Epigenetic inactivation of TCF2 in ovarian cancer and various cancer cell lines

Transcription factor 2 gene (TCF2) encodes hepatocyte nuclear factor 1β (HNF1β), a transcription factor associated with development and metabolism. Mutation of TCF2 has been observed in renal cell cancer, and by screening aberrantly methylated genes, we have now identified TCF2 as a target for epigenetic inactivation in ovarian cancer. TCF2 was methylated in 53% of ovarian cancer cell lines and 26% of primary ovarian cancers, resulting in loss of the gene's expression. TCF2 expression was restored by treating cells with a methyltransferase inhibitor, 5-aza-2′deoxycitidine (5-aza-dC). In addition, chromatin immunoprecipitation showed deacetylation of histone H3 in methylated cells and, when combined with 5-aza-dC, the histone deacetylase inhibitor trichostatin A synergistically induced TCF2 expression. Epigenetic inactivation of TCF2 was also seen in colorectal, gastric and pancreatic cell lines, suggesting general involvement of epigenetic inactivation of TCF2 in tumorigenesis. Restoration of TCF2 expression induced expression of HNF4α, a transcriptional target of HNF1β, indicating that epigenetic silencing of TCF2 leads to alteration of the hepatocyte nuclear factor network in tumours. These results suggest that TCF2 is involved in the development of ovarian cancers and may represent a useful target for their detection and treatment.

Ovarian cancer is the most deadly of gynecological malignancies, with an overall 5-year survival rate of o30%. In large part, this is because the disease usually presents at an advanced stage, as there are no overt symptoms at early stages. Ovarian cancers are morphologically and biologically heterogeneous and associated with distinct genetic alterations. For example, the serous type frequently shows p53 mutations, the mucinous type K-ras mutations, and the endometrioid type b-catenin mutations. The molecular pathogenesis of ovarian cancer is not fully understood, however, and it is hoped that a better understanding of the molecular mechanisms underlying the tumorigenic process will lead to earlier diagnosis, novel therapies and, ultimately, better outcomes.
DNA methylation can be a good molecular marker by which to identify genes inactivated in cancer (Jones & Baylin, 2002;Laird, 2003;Issa, 2004;Costello, 2005), and several techniques, including restriction landmark genome scanning and methylation-sensitive arbitrary primer PCR, have been used to screen for aberrantly methylated DNA fragments in tumour cells (Zardo et al, 2002;Yu et al, 2005). In addition, we have developed a novel methylation screening technique, methylated CpG island amplification (MCA), which we have used to identify DNA fragments that are hypermethylated in cancer (Toyota et al, 1999). And by coupling MCA to representational difference analysis (RDA) we have been able to identify those genes inactivated by hypermethylation (Toyota et al, 1999;Ueki et al, 2001;Obata et al, 2003;Palmisano et al, 2003;Yamada et al, 2004). When we used MCA in combination with RDA to screen for genes methylated in an ovarian cancer cell line, we observed for the first time that transcription factor 2 gene (TCF2) is a target for epigenetic inactivation.
TCF1 and TCF2, respectively, encode the transcription factors hepatocyte nuclear factor 1a and b (HNF1a and HNF1b), two POU-homeodomain proteins that bind to an essential element in the proximal promoter sequences of albumin and many other tissue-specific genes and are associated with development, metabolism and cancer (Cereghini et al, 1987;Courtois et al, 1987;Frain et al, 1989;Rey-Campos et al, 1991). For instance, TCF2 defects cause developmental abnormalities in multiple tissues, including kidney and pancreas (Horikawa et al, 1997;Haumaitre et al, 2005), and Bluteau et al (2002) identified biallelic mutations of TCF1 in hepatic adenomas (Bluteau et al, 2002). Subsequently, mutations of TCF1 also were found in colorectal cancer with MSI (Laurent-Puig et al, 2003) and endometrial tumours (Rebouissou et al, 2004), while Rebouissou et al (2005) also identified mutations of TCF2 in renal carcinomas.
With that as background, our aim in the present study was to evaluate TCF2 methylation in panels of ovarian cancer cell lines and primary ovarian cancers. Our findings indicate that TCF2 expression is lost in ovarian cancer cell lines that show methylation and that expression is restored by treating cells with a methyltransferase inhibitor, confirming epigenetic silencing of the gene. Moreover, restoration of TCF2 expression induced expression of HNF4a, a transcriptional target of HNF1b, indicating that epigenetic silencing of TCF2 leads to alteration of the hepatocyte nuclear factor network in tumours. These results suggest that epigenetic inactivation of TCF2 is involved in development of ovarian cancer.

MCA/RDA
MCA was carried out as described previously (Toyota et al, 1999). We used the MCA amplicon from the OVCAR-3 ovarian cancer cell line as a tester and that from normal ovarian tissue as a driver. After two rounds of subtraction by RDA, the PCR products were digested with XmaI and ligated to pBluescriptII (Stratagene). The inserts were then sequenced using an ABI3100 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Database analysis was performed using Blast and BLAT.

Bisulphite treatment of genomic DNA
Genomic DNA was isolated from cell lines and primary tissue samples (cancerous and normal) using the phenol -chloroform method and then treated with sodium bisulphite as described previously (Clark et al, 1994). Briefly, 2 mg of DNA were denatured for 10 min at 371C in 2 M NaOH, after which 30 ml of 10 mM hydroquinone (Sigma Chemical Co, St Louis, MO, USA) and 520 ml of 3 M sodium bisulphite were added. The mixture was then incubated for 16 h at 501C, and the modified DNA was purified using a Wizard DNA Purification System (Promega, Madison, WI, USA). After treating the DNA with NaOH a second time, the resultant DNA precipitate was resuspended in 20 ml of TE buffer and stored at À801C until use.

Direct-sequencing of Bisulphite-PCR products
The PCR products obtained with COBRA were used for directsequencing. Briefly, 10 ml of bisulphite-PCR product were electrophoresed in 1% SeaPlaque agarose gel, excised, purified using a PCR Purification System (Promega, Madison, WI, USA) and eluted in 50 ml of water. Samples (5 ml) were then used as templates for the sequencing reaction, which was carried out using a BigDye terminator cycle sequencing kit (Applied Biosystems, Foster City, CA, USA) with an ABI PRISM 3100 sequencer according to the manufacturer's guidelines (Applied Biosystems).

Reverse transcription -polymerase chain reaction
Expression of TCF2 and HNF4a was analyzed using reverse transcription -polymerase chain reaction (RT-PCR). Total RNA was extracted from cell lines using Trizol (Life Technologies Inc., Gaithersburg, MD, USA) according to the manufacturer's instructions. The RT reaction was carried out with 2 mg of total RNA using a SuperScript II First-Strand Synthesis System (Invitrogen Inc., Grand Island, NY, USA) with random primer. PCR was carried out in solution containing 1 Â PCR buffer (TaKaRa), 200 mM each deoxynucleotide triphosphate, 2.5 pmol of each primer, 1 U of ExTaq polymerase (TaKaRa) and 5% (v/v) DMSO. The oligonucleotide primer sequences used were 5 0 -CTATGACACACCTCC CATCCTCAAG-3 0 (TCF2RT-F), 5 0 -GTCTGGTTGAATTGTCGGAG GATCT-3 0 (TCF2RT-R), 5 0 -CGGCAGTGCGTGGTGGACAAAGAC-3 0 (HNF4-aRT-F), and 5 0 -CACACACATCTGCGATGCTGGCAAT-3 0 (HNF-4aRT-R). The PCR cycling protocol consisted of 1 cycle at 951C for 5 min and 32 cycles at 951C for 30 s, 601C for 30 s and 721C for 45 s. The housekeeping gene GAPDH served as an internal control to confirm the success of the RT reaction. The PCR products were subjected to 2.5% agarose gel electrophoresis.
For quantitative analysis, real-time PCR was carried out using a TaqMan PCR system (Applied Biosystems). PCR was carried out in 50 ml of solution containing 1 ml of cDNA, 2.5 pmol of each primer and 25 ml of TaqMan PCR mixture (Applied Biosystems). The PCR cycling protocol consisted of 1 cycle at 951C for 5 min and 40 cycles at 951C for 30 s and 601C for 1 min. Fluorescent signals were detected using an ABI 7000 Prism 7000 (Applied Biosystems), and the accumulation of PCR product was measured in real time as the increase in the fluorescent signal. Data were analysed using ABI Prism 7000 SDS Software (Applied Biosystems). The probes and primer sets were ABI Assay on Demand and Ref Hs00172123 for TCF2 and Hs00230853 for HNF4a (Applied Biosystems).

Mutational analysis
In all, 10 pairs of primers for individually amplifying and screening exons 1 -9 of TCF2 were used for amplifying genomic DNA. Spanning the entire coding region of the gene, these primers were specified by a Human BLAT Search (Chr17q12,position: 33,120,179,182), and their sequences are available upon request. PCR was carried out in a 50-ml reaction mixture containing 1 Â PCR buffer (TaKaRa), 1 mM primers, 0.25 mM dNTP mixture and 1.0 U of Hot Start Taq polymerase (TaKaRa). The amplified PCR products were electrophoresed in 1% Seaplaque gels, excised, purified using a Gel purification system (Qiagen, Hilden, Germany) and sequenced. The primer sequences and PCR conditions used for mutational analysis are available upon request.

Chromatin immunoprecipitation and PCR
Chromatin immunoprecipitation was carried out as described previously (Ueno et al, 2004). Briefly, cells were incubated in 1% formaldehyde for 10 min at 371C. The nuclei were then collected, sonicated to yield fragments ranging in size from 300 to 2000 bp, and immunoprecipitated with anti-histone H3 antibody (Upstate Biotechnology, Lake Placid, NY, USA), which specifically recognizes the diacetylated lysine residues (lysines 9 and 14) of histone H3. The chromatin was recovered using protein A Sepharose. Realtime PCR was then performed using 1 ml of chromatin DNA as shown above. The primer sequences used were 5 0 -TGGATTT GGGGTTTGCTTGTGA-3 0 (TCF2ChIP-F) and 5 0 -GACGTGAGCTT GGACACCATTTTC-3 0 (TCF2ChIP-R). Standard curves relating initial template copy number to fluorescence and amplification cycle were generated using the amplified PCR product as a template, and were used to calculate the DNA copy number in each sample. Ratios of the intensities of the TCF2 and GAPDH signals were used as a relative measure of the level of TCF2 expression in each specimen.

Statistical analysis
The statistical analysis was carried out using StatView software (SAS Institute Inc., Cary, NC, USA). Fisher's exact test (two-sided) was used to determine the association between TCF2 methylation and clinicopathological features. Values of Po0.05 were considered significant. Overall survival time was defined as the period between the diagnosis of ovarian cancer and the time of death. Differences between survival curves were assessed by Kaplan -Meier analysis using the log-rank test.

Identification of TCF2 as a target for epigenetic inactivation in ovarian cancer
Using MCA coupled with RDA, we recovered DNA fragments located within intron 1 of TCF2 ( Figure 1A). The 5 0 region of TCF2 contains a CpG rich region that fulfills the criteria for a CpG island (observed CpG: expected CpG ¼ 0.821, GC content 58.9%, length 3.4 kb). Then when we used COBRA to examine the methylation status of this region in 15 ovarian and one cervical cancer cell line ( Figure 1B), we found it to be densely methylated in eight (53%) of the ovarian cancer cell lines, but not in the cervical cancer cell line. Moreover, using the same methodology, we determined that TCF2 was aberrantly methylated in 26 of 98 (26.5%) primary ovarian cancers, but not in any of the four samples of normal ovarian tissue analysed ( Figure 1C). Apparently, methylation of TCF2 is a cancer-specific event and is not a cell line-specific one.
To confirm whether the methylation detected by COBRA reflects the overall level of methylation in the 5 0 region of TCF2, we carried out bisulphite-sequencing of the gene (Figure 2A and B). Again, methylation of TCF2 was not detected in normal ovarian tissues. On the other hand, the ovarian cancer cell lines shown by COBRA to be methylated also showed dense methylation over all their CpG islands, while those shown by COBRA not to be methylated showed no methylation of their CpG islands (Figure 2A and B). To determine whether methylation of TCF2 is correlated with gene silencing, we used RT -PCR to evaluate the gene's expression in 15 ovarian and one cervical cancer cell lines (Figure 3). No expression of TCF2 was detected in any of the nine cell lines that showed methylation; that is, methylation was significantly associated with an absence of TCF2 mRNA (Po0.001, Fisher's exact test, twosided). By contrast, all seven cell lines that showed no methylation expressed the transcript, indicating a role for DNA methylation in the silencing of TCF2.
To determine whether TCF2 is inactivated by mutation, we first carried out direct sequence analysis of TCF2 using DNA from  ovarian cancers. Using 10 sets of primers that respectively amplified exons 1 -9, we examined 15 cell lines and 33 primary ovarian tumours, and discovered two base substitutions in the introns, which were determined to be polymorphisms (data not shown, and Supplementary Figure 1). Thus, inactivation of TCF2 appears to be caused by epigenetic changes rather than genetic ones.

Silencing of TCF2 in human tumour cell lines
We next carried out COBRA using DNA from a panel of human cancer cell lines to determine the frequency with which TCF2 was methylated in other tumour types ( Figure 4A). Methylation of TCF2 was detected in six of eight (75%) colorectal, four of eight (50%) gastric and six of seven (85.7%) pancreatic cancer cell lines ( Figure 4B). Moreover, as in ovarian cancer, expression of TCF2 was absent in cell lines that showed methylation. One colorectal cancer cell line, DLD-1, also did not express TCF2, although this cell line did not show methylation.
The role of DNA methylation and histone deacetylation in the silencing of TCF2 The role of DNA methylation in the silencing of TCF2 was further confirmed by our finding that treating cells with the methyltransferase inhibitor 5-aza-dC (1 mM) restored gene expression. Of four methylated cell lines tested, all showed restored TCF2 expression following treatment with 5-aza-dC ( Figure 5A). By contrast, expression of TCF2 was not restored by the histone deacetylase inhibitor TSA, but when combined with a low dose of 5-aza-dC, TSA exerted a synergistic effect augmenting gene expression ( Figure 5B). Using ChIP assays to examine the acetylation status of histone H3, we found that DNA methylation was associated with a low level of histone acetylation, indicating that histone deacetylation, too, plays a role in the silencing of TCF2 ( Figure 5C).

The role of TCF2 methylation in the regulation of HNF4a
We next evaluated the expression of HNF4a, one of the transcriptional targets of HNF1b, in cell lines with and without TCF2 methylation ( Figure 6A). Expression of HNF4a was detected in all cancer cell lines that expressed TCF2. By contrast, cell lines that showed TCF2 methylation showed little or no HNF4a expression. Notably, treating cells with 5-aza-dC upregulated expression of HNF4a ( Figure 6B and C). As the 5 0 region of HNF4a does not contain a CpG island, the effect of 5-aza-dC on the gene's expression is likely not associated with demethylation of its promoter, but instead underscores the importance of TCF2 expression, and thus HNF1b, in the regulation of HNF4a.
Clinicopathological features of primary ovarian cancer with or without TCF2 methylation Finally, we examined the clinicopathological features of ovarian cancers with and without TCF2 methylation (Table 1). Aberrant methylation of TCF2 was detected in 12 of 29 (41.3%) serous tumours, three of 12 (25.0%) mucinous tumours, eight of 28 (28.6%) endometrioid tumours and three of 12 (25.0%) others (undifferentiated, mixed type), but in none of the 17 (0%) clear cell tumours. Methylation of TCF2 was detected significantly less frequently in the clear cell phenotype than the other types (0% vs 32.1%, Fisher's Exact test, P ¼ 0.005), but there was no significant association with any other clinicopathological feature (patients' age, FIGO clinical stage, and pathological grade). There was also no correlation between the patients' prognoses and the TCF2 methylation status (data not shown, and Supplementary Figure 2 Figure 3 RT -PCR analysis of TCF2 mRNA expression in ovarian cancer cell lines. Controls consist of carrying out PCR reactions without reverse transcription (À). The cell lines examined are shown above the gels. Expression of GAPDH was evaluated to confirm the integrity of cDNA.

DISCUSSION
The transduction network associated with tumorigenesis and the progression of ovarian cancer is not fully understood, although numerous studies indicate that DNA methylation plays a key role in human neoplasia (Jones & Baylin, 2002;Laird, 2003;Issa, 2004;Costello, 2005). Although methylation of several cancer-related genes has been reported in ovarian cancer (Yu et al, 1999;Terasawa et al, 2004), the methylated genes have not yet been fully characterised. Using the MCA genome screening technique coupled with RDA, we identified TCF2 as a novel target of epigenetic inactivation through methylation in ovarian cancer. The Methylation of TCF2 was cancer-specific and, together with histone deacetylation, may be involved in tumorigenesis in a subset of ovarian cancers. It has been reported that patients with germ line mutation of TCF2 develop renal tumours (Bellanne-Chantelot et al, 2004), and TCF2 mutations have been identified in renal cell cancer (Rebouissou et al, 2005). On the other hand, no TCF2 mutation has been previously observed in ovarian cancer (Rebouissou et al, 2005), which is consistent with our present finding that TCF2 is not mutated in ovarian cancer. Apparently, TCF2 expression is silenced by an epigenetic rather than genetic mechanism. That methylation of TCF2 was also detected in colorectal, gastric and pancreatic cancer cell lines suggests TCF2 is prone to methylation in many types of cancer and may thus play a general role in tumorigenesis. In addition, one ovarian cancer cell line, TOV112D, and one colorectal cancer cell line, DLD-1, did not express TCF2, although little or no methylation was detected in tumours. Genes silenced in cancer are often not expressed in cancer cell lines (Suzuki et al, 2002), and their silencing may be caused by mechanisms other than methylation, for example by the absence of a transcription factor or histone modification (Kondo et al, 2004).
The methylation profile of ovarian cancer may be a useful diagnostic marker with which to predict recurrence, resistance to chemotherapy and survival (Balch et al, 2004). We and others have reported that TMS1, 14-3-3sigma and WT1 genes are methylated more frequently in clear cell tumours than in other tumour types (Kaneuchi et al, 2004;Terasawa et al, 2004;Kaneuchi et al, 2005), whereas methylation of SFRP1, 18S and 28S ribosomal DNA genes is more frequently detected in nonclear cell tumours, such as the serous type (Takada et al, 2004;Chan et al, 2005). Although the molecular mechanisms responsible for the differing methylation profiles in different tumour types remains unclear, gene methylation may have relevance in the clinical situation. For instance, clear cell ovarian cancers are reportedly more resistant to chemotherapy than other types (Sugiyama et al, 2000), and this difference in sensitivity may be caused in part by differences in the genes methylated in clear cell and nonclear cell tumours. In fact, Teodoridis et al (2005) reported that methylation of genes involved in DNA repair was significantly associated with response to chemotherapy, while Chan et al (2005) reported that methylation of 18S and 28S ribosomal DNA genes is associated with longer progression-free survival in ovarian cancer. Taken together, these results suggest that differences in the epigenetic signatures between clear cell and other types of ovarian cancer may be a useful diagnostic indicator.
The specific role of TCF2 methylation in the development and progression of ovarian cancer remains unknown, but TCF2 mutations are known to affect expression of downstream genes such as HNF4a, PKHD1 and UMOD Rebouissou et al, 2005). We showed expression of HNF4a to be well correlated with that of TCF2 and to be upregulated by inducing TCF2 expression using a methyltransferase inhibitor, which suggests alterations in the hepatocyte nuclear factor network can be reversed by inducing TCF2 through demethylation of the gene. Consistent with that idea, Lazarevich et al (2004) reported that expression of HNF4a is downregulated in hepatocellular carcinoma and its restoration suppresses tumour growth. It has also been reported that HNF4a is downregulated in other types of tumours (Lemm et al, 1999) and that it induces expression of  endothelial Fas ligand (FasL), which prevents cancer cell transmigration (Osanai et al, 2002). Collectively then, the summarised findings suggest that impairment of the HNF1b/ HNF4a signalling network may lead to tumour formation. Similar epigenetic inactivation of a critical regulator of a transcriptional network has been reported for several other genes, including estrogen receptor (Leu et al, 2004) and GATA (Akiyama et al, 2003). Epigenetic inactivation of such transcription factors may affect multiple downstream genes, causing aberrant alteration of signalling pathways involved in a wide array of cellular functions, including growth, differentiation and apoptosis. Conversely, restoration of such transcription factors may lead to upregulation of genes throughout the entire signalling network and restoration of proper function. That said, further study will be necessary to definitively determine the role of epigenetic inactivation of TCF2 in the molecular mechanism of tumorigenesis and progression of ovarian cancer and its utility as a therapeutic target.