Decreased Fidelity in Replicating DNA Methylation Patterns in Cancer Cells Leads to Dense Methylation of a CpG Island

N. Watanabe ■ E. Okochi-Takada ■ Y. Yagi ■ J.-I. Furuta ■ T. Ushijima (K)

Carcinogenesis Division, National Cancer Center Research Institute, 5-1-1 Tsukiji, Chuo-ku, 104-0045 Tokyo, Japan [email protected]

1 Introduction 200

2 Fidelity in Normal Mammary Epithelial Cells 201

3 Decreased Fidelity in Gastric Cancer Cells 204

4 Decreased Fidelity and Induction of Dense Methylation 204

5 Molecular Basis for CIMP, and Variation of the Fidelity Among CGIs ... 207

6 Epilogue 208

References 209

Abstract Cancer cells that have a large number of aberrantly methylated CpG islands (CGIs) are known to have CpG island methylator phenotype (CIMP), and decreased fidelity in replicating methylation patters has been analyzed as an underlying mechanism. First we developed a method to analyze the number of errors in replicating CpG methylation patterns in a defined period. A single cell was expanded into 106 cells, and the number of errors during the culture was measured by counting the deviation from the original methylation patterns. It was shown that methylated status of a CpG site was more stably inherited than unmethylated status, suggesting that the genome is constantly exposed to de novo methylation. Promoter CGIs showed higher fidelities than CGIs outside promoter regions. We then analyzed error rates in two gastric cancer cell lines without CIMP and two with CIMP for five promoter CGIs. Two CIMP(-) cell lines showed error rates smaller than 1.0X10-3 errors per site per generation (99.90%-100% fidelity) for all the five CGIs. In contrast, AGS cells showed significantly elevated error rates, mainly due to increased de novo methylation, in three CGIs (1.6- to 3.2-fold), and KATOIII cells showed a significantly elevated error rate in one CGI (2.2-fold). Presence of densely methylated DNA molecules was observed only in KATOIII and AGS. These data demonstrated that some cancer cells have decreased fidelity in replicating CpG methylation patterns that underlie CIMP.

Introduction

DNA methylation is inherited upon cell division, and methylation of CpG islands (CGIs) in gene promoter regions is known to suppress the expression of the downstream genes (Jones and Baylin 2002; Bird 2002). Aberrant methylation of promoter CGIs of tumor-suppressor genes is known to be deeply involved in carcinogenesis. At the same time, it has recently been recognized that various CGIs, not only those of tumor-suppressor genes, are methylated in cancer cells (Costello et al. 2000; Sato et al. 2003; Ushijima 2005). Some cancers are known to have methylation of multiple CGIs, and this phenotype was designated as CGI methylator phenotype (CIMP) (Toyota et al. 1999; Issa 2004). When a cancer has a CIMP, it has been proposed that a number of important genes are inactivated due to methylation of promoter CGIs, and that this will have a significant impact on the behavior of cancers with CIMP. In fact, our recent study demonstrated that neuroblastomas with CIMP are associated with a significantly and markedly lower probability of survival (Abe et al. 2005).

On the other hand, Yamashita et al.could not find the presence of a distinct phenotype with methylation of multiple CGIs, based on their methylation analysis of seven CGIs of known tumor-related genes and 30 Not I sites randomly selected from thegenome (Yamashita etal. 2003). This raised several issues that we have to consider when analyzing the presence of CIMP. First, as described in the first report by Toyota et al. (1999), selection of appropriate CGIs is important. Methylation of appropriate CGIs should not cause selection of cells with their methylation, because selection can cause an apparent increase of cells with methylation. Also, appropriate CGIs should not be methylated in non-cancerous tissues, since CIMP refers to abnormal cellular capacity to induce methylation of CGIs, and not to age-dependent methylation (Issa et al. 1994,2001). Second, analysis of appropriate regions within a CGI is necessary. Methylation statuses within a CGI are nothomogeneous (Ushijima 2005). A totally different methylation profile can be obtained when a core region within a CGI is analyzed and when non-core regions within a CGI are analyzed. Third, and most importantly, analysis on the dynamic speed of occurrence (rate) of methylation of CGIs or CpG sites in a defined period of time is necessary (Ushijima and Okochi-Takada 2005). Most studies so far have analyzed the number of aberrantly methylated CGIs, which is dependent upon multiple factors, including the rate of occurrence of methylation, the number of past events of clonal selection, and the number of methylated CGIs in the precursor cell.

To analyze possibly increased rates of occurrence of methylation in cancer cells, information on the rate in normal cells is indispensable. However, analysis of the rates of methylation errors has been limited. The methylated status of an exogenously introduced DNA was maintained with fidelity of 94% per generation per site by Southern blot analysis (Wigler et al. 1981). Pfeifer et al. developed the ligation-mediated PCR (LMPCR) method, and analyzed the efficiency of maintenance methylation (Em) and that of de novo methylation (Ed) separately, using CpG sites within a CGI in the 5' region of the PGK1 gene on the inactive X chromosome (Pfeifer et al. 1990a, b). They observed an Em of 98.8%-99.9% per site per generation and Ed of 5%. Ed of 5% corresponds to a fidelity of 95% in keeping the unmethylated status of a CpG site.

Recent advancements in bisulfite sequencing methods have enabled researchers to analyze methylation status at the nucleotide level (Clark et al. 1994). Taking advantage of bisulfite sequencing, we decided to observe a large number of CpG sites within a CGI, and measure the fidelity in maintaining their methylated or unmethylated status, and then to analyze changes in the fidelity in cancer cells.

Fidelity in Normal Mammary Epithelial Cells

Before analyzing fidelity in cancer cells, we had to establish a system in which we could measure the number of errors in replicating methylated or unmethylated statuses of individual CpG sites in a defined number of cell divisions (Ushijima et al. 2003). For this purpose, we seeded a single human mammary epithelial cell (HMEC) in a well of a 96-well plate, and expanded it up to 106 cells (Fig. 1A). From the actual count of the number of cells at harvest and the number of cells lost during two transfers, we calculated the actual number of cell divisions during the culture. Using DNA extracted from the final 106 cells, methylation statuses of individual CpG sites were examined. To exclude artifacts due to insufficient bisulfite treatment, unconversion rates were measured using unmethylated control DNA, and were confirmed to be small enough compared with error rates.

Methylation patterns of the differentially methylated region (DMR) of H19 was initially examined, since distinction of maternal and paternal alleles was possible by a polymorphism and also by the overall methylation statuses (Fig. 1B). All of the unmethylated DNA molecules (molecules 1-9 in Fig. 1) had similar methylation patterns and the T polymorphism, while all the methylated DNA molecules (molecules 10-12 in Fig. 1) had similar methylation patterns and the G polymorphism. This showed that the number of errors in replicating methylation patterns was not too large during the expansion from 1 to 106 cells, that the methylation patterns of the two alleles in the original single cell can be inferred (molecules 1-7 for the unmethylated allele, and

A 1 cell

generations

extraction

Bisulfite sequencing

B DMR of H19

hiii ni.iiMi.Uyi. i m jiiiiMiii.1 a i ooooooooooooooooo«ooooo«otoo o 2<xooooooooooooooo«ooooo«otoo o

3 CXXXXXXXXXXXXXXXX*OCXXX3«OTOO o

4CCCOOOCCCOOOCCCCO«OOCCC«OTCO o

5 oooooooodoooocxxo«ooooo«otoo o

6CCCOOOCCCCOOCCCCC«OOCCC«OTCO o

7ooooooooooooooooo«ooooo«otoo o

8 CCOOOO«CCOOOOCXXXXXX3CCC«OTOO 2

9 0®OOOO®OOOOOOO«DOOOOOOO«DTOO 4 10MSMMOMMMMMMMMOGO9 o

II MSMMOMMMMMMMMOGO* o I20NNN0IMMNNNNNNGCI 2

Total number of errors = 8

Error rate = 8/12/27 = OQ25 Fig. 1A, B The method to measure the fidelity in replicating methylation patterns. a A single cell was expanded to 106 cells, and methylation patterns in the final cell population were analyzed. b Example of an analysis. Twelve DNA molecules were analyzed for methylation patterns of the H19 DMR, and deviation from the inferred original methylation patterns was calculated (the numbers of errors are shown to the right of clones). Open and closed circles: unmethylated and methylated CpG sites, respectively. T or G is a reported polymorphism. Based on the total number of CpG sites analyzed and the observed number of errors, the error rate in the defined period was measured. Six independent cultures were analyzed for each region molecules 10 and 11 for the methylated allele), and that the number of errors in replicating methylation patterns (shown in the right of each molecule) can be measured. To obtain an accurate number of errors, six independent cultures were analyzed, and the average number of errors was calculated. Possible errors due to erroneous selection of the original methylation patterns were examined by selecting different patterns as the original methylation patterns (permutation test), and we confirmed that selection errors do not cause significant changes in the error rates.

CGIs in promoter regions

CGIs outside

Non- CGI in CGIs promoter

CGIs in promoter regions

CGIs outside

Non- CGI in CGIs promoter

Un methylated

Methylated

Fig. 2 Error rates in various regions of the genome in normal human mammary epithelial cells. The numbers of errors per cell division per CpG site are shown. Unmethylated regions showed higher error rates than methylated regions. This was true even when the unmethylated allele (marked with *) and methylated allele (shown by **) of H19 were compared. CGIs in promoter regions showed lower error rates than CGIs outside promoter regions

The analysis was expanded to five CGIs in promoter regions, three CGIs outside promoter regions, CpG sites outside CGIs (non-CGIs), and a normally methylated CGI in a promoter region (Fig. 2). When unmethylated regions and methylated regions were compared, it was clear that error rates were higher in unmethylated regions. Even limited to DMR of H19, the unmethylated allele showed a higher rate of errors. This showed that keeping the unmethylated status of CpG sites is much more prone to errors than keeping the methylated status of CpG sites. This finding was reasonably explained using the assumption that the genome is constantly exposed to pressure of de novo methylation, which is in good accordance with a pioneering finding (Pfeifer et al. 1990b).

When CGIs in promoter regions and CGIs outside promoter regions were compared, the former had lower error rates. Since methylation of promoter CGIs leads to silencing of the downstream genes and is potentially harmful to a cell, it appeared that CGIs in promoter regions were protected from de novo methylation in a safer manner than CGIs outside.

The measurement system does not take account of errors in the very early stages of culture (founder errors), and cannot make clear distinction between a failure in maintaining methylated status and that in maintaining unmethy-lated status. However, the effect of founder errors was considered very small because the variation among six independent experiments was reasonably small. Since error rates in unmethylated regions and methylated regions were clearly different, distinction of the two types of errors is important, and the development of a new system that can distinguish them is necessary.

Decreased Fidelity in Gastric Cancer Cells

Since the system seemed to be working, we shifted to analysis of cancer cells (Ushijima et al. 2005). For this purpose, we chose two gastric cancer cell lines without CIMP (HSC39 and HSC57) and two with CIMP (KATOIII and AGS) (Kaneda et al. 2002). The fidelity was analyzed for five promoter CGIs of five genes: bA305P22.2.3 (bA305P), FLJ32130,ahomologofRIKEN2210016F16 (RIKEN2210016; currently C9orf64), E-cadherin, and cyclophilin A. Since cancer cells might have aneuploidy of the genes analyzed, the copy numbers were analyzed in all the four gastric cancer cell lines by fluorescence in situ hybridization (FISH) and Southern blot analysis. For each CGI, three times as many clones (or more), vs the number of alleles, were analyzed. As was the case in normal mammary epithelial cells, the six experiments were repeated. As a result, we sequenced 1,495 clones.

Gastric cancer cell lines without CIMP (HSC39 and HSC57) showed error rates smaller than 0.02 errors/CpG site per observed generation for all the five CGIs (Fig. 3). This corresponded to fidelities of 99.90%-100%. In contrast, KATOIII and AGS showed significantly elevated error rates, mainly due to increased de novo methylation, in one CGI (2.2-fold) and in three CGIs (1.6-to 3.2-fold), respectively. This showed that the two gastric cancer cell lines with CIMP had decreased fidelity in replicating methylation patterns that produced scattered methylation of a CGI (Fig. 4B). Interestingly, the decreased fidelity was prominent in specific CGIs, such as promoter CGIs of bA305P and RIKEN2210016.

Decreased Fidelity and Induction of Dense Methylation

The next question was whether or not the scattered methylation induced by the decreased fidelity really leads to induction of methylation of an entire CGI

bA305P

FLJ32130

RIKEN2210016

0.10

0.10

0.05

0.10

0.05

r^rilrEEn ifi.

E-cadherin

Cydophilin A

0.10

0.10

0.05

O UJ

Fig. 3 Error rates of two gastric cancer cell lines without CIMP (HSC39 and HSC57) and two with CIMP (KATOIII and AGS) in five promoter CGIs. The numbers of errors per CpG sites in 21.6-23.1 generations are shown (note that the unit is different from Fig. 2). KATOIII showed increased error rates in bA305P, and AGS showed increased error rates in bA305P, RIKEN2210016, and E-cadherin

(dense methylation; Fig. 4E). As for the role of scattered methylation, Song et al. reported that both "seeds of methylation," which they created by HpaIl methylase, and decreased gene expression were important for induction of dense methylation of a promoter CGI (Song et al. 2002). The finding was further confirmed in the authors' following report (Stirzaker et al. 2004). Encouraged by these reports, we decided to detect densely methylated DNA molecules by selective amplification of such molecules by methylation-specific PCR (MSP) (Herman et al. 1996). MSP is known to be capable of detecting a small number of methylated DNA molecules embedded in an excess amount of unmethylated DNA molecules.

0 0

Post a comment