Reciprocal, methylation-dependent binding of Zfp57 and Gzf1 safeguards Dlk1-Dio3 imprinting during developmental reprogramming

Abstract

Genomic imprinting secures parent-specific gene expression through differential DNA methylation at imprinted control regions (ICRs). However, how unmethylated alleles resist de novo methylation remains unclear. Using an allelic Dlk1-Dio3 ICR methylation reporter and genome-wide loss-of-function screening, we identify the zinc finger protein GZF1 that binds the unmethylated maternal ICR and protects it from de novo methylation via a regulatory element containing GZF1 and ZFP57 motifs that mediates mutually exclusive, methylation-dependent binding. Loss of either factor causes reciprocal imprinting failure: Gzf1 loss induces maternal allele methylation, H3K4me3 depletion, and silencing of maternal transcripts, whereas Zfp57 loss results in maternalization. Remarkably, GZF1 protects the unmethylated ICR from de novo methylation in both oocytes and embryos, and its loss leads to perinatal death consistent with paternalization of the maternal allele. Together, our findings establish a reciprocal mechanism that maintains parental epigenetic asymmetry across both imprint establishment and embryonic reprogramming.

The protein GZF1 binds the unmethylated paternal inherited copy of an imprinted gene region and protects it from silencing, acting reciprocally with ZFP57. Loss of GZF1 disrupts genomic imprinting and causes growth defects in mice.

Introduction

Genomic imprinting orchestrates the monoallelic expression of certain genes exclusively according to their parental origin, producing epialleles with distinct expression profiles despite identical genetic sequences^1,2. Imprints are established during germline development by differential DNA methylation in sperm and oocytes. These marks resist global demethylation after fertilization, forming parent-specific differentially methylated regions (DMRs), also known as Imprinting Control Regions (ICRs), which secure the precise gene dosage required for normal embryonic development^2–7. A defining feature of this process is the recognition of the methylated allele, which hinges on an intricate interplay between DNA methylation and histone modifications, orchestrated by trans-acting factors^8,9. Among these factors are the KRAB zinc finger proteins ZFP57 and ZNF445, which recognize methylated CpG-containing hexanucleotide motifs at most ICRs^10–13, subsequently recruiting TRIM28 (KAP1) and SETDB1 to deposit H3K9me3 repressive marks^9,14. This epigenetic state is faithfully transmitted across cell divisions by actively recruiting the DNA methylation maintenance machinery involving UHRF1 and DNMT1⁴.

Alongside the required maintenance of methylated imprint alleles until the blastocyst stage, it is equally crucial to protect unmethylated alleles from the wave of de novo DNA methylation that follows implantation, to ensure proper parent-of-origin monoallelic expression^5,15. Oocyte-derived (maternally methylated) ICRs are CpG-rich and reside within gene promoters. Although the exact mechanism remains unclear, these regions are likely protected from de novo DNA methylation by the binding of trans-acting factors, such as transcription factors (TFs) and components of the Trithorax or Polycomb complexes, which antagonize the activity of DNMT3A and DNMT3B. This protection is similar to that observed at other CpG-rich promoters across the genome^16–18. By contrast, sperm-derived (paternally methylated) ICRs are less CpG-dense and reside in intergenic regions lacking robust promoter-driven transcription, rendering the mechanisms that safeguard these loci from de novo methylation less well defined^2–4. Proposed mechanisms—ranging from R-loop formation^19,20 and bivalent chromatin states^18,21, to short-lived noncoding RNAs^22,23—have been demonstrated primarily in cell lines and their relevance to normal development remains to be determined. In addition to post-implantation protection, the mechanisms preventing methylation of paternal ICRs during oocyte growth remain poorly understood. While maternal ICRs are established by the de novo methyltransferase complex DNMT3A-DNMT3L, paternal ICRs are refractory to its activity. In analogy to factors like PGC7/Stella^24,25 and ZFP57^9,10, which recognize and protect methylated imprints after fertilization, it is plausible that sequence-specific factors similarly operate during oogenesis to selectively preserve the unmethylated state of paternal ICRs²⁶.

To identify potential factors that may be involved in protecting the unmethylated state of paternal ICRs, we focused on the paternally imprinted Dlk1–Dio3 locus, which harbors an intergenic DMR (IG-DMR) that functions as an ICR regulating parent-of-origin expression of multiple genes across this ~1 Mb cluster. The IG-DMR was previously shown to be functionally subdivided into two major elements: a CpG island (hereafter termed IG-CGI) at the 5′ portion, and a downstream, low-density CpG region termed the Transcriptional Regulatory Element (IG-TRE), which regulates expression of the long noncoding-RNA Gtl2^27,28 (Fig. 1a). Through targeted genetic mutations, we previously demonstrated that the germline-derived IG-CGI functions as an upstream element controlling the methylation state of both the IG-TRE and the downstream Gtl2 DMR in cis²⁹. The mechanism that maintains the methylated paternal IG-CGI allele has been extensively studied, primarily involving the sequence-specific binding of ZFP57 at this region^10–13. Insights into the regulation of the unmethylated maternal IG-CGI allele have come primarily from studies in mouse embryonic stem cells (mESCs), suggesting a role for locally produced noncoding RNAs²³ and the recruitment of Tet demethylating enzymes^27,30 in preventing de novo methylation. However, the relevance of these maternal allele-associated mechanisms to IG-DMR regulation during normal development remains to be elucidated.

Fig. 1. Genome-wide CRISPR screen identifies regulators of allele-specific methylation at the Dlk1-Dio3 IG-DMR.

a Screening workflow. Top, structure of the Dlk1-Dio3 cluster on the paternal (blue) and maternal (red) chromosomes showing the intergenic differentially methylated region (IG-DMR) and adjacent CpG islands (CGIs). Open circle lollipops schematically represent unmethylated CpG and closed indicate methylation. Bottom-left, dual-color reporter cassette in which the methylation-sensitive minimal Snrpn promoter drives tdTomato (Tom) from the maternal allele and GFP from the paternal allele. Bottom-center, mouse ESCs (P0) are transduced with a genome-wide Brie gRNA lentiviral library³³. Bottom-right, reporter-defined populations are isolated by FACS after six passages (P6) for DNA sequencing. b Representative flow-cytometry plot at passage 6 (P6; n = 2 independent transductions) resolving Tom⁺GFP^-, Tom⁺GFP⁺, and Tom^-GFP^- sub-populations. c, e Volcano plots of gene-level enrichment (positive selection), aggregating the multiple sgRNAs per gene with the MAGeCK robust rank aggregation (α-RRA) algorithm⁵⁴(one-sided; Benjamini–Hochberg false-discovery-rate correction for multiple comparisons). Dashed lines, log₂ fold-change = 1 and P = 0.05. Genes labeled in black were selected for further validation. d, f Reporter composition after individual knock-out validation as indicated. Additional tested genes are presented in (Supplementary Fig. 2c, d). Source data are provided as a Source Data file.

Here, we used allele-specific IG-CGI methylation reporter mESCs^31,32 to perform an unbiased CRISPR loss-of-function screen and identified Gzf1, a zinc finger protein that binds the maternal IG-DMR in a sequence- and methylation-state-dependent manner and protects it from de novo methylation. In vivo, Gzf1 knockout (KO) mice show substantial perinatal loss and IG-DMR hypermethylation consistent with paternalization of the maternal allele, while surviving animals display intermediate methylation at the Gtl2-DMR and postnatal growth retardation. We further find that Gzf1-null oocytes acquire heterogeneous hypermethylation at the IG-CGI, supporting a role for Gzf1 in safeguarding the unmethylated maternal allele during oogenesis. Together, our findings support a model in which two sequence-specific factors - Zfp57 and Gzf1 - safeguard distinct epigenetic configurations on each parental allele, ensuring stable imprinting through opposing waves of epigenetic remodeling.

Results

A CRISPR screening platform to identify putative trans-regulators of the IG-CGI

We previously established a reporter system that utilizes the methylation-sensitive Snrpn minimal promoter to drive fluorescent proteins, enabling faithful readout of endogenous methylation dynamics at genomic regions of choice, in living cells³¹. In this study, we sought to utilize this system as an experimental platform for unbiased screening of factors regulating the allelic methylation state of the IG-CGI. To this end, we used cells in which the reporter was inserted downstream of the IG-CGI in F1 hybrid (129×CAST) mESCs, enabling allele-specific readout of DNA methylation dynamics. Thus, in these cells, tdTomato (Tom) expression indicates an unmethylated maternal (129) allele, and GFP expression indicates an unmethylated paternal (CAST) allele³². Intact imprinting is therefore reflected by Tom⁺/GFP⁻ expression, biallelic hypomethylation by Tom⁺/GFP⁺, and biallelic hypermethylation by loss of both fluorescent signals (Tom⁻/GFP⁻) (Fig. 1a). Previously, it had been shown that upon continuous culture, a small fraction of cells became double-positive or double-negative, reflecting naturally occurring loss of imprinting in mESCs. Yet, these subpopulations did not exhibit a growth advantage over successive passages³² (Supplementary Fig. 1a). To confirm the reporter’s responsiveness to DNA methylation loss, we sequentially infected dual reporter mESCs with viruses encoding Cas9 and a gRNA targeting Dnmt1 (Supplementary Fig. 1b). Following Dnmt1 disruption, the fraction of GFP expressing cells gradually increased over successive passages, consistent with replication-dependent passive demethylation of the paternal allele (Supplementary Fig. 1c). These results provided initial indication that the dual-reporter system could be used to capture the dynamics of allele-specific IG-CGI methylation changes and can be applied to high-throughput CRISPR loss-of-function screening.

Genome-wide loss-of-function screen recovers core regulators of IG-CGI methylation

We next performed a genome-wide CRISPR/Cas9 KO screen to identify factors required to maintain Methylation of the IG-CGI on the paternal allele. Reporter-expressing mESCs were transduced with Cas9 and a whole genome gRNA library, which targets 19,674 genes in the mouse genome using 4 different gRNAs per gene³³. We detected 99.5% of gRNAs in both the input plasmid library and in the infected cells at 48 h post-infection (P0 cells) (Supplementary Fig. 2a). The transfected cells did not show a pronounced shift in the distribution of fluorescent subpopulations after six passages - as expected, due to the small number of genes hypothesized to act as imprint maintainers (Fig. 1b). Following 6 passages, we sorted the cells according to three reporter states: Tom⁺GFP⁻, Tom⁺GFP⁺, and Tom⁻GFP⁻. Subsequently, genomic DNA was extracted, and gRNA frequencies quantified by next-generation sequencing (NGS) (Supplementary Fig. 2b). In the Tom⁺GFP⁺ fraction, gRNA enrichment analysis by MAGeCK (see Methods) found Zfp57, Uhrf1, and Dnmt1 to be the highest enriched genes. Significantly, these three genes are all well known for their function in DNA methylation maintenance at the IG-CGI^9,10,34–37 (Fig. 1c, Supplementary Data 1, 2). To further examine and validate candidate factors, we separately infected reporter cells with viruses encoding individual gRNAs. Following puromycin selection for viral integration, polyclonally infected cultures were examined for GFP expression over multiple passages. Whereas targeting Dnmt1 resulted in robust activation of the paternal GFP allele (up to 56%), reflecting incomplete Dnmt1 disruption in the polyclonal population, additional gene candidates selected for their putative roles in DNA-binding or chromatin-modifying functions did not result in reporter activation (Fig. 1d and Supplementary Fig. 2c).

Gzf1 and Qrich1 are candidate regulators of the unmethylated IG-CGI allele

Next, we examined potential factors that may safeguard the IG-CGI maternal allele from acquiring de novo methylation. To this end, we sorted the Tom⁻GFP⁻ subpopulation from our pooled screen, corresponding to maternal allele hypermethylation, and examined gRNAs enriched in this fraction compared to the unsorted pool (Fig. 1e, Supplementary Data 3, 4). From these, we prioritized candidate genes implicated in DNA-binding or chromatin-modifying functions for validation through infection with individual gRNAs, as above. Of the nine genes tested, GDNF-inducible zinc finger protein 1 (Gzf1) and Glutamine-rich protein 1 (Qrich1) consistently drove a robust shift toward Tom⁻GFP⁻ reporter activity (Fig. 1f and Supplementary Fig. 2d). Notably, the genetic screen did not identify any of the Tet genes, which were recently implicated in protecting the IG-CGI maternal allele from de novo methylation^27,30. One potential explanation for this result is functional redundancy among Tet family members. Gzf1 disruption increased the Tom⁻GFP⁻ population to ~54% by passage six, while Qrich1-KO caused a 23% increase, indicating potential involvement of both factors in safeguarding the maternal allele. As Gzf1 has been suggested to function as a sequence-specific transcriptional repressor³⁸, we focused our efforts on characterizing its role in regulating imprinting at the Dlk1-Dio3 locus.

Gzf1 loss paternalizes the Dlk1-Dio3 maternal allele, whereas Tet enzymes are dispensable

To test the impact of Gzf1 on IG-DMR methylation, we generated stable Gzf1-KO clones in the reporter system, which silenced the maternal Snrpn-Tom reporter (Fig. 2a and Supplementary Fig. 3a, b). To examine whether this silencing is associated with de novo methylation, we infected Gzf1-KO cells with a viral vector encoding a gRNA targeting Dnmt1. This resulted in the activation of either both or individual GFP and Tom reporters over consecutive passages, consistent with passive loss of methylation (Fig. 2a). To directly associate Gzf1 loss with de novo methylation of the IG-CGI, we established Gzf1-KO clones in a reporter-independent cell line. Bisulfite conversion followed by either Sanger sequencing or NGS revealed hypermethylation at the IG-CGI in Gzf1-KO cells (albeit not to full levels, potentially reflecting the contribution of additional trans-acting factors or technical effects), compared to the normal intermediate methylation levels observed in WT control, or the expected hypomethylation in Dnmt1-KO cells (Fig. 2b and Supplementary Fig. 3c). Quantitative RT-PCR (qPCR) analysis revealed that Gzf1-KO resulted in repression of maternally expressed genes at the Dlk1-Dio3 locus (Fig. 2c), accompanied by hypermethylation of both the IG-TRE and downstream Gtl2-DMR (Fig. 2d). To determine the allelic expression status of Dlk1, a gene not expressed in mESCs, we differentiated the cells to embryoid bodies. Sanger sequencing of a Dlk1 cDNA amplicon containing an informative single-nucleotide polymorphism (SNP) revealed a switch from monoallelic expression in WT, to biallelic expression in Gzf1-KO cells (Fig. 2e). Together, these results are consistent with Gzf1 loss mediating paternalization of the maternal Dlk1-Dio3 allele (Fig. 2f).

Fig. 2. Gzf1 safeguards the maternal allele at the Dlk1-Dio3 IG-DMR.

a Stacked bars show reporter state composition in Gzf1-KO clone (P10) and bulk population of Gzf1-KO cells targeted by Dnmt1 gRNA (dKO) across several passages (P2–P9). b CpG methylation across the IG-CGI measured by bisulfite-amplicon sequencing in wild-type (WT; control), Dnmt1-KO, Tet-tKO and reporter-free Gzf1-KO ESCs (one ES-cell line per genotype). Left, each symbol is the average methylation at a single CpG plotted in genomic order. Right, boxes show the per-CpG methylation distribution for each line: center line, median; box, 25–75th percentiles (IQR); whiskers, Q1 −1.5× IQR to Q3 +1.5× IQR; points, outliers. c Relative mRNA expression (qRT-PCR) of maternal genes Gtl2, Rian and Mirg in Gzf1-KO and Tet-tKO cells. Fold change (2^−ΔΔCt) normalized to WT controls (dashed line = 1). Bars show mean ± s.d. of biological replicates (WT n = 3; Gzf1-KO n = 5, pooled from two independent clones; Tet-tKO n = 2). Two-tailed Welch’s t tests compared each mutant with WT; P values were Holm-adjusted within each gene. Significance: ***P < 0.001, *P < 0.05; ns, not significant (Supplementary Data 5). d CpG methylation within the IG-TRE DMR (left) and Gtl2-DMR (right) in WT and Gzf1-KO ESCs (one ES cell line per genotype). Each symbol, percent methylation at a single CpG plotted in genomic order; boxes: center line, median; box, 25–75th percentiles (IQR); whiskers, Q1 −1.5× IQR to Q3 +1.5× IQR; points, outliers. e Allelic Dlk1 expression by Sanger sequencing of an informative SNP-containing Dlk1 cDNA, after 3-week embryoid-body differentiation of WT and Gzf1-KO ESCs. Chromatograms show paternal-only expression in WT and biallelic expression in Gzf1-KO. f Schematic model for Gzf1 loss: hypermethylation of the maternal IG-DMR converts the locus to a paternal-like state, silencing maternal non-coding transcripts while activating Dlk1. g Reporter dynamics during DNA-methylation erasure and recovery. WT and Gzf1-KO ESCs were treated with 5-Azacytidine for 5 days (P + 2), Tom⁺GFP⁺ cells were isolated by FACS, and cultures were followed drug-free up to P + 8; stacked bars summarize the composition of each population. Source data are provided as a Source Data file.

Given previous reports that implicated Tet genes in regulating demethylation of the IG-DMR maternal allele^27,30, we next asked whether Gzf1 operates in concert with Tet at this locus. We utilized our previously established mESCs harboring triple knockouts (tKOs) of Tet1, Tet2, and Tet3³⁹ and assessed methylation and expression at the Dlk1-Dio3 locus. Unlike in the case of Gzf1 or Dnmt1-KO cells, Tet-tKO cells exhibited WT-comparable methylation levels at the IG-CGI, as well as normal expression of Gtl2, Mirg, and Rian (Fig. 2b, c). Furthermore, analysis of DNA methylation levels across imprinted DMRs using whole-genome bisulfite sequencing maps we previously established from Tet-tKO chimeric mouse embryos³⁹, revealed comparable intermediate methylation levels between mutant and WT cells (Supplementary Fig. 3d). Our data thereby support the notion that Tet enzymes are dispensable for maintaining the non-methylated state associated with the maternal IG-DMR allele. Rather, they suggest that GZF1 protects the IG-CGI from acquiring de novo methylation by hindering DNMT3 activity.

Remethylation kinetics reveal protective role of Gzf1

We next tested whether Gzf1 can prevent remethylation following global demethylation. To this end, we treated both WT and Gzf1 KO reporter cells with 5-azacytidine to induce genome-wide loss of methylation and then monitored the reporters’ activity after drug withdrawal (Fig. 2g). This experimental setup mimics a transient loss of DNMT function and allows assessment of whether GZF1 is sufficient to protect unmethylated alleles once the methylation machinery becomes active again. Treatment of WT Tom⁺GFP⁻ cells with 5-azacytidine resulted in robust activation of the GFP reporter after two passages. The Tom⁻GFP⁻ Gzf1-KO starting population showed a similar trend following treatment, except for the prevalence of single-positive Tom⁺ or GFP⁺ populations, likely reflecting stochastic methylation in the absence of Gzf1 (Fig. 2g). We next sorted the Tom⁺GFP⁺ cell population from each genotype, withdrew 5-azacytidine to release DNMT inhibition, and monitored reporter dynamics over eight passages. While in WT cells the biallelic activity of both reporters remained stable over time, in Gzf1-KO cells reporter expression was rapidly repressed, leading to near-complete silencing of GFP and Tom by passage eight. These results suggest that, following demethylation, GZF1 may bind both alleles thereby protecting them from de novo methylation by DNMT. In the absence of GZF1, the IG-CGI acquires de novo methylation, which facilitates ZFP57 binding and the recruitment of the canonical maintenance machinery that preserves the methylated state, as previously described^10–13.

GZF1 binds unmethylated TACGGTA sequences at the IG-DMR

To gain insight into the mechanism of action of GZF1, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) for either the endogenous GZF1 protein or an epitope-tagged version (GZF1-Flag), which we ectopically expressed in the cells. Both approaches converge on a discrete GZF1-binding region ~440 bp upstream of the IG-CGI (Fig. 3a and Supplementary Fig. 4a). Similar to the IG-CGI and IG-TRE, the GZF1 binding region exhibited intermediate methylation levels in mESCs and during gastrulation but appeared hypermethylated in somatic cells (Fig. 3a and Supplementary Fig. 3d). We next investigated whether GZF1 exhibits allele-specific binding at this locus, leveraging the high prevalence of SNPs in the F1 hybrid background to distinguish maternal (129) from paternal (CAST) alleles. Utilizing five informative SNPs within the GZF1-binding region, we found that GZF1 ChIP-seq reads align predominantly to the unmethylated maternal allele (Fig. 3b, Supplementary Fig. 4b, and Methods). ChIP-PCR sequencing of the GZF1-binding region following Dnmt1-KO revealed a switch from monoallelic binding in WT cells to biallelic binding in the mutant cells (Fig. 3c). Our data therefore support the conclusion that GZF1 binds exclusively to the maternal allele of the IG-DMR, and that this binding is influenced by the allelic methylation state.

Fig. 3. GZF1 binds the unmethylated maternal IG-DMR and opposes ZFP57-mediated paternal maintenance.

a Schematic of the IG-DMR. Top, Dlk1-Dio3 architecture: IG-CGI, IG-TRE and Gtl2-DMR. Split lollipops, intermediate methylation. Bottom, 2-kb zoom of the GZF1 binding site (GZF1-BS). Tracks: WGBS in spermatogonia⁶¹, MII oocytes⁶², ESCs and intestine⁶³; ChIP-qPCR (h) and bisulfite PCR amplicons (i); GZF1-Flag ChIP-seq; GZF1 motifs (open circles, see d); ZFP57 motifs (filled circles); Scissors, sgRNA sites flanking the excised region; genotyping primers, asterisks indicate 129/CAST SNPs (g). b Allele-specific GZF1-Flag ChIP-seq in 129× CAST F1 ESCs (RPGC). c Sanger sequencing of GZF1 ChIP-PCR: allelic bias in WT and biallelic recovery in Dnmt1-KO. d Protein-binding microarray (PBM) identifies consensus motif (TACGGTA). e PBM binding intensity for unmethylated, methylated and scrambled probes (n = 25 spots per probe; boxes: center line, median; box, 25–75th percentiles (IQR); whiskers, Q1 −1.5× IQR to Q3 +1.5× IQR; points, outliers). Two-sided Wilcoxon rank-sum, Holm-corrected: unmethylated vs methylated P = 2.2 × 10⁻¹³; unmethylated vs scrambled P = 4.7 × 10⁻¹⁴; methylated vs scrambled P = 1.5 × 10⁻⁷ (***P < 0.001). f Flow cytometry of Gzf1-KO ESCs nine passages after Gzf1 re-expression. g Sanger sequencing of deletion-bearing amplicons from sorted Tom⁻GFP⁻ cells. h ZFP57 and GZF1 ChIP-qPCR at the GZF1-BS; enrichment over a nearby locus. i IG-CGI methylation in WT, Zfp57-KO and Zfp57-Gzf1 dKO ESCs (one ES-cell line per genotype). Dots, individual CpGs; box elements as defined in (e). j qRT-PCR of Gtl2, Mirg and Rian in Zfp57-KO and Zfp57-Gzf1 dKO ESCs, normalized to WT (2^−ΔΔCt). Mean ± s.d.; n = 3 WT; 6 Zfp57-KO; 6 dKO. Two-tailed Welch’s t tests versus WT, Holm-adjusted within each gene. ***P < 0.001, *P < 0.05; ns (Supplementary Data 5). k H3K4me3 CUT&Tag (CPM) and GZF1-Flag ChIP (RPGC) at the IG-DMR in WT and Gzf1-KO ESCs. l Volcano plot of putative GZF1 interactors by mass spectrometry (Flag pulldown; GZF1-Flag expressing vs non-expressing cells; n = 3). Dashed line, log₂ fold-change = 1; selected interactors labeled. Source data are provided as a Source Data file.

To test whether GZF1 binding specificity reflects a direct sensitivity to DNA methylation, differences in chromatin makeup, or sequence variation between mouse strains, we turned to an in vitro protein-binding microarray (PBM) assay, which isolates sequence- and methylation-level interactions from chromatin context^40–42 (Methods). Introduction of GZF1 protein to a PBM array containing all possible DNA 9-mers, revealed a palindromic consensus motif, 5′-TACGGTA-3′, which contains a CpG dinucleotide (Fig. 3d). To assess whether strain-specific polymorphisms might underlie GZF1 binding preference for the maternal allele, we tiled the highest-affinity sequence into four 60-mer fragments representing both 129 and CAST variants but observed no difference in binding (Supplementary Fig. 4c). In contrast, when comparing GZF1 binding upon introduction to both synthesized methylated and unmethylated TACGGTA probes, we observed markedly reduced affinity for the methylated variant (Fig. 3e, Supplementary Data 5). Systematic dinucleotide substitutions across the 11-bp consensus revealed the same pattern: introducing CpG pairs weakened binding at nearly every position, with methylated CpGs having the greatest impact and central substitutions being most disruptive, underscoring both methylation sensitivity and a palindromic core (Supplementary Fig. 4d). Taken together, these results demonstrate that GZF1 preferentially binds the unmethylated maternal IG-DMR allele, rather than reflecting strain-specific binding.

In line with the above findings, stable ectopic expression of Gzf1 in Tom⁻GFP⁻ Gzf1-KO reporter cells, did not lead to reporter reactivation over nine passages (Fig. 3f), supporting the notion that DNA methylation prevents GZF1 binding to its cognate motif. We next tested whether the binding region is required in cis by excising the interval encompassing the GZF1 ChIP-seq peak and its associated motifs (Fig. 3a). Bulk analysis of reporter transfected cells showed pronounced expansion of the Tom⁻/GFP⁻ population, recapitulating the biallelic hypermethylation signature of Gzf1 knockout (Supplementary Fig. 4e). We sorted this population and determined the allelic origin of the deletion by Sanger and long-read sequencing across informative SNPs, which revealed that deletion-bearing reads originated preferentially from the maternal 129 allele (Fig. 3g and Supplementary Fig. 4f). Together, these data establish that GZF1 occupancy at the binding site protects the maternal IG-DMR from de novo methylation and depends on recognition of a specific DNA sequence sensitive to the local methylation state.

Gzf1 and Zfp57 facilitate distinct epi-allelic states at the IG-DMR

The GZF1 binding site characterized here also harbors a TGCCG motif, previously identified as a ZFP57 recognition motif (Fig. 3a). Given that GZF1 and ZFP57 exhibit opposing binding preferences (unmethylated versus methylated alleles, respectively), this raised the intriguing possibility that both factors target the same region on opposite alleles to stabilize parent-specific methylation states. To confirm ZFP57 binding at the GZF1 binding site, we performed ChIP-qPCR using an antibody against ZFP57 and observed ~10-fold enrichment over a nearby control locus (Fig. 3h). Notably, this convergence in binding between the two factors is largely restricted to the IG-DMR, with minimal overlap observed across other imprinted DMRs genome-wide (Supplementary Fig. 4g). To investigate possible epistatic relationship between the two genes, we generated mESC lines harboring either a Zfp57-KO or a Zfp57-Gzf1 double KO and examined IG-CGI methylation and gene expression from the Dlk1-Dio3 locus (Supplementary Fig. 4h). In both genotypes, we detected hypomethylation of the IG-CGI accompanied by increased expression of maternally expressed genes Gtl2, Mirg, and Rian, compared to WT controls (Fig. 3i, j). These findings place Zfp57 at the top of the hierarchy in maintaining methylation at the IG-DMR. Despite the de novo methylation of the IG-CGI observed upon Gzf1-KO, in the absence of Zfp57, the methylated state cannot be stabilized, leading to biallelic hypomethylation and loss-of-imprinting.

As ZFP57 binding to the methylated paternal IG-CGI triggers a cascade of changes leading to compact chromatin marked by H3K9me3⁹, we next sought to investigate which chromatin marks are associated with GZF1 binding on the unmethylated allele. CUT&Tag profiling of the active histone mark H3K4me3 in Gzf1-KO and WT mESCs showed marked H3K4me3 peaks enriched at both the GZF1 binding site and the IG-CGI, as well as at the promoter-associated Gtl2 DMR in WT cells. These peaks were completely absent in Gzf1-depleted cells (Fig. 3k and Supplementary Fig. 4i). This loss of H3K4me3 coincides with a gain of DNA methylation at the IG-DMR in Gzf1-KO cells (Fig. 2b, d), suggesting that Gzf1 may act in concert with epigenetic modifiers to establish an active chromatin state on the maternal allele. To identify potential GZF1 protein partners, we performed co-immunoprecipitation followed by mass spectrometry (Co-IP/MS) in GZF1–Flag expressing cells. Among the top potential interactors was the Lysine Demethylase 3A (KDM3A), which specifically demethylates H3K9⁴³ (Fig. 3l, Supplementary Data 6). Interestingly, the analysis also identified QRICH1, which was also found by the original CRISPR screen, and further validated to activate the IG-CGI paternal reporter upon KO (Fig. 1f). These findings suggest that GZF1 may act through a complex that facilitates an active chromatin state on the maternal IG-DMR, inhibiting DNMT3 activity^17,44 in this region.

Genome-wide analyses in mESCs reveal limited effects of Gzf1 outside the Dlk1–Dio3 locus

We next investigated whether Gzf1 exerts broader effects on gene expression and epigenetic regulation across the genome. For this purpose, we performed bulk RNA-seq on WT (n = 4) and Gzf1-KO (n = 5) mESC lines, followed by differential gene expression analysis. The results showed overall high concordance between WT and mutant transcriptional profiles, demonstrating that Gzf1 loss does not induce global gene expression changes in mESCs (Supplementary Fig. 5a). Notably, the few genes that did show significant differential expression, including Gtl2, were all downregulated in Gzf1-KO clones compared to WT, albeit mildly, consistent with Gzf1 functioning as a transcriptional activator. To further characterize the genomic features of Gzf1, we analyzed GZF1 ChIP-seq data and identified 1,110 high-confidence GZF1-binding peaks across the genome (Supplementary Fig. 5b, Supplementary Data 7). Notably, some of these peaks were localized to transcription start sites of genes that exhibit significant downregulation upon Gzf1 loss in mESCs (e.g. Rbmxl1, Slc10a7, Rab39b, and Hectd3), highlighting a potential direct role for Gzf1 in regulating their expression (Supplementary Fig. 5c). Analysis of the high-confidence binding peaks enabled inference of a de novo consensus 5′-TACGGTA-3′ sequence for GZF1, matching the motif independently identified by PBM in this study (Supplementary Fig. 5d, e). By contrast, we saw no such enrichment for a previously reported GZF1-responsive element (GRE; TGCGCGTCTATA)³⁸. Our analysis did not identify any association with specific repeat sequences (Supplementary Fig. 5f), particular enrichment for DNA methylation state (Supplementary Fig. 5g), or with particular histone modifications such as H3K4me3 (Supplementary Fig. 5h and Supplementary Data 9).

Together, these findings position GZF1 as a locus-specific regulator of Dlk1-Dio3, functionally distinct from the broadly acting ZFP57. Notably, whereas GZF1 has been characterized as a transcriptional repressor³⁸, its loss at this locus leads to reduced H3K4me3 and decreased expression, suggesting a context-dependent role in transcriptional activation. Yet, we cannot exclude additional effects or DNA-binding sites in other cellular contexts or developmental stages.

Gzf1 loss causes maternal allele paternalization and perinatal lethality in mice

Next, we sought to test whether Gzf1 fulfills similar functions during mouse development as those observed in mESCs. Consistent with a potential in vivo role, Gzf1 shows moderate, ubiquitous expression across adult tissues (Supplementary Fig. 6a, b). We generated Gzf1-KO mice by CRISPR-Cas9-mediated excision of exon 3 in zygotes (Fig. 4a, b). Modified embryos were transferred to foster mothers. Male founders carrying the exon 3 deletion were bred with WT C57BL/6 mice to obtain homozygous Gzf1-KO offspring (Supplementary Fig. 6c). Crosses expected to generate homozygous mutants showed significantly reduced litter sizes across developmental timepoints (Fig. 4c). To determine whether this reflected embryonic or postnatal loss, we first examined genotype distribution during gestation. Homozygous Gzf1-KO embryos were recovered at expected Mendelian ratios at embryonic days (E) 13.5–15.5 (Fig. 4d), indicating that Gzf1 loss does not compromise prenatal survival.

Fig. 4. Gzf1 loss causes perinatal depletion, postnatal growth retardation and disrupted imprinting at Dlk1-Dio3.

a CRISPR-Cas9 strategy to generate Gzf1-KO mice. b Targeting validation by PCR and Sanger sequencing (464-bp deletion shown in red arrowheads; chromatograms show deletion junctions). c Litter size across stages and crosses (Het×Homo E13.5–15.5, n_litters = 4; Het×Homo postnatal, n_litters = 6; Het×Het postnatal, n_litters = 9; Homo♀×WT♂ postnatal, n_litters = 4). Mean ± s.e.m.; dots, individual litters. Two-sided Welch’s t tests, Holm-adjusted (E13.5–15.5 versus the three postnatal groups, P = 0.006, 0.02, 0.006; ***P < 0.001, *P < 0.05; ns; Supplementary Data 5). d Relative genotype distribution of Het×Homo offspring at E13.5–15.5 (n_embryos = 31) (n embryos indicated in box); dashed line, expected 1:1 ratio. Two-sided exact binomial test. DNA methylation in E10.5 embryo trunks by bisulfite amplicon sequencing (n = 2 pooled embryos per genotype) at e IG-CGI; f Gtl2-DMR; g IG-TRE. Dots, average methylation per CpG (WT, light; Gzf1-KO, dark); boxes: center line, median; box, 25–75th percentiles (IQR); whiskers, Q1 −1.5× IQR to Q3 +1.5× IQR; points, outliers. h Expression of Dlk1-Dio3 transcripts in E10.5 trunks relative to WT (n = 6 per genotype, from 2 embryos each). Mean ± s.d.; two-sided Welch’s t test, Holm-adjusted (Supplementary Data 5). i Genotype distribution of Het×Homo offspring postnatally (n_pups= 22); presentation as in (d); two-sided exact binomial, P = 0.004 (*). j Representative WT and Gzf1-KO females at 3 weeks showing growth retardation. k Body weight of females at P25 (n_mice = 4 WT, 7 Het, 7 Homo, 5 Het from Homo♀×WT♂); box elements as defined in (e), points, individual mice; two-sided Welch’s t-tests, Holm-adjusted (WT vs Homo P = 8 × 10⁻⁴, Het vs Homo P = 9 × 10⁻⁵). DNA methylation at (l) IG-CGI and (m) Gtl2-DMR in one adult WT and one surviving Gzf1-KO mouse, box elements as in (e). n Long-read IG-CGI methylation in pooled metaphase-II oocytes; rows, reads; columns, CpGs; blue, unmethylated; red, methylated (WT, n = 1178 reads; Gzf1-KO, n = 2258 reads). Source data are provided as a Source Data file.

To assess the impact of Gzf1 loss on imprinting at the Dlk1-Dio3 locus, we analyzed embryos at E10.5, when both the IG-DMR and Gtl2 DMR normally exhibit parent-of-origin monoallelic methylation³². Methylation analysis revealed that the IG-CGI, Gtl2-DMR, and IG-TRE were hypermethylated in Gzf1-KO embryos compared to the intermediate methylation levels observed in littermate controls (Fig. 4e–g). We next tested whether these changes affected gene expression in mutant embryos. qPCR performed on trunk tissue from E10.5 Gzf1-KO embryos and littermate controls revealed marked downregulation of the maternally expressed Gtl2, Mirg, and Rian transcripts, along with more than a 2-fold increase in Dlk1 expression (Fig. 4h). Together, these results mirror the pattern observed in Gzf1-deficient mESCs (Fig. 2b–d), supporting the conclusion that Gzf1 loss leads to paternalization of the maternal Dlk1-Dio3 allele both in vitro and in vivo.

In contrast to the normal Mendelian ratios observed prenatally, genotype analysis of postnatal pups revealed marked depletion of homozygous KO animals (Fig. 4i, Supplementary Fig. 6d), indicating that most Gzf1-KO mice are lost during the perinatal period. This is in agreement with previous studies showing that paternalization of the maternal allele leads to perinatal lethality⁴⁵. Indeed, methylation analysis of a dead homozygous P1 pup revealed hypermethylation at the IG-CGI, consistent with paternalization of the maternal allele (Supplementary Fig. 6e). Interestingly, a small number of homozygous KO animals survived the perinatal transition and displayed clear postnatal growth retardation (Fig. 4j). This was evident at both P25 and P50, with KO mice showing significantly reduced body weight compared to WT and heterozygous controls (Fig. 4j, k and Supplementary Fig. 6f). To investigate the basis for this survival, we considered the possibility that it reflects incomplete or heterogeneous methylation across the locus, as previously reported²⁹. Consistent with this, a surviving homozygous pup at P25 displayed intermediate methylation at the Gtl2-DMR while retaining complete methylation at the IG-CGI (Fig. 4l, m). The precise mechanism underlying this survival remains to be fully elucidated.

Gzf1 safeguards the IG-DMR from de novo methylation during oogenesis

The survival of a small subset of Gzf1-KO females provided a unique opportunity to examine oocytes that developed in the absence of Gzf1 during oogenesis. Long-read nanopore sequencing revealed that, in contrast to unmethylated WT oocytes, Gzf1-KO oocytes acquired robust hypermethylation at the IG-CGI, which was high overall but heterogeneous across individual oocytes, with a subset retaining incomplete methylation patterns (Fig. 4n). If methylation is acquired in Gzf1-KO oocytes, embryos derived from these oocytes would be expected to display a maternal effect and resemble homozygous KOs despite inheriting a WT paternal Gzf1 allele. To test this, we crossed homozygous KO females with WT males. These crosses resulted in markedly reduced litter sizes, consistent with substantial perinatal loss (Fig. 4c). The few surviving heterozygous offspring displayed normal body weight and WT-like IG-CGI methylation (Fig. 4k and Supplementary Fig. 6g), suggesting that a subset of oocytes with incomplete methylation can be rescued by the paternally inherited Gzf1 allele. More broadly, these data demonstrate that Gzf1 is required both to protect the maternal IG-DMR from de novo methylation during oogenesis and to maintain its unmethylated state during embryonic development.

Discussion

Our study identifies GZF1 as a sequence- and methylation-sensitive factor that prevents de novo methylation of the maternal Dlk1-Dio3 allele, acting in coordination with the previously characterized factor ZFP57¹⁰. We define a region ~440 bp upstream of the IG-CGI that contains adjacent TACGGTA (GZF1) and TGCCGC (ZFP57) motifs. ZFP57 recognizes a methylated TGCCGC motif and initiates the recruitment of TRIM28 and SETDB1, along with H3K9me3 deposition, thereby establishing and maintaining a repressive chromatin state at the paternal allele^9,10,12–14. In contrast, GZF1 preferentially binds the unmethylated TACGGTA motif on the maternal allele. This motif differs from a previously reported GZF1 responsive element (TGCGCGTCTATA)³⁸, and likely reflects the binding preference of full-length GZF1 in this context. This binding promotes a permissive chromatin environment, potentially through concerted activity with the H3K9 demethylase KDM3A⁴³, the recently identified QRICH1-SEPHS1 zinc-finger coregulator complex⁴⁶ and by promoting H3K4me3¹⁷ (Fig. 5a). Loss of GZF1 leads to de novo methylation of the IG-CGI, converting the maternal allele into a ZFP57-competent state and initiating repressive chromatin formation via the ZFP57 pathway. Once methylation is established in a Gzf1-KO background, reintroducing GZF1 fails to restore the unmethylated state, highlighting the irreversibility of this switch. Loss of ZFP57, in turn, results in demethylation of the paternal allele and acquisition of maternal-like features. In DNMT1-deficient cells, demethylation of both alleles permits biallelic GZF1 binding. When methyltransferase activity resumes, GZF1’s occupancy blocks remethylation and preserves the hypomethylated state. Furthermore, our findings argue against active Tet-mediated demethylation in maintaining the unmethylated IG-DMR allele^27,30, as WT-comparable intermediate methylation levels are observed following KO of all three Tet enzymes in cells and embryos. Together, these findings support a model of reciprocal, sequence- and methylation- dependent regulation at the Dlk1-Dio3, in which ZFP57 stabilizes the methylated paternal allele, whereas GZF1 protects the unmethylated maternal allele from de novo methylation.

Fig. 5. Model for GZF1-mediated protection of the maternal IG-DMR and comparison with the literature.

a Proposed model for maintenance of imprinting in the Dlk1-Dio3 locus. On the maternal chromosome (top), an unmethylated element upstream of the IG-CGI (440 bp upstream of the CpG rich IG-CGI) harbors a palindromic TACGGTA motif. GZF1 binding alongside putative partners sustains local H3K4me3 (green) and inhibits de novo methylation by DNMT3A/B. On the paternal chromosome (bottom) the element is methylated, preventing GZF1 binding but permitting ZFP57 recruitment and maintenance methylation. Loss of GZF1 triggers de-novo methylation of the maternal element, ectopic ZFP57 binding, and ‘paternalization’ of the locus; once methylated, the site is refractory to GZF1 re-binding. b Genetic and epigenetic interventions at the maternal IG-DMR. Schematics (left-to-right within each box) show the IG-CGI sub-elements (GZF1-BS, IG-CGI, IG-TRE) and the downstream Gtl2 gene cluster. Top, wild-type configuration. Middle panels: complete IG-DMR deletion (Δ, Lin et al.⁴⁵) causes perinatal lethality, whereas IG-CGI excision (Weinberg-Shukron et al.²⁹; Saito et al.⁴⁷) leaves the maternal locus functional. Bottom, Gzf1-KO (this study) causes perinatal lethality and postnatal growth retardation in surviving pups, driven by stochastic de novo hypermethylation of the IG-CGI.

Cis and trans manipulations of the IG-DMR result in distinct developmental phenotypes and reveal a functional hierarchy among regulatory elements within this region (Fig. 5b). Lin et al. removed the entire maternal IG-DMR, including the upstream GZF1 binding site, and observed perinatal lethality with silencing of maternal transcripts and biallelic expression of Dlk1⁴⁵. By contrast, the smaller IG-CGI deletion generated by Weinberg-Shukron et al.²⁹, and Saito et al.⁴⁷, which left the upstream GZF1 binding site intact, produced neither molecular nor phenotypic defects, placing the GZF1 site functionally upstream of the IG-CGI. The Gzf1-KO generated in this study further refines this picture, inducing epigenetic changes at the locus without deleting the underlying DNA sequences that may comprise binding sites for yet uncharacterized trans regulators. In line with Lin et al., GZF1 loss leads to paternalization of the maternal allele and is associated with substantial perinatal lethality, linking disruption of this regulatory region to both molecular and developmental outcomes^48–50. The incomplete penetrance of the phenotype, reflected by the recovery of a small number of growth-retarded homozygotes, suggests that epigenetic perturbation of the locus generates variable outcomes rather than a fully deterministic switch. As we have previously shown²⁹, survival in such settings may arise from incomplete methylation, leading to variable dosage imbalance across the locus.

Genomic imprinting relies on the maintenance of parent-of-origin-specific epigenetic states following fertilization. Yet, in principle, similar mechanisms could also act in the germline to establish this asymmetry by protecting loci from de novo methylation. Directly probing the role of Gzf1 during oogenesis is challenging, given the perinatal lethality of homozygous mutants and the requirement for germline-specific perturbations. By taking advantage of the rare surviving Gzf1-null females, we were able to examine oocytes that developed in the absence of Gzf1 and thereby assess its role during oogenesis. Strikingly, these analyses indicate that Gzf1 is required to safeguard the unmethylated state of the maternal allele already during oocyte growth, extending its function beyond embryonic reprogramming. These findings support a model in which parent-specific epigenetic memory is maintained through selective protection of both methylated and unmethylated alleles, mediated by sequence-specific factors that act in a methylation-sensitive manner across developmental transitions. The coordinated interplay of the pan-imprint factor Zfp57 and locus-specific factors like Gzf1 may reflect a general regulatory logic underlying imprint establishment across genomic loci. The CRISPR-based screening strategy coupled with locus-specific methylation readout developed here offers a framework for discovering additional trans- acting regulators of epialleles. Although GZF1 itself may represent a locus-specific solution, the underlying requirement to protect unmethylated ICR alleles from de novo methylation is likely to be broadly conserved. The rapid evolutionary divergence of both C2H2 zinc finger proteins and non-coding regulatory sequences at imprinted loci raises the possibility that functionally analogous mechanisms have evolved independently at other ICRs and across mammals.

Methods

Animals

All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the Weizmann Institute of Science (protocols 03540425-1 and 08241020-2) and conducted in strict accordance with institutional guidelines. Mice were housed under controlled conditions with 12-h light–dark cycles and provided ad libitum access to standard chow and water. C57BL6/Jrcc mice used in this study were obtained from Envigo or bred in-house. Timed matings were set up, plugs were checked in the following morning, and embryonic tissues were collected at embryonic day 10.5 (E10.5).

mESCs cell culture

V6.5 mouse embryonic stem cells (mESCs) were cultured on irradiated mouse embryonic fibroblasts (MEFs) with standard ESCs medium: (500 ml) DMEM supplemented with 10% FBS (Hyclone), 10 μg recombinant leukemia inhibitory factor (LIF), 0.1 mM beta-mercaptoethanol (Sigma-Aldrich), penicillin/streptomycin, 1 mM L-glutamine, and 1% nonessential amino acids (all from Invitrogen). 5-azacytidine (Sigma) was administered for 2 passages (5 days) in a final concentration of 2 μM. Cell media was exchanged every day, supplemented with fresh 5-azacytidine.

Flow cytometry

Single-cell suspensions were generated by trypsinizing mouse ESC cultures and passing them through a 40 µm nylon mesh. All analytical and preparative runs were performed on a BD FACSAria III fitted with an 85 µm nozzle; samples and collection tubes were maintained at 4 °C throughout. Data were recorded with FACSDiva and analyzed in FCS Express 7 RUO (v7.24.0030). The gating workflow was as follows: first, debris was excluded on forward- vs side-scatter area (FSC-A × SSC-A); second, doublets were removed on FSC-A × FSC-H. Positivity thresholds for GFP and tdTomato were set with WT (non-fluorescent) cells, and these gates were then applied unchanged to every sample in the experiment (Supplementary Fig. 7). A minimum of 1 × 10⁴ and typically up to 2 × 10⁴ singlet events were analyzed per sample. Sorted fractions were deposited either into ice-cold ESC medium for continued culture or into pre-chilled 1 × PBS for downstream DNA/RNA isolation.

Genotyping

For genotyping, DNA was extracted from the tail tip using a solution containing NaOH 1 M and EDTA 0.5 M pH8.0 in DDW, incubated for 1 h at 95 °C and neutralized in Tris-HCl 40Mm pH5.0. Alternatively, DNA was isolated by lysing cells in lysis buffer (0.1 M Tris buffer, 0.2 M NaCl, 0.005 M EDTA, 0.2% SDS) with 10 mg/ml Proteinase K at 55 °C, precipitated with Isopropanol, washed with 70% Ethanol and resuspended in TE buffer. Genotyping of mouse strains and alleles was done by PCR; primers are listed in Supplementary Data 8.

Bisulfite conversion, PCR, Sanger sequencing and NGS

Bisulfite conversion of DNA was established using the EZ DNA Methylation-Lightning Kit (Zymo Research cat#D5031) following the manufacturer’s instructions. The resulting modified DNA was amplified by the first round of nested PCR, following a second round using loci-specific PCR primers (primers are listed in Supplementary Data 8). The first round of nested PCR was done as follows: 94 °C for 4 min; 55 °C for 2 min; 72 °C for 2 min; Repeat steps 1–3 1×; 94 °C for 1 min; 55 °C for 2 min; 72 °C for 2 min; Repeat steps 5–7 35×; 72 °C for 5 min; Hold 12 °C. The second round of PCR was as follows: 95 °C for 4 min; 94 °C for 1 min; 55 °C for 2 min; 72 °C for 2 min; Repeat steps 2–4 35×; 72 °C for 5 min; Hold 12 °C. For Bisulfite-Sanger the resulting amplified products were gel-purified, subcloned into a pGEM®-T Easy cloning vector (Promega cat#A1360), and Sanger sequenced. For Bisulfite-NGS the resulting amplified products were cleaned by Zymo C&C (D4004). Premium PCR Sequencing was performed by Plasmidsaurus using Oxford Nanopore Technology.

Fastq reads were aligned with Bismark v0.24.2 to the appropriate reference genome for each experiment; all references were pre-indexed with bismark_genome_preparation. CpG methylation was extracted using the Bismark methylation extractor. For every CpG site, percent methylation was calculated as the fraction of methylated reads among all reads at that position, and sites with coverage below 20 were discarded. Per-CpG percent-methylation values were averaged per genotype at each genomic position and plotted in genomic order, and their distribution per sample was summarised as box plots. For each genotype, methylation was profiled in a single biological sample (one isogenic ES-cell line for cell-based experiments, or one animal/pooled-embryo sample for in vivo experiments, as specified in each figure legend). Genotypes were compared descriptively from the per-CpG methylation values and their box-plot distributions; the individual CpGs are within-sample measurements.

CRISPR screen and analysis

LentiCas9-Blast was a gift from Feng Zhang (Addgene plasmid no. 52962)⁵¹. LentiCas9-Blast was cloned such that Neomycin resistance replaced the Blasticidin resistance, to allow for antibiotic selection in the presence of DR-4 inactivated MEF feeder layer. Lenti-Cas9-NeoR was packaged in 3rd generation lentivirus followed by ultra centrifugation and froze in 10 μl aliquots of 10⁹ lentiviral titters as previously described⁵². RGM mESCs were infected with the Lenti-Cas9-NeoR plasmid and then were selected for 3 passages (1 week). RGM-Cas9 were propagated in bulk, to allow for an averaged Cas9 activity across the different integration sites.

Mouse Brie CRISPR knockout in lentiGuide-Puro pooled library was a gift from David Root and John Doench (Addgene viral prep no. 73633-LV)³³. Following an initial pilot experiment to assess viral titters, a total of 120 million RGM-Cas9-NeoR mESCs were transduced with the Brie lentivirus library containing 78,637 sgRNAs at a multiplicity of infection of 0.3, resulting in infection of 30% of the cells and hence leading to a 500-fold coverage of the sgRNA library size. An MOI of 0.3 ensures a high enrichment in the proportion of cells that are infected with only one viral particle and therefore carry a single mutation. Infected cells were selected with puromycin (Sigma) for 6 passages (2 weeks). 80 million cells were counted in each passage to ensure proper complexity until the end of the experiment. We performed two independent screen experiments. In each experiment the cells were collected for DNA preparation at several time points: P0 (after initial dose of Puromycin for 1 day, in 3 biological repeats, 40 M cells each), just prior to sorting at P6 (Unsorted or US, in 3 biological repeats, 40 M cells each), and post sorting controls according to RGM fluorescence (Tom⁺GFP^- in 5 biological repeats, 1 M per sample, Tom⁺GFP⁺, in 1 sample of 200–300 K cells and Tom^-GFP^- in 1 sample of 1.5–2 M cells).

Genomic DNA was extracted from each time point and condition and the gRNA library was amplified by PCR as previously described⁵³. Deep sequencing on an Illumina Novaseq 6000 platform was used to monitor the library composition. The guide composition between the Tom⁺GFP⁺ and Tom^-GFP^- and US was compared. We used MAGeCK package⁵⁴ to identify top enriched and top depleted sgRNAs in fluorescent samples vs. US control (Supplementary Data 2, 4), and to rank the corresponding top candidate genes (Supplementary Data 1 and 3). Positive hits (genes whose sgRNAs were enriched in the fluorescent sample) were plotted by ggplot2, after excluding genes that were enriched between the US and P0 to avoid genes that confer growth advantage (e.g. tumor suppressors), and genes that were enriched between the sorted Tom⁺GFP^- and the US to exclude FACS biases.

Knockout of candidate genes in pooled ESCs

For nine top candidate genes from each sorted population - Tom⁺GFP⁺ and Tom^-GFP^- - individual targeted sgRNAs were cloned into the lentiGuide-Puro plasmid (Addgene no. 52963) by BsmBI digestion. These constructs were used to produce lentivirus and transduce RGM-Cas9 cells, followed by puromycin selection for 6 passages and FACS analyses every two passages. The guide RNA sequences used are listed in Supplementary Data 8.

Gzf1 ectopic expression

Mouse embryonic stem cells (mESCs) were transfected with a custom-generated VectorBuilder plasmid encoding full-length mouse Gzf1 fused at its C-terminus to a Flag epitope under the control of an EF1α promoter, alongside a CMV promoter-driven BFP2 fluorescent reporter. Transfection was performed using Mirus TransIT reagent according to the manufacturer’s instructions. At 48 h post-transfection, cells expressing high levels of BFP2 were sorted by fluorescence-activated cell sorting (FACS), and stable expression was monitored by assessing BFP2 fluorescence at subsequent passages.

CUT&Tag

Cleavage Under Targets and Tagmentation (CUT&Tag) was performed as previously described⁵⁵ with minor modifications. Briefly, ~1 × 10⁵ wild-type and Gzf1-knockout mouse embryonic stem cells (mESCs) were harvested and underwent nuclei extraction as previously described⁵⁶. Later, cells were resuspended in Nuclear Extraction (NE) buffer (20 mM HEPES pH 7.5, 10 mM KCl, 0.5 mM spermidine, 0.1% Triton X-100, 10% glycerol, and 1 mM PMSF) and incubated on ice. Cells were immobilized on concanavalin A–coated magnetic beads by incubation at room temperature for 15 min, followed by washes in Wash Buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM spermidine, and 0.1% BSA).

Primary antibody against H3K4me3 (cellsignal #9751) was added at a 1:50 dilution and incubated overnight at 4 °C. After washing, a secondary antibody (Cat No. ABIN101961) at a 1:100 dilution was incubated with the cells for 30 min at room temperature. Subsequently, protein A-Tn5 transposase fusion proteins were added, and the samples were incubated for 60 min at room temperature. Tagmentation was initiated by incubating the samples with MgCl₂-containing buffer at 37 °C for 1 h. Reactions were terminated by adding SDS, EDTA, and Proteinase K, followed by incubation at 56 °C for 60 min.

DNA was purified using magnetic bead-based purification and libraries were amplified by PCR. The final libraries were quality-checked, pooled, and sequenced on an Illumina platform to generate paired-end sequencing reads. Sequencing data were processed and analyzed using standard bioinformatics pipelines as described previously⁵⁵.

Paired-end reads were truncated to 36 nt with Cutadapt v2.8 to remove low-quality 3′ bases and improve mappability. Trimmed reads were aligned to the GRCm38/mm10 reference with Bowtie2 v2.5.4 with the very-sensitive-local preset; only properly paired alignments with MAPQ ≥ 30 were retained. BigWig files were produced with deepTools bamCoverage v3.3.1 (10 bp bins, CPM normalisation, read-centered, 50 bp smoothing) and used for qualitative comparison of wild-type and Gzf1-knock-out samples. All bigwigs were autoscaled to a control Cut&Tag of the same background with IgG control secondary antibody.

Genome-wide H3K4me3 differential analysis

CUT&Tag reads were aligned to mm10 with Bowtie2 v2.5.4 (--very-sensitive-local, MAPQ ≥ 30), sorted with SAMtools, and imported into the misha genomic database. H3K4me3 peaks were called independently for WT and Gzf1-KO samples by summing reads in a sliding ±1 kb window and retaining windows in the top 1% of signal. A union peak set was generated by merging peaks across both genotypes. Coverage at each union peak was quantified for both samples and normalized to counts per million (CPM). Differential peaks were defined as those with |log₂ fold-change| > 1.5. To assess whether H3K4me3 changes were directly associated with GZF1 binding, peaks within 5 kb of high-confidence GZF1 ChIP-seq sites were identified and further filtered for the presence of the GZF1 consensus motif and concordant gene expression changes from RNA-seq.

Gene expression by bulk RNA seq

WT and Gzf1-KO mESCs were harvested after depletion of feeder mouse embryonic fibroblasts (MEFs). Total RNA was isolated using TRIzol reagent (Invitrogen) followed by purification with the Direct-zol RNA Miniprep Kit (Zymo Research), according to the manufacturers’ protocols. Libraries for bulk RNA-seq were prepared using a standard poly(A)+ selection protocol as described previously⁵⁷. cDNA synthesis, library construction, and indexing were performed following manufacturer recommendations. Libraries were sequenced on an Illumina Novaseq 6000 platform to generate paired-end reads.

Bulk RNA-sequencing reads were quality-checked and trimmed before alignment to the mouse reference genome (mm10) using STAR with default parameters. Differential expression between Gzf1-KO (n = 5) and WT (n = 4) samples was tested using DESeq2 with default parameters (Wald test, median-of-ratios normalization). For visualization, per-sample log2 expression fractions were averaged within each group, and genes were called differentially expressed if the absolute difference in mean log2 expression exceeded 1.1 and the DESeq2 p-value was below 0.05.

Generation of clonal knockout mESC lines by CRISPR/Cas9

Generation of clonal KO mESC lines (Gzf1, Dnmt1, Zfp57, and double knockouts: Gzf1-Dnmt1, Gzf1-Zfp57) was performed using CRISPR/Cas9 editing. Specific gRNAs (listed in Supplementary Data 8) were cloned into the pX330-U6-Chimeric_BB-CBh-hSpCas9 plasmid (Addgene #98750) and transiently transfected into mESCs using Mirus TransIT reagent. Gzf1 knockout utilized two gRNAs targeting the intronic regions flanking exon 3 (95 bp), generating a deletion that resulted in a frameshift mutation. The Zfp57 knockout employed two gRNAs, while the Dnmt1 knockout utilized a single sgRNA followed by sorting for Tom⁺GFP⁺ cells. Double-knockout lines were created sequentially: for Gzf1-Zfp57, Zfp57 was targeted in Gzf1-knockout clones, followed by clonal selection; for Gzf1-Dnmt1, Dnmt1 was targeted in bulk in the Gzf1-knockout background without clonal selection. Single-cell-derived clones were manually picked and screened by PCR. Homozygous Gzf1-KO clones were confirmed by qPCR, demonstrating >95% transcript loss relative to WT cells. Previously generated³⁹ Dnmt1 and Tet triple knockout (Tet-tKO) lines were used.

Deletion of the GZF1-binding region and allelic assignment

To delete the GZF1-binding region, two sgRNAs flanking the ChIP-seq-defined binding peak were designed to excise a 1276-bp region in RGM reporter mESCs. sgRNAs were cloned and transfected as described above. Two passages after transfection, cells were analyzed by FACS and the Tom⁻/GFP⁻ population was sorted in bulk. Genomic DNA was extracted from sorted cells and amplified by PCR using primers flanking the deletion that span two informative SNPs distinguishing the maternal 129 allele (SNP1: A; SNP2: A) from the paternal CAST allele (SNP1: G; SNP2: C). PCR products were separated by gel electrophoresis, yielding a ~300-bp deletion band and a ~1.5-kb wild-type band. For Sanger sequencing, the deletion band was gel-excised and sequenced across the SNP positions, confirming that the deletion occurred predominantly on the maternal 129 allele. For long-read sequencing, the unseparated PCR product was sequenced by nanopore sequencing (plasmidsaurus). Reads shorter than 700 bp were extracted as the deletion-band fraction based on the bimodal read-length distribution. Reads were aligned to a custom 129-haplotype reference sequence carrying the deletion using minimap2 with the --eqx flag, and each read spanning both diagnostic SNPs was classified as concordant 129, concordant CAST, or discordant based on SNP concordance. To confirm the absence of reference alignment bias, reads were independently aligned to a CAST-haplotype reference, yielding near-identical allelic ratios (83.9% vs 82.4% for 129). Of 3238 classified reads, the deletion was carried predominantly on the maternal 129 allele (2717 reads, 83.9%), with 455 reads (14.1%) mapping to the paternal CAST allele and 66 reads (2.0%) discordant across the two SNPs, consistent with long-read sequencing error. The minority of paternal CAST-allele deletions likely reflects cells with homozygous deletions, which cannot be distinguished from maternal-only events in this assay.

Chromatin immunoprecipitation–sequencing

mESCs (1 × 10⁷ cells per assay) were fixed in 1 % formaldehyde (5 min, RT) and quenched with 0.125 M glycine (5 min); the pellets were washed twice in ice-cold PBS containing EDTA-free protease inhibitors, aliquoted into LoBind tubes, snap-frozen, and stored at −80 °C.

Frozen pellets were thawed on ice and lysed for 15 min in 50 mM Tris–HCl pH 8.0, 150 mM NaCl, 1 % Triton X-100, 0.1% sodium deoxycholate, 5 mM CaCl₂, plus inhibitors. Nuclei were collected (845 × g, 10 min, 4 °C), resuspended in 10 µL of the same buffer supplemented with 17.5 U micrococcal nuclease (Worthington LS004798), chilled 10 min on ice, and digested at 37 °C for 15 min. Digestion was stopped with 20 mM EDTA, samples were chilled 30 min on ice, and lysates were cleared (20,000 × g, 10 min, 4 °C). The supernatant was adjusted to 180 µL with RIPA buffer (10 mM Tris–HCl pH 8.0, 140 mM NaCl, 1 mM EDTA, 0.1% SDS, 0.1% sodium deoxycholate, 1% Triton X-100).

Chromatin was immunoprecipitated for 2 h at 4 °C with either 10 µg anti-Flag M2 (Sigma F1804) or 5 µg anti-GZF1 (LSBio LS-C429273), followed by 1 h capture with 20 µL Protein G Dynabeads (Invitrogen). Beads were washed sequentially at 4 °C: six times in RIPA, three times in high-salt RIPA (500 mM NaCl), three times in LiCl buffer (10 mM Tris pH 8.0, 250 mM LiCl, 0.5 % NP-40, 0.5% sodium deoxycholate, 1 mM EDTA), and three times in 10 mM Tris pH 7.5.

Library construction was performed on-bead as previously described⁵⁸: end-repair (T4 polymerase + T4 PNK), A-tailing (Klenow exo⁻), and ligation of indexed Illumina adapters (Quick Ligation Kit, NEB). Crosslinks were reversed in 300 mM NaCl, 0.6 % SDS with RNase A (30 min, 37 °C) followed by proteinase K digestion (2 h, 37 °C; overnight 65 °C). DNA was purified with 2× SPRI beads, amplified with KAPA HiFi HotStart (14 cycles), size-selected with 0.8× SPRI, quantified, pooled, and sequenced (51 bp or 61 bp paired-end) on NovaSeq 6000. Demultiplexing and base-calling were performed with bcl2fastq.

ChIP-seq was performed in two independent experiments. In the first, Cas9-expressing mESCs with lentiviral GZF1-Flag were immunoprecipitated with both anti-Flag and anti-GZF1 antibodies. Concordant binding profiles between the two antibodies confirmed the specificity of the commercial anti-GZF1 antibody (Supplementary Fig. 4a). The anti-Flag dataset from this experiment was used for genome-wide characterization of GZF1-bound sites, including de novo motif discovery, repeat-element annotation, and CpG-methylation context (Supplementary Fig. 5d–h). For these analyses, GZF1 ChIP-seq, H3K27ac ChIP-seq (anti-H3K27ac; Thermo Fisher Scientific, MA5-23516) and ATAC-seq signal were quantified over GZF1 peaks; peaks with high GZF1 signal were defined as high-confidence sites, and H3K27ac peaks served as the comparison set for the motif-score and CpG-methylation analyses (Supplementary Fig. 5e, g). In the second, ChIP-seq was performed in V6.5 and RGM reporter mESCs with lentiviral GZF1-Flag (anti-Flag antibody) and in C57BL/6 WT versus Gzf1-KO mESCs (anti-GZF1 antibody). Peaks identified independently in the V6.5 and RGM backgrounds were intersected to define 1110 high-confidence GZF1 sites; these sites showed no enrichment in Gzf1-KO cells, confirming GZF1-dependent signal. Across all experiments, paired-end reads were adapter-trimmed with Cutadapt (minimum length 20 bp) and aligned to mm10 with Bowtie2 v2.5.4 (--very-sensitive, --dovetail, -k 1). Reads with MAPQ < 10 were discarded and PCR duplicates were removed with samtools markdup. Deduplicated alignments were converted to 10 bp bigWig coverage tracks using deepTools bamCoverage v3.3.1, normalised to reads per genomic content (RPGC; effective genome size 2,652,783,500). Independently, read counts were imported into a misha genomic database, where coverage at candidate peaks was quantified as summed read counts. Peaks were defined independently for every sample as 300 bp windows whose summit coverage exceeded the sample-specific 99.985th percentile. Intervals overlapping RepeatMasker elements were removed (with the exception of the analysis for repeats in Supplementary Fig. 5d). Within each experimental pair, peaks were clustered by signal intensity (k-means, k = 6), and those belonging to the highest-signal clusters with log₂(counts + 1) > 5 in the GZF1-expressing sample were retained as GZF1-associated peaks. De novo motif discovery was performed using PREGO (regress_pwm; five candidate motifs) on sites with log₂(counts + 1) ≥ 5.9, and the resulting position-specific scoring matrix was used to score motif enrichment across all peaks.

For allele-specific analysis, a custom N-masked reference genome was generated with SNPsplit_genome_preparation⁵⁹ from the GRCm38/mm10 assembly, using a VCF file containing SNPs from the Mouse Genome Project (mgp.v5.merged.snps_all.dbSNP142). Dual-hybrid mode was configured with 129S1/SvImJ and CAST/EiJ as the parental strains. Paired-end reads were aligned to the N-masked reference with Bowtie2 v2.5.4, sorted and indexed with SAMtools v1.21. Allele assignment was performed with SNPsplit v0.5.0 in dual-hybrid, paired-end mode, producing 129S1-, CAST-, and unassigned BAM files; only allele-assigned reads were retained. PCR duplicates were removed from allele-split BAMs with samtools markdup. Allele-specific bigWig tracks were generated from the deduplicated BAMs using deepTools bamCoverage v3.3.1 (10 bp bins) and normalized to reads per genomic content (RPGC; effective genome size 2,652,783,500). All bigwigs were autoscaled to a control ChIP of the same background without GZF1-Flag ectopic expression.

Reanalysis of ZFP57 ChIP-seq data

Published ZFP57 ChIP-seq data from wild-type and Zfp57-knockout ESCs (Shi et al.)⁶⁰ were downloaded from the NCBI Sequence Read Archive (SRA). Paired-end reads were adapter-trimmed and aligned to mm10 with Bowtie2 v2.5.4 (--very-sensitive, --dovetail, -k 1), sorted with SAMtools, and imported into the misha genomic database. Candidate peaks were defined per sample using a 99.985th percentile coverage threshold, merged across samples, filtered against RepeatMasker annotations, and normalized to 300 bp. Signal was quantified as summed read counts, and ZFP57-bound sites were defined by log₂(1 + WT counts) − log₂(1 + KO counts) > 2 together with log₂(1 + WT counts) > 5. Overlap with GZF1 peaks and signal at known imprinted DMRs were then evaluated.

ChIP-PCR and ChIP-qPCR

ChIP-PCR was performed to determine allele-specific binding of GZF1 at the IG-DMR locus in hybrid (129 × CAST) mESCs. Following chromatin immunoprecipitation using an anti-GZF1 antibody as described above, purified DNA fragments were amplified by locus-specific PCR (primers listed in Supplementary Data 8) and subjected to Sanger sequencing. SNP variants distinguishing the maternal (129) and paternal (CAST) alleles were then assessed, enabling determination of allele-specific occupancy by GZF1.

For ChIP-qPCR, chromatin immunoprecipitation was similarly performed using anti-Flag (GZF1-Flag expressing cells) and anti-ZFP57 antibodies. Enrichment at GZF1 and ZFP57 binding sites was quantified by qPCR (primers listed in Supplementary Data 8), using SYBR Green Master Mix on a QuantStudio 5 Real-Time PCR platform. Fold enrichment was calculated relative to input and normalized against an adjacent genomic locus lacking GZF1/ZFP57 binding (Supplementary Data 8), as well as against β-actin as additional control. Experiments were conducted with technical duplicates.

Reverse transcription of RNA and quantitative real-time PCR

For isolation of RNA, Direct-zol RNA Miniprep Kit (Zymo Research cat.no. R2052) was used, according to the provider’s instructions. Subsequently, 1 μg of RNA served as the starting material for reverse transcription, carried out using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems cat.no. 4368814). For qPCR, Fast SYBRTM Green Master Mix (Applied Biosystems cat. no. 4385610) was used on a QuantStudio™ 5 Real-Time platform (Applied Biosystems cat.no. A34322) and experiments were performed in two or three technical repeats. Relative expression values were normalized to the expression levels of β-Actin and WT controls (representing ΔΔct).

Immunoprecipitation and on-bead digestion for mass spectrometry

GZF1-Flag ectopically expressing or control V6.5 cells were washed and lysed in RIPA buffer supplemented with 1:250 protease inhibitors (PI cocktail, Sigma-Aldrich), 1 mM PMSF (Sigma-Aldrich), and phosphatase inhibitors (PhosSTOP, Sigma-Aldrich). Lysis was performed on ice for 30 min, followed by centrifugation at 15,000 × g for 15 min at 4 °C. Protein concentrations were measured using the BCA Protein Assay (ThermoScientific). For anti-Flag affinity pulldown, 1000 μg of total protein was diluted to a final volume of 900 μL in binding buffer (50 mM Tris-HCl, pH 7.4; 150 mM NaCl; 1 mM EDTA; 1% Triton X-100, 1:250 PI). Samples were incubated overnight at 4 °C with 20 μL of anti-Flag M2 magnetic beads (Sigma-Aldrich) under gentle rotation. Beads were then washed three times with TBS containing protease inhibitors. After the final wash, beads were resuspended in TBS and subjected to on-bead digestion as described below.

Sample preparation, liquid chromatography mass spectrometry and data processing

Immunoprecipitation followed by on-bead digestion was performed in four biological replicates per condition (GZF1-Flag-expressing and control V6.5 mESCs). Anti-Flag immunoprecipitates were resuspended in 8 M urea in 0.1 M Tris-HCl (pH 7.9), reduced with 5 mM dithiothreitol and alkylated with 10 mM iodoacetamide, diluted to 2 M urea with 50 mM ammonium bicarbonate, and digested with trypsin (Promega) overnight at 37 °C. Digestion was stopped with 1% trifluoroacetic acid, and the resulting peptides were desalted (Oasis HLB, Waters), vacuum-dried and stored at −80 °C. Peptides were analyzed by nanoflow liquid chromatography (nanoAcquity; Waters) coupled to high-resolution, high-mass-accuracy mass spectrometry (Q Exactive HF-X; Thermo Fisher Scientific). Samples were trapped on a Symmetry C18 column (180 µm × 20 mm, 5 µm) and separated on a T3 HSS nano-column (75 µm × 250 mm, 1.8 µm) at 0.35 µL/min, using a gradient of 4–27% acetonitrile (0.1% formic acid) over 50 min, 27–90% over 5 min and 90% for 5 min. Each sample was analyzed separately in random order in data-dependent (discovery) mode using a Top10 method, with MS1 acquired at 120,000 resolution (375–1650 m/z, AGC 3 × 10⁶, maximum injection time 50 ms) and MS2 at 15,000 resolution (1.7 m/z isolation window, AGC 1 × 10⁵, 20 s dynamic exclusion, maximum injection time 60 ms). One GZF1-Flag replicate was identified as an outlier during quality control and excluded, yielding n = 3 for GZF1-Flag and n = 4 for control in the final analysis. Raw data were processed with MetaMorpheus v1.0.2 and searched against the murine UniProt proteome database (January 2024 xml version, including known PTMs) appended with common lab protein contaminants (17,417 entries). Search settings were: trypsin with up to two missed cleavages; minimum peptide length of 7; carbamidomethylation of C as a fixed modification and oxidation of M as a variable modification (maximum two modifications per peptide); and precursor and fragment mass tolerances of ±5 ppm and ±20 ppm, respectively. Peptide-spectrum matches, peptides and proteins were filtered to a 1% false-discovery rate. Quantification was performed using the embedded FlashLFQ and protein-inference algorithms, and the resulting LFQ (label-free quantification) intensities were used for further calculations in Perseus v1.6.2.3. Decoy hits and contaminants were filtered out, the LFQ intensities were log transformed, and only proteins that had at least 3 valid values in at least one experimental group were kept. The remaining missing values were imputed by random numbers drawn from a normal distribution with a downshift of 1.8 standard deviations and a width of 0.3, simulating low-abundance values below the detection limit. A two-sided Student’s t test was performed to identify the proteins that are differentially expressed.

Tissue expression analysis

Gzf1 expression across adult mouse tissues was obtained from the FANTOM5 expression atlas (E-MTAB-3579) as transcripts per million (TPM).

Generation of Gzf1 knockout mice by zygote electroporation

To generate Gzf1-KO mice, two synthetic guide RNAs targeting the Gzf1 coding sequence (see Supplementary Data 8) were complexed with recombinant Cas9 protein to form ribonucleoprotein (RNP) complexes. All procedured continued according to the NepaGene (NEP21) protocol for electroporation. RNPs were assembled in vitro by incubating Cas9 with each gRNA at equimolar ratios in TE buffer for 10 min at room temperature.

Zygotes were harvested from superovulated 3-week-old C57BL/6 females mated with C57BL/6 males. Zygotes were collected and washed in Opti-MEM medium. Approximately 50–100 zygotes per batch were electroporated in a 1 mm gap cuvette containing 20 μL of the RNP mixture using the NEPA Gene electroporation system under standard square-wave conditions. Following electroporation, embryos were washed and cultured in KSOM + AA medium at 37 °C and 5% CO₂ until the 2-cell stage, then transferred into the oviducts of pseudopregnant CD-1 females.

Breeding colony, litter size and genotype distribution analysis

To assess the in vivo consequences of Gzf1 loss, three breeding schemes were established: (1) Het × Homo (heterozygous × homozygous KO), (2) Het × Het (heterozygous intercross), and (3) Homo♀ × WT♂ (homozygous female × WT male). Litters were assessed at two windows: embryos were collected from timed matings at E13.5–15.5, or pups were genotyped postnatally. Postnatal genotyping was performed either at the time pups were found dead or at weaning (P21) for surviving animals; reported counts therefore include both pre-weaning deaths and weaning-age survivors. Total litter size was recorded at the time of initial assessment (Het × Homo E13.5–15.5, n = 4 litters; Het × Homo postnatal, n = 6 litters; Het × Het postnatal, n = 9 litters; Homo♀ × WT♂ postnatal, n = 4 litters). Pups were genotyped by PCR as described above.

Litter sizes were compared across groups using planned pairwise Welch’s t tests with Holm correction for multiple comparisons. Genotype distributions in Het × Homo crosses were tested against the expected 1:1 (Het:Homo) ratio using two-sided exact binomial tests, separately at E13.5–15.5 (n = 31 embryos) and postnatally (n = 22 pups). Genotype distributions in Het × Het crosses were tested against the expected 1:2:1 (WT:Het:KO) Mendelian ratio using two-sided exact binomial tests per genotype (WT vs 0.25; Het vs 0.50; KO vs 0.25), and an omnibus chi-square goodness-of-fit test was also performed (n = 52 pups). All statistics are reported in Supplementary Data 5.

Weight, gene expression, and methylation assessments in Gzf1-KO mice

Female littermates were weighed at postnatal day 25 (P25) and day 50 (P50). Sample sizes were WT (n = 4, 4), heterozygous (Het; n = 7, 6), Gzf1-KO (KO; n = 7, 5), and Het from KO♀ × WT♂ crosses (n = 5, P25 only) for P25 and P50, respectively. Within each age cohort, planned pairwise comparisons were performed using two-tailed Welch’s t tests with Holm adjustment; significance codes are ***P < 0.001, *P < 0.05, ns ≥ 0.05 (Supplementary Data 5).

For molecular analyses at E10.5, embryos were collected at embryonic day 10.5 and trunk tissue was dissected. For RNA extraction, embryonic tissue was homogenized using a plastic microtube pestle tissue grinder (Bar Naor, cat. no. BN14155390H), followed by RNA isolation using the RNeasy Mini Kit (Qiagen, cat. no. 74104) according to the manufacturer’s instructions. Gene expression analysis by qPCR was performed as described above (n = 6 per genotype, from 2 embryos with 3 technical replicates each). Genomic DNA was separately extracted from embryonic trunk tissue and subjected to bisulfite conversion, PCR amplification, and methylation assessment via next-generation sequencing (Bisulfite-NGS) at the IG-CGI, Gtl2-DMR and IG-TRE, performed as detailed above (n = 2 pooled embryos per genotype).

For methylation analysis of non-surviving neonates, tissue was collected from Gzf1-KO pups found dead at birth (P1). Genomic DNA was extracted and subjected to bisulfite-NGS at the IG-CGI as above (n = 1 KO pup). For methylation analysis of surviving adult mice, tail biopsies were collected from adult WT and Gzf1-KO animals. Genomic DNA was extracted and subjected to bisulfite-NGS at the IG-CGI and Gtl2-DMR as above. For the maternal-effect cross, heterozygous pups derived from KO♀ × WT♂ matings were genotyped by PCR and tail biopsies were subjected to bisulfite-NGS at the IG-CGI (n = 5 pups). Methylation profiles were compared with those of WT adult mice (n = 2). For Gzf1 expression across organs, brain, liver, heart, kidney and muscle were dissected from P1 WT and Gzf1-KO pups (n = 1 per genotype). RNA was extracted and Gzf1 transcript levels were measured by RT-qPCR as described above, normalized to β-actin and expressed relative to WT (2⁻ΔΔCq).

Methylation assessment in Gzf1-KO oocytes

Three-week-old WT and Gzf1-KO female mice were superovulated by injection of pregnant mare serum gonadotropin (PMSG) followed by human chorionic gonadotropin (hCG) each at 5 IU as hormonal priming 48 h apart. The following day, ~10–20 metaphase-II oocytes per mouse were collected from oviductal ampullae and denuded of cumulus cells by 20 min Hyaluronidase (60 μl in 2 ml M2) treatment followed by 10 washes in M2 medium. The oocytes were then treated with Tyrode’s solution to remove the zona pellucida and were further washed with M2 medium for an additional 10 washes. DNA from pooled oocytes underwent bisulfite conversion using the EZ DNA Methylation-Lightning Kit (Zymo Research), followed by nested PCR and next-generation sequencing (NGS) as described above. For single-read methylation visualization, individual bisulfite sequencing reads spanning the IG-CGI were extracted and the methylation state of each CpG site was scored as methylated or unmethylated. Reads were displayed as a heatmap in which each row represents a single read and each column a CpG position (5′ → 3′).

Protein cloning and expression

Linear DNA containing a T7 promoter and the Gzf1 cDNA sequence was expressed with a C-terminal Flag tag using the PURExpress In Vitro Protein Synthesis Kit (New England Biolabs) following the manufacturer’s protocol. For binding reactions, the resulting protein solution was diluted 1:24 in the binding mixture.

Protein binding microarray (PBM) experiments

The binding experiments were performed as described previously^40,41. Briefly, double-stranded microarrays were blocked with 2% (w/v) nonfat milk for 1 h. Following the blocking step and washes, the arrays were incubated for 1 h with a protein mixture in PBS buffer containing GZF1-Flag protein, 2% nonfat milk, 0.2 mg/mL bovine serum albumin, 50 ng/µL salmon testes DNA, 0.2% Triton X-100, 0.02% TX-100, 1 μM ZnCl2 and 1 mM DTT. After the binding step, bound protein was tagged with a Flag monoclonal antibody (Sigma-Aldrich) diluted 1:30 in an antibody mix containing 2% nonfat milk, PBS, and 1 mM DTT. This was followed by incubation with an Anti-mouse IgG (H + L) F(ab’)2 fragment Alexa Fluor® 647 conjugate (Cell Signaling Technologies) diluted 1:60 in the same buffer conditions. The fluorescence signal was scanned at 635 nm and quantified using a GenePix 4400 A microarray scanner and GenePix Pro analysis software. Signal was spatially adjusted according to their positions on the microarray, as previously described⁴².

Sequence libraries and microarray preparation

Three different sets of sequence libraries were synthesized and utilized on three microarrays for this study.

1. GZF1 Binding Motif Characterization: To characterize the GZF1 binding motif, we used a previously described universal design^41,42, consisting of ~15,000 unique single-stranded 60-base oligonucleotides. This DNA library contained de Bruijn sequences representing all possible 9-mers, with a constant 3’ end. Custom-designed microarrays were manufactured by Agilent Technologies and converted to double-stranded DNA via primer extension as previously described⁴⁰, using Thermo Sequenase™ DNA polymerase, dNTPs, and 10X reaction buffer (Cytiva). The reaction was carried out for 2 h with a temperature gradient ranging from 85 °C to 60 °C. Position weight matrices (PWMs) and sequence logos were generated using the “seed and wobble” algorithm⁴².

2. Methylation effect on binding: To determine the direct effect of methylation, custom 50-base-long self-complementary single-stranded DNA sequences were designed containing the GZF1 binding motif, either methylated or unmethylated at its CpG position (see Supplementary Data 8 for exact sequences). The motif sequence was selected based on the binding logo obtained from the universal design described in (i). Additionally, a scrambled variant of the motif (containing the same bases but in a random order) was selected as a negative control. The microarrays were then manufactured by LC-Sciences. The single-stranded DNA was incubated in Nuclease-Free Duplex Buffer (Integrated DNA Technologies) for 1.5 h at a temperature gradient (70 °C to 35 °C) to self-hybridize into a hairpin double-stranded form. Each sequence was present in 25 replicates, and signals were measured for all replicates (see Supplementary Data 5).

3. Effect of single base variations in the binding site: To estimate the effect of SNPs on GZF1 DNA binding, sequences were taken from a 200 bp ChIP-seq peak and divided into four distinct 60 bp segments (1–4), each containing the corresponding SNPs distinguishing the CAST and 129 alleles (see Supplementary Data 5). Sequences were obtained from Integrated DNA Technologies containing a 5’ amino modifier C6. Microarrays were spotted in-house using a SCIENION S12 ultra-low volume dispensing system onto 2D epoxy-coated glass slides. For each sequence, values were normalized to the median of the CAST allele, resulting in four separate normalization baselines. Each sequence was present in 50 replicates.

Statistics and reproducibility

Methods for statistical tests, sample sizes (n), and definitions of error bars and box-plot elements are indicated in the figure legends and Supplementary Data 5. P-values < 0.05 were considered significant. qRT-PCR fold changes, litter sizes and body weights were compared by two-tailed Welch’s t tests with Holm correction; Mendelian ratios by two-tailed exact binomial tests and a chi-square goodness-of-fit test; protein-binding-microarray intensities by two-tailed Wilcoxon rank-sum tests; GZF1 interactors by two-tailed Student’s t test; differential expression by DESeq2 (Wald test); and CRISPR-screen enrichment by MAGeCK. No statistical method was used to predetermine sample size, and genomic regions of low mappability were excluded from the analysis.

The genome-wide CRISPR screen was performed in two independent experiments, in line with common practice in the field, and bulk RNA-seq was performed on four wild-type and five Gzf1-KO biological replicates. GZF1 ChIP-seq was profiled across independent ES cell lines and clones (V6.5, RGM (129 × CAST F1) and Cas9-derived) using two antibodies (anti-Flag for tagged GZF1 and anti-GZF1 for the endogenous protein), together with no-Flag and Gzf1-KO controls; the concordant binding observed across these lines and antibodies establishes reproducibility. The GZF1-Flag interactome was profiled by immunoprecipitation-mass spectrometry in four biological replicates per condition; one GZF1-Flag replicate was identified as an outlier during quality control and excluded, yielding n = 3 GZF1-Flag and n = 4 control.

DNA methylation was profiled by targeted bisulfite-amplicon sequencing (percentage of methylated reads per CpG, at least 20× coverage), with one sample per genotype, the individual ES cell line (isogenic to its wild-type parent), embryo pool, or animal, alongside wild-type littermate controls of the same genetic background. As is standard for targeted differentially-methylated-region profiling, per-CpG methylation is summarized in box plots and is corroborated by allele-specific expression and RT-qPCR in biological replicates. Heterozygous pups (Supplementary Fig. 6) were profiled on five independent animals, whereas single samples were used for the dead-pup and adult-tissue methylation; as the intention was to demonstrate the in vivo phenotype, one sample was sufficient for these analyses. RT-qPCR experiments were performed in three technical replicates; one Rian/Mirg measurement in Fig. 4h was excluded owing to high variance among its technical replicates, indicating an unreliable quantification. The experiments were not randomized and the Investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information ^{(3.6MB, pdf)}

41467_2026_76890_MOESM2_ESM.pdf ^{(90.8KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1 ^{(1.8MB, xlsx)}

Supplementary Data 2 ^{(8.5MB, xlsx)}

Supplementary Data 3 ^{(1.8MB, xlsx)}

Supplementary Data 4 ^{(8.7MB, xlsx)}

Supplementary Data 5 ^{(1.9MB, xlsx)}

Supplementary Data 6 ^{(733.9KB, xlsx)}

Supplementary Data 7 ^{(167.3KB, xlsx)}

Supplementary Data 8 ^{(23.8KB, xlsx)}

Supplementary Data 9 ^{(42.6KB, xlsx)}

Reporting Summary ^{(110.6KB, pdf)}

Transparent Peer Review file ^{(1.6MB, pdf)}

Source data

Source Data ^{(2.4MB, xlsx)}

Author contributions

Conceptualization: A.T., L.H., and Y.S.; Experimentation, data analysis, and interpretation: A.T., L.H., R.S.R., Y.M., R.B.Y., L.S. and Y.S.; Computational analysis: L.H., A.L., R.S.R., H.R., Amos T.; PBM experiments and analysis: I.M., A.A.; Transgenic manipulation and animal handling: A.H.O.; Mass spectrometry: D.S.; Histone PTM profiling: A.T., A.W.; Manuscript Writing: A.T., L.H., and Y.S. with input from other authors.

Peer review

Peer review information

Nature Communications thanks Maxim Greenberg, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Y.S. is the incumbent of the Louis and Ida Rich Career Development Chair and is supported by the European Research Council (ERC_CoG EmbryoCellEnsemble), the ISF (2824/24), the Minerva Foundation, the Helen and Martin Kimmel Stem Cell Institute, and research grants from the Estate of Hermine Miller, from Asher Pertman and Wayne Pertman, and the Dr. Beth Rom-Rymer Stem Cell Research Fund. This research was further supported by the EKARD Research School of Biological Science, John L. and Vera Schwartz, the Abisch-Frenkel Foundation for the Promotion of Life Sciences, the Dr. Dvora and Haim Teitelbaum Endowment Fund, and Magnus Konow in honour of his mother Olga Konow Rappaport.

Data availability

The sequencing data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under SuperSeries accession GSE337304, comprising bisulfite-amplicon sequencing GSE337139, the genome-wide CRISPR screen GSE337141, ChIP-seq GSE337143, bulk RNA-seq GSE337302 and H3K4me3 CUT&Tag GSE337303. Protein-binding microarray (PBM) data are available under GEO accessions GSE337745 and GSE324373. The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository under accession PXD080456. Publicly available datasets re-analyzed in this study are available under GEO accessions GSE49624, GSE42836 and GSE123942, and ENA accession ERP000689. Cell lines and mouse strains generated in this study are available from the corresponding author under a material transfer agreement with the Weizmann Institute of Science. Source data are provided with this paper.

Code availability

Data analysis was performed using custom R scripts and publicly available packages and softwares. All codes generated in this study are available upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76890-w.