Dppa2 and Dppa4 counteract de novo methylation to establish a permissive epigenome for development

Early mammalian development entails genome-wide epigenome remodeling, including DNA methylation erasure and reacquisition, which facilitates developmental competence. To uncover the mechanisms that orchestrate DNA methylation dynamics, we coupled a single-cell ratiometric DNA methylation reporter with unbiased CRISPR screening in murine embryonic stem cells (ESCs). We identify key genes and regulatory pathways that drive global DNA hypomethylation, and characterize roles for Cop1 and Dusp6. We also identify Dppa2 and Dppa4 as essential safeguards of focal epigenetic states. In their absence, developmental genes and evolutionarily young LINE1 elements, which are specifically bound by DPPA2, lose H3K4me3 and gain ectopic de novo DNA methylation in pluripotent cells. Consequently, lineage-associated genes and LINE1 acquire a repressive epigenetic memory, which renders them incompetent for activation during future lineage specification. Dppa2/4 thereby sculpt the pluripotent epigenome by facilitating H3K4me3 and bivalency to counteract de novo methylation, a function co-opted by evolutionarily young LINE1 to evade epigenetic decommissioning. Coupling of a single-cell ratiometric DNA methylation reporter with an unbiased CRISPR screen in ESCs identifies key genes and regulatory pathways that drive global DNA hypomethylation and establishes Dppa2 and Dppa4 as essential safeguards of focal epigenetic states.

M ammalian fertilization is accompanied by widespread epigenetic remodeling of inherited genomes, including global DNA demethylation and reorganization of chromatin landscapes [1][2][3][4] . This epigenetic resetting equalizes the distinct parental epigenomes and also correlates with the emergence of naive pluripotency, implying that epigenome remodeling is central to the establishment of developmental 'competence' . Such competence confers the capacity of the genome to transcriptionally respond to future inductive signals for multiple lineages. This is particularly critical for lineage-associated genes that must be transiently repressed during pluripotent phases whilst remaining competent (primed) for robust activation in subsets of forthcoming cell fates 5,6 . Indeed, the importance of a permissive epigenome is supported by observations of impaired or reduced developmental competence after somatic cell nuclear transfer (SCNT) or in induced pluripotent stem cells, which are susceptible to incomplete epigenetic resetting 7,8 . Investigating the complex mechanisms that underpin epigenome (re)programming is therefore an important focus towards understanding developmental potency.
Several lines of evidence indicate that resetting DNA methylation (DNAme) during development is mediated by parallel mechanisms 9 . Amongst these, repression of the maintenance DNA methylation machinery is central and appears to occur through post-translational regulation of UHRF1 (refs. 10,11 ), at least in part via STELLA activity [12][13][14] . This is further supported by PRDM14, which suppresses de novo methylases and is necessary for DNA hypomethylation in naive pluripotent cells 15,16 . In parallel, replication-independent DNAme erasure occurs on both the maternal and paternal genome 1 . Counterintuitively, de novo methylation remains active throughout epigenetic reprogramming but is offset, in part, via TET proteins 17 . These collective mechanisms contribute towards resetting the epigenome, but also present an opportunity for transposable elements (TE), such as LINE1, to mobilize due to epigenetic derestriction. Such LINE1 activation has been linked with key developmental events [18][19][20] , but could also represent a hazard to the genome if left unrestrained 21,22 . Epigenetic (re)programming therefore probably strikes a balance between genome-wide resetting to a competent state for development, and targeted regulation. Nevertheless, a complete understanding of the mechanisms that crosstalk to remodel the epigenome, how they interact to balance focal and global effects, and the full repertoire of genes involved, is lacking.
Here we coupled a single-cell ratiometric reporter of cellular DNA methylation status with CRISPR screening to unbiasedly identify the gene networks that underpin DNAme remodeling. In doing so we identify upstream regulators of global DNAme erasure in pluripotent cells. We also identify Dppa2 and Dppa4 as key genes that safeguard against focal de novo DNA methylation and epigenetic silencing at lineage-associated genes by integrating chromatin states, and consequently confer developmental competence. Remarkably, full-length LINE1 elements appear to have exapted this Dppa2/4 function to escape epigenetic surveillance and enable competence for precocious activation, potentially highlighting an evolving genomic conflict. translation factors) or inherent cell-cell variance (for example, cell cycle stage).
To test ratiometric eRGM, we developed a model of developmentally induced DNA demethylation. Here, ESCs are maintained in a titrated 2i/L (t2i/L) condition (see Methods) to promote high global levels of DNA methylation (range, 64-58%), and are then transitioned to 2i/L status to induce global demethylation (range, 30-44%; P = 0.0002) (Fig. 1b). Importantly, global DNA demethylation after switching from t2i/L to 2i/L occurs without changes in cell identity, as judged by the transcriptome, which is in contrast to the switch from conventional serum/leukemia inhibitory factor (LIF) to 2i/L that constitutes a major transcriptional shift [24][25][26] ( Fig. 1b and Extended Data Fig. 1b). Moreover, the induced DNA hypomethylation pattern is well correlated with developmentally imposed DNA demethylation in vivo (Extended Data Fig. 1c). Thus, the t2i/L→2i/L model specifically captures an authentic global epigenetic transition, including global DNA demethylation, without changes in cell identity that could confound a screen for epigenome regulators.
We next examined the capacity for detection of DNA demethylation events in single cells by generating independent ESC lines carrying the ratiometric eRGM system. In t2i/L, eRGM was silenced in >95% of cells, consistent with high global DNA methylation. In contrast, eRGM exhibited a progressive activation concomitant with induced DNA methylation erasure in 2i/L, leading to eRGM activation in 12% of single cells after 3 days (d), 67% after 6 d and in >95%   Fig. 1 | Developmental model and ratiometric reporter for DNA demethylation cRiSPR screening. a, schematic of the ratiometric eRgM real-time DNAme reporter. gFP is OFF in hypermethylated cells but expressed following hypomethylation, driven by a methylation-dependent imprinted promoter downstream of a DNAme sensor. mCherry remains active, establishing a single-cell ratio score. b, PCA of the transcriptomes (RNA-seq) of s/L (serum/ Lif)-, t2i/L-and 2i/L-cultured EsCs and EpiLCs, shaded by global DNA methylation level as determined by LuMA. c-e, Representative single-cell ratiometric scores of eRgM (n = 250 cells) during transition from t2i/L to 2i/L (c), after tAM-induced Dnmt1 KO in t2i/L (d) or after EpiLC differentiation (e). Bars indicate median with 95% confidence intervals. upper panels show corresponding changes in global DNA methylation, shown as mean ± s.d. of duplicate independent experiments. f, significance scores (RRA) (see Methods for details) of CRIsPR knockout (KO) screen candidates required for eRgM (LINE1) activation after DNA demethylation transition from t2i/L to 2i/L. g, stRINg clustering of significant candidates (FDR < 0.05) required for eRgM activation from independent reprogramming screens. of cells following complete DNA hypomethylation at 12 d (Fig. 1c). Independent eRGM lines exhibited consistent response to induced hypomethylation (Extended Data Fig. 1d). Notably, Ef1α-mCherry did not alter expression during this transition, enabling its use as a ratiometric normalizer (Extended Data Fig. 1d). To further confirm that eRGM directly reports cellular DNA methylation status, we used ESCs wherein tamoxifen (TAM) drives Cre-mediated deletion of Dnmt1 (cDKO) and, consequently, global DNA demethylation occurs independent of culture conditions 27 . Following TAM exposure, we observed a strong and progressive activation of eRGM amongst single cells concomitant with cDKO-induced DNA hypomethylation (Fig. 1d).
Finally, we tested whether eRGM can also respond reciprocally to acquisition of DNA methylation by inducing differentiation of hypomethylated ESCs (in 2i/L) into hypermethylated epiblast-like cells (EpiLCs; global DNAme 33→75%). Here the reporter initiated rapid silencing in parallel with induction of DNA hypermethylation (Fig. 1e). We conclude that the enhanced ratiometric reporter of genomic DNAme (eRGM) represents a single-cell readout for dynamic transitions of cellular DNA methylation status.

A CRISPR screen for regulators of dynamic DNA methylation.
To identify factors critical for DNA methylation resetting, we generated independent ESC lines carrying ratiometric eRGM and spCas9 and introduced into them a CRISPR knockout guide RNA library 28 . To validate the strategy, we isolated ESCs that activated eRGM under hypermethylated (t2i/L) conditions, which is predicted to identify factors necessary to maintain DNA methylation and/or epigenetic silencing. This revealed that top hits comprised the key machinery for maintenance of DNA methylation, including Dnmt1 (rank 5, false discovery rate (FDR) = 0.00049) and Uhrf1 (rank 48, FDR = 0.066), unbiasedly confirming eRGM sensitivity to DNA hypomethylation (Extended Data Fig. 1e). We also identified chromatin-mediated silencers including Setdb1 (rank 51, FDR = 0.073) and the HUSH complex (Mphosph8, rank 6; Morc2a, rank 9; Fam208a, rank 13). These data support eRGM specificity for the detection of developmental epigenome regulators, including those of cellular DNA methylation status.
We next aimed to identify factors contributing to resetting the epigenome at focal or global scales. We induced global DNA demethylation and isolated individual ESCs that failed to ratiometrically activate eRGM, indicative of a failure to undergo epigenetic resetting (Fig. 1f). Importantly, this population was highly enriched for knockout of Prdm14 (rank 16, FDR = 0.0006), the key regulator known to instruct global DNA demethylation 15 , as well as its heterodimeric cofactor Cbfa2t2a (rank 20, FDR = 0.0006) 29,30 , supporting the sensitivity of the strategy for identification of reprogramming factors (Fig. 1f). Moreover, screens of independent eRGM ESC lines identified highly correlated (P = 0.01, Spearman relative ranking algorithm (RRA)) candidates (Extended Data Fig. 2a), suggesting that this system is robust. We therefore intersected significant hits (FDR < 0.05, fold-change > 3) from independent screens to define 56 core candidate genes linked with resetting the epigenome (Supplementary Table 1).
To validate the CRISPR screen hits, we generated knockout ESC populations for 24 selected candidates and transitioned them to hypomethylated conditions. Strikingly, knockout of each candidate resulted in a degree of impaired eRGM activation, implying altered epigenome remodeling in their absence (Fig. 2a). This effect was robust, since we generated additional knockouts in an independent eRGM line with similar outcomes (Extended Data Fig. 2c). Interestingly, the response kinetics of eRGM during transition to 2i/L varied amongst candidate knockout. For example, Jak1, Dppa2, Dppa4 and Brd4 mutants failed to activate eRGM per se, indicating a general block. In contrast, other candidate knockouts including Dusp6, Kdm3a, Nufip2 and Cop1 exhibited late-onset heterogeneous activation amongst single cells, implying delayed demethylation dynamics and reduced robustness in their absence. (Fig. 2b and Extended Data Fig. 2d). These validations suggest that candidate factors influence both the kinetics and absolute response of eRGM.
We next used luminometric CpG methylation assay (LUMA) to quantitatively assess global DNA methylation levels. Consistent with eRGM, we found that knockout of 20 of 24 candidate factors resulted in impaired global DNA demethylation across independent mutant lines (Fig. 2c). Amongst these is the known epigenetic regulator Prdm14, which maintained 58-64% global DNAme relative to hypomethylated wild-type (WT) control (39%), as well as Cbfa2t2 (52-54%). Novel candidates that exhibited substantially elevated DNAme following knockout and transition to 2i/L include the phosphatase Dusp6 (56-60%), the tyrosine kinase Jak1 (65-70%), the epigenetic regulator Brd4 (59-59%) and the E3 ubiquitin ligase Cop1 (54-56%). These data suggest that our screen is sufficient to identify critical components of gene regulatory networks that contribute to driving complete DNA demethylation in naive ESCs.
Dusp6 and Cop1 promote global DNA hypomethylation. To further investigate the role of candidates Dusp6 and Cop1 in epigenetic transitions, we generated independent clonal knockout ESC lines. DUSP6 is a phosphatase that acts downstream of MEK to attenuate the ERK signal cascade, whilst COP1 mediates ubiquitination and proteasomal degradation of target proteins 32,33 . We used enzymatic methyl-sequencing (EM-seq) 34 , an enhanced equivalent of bisulfite-sequencing (BS-seq), to chart the global DNA methylome in WT, Dusp6 -/and Cop1 -/naive ESCs, which confirmed that mutant lines remain hypermethylated in in 2i/L (Dusp6 -/-67%, Cop1 -/-58%) (Fig. 2d). Notably, elevated DNAme is distributed equivalently across genomic features including promoters, repeats and intergenic regions, indicating a general impairment of DNA demethylation rather than failure in locus-specific resetting (Fig. 2d,e).
Mechanistically, both Cop1 -/and Dusp6 -/-ESCs exhibited transcriptional upregulation of the de novo methylation machinery (Dnmt3a, Dnmt3b, Dnmt3L), whilst Dusp6 -/cells additionally downregulate Stella, which together may contribute to impaired global DNA demethylation (Fig. 2f). Consistent with elevated DNAme in mutants, we observed strong repression of DNA methylation-dependent (germline) genes whilst naive genes were largely unaffected, implying no underlying change to pluripotency networks (Fig. 2f). However, we did observe inappropriate expression of some early developmental genes in Cop1 -/and Dusp6 -/-ESCs, and their transcriptomes clustered separately by principle component analysis (PCA), which may partly reflect their disrupted epigenetic state (Fig. 2g). Taken together, our screen identifies and validates genes and pathways involved in the promotion of genome-scale DNA methylation transitions.
Dppa2/4 protect against aberrant de novo DNA methylation. Our screen is designed to identify both global and focal epigenome regulators. Amongst the eRGM screen hits, the paralogs Dppa2 and Dppa4 (hereafter, Dppa2/4) consistently ranked in the top five, suggesting a role in modulation of epigenome dynamics. However, in contrast to other candidates (for example, Dusp6, Cop1 and Jak1), deletion of Dppa2 or Dppa4 did not affect genome-scale DNA demethylation (Fig. 2c). This could imply that, rather than a global influence, Dppa2/4 modulate the methylation landscape at specific genomic features during pluripotent phases, when they are specifically expressed (Fig. 3a).

Pluripotency Early differentiation
Signaling targets   many gene promoters that usually remain strictly unmethylated at all developmental stages (Fig. 3d). This indicates that, rather than impaired DNA demethylation per se in Dppa2/4 mutants, there is aberrant de novo methylation activity that could establish DNA methylation 'epimutations' .
To determine whether such epimutations persist during differentiation, we profiled EpiLCs, which correspond to a formative state that has undergone genomic remethylation. We observed that the hypermethylated sites established in Dppa2 -/or Dppa4 -/-ESCs are retained in EpiLCs, whilst additional loci also acquire aberrant de novo methylation, including promoters (Extended Data Fig. 3b). Indeed, direct analysis identified 354 differentially methylated promoters (DMPs) in Dppa2/4 mutants. Gene ontology analysis revealed these DMPs are enriched specifically for developmental processes (multicellular organism development FDR = 0.0053; developmental process FDR = 0.024; anatomical structure development FDR = 0.01) (Extended Data Fig. 3c). For example, Hand1, Tnxb and Gnmt1 are key nodes for lineage-restricted cells and usually remain demethylated in all tissues, but acquire significant promoter hypermethylation (>90%) in Dppa2 -/and Dppa4 -/-ESCs and EpiLCs ( Fig. 3d and Extended Data Fig. 3d). Intriguingly, in addition to developmental gene promoters, we observed that DMRs are enriched specifically for the 5' end of full-length (>5 kb) LINE1 elements (indicative of evolutionarily young LINEs), but not for truncated LINEs and long terminal repeat (LTR) elements (Extended Data Fig. 3b). Direct analysis identified 1,131 differentially methylated LINE1 (DML), of which >80% are L1Md_T (Fig. 3d). We used pyrosequencing to verify that LINE (L1Md_T), as well as the promoters of Hand1, Nkx2-5 and Col16a1, acquire DNA methylation in Dppa2 -/naive ESCs (Fig. 3e), confirming a disrupted epigenomic landscape.   Because DMRs are focal rather than global, we next asked whether they reflect localized DPPA2/4 activity. We performed CUT&RUN for DPPA2 binding in WT cells and observed that genomic occupancy is strikingly increased over sites that become hypermethylated DMRs in Dppa2/4 -/cells, suggesting that DPPA2 may act proximally to sculpt the DNA methylome (Fig. 3f). Indeed, DPPA2-binding peaks (n = 28,338) are significantly enriched specifically over gene promoters (P < 0.001) and the 5' end of full-length LINE1 elements (>5 kb) (P < 0.001) (Fig. 3g), consistent with DMR associations. Overall, promoters and LINE account for >65% of DPPA2 genomic occupancy. The latter enrichment is specific for full-length LINE, since DPPA2 is not enriched at truncated LINEs (Fig. 3h). Amongst promoters, DPPA2 exhibits a preference for CpG-dense loci, which is reflected by its GC-rich binding motifs and preference for CpG island (CGI) promoters ( Fig. 3i and Extended Data Fig. 3e). Moreover DPPA2-binding profiles are highly correlated between ESCs and EpiLCs, implying that the additional DMR in the latter reflects their higher de novo activity rather than DPPA2 redistribution (Extended Data Fig. 3f). Notably, DPPA2 binding was observed at the sensor region used for eRGM (Extended Data Fig. 3g), explaining why a focal DNAme modulator was identified in the screen. In summary, Dppa2 and Dppa4 have a nonredundant role in protecting a target subset of developmentally associated promoters and full-length LINE elements from acquiring de novo DNA hypermethylation during naive and formative pluripotency phases, when Dppa2/4 are specifically expressed.

DPPA2/4 binding establishes a permissive chromatin state.
To understand the broader chromatin features associated with DPPA2 binding, and susceptibility to hypermethylation, we used CUT&RUN to profile H3K4me3, H3K27me3 and H3K9me3. DPPA2 occupancy correlates with strong H3K4me3 enrichment in WT cells, across all binding sites (Extended Data Fig. 4a). H3K27me3 is also enriched at a subset of DPPA2-bound sites, establishing bivalent states, but H3K9me3 is largely depleted. Strikingly, H3K4me3 enrichment at DPPA2-bound promoters occurs irrespective of expression state in both ESCs and EpiLCs (Fig. 4a), implying that DPPA2 may directly target H3K4me3, rather than H3K4me3 reflecting expression status of DPPA2-bound sites. Importantly, hypermethylated loci in Dppa2/4 mutants (DMR) correspond to genomic regions that are H3K4me3-enriched and DPPA2-bound in WT (Extended Data Fig. 4b). Taken together this suggests a potential connection between DPPA2 occupancy, H3K4me3 and DNA methylation status.
To investigate this further, we assayed H3K4me3, H3K27me3 and H3K9me3 in Dppa2 -/and Dppa4 -/cells. Remarkably, deletion of Dppa2 or Dppa4 resulted in a dramatic loss of H3K4me3 across a significant subset of DPPA2-bound sites in both ESCs and EpiLCs, whilst the remaining sites were apparently unaffected (Fig. 4b). The subset of DPPA2 sites that lost H3K4me3 were enriched for full-length LINE1 elements and promoters (Fig. 4b).
Moreover, the effect on H3K4me3 was specific, since there was no significant change of H3K9me3 in mutants whilst H3K27me3 was reduced at some loci such as Txnb but relatively unaffected at most (Fig. 4c,d and Extended Data Fig. 4c-e). In general, loci that lost H3K4me3 and gained DNAme in Dppa2/4 mutants are associated with specific absolute levels of H3K4me3 enrichment in WT cells-intermediate for promoters and (relatively) high for full-length LINE1 (Fig. 4e). Because H3K4me3 can directly impair de novo DNA methylation 35,36 , the dramatic depletion of H3K4me3 in Dppa2/4-mutants may enable aberrant DNA hypermethylation. In support of this, DMPs and DML correspond to loci that exhibit reduced H3K4me3 (Fig. 4c,d). Furthermore, by investigating all promoters and LINE1, we observed a marked negative correlation (P < 2.2 × 10 -16 ) between progressive H3K4me3 loss and DNAme gain following Dppa2 knockout (Fig. 4f).
In summary, abrogation of Dppa2/4 is linked with depletion in H3K4me3 at a specific subset of DPPA2 target loci, which directly correlates with acquisition of aberrant DNA hypermethylation. Dppa2/4 could therefore integrate chromatin states to safeguard the pluripotent epigenome, particularly at developmentally associated genes and LINE1 elements.
Dppa2/4 ensure a competent epigenome for developmental expression. To understand the relevance of Dppa2/4-mediated epigenome surveillance for developmental competence, we assessed the transcriptome. Dppa2/4 -/-ESCs expressed an unperturbed naive pluripotency network and underwent apparently normal exit from pluripotency, since naive markers (Nanog, Klf2 and Prdm14) were appropriately downregulated whilst formative markers (Fgf5, Wnt3, Dnmt3a) were upregulated as expected (Fig. 5a). Moreover, there was no difference in expression of the DNA methylation machinery or chromatin-modifying genes which, taken together, implies that Dppa2/4 do not have an overarching influence on naive, early differentiation or epigenome gene regulatory networks (Fig. 5a).
Globally, Dppa2/4 -/naive ESCs exhibit a gene expression signature distinct but broadly comparable to WT, with 269 and 245 differentially expressed genes (DEG) in Dppa2 and Dppa4 knockout, respectively, primarily downregulated (85 and 82%, respectively) ( Fig. 5b and Extended Data Fig. 5a). Induction of EpiLCs leads to a more divergent transcriptome as judged by PCA (Fig. 5c), with 801 and 611 DEG, again preferentially downregulated. Significantly, gene ontology indicated that these DEG in EpiLCs specifically relate to developmental processes (single multicellular organism process FDR = 0.000004, cell differentiation FDR = 0.00012) (Extended Data Fig. 5b), which reflects a general failure to activate genes involved in lineage-specific functions, particularly mesendoderm regulators. For example, Hand 1, Cldn9, Tnxb and others all failed to initiate primed expression in mutant EpiLCs ( Fig. 5a and Extended Data Fig. 5c). This could be linked with the ectopic promoter DNA methylation acquired in the preceding ESC state. Indeed, the collective DMP geneset (n = 354), which comprises many of the same mesendoderm genes including Hand1, Tnxb, Ttl9, Cldn9 and Gnmt, is significantly upregulated in WT EpiLCs (P = 0.018) consistent with priming of developmental genes but, strikingly, fails to initiate activation in either Dppa2 -/-(P = 0.29) or Dppa4 -/-EpiLCs (P = 0.40), suggesting they have lost competence for expression (Fig. 5d).
To understand whether the impaired expression of mesendoderm genes in EpiLCs represents a delay in their activation or stable silencing, we induced endoderm differentiation for 12 d. Dppa2 -/cells appeared morphologically equivalent to WT and activated master endoderm regulators, including Emb and Foxa1, with comparable dynamics, indicating no general impairment in differentiation (Fig. 5e). However, endoderm-associated genes, including Gnmt, Nkx2-5, Col16a1 and Hand1, exhibited a highly significant failure to activate in mutants, even after 12 d of endoderm induction, implying an absolute blockade in their response (Fig. 5f). Importantly, H3K27me3 H3K9me3 ESCs EpiLCs

Dppa2/4 -/-EpiLCs
ESCs d3 d6 d12 Dppa2 is rapidly downregulated after 3 d of endoderm differentiation, but impaired gene upregulation manifests at later timepoints, suggesting a memory of previous DPPA2 activity (Fig. 5g). Indeed, pyrosequencing revealed that ectopic promoter DNA methylation established in ESCs propagates through to day-12 endoderm (Fig. 5h). Together, this indicates that the absence of Dppa2/4 in pluripotent phases leads to impaired competence for gene activation during later differentiation. Importantly, cell fate transition per se appears unperturbed, but rather specific genes within the developmental program are rendered stably epigenetically silenced.
We next asked whether repetitive element activation is also affected by Dppa2/4, since many evolutionary young LINE1 become hypermethylated and lose H3K4me3 in their absence (Figs. 3d and 4c). We observed a particularly striking downregulation of full-length (>5 kb) LINE families (L1Md_T, L1Md_A and L1Md_Gf) in Dppa2/4 mutant ESCs and EpiLCs (Fig. 5i and Extended Data Fig. 5d). We confirmed this using independent quantitative PCR with reverse transcription (RT-qPCR), which showed that disruption of Dppa2 in ESCs and EpiLCs leads to extensive repression of L1Md_T (Extended Data Fig. 5e) whilst MERVL is also strongly repressed, consistent with recent reports 37 . IAP and other LTR elements were largely unaffected. These data suggest that the same Dppa2/4-dependent system that maintains epigenetic competence at developmental promoters may have been co-opted by LINE1 elements to evade epigenetic silencing in pluripotent phases.
H3K4me3 and DNAme interact to confer functional epigenetic memory. Last, we investigated whether induced DNA methylation and H3K4me3 loss is functionally instructive for subsequent gene-silencing memory. We noted that depletion of promoter H3K4me3 and gain of DNAme are both correlated with gene repression in Dppa2 -/cells (Extended Data Fig. 6a). Moreover, altered DNAme and H3K4me3 are also directly anticorrelated (Fig. 4f), implying a hierarchy of robust molecular changes following Dppa2/4 abrogation. To determine whether acquired DNA methylation (and H3K4me3 loss) could instruct gene repression, we deleted Dnmt1 in Dppa2 -/-ESCs to generate compound mutants (Dnmt1 -/-Dppa2 -/-) that are hypomethylated and predicted to erase ectopic DNAme at developmental promoters and LINE1. Remarkably, analysis of the DMP geneset that acquired aberrant promoter methylation and silencing in Dppa2 -/-EpiLCs revealed that additional deletion of Dnmt1 partially rescues their activation block in EpiLCs. This effect is significant (P = 0.024) among genes with CGI promoters, but not among non-CGI promoters (P = 0.23) (Fig. 6a). Moreover, we observed reactivation of L1Md_T elements in Dnmt1 -/-Dppa2 -/-ESCs and EpiLCs (Fig. 6b). These data imply that ectopic DNAme in Dppa2 -/cells is instructive, at least at some CpG-dense genes and LINE1, and directly impairs their response to inductive activating signals.
We next investigated whether the depletion of H3K4me3 in Dppa2 mutant cells is also affected in Dnmt1 -/-Dppa2 -/mutants and, surprisingly, observed reinstatement of H3K4me3 at a subset of promoters and most LINE1 elements (Fig. 6c,d and Extended Data Fig. 6b). This indicates a potentially complex interplay whereby absence of Dppa2/4 leads to loss of H3K4me3, enabling aberrant DNA methylation, but that subsequent removal of ectopic DNAme tips the balance back, allowing H3K4me3 to reaccumulate through Dppa2/4-independent mechanisms (Extended Data Fig. 6c). More generally, we show that altered DNAme and H3K4me3 in the absence of Dppa2/4 can propagate through differentiation to manifest as instructive gene silencing at future developmental stages, long after the epimutation is established. Dppa2/4 therefore act as a safeguarding system during dynamic epigenome remodeling phases to ensure epigenetic competence for impending multilineage development.

Discussion
Here, we have established a ratiometric reporter of DNA methylation (eRGM) that is enhanced to enable unbiased CRISPR screening. By coupling this with an ESC model of developmental DNAme reprogramming, we identify and validate epigenome modulators that influence global DNA demethylation events and also focal DNAme states (Dppa2 and Dppa4). The global regulators relate to diverse pathways such as m6A RNA methylation (for example, Mettl3, Virma and Ytdhf2), LIF signaling (for example, Lifr, Jak1 and Stat3) and E3 ubiquitin ligases, which presumably exert influence through acting as upstream regulators as recently reported for Nudt21 (ref. 38 ). Amongst these, we show that the phosphatase Dusp6 is necessary for completion of global DNA demethylation in naive ESCs. DUSP6 functions to attenuate MEK/ERK signaling 32 , which is linked with DNA methylation 26,39 , suggesting a probable connection. Indeed X-linked DUSP9 contributes to female-specific ESC hypomethylation by influencing MEK/ERK 40 , and DUSP6 could play a comparable, but nonredundant, role in thresholding MEK/ ERK more generally in pluripotent cells to promote epigenome erasure. Interestingly another screen hit, Med24 (rank 3), also impacts MEK/ERK signaling 41 . Mechanistically, Dusp6 may function via modulation of the de novo methylation machinery and/or Stella. The mechanism through which the E3 ubiquitin ligase Cop1 modulates global epigenetic state is less clear, but could relate to regulation of the stability of proteins involved in maintaining DNAme, such as UHRF1 (ref. 11 ).
In addition to global regulators we identify Dppa2/4, which we show guards against ectopic de novo methylation activity at key genomic sites during phases of both DNAme erasure (in naive ESCs) and remethylation (EpiLCs). Previous studies have shown that Dppa2/4 overexpression enhances induced pluripotent stem cell generation, and they are linked with facilitating the two-cell program via modulation of Dux, suggesting broad functional roles 37,42,43 . Nevertheless, Dppa2/4 mutant mice undergo normal embryogenesis but die perinatally due to aberrant gene repression in lung, where Dppa2/4 are not expressed 44 , implying that the phenotypically relevant activity of Dppa2/4 is ensuring that lineage-associated genes are appropriately primed during early development. We dissect this molecularly by demonstrating that the absence of Dppa2 or Dppa4 leads to a marked loss of H3K4me3 and parallel acquisition of de novo DNA methylation at developmental genes and LINE1 elements, which propagates to manifest as epigenetic silencing in lineage-restricted cells. The equivalence of DPPA2 and DPPA4 probably reflects that they reciprocally stabilize each other (Extended Data Fig. 3), whilst the unusual association of DPPA2 outside classical pluripotency networks could underpin its distinct role 45 .
Mechanistically, several lines of evidence suggest that DPPA2 targets H3K4me3. First, DPPA2 genomic binding sites are highly H3K4me3 enriched, irrespective of underlying transcription. Second, H3K4me3 is lost at a subset of sites following Dppa2 deletion; and third, DPPA2 is reported to interact with the H3K4me methylase MLL2 (ref. 46 ). Because H3K4me3 restricts the recruitment of de novo DNA methyltransferases 35,36 , the depletion of H3K4me3 at these loci may enable access for ectopic DNA methylation to follow 47 . Indeed, we observe a striking correlation between the degree of H3K4me3 loss and DNAme gain in Dppa2/4 mutants. Consistently, a Dnmt3a engineered to tolerate H3K4me3 enables aberrant de novo methylation at developmental genes 48 , supporting a model whereby Dppa2-dependent H3K4me3 protects against DNAme. Such a system is probably necessary due to widespread de novo methylation activity throughout developmental (re)programming phases 17 , which targets specific genomic compartments such as TE, but must also be restrained 49 . Notably, by counteracting de novo methylation, DPPA2/4 may also facilitate H3K27me3 accumulation and bivalency because a subset of loci exhibit H3K27me3 depletion in Dppa2/4 mutants. This could reflect either direct loss of targeting by DPPA2/4 or inhibition of PRC2 activity by acquired DNAme [50][51][52] . Functionally, the acquisition of ectopic DNAme and loss of H3K4me3 appear to be instructive for epigenetic silencing following differentiation, at least at some CpG-dense promoters. Indeed, erasure of acquired promoter DNAme partially rescues expression defects and also reinstates H3K4me3. This suggests a switch-like interdependency whereby DPPA2-dependent H3K4me3 impairs DNAme, but acquired DNAme reciprocally prevents H3K4me3 accumulation, potentially underpinning stable transcriptional memory at target genes.
In addition to maintaining epigenetic competence at developmental genes, we observe that evolutionarily young LINE1 elements rely directly on Dppa2 for H3K4me3 and their activity. This may reflect a strategy of successful LINE1 elements that have acquired/co-opted DPPA2-binding sites at their 5' ends to protect against host-directed epigenetic silencing. This would in turn enable expression of full-length LINE1s during early development when Dppa2/4 are expressed, which aligns well with the optimal period for retrotransposition 53 . This scenario would represent a genomic conflict, whereby Dppa2/4 activity is critical for epigenetic competence of lineage-associated genes, and therefore essential for viability, but also renders full-length LINE1s transcriptionally competent-a potential threat to genome integrity. An alternative scenario is that Dppa2/4-mediated activation of LINE1 reflects an important developmental role for the host genome. For example, LINE1 activation has been linked with establishment of zygotic chromatin accessibility, X-chromosome inactivation and regulation of the two-cell program [18][19][20] . Consequently, Dppa2/4-dependent LINE1 activation could represent exaptation by host systems to exploit LINE1 functionality at critical developmental points. In any case, Dppa2/4 are placed at the center of an epigenetic competence circuit in pluripotent cells that facilitates expression of both LINE1 and developmental genes.
In summary, we characterize upstream gene networks that influence global DNA methylation erasure, and additionally uncover a complementary pathway that protects against the counterforce of uncontrolled de novo methylation during (re)programming to ensure developmental competence.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41594-020-0445-1. It also reports on specific regulators of focal DNAme that operate at Dazl. The antisense orientation of Dazl ensures that it bears no promoter activity on the reporter per se, but simply instructs the DNAme status of the adjacent Kcnq1ot1 imprinted promoter in line with global levels. To establish a ratiometric system, an additional construct was generated by cloning the methylation-insensitive EF1a promoter upstream of an mCherry::H2B cassette into a piggyBac vector to generate pPB-EF1a-H2B::mCherry. Correct assembly and sequences was confirmed by tiled Sanger sequencing, and vectors were amplified and purified by endotoxin-free midi-preparations.

Generation of eRGM ESC lines. Embryonic stem cell lines carrying floxed
Dnmt1 alleles 27 and WT ESCs were transfected with pPB-asDazlsensor-Kcnq1ot 1-H2B::GFP (in silico DNA methylated with M.SssI), pPB-EF1a-H2B::mCherry, pPB-spCas9-Hygro 56 and PBase using Lipofectamine 3000. Transfected cells were selected for spCas9 integration in titrated 2i/L using Hygromycin (250 µg ml -1 ) for 5 d, and clonally derived cell lines were subsequently isolated and expanded. Clonal ESC lines were tested to confirm single-copy integrations by qPCR on genomic DNA, and their response to DNA demethylation was confirmed by evaluation of eRGM (GFP and mCherry) expression using flow cytometry after culture for 7 d in either t2i/L (hypermethylated) or 2i/L (hypermethylated). Further confirmation of eRGM response was determined by the addition of TAM (800 nM) for 6 d in t2i/L to induce conditional Dnmt1 knockout and DNA hypomethylation. Clonal ESC lines exhibiting the best dynamic range of eRGM response were selected, and two independent lines (eRGM nos. 1 and 2) were used for CRISPR screening.
CRISPR screen. Lentiviral particles carrying the Brie gRNA library 28 were produced by transfection of Lenti-X HEK 293T with pPax2 plasmid, pMD2.G plasmid and the Brie library plasmid, with Lipofectamine 3000 in a BSL2 tissue culture facility. Lentivirus-containing supernatant (medium) was harvested at 48 and 72 h after transfection, and clarified by filtering through a 0.22-μm, low-protein-binding unit. Viral particles were concentrated using a Lenti-X concentrator, according to the manufacturer's instructions, and resuspended in NDIFF 227. Lentiviral activity and efficiency were determined by transduction of ESCs across a titration curve, and assay of cell survival following puromycin selection for virally encoded integration of a resistance cassette. To generate knockout library cell lines, 7 × 10 7 ESCs of eRGM nos. 1 and 2 cultured in t2i/L were transduced with a predetermined number of lentiviral particles carrying the Brie genome-wide CRISPR knockout single guide RNA library (n = 78,637) 28 to ensure ~45% infection efficiency (>400-fold guide RNA coverage). Transduced cells were selected for with puromycin (1.2 µg ml -1 ) for 7 d in t2i/L. The minimum population of cells was maintained at >3.2 × 10 7 during passaging (>400-fold coverage) to ensure maintenance of library coverage. To initiate the screen knockout library, eRGM cell lines were transitioned into 2i/L for 12 d to drive extensive DNA hypomethylation. At 12 d, GFP-negative cells (defined as the lowest 1% of GFP expression) that also maintained normal mCherry expressiontogether indicative of incomplete epigenetic resetting-were purified by flow cytometry (291,248 and 237,121 for eRGM nos. 1 and 2, respectively). We additionally collected total unsorted cells (>3 × 10 7 ) from both t2i/L and 2i/L, and GFP-positive cells (top 1%) from t2i/L (indicative of loss of epigenetic silencing), as controls. Genomic DNA was isolated from purified populations using the Quick-DNA microprep plus kit (Zymo Research, no. D3020) or a DNeasy blood and tissue kit (Qiagen, no. 69504). Integrated gRNAs from each population were amplified from genomic DNA using custom primers with the P7 flow cell o ve rh angs: 5 '-CA AG CA GA AG AC GG CA TA CG AG AT NN NN NN NN GT GA CT GG  AG TT CA GA CG TG TG CT CT TC CG ATCTTCTACTATTCTTTCCCCTGCAC  TGT-3' (8 base- Table 1 for the full list of gRNA used), and selected with puromycin (1.2 µg ml -1 ) for 60 h. Transfected cells were subsequently seeded at low density (1,000 cells per 9.6 cm 2 ) for single clone isolation. After clonal expansion, successful homozygous knockout lines (carrying frame-shifting indels) were confirmed by Sanger sequencing using the tracking of indels by decomposition tool 57 , by immunoblot and via functional assays. To generate population-scale knockout of multiple candidate factors (n = 24), gRNAs targeting the gene(s) of interest were cloned into a piggyBac vector containing the enhanced gRNA cassette 58 . This was cotransfected with PBase into independent eRGM lines carrying spCas9 activity using Lipofectamine 3000, following the manufacturer's recommendations. ESCs were selected for successful integration of gRNAs for 7 d with puromycin (1,2 µg ml -1 ), which drives iterative targeting of the gene of interest until indel formation is induced. We assayed the population for successful knockout by immunoblot and flow cytometry, and typically observed that >95% of individual cells within each population carried homozygous functional knockout.
Flow cytomtery. Cells were gently dissociated into single-cell suspension using TrpLE and resuspended in PBS+1% FBS (fluorescent activated cell sorting (FACS) medium) and filtered. FACS was performed using a FACS Aria III (Becton Dickinson) and FACS Diva software. For flow analysis, samples were run on Attune NxT (Thermo Fisher Scientific). Data were analyzed using FlowJo v.10.5.3 (Tree Star).
LUMA. LUMA was used to measure global DNA (CpG) methylation levels.
Briefly 200-500 ng of purified genomic DNA was split equally and subjected to two parallel 4-h restriction digests at 37 °C: digestion A, HpaII/EcoRI; digestion B; MspI/EcoRI, in which EcoR1 is included as an internal reference. An equal volume of annealing buffer was added, and samples were loaded into a PyroMark Q24 Advanced pyrosequencer to quantitate the protruding ends from each digestion, using the dispensation order GTGTGTCACACAGTGTGT. Percentage DNA methylation was calculated by comparing the EcoRI normalized HpaII signal intensity ratio to the normalized MspI signal intensity ratio using the formula Immunoblot. Cellular protein was extracted using RIPA buffer (Sigma, no. R0278) with protease inhibitors (Roche, no. 4693159001) at 4 °C for 30 min. After centrifugation at full speed, cell lysis supernatant was collected and Bolt LDS sample buffer (Thermo Fisher Scientific, no. B0007) and Bolt reducing agent (Thermo Fisher Scientific, no. B0004) were added to the samples. These were heated at 70 °C for 10 min and loaded onto 4-12% Bis-Tris gel (Thermo Fisher Scientific, no. NW04125BOX). Proteins were separated by 150-V electrophoresis for 30 min and blotted onto a polyvinylidene difluoride membrane using the iBlot Dry 2 blotting system. The membrane was blocked in 5% milk/PBS for 1 h at room temperature, followed by incubation with primary antibody (dilution 1:500-1:1,000; see Nature Research Reporting Summary for antibodies used) with 5% milk/PBS at 4 °C overnight and agitation. After washing twice with PBS/0.1% Tween, the membrane was incubated for 1 h at room temperature with a horseradish peroxidase-linked secondary antibody diluted 1:10,000 in 5% milk/PBS. The membrane was washed three times with PBS 0.1% Tween, and Pierce ECL immunoblot plus solution (Thermo Fisher Scientific, no. 32132) was added to the membrane for 5 min before imaging using the ChemiDoc XRS+ system (Bio-Rad).
RT-qPCR. Total RNA was isolated with RNeasy (Qiagen) and used to synthesize complementary DNA with a mixture of random hexamers and reverse