De novo DNA methylation of endogenous retroviruses is shaped by KRAB-ZFPs/KAP1 and ESET

Endogenous retroviruses (ERVs) account for close to 10% of mammalian genomes (Waterston et al., 2002) and are both drivers of evolution and threats to genetic integrity because of their ability, first, to retrotranspose and, second, to alter the expression of neighbouring genes through cis-acting transcriptional influences. In mice, ERVs display residual retrotransposition activity, leading to polymorphic integrations and differential gene regulation between mouse strains (Qin et al., 2010; Takabatake et al., 2008; Zhang et al., 2008). Moreover, around 10% of spontaneous mutations in inbred mice are linked to ERVs (Maksakova et al., 2006), for example dactylaplasia that results from a MusD neo-insertion (Friedli et al., 2008; Kano et al., 2007). In humans, the non-LTR retroelement LINE1 (L1) retains some activity and, remarkably, individuals with Rett syndrome caused by mutations in the DNA methylation-binding protein MeCP2 are more prone to L1 retrotransposition (Muotri et al., 2010).

ERVs are duly inactivated during early embryogenesis by histone modifications and de novo DNA methylation (reviewed by Rowe and Trono, 2011). This process counteracts the prior genome-wide erasure of DNA methylation that begins at the zygote phase, leaving transposons in a low methylated state, including L1s and, to a lesser extent, intracisternal A-type particles (IAPs) (Feng et al., 2010; Lane et al., 2003). De novo DNA methylation is key to ERV repression from plants to mammals, and involves methylation of cytosine residues at CpG dinucleotides by the enzymes DNMT3A and DNMT3B, which act in conjunction with their catalytically inactive co-factor DNMT3L (Chedin et al., 2002; Okano et al., 1999; Suetake et al., 2004). Later in development and in adult tissues, levels of de novo methyltransferases are reduced compared with those in pre-implantation embryos (Carlson et al., 1992), yet pre-established DNA methylation patterns are perpetuated during DNA replication by the maintenance DNA methyltransferase DNMT1, which is recruited to hemi-methylated DNA by UHRF1 (Bestor et al., 1988; Bostick et al., 2007; Sharif et al., 2007). DNA methylation is particularly crucial for constraining the transcription of some ERVs, including IAPs, the expression of which become uncontrolled following inactivation of DNA methylation machinery (Bourc'his and Bestor, 2004; Chen et al., 2003; Gaudet et al., 2004; Walsh et al., 1998). However, IAPs are still silenced in DNMT1 knockout ES cells unless they are differentiated (Hutnick et al., 2010), highlighting the particular importance of DNA methylation later in development. Interestingly, alterations in ERV DNA methylation patterns impact on the expression of neighbouring genes (Duhl et al., 1994; Macfarlan et al., 2011; Macfarlan et al., 2012; Michaud et al., 1994; Rebollo et al., 2011).

DNA methylation is conditioned by the density and spacing of CpG dinucleotides, can be affected by DNA-binding proteins such as CTCF and REST, and is influenced by the histone code, with methylation of histone 3 at lysine 4 (H3K4) preventing the docking of the DNMT3A-DNMT3L complex (Glass et al., 2009; Jia et al., 2007; Lienert et al., 2011; Okitsu and Hsieh, 2007; Ooi et al., 2007; Stadler et al., 2011; Weber et al., 2007). The factors and pathways that recruit de novo DNA methyltransferases to specific genomic targets, notably ERVs, remain largely undefined, although it has recently been shown that DNMT3L and the lysine methyltransferase G9a are factors required for the initiation of proviral de novo DNA methylation (Leung et al., 2011; Ooi et al., 2010).

Here, we have investigated the role played in this process by KAP1 (KRAB-associated protein 1, also known as tripartite motif protein 28 – TRIM28 – or TIF1β), a co-factor to the sequence-specific DNA-binding KRAB zinc-finger proteins (ZFPs), a large family of tetrapod-restricted transcriptional factors that co-evolved with ERVs (Emerson and Thomas, 2009; Friedman et al., 1996; Thomas and Schneider, 2011; Urrutia, 2003). KAP1 has previously been demonstrated to maintain ERVs in a silent state in embryonic stem (ES) cells via the histone methyltransferase ESET (also known as SETDB1) and secondary H3K9 trimethylation (Matsui et al., 2010; Rowe et al., 2010). It is also known that the artificial tethering of a KRAB domain to the vicinity of a promoter can lead to its DNA methylation in transgenic mice but not in ES cells (Wiznerowicz et al., 2007), and that maintenance of imprinting marks depends on ZFP57 and KAP1 (Li et al., 2008; Messerschmidt et al., 2012; Quenneville et al., 2011; Zuo et al., 2012). Of relevance here, the retrovirus murine leukemia virus (MLV) is known to be recognized by the mouse-specific KRAB-ZFP ZFP809 through its proline primer-binding site (PBS Pro) sequence and undergoes KAP1-dependent silencing in embryonic carcinoma (EC) cells (Wolf et al., 2008; Wolf and Goff, 2007; Wolf and Goff, 2009), although it is not known whether these factors are sufficient to promote DNA methylation. MLV-based vectors are themselves prone to DNA methylation weeks after their integration in ES or EC cells, through a process that is independent of the presence of a PBS Pro sequence (Leung et al., 2011; Niwa et al., 1983; Wang et al., 1998), as also documented for the cytomegalovirus (CMV) promoter (Meilinger et al., 2009).

Determining how specific ERVs are targeted for de novo DNA methylation is technically challenging, owing to their genomic representation in multiple copies of often differentially methylated status in ES cells and early embryos. To circumvent this difficulty, we inserted previously identified genomic targets of KAP1 derived either from the proximal region of IAPs or from PBS Pro (present in some 146 endogenous MLVs in the C57Black6 mouse genome) into lentiviral vectors, either upstream or downstream of a promoter driving a GFP reporter. In both configurations, these introduced retroviral elements were sufficient to induce not only the silencing but, in parallel, de novo DNA methylation of the flanking promoter. Methylation was remarkably rapid with an adjacent retroviral promoter (MND), occurring within 48 hours, but slower with a cellular promoter (PGK), occurring within 6 days. For PBS Pro, we could demonstrate that DNA methylation was conditioned by the expression of ZFP809. We then determined that de novo DNA methylation of introduced ERVs in ES cells was dependent on both KAP1 and ESET, and that sequence-specific ERV methylation could be recapitulated in vivo and maintained through development, with ERV sequences directing DNA methylation of an adjacent retroviral or cellular promoter. Finally, we documented that methylation footprints of genomic ERVs are affected by KAP1 loss in early embryos. These data reveal that KAP1 shapes DNA methylation signatures at ERV-containing loci in early mouse development.

All MND vectors have been described previously (Rowe et al., 2010). For ERV-directed repression of a cellular promoter, the transfer vector pRRLSIN.cPPT.PGK-GFP.WPRE (available from Addgene) was employed, which contains a human phosphoglycerate kinase 1 (PGK) promoter to distinguish it from the endogenous mouse promoter during methylation analysis. Pro and B2 sequences were cloned into this vector by annealing primers upstream (into the XhoI site) or downstream (between the BamHI and AgeI sites) of the promoter in the antisense orientation. Or IAP1 or IAP4 [described previously (Rowe et al., 2010)] were cloned upstream of the promoter into the XhoI site in the antisense orientation. Cloning was verified by sequencing (Microsynth). Vectors were produced by transient transfection of 293T cells with the transfer vector, packaging and VSVG envelope plasmids (Barde et al., 2010), and titrated on 3T3 fibroblasts.

Mouse ES cells were cultured in Glasgow Minimum Essential Media (GMEM, Sigma: G5154) with sodium pyruvate (1 mM; Sigma S8636), MEM non-essential amino acids (1×; Gibco 11140-035), L-glutamine (2 mM; Gibco 25030-024), 2-mercaptoethanol (0.1 mM; Sigma), ES cell tested FBS (Gibco 16141-079) and leukaemia inhibitory factor (LIF, 50 units/500 ml; Chemicon ESG1107). Cells were grown on 0.2% gelatin (Sigma: 48723-500G-F)-coated plates and split every 2 days. ES cell lines used were two KAP1_loxP/loxP lines called ES3 and ES6, and their derived KAP1-conditional knockout cell lines transduced with tamoxifen (4-0HT)-inducible cre (Rowe et al., 2010). Knockout cells were collected 3-4 days after treatment with 4-OHT (overnight at 1 μM; Sigma H7904) as stated. G9a parental or stable knockout ES cells and Eset-inducible knockout ES cells were from Yoichi Shinkai (Dong et al., 2008; Matsui et al., 2010; Tachibana et al., 2008) (The RIKEN Institute, Japan). Dnmt1^−/−, Dnmt3a^−/−, Dnmt3b^−/− (triple knockout or TKO) ES cells were from Masaki Okano (Tsumura et al., 2006) (The RIKEN Institute, Japan). F9 EC (embryonic carcinoma) cells, primary mouse embryonic fibroblasts (MEFs), 3T3 fibroblasts and 293T cells were also used where stated. Vector titres and GFP repression were measured by flow cytometry and for KAP1 and ESET knockout experiments, cells were stained with an anti-SSEA-1 PE-conjugated antibody or isotype control (BD Pharmingen: 560142 and 555584) and SSEA-1 high-expressing cells gated for undifferentiated cells.

Genomic DNA was converted (200 ng/sample) using an EpiTect Bisulfite kit (Qiagen: 59104) and used for PCR (primers, one tagged with biotin, were designed using PyroMark Assay Design Software 2.0). PCR products were checked by Sanger sequencing and verified on agarose gels for each experiment before immobilizing on 96-well plates using a Vacuum Prep Workstation and pyrosequencing using PyroMark Gold Reagents (Qiagen 972804; Center for Integrative Genomics, University of Lausanne, Switzerland). Results were analysed using Pyro Q-CpG Software. Primer sequences are in supplementary material Table S1.

Cells were washed with ice-cold PBS and resuspended in radioimmunoprecipitation buffer to prepare total cell extracts. Protein was quantified by BCA Protein Assay Reagents (Pierce) and normalized for loading on a 10% denaturing SDS-polyacrylamide gel. Wet transfer was performed and primary antibodies used were: anti-KAP1 (Mouse mAb, MAB3662, Chemicon), anti-IAP GAG [a kind gift from Bryan Cullen, Duke University, NC, USA (Bogerd et al., 2006)] and anti-PCNA (mouse mAb, clone PC10, catalog number NA03, Calbiochem).

Cells were seeded at 1×10⁵ per well in a 12-well plate. Then, 100 μl transfection mix was prepared per well [containing 6 μl FuGene 6 (Promega) and 1 μg DNA in serum-free media] and incubated at room temperature for 15 minutes before adding drop-wise to cells.

RNA was extracted and qRT-PCR performed as described previously (Rowe et al., 2010). Primer specificity was confirmed by dissociation curves and samples normalized to Titin, Gapdh or Eef1a1 in Fig. 2D because it reacts with both human and mouse Eef1a1. For in vivo experiments, genomic DNA was extracted from embryos to measure DNA copy numbers by performing Taqman qPCR for HIV Gag, GFP and Titin as a normalizer, in comparison with a titration curve of plasmid containing sequences of HIV Gag, GFP and Titin. A cell line with a known number of vector copies was used as a control. Results presented are means of values obtained for HIV Gag and GFP primers. See supplementary material Table S1 for primer sequences.

Lentiviral vectors for transgenesis were prepared using Episerf medium (Invitrogen 10732022); the particle concentration was obtained by p24 ELISA (Perkin-Elmer: NEK050B001KT) and the infectious titre determined on HCT116 cells using GFP flow cytometry. Vectors for each transgenesis experiment were produced and titred in parallel in order to be in the same range: Fig. 4 titres, 2×10⁹ (IAP2 MND) and 4×10⁹ (MND); Fig. 5 titres, 1.4×10⁹ (PGK Pro) and 1.2×10⁹ (PGK); supplementary material Fig. S4B titres, 2.3×10⁹ (IAP4 PGK) and 2.4×10⁹ (IAP1 PGK). Transgenesis was performed by perivitelline injection of vectors into fertilized oocytes that were transferred to foster mothers (strain B6D2F1/J) and then recovered at E13. Photographs were taken using the same saturation, gain and exposure settings for all embryos, and GFP results displayed all represent the same image processing settings.

GraphPad Prism version 4.00 (www.graphpad.com) was used for all statistical analyses. For all DNA methylation analyses, two groups (as detailed in the figure legends) were compared across multiple CpG positions using paired two-tailed t-tests (except where stated). Any additional statistical analyses are specified in the figure legends.

As repeat-derived sequences are largely methylated in ES cells, we set up a method to measure the de novo DNA methylation directed at ERV-specific elements by cloning these into lentiviral vectors, a gene transfer system itself not prone to spontaneous methylation, compared with MLV-based vectors (Wang et al., 1998). To quantify levels of de novo DNA methylation, we used bisulfite pyrosequencing and verified this technique with primers for endogenous ERVs and the Oct4 promoter, confirming the latter to be methylated in mouse embryonic fibroblasts (MEFs) but not in ES cells; conversely, the bulk of ERVs harboured this modification in both cell types (Fig. 1A). In ES cells transduced with lentiviral vectors, we confirmed that PBS Pro and previously described KAP1-sensitive IAP elements mediated repression of an adjacent MND retroviral promoter, whether placed upstream or downstream (Fig. 1B; supplementary material Fig. S1A), as previously reported (Rowe et al., 2010). Bisulfite analyses then further revealed that de novo DNA methylation systematically correlated with repression, and that it surprisingly was present at 4 days post-transduction (Fig. 1C). A time-course study showed that this process was already under way by day 3 and then progressively increased, with mean levels across the promoter reaching around 75% at day 7 (supplementary material Fig. S1B). Control vectors, whether they contained no KAP1-targeted retroviral sequence, the point-mutated B2 PBS Pro derivative or the previously identified KAP1-insensitive IAP1 element (Rowe et al., 2010), underwent neither repression nor DNA methylation. Moreover, in 3T3 fibroblasts, no ERV-directed promoter methylation was detected, suggesting this process may be restricted to ES cells (Fig. 1C).

KAP1 has previously been shown to repress ERVs in ES cells and in early embryos (Rowe et al., 2010) and is known to be recruited to specific DNA sequences through KRAB-ZFPs (Friedman et al., 1996; Thomas and Schneider, 2011). One characterized KRAB-ZFP is ZFP809, which targets KAP1 to the PBS Pro sequence (Wolf and Goff, 2007; Wolf and Goff, 2009). As we have previously found ERVs to be regulated by KAP1 in mouse ES cells but not in MEFs (Rowe et al., 2010), we hypothesized that this difference may correlate with the expression profiles of KRAB-ZFPs, especially as KAP1 is itself still present in MEFs. We therefore used mRNA-sequencing data from ES cells and MEFs to compare expression levels of KRAB-ZFPs between these cell types. Although most KRAB-ZFPs were enriched in ES cells compared with MEFs (although there is no evidence that any of them repress ERVs) (with mean expression levels at 3.421 and 1.4242, respectively), ZFP809 displayed similar levels in the two cell types (Fig. 2A; data not shown). Correspondingly, when we compared the silencing of the PBS Pro and IAP vectors in MEFs, we found the former but not the latter underwent repression in these cells (Fig. 2B), consistent with the lack of activation of endogenous IAPs in MEFs upon KAP1 removal (Rowe et al., 2010). Indeed, we propose that IAPs are already stably methylated in MEFs so it is not necessary for their cognate KRAB-ZFP to be maintained. PBS Pro-induced repression was accompanied by de novo DNA methylation in MEFs, albeit at lower levels than in ES cells (Fig. 2C). In order to demonstrate that this mechanism is determined by the presence of the KRAB-ZFP, we complemented human 293T cells with mouse ZFP809, leading to repression but also to significant PBS Pro-dependent DNA methylation of the adjacent promoter (Fig. 2D-F). This level of DNA methylation, however, is not sufficient to maintain silencing, which is reversible (supplementary material Fig. S2).

In order to determine the role of KAP1 in ERV de novo DNA methylation, we used a previously described conditional knockout ES cell line (Rowe et al., 2010). As Kap1 deletion is ultimately lethal in early embryos and in ES cells (Rowe et al., 2010) (supplementary material Fig. S3), experiments were performed within 3 days of inducing this process (a time-point established from supplementary material Fig. S1B). At that point, both repression and DNA methylation of the lentivirally introduced, IAP- or Pro PBS-flanked promoter could be documented in control cells, whereas both processes were abrogated by KAP1 removal (Fig. 3A-C). Similar experiments in Eset-conditional knockout ES cells demonstrated that this histone methyltransferase is also required for ERV-directed repression and de novo DNA methylation (Fig. 3D-F). Both KAP1 and ESET are thus necessary not only for the maintenance of proviral silencing (Leung et al., 2011; Matsui et al., 2010; Rowe et al., 2010), but also required initially for de novo DNA methylation of introduced ERV sequences.

De novo DNA methylation patterns, which are established early in embryogenesis, are maintained later in development through the action of DNMT1 (Gaudet et al., 2004; Walsh et al., 1998). We investigated whether DNA methylation marks promoted by KAP1 in ES cells could also be induced in vivo, and whether in this setting they were perpetuated until late in embryogenesis. To achieve this, control and KAP1-sensitive IAP sequence-containing MND-GFP lentiviral vectors were injected into fertilized mouse oocytes, and the resulting embryos were collected at E13 to assess integration, repression and promoter methylation. Although all embryos containing the control vector were bright or dull green, none of the IAP LV-harbouring embryos expressed GFP to detectable levels, whether examined by microscopy or by flow cytometry (Fig. 4A,B). Of note, MEFs transgenic for the non-repressed promoter displayed variability in GFP intensity within the population, likely owing to KAP1-independent silencing mechanisms, including those relating to integration site differences (Barklis et al., 1986; Ellis, 2005). Strikingly, DNA methylation at E13 mirrored data obtained in ES cells, with high promoter methylation for all IAP vector-harbouring embryos, whereas DNA methylation of the lentivirally introduced promoter was lower and more variable in the absence of an adjacent KAP1-tethering element (Fig. 4C). These results suggest that in early embryos, KAP1 machinery acts as a lock to ensure active DNA methylation and thus robust ERV silencing throughout development.

As cellular promoters can lie in close proximity to ERVs in the genome and ERV methylation status can influence cellular gene expression (Duhl et al., 1994; Macfarlan et al., 2011; Macfarlan et al., 2012; Michaud et al., 1994; Rebollo et al., 2011), we assessed whether our KAP1-dependent ERV sequences could also direct de novo DNA methylation of a cellular promoter. PBS Pro could induce some repression of a juxtaposed PGK (phosphogylcerate) promoter, albeit less efficiently than for MND, and with only minimal, if any, DNA methylation at day 3 post-transduction (Fig. 5A). However, by day 6, repression was clear (around sevenfold, supplementary material Fig. S4) so we used this time-point to demonstrate that both PBS Pro and IAP sequences could direct de novo DNA methylation of the cellular promoter with levels further increasing after 12 days (Fig. 5B). Furthermore, PBS Pro or a KAP1-sensitive IAP sequence induced repression and DNA methylation of the adjacent PGK promoter when lentiviral vectors containing these elements were used to generate transgenic embryos (Fig. 5C; supplementary material Fig. S4B).

Our Southern blot analyses previously failed to detect significant global loss of DNA methylation at ERVs upon Kap1 deletion (Rowe et al., 2010) but, here, we revisited this issue using the more-sensitive bisulfite pyrosequencing technique to examine ERVs in ES cells and early embryos. As expected, IAPs became highly overexpressed following KAP1 removal in ES cells, which here we record to translate as a marked accumulation of IAP GAG protein (Fig. 6A). We then measured IAP methylation, using the Oct4 promoter as a control (Fig. 6B): whereas global IAP LTRs were unaffected in Kap1-knockout MEFs, they displayed a mild but significant decrease in knockout ES cells in line with the minimal decrease previously observed at ERVs in Eset-knockout ES cells (Matsui et al., 2010). A comparatively greater loss of IAP DNA methylation was measured in an ES cell line stably deleted for the G9a histone methyltransferase (Dong et al., 2008; Tachibana et al., 2008). However, this cell line grows continuously, allowing cumulative decreases in DNA methylation, whereas Kap1-deleted cells divide only a few times before dying. The overexpression of IAPs in Kap1-deleted cells is most probably a consequence of the removal of histone repressive marks such as H3K9me3 (Matsui et al., 2010; Rowe et al., 2010), especially as we confirmed with conventional bisulphite sequencing at one IAP locus that no molecules are completely or even largely demethylated (supplementary material Fig. S5). However, it is possible that such a small decrease in DNA methylation contributes to the phenotype, particularly as we find that 5-Aza treatment of MEFs induces fairly conservative decreases in IAP DNA methylation (supplementary material Fig. S6), yet a massive upregulation of their transcription (Fig. 7B). Interestingly, we also found, using two different primer sets, that DNA methylation levels at ERVs were mildly but significantly lower in E5.5 Kap1 knockout embryos compared with wild-type littermates (Fig. 6C; supplementary material Fig. S7). Although only 1 out of 23 further knockout embryos was obtained (at E4.5), probably owing to the high lethality of this mutation, this embryo also displayed less DNA methylation at IAPs than its two control littermates (P=0.0002 and 0.0014, respectively, data not shown). Altogether, these data support a role for KAP1 in genome-wide ERV DNA methylation during early embryogenesis.

Finally, in order to address the relevance of DNA methylation of ERV sequences and endogenous ERVs in development, we derived MEFs from transgenic embryos in which our ERV-derived proviruses were integrated. Treatment with the DNA methyltransferase inhibitor 5-aza (but not the HDAC inhibitor, TSA) was sufficient to relieve silencing and this was true for both ERV sequences newly introduced within the context of lentiviral vectors and endogenous IAPs that were upregulated by up to around 700-fold (Fig. 7). To determine whether DNA methylation already played a role in silencing ERVs at an earlier developmental stage, we also examined DNMT triple knockout (TKO) ES cells (Tsumura et al., 2006), in which all three DNA methyltransferases are deleted (supplementary material Fig. S8). Levels of IAP GAG protein and of IAP and MERV-L RNA were significantly higher in these than in control ES cells (supplementary material Fig. S8A,B), consistent with a previous study (Matsui et al., 2010). It was recently reported that in Dnmt1 KO ES cells and in another ES cell line where all DNMT proteins were depleted (by knockout of Dnmt3a/3b and knockdown of Dnmt1) (Meissner et al., 2005), IAP protein was detected only upon LIF (leukemia inhibitory factor) withdrawal for six or more days, a time point at which OCT4 protein was undetected in differentiated cultures by immunofluoresence (Hutnick et al., 2010). However, the TKO ES cells we have used here were maintained in standard ES cell conditions (see Materials and methods) and were largely undifferentiated as indicated by comparable levels of OCT4 protein to wild type (supplementary material Fig. S8A) as well as high surface expression of SSEA1 (supplementary material Fig. S8C), in contrast to ES cells cultured without LIF, which downregulated this marker (supplementary material Fig. S8D). In addition, these cells have previously been found to conserve their self-renewal ability (Tsumura et al., 2006). We still cannot exclude the possibility that some differentiating cells contribute more prominently to the observed IAP upregulation. Importantly, the overexpression of these elements in ES cells (around 40-fold) is modest compared with that seen in MEFs (Fig. 7), supporting a crucial role for DNA methylation in ERV silencing later in development (Gaudet et al., 2004; Walsh et al., 1998), in contrast to in ES cells where KAP1-dependent histone methylation is present (Hutnick et al., 2010; Karimi et al., 2011; Matsui et al., 2010; Rowe et al., 2010) (reviewed by Leung and Lorincz, 2012; Rowe and Trono, 2011). Indeed, both 5-aza treatment and Kap1 knockout need to be performed in order to induce such a dramatic upregulation of IAPs in ES cells (around 400-fold) (Rowe et al., 2010).

The present work demonstrates the involvement of sequence-specific KRAB-ZFPs, KAP1 and ESET in establishing early embryonic DNA methylation patterns at ERV sequences that are subsequently maintained through embryogenesis. This suggests that these factors are required during early development first for the histone-based silencing and then for the de novo methylation of genomic ERVs, even though the evidence provided here is limited to newly introduced ERV sequences. Our results support a model (Fig. 6D) whereby sequences located in the 5′ region of ERV genomes recruit specific KRAB-ZFPs, KAP1, ESET and most likely other chromatin modifiers, the conjugated action of which leads to transcriptional repression and de novo promoter methylation. This phenomenon implies the secondary recruitment of de novo DNA methyltransferases, consistent with the documented immunoprecipitation of KAP1 with DNMT3A and DNMT3B (Li et al., 2008; Quenneville et al., 2011; Zuo et al., 2012), and the reported interaction of DNMT3A with the B-Box coiled coil domain of KAP1 (Zuo et al., 2012). As ERV DNA methylation requires ESET, it could be that this enzyme directly recruits DNMT3AB and DNMT3L. In that respect, histone methyltransferases have previously been revealed to act as docking sites for DNA methylation machinery (Dong et al., 2008; Tachibana et al., 2008) and, importantly, ESET has been proposed to bind directly to DNMT3A and tether it to promoters, leading to their DNA methylation-mediated silencing (Li et al., 2006). Alternatively, de novo DNA methylation may result indirectly from a lack of transcription and active marks such as H3K4me3 (Jia et al., 2007; Ooi et al., 2007), owing to the recruitment of H3K9me3 and histone deacetylation at ERVs.

We reveal here that KAP1-recruiting ERV sequences are able to trigger repression and de novo DNA methylation of a strong cellular promoter, as well as of a retroviral promoter, at least within the context of newly introduced lentiviral vectors. The slower kinetics observed with the PGK promoter, compared with its MND counterpart, may be due to partial protection of its CpG-dense core by transcription factors, as recently proposed (Lienert et al., 2011; Stadler et al., 2011). These data indicate that KAP1-mediated silencing of ERVs may dampen expression from nearby cellular transcription units by inducing promoter CpG methylation. This is supported by our observation that the methylation status of CpG islands correlates with their immediate proximity to KAP1-recruiting sites in ES cells (Quenneville et al., 2012). Relevant to ERVs, in a recent study, polymorphic copies of ERVs across mouse strains were employed to determine whether histone and DNA methylation marks could spread from ERVs to cellular promoters. Results showed that, whereas H3K9me3 marks could spread several kilobases, it was infrequent for DNA methylation to spread from an IAP to neighbouring gene promoters, except where genes lie within 500 bp of an IAP (Rebollo et al., 2011). Our results now implicate a KAP1/ESET complex in this mechanism. As DNA methylation rarely spreads from ERVs as far as nearby promoters in the genome, in future work it will be interesting to investigate the possible presence of barrier elements protecting cellular genes from the spread of repressive marks, including DNA methylation, which is nucleated at ERV sequences.

Here, we observed significant DNA methylation of a PBS Pro-, retroviral promoter-containing lentiviral vector in MEFs, which express similar levels of ZFP809 to ES cells, as well as in 293T cells engineered to express high levels of this KRAB-ZFP. By contrast, a KAP1-restricted IAP sequence induced neither repression nor de novo DNA methylation of the MND promoter in MEFs, suggesting that the cognate KRAB-ZFP is not expressed in these targets. The matching KRAB-ZFP(s) could be any of those enriched in ES cells except ZFP568, which has already been shown not to be responsible for repressing IAPs in vivo (Shibata et al., 2011). The induction of only low methylation in MEFs and 293T cells, compared with ES cells, by ZFP809 explains why the repression is reversible, as previously noted when KRAB is artificially tethered to a promoter in cell lines (Barde et al., 2009; Groner et al., 2010; Quenneville et al., 2012; Wiznerowicz and Trono, 2003).

Surprisingly, we found very little decrease in DNA methylation levels at resident ERVs in KAP1 knockout embryos, despite the role we have demonstrated for KAP1 in shaping DNA methylation patterns at ERV sequences in embryogenesis. It could be that the loss of DNA methylation that follows the loss of repressive histone marks cannot be measured accurately owing to the earlier lethality of knockout embryos (at or before E5.5) (Cammas et al., 2000) or that the majority of DNA methylation at genomic ERVs is KAP1 independent at this stage. Along these lines, as IAPs are relatively resistant to genome-wide DNA demethylation in preimplantation embryos (Lane et al., 2003), methylation may be maintained without requiring de novo methyltransferase activity. Importantly, this and previous studies (Matsui et al., 2010; Rowe et al., 2010) indicate that histone modifications, rather than DNA methylation, are primarily responsible for silencing ERVs in ES cells, as IAPs become immediately and markedly upregulated upon KAP1 or ESET removal in ES cells, in the absence of any significant change in their DNA methylation status. It is noteworthy that ERVs exhibit a range of DNA methylation rates in ES cells, suggesting that some, including many IAPs, may escape demethylation, whereas others may become demethylated but rapidly re-methylated following their KRAB/KAP1 recognition. Moreover, the impact of DNA methylation per se on transcription, at least during this period, needs to be further investigated, as blocking histone-based repression of IAPs suffices to induce their upregulation. It is possible that subtle alterations at specific positions, which are crucial to block transcription, have so far escaped analyses, or that DNA methylation does not have the same consequences in ES and somatic cells: for example, if some factors important for repressing transcription at DNA methylated loci are expressed in differentiated but not in undifferentiated cells.

Our in vivo experiments with vectors reveal that ERV methylation patterns are faithfully copied late into development, leading to extremely high DNA methylation levels (upwards of 80%) that mimic natural levels at IAP LTRs resident in the genome. The self-perpetuating nature of DNA methylation alleviates the need for the continuous expression of cognate ERV-binding KRAB-ZFP repressors. It also explains why the maintenance of DNA methylation at ERVs throughout development is crucial for suppressing their transcription, notably for IAPs (Walsh et al., 1998). In summary, the present work implies that ERV sequence recognition by KRAB-ZFPs/KAP1 and the lysine methyltransferase ESET is necessary to target ERV-containing loci for rapid de novo DNA methylation in early development, leading to stable site-specific DNA methylation signatures across the genome.

We thank P. V. Maillard for advice; A. Reymond and J. Chrast (from the Center for Integrative Genomics, University of Lausanne, Switzerland) for use of their pyrosequencer; and Carine Gubelmann and Andrea Corsinotti for the ZFP809 plasmid. We also thank D. Schubeler for supplying us with the G9a knockout and parental ES cells, originally from Y. Shinkai (The Riken Institute, Japan), who provided us with the Eset-inducible knockout ES cells; and B. Cullen (Duke University) for the anti-IAP GAG antibody. We thank F. Spitz (EMBL, Heidelberg), in whose lab T.A. performed the experiment depicted in supplementary material Fig. S5.