Roles for Dnmt3b in mammalian development: a mouse model for the ICF syndrome

ICF syndrome is a rare autosomal recessive disorder characterized by immunodeficiency, centromeric instability and facial anomalies(Ehrlich, 2003; Wijmenga et al., 2000). Individuals with this disease show combined immunodeficiency, including absence or severe reduction of at least two classes of immunoglobulin and a reduced number of T cells, making them prone to infections and death before adulthood. Centromeric instability is characterized by the formation of radiated chromosomes (chromosome 1, 9 and 16) due to demethylation at cytosine residues in classical satellites 2 and 3 at juxtacentromeric regions of these chromosomes. Facial anomalies include hypertelorism, flat nasal bridge, low set ears, protrusion of the tongue, epicanthal folds, micrognathia and high forehead. Most individuals also exhibit growth and mental retardation. Several studies have demonstrated that ICF syndrome is caused by homozygous or compound heterozygous mutations of the DNA methyltransferase 3B(DNMT3B) gene (Hansen et al.,1999; Okano et al.,1999; Xu et al.,1999).

Dnmt3b is one of the three active DNA cytosine methyltransferases identified in human and mouse (Okano et al., 1998; Xie et al.,1999). Dnmt3a and Dnmt3b have high structural similarity, and both can carry out de novo methylation in embryonic stem cells and during embryonic development (Okano et al.,1999). Although these two enzymes exhibit overlapping functions during early development, each has distinct expression patterns, genomic targets and functions (Chen et al.,2003; Okano et al.,1999). Although Dnmt3a deficient mice develop to term and appear to be normal at birth, Dnmt3b deficient mice are embryonic lethal. Recently, Dnmt3a, but not Dnmt3b, has been shown to be essential for the establishment of methylation imprints during gametogenesis(Kaneda et al., 2004). Genetic analysis of Dnmt3a and Dnmt3b function in embryonic stem cells has revealed that Dnmt3a and Dnmt3b, and their different variants, have shared as well as distinct DNA targets(Chen et al., 2003). However,the specific functions of Dnmt3a or Dnmt3b in vivo have not been fully analyzed yet.

To investigate the function of Dnmt3b in mouse development and to determine whether Dnmt3b mutations result in phenotypes similar to those of individuals with ICF syndrome, we generated mice with point mutations in Dnmt3b corresponding to the mutations found in human patients (ICF mice). Our studies of Dnmt3b null and ICF mutant mice show that Dnmt3b is essential for mouse embryonic development, and that the ICF mice exhibit phenotypes that resemble some of the symptoms of the human ICF syndrome. We also demonstrate that Dnmt3b is essential for the survival of T cells in the thymus of newborn mice.

Construction of the GFP-Dnmt3b1, GFP-Dnmt3b2, GFP-Dnmt3b3, myc-Dnmt3a and myc-Dnmt3b1 vectors has been described previously(Chen et al., 2002; Hata et al., 2002). Human DNMT3B cDNA was synthesized by RT-PCR from total RNA of NT-2 cells,by using the oligonucleotides 5′-ATGAAGGGAGACACCAGGC-3′ and 5′-GCCTGGCTGGAACTATTCAC-3′ as primers, and subcloned into pCAG-IRESblast vector. The cDNAs of ICF mutants were generated by a PCR-based site-directed mutagenesis experiment, by using mouse Dnmt3b1 or human DNMT3B cDNA as template, and subcloned into the pEGFP-C1 (CLONTECH)and/or the pCAG-IRESblast vector. The Dnmt3b A609T and D823G knock-in vectors were generated by sequentially subcloning Dnmt3b fragments and a floxed IRES-βgeo cassette, in which IRES-βgeo is flanked by loxP sites, into pBluescript II SK. The fragments containing A609T and D823G mutations were generated by PCR using a bacterial artificial chromosome clone as a template and the following oligonucleotides as primers:5′-CTGGAGCTGCTATATGTGCC-3′,5′-GGAAAAGTACATTACCTCCGA-3′,5′-CACAGACTTCGGAGGTAATG-3′ and 5′-TTGGTGATTTTCCGGACGTC-3′ for A609T; and 5′-CAGACAGGGCAAAAACCAGC-3′ and 5′-CCGCGGCCCATGTTGGACACGCC-3′ for D823G. The other Dnmt3bfragments were obtained from a bacterial artificial chromosome clone. The identities of all constructs were verified by DNA sequencing.

Transient transfection was carried out in COS-7 cells for immunoprecipitation assay and NIH3T3 cells for fluorescence microscopy analysis, using LipofectAMINE PLUS reagent (Invitrogen). Immunoprecipitation,immunoblotting and fluorescence microscopy analyses were performed as previously described (Chen et al.,2002; Hata et al.,2002). Stable transfection of Dnmt3b-expression vectors in ES cells was performed according to a procedure described previously(Chen et al., 2003). The antibodies used in these experiments were anti-myc (Roche), anti-GFP (Roche),anti-Dnmt3b, anti-Dnmt3a (clone 64B1446; Imgenex), and anti-α-tubulin(Ab-1; Oncogene Research Products).

The Dnmt3b A609T and D823G targeting vectors were transfected into ES cells via electroporation, and transfected cells were selected with G418,as previously described (Li et al.,1992). Clones with homologous recombination were identified by Southern blot analysis. Genomic DNA from ES cell clones was digested with BamHI, blotted and hybridized to an external probe. The wild-type,null, A609T and D823G alleles give fragments of 17 kb, 5.7 kb, 6.5 kb, and 14 kb, respectively (Fig. 2D). Chimeric mice and F1 heterozygotes were generated from multiple A609T or D823G mutant ES cell lines as described (Li et al., 1992). Mutant mice were maintained on a 129SvJae×C57BL/6 hybrid background and analyzed.

Genomic DNA isolated from ES cells, E12.5 embryos, tissues of newborns, and tails of adult mice was digested with methylation-sensitive restriction enzymes and analyzed by Southern hybridization as previously described(Chen et al., 2003; Lei et al., 1996). The probes used for methylation analysis were: pMR150 for minor satellite repeats, pMO for endogenous C-type retroviruses, and an oligonucleotide probe for major satellite repeats.

For histological analysis and TUNEL assay, tissues were fixed in 10%buffered formalin and processed into paraffin wax-embedded sections using routine procedures. For general morphology, deparaffinized sections were stained with Hematoxylin and Eosin using standard procedures. TUNEL assay was performed using the ApopTag Plus Fluorescein In Situ Apoptosis Detection Kit(Serologicals Corporation), by following the manufacturer's protocol. For skeletal staining, mice were skinned, placed in 1.5% KOH and macerated. After 4-5 days, the mice were stained in Alizarin Red for 3 days, and then placed in 1.5% KOH for 1 day to remove the excess stain. The stained bones were then transferred to glycerol.

Cells from the thymus, spleen, and bone marrow were prepared and analyzed for the expression of surface differentiation antigens as described(Georgopoulos et al., 1994; Winandy et al., 1995). Flow cytometric analysis was performed using a FACScan flow cytometer (BD). All antibodies used for staining were from PharMingen.

To address how the ICF mutations affect Dnmt3b function, we introduced six ICF mutations into mouse Dnmt3b cDNA(Fig. 1A) in expression vectors. A609T, L670T, V732G, A772P and D823G are missense mutations corresponding to the human ICF mutations A603T, L664T, V726G, A766P and D817G,respectively (Okano et al.,1999; Wijmenga et al.,2000; Xu et al.,1999). STP813ins is a 9 bp insertion at codon 813 encoding three amino acids, serine, threonine and proline, corresponding to the change found in a patient with ICF syndrome (Okano et al., 1999).

First, the ability of Dnmt3b mutant proteins to interact with Dnmt3a was examined. Immunoprecipitation analysis was performed using GFP-Dnmt3b fusion proteins and myc-tagged Dnmt3a (Fig. 1B). As previously reported, Dnmt3b1 was co-immunoprecipitated with Dnmt3a, as well as with Dnmt3b1 itself(Kim et al., 2002). We also showed the binding of endogenous Dnmt3a to endogenous Dnmt3b in ES cells by co-immunoprecipitation assay (see Fig. S1 in the supplementary material). The A609T mutation disrupted the interaction with Dnmt3a, as well as with Dnmt3b1. The other ICF mutations did not affect this interaction. Further mutagenesis analysis revealed that the region of Dnmt3b required for its interaction with Dnmt3a is in the amino-terminal region of the catalytic domain, including alanine 609 (see Fig. S1 in the supplementary material), suggesting that the A609T mutation alters the conformation of the Dnmt3a interaction domain.

We next examined the subcellular localization of these mutant proteins using GFP fusion proteins. As described before(Bachman et al., 2001; Chen et al., 2004), wild-type Dnmt3b1 and Dnmt3b2 displayed punctate nuclear localization patterns with two major types of nuclear foci: large foci, which corresponded to pericentric heterochromatin; and small foci, which were distributed throughout the nucleus, excluding the nucleoli. Both types of foci were present in the majority of transfected cells (pattern A); however, only the small foci were visible in some transfected cells (pattern B). By contrast, Dnmt3b3, which lacks part of motif IX in the catalytic domain and has no enzymatic activity,displayed diffuse nuclear localization patterns with or without accumulation in pericentric heterochromatin (patterns C and D, respectively). Four of the ICF mutations, A609T, V732G, STP813 and D823G, exhibited obvious changes in localization patterns when compared with wild-type Dnmt3b1. A609T showed no accumulation in pericentric heterochromatin (∼75% of transfected cells showed pattern B and the rest showed pattern D), suggesting that Dnmt3a-Dnmt3b heterodimerization and/or Dnmt3b homodimerization may be required for targeting Dnmt3b to pericentric heterochromatin. V732G, STP813 and D823G showed diffuse patterns similar to those of Dnmt3b3, indicating that these mutations disrupt the association of Dnmt3b with the small type of nuclear foci. Although the identity of these foci remains to be determined, it is possible that these structures correspond to heterochromatin regions, which usually consist of repetitive DNA sequences, including satellite repeats. Failure to target Dnmt3b to heterochromatin may thus contribute to demethylation of satellite DNA, a hallmark of ICF syndrome. The other two mutations, L670T and A772P, did not affect the localization patterns of Dnmt3b, although a minor increase in the number of cells exhibiting pattern B was observed with the L670T mutation (Fig. 1C).

We then examined the effects of the ICF mutations on the methyltransferase activity of Dnmt3b. Mouse Dnmt3b1, A609T, D823G and PC(Dnmt3b1 with its PC motif mutated), as well as human DNMT3B1, A603T and D817G, cDNAs were introduced into highly demethylated Dnmt3a^-/-Dnmt3b^-/- ES cells. Individual clones that expressed different levels of these proteins, as determined by immunoblotting analysis, were obtained(Fig. 1D). DNA methylation patterns were examined using genomic DNA isolated from these clones. As reported previously (Chen et al.,2003), expression of wild-type Dnmt3b1 substantially restored the methylation levels of all the repetitive sequences examined. We found that human DNMT3B1 could also restore the methylation levels of the mouse endogenous C-type retroviral DNA (Fig. 1E), and of the major and minor satellite repeats (data not shown). By contrast, mouse A609T and D823G, as well as human A603T and D817G,proteins failed to restore the DNA methylation of these repetitive sequences,suggesting that these ICF mutants have little or no methyltransferase activity.

Taken together, our results suggest that different ICF mutations result in a loss of function via different mechanisms; some by disrupting protein-protein interactions and others by altering protein localization.

Two ICF mutations, A609T, which disrupts the interactions with Dnmt3a and Dnmt3b1, and D823G, which alters protein localization, were introduced into the mouse Dnmt3b gene by homologous recombination, using a loxP-flanked IRES-βgeo cassette as a selection marker(Fig. 2A). The IRES-βgeo cassette was removed by crossing the heterozygous mice with EIIa-Cre that expresses Cre recombinase in the unfertilised egg of the female germline(Lakso et al., 1996). We have previously generated a Dnmt3b null allele in which the region of the catalytic domain that encompassed the highly conserved PC and ENV motifs was deleted (Okano et al., 1999). The three different mutant alleles of Dnmt3b are shown in Fig. 2B. These alleles could be distinguished by Southern blot analysis(Fig. 2C), and the point mutations were confirmed by sequence analysis of genomic DNA extracted from the tails of the heterozygous mice (data not shown). Mice carrying A609T and D823G mutations produced full-length proteins at levels similar to those of wild-type mice, whereas no Dnmt3b protein was detected in Dnmt3b^-/- mice, as determined by immunoblotting of E12.5 embryo extracts with Dnmt3b antibody (Fig. 2D). The expression of Dnmt3a and Dnmt3a2 proteins was not affected by the Dnmt3b mutations(Fig. 2D). We tried to confirm the effects of the A609T and D823G mutations on Dnmt3b subcellular localization in vivo. Immunostaining of wild-type and mutant embryos at E8.5 and E12.5, as well as of wild-type embryonic fibroblasts, only detected weak signals of a diffuse nuclear pattern (data not shown), consistent with the observations that the level of Dnmt3b is low (as compared with that in ES cells) and that the major isoform is Dnmt3b3 at these developmental stages(Fig. 2D).

Mice heterozygous for the Dnmt3b A609T, D823G or null mutation appeared to be grossly normal, fertile and indistinguishable from their wild-type littermates. As reported previously, Dnmt3b^-/-mice are embryonic lethal (Okano et al.,1999), so no homozygous newborn mice were found(Fig. 3A). By contrast, normal Mendelian frequencies of mice with T/-, G/-, T/T, G/G and T/G genotypes were found at birth, but most of the mutant mice died within 24 hours after birth. A few mice survived longer than 24 hours, and only four mice (three T/T mice and one T/G mice) were found at weaning age (about 3 weeks). These differences in lethality between Dnmt3b^-/- and the ICF mutants indicate that the ICF mutations are not complete loss-of-function alleles.

To determine whether the ICF mutations in Dnmt3b affected global methylation of genomic DNA, we examined the methylation status of three repetitive sequences: major satellite repeats, minor satellite repeats and C-type retrovirus. We found that these repeats were demethylated in Dnmt3b mutant embryos, and in newborn and adult mice. DNA methylation levels in ICF mutant mice were significantly higher than in Dnmt3b^-/- mice, confirming that the ICF mutations resulted in a partial loss of function of Dnmt3b. No difference between A609T and D823G was detected (Fig. 3B-D). Hypomethylation was observed in all tissues that we tested in newborn mice(Fig. 3D), although the expression of Dnmt3b is regulated during embryogenesis and restricted in some tissues at later stages. Moreover, DNA in adult ICF mice was hypomethylated to a similar extent as that in embryos and newborns(Fig. 3C).

To investigate the role of Dnmt3b in mouse development, we first analyzed Dnmt3b^-/- embryos. Dissection of embryos revealed that most of the Dnmt3b^-/- mice were lethal between 13.5 days post-coitum (dpc) and 16.5 dpc (Fig. 4A), showing progressive necrosis. All of the live Dnmt3b^-/- embryos that were recovered at this stage showed subcutaneous edema and liver atrophy (Fig. 4B,C,F), which became apparent at 13.5 dpc. Some of the Dnmt3b^-/- embryos showed ectopic hemorrhage in the head region (Fig. 4B). Furthermore,genotyping revealed that Dnmt3b^-/- embryos were slightly underrepresented from the expected Mendelian ratio at 11.5-13.5 dpc[Fig. 4A, Dnmt3b^+/+:Dnmt3b^+/-:Dnmt3b^-/-= 57(27%): 100(48%): 33(14%)], suggesting an earlier lethality of Dnmt3b^-/- embryos in a small population (about 30%). Occasionally, we recovered abnormal Dnmt3b^-/- embryos between 9.5 and 12.5 dpc (Okano et al.,1999), but the cause of lethality was unclear.

Serial sections of the thoracic region of Dnmt3b^-/-embryos revealed that the ventricular septum was not closed in the heart of Dnmt3b^-/- embryos at 14.5 and 15.5 dpc(Fig. 4D, and data not shown),although ventricular septum closure is normally completed by 13.5 dpc in mice(Webb et al., 1998). Other histological features of the heart, such as trabeculation, were normal in Dnmt3b^-/- embryos. We also found hemorrhaging in the middle of the dorsal root ganglia and in the limb region(Fig. 4E, and data not shown),which suggests defects of blood vessel formation or maintenance in Dnmt3b^-/- embryos. We hypothesize that the ventricular septum defect and blood vessel abnormalities account for the subcutaneous edema of Dnmt3b^-/- embryos.

The fetal liver of Dnmt3b^-/- embryos was much smaller(about one fifth) than that of normal littermates(Fig. 4F). Because the fetal liver is a major hematopoietic organ at this stage, we examined expression of the late erythroid marker TER-119 and the hepatocyte marker albumin in the liver of Dnmt3b^-/- embryos by immunohistochemistry (data not shown). However, we did not see a dramatic change in the erythroblast or hepatocyte populations, suggesting that a defect in hematopoiesis was not the sole reason for this phenotype. We also did not observe significant differences in BrdU incorporation or TUNEL staining in the mutant fetal liver(data not shown), suggesting that the proliferation and apoptosis of hepatocytes was normal.

In contrast to Dnmt3b^-/- animals, mice with ICF mutations (T/T, G/G, T/G, T/- and G/-) developed to term and were alive at birth. These mice showed no sign of edema or liver atrophy, although some exhibited hemorrhage in the head region. The gross anatomy of the mutant mice appeared to be normal, although the body size of the mutant mice was much smaller than that of the control mice (Fig. 5A,B). The body weight of the mutant mice was less than two-thirds of that of the normal littermates at birth, and the adult mutant mice that survived remained smaller (Fig. 5C).

All four types of adult ICF mice showed characteristic facial anomalies,such as shorter nose and wider nasal bridge, as shown in Fig. 5D. These features are similar to the hypertelorism and the flat nasal bridge that are frequently seen in individuals with ICF syndrome. The craniofacial defects were further characterized after skeletal preparations were made(Fig. 5E). The calvarium of adult T/T mice showed abnormally shaped frontal bones. The frontal bones of ICF mice were wider than those of normal littermates, resulting in a wider distance between the eyes. The lateral view revealed that the frontal and parietal bones were round shaped (data not shown). The nasal bone of T/T mice was significantly shorter, whereas the axial length of the calvarium of the mutant mice was the same as that of the control mice. The skeletons of newborn ICF mice showed a similar defect to that of the adults(Fig. 5F). An enlarged frontal fontanel was present in the mutant, indicating that ossification was delayed at the newborn stage. These craniofacial defects were observed in all newborn and adult mice that had T/T, G/G, T/G, T/- or G/- mutations.

To investigate whether Dnmt3b mutant mice had immune defects as seen in human ICF patients, the bone marrow and thymus of the newborn mice were analyzed. B cell populations in the bone marrow of ICF mice (P0-P1) were analyzed by flow cytometric analysis using anti-B220 and anti-IgM antibodies. No significant differences in these precursor and mature B cell populations were observed between ICF mice and littermate controls (data not shown). The thymus of ICF mice was then examined for T cell developmental defects. The number of thymocytes in newborn (P0) ICF mice was similar to that of their littermate controls, and flow cytometric analysis of CD4, CD8 and TCRβexpression revealed no developmental defects in the thymus (data not shown),suggesting that mutations in Dnmt3b do not affect the development of T cells. The thymuses of P1 ICF mice (born 24 to 48 hours before analysis)were, however, much smaller than those of their littermate controls, and the number of total thymocytes was reduced about 10-fold(Fig. 6B). Analysis of CD4/CD8 subpopulations by flow cytometry revealed that CD4⁺CD8⁺double-positive (DP) cells were reduced, and the CD4^-CD8^- double-negative (DN) cells were relatively increased (Fig. 6A). The absolute numbers of thymocytes in all of these subpopulations in ICF mice were reduced (Fig. 6B). Analysis of cell death profiles by AnexinV/propidium iodide staining revealed that the majority of P1 ICF mutant thymocytes were dead or dying (+/+, 9%; T/-, 55%; Fig. 6A). These changes were observed in all of the ICF mice we analyzed, including those with T/-(Fig. 6A,B), G/-, T/T and G/G(data not shown) mutations, although the number of total thymocytes varied among different individuals (probably because of the difference in age, which ranges from 24 to 48 hours).

Examination of Hematoxylin/Eosin-stained histology sections showed that the thymus of P1 ICF mice contained a large number of fragmented and condensed nuclei (data not shown). TUNEL staining confirmed that the majority of the thymocytes were undergoing apoptosis in Dnmt3b mutant animals(Fig. 6C). No difference in the TUNEL staining of spleen, liver, lung, brain, stomach, gut or kidney was detected between ICF mice and controls(Fig. 6C and data not shown). A dramatic increase in thymocyte apoptosis in P1 ICF mice was also detected by DNA ladder analysis (Fig. 6D). Interestingly, DNA laddering was not observed in the thymus of P0 newborn mice, suggesting that T cell death was induced after birth. These data were also supported by an Annexin V binding analysis, which showed a slight increase in apoptotic cells at the P0 stage and no difference in the number of apoptotic cells obtained from 19 dpc embryos (data not shown). A small degree of DNA laddering was also observed in the spleen of P1 mutant mice, but was not observed in other tissues, including liver, kidney, brain and tail(Fig. 6D and data not shown). Moreover, apoptosis was observed in the thymus of Dnmt3b G/G,p53^-/- double mutant mice, indicating that the apoptosis is p53 independent (Fig. 6D). These data suggest that Dnmt3b is essential for maintaining the survival of T cells in the thymus of perinatal mice.

Our previous studies have shown that inactivation of both Dnmt3aand Dnmt3b by gene targeting abolishes de novo methylation of the genome during early post-implantation development and causes embryonic lethality at the late gastrula stage(Okano et al., 1999). However,the function of Dnmt3a or Dnmt3b alone in development remains largely unknown. In this study, we showed that mice homozygous for a null mutation of Dnmt3b die at the late gastrula stage, whereas mice carrying missense mutations of two human ICF syndrome alleles develop to term. Analysis of Dnmt3b mutant mice has revealed that Dnmt3b is essential for the development of the fetal heart, liver and craniofacial structures, and for the survival of T cells at the newborn stage.

ICF syndrome is a rare recessive genetic disorder. Individuals with ICF syndrome can survive to adulthood, but display various defects, most notably immune deficiency and craniofacial abnormalities(Ehrlich, 2003; Wijmenga et al., 2000). Because Dnmt3b null mutant mice die during fetal development(Okano et al., 1999), it is believed that individuals with ICF syndrome must have residual DNMT3B activity and that a complete loss of function of DNMT3B is probably incompatible with fetal development in humans. Indeed, most ICF patients carry missense mutations in or near the catalytic domain of DNMT3B and none were homozygous for nonsense alleles (Hansen et al.,1999; Okano et al.,1999; Shirohzu et al.,2002; Wijmenga et al.,2000; Xu et al.,1999), suggesting that the complete loss of function of DNMT3B is lethal. This also implies that most missense mutations in individuals with ICF may not be complete loss-of-function alleles. This notion was supported by the in vitro analysis of six different missense mutations in the catalytic domain of Dnmt3b (Gowher and Jeltsch,2002). The catalytic activities of the mutant Dnmt3b was shown to be reduced 10- to 50-fold, but was still detectable. However, in vivo studies using cultured mammalian cells expressing Dnmt3b missense mutants characteristic of ICF syndrome failed to detect de novo methylation of episomal DNA (Xu et al.,1999).

In this study, we tested the function of several Dnmt3b mutations in mouse ES cells. We showed that Dnmt3b cDNA containing characteristic ICF mutations failed to induce detectable de novo methylation of repetitive sequences after being introduced into mouse ES cells lacking endogenous Dnmt3a and Dnmt3b (Fig. 1E). However,introduction of two of the ICF mutations into the mouse Dnmt3b gene did not result in embryonic lethality. The fact that these mice survive to term and adult stages, and show significantly higher levels of genomic methylation than do Dnmt3b null embryos, strongly indicates that these ICF mutations are not null, but are hypomorphic alleles. The reason for the difference between the cell culture and in vivo studies is unknown. It is possible that the introduction of the ICF mutations into the endogenous Dnmt3b gene by homologous recombination may allow more precise regulation of Dnmt3b than the introduction of mutant cDNA into ES cells. Dnmt3b may also function more efficiently in whole embryos than in cultured cells.

We showed that all Dnmt3b^-/- mice die during embryonic development between 14.5 and 16.5 dpc as a result of multiple developmental defects, whereas mice expressing point mutations in one or both alleles survive to term. The residual Dnmt3b activity in the ICF mutants is therefore sufficient to allow mutant mice to escape from some of the early developmental defects shown in Dnmt3b null embryos.

The embryonic lethality of Dnmt3b^-/- mice might be caused by several different factors. Dnmt3b may normally repress the expression of genes that control cell growth, and dysregulated activation of these genes may lead to inappropriate growth arrest. We have shown that MEF cells derived from Dnmt3b^-/- embryos undergo premature senescence (Dodge et al.,2005). It is also possible that the multiple cardiovascular defects observed in Dnmt3b^-/- mice might lead to fetal death, although the ventricular septal defect alone does not cause embryonic lethality. Dnmt3b is highly expressed in the placental tissues and disruption of Dnmt3b may lead to placental defects and fetal death.

The Dnmt3b hypomorphic mutations (ICF mutations) lead to a partial loss of de novo methyltransferase activity and a decrease in overall methylation of genomic DNA. Most of the Dnmt3b hypomorphic mutants die shortly after birth but the cause of the newborn lethality remains unknown. We found hemorrhages in the head region in some mice, similar to what was observed in Dnmt3b^-/- embryos. Unlike Dnmt3b^-/-embryos, no ventricular septum malformation was detected, although some mice showed thickening of the myocardium. Other organs, including brain, lung,liver, kidney, stomach, gut and spleen, appeared to be histologically indistinguishable from their normal littermates. Several knockout mice with global alterations of DNA methylation die shortly after birth. Conditional deletion of Dnmt1 in neural progenitor cells (nestin promoter Cre) leads to perinatal lethality due to respiratory failure in the pups(Fan et al., 2001). Targeted disruption of the Lsh gene, which encodes an SNF2 family protein,also causes global hypomethylation and perinatal lethality(Dennis et al., 2001; Geiman and Muegge, 2000). Lsh has been shown to regulate DNA methylation and histone modification(Dennis et al., 2001; Yan et al., 2003a; Yan et al., 2003b). It remains to be determined how Lsh and DNA methylation may regulate developmental processes that are essential for postnatal survival.

The two hypomorphic Dnmt3b mutant strains also showed defects in lymphocyte homeostasis. We observed extensive apoptosis of T cells in the thymus of P1 newborn pups. T cell apoptosis appears to start after birth, as the thymocyte profiles were normal in embryonic and newborn (P0) mice. This suggests that de novo DNA methylation plays a crucial role in suppressing T cell apoptosis in the newborn. As Dnmt3b functions primarily as a de novo methyltransferase, and DNA methylation has been shown to regulate the expression of genes including the T cell cytokine genes IL2, IL4 and interferon (INF)(Bruniquel and Schwartz, 2003; Fitzpatrick et al., 1998; Makar et al., 2003), it is possible that alterations in the expression of these genes may trigger cell death.

Recent studies in mice have suggested an important role for DNA methylation in lymphocytes. Mice carrying a hypomorphic Dnmt1 mutation, which causes genome-wide hypomethylation in all tissues, have been shown to develop T cell lymphomas that exhibit chromosomal instabilities(Gaudet et al., 2003). Conditional deletion of Dnmt1 in early double-negative thymocytes leads to impaired survival of TCRαβ+ cells(Lee et al., 2001), suggesting that Dnmt1 is required for the maintenance of mature T cells. The importance of DNA methylation in lymphocyte survival is also suggested by the phenotype of Lsh deficient mice. Targeted deletion of Lsh results in global genomic hypomethylation and perinatal lethality(Dennis et al., 2001; Geiman et al., 2001). Injection of Lsh^-/- fetal liver cells into Rag2^-/- mice to reconstitute lymphoid development caused a reduction in T cells and B cells compared with controls (Geiman and Muegge,2000). These results suggest that DNA methylation plays a crucial role in the maintenance and normal function of lymphocytes. The specific target genes regulated by DNA methylation during T cell differentiation,however, remain largely unknown.

DNMT3B mutations primarily affect the lymphocytes of individuals with ICF syndrome. Characteristic symptoms include agammaglobulinemia and combined immunodeficiency (Ehrlich,2003). ICF syndrome patients also exhibit hypomethylation of satellite 2 repeats in the pericentromeric heterochromatin, and rearrangements of chromosomes 1, 9 and 16 via hypomethylated satellite 2 in lymphocytes to form multiradiate chromosomes (Jeanpierre et al., 1993; Xu et al.,1999). It should be noted that chromosomal rearrangements have been observed in bone marrow cells from only one of four patients studied(Fasth et al., 1990; Hulten, 1978; Smeets et al., 1994; Turleau et al., 1989), and have never been detected in fibroblast cells derived from four ICF syndrome patients (Brown et al., 1995; Carpenter et al., 1988; Maraschio et al., 1988; Tiepolo et al., 1979). Facial anomaly is another characteristic symptom afflicting individuals with ICF. However, many of the ICF syndrome symptoms are rare and difficult to study because of a limitation in the number of patients. The development of mouse models are thus of great importance.

In this study, we describe mice with two independent ICF point mutations(A609T and D823G) that show developmental defects including T cell defects,facial anomaly and low body weight, which are common symptoms of human ICF syndrome. In addition, genome-wide hypomethylation was observed in both the ICF-like mice and human patients. In the Dnmt3b mutant mice, all repetitive sequences tested were found to be hypomethylated, although the methylation status of single genes has not been analyzed. In ICF patients,hypomethylation of satellite 2, satellite 3, and non-satellite repeats D4Z4 and NBL2 has been observed (Jeanpierre et al., 1993; Kondo et al.,2000; Xu et al.,1999). The striking resemblance of ICF-like Dnmt3b mutant mice to individuals with ICF suggests that these mice will serve as good models for understanding the etiology of ICF syndrome and will aid in the identification of target genes that are regulated by DNA methylation during development.

We thank H. Lei and L. Yu for excellent technical assistance. This work was supported by grants CA82389 and GM52106 from the National Institutes of Health(to E.L.). Y.U. is supported by the Japan Society for the Promotion of Science. T.C. is a recipient of a long-term fellowship from the Human Frontier Science Program.