De novo methylation of cytosines in CpG dinucleotides in genome DNA is a developmentally regulated epigenetic process shown to play a pivotal role in the regulation of gene expression in mammals. However, little is known about de novo methylation in nonmammal vertebrates. Given the availability of forward and reverse genetic approaches, we expect that zebrafish will be an attractive organism to study de novo methylation in nonmammal vertebrates. Although the presence of genome-wide changes in methylation in zebrafish is still in debate (Macleod et al., 1999; Martin et al., 1999; Mhanni and McGowan, 2004), we have revealed previously that the CpG island of ntl becomes methylated during early embryogenesis (Yamakoshi and Shimoda, 2003a). Because no de novo methylation was observed in five other zebrafish genes examined, ntl methylation must occur in a gene-specific manner. To understand the significance and mechanism of gene-specific de novo methylation, which is little understood in any organisms, we attempted to isolate the DNA methyltransferase gene responsible for de novo methylation of ntl.
In mammals, three distinct families of DNA methyltransferase genes, termed Dnmt1, Dnmt2, and Dnmt3, have been identified. Dnmt1 shows marked preference for hemimethylated DNA and acts primarily as a maintenance methyltransferase (Bestor, 1992).
Dnmt2 contains all the highly conserved methyltransferase motifs, but lacks the N-terminal domain of Dnmt1. No methyltransferase activity has been detected with recombinant Dnmt2, and inactivation of Dnmt2 in mouse embryonic stem (ES) cells had no effect on either de novo methylation or maintenance methylation (Okano et al., 1998a). Thus, it remains an open question as to whether Dnmt2 encodes a functional methyltransferase.
The Dnmt3 family consists of two related genes, termed Dnmt3a and Dnmt3b, both of them act as de novo methyltransferases (Okano et al., 1998b, 1999). They show striking sequence similarity and have shorter N-terminal domains with low homology to that of Dnmt1.
In zebrafish, three possible methyltransferase genes were identified previously: zebrafish dnmt1 (Martin et al., 1999; Mhanni et al., 2001), zebrafish dnmt2 (Dong et al., 2001), and dnmt3, a homologue of the murine de novo methyltransferase genes Dnmt3a and Dnmt3b (Xie et al., 1999). Among these zebrafish genes, complete cDNA structure was determined only for dnmt1 (Mhanni et al., 2001). In this study, we suppressed the function of these three genes, together with other five candidate genes, and we found that one of the genes, termed dnmt7, is required for the de novo methylation of ntl.
RESULTS AND DISCUSSION
We first searched for possible de novo methyltransferase genes in the zebrafish genome database (http://www.sanger.ac.uk/Projects/D_rerio/) using the tblastx program. As a query, we used the partially determined peptide sequence of the zebrafish dnmt3, and found five other genes that contain the catalytic domain of methyltransferase (Kumar et al., 1994). We referred to the five genes as dnmt4, dnmt5, dnmt6, dnmt7, and dnmt8.
To clone full-length cDNAs of the five new genes together with partially determined dnmt2 and dnmt3 genes, we performed 5′ and 3′ rapid ampification of cDNA ends (RACE) using gene-specific primers chosen from the upstream region of the conserved catalytic domain of each gene. Single polymerase chain reaction (PCR) products were obtained from each RACE except for dnmt8, showing that all the seven genes were transcribed in the 1-day-old embryos from which total RNA for the RACE was extracted. We sequenced the RACE products and reconstituted cDNA structures of the seven genes (Fig. 1A). All the genes except dnmt2 encode in common the PWWP domain and C-terminal catalytic domain, both of which are present in the murine de novo DNA methyltransferases Dnmt3a and Dnmt3b (Okano et al., 1998b). We thus regarded these genes as members of the Dnmt3 family. Phylogenetic trees for the Dnmt2 and Dnmt3 families are shown in Figure 1B and C, respectively. Dnmt3 and Dnmt6 are closely related to Dnmt5 and Dnmt8, respectively (Fig. 1C), owing to their similarity in the N-terminal domains (data not shown). Dnmt3, which was designated due to its similarity to the murine Dnmt3 family, was subsequently found to be less-closely related than Dnmt4, Dnmt6, or Dnmt8 (Fig. 1C). It should be noted that Dnmt3 and Dnmt7 have a calponin-homology (CH) domain in their N-terminal regions (Fig. 1A), known to bind to microtubules (Bu and Su, 2003). No other known methyltransferase genes carry this domain. The function of the CH domains in Dnmt3 and Dnmt7 remains to be determined.
We then suppressed the activity of dnmt1, 2, 3, 4, 5, 6, 7, and 8 genes by injecting antisense morpholino oligonucleotides (MOs) into one-cell zebrafish embryos. All the MOs targeted the presumptive translation initiation codon of each gene. The methylation status of ntl was surveyed 48 hours postfertilization (hpf), by which time nearly half of CpG dinucleotides in the ntl CpG island were methylated in normal development (Yamakoshi and Shimoda, 2003a). We found that two MOs, dnmt1-MO and dnmt7-MO, reduced the methylation level at the CpG island to 60% and 20% of those of control experiments, respectively (Fig. 2A). No other MOs changed the level of de novo methylation.
To confirm the specificity of dnmt1-MO and dnmt7-MO, we first tested negative-control MOs for dnmt1 and dnmt7, which carry mismatched nucleotides in their original sequences, and found that neither of them affect de novo methylation of ntl (Fig. 2A). We then examined three additional MOs, two for dnmt1 (dnmt1 5′-untranslated region [UTR]-MOa and dnmt1 5′UTR-MOb) and one for dnmt7 (dnmt7 5′UTR-MO). They are complementary to the 5′UTR region of dnmt1 or dnmt7, but do not overlap with the original dnmt1-MO or dnmt7-MO. Contrary to our expectations, the two new dnmt1 5′UTR-MOs showed no effect on ntl methylation (data not shown). On the other hand, the dnmt7 5′UTR-MO reduced the methylation level of ntl, as shown in Figure 2B. Furthermore, the reduced methylation level was partially restored in dnmt7 5′UTR-MO–injected embryos by the introduction of dnmt7 mRNA (Fig. 2B), which is insensitive to the MO. No restoration of methylation level was observed by coinjection of dnmt1 mRNA. Based on these results, we judged that the MOs against dnmt7 worked in a gene-specific manner and that dnmt7 was involved in ntl methylation. We did not further pursue the specificity of dmt1-MO in this study.
Incomplete repression of ntl methylation by the dnmt1-MO or the dnmt7-MOs (Fig. 2A,B) may be due to the activity of the maternal Dnmt1 and Dnmt7 proteins that morpholinos are unable to eliminate. In fact, the presence of Dnmt1 protein in unfertilized eggs has been suggested (Mhanni and McGowan, 2002).
Because Dnmt1 is the zebrafish authologue of the mammalian maintenance methyltransferase, we suspected that the effect of dnmt1-MO was not specific to the ntl CpG island but rather reduced the methylation level over the entire genome, a phenomenon observed in organisms in which the activity of maintenance methyltransferases are suppressed (Li et al., 1992; Stancheva and Meehan, 2000). To test this possibility, we performed Southern blot analysis using two repetitive sequences as probes, DANA (mermaid) and Type I satellite DNA (Fig. 3A). DANA is dispersedly located in zebrafish chromosomes (Izsvàk et al., 1996; Shimoda et al., 1996), and Type I DNA is tandem-duplicated at centromeric regions (He et al., 1992; Ekker et al., 1992). Using both probes, we observed global hypomethylation in the genomes of dnmt1-MO–injected embryos (Fig. 3A). Bisulfite sequencing of these two repetitive sequences showed that dnmt1-MO reduced the methylation levels at CpG dinucleotides in a DANA and Type I sequences to 73% and 51% of those of the control sample, respectively (Fig. 3B). Furthermore, reduction of the methylation level was also detected at the 3′ region of ntl (Fig. 2C) where CpG dinucleotides are heavily methylated regardless of the developmental stages in wild-type zebrafish (Yamakoshi and Shimoda, 2003a). Because dmt1-MO induced hypomethylation in all the genomic regions we surveyed, a common phenomenon observed in organisms in which maintenance methyltransferases are defective (Li et al., 1992., Stancheva and Meehan, 2000), it is likely that zebrafish Dnmt1 is a maintenance methyltransferase.
In contrast to dnmt1-MO, dnmt7-MO did not induce genome-wide hypomethylation or lower the methylation level at the ntl 3′ region (Figs. 2B, 3A,B). These results indicated that dnmt7 is a gene that selectively methylates the ntl CpG island de novo. Presumably Dnmt1 maintains the ntl methylation pattern established by Dnmt7, and the severest reduction of ntl methylation by coinjection of dnmt1-MO and dnmt7-MO (Fig. 2A) is possibly caused by failure in the both process: de novo and maintenance methylation of the ntl CpG island. Based on these results, we decided to focus on dnmt7 for further study.
The expression pattern of dnmt7 is shown in Figure 4. dnmt7 transcripts were detectable as early as the 4-cell stage (Fig. 4A), indicating that they are maternally supplied. From blastula to mid-segmentation stages, dnmt7 is expressed ubiquitously in embryos. At the end of the segmentation stage, 24 hpf, a slightly higher expression was observed in the brain (Fig. 4I), and the highest expression was detected in the tectum (Fig. 4H). The transcripts were almost undetectable by 2 days postfertilization (Fig. 4J). The drastic reduction of dnmt7 mRNA that occurred between day 1 and day 2 was confirmed by reverse transcriptase-PCR (RT-PCR; Fig. 5). The expression pattern shown in Figure 4 suggests that ntl methylation by Dnmt7 occurs globally in early zebrafish embryos. This suggestion is in agreement with our previous observation that, apart from sperm, the ntl CpG island was methylated in the genomes of all adult organs and tissues examined (Yamakoshi and Shimoda, 2003a). In addition, the strong expression of dnmt7 in the brain may be related to the highest degree of ntl methylation observed in the brain of an adult zebrafish (Yamakoshi and Shimoda, 2003a).
Although maternal dnmt7 transcripts are supplied in cleavage-stage embryos and zygotic dnmt7 is expressed throughout the blastula and gastrula stages, de novo methylation of ntl becomes evident ∼16 hpf (Yamakoshi and Shimoda, 2003a), well after the end of gastrulation (∼10 hpf). The absence of ntl methylation, despite the presence of dnmt7 mRNA, may be caused by demethylation activity in early zebrafish embryos, which is detected for at least 6 hr after fertilization (Collas, 1998). The demethylation activity may be gradually overridden by increasing methyltransferase activity of Dnmt7.
We then examined whether the timing of ntl methylation could be altered by overexpression of dnmt1 and dnmt7. Although we investigated the degree of ntl methylation at three different time points after injection of dnmt1 or dnmt7 mRNA into zebrafish eggs (10 hpf, a few hours before the onset of ntl methylation; 16 hpf, a few hours after the onset of ntl methylation; and 30 hpf), no significant changes in ntl methylation were observed between dnmt1- or dnmt7-overexpressing embryos and control embryos (data not shown).
Except for ntl, plasmid DNA microinjected into zebrafish eggs is the only other known example in which de novo methylation occurs in zebrafish (Collas, 1998). Because the de novo methylation of microinjected DNA starts around 12 hpf (Collas, 1998), as does that of ntl (Yamakoshi and Shimoda, 2003a), we asked if Dnmt7 was also involved in methylation of plasmid DNA. We coinjected both pEGFP (enhanced green fluorescent protein) plasmid and dnmt7-MO into fertilized eggs, and examined the methylation status of the plasmid 2 days later. We found that de novo methylation occurred in the injected plasmid DNA in the presence of dnmt7-MO (Table 1). Of interest, methylation occurred not only in CpG, but also in CpT, CpA, and CpC whose cytosines are not methylated in the zebrafish endogenous genes we have investigated to date (data not shown). Therefore, de novo methylation of injected exogenous DNA differs from that of ntl in terms of enzymes involved and cytosines targeted.
Table 1. dnmt7 Is Not Involved in Methylation of Plasmid DNAa
Plasmid DNA microinjected into fertilized zebrafish eggs was recovered 24 hr later, and the methylation status of the DNA was analyzed by bisulfite sequencing (n = 10). Note that methylcytosines are not limited to CpG dinucleotides.
Total numbers of CpN dinucleotides included in the region we surveyed by bisulfite sequencing.
We then checked variation of methylation status between samples of normal and MO-treated embryos. We measured methylation of plasmid DNA and genomic sites all recovered from the same embryos and tried this experiment three times independently (Table 2). We found that, at each site we surveyed, mean values of methylated cytosines in 10 independent clones were close each other in the three independent experiments, although the number of methylated cytosines varied clone by clone. These results showed that changes in methylation level at the regions we surveyed could be reproducibly induced by the MOs we used.
Table 2. Comparison of Three Independent Experiments for Measuring Methylation Changes Induced by dnmt1-MO or dnmt7-MOa
We microinjected plasmid DNA with or without morpholino oligonucleotide (MO) into zebrafish eggs and recovered DNA from the eggs 48 h later. The methylation status of the DNA was analyzed by bisulfite sequencing. Shown are the percentages of methylated cytosines. Values are averages ± standard errors (n = 10), except for Type I repeat, for which only averages are indicated (n = 50).
44 ± 8.5
96 ± 5.0
97 ± 3.2
2.2 ± 0.5
37 ± 10
98 ± 4.4
90 ± 8.6
2.9 ± 2.0
51 ± 11
95 ± 8.2
93 ± 8.9
2.8 ± 1.3
26 ± 6.7
64 ± 9.1
65 ± 22
2.5 ± 1.2
24 ± 8.1
52 ± 12
71 ± 20
2.1 ± 1.4
21 ± 9.0
60 ± 10
59 ± 10
2.9 ± 1.8
7.0 ± 6.6
94 ± 4.8
92 ± 5.8
3.0 ± 0.9
8.2 ± 6.2
95 ± 4.7
95 ± 6.0
2.6 ± 0.8
4.0 ± 3.4
96 ± 3.9
96 ± 8.5
2.0 ± 1.3
The biological significance of ntl methylation by Dnmt7 remains unclear because dnmt7-MO–injected embryos normally develop and no difference in ntl expression was observed between wild-type and dnmt7-MO–injected embryos at any stages examined (data not shown). Further analyses await mutants in which no functional Dnmt7 protein is produced.
Also unknown is how Dnmt7 selectively methylates ntl. In plants (Chan et al., 2004; Zilberman et al., 2004) and, recently, in human cells (Kawasaki and Taira, 2004; Morris et al., 2004), correlation has been suggested between DNA methylation and short RNA. DNA methylation may be targeted to identical sequences by short RNAs derived by means of Dicer cleavage of dsRNA. To test whether short RNA is involved in ntl methylation, we tried to suppress the function of zebrafish Dicer1 using MO because dicer1-MO–injected zebrafish embryos are defective in the maturation of short RNA (Wienholds et al., 2003). Although dicer1-MO–injected embryos showed the same dicer mutant phenotypes as reported by Wienholds et al. (2003), the de novo methylation of ntl occurred normally (data not shown). Furthermore, we were unable to detect an antisense strand of ntl mRNA or short RNA homologous to ntl (data not shown). Thus, we have not so far obtained any data suggesting the involvement of RNA in ntl methylation. Small RNA-independent DNA methylation has been shown in Neurospora (Freitag et al., 2004).
An alternative possibility accounting for ntl methylation is DNA–DNA pairing. A precedent for DNA–DNA pairing as a trigger for DNA methylation is provided by the methylation induced premeiotically (MIP) phenomenon in filamentous fungi, Ascobolus immerses (Selker, 1997). MIP is pairing-dependent silencing and methylation of duplicated DNA regions longer than ∼400 bp. By analogy, we are interested in the possibility that DNA pairing between ntl and cryptail, a ntl-pseudogene-like DNA (Yamakoshi and Shimoda, 2003b), occurs by means of the transient formation of double-stranded DNA ends in an ntl intron (Yamakoshi et al., 2005) and that the resultant heteroduplex becomes a target of Dnmt7.
We found a gene, dnmt7, that specifically methylates ntl in the zebrafish genome. The identification of dnmt7 will be an important step toward understanding how local methylation takes place in the genome because, as far as we know, dnmt7 is the first de novo methyltransferase gene shown to be involved in gene-specific methylation in any organisms. Biochemical analyses of the Dnmt7 protein and genetic screenings for other factors involved in ntl methylation will help us to understand the mechanism and biological significance of de novo methylation of ntl.
Cloning of Seven Methyltransferase Genes
One microgram of total RNA from 1-day-old zebrafish embryos was used for reverse transcription to generate first-strand cDNA. RACE was performed using a SMART RACE cDNA Amplification Kit (Clontech) following the manufacturer's instructions. Gene specific primers used for 5′ and 3′RACE were as follows: for dnmt2, CATCTCCCTGCAAACCAATCCTG (5′ RACE) and TTTGACATGATTTTGATGAGCCCACC (3′ RACE); for dnmt3, CCAGCAAGCTAACAATTAATTTCCCGA (5′ RACE) and CCAACTGAGAAGTACTGGCAAACAGAGC (3′ RACE); for dnmt4, GGTAATGTTGCGAACATCATGCACATA (5′ RACE) and GACGGCCAATCAGAGTGCTTTCG; (3′ RACE); for dnmt5, ATCTTCTACTTCCATTGGCTCAACGTCTTC (5′ RACE) and GCAGTTCTTATAAAAGTGCTCCATCACATC (3′ RACE); for dnmt6, TTGGTCGTGATTGTTGGCAAAGAAGTGCTG (5′ RACE) and AGTGTGTGGATTTGTTGGTGGGTC (3′ RACE); for dnmt7, CCTTCACATATTCAATCTTGCCCTCATGTT (5′ RACE) and TGTATCTTTCTATTCCAGCTCATAAGCGCC (3′ RACE); for dmt8, GAGGATGCTCAAGTGTGACGTTCAGAC (5′ RACE) and AGGAACCCGAGAGAGACTCATGTTAGAA (3′ RACE).
Synthesis of dnmt1 and dnmt7 mRNA
The open reading frames of dnmt1 and dnmt7 cDNA were PCR amplified from first-strand cDNA (derived from bud-stage embryos) and plasmid DNA carrying full-length dnmt7 cDNA, respectively. To minimize PCR error, KOD DNA polymerase (TOYOBO) was used for this PCR, and sequences of both strands of each PCR product were confirmed. Primer sets used were TAGAATTCCCATGCCTACCAAGACCTCATTGT (forward) and CTCTCGAGTTAGTCAGAGAGCTCCATTTTCTCCT (reverse) for dnmt1, GCGGATCCACCATGGCTACAAATGTTAGTCTG (forward) and GACTCTAGATCACTCGCAAGCAAAGTAATCCT (reverse) for dnmt7. The two PCR fragments were cloned into pCS2+, and after the digestion with NotI, the plasmids were used for mRNA synthesis by using a mMESSAGE mMACHINE kit (Ambion). Concentration of mRNA was determined by an RNA 6000 Nano Assay kit (Agilent Technologies).
MOs against dnmt genes were designed and synthesized by GeneTools. We used the same MO against dicer1 as described (Wienholds et al., 2003). Two nanograms of MOs were injected into the cytoplasm of one-cell stage eggs. The sequences of MOs were as follows: dnmt1-MO, GAGACAATGAGGTCTTGGTAGGCAT; dnmt1 5′UTR-MOa, TCAAAGGAATTGACTACTTAAAAAC; dnmt1 5′UTR-MOb, TTAAAACAACAACAGATAAAAGCGG; dnmt2-MO, AGTCGCTCCGTGTTCTCCATATTTC; dnmt3-MO, CTCCGATCTTTACATCTGCCACCAT; dnmt4-MO, TTATTTCTTCCTTCCTCATCCTGTC; dnmt5-MO, GAACGTCTTCTAGGCATTTCAAGCT; dnmt6-MO, TGGTCCTCCATTGAGTTCATCACAG; dnmt7-MO, CAGACTAACATTTGTAGCCATCTTC; dnmt7 5′UTR-MO, ACCTTGGTCCCGGAGACCTCGCTAC; dnmt8-MO, GGTGGTCCATTGTGTCACTCATTTG; dicer1-MO, CTGTAGGCCAGCCATGCTTAGAGAC. The sequences of control MOs for dnmt1-MO and dnmt7-MO were as follows: dnmt1-mmMO, GAcACAtTGAGcTCTTGGaAGGgAT; dnmt7-mmMO, CAcACTAtCATaTGTAcCCATgTTC. Small letters indicate mismatch nucleotides.
In Situ Hybridization
Whole-mount in situ hybridization using digoxigenin (DIG) -labeled riboprobes (Roche) was carried out as previously described (Schulte-Merker et al., 1994). A 3.8-kb dnmt7 fragment and SP6 RNA polymerase were used for the synthesis of the dnmt7 antisense strand probe. An antisense strand of ntl cDNAs was synthesized as previously described (Schulte-Merker et al., 1994). Probes were synthesized using mMESSAGE mMACHINE (Ambion).
Genomic and plasmid DNA used for methylation analysis were extracted from 2-day-old embryos into which each MO was injected. To convert unmethylated cytosines into uracils by bisulfite, DNA was processed using an EZ DNA Methylation Kit (Zymo Research). The ntl CpG island and 3′ non-CpG island regions were PCR amplified with pairs of primers, GTTGTTAAAGTAATAGTATTTAATGGGATT (forward) and CACTTAATATAATATCAAATCTCAACTTAC (reverse), and TTGTATTATAGTTAGTATTTAAGTTTGTGG (forward) and CAAAATCAAATCCATAACTACAACATCAAT (reverse), respectively. To survey the methylation status of injected pEGFP (Clontech), a coding region of the GFP gene was PCR amplified using a pair of primers, TGGGGTATAAGTTGGAGTATAATTATAATA and AACTCCAACAAAACCATATAATC. A DANA/mermaid element located upstream of the zebrafish ependymin gene was amplified by nested PCR under the condition of 2 min at 94°C, 30 × (30 sec at 94°C, 30 sec at 51°C, 1 min at 72°C). The primer sets used for the first and second PCR were TAAATGGTTAAGTGGTTGAGTTTGTT (forward)/CAAACACAAAACCATCAAAAAATATAC (reverse) and TTTTAAAAGTTGTTGATATGAAATTTAATT (forward)/CACCAAAATATTTCAATACTTTCT (reverse), respectively. Type I repetitive sequences were amplified with the same PCR condition as the DANA/mermaid. The template genomic DNA used for this PCR was digested genomic DNA with Sau3AI before bisulfite treatment. The primer set used was ATAAAATGTATTATTTTTTTTTGTT (forward) and AACTAAATCAAATTTTATTACATTCTA (reverse).
Genomic DNA was extracted from 2-day-old embryos into which dnmt1-MO or dnmt7-MO was injected. After digestion with HpyCH4IV or Sau3AI, DNA fragments were separated on 1.2% agarose gel and transferred and fixed onto Biodyne B (Pall) according to the manufacturer's instructions. To create DANA/mermaid and Type I repeat probes, we first cloned them in plasmid vectors. A DANA element located upstream of the zebrafish ependymin gene was PCR amplified as described (Shimoda et al., 1996) and cloned into pGEM-T Easy vector (Promega). To clone Type I repetitive DNA, we first digested zebrafish genomic DNA with Sau3AI. The digested fragments were then directly cloned into the BamHI site of pBluescript SK and used for the transformation of Escherichia coli. Plasmids isolated from 50 transformants were sequenced, and a plasmid carrying Type I repeat was identified. DANA and Type I repeat fragments excised from the vectors were used for probe synthesis with an AlkPhos direct labeling kit as described by the manufacturer (Amersham Pharmacia Biotech). Hybridization, washing unhybridized probes, and detection of signals were also performed following the manufacturer's instructions.
One microgram of total RNA isolated from zebrafish embryos was treated with RNAase free DNase RQ1 (Promega) and reverse transcribed by SuperScript II (Invitrogen) to synthesize first-strand cDNA. cDNAs were subjected to serial dilutions and with control primers for EF1α (forward, AAGAAGCTTGAAGACAACC; reverse, TTCTGTGCAGACTTTGTG) to estimate the linear range of PCR amplification. A fragment of dnmt7 cDNA was PCR amplified with a pair of primers, GGAGTACCAGCTTGCTGAGG (forward) and ATCCCATTTTTGATGGGTGA (reverse), under the condition of 2 min at 94°C, 25 × (30 sec at 94°C, 30 sec at 55°C, 30 sec at 72°C). Because both of the primer sets flank an intron, amplified fragments from cDNA can be distinguished from those from contaminated genomic DNA by size difference. PCR products were run on a 2% agarose gel with a DNA-size marker (New England BioLabs).
We thank Dr. H. Siomi for providing us with facilities during the early stage of this study; H. Yakushi, F. Unemi, T. Nagami for fish maintenance; and Dr. M. Ekker for technical advice on cloning the Type I repeat.