters, 2007]. Gene expression profiling of tumor and normal B-cells have identified two major subgroups of DLBCL, germinal center B-cell-like DLBCL (GCB-DLBCL) and activated B-cell-like DLBCL (ABC-DLBCL), which have distinct clinical outcome [Alizadeh et al., 2000; Lossos and Morgensztern, 2006; Rosenwald et al., 2002; Wright et al., 2003]. In addition, these subgroups show differences in chromosomal abnormalities [Bea et al., 2005; Chen et al., 2006; Lenz et al., 2008; Rosenwald et al., 2002; Tagawa et al., 2005] and miRNA expression [Lawrie et al., 2007; Malumbres et al., 2009]. Recent evidence that ABC-DLBCL and GCB-DLBCL cases also show different responses to chemotherapeutic regimens raises the possibility that genetic testing could inform more personalized medical interventions in the future [Dunleavy et al., 2009].
In addition to these extensive genetic characterizations, DNA methylation profiling studies of different B-cell lymphomas are now being conducted [Guo et al., 2005; Killian et al., 2009; Martin-Subero et al., 2009; Pike et al., 2008; Shi et al., 2007; Yang et al., 2003]. Previously, we used qualitative CpG island microarray and PCR-based assays to DNA methylation in DLBCL [Pike et al., 2008]. In these studies, we identified CpG islands that were frequently methylated in DLBCL and found that the bulk of these CpG islands were proximal to genes that were poorly expressed in all tumors, regardless of methylation status [Pike et al., 2008]. However, we could not determine if the observed DNA methylation was acquired during tumorigenesis or if it was associated with the silencing of genes normally expressed in B-cell populations thought to give rise to these tumors.
In this study, we use quantitative BeadArray assays to evaluate the DNA methylation status of a novel expanded set of CpG loci proximal to carefully chosen cancer-related genes in a larger cohort of DLBCL and purified normal B-cell populations of origin. This now allows us to identify DNA methylation acquired de novo during B-cell tumorigenesis and begin to investigate the functional relevance DNA methylation has on gene expression in B-cells. This focused epigenetic evaluation provides an important counterpart to studies aimed at determining the functional significance of de novo mutations acquired during B-cell tumorigenesis.
Frozen tissue samples from diagnostic tumor biopsies are obtained from patients before anthracycline-based chemotherapy at the University of Nebraska Medical Center. DNA and RNA samples are from 46 specimens, 21 from patients with ABC-DLBCL, 25 from patients with GCB-DLBCL. There was consensus central pathology re-review of the specimens to confirm the diagnosis of DLBCL and the samples had >75% tumor cells. Clinical information has been obtained from all patients according to a protocols approved by the University of Nebraska Medical Center Institutional Review Board.
Germinal center B-cells were purified from healthy lymph node tissues obtained from adult human donors. Lymph node tissues were minced and passed through 30 micrometer nylon mesh to obtain single cell suspensions. B-cells were enriched from the cell suspensions by positive selection with CD19 multisort kit (Miltenyi Biotec, Auburn, CA). Germinal center B-cells were isolated from the total B-cells by sorting IgD−CD20+CD38+ B-cells using direct fluorescence-activated cell sorting (FACS). In keeping with previous studies [Liu and Banchereau, 1997], analytical FACS analyses confirmed that a greater than 85% purity. Human B-cells were purified from peripheral blood obtained from healthy adult donors. Monocytes were isolated using Ficoll density gradients. CD19+ B-cells were isolated by positive selection using the CD19 multisort kit (Miltenyi Biotec, Auburn, CA). The purity was greater than 85% [Allman et al., 1996].
The isolated CD19+ B-cells were cultured at 2×106 cells/ml in RPMI containing 10% fetal calf serum (FCS), 1μg/ml glutamine, 1μg/ml each of penicillin and streptomycin, and 5×105 mol/L 2-mercaptoethanol. Activated peripheral B-cells were obtained by exposing purified cultured CD19+ B-cells to one of three types of mitogenic agents, 25μg/ml LPS, 1μg/ml IL4 plus 1μg/ml CD40 ligand, or 20μg/ml anti-IGM-Ab, for 44 hours. These treatments were used in the prior studies to obtain an activated B-cell gene expression signature [Alizadeh et al., 2000].
The DNA methylation levels of 1536 specific CpG sites in 371 genes (1~9 CpG sites per gene) were measured on the Illumina (San Diego, CA) GoldenGate BeadArray® platform, as previously described [Bibikova et al., 2006]. Briefly, this primer extension-ligation assay monitors the extension of oligonucleotide primers designed to bind to methylated or unmethylated CpG sites in bisulfite-converted templates. Extended templates are PCR amplified, labeled with fluorescent dyes, and hybridized to the BeadArrays. The relative hybridization signals from PCR products derived from primers specific for methylated and unmethylated templates are used to calculate relative fractions of DNA methylation in the original templates.
The GoldenGate DNA methylation assays measure the DNA methylation levels of a given locus as β-values ranging from 0 (no DNA methylation detected) to 1 (complete DNA methylation). Following the nomenclature established in prior Illumina GoldenGate BeadArray DNA methylation studies, we defined a hypermethylated locus as having β>0.8 and a hypomethylated locus as having β<0.2 [Killian et al., 2009]. We deemed loci with a β>0.4 (i.e. loci with any measureable level of DNA methylation in our assay system) as being “methylated”. We chose β>0.4 since it is 0.2 units above the upper threshold of the hypomethylated category. Since the GoldenGate assays are reported to be able to distinguish 0.15 β-units of DNA methylation [Bibikova et al., 2006], this provides a reasonable definition of the lower limits of β-values that reflect the presence of DNA methylation.
We applied multiple quality control metrics to analyze high quality DNA methylation data. Assays showing β>0.6 in either of two whole genome amplification (WGA) fully hypomethylated controls or β<0.4 in either of two M.SssI-treated fully methylated controls were excluded from further analysis. Likewise, CpGs associated with imprinted genes (based on the publicly available databases http://igc.otago.ac.nz/home.html and http://www.geneimprint.com/site/genes-by-species and [Murphy and Jirtle, 2003], as described in [Martin-Subero et al., 2009]), X-chromosomal genes, and single-nucleotide polymorphism known to affect the performance of the Illumina GoldenGate BeadArray DNA methylation platform [Byun et al., 2009] were excluded from the analysis.
Bisulfite conversion was performed on 0.5–1μg tumor DNA samples using EZ DNA methylation kit (Zymo Research, Orange, CA). The converted DNA samples served as templates for PCR reactions using primers were designed using Methprimer software [Li, 2007; Li and Dahiya, 2002] (Supplemental Table 1). The resultant amplicons were sub-cloned and at least 12 colonies were sequenced for each selected CpG site. Sequence analysis and visualization was completed using BiQ Analyzer Software [Bock et al., 2005].
Total RNA from normal GC B-cells was isolated and subjected to gene expression profiling analysis on Human Genome U133Plus2 Arrays (Affymetrix, Santa Clara, CA) according to the manufacturer's recommended protocols. We report normalized log2-transformed gene expression data for probe tilings with minimal cross-hybridization potential (i.e. disregarding `_x_at' probe sets designated by Affymetrix, Santa Clara, CA) that interrogate NCBI-designated Reference Sequence (RefSeq) transcripts associated with genes proximal to or containing candidate de novo DNA methylation in GCB-DLBCL (Supplemental Table 2). All scaled fluorescent intensity values and .cel files from normal B-cells are available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) repository (http://www.ncbi.nlm.nih.gov/geo/) under Series Accession Number GSE16712.
Our initial goal was to identify de novo DNA methylation in DLBCL that are acquired during B-cell tumorigenesis. To address this issue, we conducted a wide-spread survey of DNA methylation in cancer-related genes in a cohort of GCB-DLBCL (N=25), ABC-DLBCL (N=21), purified normal germinal center B-cells (N=6, DNA obtained from five donors with two lymph nodes obtained from one donor), and activated peripheral B-cells (N=6, cultured under 3 different stimulation conditions each - see Methods). We used Illumina GoldenGate BeadArray® assays to measure the DNA methylation levels of over 1,500 carefully selected cancer-related CpG loci as β-values ranging from 0 (no DNA methylation detected) to 1 (complete DNA methylation) (see Methods).
Although we have acquired DNA methylation data on all of 1,500 loci, we focused our attention on 690 autosomal loci that passed rigorous quality control metrics and are not subject to the confounding effects of imprinting (see Methods). These allowed us to conduct for a focused and rigorous analysis DNA methylation and gene expression profiles. Due to technological improvements and the focused nature of the current study, we profile a different group of CpG loci from our previous studies that investigated a discovery set of only 14 DLBCL tumors [Pike et al., 2008]. All β-values for these 690 loci are available in Supplemental Table 3, with complete annotation provided in Supplemental Table 4.
In Fig. 1, we present an unsupervised hierarchical clustering analysis of DNA methylation data from 690 autosomal loci proximal to (defined hereafter as being located within 1,500-bp of the transcription start site, as detailed in Supplemental Table 3) or within 238 unique cancer-related genes (see Methods). With only four exceptions (i.e. GCB-DLBCL tumors 4, 8, 10, and 16), 91.3% (42/46) of the tumors clustered separate from the normal B-cells (N=24, including all activated and germinal center B-cells). However, the ABC-DLBCL and GCB-DLBCL tumors could not be distinguished based on this unsupervised clustering method. As discussed in our previous study [Pike et al., 2008], this could be due to the non-concordance of genes whose expression levels can be used to differentiate the ABC-DLBCL and GCB-DLBCL subgroups and the CpG loci interrogated by the BeadArrays. Likewise, GC B-cells could not be distinguished from the activated peripheral blood B-cell cultures, generated using three different established protocols, in our DNA methylation-based hierarchical clustering analyses (Fig. 1).
From these hierarchical clustering analyses, it was apparent that methylated loci in DLBCL were either frequently methylated in normal B-cell populations or acquired de novo during tumorigenesis (Fig. 1 and Supplemental Fig. 1). Below, we will discuss our efforts to identify methylated loci that fall into these two categories and relate them to existing gene expression and copy number data.
To identify robust DNA methylation in GCB-DLBCL tumors that derive from the epigenetic state of normal GC B-cell populations, we focused on candidate loci in which every allele present was methylated (β>0.8) in >95% of GCB-DLBCL tumors and in >95% of normal B-cell preparations. In the latter case, we combined data from all purified germinal center (GC) B-cells and activated B-cells (N=24) since they could not be distinguished from one another by hierarchical clustering analyses (Fig. 1). This observation is especially striking since these samples were obtained from different donors. This suggests that the cancer-related CpG loci we examined have robust B-cell methylation signatures and that B-cell subtype-specific DNA methylation profiles, at least for the most abundant B-cell species, are far less frequent.
Overall, we identified 16 CpG loci that our criteria for being methylated in both normal and cancerous B-cells. These were proximal to or within 13 genes (ASCL2, BRCA2, C4B, COL1A2, CTBP1, DDX17, HFE, MC2R, PPP2R1B, SMARCA3 (aka HLTF), TGFB1, UBB, and VHL) (Supplemental Table 5). These genes did not appear to have highly related specific functions as GeneOntology (GO) analyses did not identify any enriched functional categories (≥2 genes, P<0.01).
We note that the GoldenGate BeadArray platform evaluates the relative levels of DNA methylation of all alleles present in a sample. In normal diploid B-cells, we can assume that a β-value of 1 reflects DNA methylation of both autosomal alleles. However, in a tumor sample, a β-value of 1 could reflect DNA methylation of all copies of a haploid locus (due to a deletion), polyploid locus (due to an amplification), or diploid locus (no copy number change). Nevertheless, based on array-CGH (comparative genomic hybridization) analyses of 17 GCB-DLBCL in our cohort, two of the CpG loci (MC2R and VHL) that were methylated in GCB-DLBCL and normal B-cells showed both a deletion and DNA methylation (β>0.8) in one tumor sample each (Supplemental Table 5). Intriguingly, loci proximal to or within ASCL2, BRCA2, C4B, COL1A2, HEF, MCR2, PPP2R1B, and UBB (all methylated in GCB-DLBCL tumors and normal B-cells) showed DNA methylation (β>0.8) and an increase in gene copy number in one or two GCB-DLBCL tumors profiled by array-CGH (Supplemental Table 5). This suggests that DNA methylation present in normal B-cells was retained after amplification that occurred during tumorigenesis.
Next, we focused on the differences between ABC-DLBCL and normal B-cells (N=24, including all activated B-cells and germinal center B-cells). These differences were less well-defined since activated B-cell cultures may not necessarily represent the epigenetic state of normal B-cell populations that result in ABC-DLBCL (which are currently unknown), even though their gene expression profiles show similarities [Alizadeh et al., 2000; Rosenwald et al., 2002; Wright et al., 2003]. A total of seven loci proximal to or within six genes were hypermethylated (β>0.8) in >95% of ABC-DLBCL and >95% of all normal B-cell samples (Supplemental Table 6). These included ASCL2, C4B, COL1A2, DDX17, MC2R, and RARRES1. Interestingly, all genes but RARRES1 were also methylated in ≥95% of the GCB-DLBCL tumors. Again, the specific functions of this subset of genes were not highly related given that GO analyses did not identify any enriched functional categories (≥2 genes, P<0.01).
To identify de novo DNA methylation frequently acquired during GCB-DLBCL tumorigenesis, we screened for loci that were methylated (β>0.4) in ≥20% of GCB-DLBCL tumors (i.e. at least 5) and hypomethylated (β<0.2) in >95% of all normal B-cell samples. We chose less stringent criteria for DNA methylation in tumors (β>0.4) since we wanted to include all loci in which de novo DNA methylation occurred in a single allele in the absence of amplification. Overall, we detected 81 such candidate loci proximal to or within the following 42 genes: ABO, ADAMTS12, APBA1, APP, ASCL2, CAV1, CCNA1, CDH3, COL1A2, CSF1, CSPG2 (aka VCAN), CYP1A1, EPHA3, EPO, EYA4, F2R, FN1, HRASLS, IGFBP7, IGSF4 (aka CADM1), IL17RB, IMPACT, LAT, LOX, LRP2, MT1A, MYCN, MYOD1, NEFL, NPY, PAX6, POMC, PROK2, PTPRO, RAB32, RARB, RASGRF1, RET, SEZ6L, TMEFF1, TNFRSF10D, and ZMYND10 (Supplemental Table 7). As above, the specific functions of these genes were not highly related given that GO analyses did not identify any enriched functional categories (≥2 genes, P<0.01).
Based on previous array-CGH analyses of 17 GCB-DLBCL tumors in our cohort, four of the loci (EYA4, NEFL, RAB32 and ZMYND10) with common de novo DNA methylation in GCB-DLBCL tumors showed both a deletion and DNA methylation (β>0.4) in one or two GCB-DLBCL tumors (Supplemental Table 7). Intriguingly, loci with de novo DNA methylation proximal to or within ADAMTS12, ASCL2, CAV1, CCNA1, CDH3, COL1A2, EPO, FN1, HRASLS, IGSF4 (aka CADM1), IMPACT, LRP2, MT1A, MYCN, NEFL, NPY, PAX6, POMC, and PTPRO showed DNA methylation (β>0.4) and an increase in gene copy number in one or two of the GCB-DLBCL tumors previously profiled by array-CGH (Supplemental Table 7). However, based on this data, we cannot determine if de novo DNA methylation occurred prior to amplification and was maintained or if de novo DNA methylation occurred independently at all loci after amplification. The former hypothesis provides a more parsimonious explanation since it requires the fewest de novo epigenetic changes in each tumor.
As in our above analysis of de novo DNA methylation in GCB-DLBCL tumors, we screened for loci that were methylated (β>0.4) in ≥20% of ABC-DLBCL tumors and hypomethylated (β<0.2) in >95% of all normal B-cells (N=24, including all activated and GC B-cells). Overall, we found 67 loci meeting the de novo DNA methylation criteria either proximal to or within 32 unique genes (ABO, ADAMTS12, APBA1, APP, ASCL2, CCNA1, COL1A2, CSF1, CSPG2 (aka VCAN), CYP1A1, EPHA3, EPO, EYA4, HRASLS, IGFBP7, IGSF4 (aka CADM1), IMPACT, LOX, LRP2, MT1A, MYCN, MYOD1, NEFL, NPY, PAX6, PROK2, PTPRO, RAB32, RARB, RASGRF1, RET, and SEZ6L) (Supplemental Table 8). A total of 64 loci proximal to or within 32 unique genes showed de novo DNA methylation in both GCB-DLBCL and ABC-DLBCL. All 32 genes proximal to or encompassing loci showing common de novo DNA methylation in ABC-DLBCL also showed common de novo DNA methylation in GCB-DLBCL.
There is literature precedent for DNA methylation occurring proximal to a subset of the above-stated genes in lymphoma. For example, de novo methylation of loci proximal to or within CCNA1, EYA4, IGFBP7, LOX, MT1A, MYOD1, NEFL, PROK2, and RAB32 was also observed in mature aggressive B-cell NHLs (maB-NHLs), including molecular Burkitt's lymphoma and DLBCL [Martin-Subero et al., 2009]. These loci comprised part of the evidence for the proposal that de novo methylation in maB-NHLs are enriched for polycomb repressive complex (PRC) targets and suggestion that maB-NHLs originate from cells with stem cell features or stemness acquired during lymphomagenesis by epigenetic remodeling [Martin-Subero et al., 2009]. We also note agreement with our prior observation of frequent DNA methylation of a locus proximal to MYOD1 in both ABC-DLBCL and GCB-DLBCL, based on Methylight assays [Pike et al., 2008].
Furthermore, we previously proposed that epigenetic processes affect IGSF4 expression in DLBCL due to the observed proportional reductions of IGSF4 expression with increasing DNA methylation levels of a nearby locus [Pike et al., 2008]. DNA methylation of loci proximal to IGSF4 in DLBCL was also previously reported by others [Martin-Subero et al., 2009]. Thus, we propose that further functional analyses into the epigenetic regulation of IGFS4 expression in DLBCL are warranted.
Intriguingly, loci proximal to ADAMTS12, ASCL2, CCNA1, EYA4, IGFS4 (aka CADM1), IMPACT, LOX, MYOD1, NEFL, NPY, PEX6, and RAB32 have been reported to show robust increases in DNA methylation in follicular lymphoma (FL) relative to normal follicular hyperplasia (FH) [Killian et al., 2009]. We propose that the shared common de novo methylation events in DLBCL and FL are related to their B-cell origins. In fact FL commonly transforms into DLBCL (t-FL) [Davies et al., 2007]. However, it should be noted that t-FL has a gene expression phenotype similar to GCB-DLBCL, but not ABC-DLBCL [Davies et al., 2007]. This makes the interpretation of the epigenetic profiles amongst these diseases challenging.
Based on array-CGH analyses of seven ABC-DLBCL tumors in our cohort, eight of the loci (ABO, CSF1, EYA4, IGFBP7, LRP2, MYCN, RAB32, and RASGRF1) that showed de novo DNA methylation in ABC-DLBCL tumors also showed both a deletion and DNA methylation (β>0.4) in one or more ABC-DLBCL tumors (Supplemental Table 8). Intriguingly, loci with de novo DNA methylation proximal to or within ADAMTS12, COL1A2, EPHA3, HRASLS, IMPACT, NPY, PROK2, PTPRO, RARB, and RET also showed DNA methylation (β>0.4) and an increase in gene copy number in one or two of the ABC-DLBCL tumors previously profiled by array CGH (Supplemental Table 8).
Although ABC-DLBCL and GCB-DLBCL tumors could not be distinguished by our hierarchical clustering analysis (Fig. 1), we uncovered a subset of loci showing robust differences in DNA methylation levels across these two DLBCL subtypes (Table 1). We found 19 loci showing differential DNA methylation (Benjamini and Hochberg corrected Wilcoxon P<0.05 and >0.4 units difference in geometric means of ABC-DLBCL and GCB-DLBCL β-values). These loci are proximal to or within 16 genes (ABCC5, APBA2, CAPG, CD44, DAD1, DMP1, EPM2A, GP1BB, IL13, IRF5, LTB4R, RUNX3, SEZ6L, SMARCA3 (aka HLTF), TMEFF1, and TSP50). Based on BeadArray probe identifiers, none of these loci showed common de novo DNA methylation for either ABC-DLBCL or GCB-DLBCL.
Although the BeadArray DNA methylation platform has been extensively characterized [Bibikova and Fan, 2009; Bibikova et al., 2006; Killian et al., 2009; Martin-Subero et al., 2009], we sought to evaluate independently its performance in our study. We performed bisulfite sequencing analysis of nine loci with differential DNA methylation between ABC-DLBCL and GCB-DLBCL or de novo DNA methylation in tumors (Table 2 and Supplemental Fig. 2). These included CpG islands proximal to or within CCNA1, DBC1, IGFBP1, IRF5, MYOD1, RAB32, RUNX3, and TSP50 (Table 2). Overall, the results from the bisulfite sequencing and BeadArray analyses were in excellent agreement (Table 2 and Supplemental Fig. 2). Furthermore, they demonstrate that the DNA methylation status of CpG dinucleotide(s) interrogated by the BeadArray assays largely reflects the overall DNA methylation status of flanking CpG sites.
Next, we sought to determine the relationships between common de novo DNA methylation in GCB-DLBCL and the expression levels of proximal or encompassing genes in normal GC B-cells. Through Affymetrix GeneChip analyses, we were able to obtain expression data for 39 of the 42 genes proximal to or containing loci showing common de novo DNA methylation in GCB-DLBCL. We found that >87% (34/39) of the de novo DNA methylation in GCB-DLBCL occurred proximal to or within genes that were poorly expressed in normal GC B-cells (i.e. below the upper 50th percentile of gene expression scores – see Methods).
In fact, the five exceptions (APBA1, EPO, IGFBP7, IGSF4 (aka CADM1), and RAB32) showed only modest expression in normal GC B-cells. None of the gene expression scores ranked within the upper 25th percentile for any normal GC B-cell sample (Additional File 9). However, we have previously noted that IGSF4 gene expression levels in DLBCL tumors were inversely proportional to the DNA methylation levels of a proximal CpG island [Pike et al., 2008]. IGSF4 is a tumor suppressor gene that encodes an immunoglobulin superfamily member protein important for cell adhesion and motility. It is inactivated in multiple cancers including non-small cell lung cancer (NSCLC) [Murakami, 2005] and nasal NK/T-cell lymphoma [Fu et al., 2009]. To our knowledge, APBA1, EPO, IGFBP7, or RAB32 methylation has not yet been highlighted in lymphoma. However, the DNA methylation status of a locus proximal to IGFBP7 is inversely associated with IGFBP7 gene expression in cultured human colon cancer cell lines [Lin et al., 2007]. Likewise, it has been reported that RAB32 expression is down-regulated in response to DNA methylation in a group of primary colon cancer and colon cancer cell lines [Mori et al., 2004].
The combined analysis of DNA methylation profiles in DLBCL and normal B-cell populations provides an opportunity to discriminate de novo DNA methylation arising during tumorigenesis from DNA methylation occurring in normal B-cells. A subset of de novo DNA methylation events could be of functional significance in the development and/or progression of disease. However, the identification of functional de novo DNA methylation will be challenging, even when both tumor and normal progenitor cell DNAs can be obtained from the same individual to help control for age, gender, and/or environmental exposures [Foley et al., 2009; Fraga et al., 2005; Woo et al., 2009].
In our attempts to identify epigenetic changes of functional significance, we observed that de novo DNA methylation occurred most frequently in CpG islands proximal to or within genes that are already poorly expressed in normal B-cell populations. This supports a model in which de novo DNA methylation is most frequently involved in the maintenance, as opposed to the initiation, of gene silencing [Weber and Schubeler, 2007]. Overall, our data is consistent with similar observations of pervasive epigenetic remodeling in FL, wherein genes that are poorly expressed in normal B-cells were proximal to loci that were methylated in tumors [Killian et al., 2009]. Analyses in aggressive B-cell non-Hodgkin lymphomas have also demonstrated that genes proximal to or encompassing de novo methylated loci are expressed at low levels in lymphomas and normal hematopoietic tissues [Martin-Subero et al., 2009].
Furthermore, there is increasing evidence that the above observations can be generalized to other cancers. For example, genes proximal to or containing methylated loci in colon tumor samples have been shown to be expressed at low levels in normal colon as well as in colorectal adenocarcinomas [Keshet et al., 2006]. Functional studies in cultured human colon cancer [Bachman et al., 2003] and prostate cancer cells [Gal-Yam et al., 2008] and mouse erythroleukemia (MEL) cells [Feng et al., 2006] reached a similar conclusion that de novo DNA methylation locks in rather than initiates gene silencing (de novo repression). Gel-Yam and colleagues provide compelling evidence of frequent epigenetic switching in PC3 prostate cancer cell cultures by which DNA methylation replaces Polycomb repressive complex (PRC) marks found in normal prostate epithelial cells [Gal-Yam et al., 2008]. Given the overlap in PRC targets and the de novo methylation found in our studies and in a different survey of maB-NHLs, we agree with the speculation that epigenetic switching can frequently occur in DLBCL [Martin-Subero et al., 2009]. However, rigorous functional studies in primary tumor cells are needed to validate these speculations.
Nevertheless, it has been recently reported that the methylation status of CpG island shore sequences, located on the edges of CpG islands, is strongly associated with the transcription of associated genes (within 2-kb) in normal tissues and colon cancer [Irizarry et al., 2009]. The BeadArray assays used in our study are not optimal for addressing the methylation status of CpG island shore sequences. Thus, large-scale, comprehensive analyses of DNA methylation and gene expression profiles are needed to elucidate the likely complex nature of their relationships. Although this will require further technology development, continued advances in sequencing technologies could make this more feasible in the future.
We note that there are multiple scenarios by which maintaining the silence of an otherwise normal gene (i.e. one not carrying genetic mutations that compromise its activity) would have functional significance in tumorigenesis. For example, the epigenetic silencing of genes whose primary function is to prevent the division or promote the death of cells with genetic instability could have a profound impact in cancer development. In principle, such genes could be poorly expressed in normal cells under non-stressful physiological conditions.
In our attempts to identify epigenetic changes of functional significance, we observed a limited numbers of cases in which de novo DNA methylation was associated with a chromosomal deletion. For tumor suppressor genes, the association of gene deletions and epigenetic silencing would be consistent with Knudson's two-hit hypothesis of cancer [Knudson, 1971]. It is possible that some loci showing de novo DNA methylation are proximal to functionally significant genes that are haploinsufficient or have inactivating point mutations or deletions not detected by array CGH analyses in their second copy.
Even taking into consideration the caveats discussed above, the relationships among de novo DNA methylation, gene expression, and locus copy number call into question the functional significance of de novo DNA methylation uncovered in this study and others [Martin-Subero et al., 2009]. This would be in keeping with genetic studies suggesting that `driver' mutations that are causally involved in cancer development and progression can be significantly less numerous than `passenger' mutations that have little to no functional significance [Greenman et al., 2007; Sjoblom et al., 2006]. The discrimination between epigenetic `driver' and `passenger' changes in tumors will require functional analyses. Although challenging, the development of novel technologies to introduce specific epigenetic changes into cancer and normal cells would provide a powerful tool to facilitate such studies.
Regardless of their functional significance, de novo DNA methylation could represent epigenetic biomarkers that are useful for monitoring residual disease levels in body fluids, especially during remission periods. The decreasing costs of sequencing technologies bode well for the development of personalized genetic and epigenetic biomarkers of this type based on data acquired from individual tumors. We believe the combined analysis and application of genetic and epigenetic data will lead to a more sophisticated understanding of the molecular etiology of cancer and the development of more sensitive and specific clinical biomarkers.
Grant Information : Contract grant sponsor: NIGMS, NIH; Contract grant number: GM072447
Contract grant sponsor: NHGRI, NIH; Contract grant number: HG002790
Contract grant sponsor: NCI, NIH; Contract grant number: CA36727
Contract grant sponsor: NCI, NIH; Contract grant number: CA84967
Contract grant sponsor: NCI, NIH; Contract grant number: CA114778
We thank Lou Staudt at the National Cancer Institute for thoughtful discussion and access to Affymetrix gene expression data used in tumor classification. We also thank Peter Jones and Daniel Weisenberger at the University of Southern California for thoughtful discussions and valuable feedback. This study was funded by grant GM072447 awarded by the National Institute of General Medical Sciences, National Institutes of Health, grant P50-HG002790 awarded by the National Human Genome Research Institute, National Institutes of Health, and grants CA36727, CA84967, and CA 114778 awarded by the National Cancer Institute, National Institutes of Health. TCG is a grantee of the Mantle Cell Lymphoma Research Program of the Lymphoma Research Foundation. This investigation was conducted in a facility constructed with support from Research Facilities Improvement Program Grant Number C06 (RR10600-01, CA62528-01, RR14514-01) from the National Center for Research Resources, National Institutes of Health.