Plasticity in Dnmt3L-dependent and -independent modes of de novo methylation in the developing mouse embryo

Cytosine methylation profoundly impacts genome expression and stability. In mammals, its occurrence at CpG-rich promoters is correlated with transcriptional repression of retrotransposons and various single-copy sequences, such as pluripotency and germline-expressed genes, as well as genes subject to genomic imprinting and X-chromosome inactivation. DNA methylation is essential for mammalian development and reproduction, as shown by the early embryonic lethality and sterility phenotypes of mouse models of global DNA methylation deficiency (Li et al., 1992; Okano et al., 1999; Bourc'his et al., 2001; Kaneda et al., 2004; Bostick et al., 2007). Abnormal methylation patterns are also associated with pathological states in humans, including cancer (Rakyan et al., 2011). DNA methylation patterns are set genome-wide during two main developmental stages: gametogenesis and early embryogenesis (Morgan et al., 2005; Borgel et al., 2010). These establishment phases follow a genome-wide clearance of methylation, which occurs concomitantly with the expression of pluripotency factors, in primordial germ cells and preimplantation embryos (Rougier et al., 1998; Smallwood et al., 2011; Guibert et al., 2012; Smith et al., 2012).

DNA methylation patterns are driven de novo by two DNA methyltransferase enzymes, Dnmt3A and Dnmt3B, which can target cytosines in both CG and non-CG contexts (Chédin, 2011). Expression of these proteins peaks at times of major de novo methylation in developing germ cells, early embryos and embryonic stem (ES) cells (Watanabe et al., 2002; La Salle et al., 2004; Sakai et al., 2004; Hirasawa et al., 2008). Although these enzymes may show some degree of functional redundancy, they are not interchangeable. Dnmt3A is strictly required for gametic methylation but is globally dispensable for de novo methylation in the early embryo. Conversely, gametic methylation can occur without Dnmt3B, whereas embryonic methylation cannot (Okano et al., 1999; Kaneda et al., 2004; Borgel et al., 2010; Kaneda et al., 2010). Conditional deletions of the Dnmt3a gene further highlight its involvement in methylation processes that take place later, in postnatal cell lineages (Nguyen et al., 2007; Challen et al., 2012). The reasons for this stage specificity in Dnmt3A and Dnmt3B function are unknown.

Whereas Dnmt3a and Dnmt3b orthologs are found widely among distant eukaryotic species, a third Dnmt3 member, Dnmt3-like (Dnmt3l), has evolved uniquely in mammals (Yokomine et al., 2006). This gene encodes a catalytically inert protein that is unable to methylate DNA substrates alone but serves as a regulator of the de novo methylation machinery. In biochemical assays performed in vitro and in cellular systems, human DNMT3L increases the efficiency of the methylation reaction by stimulating DNMT3A and DNMT3B catalytic activity (Chédin et al., 2002; Suetake et al., 2004; Moarefi and Chédin, 2011) and the deposition of more uniform methylation patterns (Wienholz et al., 2010). DNMT3L is moreover involved in the organization of tetrameric complexes with DNMT3A (Jia et al., 2007), leading to the formation of DNMT3L-DNMT3A filaments that polymerize along the DNA molecule in vitro (Jurkowska et al., 2008). Dnmt3L-dependent stimulation of Dnmt3 enzyme activity has been linked to increased affinity for the methyl donor S-adenosyl-L-methionine (SAM), longer residency on DNA, and improved processivity, none of these mechanisms being mutually exclusive (Kareta et al., 2006; Holz-Schietinger and Reich, 2010).

In vivo, Dnmt3l is highly expressed during gametogenesis under the control of sex-specific promoters (Bourc'his et al., 2001; Shovlin et al., 2007). The canonical transcript is produced in prospermatogonia, whereas in oocytes a longer transcript originates from an upstream promoter embedded in an intron of the neighboring Aire gene. Although different in length, these transcripts produce a protein of similar size (O'Doherty et al., 2011). Dnmt3L activity is strictly required for gametic methylation and fertility. Dnmt3l^−/− males do not produce mature sperm, in association with retrotransposon hypomethylation and reactivation during spermatogenesis (Bourc'his and Bestor, 2004). Dnmt3l^−/− females produce methylation-deficient oocytes and their Dnmt3l^−/+ progeny systematically die at mid-gestation, in association with the dysregulation of imprinted gene expression (Bourc'his et al., 2001; Smallwood et al., 2011; Kobayashi et al., 2012). Constitutive Dnmt3l mutant phenotypes mimic a conditional knockout of Dnmt3a in germ cells (Kaneda et al., 2004).

Gametic de novo methylation therefore occurs under the control of a developmentally regulated Dnmt3L-dependent pathway. However, whether embryonic de novo methylation occurs in a Dnmt3L-dependent manner is still controversial. Dnmt3L is abundantly expressed in ES cells from the same promoter used in prospermatogonia (Bourc'his et al., 2001; Shovlin et al., 2007). Dnmt3L protein is found in postimplantation embryos from E5.5 to E8.5 (Hu et al., 2008), in a pattern mirroring embryonic methylation acquisition. On the one hand, Dnmt3l deficiency in ES cells impairs both de novo methylation of newly integrated proviral DNA and maintenance of genomic methylation patterns upon multiple passages in culture (Ooi et al., 2010; Arand et al., 2012). On the other hand, Dnmt3L deficiency in living organisms does not seem to compromise somatic methylation. Indeed, although Dnmt3l^−/− mice are sterile they are perfectly viable with no overt phenotype and present normal methylation patterns in somatic tissues in pre- and postnatal life (Bourc'his et al., 2001; Schulz et al., 2010). By contrast, Dnmt3b^−/− embryos die at mid-gestation, and Dnmt3a^−/− animals develop to term but die within one month (Okano et al., 1999). Contrary to the data obtained in cellular systems, this in vivo observation strongly argues against an essential role of Dnmt3L in embryonic methylation. Alternatively, the presence of maternally inherited Dnmt3L has been proposed as a potential source of Dnmt3L to support the formation of normal methylation patterns in Dnmt3l^−/− embryos.

Whether de novo methylation can occur with or without Dnmt3L depending on the developmental context is an important issue that will shed light on the mechanisms governing methylation patterns in normal and pathological states. Here we present a detailed in vivo analysis of the regulation and requirement of Dnmt3L in gametogenesis and early embryogenesis. Our work illustrates that, contrary to gametic methylation, embryonic methylation can be achieved without Dnmt3L, potentially through the ability of the embryo to express and assemble various combinations of the de novo methylation complex.

All procedures using animals were reviewed and approved by the Institut Curie Animal Care and Use Committee. Oocytes and preimplantation embryos were recovered from ~5-week-old superovulated B6/CBA F1 Mus musculus females. Trophectoderm (TE) cells were dissected as follows. E3.5 embryos were placed individually in 5 μl M2 medium under oil, held in place by gentle suction using a holding micropipette (Humagen, Origio, France) and rotated until the inner cell mass (ICM) was at the 3 o'clock position. A large opening in the zona pellucida (ZP) was created by laser force (MTM Medical Technology, Montreux SA, Switzerland), from which the ICM was aspirated with a biopsy micropipette (Humagen). The ZP was removed by gentle suction and the TE cells were aspirated with the holding micropipette and dissected from the blastocyst using four laser pulses of 3-msecond duration. TE cells were directly frozen at −80°C. ICM cells were isolated by immunosurgery (Nagy et al., 2003). Postimplantation Dnmt3lwt (Dnmt3l^+/+) and Dnmt3l^z− (Dnmt3l^−/−) embryos were obtained from (Dnmt3l^+/− × Dnmt3l^+/−) crosses, whereas Dnmt3l^m− (Dnmt3l^−/+) and Dnmt3l^m−z− (Dnmt3l^−/−) embryos were obtained from (Dnmt3l^−/− × Dnmt3l^+/−) crosses (Bourc'his et al., 2001). At E6.5, epiblasts were separated from the extra-embryonic ectoderm (Nagy et al., 2003). At E9.5, only embryos were kept for analysis. Dnmt3l and sex genotyping were performed by PCR on extra-embryonic tissues for E6.5 and E9.5 conceptuses.

For oocytes and preimplantation embryos, ~300 diploid genomes were collected per stage, total RNA was obtained by three freeze/thaw cycles as previously described (Jeong et al., 2005) and lysis extracts were processed for cDNA conversion (Superscript III, Invitrogen). For E6.5 embryos, ten epiblasts of sex-matched wt and Dnmt3l^z− genotype were separately pooled, extracted using the RNA Picopure kit (Applied Biosystems) and treated with DNase (TURBO DNase, Ambion) prior to cDNA conversion. Quantitative amplification of cDNA was performed using the PCR primers listed in supplementary material Table S1 on an Mx4000 QPCR system according to manufacturer instructions (Stratagene).

Bisulfite genomic treatment on postimplantation embryos was performed using the Epitect kit (Qiagen). For E6.5, seven Dnmt3lwt and five Dnmt3l^z− epiblasts were pooled for DNA extraction. For E9.5, DNA from two independent embryos was treated separately for each genotype. For oocytes (300) and blastocysts (50), DNA was extracted as described (Smith et al., 2009). For oocytes, bisulfite conversion was performed on agarose-embedded DNA. For blastocysts, bisulfite conversion was carried out with the Epitect kit. Single, nested or semi-nested PCR was conducted using the primers listed in supplementary material Table S2. Amplification products were cloned using the TOPO-TA vector system (Invitrogen) and individually sequenced, or directly processed by mass array technology (Sequenom) as described previously (Popp et al., 2010). Lack of somatic contamination in oocyte samples was assessed by profiling the methylation of the paternally and maternally imprinted loci H19 DMD and Kv DMR, respectively (data not shown). Quma Analyzer software was used for sequence alignments (quma.cdb.riken.jp) and clones with identical patterns of conversion were removed from the final Pileup. Sequence logos were created using WebLogo (weblogo.berkeley.edu).

Whole-mount immunostaining was performed essentially as previously described (Santos et al., 2005), with some minor modifications. Briefly, following ZP removal, embryos were fixed in 3% paraformaldehyde for 15 minutes at room temperature, and then permeabilized with 0.5% Triton X-100 in PBS at 4°C from 3 to 20 minutes (oocyte to E4.5). After blocking in 5% BSA and 0.1% Triton X-100 in PBS, embryos were incubated with primary antibodies at 4°C overnight: mouse anti-mouse Dnmt3A (Abcam ab13888; 1:200), mouse anti-mouse Dnmt3B (Abcam ab13604; 1:100), rabbit anti-mouse Dnmt3L [from S. Tajima (Sakai et al., 2004); 1:200], mouse anti-mouse Cdx2 (Biogenex AM392; 1:50), and mouse anti-mouse Nanog (BD Biosciences 560259; 1:100). Two-dimensional images were obtained using an epifluorescence photomicroscope (Digital Module R, Leica). Confocal sections were obtained at 0.2 μm intervals on an UltraVIEW spinning disk laser-scanning confocal multifocal monophoton (PerkinElmer) using 60× and 40× objectives for E2.5 and E3.5-4.5 embryos, respectively.

For western blotting, 30 E6.5 embryos were collected from wt and Dnmt3l^z− genotypes and homogenized in 100 μl ice-cold buffer comprising 10 mM NaH₂PO₄, 150 mM NaCl, 6 mM CaCl₂, 0.5% NP40 and protease inhibitors (Roche). Protein extracts were run on 4/16% SDS-PAGE polyacrylamide gradient gels. Blocking was performed in 5% milk, 0.1% Tween 20 in PBS. Anti-Dnmt3A (Abcam ab13888; 1:500) and anti-tubulin (Oncogene CP06; 1:1000) antibodies were incubated overnight at 4°C and detected by HRP-conjugated secondary antibodies (GE Healthcare). The ECL system (Amersham) was used to detect the signal.

Little is known about how the specificity of Dnmt3l transcripts is regulated in gametes or which forms are found in early embryos before implantation. We determined the kinetics of Dnmt3L expression during preimplantation development from the ovulated oocyte (E0) to the late blastocyst (E4.5) so as to address the transition between maternally inherited and autonomous zygotic expression. For protein assessment, we performed whole-mount immunostaining with an antibody raised against the C-terminal region of mouse Dnmt3L (Sakai et al., 2004), the specificity of which we confirmed on fertilized Dnmt3l^−/− oocytes (supplementary material Fig. S1).

In the wild-type (wt) ovulated oocyte, Dnmt3L protein accumulated on meiotic chromosomes (Fig. 1A). In the 1-cell zygote, the maternal protein was distributed equally on the maternal and paternal pronuclei (supplementary material Fig. S1), with enrichment in DAPI-dense ring structures around nucleolar bodies, a staining reminiscent of pericentromeric heterochromatin. The protein persisted only briefly in the nucleus until the 2-cell stage, and was mostly absent from the 4-cell to compacted morula (E2.5) stages. At E3.5, all the blastomeres of the blastocyst regained Dnmt3L protein, with cell-to-cell variation in intensity but no obvious delineation between the inner cell mass (ICM) and the trophectoderm (TE). At E4.5, cells positive for Dnmt3L were clearly restricted to a discrete area of the expanded blastocyst. Co-staining for the TE marker caudal type homeobox 2 (Cdx2) confirmed that the Dnmt3L protein progressively disappears from the mural and the polar TE from which the placenta and other extra-embryonic membranes originate, and is instead specific to the ICM, which will give rise to the embryo proper (Fig. 1B). The Dnmt3L territory was larger than that of the pluripotency factor Nanog, suggesting that Dnmt3L expression might extend to the primitive endoderm (supplementary material Fig. S2A).

In RT-qPCR assays, the total amount of Dnmt3l mRNA followed the same developmental kinetics as observed at the protein level. We observed a sharp decrease from the oocyte to the 2-cell stage embryo, which is a common trait for maternally inherited transcripts, followed by reinitiation of expression between E2.5 and E3.5, and a dramatic increase from E3.5 to E4.5 (Fig. 1C). Moreover, there was negligible Dnmt3l transcript in the TE as early as E3.5, suggesting that transcript disappearance precedes Dnmt3L protein loss in these cells (supplementary material Fig. S2B). Using primers specific to different Dnmt3l transcript isoforms, we confirmed that the protein present at the 1-cell stage is the translation product of the oocyte transcript (Dnmt3loo), and that the zygotic transcript activated after E2.5 is the same as that expressed in prospermatogonia and ES cells (Dnmt3lzygo) (Fig. 1C).

The kinetics of Dnmt3lzygo expression is reminiscent of pluripotency-dependent transcriptional regulation. Examination of publicly available ChIP-Seq data from ES cells revealed a significant enrichment of pluripotency factor [Oct4 (Pou5f1), Sox2 and Nanog] binding just upstream from Dnmt3lzygo (P<10⁻⁹), but not in the vicinity of the Dnmt3loo promoter (supplementary material Fig. S2C) (Marson et al., 2008). Pluripotency factor availability is therefore likely to underlie the specificity of Dnmt3lzygo recruitment in pluripotent blastomeres of the preimplantation embryo.

In terms of CpG enrichment, the Dnmt3loo and Dnmt3lzygo promoters belong to an intermediate class between those with high CpG content (CpG islands) and low CpG content. As the activity of these intermediate promoters is inversely correlated with DNA methylation (Weber et al., 2007), we tested whether differential DNA methylation acts as a regulator of promoter switching and lineage restriction of Dnmt3l transcription in the peri-conception window. We analyzed the DNA methylation pattern of the Dnmt3loo and Dnmt3lzygo promoters in gametes, E3.5 preimplantation and E6.5 postimplantation embryos using a bisulfite cloning approach. We investigated the regions spanning the transcription start site (TSS) of each isoform, with Dnmt3loo containing eight CpGs and an observed/expected CpG ratio of 0.6, and Dnmt3lzygo with 16 CpGs and a CpG ratio of 0.45.

Both promoters showed sex-specific methylation in gametes, but with an opposite pattern: Dnmt3loo is specifically methylated in sperm, whereas Dnmt3lzygo is methylated in the oocyte (Fig. 2A). This pattern qualifies these regions as germline differentially methylated regions (gDMRs) of paternal and maternal origin, respectively. For the Dnmt3lzygo promoter, the region of oocyte-specific methylation extended at least 1 kb upstream from the TSS (Fig. 2A), but was previously shown to exclude the Aire promoter located 6 kb upstream (O'Doherty et al., 2011). As shown for other oocyte-methylated regions, abundant non-CG methylation was found at the Dnmt3lzygo promoter, mostly in the CA and CT context (Fig. 2B,C) (Tomizawa et al., 2011; Kobayashi et al., 2012). The dramatic reduction in methylation observed in Dnmt3l-deficient oocytes further confirmed that Dnmt3lzygo promoter methylation during oogenesis is controlled by the Dnmt3L protein itself, both in CG and non-CG contexts (Fig. 2D,E).

The gDMRs of the Dnmt3l locus were, however, lost during preimplantation, as demonstrated by the hypomethylated state of both the Dnmt3loo and Dnmt3lzygo promoters in E3.5 blastocysts (Fig. 3A). Loss of parent-of-origin methylation was associated with equal expression of the paternal and maternal alleles of the Dnmt3l transcript in hybrid E3.5 blastocysts derived from crosses between C57BL/6J and CASTEi/J mouse strains (Fig. 3B), and further confirmed in hybrid ES cells. Interestingly, although Dnmt3lzygo transcription was repressed in the TE as early as E3.5, no distinction in promoter methylation was found between the ICM and the TE at this stage (supplementary material Fig. S2D). After implantation, in E6.5 epiblasts, methylation is gained at both the Dnmt3loo and Dnmt3lzygo promoters (~75% methylated), as differentiation occurs and pluripotency is lost (Fig. 3A).

Zygotic Dnmt3L expression is concomitant with the wave of de novo methylation that occurs around the time of implantation. The timing of methylation initiation has not been precisely determined, but most data converge toward a window from E4.5 to E6.5, with locus-to-locus heterogeneity (Dean et al., 2001; Borgel et al., 2010). To functionally assess the role of Dnmt3L in this process, we analyzed the kinetics of methylation acquisition in postimplantation embryos lacking zygotic Dnmt3l (Dnmt3l^−/−, referred to here as Dnmt3l^z−), as generated from heterozygous crosses (Dnmt3l^+/− × Dnmt3l^+/−), compared with their wt littermates. Various genomic sequences that are known to acquire methylation around the time of implantation were analyzed after bisulfite conversion in dissected epiblasts at E6.5 for the beginning of de novo methylation and in E9.5 embryos for a readout at the end of the process; these comprised promoters of germline-specific and pluripotency genes, secondary or somatic DMRs of imprinted loci, and repetitive sequences such as LINE-1 elements and major satellite DNA. We included intracisternal A-particle (IAP) elements as a control of maintenance methylation, as the methylation of these sequences has been reported to be globally unaltered throughout development (Lane et al., 2003; Guibert et al., 2012).

We found that the germline-expressed Scp3 (Sycp3 – Mouse Genome Informatics) and Dazl genes acquired complete methylation as early as E6.5 in wt epiblasts, confirming previous data (Borgel et al., 2010). CpG methylation levels of 90% were observed at E6.5 for both promoters and were maintained until E9.5 (Fig. 4A). A dramatic reduction in methylation was observed in Dnmt3l^z− embryos at E6.5, but significant methylation differences disappeared by E9.5. For Scp3, values of 89.1% and 20% CpG methylation (P<10⁻⁹⁰, Fisher's exact test) were observed at E6.5 and values of 87.9% and 75.3% (P=1.00) at E9.5 in wt and Dnmt3l^z− embryos, respectively (Fig. 4A). Examination of individual alleles at E6.5 indicated that Dnmt3L deficiency leads to a reduction in the number of alleles with methylated CpGs, but, even more dramatically, in the number of methylated CpGs per allele (Fig. 4B). In wt embryos, 100% of Scp3 alleles had at least one methylated CpG, with an average of 16.9 methylated CpGs per allele (out of 19 analyzed CpGs). In the Dnmt3l^z− background, 80% of alleles had at least one methylated CpG but with a much lower average of 3.8 methylated CpGs per allele.

A significantly lower methylation density in E6.5 Dnmt3l^z− embryos was also observed for three secondary, somatically acquired imprinted DMRs associated with the promoters of the Cdkn1c, Igf2r and Igf2 genes (P<10⁻⁵) (Fig. 4A; supplementary material Fig. S3A). At E9.5, methylation levels were close to the expected 50% (in agreement with only one of the parental alleles being methylated) both in wt and mutant embryos, except for Igf2 DMR1, which remained significantly hypomethylated in mutants (36% versus 19.9%; P=0.0005). For genes associated with pluripotency, the Oct4 promoter exhibited a methylation level below 10% at both stages in wt, with increased methylation in E9.5 Dnmt3l^z− embryos (8.4% versus 23.2%; P<10⁻⁴). The Dnmt3lzygo and Dnmt3loo promoters were equally methylated in the presence or absence of Dnmt3L, at E6.5 and E9.5. The embryonic context therefore differs from that of the oocyte, where Dnmt3lzygo methylation was strictly dependent upon Dnmt3L. Finally, methylation of repetitive sequences was globally unaltered in Dnmt3l^z− embryos, and this was observed both for clustered centromeric repeats (major satellite DNA) and IAP elements. However, for transposons of the LINE-1 family, some discrepancies were observed among subclasses of active elements, which can be identified by specific monomeric repeats in their 5′ UTR. The type A subclass was equally densely methylated in wt and Dnmt3l^z− embryos at both stages, whereas the Tf subclass exhibited a delay in acquiring methylation in E6.5 mutant embryos, but gained normal methylation levels by E9.5.

In summary, we provide evidence that the absence of Dnmt3L slows the establishment of embryonic methylation patterns. To rule out a developmental delay as a potential cause of reduced methylation, we confirmed that Dnmt3l^z− embryos are morphologically indistinguishable and present the same growth rate as their wt counterparts at E6.5 (supplementary material Fig. S4), implying a direct effect of Dnmt3L on de novo methylation efficiency. Functionally, the temporary hypomethylation of the germline Scp3 gene and the LINE-1 Tf retrotransposons does not lead to increased transcription of these sequences in E6.5 Dnmt3l^z− embryos (Fig. 4C; supplementary material Fig. S3B). This was an important point to address because reactivation of these sequences has been previously linked to genomic instability (Bourc'his and Bestor, 2004; Borgel et al., 2010). Finally, the hypomethylation phenotype is largely compensated a few days later, so that Dnmt3l^z− embryos attain normal methylation levels by E9.5, continue development, are born in the expected Mendelian ratios (supplementary material Table S3), and progress throughout adulthood with normal somatic methylation patterns and no overt phenotype aside from sterility (Bourc'his et al., 2001).

Next, we investigated the potential developmental compensation of zygotic Dnmt3L deficiency, which would allow Dnmt3l^z− embryos to successfully acquire methylation at some genomic sequences at E6.5 and to fully recover normal methylation patterns at E9.5. In the homozygous state the Dnmt3l mutation prevents the production of a functional protein, from both the zygotic and oocyte promoters (Shovlin et al., 2007). However, it has been argued that the maternal store of Dnmt3L could be maintained throughout preimplantation development and could participate in de novo methylation, even in minute amounts. To functionally assess the contribution of maternal Dnmt3L to embryonic methylation, we generated embryos deficient for both maternal and zygotic Dnmt3l expression (Dnmt3l^m−z−) by crossing Dnmt3l^−/− females with Dnmt3l^+/− males (Fig. 5A). Embryos derived from Dnmt3l^−/− females (Dnmt3l^m−) die at E10.5 (Bourc'his et al., 2001). We found that Dnmt3l^m−z− embryos did not progress past this stage and were phenotypically similar to their Dnmt3l^m−z+ littermates (Dnmt3l^m−), confirming the minor to zero impact of the zygotic mutation at this stage of development, as compared with the severe effect of the maternal Dnmt3l mutation.

We then analyzed the methylation status of sequences that we had previously found to be properly methylated at E9.5 in the absence of zygotic Dnmt3l. Using bisulfite sequencing, we assessed the Scp3 promoter, Igf2r DMR1 and the Dnmt3lzygo promoter, and a methylation-sensitive Southern blot approach was used to analyze three classes of repetitive sequences (LINE-1 type A, IAP and major satellite DNA). None of the analyzed loci differed significantly in methylation in the absence of maternal or zygotic Dnmt3L, nor, most importantly, in the absence of both maternal and zygotic Dnmt3L, as compared with wt embryos at E9.5 (Fig. 5B,C). This convincingly illustrates that the maternal form of Dnmt3L inherited from the oocyte does not play a role in setting up methylation patterns around the time of implantation. These results provide ultimate proof that, in striking contrast to the situation in germ cells, de novo methylation can be achieved in the embryo without any source of Dnmt3L.

Finally, we assessed which of the de novo methylation enzymes, Dnmt3A and Dnmt3B, is available at the time of methylation pattern formation in the early embryo, in the presence or absence of Dnmt3L. By RT-qPCR and immunofluorescence, we confirmed that Dnmt3B is the first active DNA methyltransferase to be transcribed in the embryo as early as E2.5, while Dnmt3lzygo transcription was not yet activated. At E3.5 to E4.5, the Dnmt3B protein is highly expressed. By contrast, Dnmt3A expression only becomes obvious at E4.5 in the ICM of the late blastocyst, with a timing of activation more similar to that of Dnmt3lzygo (Fig. 6A,B).

Interestingly, upregulation in Dnmt3a expression was observed in response to zygotic Dnmt3l deficiency at E6.5, as demonstrated by a twofold increase in transcription in Dnmt3l^z− epiblasts compared with wt littermates (P<0.01) detected by RT-qPCR (Fig. 6C). Dnmt3b transcript levels were comparatively unaffected. The specific upregulation of Dnmt3a could not be explained by a loss of promoter methylation in the Dnmt3l^z− context: the two alternative CpG-rich promoters of this locus – a 5′ promoter that gives rise to Dnmt3a and an intragenic promoter that gives rise to the Dnmt3a2 isoform – are constitutively unmethylated at E3.5, E6.5 and E9.5 in wt embryos, as suggested from publicly available methylated DNA immunoprecipitation (MeDIP)-chip and reduced representation bisulfite sequencing (RRBS) datasets (Borgel et al., 2010; Smith et al., 2012) (supplementary material Fig. S5). Importantly, increased steady-state levels of Dnmt3a transcripts resulted in higher levels of Dnmt3A protein in Dnmt3l^z− epiblasts. Notably, Dnmt3A2, which is the most active form of Dnmt3A and abundant in ES cells and early postimplantation embryos (Nimura et al., 2006), was clearly overexpressed in E6.5 Dnmt3l^z− embryos, as revealed by western blotting (Fig. 6D). Although the details of the compensatory feedback of Dnmt3L deficiency on Dnmt3A expression are not known, a greater availability of Dnmt3A protein provides a likely mechanism for the normal acquisition of embryonic methylation patterns in the absence of a Dnmt3L stimulatory effect.

Although Dnmt3L is strictly required for de novo methylation in germ cells, clear evidence was still lacking as to whether its stimulatory function is necessary in the developing embryo. Our in vivo study demonstrates that a promoter switch underlies the autonomous and time-staged expression of Dnmt3L in the embryo, which appears to have an innate role in accelerating the de novo methylation process. However, we found that the absence of Dnmt3L can be compensated for by other factors. Compared with the germline, the early embryo is more plastic in terms of the mechanisms by which DNA methylation patterns can be acquired.

Although the existence of alternative Dnmt3l promoters was known (Shovlin et al., 2007), we reveal here that these promoters are transmitted to the embryo with opposite allelic states of DNA methylation acquired in the parental gametes: Dnmt3lzygo is methylated on the maternal allele whereas Dnmt3loo is methylated on the paternal allele. Recent genome-wide methylation studies have highlighted certain genomic features that are conducive to oocyte methylation: high CpG density and intragenic location (Chotalia et al., 2009; Smallwood et al., 2011; Kobayashi et al., 2012). The maternal Dnmt3lzygo gDMR fulfills these two criteria, as it is located within the transcription unit defined by the 5′ Dnmt3loo promoter, which is specifically active in the oocyte. Although subject to parent-specific methylation at fertilization, the two promoters of Dnmt3l are not imprinted, as methylation is not maintained during embryonic cleavage. Accordingly, we found no evidence of binding sites at these loci for Zfp57, which is likely to be an obligate factor for the protection of gamete-inherited methylation (Quenneville et al., 2011; Messerschmidt et al., 2012; Proudhon et al., 2012).

DNA methylation has previously been shown to regulate Dnmt3l transcriptional output and promoter usage in the context of ES cells and adult tissues, as well as in vitro transfection studies (Aapola et al., 2004; Hu et al., 2008; O'Doherty et al., 2011). Here we confirm the mutually exclusive relationship between expression and promoter methylation in the oocyte, where the Dnmt3loo promoter is active and unmethylated and the Dnmt3lzygo promoter is inactive and methylated. This is also true for the postimplantation embryo, where the two promoters are silenced and methylated. However, we identified two situations in which transcription control is achieved without DNA methylation: (1) in the E3.5 blastocyst, where both Dnmt3l promoters are unmethylated but only the Dnmt3lzygo transcript is found; and (2) in the TE, where the Dnmt3lzygo promoter is unmethylated but the corresponding transcript is not produced. If transcriptional regulation is involved, it occurs in a DNA methylation-independent manner and/or additional factors are required to activate hypomethylated promoters in these contexts. In support of this hypothesis, the Dnmt3lzygo promoter is physically bound by pluripotency factors in ES cells (Marson et al., 2008), which fits with our in vivo observation of Dnmt3L restriction to the pluripotent part of the embryo. Our study highlights a two-step regulatory model whereby clearance of parent-inherited methylation could potentiate promoter activity, then embryonic pluripotency factors further specify this activity. This double security system might prevent spurious activation of key factors, such as the DNA methylation effectors, to provide efficient spatiotemporal control of genome activity and lineage decision. This type of regulation was formerly reported for the Nanog gene (Farthing et al., 2008). Our study suggests that it could apply more generally to targets of pluripotency factors.

The intricate regulation of Dnmt3l expression may attest to its important function during embryonic development. Our work formally demonstrates that Dnmt3L is indeed involved in de novo methylation at implantation. The germline-expressed Scp3 and Dazl genes, some somatic imprinted DMRs and LINE-1 Tf elements are significantly, although not completely, hypomethylated in Dnmt3l^z− embryos at E6.5, within the early steps of methylation acquisition. The most potent effect we observed is on methylation density per DNA strand, thus providing developmental evidence to support the proposal that Dnmt3L enhances the deposition of long methylation tracts (Wienholz et al., 2010), potentially by stimulating the processivity and/or the catalytic rate of Dnmt3 enzymes once they are loaded on DNA (Kareta et al., 2006; Holz-Schietinger and Reich, 2010; Chédin, 2011). However, Dnmt3L-dependent stimulation is not universal, as other sequences are properly methylated at E6.5 in the absence of Dnmt3L. Moreover, Dnmt3L-dependent effects appear to be time specific, as full methylation patterns are globally recovered in Dnmt3l^z− embryos at E9.5. Taken altogether, our work suggests that Dnmt3L facilitates the formation of methylation patterns in the embryo in its initiation phase. Only by undertaking a developmental analysis of ongoing de novo methylation at individual loci was it possible to reveal such subtle spatiotemporal details of the influence of Dnmt3L on embryonic de novo methylation.

Differential Dnmt3L-dependent stimulation might reflect differential Dnmt3A or Dnmt3B target specificity. On the one hand, expression and locus-specific analyses have identified Dnmt3B as the main enzyme responsible for de novo methylation in the embryo, although some rare sequences require Dnmt3A (Okano et al., 1999; Watanabe et al., 2002; Borgel et al., 2010). On the other hand, Dnmt3L can potentiate both Dnmt3A and Dnmt3B activity in vitro, but Dnmt3A appears to be the favored partner in vivo and in particular in ES cells (Nimura et al., 2006). Of direct relevance to our results, de novo methylation of Dazl and Dnmt3lzygo promoters was demonstrated to be strictly Dnmt3B dependent (Hu et al., 2008; Borgel et al., 2010). The fact that these sequences are either hypomethylated or normally methylated in E6.5 Dnmt3l^z− embryos does not support preferential Dnmt3L stimulation of Dnmt3A or Dnmt3B. Differential methylation kinetics provides another potential explanation for the observed difference in Dnmt3L requirement. In this regard, X-linked CpG islands that acquire methylation slowly upon X-chromosome inactivation were shown to require the activity of the Smchd1 chromatin protein, whereas fast-acquiring sequences do not (Gendrel et al., 2012). Such a relationship might not apply to Dnmt3L, however. For example, although the Scp3 promoter and Cdkn1c DMR are both hypomethylated in E6.5 Dnmt3l^z− embryos, Scp3 is a fast-acquiring sequence that is already completely methylated in E6.5 wt embryos, whereas the Cdkn1c DMR is a slow-acquiring sequence, attaining full methylation only by E9.5. More generally, the process of de novo methylation has been shown to be influenced by local CpG content, gene density, transcription and histone H3K4 methylation states (Ooi et al., 2007; Weber et al., 2007; Borgel et al., 2010; Lienert et al., 2011; Gendrel et al., 2012). Our study represents an indispensable proof of principle for further investigation of the sequence elements and/or chromatin contexts genome-wide that require Dnmt3L for a fast rate of DNA methylation acquisition in the early developing embryo.

The most important finding of our work is the formal demonstration that embryos can attain normal genomic methylation levels in the absence of any source of Dnmt3L, either maternal or zygotic in origin. This situation is at odds with the obligate requirement of Dnmt3L for full methylation in gametes (Bourc'his et al., 2001). This unique property of the embryo occurs in the context of higher levels of the Dnmt3A enzyme, and in particular of the most active Dnmt3A2 isoform. By contrast, Dnmt3L deficiency has no effect on Dnmt3 enzyme levels in prospermatogonia (Niles et al., 2011), and, although it leads to upregulation of Dnmt3b transcripts in oocytes the amount of Dnmt3B protein is unchanged (Lucifero et al., 2007). Increased Dnmt3A availability might result from transcriptional and/or post-transcriptional mechanisms. Because both Dnmt3a promoters are constitutively methylation free in wt peri-implantation embryos, methylation changes in Dnmt3l^z− embryos should not affect Dnmt3a expression levels directly (Borgel et al., 2010; Smith et al., 2012). Of note, several microRNA families have been shown to modulate Dnmt3a and Dnmt3b expression. Downregulation of miR-29 could cause higher levels of Dnmt3A in the Dnmt3l mutant background (Takada et al., 2009). Alternatively, increased Dnmt3A levels could be the indirect consequence of upregulation of the miR-290 species, which target the Dnmt repressor Rbl2 (Benetti et al., 2008; Sinkkonen et al., 2008).

Although we cannot conclude that increased Dnmt3A availability is responsible for the acquisition of normal methylation patterns, it could mechanistically counterbalance Dnmt3L depletion. In biochemical assays, maximal Dnmt3L-mediated stimulation is achieved at a 1:1 molar stoichiometry of Dnmt3L and Dnmt3A, but increasing Dnmt3A protein alone can also lead to increased recruitment of methyl group donors (SAM). Moreover, although Dnmt3L interacts with Dnmt3A2 to form tetrameric structures, Dnmt3A2 is capable of auto-assembling into large polymers of up to 30 monomers in solution and likely also in vivo (Kareta et al., 2006; Holz-Schietinger and Reich, 2010). Our work therefore illustrates that the embryo may be flexible in its ability to use various combinations of Dnmt3A, Dnmt3B and Dnmt3L homo- and heterocomplexes, depending on relative partner availability. By contrast, the assembly of Dnmt3L-associated complexes is strictly required for proper methylation in the germline (Bourc'his et al., 2001; Bourc'his and Bestor, 2004; Kaneda et al., 2004; Kaneda et al., 2010). Interestingly, DNA replication might play an enhancing role in de novo methylation and alleviate the need for Dnmt3L stimulation: indeed, whereas embryonic methylation occurs in rapidly cycling cells, gametic methylation occurs in non-dividing cell types during male and female germ cell development (Bourc'his et al., 2001; Bourc'his and Bestor, 2004).

To conclude, whereas abnormal methylation patterns are usually associated with dramatic outcomes, the slight delay in acquiring full methylation has no somatic impact on Dnmt3l^z− individuals, underscoring the efficiency and precision of Dnmt3L-independent methylation in the developing embryo.

We thank Anne-Valérie Gendrel for assistance with Sequenom analysis; Renata Kozyraki for help with protein analysis in early embryos; Shoji Tajima for providing the anti-Dnmt3L antibody; Mickael Weber for sharing the MeDIP-chip data on Dnmt3a promoters; the PICT IBiSA Genetics and Developmental Biology Imaging Facility for help with microscopy; and Isabelle Grandjean and Colin Jouanneau for excellent mouse husbandry.