We have demonstrated binding between human DNMTs involving the N-terminal regulatory regions. The biological significance of this association may be involvement in gene silencing via transcriptional repression or DNA methylation. Indeed, functional co-operation of human de novo methyltransferase DNMT3b and maintenance methyltransferase DNMT1 is essential for almost all of the methylation in the colorectal cancer cell line HCT116 (Rhee et al., 2002). Knockout of either DNMT3b or DNMT1 resulted in minimal effects on genomic methylation. However, a double knockout of DNMT1 and DNMT3b led to loss of most of the genomic methylation, despite the presence of the de novo methyltransferase DNMT3a. The loss extended to the repeated DNA sequences, such as satellite 2 and the Alu family repeats. Loss of insulin-like growth factor II (Igf2) imprinting and abrogation of silencing of the tumor suppressor gene p16INK4a were also observed in double-knockout cell lines. In mouse cells, functional co-operation is required between DNMT1 and DNMT3a and/or DNMT3b for methylation of a selected class of sequences, including abundant LINE-1 promoter sequences (Liang et al., 2002). This suggests that both de novo and maintenance enzymes must co-operate to carry out the maintenance of methylation and the silencing of the genes in the mammalian genome.
Methylation also plays an important role in establishing epigenetic patterns during mammalian development. In pre-implantation embryos, the mammalian genome is believed to be essentially unmethylated. At this stage, DNMT1 is transcribed and enzyme is made in the cytoplasm, but it is not targeted to the nucleus (Mertineit et al., 1998). However, 5-methylcytosine staining of early mouse embryos at different stages indicated the presence of methylated DNA (Santos et al., 2002). Transcripts of DNMT3a and DNMT3b could be detected as early as embryonic day 6.5, with the peak activity of DNMT3a at day 10.5 (Okano et al., 1999). These authors speculated that de novo enzymes are most active during the early developmental stages, where they establish the pattern of DNA methylation. In later developmental stages, DNMT1 activity is high and DNMT3a and DNMT3b levels go down. In fact, for DRM2 of mouse Igf2, an imprinted gene, all three enzymes were required for the correct establishment of methylation (Okano et al., 1999). Many human cell lines express hDNMT3a as well as hDNMT3b (Figure 2A), although it is not known whether this is merely an artifact of the establishment process. Although all the mammalian DNMTs recognize the same CG sites and modify them, the relationship between them, their mode of action and precise target recognition is still unknown. Recently, it has been demonstrated that DNMT3a and DNMT3b can modify non-CG sites such as CA, CT and to a lesser extent the 5′ C of the CC dinucleotide (Aoki et al., 2001; Gowher and Jeltsch, 2001; Yokochi and Robertson, 2002). Murine DNMT3a, the most studied de novo methyltransferase, shows a 3-fold preference for de novo methylation as compared with maintenance activity, and exhibits strand asymmetry (Lin et al., 2002; Yokochi and Robertson, 2002). One key question is how the unmethylated pre-implantation genome acquires its DNA methylation pattern during the short period of embryonic cell division. One possibility is that methylation spreading might be involved, with a small amount of de novo methylation by hDNMT3a and/or hDNMT3b followed by rapid allosteric activation of hDNMT1, a property encoded in the N-terminal region (Bacolla et al., 1999). Our binding results suggest that either hDNMT3a or DNMT3b could recruit DNMT1 and facilitate the rapid establishment of the de novo methylation pattern. Co-localization of hDNMT3b and hDNMT1 in the nucleus supports such a hypothesis. Since DNMT3a and DNMT3b are active during pre-implantation, both are likely to be involved in an initial wave of de novo methylation. With different target selectivity, both enzymes would be able to establish the initial blueprint of DNA methylation at CN (N is any nucleotide) sites. During this time, DNMT1 resides outside the nucleus, perhaps as a truncated version of DNMT1 (DNMT1°), as observed by Mertineit et al. (1998) (Figure 7A). After the pre-implantation stage, DNMT1 is transported into the nucleus. Nuclear DNMT1 co-localizes with DNMT3b and perhaps binds less strongly to DNMT3a. The presence of DNMT1 and DNMT3b in the nucleus brings out a wave of de novo and maintenance methylation. DNMT3a, which is capable of non-CG methylation, is perhaps escorted out of the nucleus, since further non-CG methylation is either not essential or may even be deleterious for later development (Figure 7B). Thus, co-operation between de novo and maintenance methyltransferases is crucial for the establishment of methylation. This co-operation might be lost during abnormal growth and development, such as in cancer.
Figure 7. Model for DNA methylation. (A) mCpN methylation by DNMT3a and DNMT3b. DNMT3s can catalyze mCG, mCA, mCT and mCC methylation in vivo, either individually or together. This type of methylation is predominant in the pre-implantation stage of mammalian development. A shorter DNMT1, lacking 118 amino acids, stays in the cytoplasm, except at the 8-cell stage of the embryo (Mertineit et al., 1998). Lack of this N-terminal region might lead to an unfavorable interaction between de novo and maintenance methyltransferases. (B) DNA methylation by DNMTs. Both DNMT1 and DNMT3b co-localize in the nucleus. DNMT3a may interact with either DNMT3b or the full-length DNMT1. Excess DNMT3a is escorted out to the cytoplasm, or cytoplasmic DNMT3a is not allowed into the nucleus. DNMT1 ensures CG methylation in the cell. The nuclear membrane is shown as a dotted circle.
Download figure to PowerPoint
Thus, a model emerges in which DNMT3a and DNMT3b, in the absence of DNMT1, lay down an initial pattern of methylation that is spread once DNMT1 joins the methylation complex in the nucleus, perhaps via allosteric activation by methylated DNA. It is now crucial to determine what are the signals that cause DNMT3a and DNMT3b to set this initial pattern. It seems unlikely that DNA sequence alone is sufficient, and so chromatin structure in collaboration with various protein factors is probably responsible. Thus, accessibility of the target bases may be the key factor in establishing the methylation blueprint of the mammalian genome.