Abstract

Introduction

Increases in methylation of certain DNA sequences and decreases in others are very frequent in human cancers. However, the relationship between cancer-associated hypermethylation of some parts of the genome and hypomethylation of others is still unresolved. In human cancers, hypermethylation is often observed at some 5’ gene or promoter regions that are largely unmethylated in normal somatic tissues (reviewed in Jones & Baylin, 2002). For many of these genes, this hypermethylation has been linked to transcriptional silencing. In addition to these increases in DNA methylation, an overall decrease in the 5-methylcytosine content of the genome relative to various normal postnatal somatic tissues (global DNA hypomethylation) has been described in disparate types of cancers, indicating cancer-specific, and not tissue-specific differences (Gama-Sosa et al., 1983; Ehrlich, 2002). Some cancer-linked global DNA hypomethylation can be attributed to a decrease in satellite DNA methylation, as we have demonstrated for breast cancers, Wilms tumors, and ovarian epithelial carcinomas (Narayan et al., 1998; Jackson et al., 2004; Qu et al., 1999a). Single-copy sequence hypomethylation in tumors is also observed (Feinberg & Vogelstein, 1983; Feinberg et al., 2002; Grunau et al., 2005; Okada et al., 2005). In numerous studies in which both DNA hypomethylation and local DNA hypermethylation (usually CpG islands in 5’ gene or promoter regions) were investigated, these opposite epigenetic changes have been observed in the same tumors (Ehrlich et al., 2002; Narayan et al., 1998; Santourlidis et al., 1999). However, it is not clear whether these two sorts of cancer-associated epigenetic events share a mechanistic basis or etiology.

There might be cross-talk between demethylation and de novo methylation pathways during tumorigenesis, which could make one dependent on the other. For example, CpG island demethylation, which can occur in non-embryonal human cells (Qu & Ehrlich, 1999), might be used as a type of epigenetic repair in normal postnatal cells to attempt to compensate for physiologically inappropriate methylation of CpG islands overlapping promoters of tumor suppressor genes. As is true for conventional DNA repair pathways, this epigenetic repair pathway might be only partially effective and might demethylate many more sequences than just the incorrectly methylated CpG islands. Also, there is evidence that DNA hypermethylation might sometimes be a consequence of hypomethylation, as observed in the hypermethylation that followed hypomethylation of the p53 gene in rats fed a folate/methyl-deficient diet (Pogribny et al., 1997). Studies of azaCR-treated cultured mammalian cells and DNMT antisense-expressing Arabidopsis transgenic lines in which gene hypermethylation was observed against a background of global DNA hypomethylation are also consistent with this hypothesis (Broday et al., 1999; Jacobsen & Meyerowitz, 1997). Lastly, we have recently shown that a DNA repeat can be subject to cancer-associated demethylation and de novo methylation even within the same region on the same DNA molecules (Nishiyama et al., 2005 and unpub. data). In this study, we investigated whether 5’ gene/promoter hypermethylation and global or satellite DNA hypomethylation are linked in ovarian epithelial tumors. For this analysis, we compared ovarian epithelial tumors of varying degrees of malignancy, namely, carcinomas, low malignant potential (LMP, borderline malignant) tumors, and cystadenomas (benign tumors) as well as various control tissues.

We analyzed the relationships of cancer-linked DNA hypermethylation and hypomethylation using 55 different gene loci, three classes of repetitive elements, and global DNA hypomethylation measurements. With all these variables we could potentially investigate hundreds of separate associations. Large numbers of statistical tests increase the likelihood of obtaining P values below 0.05 merely by chance. To correct for this problem of multiple comparisons, we set the false discovery rate for each statistical test at 5% (Benjamini et al., 2001).

We also tested whether there is an association of hypermethylation or hypomethylation of the above-mentioned DNA sequences with RNA levels for ten different DNMT isoforms because cancer-linked hypomethylation or hypermethylation might be explained by differences in expression of different types of DNA methyltransferases (DNMTs). Several studies have reported an increase in the mRNA levels for DNMTs in cancer cells (el-Deiry et al., 1991; Robertson et al., 1999; Robertson, 2001), although this is controversial (Kimura et al., 2003; Lee et al., 1996; Eads et al., 1999). Our study included DNMT3B4 RNA, which encodes a presumably non-catalytic protein and was reported to be significantly overexpressed in hepatocellular carcinomas exhibiting satellite 2 (Sat2) DNA hypomethylation (Saito et al., 2002).

Results

Hypomethylation of satellite DNA and global DNA hypomethylation in ovarian tumors

We previously showed extensive satellite DNA hypomethylation in several types of human cancers (Narayan et al., 1998; Qu et al., 1999a; Qu et al., 1999b). In order to compare gene-region hypermethylation with satellite DNA hypomethylation in the ovarian tumors, we next analyzed methylation at satellite α throughout the centromeres (all Satα), Satα in chromosome 1 (Chr1 Satα), and satellite 2 in the largest juxtacentromeric heterochromatin region (Chr1 Sat2 at 1qh) using digests made with the CpG methylation-sensitive BstBI. Southern blotting revealed that carcinomas and LMP tumors were frequently hypomethylated in these satellite DNAs relative to all tested normal postnatal somatic tissues (Figs. 1 and ​and2).2). The percentages of tumors displaying strong hypomethylation (score 3 or 4 on a scale of 0 – 4) in carcinomas, LMP tumors, and cystadenomas were 42, 5, and 0, for Chr1 Sat2; 32, 5, and 0, for Chr1 Satα; and 21, 5, and 0, for all Satα, respectively. Differences in the extent of satellite DNA hypomethylation or global DNA hypomethylation vs. the malignant potential were highly significant after adjustment for multiple comparisons (Fig. 2). These results were similar to those seen for Chr1 Sat2 in a much smaller study of ovarian epithelial tumors (Qu & Ehrlich, 1999). Some carcinomas had hypomethylation levels as extreme as in sperm (hypomethylation score, 4). Among all the ovarian tumors, Chr1 Sat2, Chr1 Satα, all Satα hypomethylation, and global DNA hypomethylation were significantly associated with one another in each pair-wise combination (P<0.005 except P=0.009 for global DNA hypomethylation vs. all Satα hypomethylation). In a comparison of cystadenomas vs. control somatic tissues, Chr1 Satα and all Satα, but not Chr1 Sat2, displayed significant, but mostly weak (Fig. 1), hypomethylation (P<0.001). This may reflect atypical tissue-specific differences or very early epigenetic changes during tumorigenesis. Therefore, Sat2 hypomethylation appears to be a more specific cancer marker than Satα hypomethylation, as in our recent study of breast cancers (Jackson et al., 2004).

Open in a separate window

Fig. 1

Southern blot analysis of satellite DNA hypomethylation in ovarian epithelial tumors

Examples of satellite DNA hypomethylation in BstBI digests of ovarian carcinomas in a single blot hybridized three times. Normal postnatal somatic DNAs and sperm DNA are the hypermethylated and hypomethylated standards, respectively. (A) juxtacentromeric Chr1 Sat2 probe and (B) centromeric Chr1 Satα probe under high-stringency hybridization conditions; (C), centromeric Chr1 Satα probe under low-stringency conditions that allow hybridization to DNA from most of the centromeres. Scoring of hypomethylation was by phosphorimager quantitation of <4-kb vs. >4-kb signal and visual comparison of bands b, c, d, e, and f to region a.

Open in a separate window

Fig. 2

Comparison of satellite DNA hypomethylation and global DNA hypomethylation in ovarian epithelial tumors of different malignant potential (A)

The levels of satellite DNA hypomethylation (score 0–4) and (B) the percent of total DNA cytosine residues that were methylated are plotted for individual tumors. Not shown are three LMP tumors and six cystadenomas analyzed only for satellite DNA methylation. The horizontal line is the global DNA hypomethylation threshold, i.e., the minimum percentage of methylated cytosine residues found in diverse normal postnatal somatic tissues (3.43; range for normal tissue types, 3.43–4.1). As a conservative estimate of an overall deficiency of m⁵C in tumor DNAs, global methylation levels of <3.43 were considered hypomethylated. (C) The percentage of tumors displaying global DNA hypomethylation or strong-to-moderate satellite DNA hypomethylation (score 2, 3, or 4) relative to all the somatic controls is shown. P-values for hypomethylation scores vs. degree of malignancy are given. For all of these DNA hypomethylation parameters, except the all-Satα hypomethylation, there was a significant association with the degree of malignancy after adjustment for multiple comparisons.

Discussion

We rigorously analyzed the association of malignancy-linked DNA hypermethylation and hypomethylation using quantitative measures of hypermethylation at gene loci (MethyLight assay), global DNA hypomethylation (HPLC) and semi-quantitative data on satellite DNA hypomethylation (Southern blot/phosphoimager analysis) on a collection of ovarian epithelial tumor samples enriched for neoplastic cells. On the 60 studied tumors, 60 MethyLight reactions covering 55 gene loci were performed which resulted in approximately 3600 total methylation data points With adjustment for multiple comparisons, we first identified six gene regions (LTB4R, MTHFR, PGR, CDH13, CDH1, and IGSF4) whose hypermethylation was significantly associated with the degree of malignancy. Only CDH1 and CDH13 were previously shown to be abnormally methylated in ovarian cancers (Kawakami et al., 1999; Rathi et al., 2002; Widschwendter et al., 2004). In addition, both genes are hypermethylated in many other types of human cancers including lung and breast (Zochbauer-Muller et al., 2001), prostate (Isaacs et al., 1995), colon (Hibi et al., 2004) and thyroid (Graff et al., 1998). PGR is hypermethylated in breast (Lapidus et al., 1996) and endometrial cancers (Sasaki et al., 2003). Hypermethylation of IGSF4 was described in cervical (Li et al., 2005) and pancreatic (Jansen et al., 2002) cancers. MTHFR had not been previously shown to be hypermethylated in ovarian cancers, but it had been linked to ovarian carcinogenesis by loss of heterozygosity (Viel et al., 1997).

After demonstrating that satellite DNA hypomethylation and global DNA hypomethylation were also significantly related to the degree of malignancy in these tumors, we looked at interrelationships between hypermethylation and hypomethylation. A multiple regression analysis indicated that these types of hypomethylation and hypermethylation are not interdependent so that neither is likely to be the result or cause of the other. Instead, they share an occasional inverse correlation, as described below. Nonetheless, both hypermethylation and hypomethylation of DNA are linked to carcinogenesis and so may have some step(s) in common in the poorly understood pathways that lead to these divergent epigenetic changes (Nishiyama et al., 2005).

Our findings on the general independence of DNA hypomethylation and hypermethylation in different parts of the genome in ovarian cancer are consistent with the following less extensive analyses: LINE1 hypomethylation vs. GST1 hypermethylation in prostate carcinomas (Santourlidis et al., 1999); satellite and global DNA hypomethylation vs. gene hypermethylation in Wilms tumors (Ehrlich et al., 2002); cancer/testes antigen gene hypomethylation vs. gene hypermethylation in gastric cancers (Kaneda et al., 2004); and in vitro DNMT acceptor activity vs. gene hypermethylation in colon neoplasms (Bariol et al., 2003). In contrast to the separate relationship of the studied DNA hypermethylation and DNA hypomethylation to malignant potential in ovarian tumors, we found that all analyzed categories of satellite DNA hypomethylation were significantly related to each other and to global DNA hypomethylation. Therefore, global hypomethylation, which is frequent in diverse types of cancer (Ehrlich, 2002), might promote tumor formation or progression by encompassing satellite DNA or DNA repeats in general.

We observed one association of satellite DNA hypomethylation with gene hypermethylation in the carcinoma-only group but it was an inverse one. CDH13 was less likely to be hypermethylated (P=0.01) when there was extensive loss of Sat2 methylation. Similarly, we previously found a significant inverse association between CDH13 hypomethylation and Sat2 hypomethylation in a different, larger group of ovarian cancers (96 tumors) and in breast cancers (43 tumors; Widschwendter et al., 2004). These findings might reflect waves of DNA demethylation during carcinogenesis that could affect some CpG-rich regions and reverse their de novo methylation (Cheng et al., 2001; Taniguchi et al., 2003).

Despite many investigations, it is still unclear to what extent altered DNMT levels are involved in cancer-linked hypermethylation or hypomethylation. Although no single change in DNMT levels could explain both cancer-linked increases of DNA methylation in some portions of the genome and decreases in others, differential changes in specific DNMT isoforms might be involved if they have sufficient regional specificity. At least some compartmentalization of human DNMT specificities can be inferred from the ICF syndrome in which a major deficiency in DNMT3B activity (Gowher & Jeltsch, 2002; Hansen et al., 1999; Okano et al., 1999) cannot be compensated by the other DNMTs and results in selective hypomethylation of certain portions of the genome, notably long regions with tandem repeats, including Sat2 (Ehrlich, 2003; Jeanpierre et al., 1993; Kondo et al., 2000). By comparing DNMT RNA levels and differential DNA methylation in cancers of the same type, we avoided the problems implicit in normal tissue-to-cancer comparisons for such cell cycle-dependent RNAs. The most notable associations between RNA levels for ten DNMT isoforms with gene hypermethylation in cancer were those between DNMT1 RNA levels and LTB4R hypermethylation (P=0.0005) or MTHFR hypermethylation (P=0.002). LTB4R and MTHFR are the only two loci with appreciable DNA methylation even in the normal tissue controls as well as significant hypermethylation related to the degree of malignancy. Therefore, increased DNMT1 activity might be involved in malignancy-related increases in methylation of genes that have appreciable methylation in normal tissues. Caveats in this analysis are that the transient changes in DNMT RNA levels would be missed and that the analysis was only at the level of RNA.

A recent study showed that DNMT1 protein levels were elevated in human breast cancer cells due to increased stability, while the DNMT1 mRNA levels were unchanged (Agoston et al., 2005). However, quantitative analysis of individual protein splice variants, such as those of DNMT3B, is difficult due to poor specificity and sensitivities in recognizing these isoforms, which are expressed at very low levels. Moreover, the detection of DNMT proteins in human cancers is insufficient to provide information regarding the catalytic activity of each isoform. We utilized highly sensitive, quantitative, real-time RT-PCR assays which are specific for each DNMT mRNA isoform.

No DNA hypomethylation variables were significantly associated with DNMT RNA levels after adjustment for multiple comparisons. From a study of hepatocellular carcinomas, it was proposed that Sat2 hypomethylation in cancer is due to a reported overproduction of an alternatively spliced form of DNMT3B RNA (DNMT3B4) that is missing part of the sequence encoding the catalytic domain (Saito et al., 2002). Because of the above-mentioned Sat2 hypomethylation when DNMT3B is mutationally inactivated in the ICF syndrome, it was hypothesized that there is an excess of DNMT3B4 protein in cancers that inhibits catalytically active variants of DNMT3B. However, in the ovarian carcinomas and LMP tumors, we saw no correlation between cancer-associated Sat2 hypomethylation and increased levels of DNMT3B4 RNA.

Our data are consistent with the hypothesis that the very frequent satellite DNA hypomethylation and global DNA hypomethylation in cancer favor carcinogenesis and are not just an epiphenomenon linked to gene hypermethylation and silencing. Cancer-linked DNA hypomethylation may have an important role in increasing cancer-predisposing chromosome rearrangements (Kokalj-Vokac et al., 1993; Hernandez et al., 1997; Eden et al., 2003), although the frequency of this hypomethylation in satellite DNA is much greater than that of satellite DNA rearrangements and is not related to aneuploidy (Ehrlich et al., 2003). Satellite DNA hypomethylation might have an indirect effect on gene expression that favors tumorigenesis, e.g., by influencing the sequestration of DNA-binding proteins in highly repeated DNA regions. Alternatively, cancer-associated DNA hypomethylation might greatly overshoot critical promoter targets but often include transcription regulatory regions, despite the finding that promoter hypomethylation in cancer appears to be much less frequent than promoter hypermethylation (Ehrlich, 2002). Our recent demonstration that satellite DNA hypomethylation is a superior and independent marker of poor prognosis in ovarian cancer patients undergoing optimal cytoreductive surgery (Widschwendter et al., 2004) highlights the importance of more studies on cancer-linked DNA hypomethylation to elucidate why hypomethylation, as well as hypermethylation, is so frequent in cancer and to carefully consider diagnostic, prognostic, and therapeutic applications of cancer epigenetics.

Materials and methods

Acknowledgments

References