INTRODUCTION

Top Abstract Introduction Materials and methods Results Discussion References

CpG methylation plays an important role in mammalian development (31; for a review, see reference 2), and it has been correlated with repression of gene expression (22; for reviews, see reference 4 and 33). Changes in the basic pattern of de novo methylation and demethylation occur throughout development. How DNA regions are targeted for de novo methylation or demethylation and how the de novo methylation and demethylation processes are mediated are not clear.

It has been hypothesized that protein factors may protect CpG islands from methyltransferase (1). However, the data to support this has been indirect. In vivo footprinting and DNA methylation studies on the phosphoglycerate kinase 1 (PGK-1) gene indicated protein-DNA contact in the promoter region of this gene on the active X chromosome but not on the inactive X chromosome (29, 30). It has been proposed by these authors that unidentified protein factors may be involved in keeping specific regions methylation free. In a second system, studies with transgenic mice have indicated that DNA sequences corresponding to Sp1 sites play an important role in protecting the CpG island of the adenine phosphoribosyltransferase (APRT) gene from methyltransferase (3, 24). However, a recent study (26) demonstrated the methylation-free status of the APRT gene in Sp1 knockout mice, and the authors suggested that the methylation-free status may be maintained by the binding of other members, such as Sp3, of the Sp1 family to the Sp1 sites. Hence, the identity of protein factors that might maintain sites of demethylation in this system remains uncertain. Study of the -actin gene in myoblasts by using transient, nonreplicating plasmids indicated that demethylation of one DNA strand at a specific site occurs within 2 h after DNA enters the cells (28). Although it was not demonstrated directly in their study, Paroush et al. (28) suggested that demethylation may involve protein-DNA interactions based on the specificity of the demethylated sites.

Two possible mechanisms for demethylation have been proposed in the above studies. In the first, a passive mechanism, demethylation is due to the failure of remethylation by maintenance methyltransferase, and hence demethylation of both DNA strands should occur in 50% of the cells after two rounds of replication (32). At least four or five rounds of replication would be required to demethylate about 95% of the DNA on both strands by this mechanism. In the second mechanism, an active mechanism, activity of a demethylase that may require cis- and trans-acting factors (reviewed in reference 38) can lead to demethylation of both strands on all the DNA without any extensive DNA replication. The hemimethylated chicken vitellogenin gene becomes symmetrically demethylated with limited DNA replication at 24 h after hormone stimulation in chicken liver (34). Transient transfection of the -actin promoter upstream of a reporter gene into a rat cell line indicated that demethylation is a two-step process with a hemimethylated intermediate (28). Studies on -crystallin demethylation suggested possible hemimethylated intermediates (10, 36). However, after several rounds of DNA replication, no hemimethylated DNA could be detected by a sensitive PCR assay in an experiment in which in vitro-methylated sequences were injected into mouse zygotes (21). These studies support the hypothesis that demethylation is a multistep process and that a passive mechanism alone without an active mechanism participating at least in some steps cannot achieve complete demethylation.

Considering that demethylation of the two DNA strands may involve different (distinguishable consecutive) mechanisms, four possible mechanisms can be postulated for the demethylation of the two DNA strands (Fig. 1): (i) protein binding could recruit a demethylase to demethylate both strands (active-active mechanism); (ii) protein binding could interfere with remethylation after replication, leading to demethylation of both strands on half of the DNA molecules after two rounds of replication (passive-passive mechanism); (iii) protein binding could recruit demethylase to demethylate one DNA strand, and then the first round of replication would lead to demethylation of both strands on half of the molecules (active-passive mechanism); and (iv) protein binding could interfere with remethylation after replication, and then the resulting hemimethylated DNA would recruit a demethylase to demethylate the second DNA strand (passive-active mechanism). While any one of these conceivable pathways may turn out to be the predominant one, it is conceivable that more than one may function during the course of cell differentiation and development at different DNA sites in different cell types.

View larger version (35K): [in this window] [in a new window]   FIG. 1.   Possible mechanisms of demethylation at specific sites. The term "passive" is used to indicate that the specific protein, such as EBNA-1, binds to its recognition sites and blocks remethylation at these sites by the maintenance methyltransferase after DNA replication. These sites become hemimethylated instead of being restored to the symmetrically methylated configuration after one round of DNA replication. The term "active" is used to indicate that a demethylase removes methyl C independently of any DNA replication. The ovals represent proteins, such as EBNA-1, binding to a specific DNA sequence. Thick lines represent methylated DNA, and thin lines represent unmethylated DNA.

It is well documented that plasmids bearing the latent replication origin of the Epstein-Barr virus (EBV), oriP, can be maintained in human cells expressing the EBV nuclear antigen EBNA-1 (17, 35, 40, 41). We have developed a stable episomal system in which oriP is used to study the dynamics of CpG methylation over time courses of several months in human cells (13). In our system, the CpG methylation patterns on the plasmids generated in vitro by using either FnuDII, HhaI, HpaII, or SssI methylase are maintained for months after transfection into human cells. This indicates that the maintenance methyltransferase can efficiently remethylate the newly synthesized strand at positions opposite the existing sites of CpG methylation. However, we have observed a few specific preferential demethylation sites on the episome. Among these few sites, three are HpaII sites in the oriP region that become demethylated very quickly after transfection into human cells expressing EBNA-1. The demethylation of these HpaII sites appears to be either simultaneous or close to simultaneous.

In the current study, the sites of demethylation in oriP were mapped by using both Southern blot analysis and bisulfite genomic sequencing methods. Furthermore, experiments were designed to explore the mechanism of demethylation in this region. We found that protein binding is required for demethylation of the oriP region. Replication alone does not lead to demethylation of the oriP without EBNA-1 binding. Furthermore, EBNA-1 binding alone also does not lead to demethylation of the oriP before replication. The HpaII sites within the EBNA-1 binding sites remained at least hemimethylated after one round of DNA replication. Moreover, the ability to identify molecules that were not replicated or were replicated once or twice in this system allows the analysis of the steps and mechanism of demethylation in the oriP region. The first-strand demethylation of the oriP region occurs by a passive mechanism, and the second-strand demethylation of these sites is probably processed through an active mechanism. This mechanism offers a logical interpretation of observations on demethylation events of endogenous genes.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and methods Results Discussion References

Plasmids. Plasmids with wild-type oriP sequences used in this study include pCLH22, p291, p291, and pHEBo. These plasmids can replicate once per cell cycle in human cells expressing EBNA-1. pCLH22 (13) contains oriP, EBNA-1, the luciferase reporter gene, the hygromycin resistance gene, and the necessary prokaryotic replication sequences. p291 has the simian virus 40 replication origin in addition to the pCLH22 sequences, but it does not have the luciferase reporter gene. pHEBo (35) has only oriP, the hygromycin resistance gene, and the necessary prokaryotic replication sequences. p291 was constructed by deleting the HindIII fragment containing the EBNA-1 coding sequence from p291. Several plasmids with defective oriP, dpm1, dpm1+2, and dpm3+4, were used in this study. These plasmids were a generous gift from J. Hearing. They have mutations in the dyad-symmetry (DS) region and were characterized previously (11). In brief, dpm1 has two nucleotides mutated in EBNA-1 binding site 1, dpm1+2 has two nucleotides mutated in each of the EBNA-1 binding sites 1 and 2, and dpm3+4 has two mutations in each of the EBNA-1 binding sites 3 and 4. All mutations, except one in site 3, generate new CpG sites in these binding sites.

In vitro DNA methylation. DNA was methylated with SssI methylase, which methylates C's at all CpG sites. The conditions used were those recommended by the manufacturer (New England Biolabs). DNA was extracted with phenol-chloroform and precipitated with ethanol after in vitro methylation. The status of methylation was confirmed by digestion with methylation-sensitive restriction endonucleases.

Cell lines and transfection. 293, a human embryonic kidney carcinoma cell line (American Type Culture Collection), and a derivative of this cell line, 293/EBNA1 (13), were used in this study. A prostate cancer cell line, PC-3, and a derivative, PC-3/EBNA1, with an integrated EBNA-1 gene driven by the cytomegalovirus promoter were also used. The EBNA-1 protein is constitutively expressed in the 293/EBNA1 and the PC-3/EBNA1 cell lines. Throughout this study, the calcium phosphate transfection method (13, 39) was used. All transfections were done in duplicate in each experiment, and all experiments were performed multiple times for confirmation.

Episome recovery and analysis. In the cases where the plasmids were able to replicate, when the transfected cells reached confluence 2.5% of the cells were replated into a 100-mm plate and the remaining cells were harvested for plasmid DNA extraction. No replating was done in cases where the plasmids were not able to replicate. All the transfection experiments were carried out without any selection for the episomal plasmid.

Bisulfite genomic sequencing. Bisulfite genomic sequencing was carried out by the method of Clark et al. (5) with minor modifications. Of the DNA harvested from each transfection, 30% was used for bisulfite genomic sequencing. DNA was digested with MboI and denatured before being treated with a final concentration of 2.3 M sodium bisulfate-0.5 mM hydroquinone at 55°C for 4 h. The bisulfite-treated DNA was purified with the Wizard DNA purification resin (Promega), treated with a final concentration of 0.3 N sodium hydroxide, and precipitated with ethanol. The primers for top-strand amplification were 5'-GTGATAGTTTATGGGGTGGGA (forward) and 5'-CAATCAAAAAAACCTATATAACTAC (reverse). The PCR conditions for the top strand were 3 min at 95°C for initial denaturation followed by 35 cycles of 40 s at 94°C, 40 s at 55°C, and 40 s at 72°C in a Robo 96 cycler (Stratagene). The primers for bottom-strand amplification were 5'-ATAACAACTCATAAAATAAAAAATAT (forward) and 5'-TTAATTAGAGGGGTTTGTGTAG (reverse). The PCR conditions were similar to those for the top-strand amplification but with a reannealing temperature of 52°C. The 245-bp PCR products were gel purified and cloned into T-vector (25) made from pBSK. The clones were sequenced with the ABI PRISM Dye terminator cycle sequencing kit (Perkin-Elmer).

	RESULTS

Top Abstract Introduction Materials and methods Results Discussion References

Demethylation occurs at the HpaII sites within the oriP region. After the SssI-methylated pCLH22 was transfected into 293/EBNA1 cells, the DNA was harvested at various times and digested with HpaII. This 12.1-kb plasmid has the hygromycin resistance gene, the EBNA-1 gene, the luciferase gene, and the oriP segment in addition to the prokaryotic replication sequences and a selectable marker. Stable maintenance of the methylation status at nearly all CpG sites over at least a 2-month interval has been routinely observed in previous experiments (13, 14). However, several specific demethylation sites were observed within a few days after transfection in a large fraction of the minichromosome population (Fig. 2B). By using region-specific probes, these HpaII sites were localized to the oriP region. Three of the four HpaII sites in the oriP region were demethylated on the minichromosome (Fig. 2B). Demethylation at these HpaII sites was also observed when a different plasmid, p291, was used for the same experiment (data not shown). These three HpaII sites are known to remain unmethylated, while the remaining viral DNA is heavily methylated in tumor cells carrying the entire EBV episome (8). Our findings demonstrate that demethylation occurs at specific sites on the minichromosome in human cells and that the demethylation is unique to oriP and is typical of any plasmid that has oriP. Furthermore, the site-specific demethylation appears to be the same for the minichromosome as for the intact EBV viral genome.

View larger version (32K): [in this window] [in a new window]   FIG. 2.   Demethylation at specific HpaII sites within oriP. (A) HpaII and HhaI sites in the oriP region. The numbers below the HpaII sites are the nucleotide numbers of the HpaII sites on pCLH22. Three HhaI sites in the region are indicated by H. The asterisks indicate the sites of demethylation. (B) Southern blot of DNA harvested 8 days after transfection. The DNA is digested with different enzymes, as indicated above each lane. The probe used is indicated in panel A. The plasmid DNA is linearized by XbaI in the lanes indicated. The 12.1-kb band in the left panel is linearized plasmid. The 9.5-kb band in the right panel is the plasmid backbone hybridized with the portion of the probe upstream from the HpaII site at position 10135. The digestion of oriP by HpaII and the site of linearization resulted in the smaller size of this fragment. The 1.3-kb and 326-bp fragments in the right panel are generated by the digestion at three HpaII sites located at positions 10135, 10461, and 11773. The HpaII site at position 10857 remained methylated and HpaII resistant.

Demethylation within the oriP region is not due to the DNA sequence alone. To determine whether the oriP sequence alone without replication or EBNA-1 binding can lead to demethylation of this region, plasmid p291 was methylated with SssI and then transfected into either 293 or 293/EBNA1 cells for comparison. Plasmid p291 will not replicate in 293 cells because of the absence of EBNA-1, but it will replicate in the 293/EBNA-1 cells. Demethylation of the HpaII sites within the oriP region did not occur when p291 DNA was harvested from 293 cells 3 days after transfection (Fig. 3). This strongly suggests that the oriP sequence is not targeted by an active demethylation process without other factors. Demethylation clearly took place in the 293/EBNA1 cells, where replication can occur, during the same time interval (Fig. 3). This indicates that demethylation in the oriP region does not occur without some aspects of the replication process (EBNA-1 binding [6, 15]), bending of the DNA at the origin [6, 15]), and synthesis of a new DNA strand). Therefore, neither the sequence of oriP nor any structural features intrinsic to its repetitive nature leads to demethylation.

View larger version (26K): [in this window] [in a new window]   FIG. 3.   Demethylation does not occur without EBNA-1 binding and DNA replication. A Southern blot of DNA harvested 3 days after transfection and digested with HindIII (for linearization) and HpaII is shown. The plasmids and cell lines used in the experiments are indicated above the panel. 293E, 293/EBNA1 cell line. The capability of plasmid replication and the presence of EBNA-1 in the cells are also indicated above each lane. The probe used is the NsiI-HpaI fragment, as indicated. For an explanation of the outline letters, see the legend to Fig. 4. The asterisk designates the HpaII site from which the partial digestion product likely derives.

Replication alone without EBNA-1 binding does not lead to demethylation. To examine whether EBNA-1 binding plays a role in the demethylation in the oriP region, several mutants with defective EBNA-1 binding sites in the DS region were used. Of the four EBNA-1 binding sites in the DS region, only one pair is required for oriP function (11). The mutant plasmid dpm1 has two point mutations in EBNA-1 binding site 1 in the DS region, and these mutations greatly reduce the binding of EBNA-1 to both sites 1 and 2. Plasmid dpm1+2 has two point mutations in EBNA-1 binding site 2 in addition to dpm1. This plasmid has been shown to replicate in mammalian cells, while the EBNA-1 protein binding to sites 1 and 2 of the DS region is abrogated, as shown in DNase I protection assays (11). Both plasmids dpm1 and dpm1+2 can replicate by using sites 3 and 4 (11). Plasmid dpm3+4 has EBNA-1 binding sites 3 and 4 mutated in the DS region, but it replicates in human cells by using the wild-type EBNA-1 binding sites 1 and 2.

View larger version (40K): [in this window] [in a new window]   FIG. 4.   Demethylation does not occur without EBNA-1 binding. (A) The diagram indicates the HpaII sites in the oriP region and the HpaII fragment sizes. Outline letters indicate the HpaII site that does not become demethylated and therefore is not digested by HpaII. All the HpaII sites can be digested by a CpG methylation insensitive enzyme, MspI, which is an isoschizomer of HpaII. (B) Southern blot of DNA harvested 10 days after transfection. All of the DNA was linearized by EcoRI digestion before being digested with HpaII (H) or MspI (M). The plasmids used are indicated above the lanes. The 1.6-kb HpaII band and the 727-bp MspI band are most likely to be derived from partial digestion of the HpaII-MspI site indicated by the asterisk in panel A. dpm1, dpm1+2, and dpm3+4 have mutations in the EBNA-1 binding sites in the DS region, as described in Materials and Methods and illustrated in Fig. 6A.

View larger version (25K): [in this window] [in a new window]   FIG. 5.   Quantitation of demethylation in oriP. The ratio of radioactivity in the 1.3-kb band (including the 1.6-kb partial-digestion band) to the combined radioactivity in the 7-, 1.3-, and 1.6-kb bands in the HpaII-digested DNA was calculated. Percentages listed in the "demethylation" column represent the fraction of plasmids having undergone demethylation in the HpaII site marked with an asterisk between EBNA-1 binding sites 1 and 2 in the dyad symmetry (see the text for details). The capability of EBNA-1 binding and plasmid replication in eukaryotic cells are indicated: +/, weak binding; ++, strong binding. The mutated EBNA-1 binding sites in each plasmid are indicated by X. Plasmids dpm1, dpm1+2, and dpm3+4 have mutations in the EBNA-1 binding sites in the DS region, as described in Materials and Methods and illustrated in Fig. 6A.

Demethylation of oriP is site specific and is not regionally specific. It is intriguing that one of the four HpaII sites within oriP remains methylated. This indicates that demethylation may occur only at CpG sites protected by EBNA-1. Bisulfite genomic sequencing of plasmid DNA harvested from 293/EBNA1 cells transfected with dpm1, dpm1+2, dpm3+4, and pHEBo was carried out to examine the CpG sites within and adjacent to the DS region. The 245-bp sequence examined includes seven CpG sites in the wild-type plasmid, pHEBo (Fig. 6A). Four of these seven sites are located within the EBNA-1 binding sites. There are 9, 11, and 10 CpG sites in dpm1, dpm1+2, and dpm3+4, respectively. The extra CpG sites in these mutants were generated by site-directed mutagenesis in the EBNA-1 binding sites within the DS region (11).

View larger version (21K): [in this window] [in a new window]   FIG. 6.   Specific CpG site demethylation in the DS region. (A) Nucleotide sequence of the oriP region analyzed by bisulfite genomic sequencing. The wild-type sequence is presented. The mutations in plasmids dpm1, dpm1+2, and dpm3+4 are indicated under the wild-type sequence. The CpG sites are underlined and indicated by a circled number. Brackets indicate sites protected by EBNA-1 in the in vitro DNase I protection assay described by Harrison et al. (11). (B) The CpG sites illustrated in panel A were examined by bisulfite genomic sequencing. The CpG sites in the four EBNA-1 binding sites are indicated by the brackets above the line. , no conversion by sodium bisulfite treatment (with, the CpG site thus remaining methylated); , conversion by sodium bisulfite treatment (with, the CpG site becoming demethylated). Some CpG sites exist only in the mutant plasmids, and circles are absent at these sites in other plasmids. Multiple clones (at least four) from each mutant of each experiment were sequenced, and sequences of the different clones were identical within each mutant. The DNA was harvested 10 days after transfection.

Demethylation of the first DNA strand in the EBNA-1 binding sites involves a passive mechanism. The sequence of oriP alone or DNA replication alone does not lead to demethylation of the EBNA-1 binding sites, as described above. To dissect the possible mechanism underlying the demethylation process, SssI-methylated and unmethylated pCLH22 plasmid was transfected into 293/EBNA1 cells and the DNA was harvested from isolated nuclei at various time points for analysis. In an experiment to determine how quickly the DNA enters the nucleus, DNA was harvested from isolated nuclei at 6 and 12.5 h after transfection. The DNA could be recovered from the nuclei within 6 h after transfection.

View larger version (57K): [in this window] [in a new window]   FIG. 7.   Replication precedes demethylation of the HpaII sites in the oriP region. A Southern blot of DNA harvested 7, 15, 19.5, 24, 40, and 66 h after transfection and digested with DpnI (D) or HpaII (H) is shown. The same probe used in the experiment in Fig. 3 that contains only the oriP region is used here. (A) DNA from cells transfected with unmethylated pCLH22; (B) DNA from cells transfected with SssI-methylated pCLH22 (me-pCLH22). The thin and thick lines above the time designation in panel B represent the possible methylation states within oriP on molecules replicated zero, one, or two times. Thick lines represent methylated DNA strands, and thin lines represent unmethylated DNA strands. The open circle on one end of the line indicates DNA that retains the bacterial dam methylation (at A of GATC). The DpnI-resistant DNA is detected in the DNA harvested at 40 and 66 h after transfection but not at any earlier time points. Plasmid DNA was not linearized in this experiment; therefore, the uncut DNA appears as nicked and supercoiled bands as indicated. The pCLH22 DNA can be digested by HpaII because it is not methylated. A 916-bp band and the smaller 396- and 326-bp, bands which appear as one band, are detected in the HpaII-digested pCLH22 DNA. The methylated pCLH22 DNA is not digestable by HpaII unless demethylation occurs. In the methylated pCLH22 DNA harvested at or before 40 h after transfection, the DNA remains uncut by HpaII. A 1.3-kb band is clearly detected in the methylated pCLH22 DNA harvested at 66 h after transfection. There are two DpnI-MboI sites within the oriP region that generate three complete digest bands at 2.8 kb, 1.0 kb, and 763 bp. Loss of dam methylation at specific sites on some unreplicated molecules, for unknown reasons, generates a strong 3.5-kb band and a weak 1.4-kb band in the transfected DNA (both methylated and unmethylated) in addition to the three bands described above when probed with the oriP fragment.

Demethylation of the second DNA strand in the EBNA-1 binding sites most probably involves an active mechanism. Only plasmids which have lost CpG methylation on both DNA strands in the oriP region will be sensitive to HpaII digestion. If demethylation of the second DNA strand is accomplished by a passive mechanism, only half of the molecules that replicated twice would have lost CpG methylation on both DNA strands in the oriP region and would be digested by HpaII (Fig. 8A). In contrast, if demethylation of the second DNA strand involves an active mechanism, all molecules replicated twice and some molecules replicated once should become demethylated on both DNA strands within oriP; therefore, these molecules would be digestable by HpaII (Fig. 8A).

View larger version (35K): [in this window] [in a new window]   FIG. 8.   Quantitative analysis of second-strand demethylation. (A) The bacterial and CpG methylation status of molecules in the harvested DNA based on passive-passive and passive-active mechanisms. Thick lines represent CpG-methylated DNA, and thin lines represent DNA demethylated at CpG sites. The open circle on one end of the line indicates DNA that retains the bacterial dam methylation. S, sensitive to enzyme digestion; R, resistant to enzyme digestion. A, plasmids that have undergone no replication; B, plasmids that replicated once; C, plasmids that have replicated twice in plasmids harvested at 66 h after transfection. (B) The fraction of DNA that is DpnI resistant (DpnI-R), DpnI-MboI double resistant (DpnI/MboI-R), and HpaII-sensitive (HpaII-S) in the plasmids harvested at 66 h after transfection. The observed value is derived from quantitation of uncut bands divided by the total hybridization in each of the digests. The percentage of the total for HpaII-sensitive molecules is derived from prediction of methylation status based on each of the two mechanisms as summarized in panel A.

	DISCUSSION

Top Abstract Introduction Materials and methods Results Discussion References

The major findings in this study are as follows. First, there are specific preferential CpG demethylation sites within the oriP region that are independent of the cell line and of the sequences outside oriP. These sites appear to be demethylated at the same time. Second, the DNA sequence of oriP does not become demethylated without DNA replication and EBNA-1 binding. Third, EBNA-1 binding is required for demethylation at these specific sites. Fourth, demethylation in oriP is not regionally specific but is specified precisely by EBNA-1 binding to its sites. Fifth, EBNA-1 binding with one round of replication does not lead to double-strand demethylation of oriP within 40 h after transfection. Finally, demethylation of the first DNA strand involves a passive mechanism, and demethylation of the second DNA strand most probably involves an active demethylase activity.

It has been reported that the oriP region is unmethylated in the otherwise highly methylated EBV genome in the Burkitt's lymphoma cell line, RaeI (8). In most cells derived from Burkitt's lymphomas, EBNA-1 is the only viral protein expressed. It has not been clear whether oriP initially becomes methylated after viral entry and then becomes demethylated some time later or whether it never gets methylated from the start. The HpaII site in the spacer region between the FR and the DS region is unmethylated in the viral genomes from Burkitt's lymphoma and nasopharyngeal carcinoma cells (7, 8, 16). However, this very same HpaII site did not become demethylated in the present study. This suggests that the oriP region of the EBV genome does not become methylated in the first place in these tumor cells. This is because EBNA-1 binding is inadequate to lead to demethylation at this particular HpaII site if the site were ever to become methylated. It is most likely that EBNA-1 binding can also protect the DNA region from de novo methylation in addition to protecting specific sites from remethylation by the maintenance methylase. It is noteworthy that the region protected from de novo methylation appears to be larger than the region protected from maintenance methylation. This may explain the observation that Sp1 sites can prevent methylation of downstream CpG sites at the APRT gene during development (24).

Although protein-DNA interaction has been suggested to be important for DNA demethylation, this study provides direct evidence that protein binding can specify demethylation sites. It has been demonstrated in this study that demethylation does not take place when EBNA-1 is absent from the cells (therefore, no EBNA-1 binding and no plasmid DNA replication occur). This indicates that oriP demethylation is not specified by the DNA sequence alone. By using oriP mutants, it is clearly shown that the mutated EBNA-1 binding sites remain methylated after many rounds of minichromosome replication. This indicates that replication alone does not lead to demethylation without EBNA-1 binding. Otherwise, demethylation should occur at mutated EBNA-1 binding sites in oriP, regardless of the lack of EBNA-1 binding at those sites. The site of demethylation is specified strictly by EBNA-1 binding based on the observation that only the CpG sites within the EBNA-1 binding sites, not the CpG sites either adjacent to or between them, are demethylated. In summary, EBNA-1 is required for demethylation of the oriP region, and its binding specifies the sites of demethylation. This leaves open the question of mechanism; namely, how does the demethylation occur?

Although the critical role of EBNA-1 in oriP demethylation is clearly defined, the involvement of replication requires more complex analysis. It is difficult to clearly dissect EBNA-1 binding and DNA replication unless plasmid replication does not occur when high EBNA-1 expression and strong binding sites are present. Several experiments were performed to obtain EBNA-1 binding without plasmid replication in the cells. However, a low level of plasmid replication was always observed in the two cell lines, 293/EBNA1 and PC-3/EBNA1, used in this study (results not shown). This is not unexpected, because the use of alternative replication initiation sites has been reported for latent replication of the EBV genome (23). The fact that EBNA-1 binding sites are the components of the functional replication origin limits our ability to directly demonstrate the requirement of replication in demethylation. However, this very feature allows us to analyze the demethylation process in a stepwise manner on each DNA strand in the time course experiments. An entirely different system, containing a protein binding site that is not involved in replication initiation, is currently being developed to address the requirement of replication directly.

The demethylases described by Vairapandi and Duker (37) and Weiss et al. (38) demethylate both DNA strands within a short time (within 6 h) in cell-free enzymatic studies. If confirmed as physiologic demethylating activities, these two enzymes are most likely to be involved in an active-active mechanism (Fig. 1). Moreover, Paroush et al. (28) observed demethylation of the first strand within 2 h after DNA enters the cells. Therefore, demethylation of both strands (making DNA sites HpaII digestable) should be observed within a short period for plasmid entering the nucleus, if the active-active mechanism is operating. If the active-passive mechanism is operating, demethylation of the first DNA strand is most likely to occur much earlier than 40 h after transfection. Therefore, both DNA strands should become demethylated on 50% of the replicated plasmids after the first round of DNA replication while the other 50% of the replicated plasmids become hemimethylated at this time (Fig. 1).

The time course experiments in this study show that replication precedes demethylation of both DNA strands. This is based on the fact that no HpaII-digestable DNA was detected in the DNA harvested at several time points after transfection including the time point (40 h) when replicated DNA was first detected. This clearly demonstrates that although EBNA-1 binding specifies the demethylation sites, binding alone does not lead to demethylation on both DNA strands, at least within 40 h after transfection (34 h in the nucleus). Therefore, demethylation of oriP did not occur through either an active-active mechanism or an active-passive mechanism within 40 h after transfection; if it had, DNA demethylation on both strands should have been observed at this time. Although one can speculate that the active demethylation machinery takes more than 40 h to act on the DNA, this is highly unlikely, based on the findings in the studies referred to above. Moreover, the active demethylase should be able to demethylated both replicated and unreplicated molecules at these HpaII sites. The fact that unreplicated molecules do not become demethylated strongly supports the conclusion that the active mechanism is not involved in the first-strand demethylation in the oriP region and replication is required. This rules out the active-active and the active-passive mechanisms as being the mechanism of the oriP demethylation; hence, the double-strand demethylases described above are most probably not involved in the oriP demethylation.

This study allows us to analyze the order and nature of the steps in oriP demethylation. It provides a compelling indication that DNA replication is required for demethylation of the first DNA strand. The fraction of HpaII-sensitive molecules (38%) and the fraction of DpnI-resistant molecules (41%) are nearly equal in the DNA harvested at 66 h after transfection. This argues that the second DNA strand is demethylated by an active demethylase after the first-strand demethylation. Otherwise, the fraction of HpaII-sensitive molecules should have been much smaller than the observed value.

Observations in this study indicate that demethylation of the oriP is a two-step process. EBNA-1 specifies the sites of demethylation by binding to DNA and interfering with remethylation by the maintenance methylase after replication. This first step generates specific hemimethylated sites, and then the second strand is demethylated by an active demethylase. Interestingly, one of the reported active demethylases, 5-methylcytosine-DNA glycosylase (18, 19), prefers hemimethylated substrates in vitro. This enzyme may target hemimethylated sites specified by protein binding. This model may explain some of the demethylation events in the genome, such as demethylation of CpGs in the regulatory region of the avian vitellogenin gene in chicken liver (34). In vivo experiments that directly address whether protein binding is required for a demethylase to demethylate hemimethylated DNA are under way.

It has been reported that the 5-methylcytosine-DNA glycosylase requires RNA with at least four nucleotides of complementarity to the hemimethylated target (9, 20). One can speculate that transcription factors or other regulatory proteins bind to the regulatory region of genes that are being activated during development and that this leads to hemimethylation at this region after one round of replication and a low level of transcription. This region then becomes demethylated on both strands, with 5-methylcytosine-DNA glycosylase targeting the hemimethylated DNA and being stabilized or activated by the RNA synthesized at the site. Hence, hemimethylated sites without at least a low level of transcription will not be demethylated by this enzyme. Regardless of how the second step of demethylation occurs, the strength of protein binding may play a critical role in targeting specific sites by protecting them from the maintenance methylase in this demethylation process. The requirement of transcription factor binding and DNA replication for demethylation in Xenopus embryos was reported (27) while this paper was in preparation. The findings in the present study are mostly in agreement with the observations reported by Matsuo et al. (27). However, the present study goes considerably further in providing detailed analyses of the sites of demethylation, the role of EBNA-1 binding, the site specificity of demethylation, and the stepwise mechanism of demethylation in mammalian cells. Although the demethylation is clearly a two-step process, this two-step mechanism may be responsible for only a subset of demethylation events. However, nothing is currently known that precludes it from being the predominant and perhaps the only pathway.