MBD4 and MLH1 are required for apoptotic induction in xDNMT1-depleted embryos

Epigenetic regulatory mechanisms involve heritable alterations in chromatin structure that govern the transcription profile and genome stability of a cell during animal development or in different disease states(Meehan et al., 2005). It is well established that modification of DNA at CpGs by the addition of a methyl group at the 5 position of cytosine affects gene expression by suppressing the function of gene regulatory elements, including promoters, enhancers,insulators and enhancer blockers (Goll and Bestor, 2005). Dynamic changes in DNA methylation patterns during development or in cancer cells contribute directly to altered transcription expression profiles (Gius et al.,2004; Hershko et al.,2003). Targeted mutation of the gene encoding the maintenance methyltransferase DNMT1 in mice results in DNA hypomethylation, transcription activation and embryo lethality that is associated with activation of apoptosis (Jackson-Grusby et al.,2001; Lei et al.,1996; Li et al.,1992; Takebayashi et al.,2007). Equally, overexpression of DNMT1 is incompatible with mouse and amphibian development (Biniszkiewicz et al., 2002; Stancheva and Meehan, 2000). Homozygous mutant Dnmt1^-/-mouse embryos can be identified as early as 8.5 days postcoitum (dpc), and by 9.5 dpc they exhibit severe developmental delay, lack visible somites and display a distorted neural tube (Lei et al., 1996). This contrasts with Dnmt1^-/- ES cells, which proliferate normally in culture unless they are induced to differentiate, although they do exhibit high rates of microsatellite instability (Guo et al., 2004; Jackson-Grusby et al., 2001). Conditional deletion of DNMT1 in mouse embryonic fibroblasts (MEFs) results in DNA demethylation and a uniform p53-dependent cell death(Jackson-Grusby et al., 2001). However, MEFs with a hypomorphic mutant Dnmt1^n/n can proliferate in culture at normal exponential rates and do not undergo senescence even after 100 generations in the absence of p53 function(Lande-Diner et al., 2007). Inactivation of p53 prolongs the proliferative capacity of Dnmt1 null fibroblasts (Jackson-Grusby et al.,2001).

Despite obvious differences in biology and early developmental strategies,the phenotype of xDnmt1-deficient Xenopus laevis embryos shows remarkable similarity to the mouse Dnmt1^-/-phenotype, as indicated by the presence of axial defects, failure to form a neural tube, and improper patterning of the somites(Stancheva and Meehan, 2000). In frogs, severe depletion of maternally expressed xDNMT1p by antisense (AS)RNA results in the premature transcription of developmentally essential genes and p53-mediated apoptosis (Stancheva and Meehan, 2000; Stancheva et al., 2001). One generalisation from the analysis of DNMT1 depletion or inhibition in embryos is that it is not essential for the survival of embryonic cells during early cleavage stages(Meehan and Stancheva, 2001). Instead the phenotypical consequences of disrupting DNMT1 function only become apparent during and after gastrulation when p53-dependent apoptosis is initiated (Stancheva et al.,2001). A partial reduction (between 35-50% depletion) in xDNMT1p levels, which is not enough to result in global changes in DNA methylation levels, is sufficient for activation of a cell death program in Xenopus laevis embryos (Dunican et al.,2008). These observations support the hypothesis that DNMT1 possesses essential functions that are independent of its role as a maintenance methyltransferase and links its absence with activation of a cellular checkpoint response (Brown and Robertson, 2007).

The initiating signals for apoptosis in DNMT1-deficient embryos are unknown, one hypothesis is that hypomethylation leads to the release of chromosomal proteins that act as sensors of DNA methylation levels and induce p53 activation (Stancheva et al.,2001). Several protein candidates can be considered as sensors,including methyl-CpG binding proteins (MBD), which bind to chromatin by a methylation-dependent mechanism (Klose and Bird, 2006). However this model in its crude form would not be compatible with the dynamic changes in DNA methylation patterns that occur during development (Morgan et al.,2005). Special cases in which normal alterations in methyl-CpG levels would be unable to trigger apoptosis would need to be in place. Another possibility is that the signal originates with xDNMT1p itself, perhaps through its association with the replication machinery or DNA repair pathways(Guo et al., 2004; Leonhardt et al., 1992; Vertino et al., 2002). Our recent observation that the xDNMT1-deficient phenotype is independent of alterations in DNA methylation levels would support this latter possibility(Dunican et al., 2008).

The functional properties of the methyl-CpG binding protein MBD4 (MED1)suggest it as a possible candidate mediator of the apoptotic response in xDNMT1-depleted embryos. Like other MBD proteins, it has been shown to function as a methylation-dependent transcription repressor, but it is also a thymine glycosylase and interacts with the mis-match repair (MMR) protein MLH1(Bellacosa et al., 1999; Hendrich et al., 1999; Kondo et al., 2005; Parsons, 2003). MBD4 thus,like DNMT1, potentially links epigenetic modification mechanisms with the MMR pathway (Parsons, 2003; Guo et al., 2004). An important function of MMR proteins is to sense DNA damage and to mediate the decision to repair the lesion or to induce apoptosis(Palii et al., 2008). The levels of several MMR proteins are reduced in Mbd4^-/-mouse embryonic fibroblasts, which can account for the diminished apoptotic response of these cells to DNA-damaging agents(Cortellino et al., 2003; Sansom et al., 2003). In wild-type cells, DNA-damage recognition by MMR factors is sufficient to trigger cell cycle arrest and apoptosis through direct interaction with signalling kinases such as ATM, ATR, CHK1 and CHK2, which ultimately activates p53 (Luo et al., 2004). In line with this observation, it has been shown that overexpression of MLH1 induces apoptosis in cell lines probably through its interaction with ATM(Luo et al., 2004; Zhang et al., 1999).

Here we show that inhibition of xMBD4 function in early Xenopusembryos reduces the phenotypical severity of xDNMT1 depletion, whereas microinjection of mouse Mbd4 mRNA induces an apoptotic response that disrupts early development. In UVA laser micro-irradiated cells, MBD4, MLH1 and DNMT1 can be colocalised at sites of DNA damage. Our results are compatible with a model in which MBD4 functions as an apoptotic-signalling molecule through its direct interactions with DNMT1 and MLH1.

Xenopus embryos were manipulated according to standard procedures(Stancheva and Meehan, 2000). Two-cell embryos were injected with antisense morpholino oligonucleotides (10 ng for DNMT1 and Kaiso, 5-10 ng for MBD4, MLH1 and p53, per embryo) or with RNA (0.2-1 ng for wild-type mMBD4 and xMLH1 per embryo) synthesised by the mMessage Machine kit (Ambion). The ability of morpholino oligonucleotides to suppress translation was checked by the TNT Quick Coupled Transcription/Translation Kit (Promega) using 1 μg of plasmid DNA and 4μg of morpholino (Fig. S6 in the supplementary material). The following morpholinos were used: xDNMT1 (xDMO1), GGACAGGCGTGAAACAGACTCGGC; xMBD4(xM4MO), TCCGGCACAGGAGCAGCCATGTTTG; xMLH1 (xMLH1MO), CAGCCGCCGAAT AACTCCCGCCATT; xKaiso (xKMO), GATCAGCTTTTTTGTCTCCATGTCT; xp53 (xp53MO),TGCCGGTCTCAGAGGAAGGTTCCAT.

DNA from wild-type and microinjected embryos was isolated and 10 μg of each sample were digested to completion with HpaII or MspI(MBI-Fermentas) and electrophoresed in a 1% native agarose gel. DNA digests were transferred to nylon membranes (Osmonics) and hybridised according to the manufacturer's protocols with a labelled 750-bp Xenopus satellite I probe (Stancheva and Meehan,2000; Stancheva et al.,2001).

Xenopus laevis MBD4 (AW641890), mouse MBD4 (BE335048 and BE457540)and Xenopus DNMT1 (CA790804) clones were used along with the Xenopus MLH1 clone DKFZp724LO463Q3. The full-length (477 amino acid)protein, the glycosylase domain (307-477 amino acids) and the methyl-CpG binding domain (55-133 amino acids) of xMBD4 were cloned into the BamHI/SalI sites of pGEX-4T-1. The RFP-PCNA construct has been described (Mortusewicz et al.,2005). The GFP-MLH1 expression vector was constructed by cloning MLH1 cDNA into the pEGFP-C1 vector. A red variant of the GFP-MLH1 expression vector was made by replacing GFP with mCherry using NheI and BsrGI. The full-length T7-tagged xMBD4 expression construct was made by cloning into the XbaI-BamHI sites of pCGT7 vector(Cazalla et al., 2005). The hDNMT1-FLAG construct was obtained from W. Nelson(Agoston et al., 2005). The GFP-MBD4 construct was obtained from A. Bird and mutated to make the catalytically dead version GFP-MBD4^D534A(Hendrich and Bird, 1998; Rai et al., 2008). The GFP-xDNMT1 fusion was made by cloning full-lenth xDNMT1 into pEGFP-C1. The MLH1-GFP construct was a gift from V. Cryns(Chen et al., 2004). xDNMT1-GST fusions were made by cloning regions corresponding to 1-130 amino acids, 124-380 amino acids and 505-600 amino acids into the pGEX-4T-1 vector(GE Lifesciences).

Published procedures for the whole-mount TUNEL staining of embryos were followed (Hensey and Gautier,1997).

GST-tagged fusions were expressed in the ER2566 E. coli strain,purified using glutathione sepharose 4B and dialysed. GST pull-downs were performed according to standard protocols using in vitro translated and ³⁵S-Met-labelled proteins. Samples were subjected to SDS-PAGE,transferred to 3MM paper, dried and visualised by phosphorimaging. HeLa cells were used for immunoprecipitation experiments. DNA constructs were transfected with Lipofectamine 2000 (Invitrogene). Cells were analysed 24 hours after transfection. Immunoprecipitation was performed using the Catch and Release v2.0 system (Upstate). Anti-T7-tagged antibody (Novagen), anti-FLAG (Upstate)and monoclonal anti-GFP (CRUK) antibodies were used.

Subcellular localisation studies using GFP and RFP fusion proteins were done according to standard procedures, using p53^-/- MEFs as recipients (Lande-Diner et al.,2007). Cells were analysed 24 hours after transfection. Positive signals were visualised with Alexa Fluor secondary antibodies. Slides were examined using a Zeiss Axioplan II fluorescence microscope with Plan-Neofluar objectives, a 100 W mercury source (Carl Zeiss, UK) and a Chroma #84000 quadruple band-pass filter set (Chroma Technology, Rockingham, VT, USA) with the excitation filters installed in a motorised filter wheel (Ludl Electronic Products, USA). Grey-scale images were captured with a Hamamatsu Orca AG CCD camera [Hamamatsu Photonics (UK)] Instruments Pentamax). Image capture and analysis were performed using in-house scripts written for IPLab Spectrum(Scanalytics Corp, Fairfax, VA, USA). Transfection of expression plasmids,microirradiation and live cell imaging was as described(Mortusewicz et al., 2005; Mortusewicz et al., 2007). For apoptosis analysis, primary mouse embryonic fibroblasts or HCT116 cells were transfected using lipofectamine, stained with PE Annexin V (BD Biosciences)8-10 hours after transfection, sorted on BD FACSAria II SORP sorter and analysed using BD FACs Diva software. The percentages of annexin-positive cells in populations of GFP-positive and GFP-negative cells were identified.

Rat DNMT1 Myc-tagged constructs were kindly provided by Kunio Shiota(Kimura and Shiota, 2003). Mouse DNMT1-GST fusions were obtained from Sara Nakielny(Jeffery and Nakielny, 2004). Immunostaining was performed according to standard techniques using p53^-/- mouse embryonic fibroblasts (MEFs) as recipients(Lande-Diner et al., 2007). Cells were analysed 24 hours after transfection. A mouse monoclonal anti-T7 tag antibody (Novagen) was used. Positive signals were visualised with Alexa Fluor secondary antibodies. Slides were examined as described above.

Recently, we demonstrated that injection of a morpholino against xDNMT1(xDMO) into two-cell embryos results in premature zygotic transcription before the mid-blastula transition (MBT) and phenotypical abnormalities that mimic the effect of AS xDnmt1 RNA injection(Dunican et al., 2008; Stancheva et al., 2001). However, the xDMO morphants do not exhibit any general changes in DNA methylation levels (Fig. 1A)(Dunican et al., 2008). Despite this, they undergo apoptosis, as evidenced by the appearance of cellular shedding from the embryo surface and by TUNEL positivity(Fig. 1B,C; Fig. 3B), features that we have previously characterised for xDnmt1 AS RNA-injected embryos(Stancheva et al., 2001). This suggests that global hypomethylation of DNA is not the trigger for apoptosis and that an alternative pathway is responsible for initiating programmed cell death when xDNMT1p levels are reduced. We focused on the possible involvement of xMBD4 in initiating apoptosis, as its murine counterpart has been implicated as a central molecule in signalling pathways that are activated in response to DNA damage and (like DNMT1) it is linked with the MMR pathway(Cortellino et al., 2003; Parsons, 2003; Sansom et al., 2003; Bellacosa et al., 1999; Hendrich et al., 1999).

We first tested by morpholino injection whether depletion of xMBD4 or its binding partner xMLH1 could alleviate the effect of xDNMT1 depletion. Fig. 2 shows a panel of control and injected embryos (Fig. 2A-F) at equivalent stage 12-13. In comparison to the wild-type embryos, the xDMO morphants show a high proportion of abnormal embryos (70%),many of which display characteristic white patches of shedding cells that are indicative of apoptotic death (Fig. 2B). Higher magnification versions of Fig. 2A-F are available for clarity (see Fig. S1 in the supplementary material). A very small proportion(2%) of the xDMO morphants survive to develop to stage 30(Fig. 2J,L), but in comparison to wild-type controls have phenotypical defects that are indicative of abnormal development, including spina bifida (compare Fig. 2J and Fig. 2K). Co-injection of xDMO with either the xMBD4 (xM4MO) or the xMLH1 (xMLH1MO) morpholinos relieves the severity of the phenotype (Fig. 2C,D). There are fewer obviously apoptotic embryos and a higher proportion (65% and 60%, respectively) have a normal appearance at the gastrula/neurula stage (Fig. 2G,H,L). Indeed, many of these double-injected embryos go on to form a high percentage (20% and 13%, respectively) of normal stage 30 embryos when compared with xDMO morphants (2%; Fig. 2L).

We had shown previously that targeted degradation of p53 levels by human papilloma virus E6 protein (HPV-18E6) can rescue the apoptotic response in AS RNA-depleted embryos (Stancheva et al.,2001). As an additional control we also co-injected with xDMO a morpholino against xp53 (xp53MO), which as predicted, rescued the apoptotic phenotype induced by xDMO injection and led to a high survival rate at gastrulation/neurulation (80%) and tadpole stage (75%; Fig. 2E,L). However, none of the surviving xDMO/xp53MO morphants had a normal appearance, and they exhibited a reduction of trunk and posterior regions with relatively small head structures by stage 30 (Fig. 2I). This is probably because, in addition to its role in DNA-damage signalling pathways, xp53 is required for proper mesoderm specification during Xenopus development(Cordenonsi et al., 2003; Takebayashi-Suzuki et al.,2003). As an additional control, we also co-injected an xKaiso morpholino (xKMO) with xDMO, which is also required for pre-MBT silencing(Ruzov et al., 2004). Unlike xM4MO, co-injection with xKMO enhances cell death by gastrulation/neurulation with no survivors at this stage (Fig. 2F). We conclude that either xMBD4 or xMLH1 is specifically necessary for the activation of the apoptotic pathway that results from xDNMT1 depletion.

Beause depletion of either xMBD4 or xMLH1 ameliorates the apoptotic response that results from a decrease in xDNMT1 levels, we hypothesised that overexpression of either protein would induce the cell death pathway. To test this possibility, we microinjected mRNAs (1 ng) for murine MBD4 (mMBD4) and MLH1 (mMLH1) into two-cell embryos, allowing them to develop until equivalent stage 12-13, and assayed them for apoptosis with the TUNEL assay. In comparison to wild-type embryos, the mMBD4 and mMLH1 mRNA induced developmental arrest as exogastrulae and resulted in hyper TUNEL-positive staining (Fig. 3A-D). For comparison, the xDMO morphants showed less intense TUNEL staining and also an exogastrulae phenotype at the equivalent stage(Fig. 3B). The effect of mMBD4 overexpression is dose dependent, at a lower amount (200 pg) the embryos went on to develop to tadpole stages but with axial defects (see Fig. S2A,B in the supplementary material); none of the high dose-injected embryos developed beyond gastrulation (Fig. 3E,I). Injection of an equivalent dose of β-galactosidase mRNA had no adverse effect on development (see Fig. S3 in the supplementary material).

As additional proof that mMBD4 overexpression results in the activation of a caspase-dependent pathway, we cultured the control and mMBD4 mRNA-injected embryos in the presence of 20 μM Z-DEVD-FMK (a caspase inhibitor), a concentration that does not affect the development of control embryos (data not shown). This ensured that a high proportion of the mMBD4 mRNA-injected embryos (55%) developed to tadpole stage, although many of them had axial defects that were similar to those observed for the low-dose injection(Fig. 3E-G; see also Fig. S2A,B in the supplementary material). In addition, the effect of mMBD4 overexpression (1 ng mRNA) could be suppressed by the presence of either xp53MO or xMLH1MO, which resulted in 27% and 18% survival, respectively, by stage 30 (Fig. 3I). Furthermore, the presence of xM4MO reduced the lethality of the mMBD4 mRNA injection, which suggests that endogenous xMBD4 is recruited in the apoptotic response pathway that is activated upon mMBD4 mRNA injection(Fig. 3H).

In agreement with these data, overexpression of either GFP-MLH1 or GFP-MBD4 in transiently transfected primary mouse embryonic fibroblasts (MEFs) resulted in activation of an apoptotic response 2-4 (for MLH1) or 2-3 (for MBD4) times above that caused by GFP alone (Fig. 3J). The glycosylase catalytic activity of MBD4 is not required to activate apoptosis, as an inactive mutant, GFP-MBD4^D534A,also induced apoptosis in MEFs (Fig. 3J). This suggests that the response is not activated through the DNA damage that is potentially mediated by the glycosylase activity of MBD4. By contrast, overexpression of GFP-MBD4 did not induce apoptosis in HCT116 cells, which lack functional MLH1(Papadopoulos et al., 1994),whereas overexpression of GFP-MLH1 did induce apoptosis in these cells(Fig. 3K). We conclude that MBD4 can signal an apoptotic response in combination with MLH1, although it is not clear how xDNMT1p depletion acts as a trigger for apoptotic activation through xMBD4/xMLH1 (see model in Fig. S2C in the supplementary material).

Because hypomethylation is not required for the apoptotic response, yet depletion of xDNMT1 is and the absence of xMBD4 or xMLH1 inhibits apoptosis,we hypothesised that the xMBD4 protein interacts directly with xDNMT1. To test this idea, we co-expressed tagged forms of either human DNMT1 (FLAG) or xDNMT1(GFP) with xMBD4 (T7) in HeLa cells and tested their interaction by immunoprecipitation (IP) and subsequent western blotting. IP of xMBD4 precipitates Xenopus and human DNMT1(Fig. 4A; data not shown). In the reverse experiment, IP of xDNMT1 or human DNMT1 brings down xMBD4(Fig. 4B). To test for a direct interaction, we made a GST fusion of xMBD4 and investigated whether it could pull down in vitro-translated xDNMT1. As a positive control, we also tested the interaction between mouse MLH1 and xMBD4(Bellacosa et al., 1999);unsurprisingly, xMBD4 interacts with MLH1 via its C-terminal glycosylase domain (Fig. 4C), whereas the N-terminal MBD domain of MBD4 mediates its interaction with xDNMT1(Fig. 4C). We also mapped the MBD4-interaction domain on DNMT1 from mouse, rat and Xenopus using a combination of IP and GST pull-down experiments (see Fig. S4A-C in the supplementary material). We could define a minimum domain of approximately 70 amino acids in the N-terminal targeting sequence (TS) region of DNMT1 that accounts for the MBD4 interaction (Fig. S4D in the supplementary material);interestingly, this overlaps but is not the same as a region in rat DNMT1 that interacts with MECP2 (Kimura and Shiota,2003).

Previous work has shown that MBD4 localisation to heterochromatin in mouse cells is methylation dependent (Hendrich and Bird, 1998). We could also demonstrate by immunofluorescence that epitope-tagged versions of xMBD4 localise at methylated heterochromatic foci in p53^-/- mouse embryonic fibroblast (MEF) cells (to avoid inducing the apoptotic response), whereas MLH1 does not(Fig. 4D; Fig. 5A). However, in the presence of xMBD4, MLH1 can be recruited to heterochromatin(Fig. 4E). DNMT1 associates with chromatin during the G2 and M phases, and this association is mediated by a specific targeting sequence that shows a strong preference for constitutive but not facultative heterochromatin(Easwaran et al., 2004). Interestingly, DNMT1 can colocalise with xMBD4 at heterochromatic foci(Fig. 5A). By contrast, MLH1 is not preferentially localised to constitutive heterochromatin in the presence of overexpressed DNMT1, although there is some overlap when cells are in G1 and S phases (Fig. 5B). Because MBD4 interacts with both MLH1 and DNMT1 by independent domains, it is possible that xMBD4 can act as a recruiter of MLH1 to chromatin sites that are coincident with methylated DNA and DNMT1 during the cell cycle. This possibility is borne out in triple transfection experiments in which GFP-tagged DNMT1 can be colocalised with Cherry-tagged MLH1 in the presence of xMBD4 (Fig. 5C).

Recent work suggests that DNMT1 participates in the MMR pathway and can be recruited to sites of DNA damage by PCNA(Guo et al., 2004; Mortusewicz et al., 2005). PCNA encircles double-stranded DNA as a trimer, forming a sliding clamp that tethers proteins involved in DNA replication and repair. We analysed the recruitment kinetics of either GFP-MLH1 or GFP-MBD4 in the presence of RFP-PCNA in HeLa cells (which do not have prominent heterochromatic foci)sensitised by BrdU incorporation after inducing localised DNA damage with a UVA laser micro-irradiation system(Mortusewicz et al., 2005). Under these conditions, various types of DNA lesion are generated leading to the recruitment of PCNA. As might be expected, this also led to the recruitment of GFP-MLH1 to damaged sites, but its accumulation lagged behind that of PCNA (Fig. 6A; see also Movies 1-4 in the supplementary material). Likewise GFP-MBD4 was recruited to PCNA foci, but its retention in the presence of PCNA was less stable than that of MLH1 (Fig. 6B; see Movies 1-4 in the supplementary material). However, when GFP-MBD4 and Cherry-MLH1 were co-transfected, both molecules were recruited to microirradiated sites and maintained there during the 5-minute observation period(Fig. 6C). In addition,RFP-DNMT1 and GFP-MBD4 could localise to sites of DNA damage(Fig. 6D). These data suggest that the interaction of these partners on chromatin is influenced by DNA damage and that, while present at these lesions, it is possible that MBD4 and DNMT1 can participate in the decision to either repair the DNA damage(including the DNA methylation profile) or activate a cell death program.

The interaction between xMBD4/xMLH1 and xDNMT1 suggests a model whereby loss of xDNMT1p releases or activates xMBD4 to initiate the apoptotic program via xp53. It also suggests that catalytically dead forms of DNMT1 might suppress the apoptotic pathway, as this role appears to be independent of changes in DNA methylation (Fig. 1) (Dunican et al.,2008). As shown previously, AS RNA-injected embryos are severely depleted in xDNMT1 protein levels during early stages (stage 5-10), resulting in their DNA becoming hypomethylated, and they exhibit high rates of embryo lethality by stage 30 even though xDNMT1p levels are restored by stage 11(Stancheva and Meehan, 2000; Stancheva et al., 2001)(Fig. 7A). Substitution of cysteine 1226 with tyrosine in the catalytic site of human DNMT1(hDNMT1^C1226Y) results in an enzyme with no methyltransferase activity (Jair et al., 2006). The presence of either the wild-type or mutant hDNMT1 mRNA substantially rescues the phenotypical deficits associated with transient depletion of xDNMT1 before the MBT, with the appearance of between 20-25% phenotypically normal embryos at stage 30 (Fig. 7B,C) compared with the low number (4%) of abnormal survivors observed following AS RNA-only injection. This suggests that hDNMT1^C1226Y can suppress activation of the apoptotic pathway in AS RNA-injected embryos, and the subsequent phenotypical abnormalities. It also supports the idea that DNA hypomethylation is not the trigger for apoptosis,as this cannot be restored by hDNMT1^C1226Y. Another feature of AS RNA-injected embryos is the premature activation of transcription prior to the MBT (stage 8.5) (Stancheva and Meehan,2000). In contrast to xDMO-rescued embryos in which xDNMT1p is depleted by 30-40% (Dunican et al.,2008), co-injection of either the hDNMT1^C1226Y or wild-type hDNMT1 (hDNMT1^WT) with xDNMT1 AS RNA was unable to restore pre-MBT silencing of xBRA and xID2 to wild-type levels(Fig. 7D). This implies that the premature transcription phenotype before the MBT (stage 8.5) and the activation of apoptosis after MBT (stage 11 and later) are independent processes, and that it is the latter mechanism that is responsible for the phenotypical appearance of xDMO and AS RNA mutant embryos. The inability to restore silencing in rescued AS RNA embryos is in line with a model in which high levels of DNMT1 are required during pre-MBT stages to maintain gene repression (Dunican et al.,2008).

In Fig. 8A, we present a model for the activation of xDNMT1-mediated apoptosis in Xenopus laevis embryos. We propose that a chromatin-associated complex of xDNMT1,xMBD4 and xMLH1 in normal embryos responds to DNA damage or replication stress by deciding either to repair the lesion or to activate an apoptotic response through the activation/release of MBD4/MLH1 from chromatin-bound xDNMT1. The MBD4/MLH1 complex signals, probably via interaction with the DNA-damage kinases ATM and ATR, activate the p53-dependent programmed cell death pathway. Depletion of xDNMT1 triggers the apoptotic response by activation of the pro-apoptotic MBD4/MLH1 complex (Fig. 8B). Removal of either xMBD4 or xMLH1 prevents the activation of apoptosis in xDMO morphants (Fig. 8C).

In previous work, we characterised the molecular pathology of xDNMT1-depleted embryos, which exhibit the typical features of programmed cell death, including DNA fragmentation, the appearance of TUNEL-positive staining,caspase activation and stabilisation of p53 levels(Stancheva et al., 2001). The induction of 14-3-3 proteins in these mutant embryos is consistent with the activation of the G2 DNA damage cell cycle checkpoint. The cell death phenotype of xDNMT1-depleted animal pole explants could be suppressed by overexpression of either the anti-apoptotic protein Bcl2 or HPV-E6, an inhibitor of p53 function. Taken together, this suggests that loss of the DNMT1 protein is linked with the cellular response to DNA damage (checkpoint defects). Exposure of cells to 5-aza also suggests that DNMT1 is a component of the DNA damage-response system. This results in the robust induction ofγ-H2AX, DNA fragmentation, and the activation of repair proteins in both ATM and ATR (DNA damage-response kinases) pathways(Palii et al., 2008). DNMT1 is also strongly colocalised with γ-H2AX in 5-aza-treated cells, which is consistent with its recruitment to sites of DNA damage(Mortusewicz et al., 2005; Palii et al., 2008). A major question that results from these observations is what is the initiating signal for `damage' in the absence of DNMT1 in cell lines and embryos? An obvious possibility is improper DNA re-methylation; however, our experiments in which partial xDNMT1 depletion by morpholino injection results in an embryonic lethal phenotype and activation of the apoptotic pathway without any global or locus-specific changes in DNA methylation suggests that this model is not applicable (Fig. 1A)(Dunican et al., 2008). Like the effect of AS RNA depletion, the xDMO morphants are TUNEL positive and the cell lethal phenotype (but not the developmental defects) is rescued by inhibiting xp53 function (Fig. 3B; Fig. 2E,I). Similarly, the apoptotic response in (Dnmt1^-/-,p53^-/-) MEFs is suppressed even though up to 10% of genes are ectopically activated, including MBD4, caspase 1 and caspase 4(Lande-Diner et al., 2007). Our recent observation demonstrating that a catalytically dead version of human DNMT1 can rescue the embryonic (apoptotic) phenotype of xDMO morphants suggested to us that the signal for apoptosis lies not with DNA methylation but with xDNMT1 itself (Dunican et al.,2008).

As DNMT1 has been connected to the MMR and the DNA damage-response pathways(Guo et al., 2004; Palii et al., 2008), we considered what might functionally link xDNMT1 directly to these pathways. We choose to focus on MBD4, as it promotes an apoptotic response to DNA-damaging agents and excises spontaneously deaminated cytosine (i.e. uracil) or methylcytosine (i.e. thymine) from G-T/U mismatches(Cortellino et al., 2003; Hendrich et al., 1999; Sansom et al., 2003). MBD4 also interacts with the MMR/tumor suppressor gene (TSG) MLH1, and inhibition of MLH1 function also causes increased apoptosis resistance with regard to DNA-damaging agents, implying possible checkpoint and apoptotic signalling functions for both proteins (Luo et al.,2004; Zhang et al.,1999; Brown et al.,2003). Our double-depletion experiments(Fig. 2) suggest that both xMBD4 and xMLH1 are required for the embryonic lethal phenotype of xDMO morphants, which we attribute to their ability to activate the apoptotic response (Fig. 1E,F; Fig. 3A-E). This suggests a possible signalling pathway that connects DNMT1 to MBD4/MLH1 and subsequently a DNA damage-sensing response. MLH1 associates with the DNA-damage kinase ATM and MMR complexes formed at sites of DNA damage facilitate the phosphorylation of checkpoint kinases 1 and 2 by ATM(Bartek and Lukas, 2003; Brown et al., 2003). ATM also directly phosphorylates p53 at serine 15, contributing to its stabilisation and increased activity as a transcription factor(Canman et al., 1998). The activation of p53 by these stimuli leads to the transcriptional modulation(both activation and repression) of p53-target genes. In addition, p53 regulates cell cycle progression by controlling the G1-S and G2-M checkpoints. Thus, p53 is an obvious candidate for mediating the DNMT1-dependent apoptotic response (Bellacosa, 2001; Levine et al., 2006). Phosphorylation of ATM is also observed after radiation-induced DNA double-strand breaks that lead to activation of the G2/M checkpoint and the double-strand break-response pathway(Shiloh, 2003). The potential colocalisation of DNMT1, MBD4 and MLH1 implies that they might participate in a cellular checkpoint that monitors potential DNA hypomethylation events by a`proxy' mechanism that detects the presence or absence of DNMT1, perhaps at or adjacent to the replication fork (see Fig. S5 in the supplementary material). The recruitment of these components in response to localised DNA damage also suggests that they can have a role in the cellular decision whether to repair the lesion or activate apoptosis (Fig. 8). We hypothesise that it is the absence of DNMT1 at this checkpoint that triggers apoptosis by releasing MLH1/MBD4 to activate ATM in mutant embryos and somatic cell lines. Interestingly, complete inactivation of DNMT1 in HCT116 cancer cells also activates G2 arrest; however, most if not all of these DNMT1 mutant cells eventually escape this arrest and exhibit mitotic defects, including broken chromosomes, chromosome congression or alignment; defects that result in mitotic catastrophe(Chen et al., 2007). It is possible that, in agreement with our model, DNMT1-deficient HCT116 cells could escape G2 arrest without the activation of apoptosis because they lack functional MLH1, and that this contributes to the observed mitotic catastrophe phenotype (Papadopoulos et al.,1994).

What is the potential importance of the mechanism that we have identified?Cancers arise from the sequential acquisition of genetic alterations in specific genes. The high number of mutations in cancer cells led to the hypothesis that an early step in tumor progression is the generation of genetic instability. The potential role of genetic instability in the initiation and progression of colorectal cancers has been well defined in the hereditary nonpolyposis colon cancer syndrome(Bodmer, 2006). Constitutive activation of the DNA-damage response is a feature of the pre-invasive stages of many human tumors, including colorectal cancer (CRC), and this checkpoint response is believed to delay or prevent the malignant transformation of early lesions (Bartkova et al.,2005). MBD4 and other MMR components are frequently mutated in CRCs that exhibit micro-satellite instability(Miquel et al., 2007). These observations suggest that DNMT1 is a crucial part of a signalling cascade that activates an inducible barrier against tumour progression and genetic instability resulting from replicative stress in pre-cancerous lesions or the perturbation of epigenetic regulatory mechanisms during development(Halazonetis et al., 2008). Cellular sensitivity to changes in DNMT1 levels is thus lost when components of the signalling cascade are either absent or mutated. In this context, it is notable that cell cycle regulation of DNMT1 is often abrogated in tumor cells and this has been suggested to be indicative of a poor prognosis(Agoston et al., 2005; De Marzo et al., 1999; Saito et al., 2003).