Our experiments using ES cells with systematic knockouts of the three known active DNA methyltransferases have revealed a remarkable cooperativity in the activities of these enzymes in maintaining methylation patterns of specific sites during cell division. Studies using cell lysates (21
) have shown that all three enzymes are capable of methylating unmethylated and hemimethylated substrates, making it difficult to discern the relative contributions of each of the enzymes in pattern maintenance in living cells. Previous experiments using cells with disabled genes for the various enzymes have focused mainly on the methylation statuses of CpG islands in the various cell types. Since the majority of methylation occurs in CpG-depleted regions (5
), we have included analysis of CpG-poor regions in addition to the CpG-rich A-repeats for this study. Our results showed that Dnmt1 by itself is capable of maintaining the methylation status in some CpG-poor class I sequences with a fairly high degree of fidelity, but not at other sequences, which we named class II. Class II sequences seem to have a higher proportion of associated repetitive elements, but this distinction is not absolute, and they resemble Igf-2 and Xist in their behavior in M1 cells (20
). We performed additional studies to investigate the nature of the difference between class I and class II sequences in more detail by selecting a representative region from each of the two classes and analyzing the methylation patterns on individual molecules. In addition, we developed a novel assay to measure the level of hemimethylation at these regions and also analyzed the kinetics of methylation at these sites following DNA replication.
Methylation of the class II retroviral sequence CII-d and the type A repeats in the LINE elements was very poorly maintained by Dnmt1 alone, or by Dnmt3a and Dnmt3b without Dnmt1. Nevertheless, the CII-d sequence was highly methylated when Dnmt3a and -3b were present in addition to Dnmt1. Indeed, the level of methylation in wild-type cells (80%) far exceeded the sum of the individual contributions by Dnmt1 (4%) and Dnmt3a and -3b (20%) (see Fig. ). This suggests a strong cooperativity between Dnmt3a and/or -3b and Dnmt1 in ensuring the methylation of these sequences. A steady-state level of methylation of 80% can be maintained either by very efficient postreplication maintenance methylation, with relatively little de novo methylation activity necessary, or by less-efficient maintenance, balanced by substantial, continuing de novo methylation of the region. The strong synergy observed between Dnmt1 and Dnmt3a and/or Dnmt3b suggested that the latter situation may be the case for the CII-d region. One way to distinguish between efficient maintenance methylation of this region and poor maintenance combined with de novo methylation was to determine the level of hemimethylation at this region.
However, no method to analyze the levels of hemimethylation at individual regions in the genome existed. We developed such an assay for the purpose of this analysis, by taking advantage of the fact that the restriction enzyme HpaII does not cut hemimethylated DNA. Therefore, the combination of resistance to HpaII cutting, together with a lack of detectable methylation on one strand in a bisulfite analysis, is indicative of a hemimethylated state at that CpG dinucleotide in the original double-stranded DNA. An unmethylated CpG dinucleotide would be cut by HpaII, while a fully methylated CpG dinucleotide would reveal the presence of methylation in the bisulfite analysis.
This technique that we developed to measure hemimethylation was successfully applied to the analysis of both the CI-f region and the CII-d region, and it showed that the latter had considerable levels of hemimethylation in wild-type cells. As a result of this high level of hemimethylation, it appears that our original determination of 80% methylation at this region in wild-type cells, based on the analysis of a single DNA strand, was actually an underestimate. Only 5% of CpG dinucleotides at this region were completely devoid of DNA methylation (see Fig. 5A). We conclude that the high level of hemimethylation at this region (30%) must reflect poor maintenance methylation, compensated by a substantial amount of ongoing de novo methylation.
Dnmt3a and/or -3b is most likely responsible for this de novo methylation in wild-type cells. This conclusion was supported by our demethylation experiments conducted with 5-aza-CdR in which Dnmt1, by itself, was unable to restore methylation in treated cells. Indeed, the methylation level continued to decrease long after drug treatment, suggesting that once the equilibrium had been reduced below a certain level, the enzyme was incapable of sufficient de novo methylation to keep the sequences methylated. Thus, a major function of the Dnmt3a and -3b enzymes, in addition to de novo methylation of incoming retroviral sequences (20
), may be to methylate endogenous repetitive DNAs so that an equilibrium level of methylation can be maintained within this sequence class. However, since some retroviral sequences and IAP sequences have been shown to be quite well maintained for methylation in M1 cells (20
), Dnmt1 may have sequence specificity with respect to de novo and maintenance methylation of some repetitive DNAs. It is also possible that some unidentified DNA methyltransferases may have sequence specificity instead of Dnmt1. As Yoder and Bestor (29
) have pointed out, the majority of methylation in the eukaryotic genome is found in parasitic sequences, making the complementary activities between the enzymes important in maintaining the transcriptional suppression of some of these potentially transposable elements.
Our experiments were conducted with ES cells known to express increased levels of Dnmt3a and Dnmt3b relative to their differentiated counterparts (21
), so that the relevance of our findings to the maintenance of methylation in somatic cells is an issue. Since expression of Dnmt3a and -3b mRNA can easily be detected in normal fetal and adult human tissues (8
), in addition to mouse tissues (21
), it seems likely that the two enzymes could play similar roles in somatic cells even though they appear to be downregulated during differentiation in vitro. For the above reasons, it will be important to conduct similar experiments in other suitable knockout cell types as they become available.
Our findings regarding the delayed methylation of sequences may have importance in understanding how the sequences are methylated in normal cells and cancer cells. We found that Dnmt1 acted in a biphasic mechanism with respect to the timing of methylation, with up to 10 to 20% of the methylation catalyzed by this enzyme being delayed for some time following DNA synthesis. On the other hand, the Dnmt3a and -3b enzymes appeared to act mainly at the time of replication, since little delayed methylation occurred on DNA which was synthesized in cells lacking Dnmt1. Our data show that the majority of Dnmt1-dependent methylation is concomitant with DNA replication, yet a wave of postreplication methylation is also observed. It will be of obvious importance to probe the exact timing by which these enzymes operate in the cell cycle, in order to understand in more detail how methylation patterns are copied and propagated within eukaryotic cells. Conceivably, Dnmt3a and -3b act immediately before Dnmt1, although it remains to be seen whether this occurs prior to the movement of the parental DNA through the replication machinery or whether the enzymes act on the newly synthesized DNA to cause de novo methylation of sequences unmethylated in previous cell divisions.
The fact that the methylation of sequences in CpG-poor class I DNA sequences (such as CI-f) was maintained in cells lacking Dnmt3a and -3b strongly suggests that Dnmt1 or an unidentified enzyme has some de novo methylation activity toward this sequence class. Otherwise, sites inadvertently left unmethylated in a given S phase would remain unmethylated in the next S phase, resulting in the inexorable loss of methylation with increasing numbers of cell divisions (22
). However, in this regard it was of obvious interest that in the 5-aza-CdR experiments, Dnmt1 by itself could not maintain the methylation of class I sequences once the level of 5-methylcytosine had been reduced below a certain point. This resulted in the continued loss of methylation with increasing cell division. The fact that methylation was maintained at wild-type levels suggests that a significant function of the Dnmt3a and/or -3b enzyme is to restore the methylation of sites “skipped over” by Dnmt1 during the previous round of DNA replication. When these results are coupled with our observations that individual DNA strands tend to have multiple sites methylated and that preexisting methylation encourages de novo methylation by Dnmt1 (6
), the most likely explanation for the maintenance failure is that the existence of methylation in sequences reinforces its propagation. For some reason Dnmt1 alone cannot keep pace with the requirement of de novo methylation of repetitive DNA sequences, indicating that a major function of Dnmt3a and/or Dnmt3b in ES cells is to ensure that this sequence class is kept modified in the genome.