Transient Down-regulation of DNMT1 Methyltransferase Leads to Activation and Stable Hypomethylation of MAGE-A1 in Melanoma Cells*

MAGE-A1 belongs to a group of germ line-specific genes that rely primarily on DNA methylation for repression in somatic tissues. In many types of tumors, the promoter of these genes becomes demethylated and transcription becomes activated. We showed previously that, although MZ2-MEL melanoma cells contain an active unmethylated MAGE-A1 gene, they lack the ability to induce demethylation of newly integrated MAGE-A1 transgenes that were methylated in vitro before transfection. In the same cells, unmethylated MAGE-A1 transgenes were protected against remethylation, and this appeared to depend on the level of transcriptional activity. We therefore proposed that hypomethylation of MAGE-A1 in tumors relies on a past demethylation event and on the presence of appropriate transcription factors that maintain the promoter unmethylated. Here, we tested this hypothesis further by examining whether induction of a transient demethylation phase in MZ2-MEL would suffice to convert a previously methylated MAGE-A1 transgene into a permanently hypomethylated and active one. For induction of the demethylation phase, we used antisense oligonucleotides targeting the three known human DNA methyltransferases. We found that down-regulation of DNMT1, but not of DNMT3A and DNMT3B, induces activation of the MAGE-A1 transgene, suggesting that DNMT1 has a predominant role for methylation maintenance in MZ2-MEL cells. By using a selectable MAGE-A1 transgene construct, we were able to isolate a cell population in which DNMT1 depletion had resulted in transgene activation. The promoter region of the transgene was almost completely unmethylated in these cells, and this active and unmethylated state was maintained for over 60 days after restoration of normal DNMT1 expression.

MAGE-A1 belongs to a group of germ line-specific genes that rely primarily on DNA methylation for repression in somatic tissues. In many types of tumors, the promoter of these genes becomes demethylated and transcription becomes activated. We showed previously that, although MZ2-MEL melanoma cells contain an active unmethylated MAGE-A1 gene, they lack the ability to induce demethylation of newly integrated MAGE-A1 transgenes that were methylated in vitro before transfection. In the same cells, unmethylated MAGE-A1 transgenes were protected against remethylation, and this appeared to depend on the level of transcriptional activity. We therefore proposed that hypomethylation of MAGE-A1 in tumors relies on a past demethylation event and on the presence of appropriate transcription factors that maintain the promoter unmethylated. Here, we tested this hypothesis further by examining whether induction of a transient demethylation phase in MZ2-MEL would suffice to convert a previously methylated MAGE-A1 transgene into a permanently hypomethylated and active one. For induction of the demethylation phase, we used antisense oligonucleotides targeting the three known human DNA methyltransferases. We found that down-regulation of DNMT1, but not of DNMT3A and DNMT3B, induces activation of the MAGE-A1 transgene, suggesting that DNMT1 has a predominant role for methylation maintenance in MZ2-MEL cells. By using a selectable MAGE-A1 transgene construct, we were able to isolate a cell population in which DNMT1 depletion had resulted in transgene activation. The promoter region of the transgene was almost completely unmethylated in these cells, and this active and unmethylated state was maintained for over 60 days after restoration of normal DNMT1 expression.
Cytosines in CpG dinucleotides are often methylated in mammalian genomes. This epigenetic modification of DNA exerts a potent repressive effect on transcription by preventing the binding of transcription factors and by recruiting methyl-CpG-binding proteins, which in turn attract repressor complexes that modify chromatin structures by deacetylating or methylating specific residues on histones (1). DNA methylation has an essential role in the inactivation of one of the two X chromosomes in female somatic cells (2), in the monoallelic silencing of parentally imprinted genes (3), in the repression of genomic parasite sequences (4), and in the regulation of tissue-specific gene expression (5)(6)(7).
DNA methylation patterns are acquired during development by a process that involves a phase of extensive demethylation in the early embryo, followed by an active phase of de novo methylation (8,9). Once established, methylation patterns are inherited through cell generations by a maintenance process that copies pre-existing methylation sites onto the newly synthesized strand (10). In mammals, the transfer of a methyl group to cytosines is catalyzed by three DNA methyltransferases, namely DNMT1, 3 DNMT3A, and DNMT3B (11). The two latter enzymes have de novo methylation activity and are responsible for the establishment of DNA methylation patterns during embryonic development (12). Consistently, DNMT3A and -3B are highly expressed in the embryo, whereas they are expressed at low levels in adult somatic cells (13). DNMT1, which has a preference for hemimethylated DNA, is the major maintenance methyltransferase (14 -18). In embryonic cell lines, however, DNMT3A and DNMT3B appear to cooperate with DNMT1 to maintain methylation patterns (19,20).
Cancer cells show altered patterns of DNA methylation (21). Normally unmethylated promoters often become densely methylated in tumor cells, and this results in silencing of critical genes such as tumor suppressor genes. In many tumors, including these same tumors, the global level of DNA methylation is usually decreased (22,23). Using a mouse model in which DNMT1 was inhibited by a genetic approach, Gaudet et al. (24) provided evidence that genome-wide hypomethylation promotes tumor formation by increasing genomic instability. Another consequence of the overall demethylation process in tumors is the activation of genes that are controlled by DNA methylation (25). This is the case for a group of germ line-specific genes that rely primarily on DNA methylation for silencing in normal somatic tissues (7,26,27). Thus, these genes become aberrantly activated in many different types of tumors, where they direct the expression of tumor-specific antigens recognized by cytolytic T lymphocytes (28). To date, more than 30 genes or gene families showing specific expression in germ line and cancer cells have been identified. They are named cancer-germ line or cancertestis genes.
The mechanisms leading to DNA hypomethylation in tumors are still unclear. In a previous report we addressed this issue by using the MAGE-A1 cancer-germ line gene as a model (29). We found that in tumor cells expressing this gene, hypomethylated CpGs are not randomly distributed over the MAGE-A1 locus but are clustered within the 5Ј-region of the gene. In the MZ2-MEL melanoma cell line for instance, where MAGE-A1 is expressed, hypomethylation within the gene is restricted to a ϳ3-kb region centered on the transcription start site. This site-specific hypomethylation does not appear to be due to a permanent targeted demethylating activity, because in vitro methylated MAGE-A1 sequences did not undergo demethylation following stable transfection into MZ2-MEL cells. Remarkably, in the same cells unmethylated MAGE-A1 transgenes were progressively remethylated at all sites except those located in the 5Ј-region of the gene. This local inhibition of de novo methylation appeared to depend on the level of transcriptional activity. Based on these observations, we proposed that MAGE-A1 hypomethylation in tumors might be the result of a past event of demethylation, which would be maintained by the presence of potent transcriptional activators (29).
In the experiments described above, protection of the MAGE-A1 promoter region against de novo methylation was observed after transfection of naked unmethylated DNA. In the present study, we investigated whether a previously methylated transgene would similarly become permanently protected against remethylation following induction of a transient demethylation phase. To this end, we introduced an in vitro methylated MAGE-A1 construct into MZ2-MEL cells. We then induced a transient demethylation phase by treating the cells for a limited period of time with antisense oligonucleotides targeted to the DNA methyltransferases (30,31). By testing different combinations of anti-DNMT1, anti-DNMT3A, and anti-DNMT3B oligonucleotides, we found that depletion of only DNMT1 resulted in transgene activation. However, anti-DNMT1 oligonucleotides produced transgene derepression in only a small fraction of the treated cells. By using a transgene construct in which the last exon of MAGE-A1 was replaced by the sequence encoding resistance to hygromycin, we were able to select the cells in which transgene activation occurred. Selected cells were then assessed for stability of demethylation and activation of the transgene.

MATERIALS AND METHODS
Plasmid Constructions-The pMAGEA1/EGFP plasmid was constructed by removing the EBFP sequence from pEBFPM1-WT plasmid (29) following digestion with NotI and AgeI and by replacing it with the EGFP coding unit released from the pEGFP-N1 plasmid (Clontech, Palo Alto, CA) following digestion with the same enzymes. For construction of the pMAGEA1/hph plasmid, the hph sequence from Ϫ12 to ϩ1077 relative to the translation initiation site was amplified by PCR from pHMR272 (32) with primers 5Ј-ACGCGTCGACGATCCGGGAGAT-ATGAAAAA and 5Ј-ATAAGAATGCGGCCGCGTTCTCGAAATC-AGCT, which carry a 5Ј-overhang containing a SalI or a NotI restriction site, respectively. On the other hand, the MAGE-A1 sequence located between position Ϫ1055 and position Ϫ3 relative to the translation start site and containing a KpnI site at position Ϫ1022 was amplified by PCR using the sense primer 5Ј-GCGTTCGGGTGAGGAACA and an antisense primer carrying a 5Ј-overhang with a SalI site, 5Ј-ACGCGTC-GACTCTCGTCAGGGCAGCA. The MAGE-A1 and hph fragments were digested with KpnI and SalI or with SalI and NotI, respectively, and were co-ligated to the pEBFPM1-WT in which part of the MAGEA1/ EBFP insert had been removed by KpnI and NotI digestion. Both pMA-GEA1/EGFP and pMAGEA1/hph plasmids contain a geneticin resistance gene for selection of transfected cells. For construction of pDNMT3A and pDNMT3B, the open reading frame of DNMT3A or DNMT3B was amplified from human MZ2-MEL cells using Prime-STAR DNA polymerase (Takara Bio Inc.). The DNMT3A PCR product was introduced into the pCDNA3 vector (Invitrogen) between the Hin-dIII (blunted) and EcoRI sites. The DNMT3B open reading frame was introduced into the pCDNA3.1 vector (Invitrogen) between the KpnI and XbaI sites. All constructs were verified by sequencing.
In Vitro Methylation and Transfections of Plasmids-pMAGEA1/ EGFP and pMAGEA1/hph DNAs were prepared using the EndoFree Plasmid Maxi kit (Qiagen, Hilden, Germany). Plasmids were methylated in vitro with the SssI methylase, which methylates cytosines in CpG dinucleotides, as previously described (29). Efficient methylation of the plasmids was confirmed by their resistance to digestion upon incubation with the methylation-sensitive restriction enzyme HpaII.
The MZ2-MEL3.1 cell line and conditions of cell culture have been described previously (33). The cells (4 ϫ 10 5 ) were inoculated in 75-cm 2 flasks 48 h before transfection. Transfections were performed using the calcium phosphate precipitation method (34) with 10 g of plasmid DNA (carrying a geneticin resistance gene) diluted in 20 g of EcoRI-digested human genomic DNA. Dilution of the plasmid in genomic DNA was previously shown to minimize the integration of multiple copies in MZ2-MEL cells (29). Transfectants were selected in medium containing 2 mg of geneticin/ml for 15-16 days. Transfected cell populations were cloned by limiting dilutions. Several clones were cultured in the absence or in the presence of 2 M 5-aza-2Ј-deoxycytidine during 3 days, and expression of the transgene was tested by RT-PCR (see RT-PCR conditions below). This led to the selection of clone MZ2-MEL.Tr4 (containing an in vitro methylated pMAGEA1/EGFP plasmid) and clone MZ2-MEL. TrHM (containing an in vitro methylated pMAGEA1/hph plasmid) in which the transgene was completely repressed but was strongly induced upon treatment with 5-aza-2Ј-deoxycytidine.
Antisense Oligonucleotide Transfections-2 ϫ 10 6 cells were seeded in 75-cm 2 flasks 24 h before transfection. Transfections were performed in 4.5 ml of OPTIMEM-1 medium (Invitrogen) containing 6.25 g/ml of Lipofectin (Invitrogen) and 50 -100 nM of each 2Ј-O-methyl phosphorothioate antisense oligonucleotides, which were commercially synthesized (Eurogentec SA, Seraing, Belgium). After 4 h of incubation, cells were transferred in complete medium. Cells were transfected in this way every other day and split 24 h after the transfections. The antisense oligonucleotide sequences were described elsewhere, with DNMT1-AS oligonucleotide referring to MG98 oligonucleotide (30,31). The sequence of the control scramble phosphorothioate oligonucleotide is 5Ј-CCTAGCAAGATCCGTCATCT, with the four nucleotides at each end carrying 2Ј-O-methyl modifications. For selection of MZ2-MEL.TrHM cells in which DNMT1 depletion resulted in transgene activation, cells were transferred in medium containing 180 g/ml of Hygromycin B (CalBiochem, San Diego, CA).
Bisulfite Genomic Sequencing-Sodium bisulfite genomic sequencing was performed as described previously (29). Primers for the first round of PCR amplification of the MAGE-A1 promoter region were as described elsewhere (7). For the second semi-nested PCR, we used the same sense primer together with 5Ј-CTAAAACRTCTTCCCRCRC-TCTAA as antisense primer.

Generation of an MZ2-MEL Clone Containing a Methylated
MAGE-A1 Transgene (MZ2-MEL.Tr4)-To produce an MZ2-MEL clone containing a methylated MAGE-A1 transgene, we transfected MZ2-MEL cells with an in vitro methylated construct carrying a 6.9-kb fragment of MAGE-A1 (Ϫ3255 to ϩ3639 relative to the transcription start site) fused to the enhanced green fluorescent protein (EGFP) transcription unit (Fig. 1A). Transfectants were selected according to their resistance to geneticin, which was conferred by the transgene vector, and several clones were isolated. Among these, we selected clone MZ2-MEL.Tr4, because the transgene in these cells was completely repressed but could be strongly activated upon treatment with the demethylating agent 5-aza-2Ј-deoxycytidine (Fig. 1B). Methylation of the MAGEA1/ EGFP transgene was maintained in untreated MZ2-MEL.Tr4 cells even after long-term culture (data not shown). This confirms our previous observations that, although MZ2-MEL cells contain an endogenous unmethylated MAGE-A1 gene, they lack the ability to induce demethylation of newly integrated MAGE-A1 transgenes.
Efficient Inhibition of DNMT1, DNMT3A, and DNMT3B Expression by Antisense Oligonucleotides-To induce demethylation of the MAGE-A1 transgene in MZ2-MEL.Tr4, we decided to transiently deplete the maintenance DNA methyltransferase(s) by treating the cells with specific antisense oligonucleotides. Although methylation maintenance has been classically attributed to DNMT1, recent evidence suggests that in some tumor cells DNMT3 methyltransferases participate in this process (36,37). We therefore tested the effect of depleting each of the three DNMTs in MZ2-MEL.Tr4. To this end, we used the phosphorothioate antisense oligonucleotides (DNMT1-AS, DNMT3A-AS, and DNMT3B-AS) previously described by Robert et al. (30,31).
We first evaluated the level of expression of the three DNMTs in MZ2-MEL.Tr4 by quantitative RT-PCR. The results showed that the amount of DNMT1 mRNA in these cells is ϳ15 and 10 times higher than that of DNMT3A and DNMT3B, respectively ( Fig. 2A). We also performed Western blot analysis on a nuclear extract from MZ2-MEL.Tr4. Although ␣-DNMT1 antibodies revealed a clear signal at a size corresponding to the DNMT1 protein (ϳ191 kDa; Fig. 2B), ␣-DNMT3A and ␣-DNMT3B antibodies did not allow detection of a band of the expected size above the background (Fig. 2B). Lack of detection of DNMT3A and DNMT3B proteins is likely related to the low level of transcription of the corresponding genes in MZ2-MEL.Tr4. The proteins became detectable in MZ2-MEL.Tr4 that were induced to express increased levels of DNMT3A (10-and 86-fold increase) or DNMT3B transcripts (15-and 61-fold increase) following transfection of different amounts of pDNMT3A and pDNMT3B expression vectors (Fig. 2B). Taken together, these data indicate that DNMT1 is the major DNA methyltransferase in MZ2-MEL.Tr4.
Although the efficiency of the DNMT-AS oligos has been demonstrated previously in different cell lines (30,31), we evaluated the ability of these antisense oligonucleotides to inhibit expression of their corresponding DNMT gene in MZ2-MEL.Tr4 cells. To this end, the cells were transfected with different combinations of the three DNMT-AS oligos or with an unrelated scramble oligonucleotide, and the level of mRNA expression of the three DNMTs was evaluated 24 h after transfection. As shown in Fig. 2C, the DNMT-AS oligos specifically inhibited their target gene with a ϳ70% reduction in the corresponding mRNA expression level. In contrast, the scramble oligonucleotide had no significant effect on DNMT expression levels. We confirmed by Western blot analysis that the reduction in DNMT1 mRNA induced by the DNMT1-AS oligo was associated with depletion of the corresponding protein (Fig. 2D).
Derepression of the MAGEA1/EGFP Transgene in DNMT1-depleted MZ2-MEL.Tr4 Cells-Having shown that the DNMT-AS oligos specifically inhibit expression of the corresponding gene in MZ2-MEL.Tr4, we tested their ability to induce derepression of the methylated MAGEA1/EGFP transgene. In the first set of experiments, activation of MAGEA1/EGFP was compared following depletion of DNMT1 or both DNMT3A and DNMT3B. The two latter enzymes were targeted simultaneously because they are known to act redundantly (12). DNA demethylation upon inhibition of methylation maintenance occurs by a passive replication-dependent process. We therefore maintained depletion of the DNMTs for several days by transfecting the cells every 48 h with the antisense oligonucleotides. MZ2-MEL.Tr4 cells that had been treated in this way during 5 and 9 days were collected and analyzed by quantitative RT-PCR to evaluate the level of expression of the MAGEA1/EGFP transgene. At day 5, MAGEA1/EGFP expression was slightly higher in DNMT1-depleted cells as compared with cells depleted of both DNMT3A and -3B or control cells (Fig. 3A). This difference became clearly significant at day 9 (Fig. 3A). However, even after 9 days of DNMT1 depletion, the level of expression of the transgene in MZ2-MEL.Tr4 was 24 times lower than that in MZ2-MEL cells that were transfected with the unmethylated pMAGEA1/EGFP plasmid. This suggested that treatment with DNMT1-AS oligos either resulted in activation of the transgene in only a fraction of the cells or induced only partial derepression of the transgene in most cells. Nevertheless, the results indicated that DNMT1 is involved in maintaining repression of the MAGEA1/EGFP transgene.
To test whether DNMT1 exerts the same function on an endogenous cancer-germ line gene, we analyzed the effect of DNMT1 and DNMT3Aϩ3B depletion on the activation of gene LAGE-2 NY-ESO-1 , which is naturally repressed in MZ2-MEL cells. Like the MAGEA1/ EGFP transgene, LAGE-2 NY-ESO-1 became activated only in DNMT1depleted MZ2-MEL.Tr4 cells (Fig. 3B).
Minimal Demethylation of the MAGEA1/EGFP Transgene following DNMT1 Depletion-We next examined whether activation of the MAGEA1/EGFP transgene following depletion of DNMT1 correlated with demethylation of the promoter region. To this end, DNA was extracted from MZ2-MEL.Tr4 cells that were either untreated or treated for 9 days with DNMT1-AS or DNMT3A-AS plus DNMT3B-AS oligos. We applied the bisulfite genomic sequencing method to analyze the methylation status of 16 CpG sites located between position Ϫ81 and ϩ161 relative to the transcription start site. Demethylation within this region was previously shown to be critical for transcription of the MAGE-A1 gene (7). MAGE-A1 sequences originating from the transgene could be specifically amplified, as they differ from the endogenous MAGE-A1 sequence by the presence of a short tag sequence inserted in the 5Ј-part of intron 1 (Fig. 1A). As expected, the MAGEA1/EGFP transgene was mostly methylated in untreated MZ2-MEL.Tr4 cells (Fig. 3C). A similar level of methylation was observed in cells depleted of DNMT3A and -3B, which was consistent with the lack of transgene activation observed in this group of cells (Fig. 3C). Surprisingly, the MAGEA1/EGFP transgene showed only limited, albeit significant (p Ͻ0.001), demethylation in DNMT1-depleted cells. Indeed, in these cells only 2 sequences of the 15 analyzed displayed Ͼ50% unmethylated CpGs (Fig. 3C). Although these data confirmed that DNMT1 is involved in methylation maintenance within the MAGEA1/EGFP transgene, they indicated that exposure to the DNMT1-AS oligo does not lead to an efficient demethylation of the transgene in most cells.

MAGEA1/EGFP Activation Is Not Increased by Simultaneous
Inhibition of DNMT1 and DNMT3A and/or -3B-We investigated the possibility that the low level of transgene demethylation and activation that we observed following DNMT1 depletion was due to the presence of DNMT3A and DNMT3B. These two enzymes might indeed exert their de novo methylation activity to rapidly remethylate the CpG sites that lost methylation by lack of the maintenance methyltransferase DNMT1. Previous studies reported that simultaneous depletion of DNMT1 and DNMT3B in tumor cell lines results in increased gene demethylation and activation, as compared with depletion of DNMT1 alone (36,37).
MZ2-MEL.Tr4 cells that had been exposed to the DNMT1-AS oligo for 9 days were divided into four groups and treated during four additional days with either DNMT1-AS alone, DNMT1-AS and DNMT3A-AS, DNMT1-AS and DNMT3B-AS, or a mixture of all three DNMT-AS oligos (Fig. 4A). The effect of these treatments on derepression of the transgene was then evaluated by measuring the level of MAGEA1/EGFP expression. As shown in Fig. 4B, additional inhibition of DNMT3A and/or DNMT3B in cells already depleted of DNMT1 did not increase the level of activation of the MAGEA1/EGFP transgene. We observed similar results for the endogenous LAGE-2 NY-ESO-1 gene (Fig. 4B). It is therefore likely that the low level of demethylation induced by DNMT1-AS treatment is due to insufficient depletion of DNMT1 in most cells rather than remethylation by DNMT3A and DNMT3B. Additionally, cells in which DNMT1 is drastically reduced might be counter selected, as it was previously demonstrated that loss of DNMT1 induces both cell cycle arrest and apoptosis (5, 38, 39).  Fig. 1A were cloned, and 15 clones were sequenced for each group. Shaded squares correspond to methylated cytosines, and empty squares represent unmethylated cytosines. The percent of unmethylated CpGs is indicated.

Selection of Cells in Which the Transgene Was Activated by DNMT1
Depletion-We next searched to isolate a subset of cells in which treatment with DNMT1-AS resulted in demethylation and activation of the transgene. Because MAGEA1/EGFP encodes a fusion protein comprising the green fluorescent protein, we initially tried to select cells in which the transgene was activated by flow-cytometric sorting of fluorescent cells. We were unable, however, to detect cells showing fluorescence above the background (data not shown). We then turned to another selection strategy and produced a MAGE-A1 construct carrying the bacterial hygromycin resistance coding sequence hph within its last exon (Fig. 5A). This construct, pMAGEA1/hph, was methylated in vitro and transfected into MZ2-MEL cells. From the transfected population, we selected a cell clone, referred as MZ2-MEL.TrHM, containing a stably repressed pMAGEA1/hph transgene that could be activated following treatment with demethylating agents. MZ2-MEL.TrHM cells were exposed to DNMT1-AS oligos during 7 days (Fig. 5B). This resulted in ϳ80% reduction in DNMT1 mRNA amount (Fig. 5C) and in moderate but significant activation of the MAGEA1/hph transcript (Fig.  5D). Bisulfite sequencing indicated that this activation was associated with partial demethylation of the transgene promoter, as evidenced by loss of Ͼ50% methylation in 5 of 14 sequences (Fig. 5E). After 7 days of treatment, DNMT1-depleted and control cells (1 ϫ 10 6 ) were transferred into medium containing hygromycin. Whereas all control cells died after 9 days of selection, cells that had been treated with the DNMT1-AS oligos showed emergence of a hygromycin-resistant population (Fig. 5B). Parallel experiments performed in limiting dilution conditions indicated that the proportion of cells surviving hygromycin selection represented 0.6% of the DNMT1-depleted cells. The hygromycin-resistant cell population showed a higher level of MAGEA1/hph expression as compared with the unselected cells (Fig. 5D). Consistently, bisulfite sequencing showed that hygromycin selection resulted in the enrichment of cells containing an unmethylated transgene, as all sequences originating from the hygromycin-resistant cell population displayed between 50 and 100% unmethylated CpGs (Fig. 5E). The expression of DNMT1 in the hygromycin-resistant population returned to a level comparable with that before treatment with DNMT1-AS oligos (Fig. 5C). Altogether, these data indicate that DNMT1 depletion resulted in transgene activation in a fraction of the cells that could be selected through the hygromycin resistance conferred by the activated transgene.
Stable Activation of MAGEA1/hph Transgene following Transient Inhibition of DNMT1-The selection of a cell population in which the MAGEA1/hph transgene has been activated following down-regulation of DNMT1 allowed us to evaluate whether hypomethylation and activation of the transgene would be maintained in the absence of further inhibition of methylation. To this end, twelve clones were derived from the hygromycin-resistant MZ2-MEL.TrHM cell population. These clones were expanded in medium lacking hygromycin, to avoid selection of MAGEA1/hph-expressing cells, and RNA was extracted after 23-29 days of culture. RT-PCR analysis indicated that MAGEA1/hph expression was still present in all clones at these days (Fig. 6A). To further assess maintenance of the MAGEA1/hph active status, three clones were cultured during 5 more weeks in medium lacking hygromycin. As shown in Fig. 6B, even after this extended period of culture, all three clones maintained expression of the transgene. Consistently, bisulfite sequencing on two of these clones revealed that hypomethylation of the transgene promoter was completely maintained over time (Fig. 6C). Altogether, these results indicate that stable hypomethylation and activation of the MAGE-A1 promoter can be acquired following a past event of demethylation.

DISCUSSION
Many human tumors show hypomethylation of their genome, but the mechanisms leading to this epigenetic alteration remain largely unknown. To address this issue, we used gene MAGE-A1 as a model, because the promoter region of this gene is a frequent target of DNA demethylation in cancer cells. This is the case in melanoma cell line MZ2-MEL, where MAGE-A1 is hypomethylated and active. We showed in previous transfection experiments that in vitro methylated MAGE-A1 transgenes do not become demethylated after integration into MZ2-MEL cells, indicating that these cells have not acquired a permanent demethylating activity targeted to the gene (29). Yet these experiments did not exclude the possibility that MZ2-MEL cells FIGURE 5. Selection of cells in which DNMT1 depletion resulted in transgene activation. A, the insert in the pMAGEA1/hph plasmid is represented as in Fig. 1A. B, schematic outline of the experiment. MZ2-MEL.TrHM cells containing a methylated pMAGEA1/hph plasmid were repeatedly treated with Lipofectin only (empty arrows) or with the DNMT1-AS oligonucleotide (black arrows). After 7 days of this treatment, 10 6 cells from each group were transferred into medium containing hygromycin (shaded area). DNA and RNA were extracted from the two groups of cells at day 7 and from the hygromycin-resistant population at day 16. C, DNMT1 expression levels in these samples were quantified by quantitative RT-PCR and are expressed relative to the level in Lipofectin-only control cells, taken as 100% reference. The data derive from two independent RT-PCR analyses. D, the same samples were tested by RT-PCR for the expression of MAGEA1/hph and ␤-ACTIN (as an internal control). E, methylation patterns (represented as in Fig. 3C) within the promoter region of the MAGEA1/hph transgene were analyzed in control cells and in DNMT1-AS-treated cells either before or after hygromycin selection.
undergo a permanent demethylation process of lesser intensity, leading to transgene demethylation in a minor (and therefore undetected) fraction of the cells. The present study suggests that this is not the case, because we did not detect any cell showing spontaneous activation of the methylated MAGEA1/hph transgene despite the selective pressure imposed by the presence of hygromycin in the medium. We conclude therefore that MAGE-A1 hypomethylation and activation in MZ2-MEL results from a past event of demethylation. It is important to note that MAGE-A1 activation was already observed in a metastatic tissue sample of patient MZ2 (40), indicating that the demethylation process occurred during tumor progression in vivo and is not a consequence of in vitro culturing.
Lack of a persistent demethylating activity seems to be a common feature of tumor cells, as evidenced by the observation that hypomethylated tumor cell lines usually do not show further demethylation of their genome during culturing (41)(42)(43). Instead, tumor cell lines often show slow but consistent de novo methylation activity (44,45). Consistently, several reports indicate that the level of expression of the DNA methyltransferases in tumor cells is similar or even higher than that in normal cells (46,47).
The involvement of an historical process of demethylation in the activation of MAGE-A1 requires that hypomethylation is maintained within the promoter region of the gene after the end of the transient demethylation phase. In support of this, previous transfection experiments revealed that MZ2-MEL cells contain transcription factors capable of inhibiting de novo methylation within the promoter region of MAGE-A1 transgenes that were unmethylated before transfection. The other parts of the transgene were progressively remethylated, reaching ϳ40% methylation 6 weeks after transfection. This indicates that MZ2-MEL cells possess significant de novo methylating activity (29). We have now shown that, following induction of a transient demethylation phase, previously methylated MAGE-A1 transgenes also become permanently protected against remethylation. These results confirm that in MZ2-MEL cells a transient phase of DNA demethylation suffices to convert a methylated and repressed MAGE-A1 gene into a permanently hypomethylated and active one.
Based on our observations we propose the following model of activation of MAGE-A1 in tumors. During the neoplastic process, the cancerous cells would undergo a transient phase of epigenetic instability leading to demethylation of the MAGE-A1 gene. Transcriptional activators would then be able to bind to the unmethylated promoter region. In cells containing factors that exert a sufficient level of transcriptional activation to protect the region against remethylation, the promoter of MAGE-A1 would be maintained unmethylated, even if at later stages the methylation/demethylation equilibrium shifts back toward de novo methylation activity. This model may explain how hypomethylation in tumors is maintained despite increased methylation activities and how hypomethylated and hypermethylated sequences can coexist.
The present study also investigated the involvement of the different DNMTs in the methylation maintenance of cancer-germ line gene promoters. The contribution of DNMT1 and DNMT3B to methylation maintenance in tumor cells has been a disputed issue. Several reports indicate that inhibition of DNMT1 in bladder (T24) and colon (HCT116) cancer cell lines suffices to induce both gene-specific and global DNA demethylation (31,48). In contrast, other reports suggest that in tumor cells other DNA methyltransferases may participate in the methylation maintenance process. In HCT116 cells for instance, disruption of the DNMT1 gene resulted in only minor DNA demethylation, whereas disruption of both DNMT1 and DNMT3B resulted in severe loss of methylation (37,49,50). In an ovarian cancer cell line, Leu et al. (36) showed that maximal demethylation of several hypermethylated tumor suppressor genes was obtained only when both DNMT1 and DNMT3B were inhibited. We have shown here that in MZ2-MEL cells depletion of only DNMT1 suffices to induce demethylation and activa-  (Fig. 2B), we derived twelve clones and cultured them in the absence of further drug selection. A, RT-PCR was used to assess MAGEA1/hph expression in these clones at days 23-29 from drug removal. Controls include MZ2-MEL.TrHM cells treated with Lipofectin only (Lipo only) and the DNMT1-AS-treated cell population that survived after 9 days of hygromycin selection (7d DNMT1-AS ϩ 9d Hygro). B, for three of the clones (1, 4, and 8), the same RT-PCR analysis was performed after an extended period of culture in the absence of hygromycin (60 days). C, methylation patterns (represented as in Fig. 3C) within the promoter region of the MAGEA1/hph transgene were evaluated in two clones (1 and 4) at two time points after removal of hygromycin. tion of MAGE-A1 transgenes and of the endogenous cancer-germ line gene LAGE-2 NY-ESO-1 . It is worth noting that DNMT1 depletion also sufficed to induce reactivation of the ASC/TMS1 tumor suppressor gene (51), which we found to be silenced and hypermethylated in MZ2-MEL cells (data not shown). DNMT1 appears therefore to have a predominant role for methylation maintenance in MZ2-MEL cells. Altogether, these observations indicate that the relative contribution of the different DNA methyltransferases to methylation maintenance varies from one tumor cell line to another. This may depend on the relative abundance of the different DNMTs, as suggested by the low level DNMT3A and DNMT3B expression in MZ2-MEL cells.
Following transient depletion of DNMT1, Ͻ1% of MZ2-MEL.TrHM cells, which contain a selectable pMAGEA1/hph transgene, survived in medium containing hygromycin. Two factors may contribute to this low level of recovery of cells with an unmethylated and active MAGEA1/ hph transgene. First, it appears that, despite the efficient inhibition (ϳ80%) of DNMT1 that we observed after antisense treatment, the enzyme was insufficiently depleted in most cells to allow transgene demethylation. This is suggested by the bisulfite sequencing results showing that after 7 days of treatment with DNMT1-AS oligonucleotides, only 36% of the transgene promoter sequences had more than 50% unmethylated CpGs (Fig. 5E). Secondly, it is likely that cells in which DNMT1 was drastically reduced, and therefore where the transgene was demethylated, were counter-selected during culturing. It has indeed been reported that loss of DNMT1 induces both cell cycle arrest and apoptosis (5,38,39).
The MZ2-MEL.TrHM cell clone, which contains a methylated MAGEA1/hph transgene conferring resistance to hygromycin upon reactivation, proved to be a powerful system to select cells in which demethylation and activation of the transgene occurred. This cell clone will be used to assess the contribution of several other factors to DNA methylation maintenance in MZ2-MEL cells. Candidate factors include methyl-CpG-binding proteins and chromatin-modifying enzymes such as histone deacetylases and histone methyltransferases, which appear to participate along with DNA methyltransferases to the maintenance of transcriptionally repressed domains (52). Identifying the different factors involved in methylation maintenance should help us to uncover the molecular mechanisms responsible for the DNA demethylation process in tumors.