Epigenetic Mechanisms Governing Female and Male Germline Development in Mammals

Recent advances in high-throughput sequencing technologies and highly sensitive methods for the analyses of chromatin-associated factors have enabled mapping of the epigenome in unprecedented resolution from small numbers of cells, including in the early embryo and developing germ cells. Such epigenomic profiles have led to many unexpected findings, including the dramatic differences in the distribution of histone PTMs as well as DNAme in the germ cells of female and male mice. These sex-specific epigenomic profiles include broad domains of H3K4me3 (bdH3K4me3) in oocytes [Dahl et al., 2016; Liu et al., 2016; Zhang et al., 2016] and profiles of H3­K36me2 that are distinct between growing oocytes and PSG, the stages in which de novo DNAme occurs in each sex, respectively [Shirane et al., 2020]. Tissue-specific disruption of chromatin-modifying enzymes in the mouse followed by epigenomic profiling has uncovered a complex interplay between these epigenetic modifiers and the marks that they deposit, and shed light on the influence of these interactions on the transcriptome. In this review, we describe: (1) the dynamics of DNAme in male and female germ cell development, (2) the mechanisms of widespread erasure and distinct pathways engaged in the establishment of DNAme in PSG and growing oocytes, and (3) the divergent distribution and crosstalk between histone PTMs, including H3K4me3, H3K27me3, and H3­K36me2/me3, and the influence of these complex interactions on the transcriptome. Finally, we discuss outstanding questions relating to the underlying mechanisms of the sexually dimorphic patterns of epigenetic marks observed in the mammalian germline and the impact of these differences on transcription from the parental genomes in the early embryo and beyond.

DNAme in mammalian cells is predominantly deposited on the fifth position of the cytosine base (5-methylcytosine, 5mC) in the context of CpG dinucleotides. Following DNA-replication, UHRF1 (also known as NP95 or ICBP90) (shown in Fig. 1a) recognizes 5mC on the parent strand of the hemimethylated template and recruits the maintenance DNA methyltransferase DNMT1 to catalyze methylation of the CpG on the newly synthesized complementary strand, thereby maintaining the DNAme pattern following DNA-replication [Bostick et al., 2007; Sharif et al., 2007]. In contrast, de novo DNAme, catalyzes the deposition of 5mC at unmethylated cytosines predominantly in the context of CG and to a lesser extent at CH (H = A, C or T) dinucleotide, is catalyzed by members of the DNMT3 family of methyltransferases, which are particularly active during embryonic and germ cell development [Greenberg and Bourc’his, 2019]. Two catalytically active DNMT3 proteins, DNMT3A and DNMT3B, have been identified in mammals, along with one catalytically inactive paralog, DNMT3L [Greenberg and Bourc’his, 2019]. A third rodent-specific catalytically active family member, DNMT3C, is expressed in the male germline [Barau et al., 2016].

In addition to the catalytic domain, DNMT3A and DNMT3B possess unique domains that interact with histone PTMs. For example, the PWWP domain of DNMT3A and DNMT3B binds specifically to di- or tri-methylated H3K36 (H3K36me2/me3) [Dhayalan et al., 2010; Dukatz et al., 2019], whereas methylation of H3K4 inhibits binding of the ADD domain (ATRX-DNMT3-DNMT3L) of DNMT3A and DNMT3B [Zhang et al., 2010], as well as DNMT3L [Jia et al., 2007; Ooi et al., 2007] (shown in Fig. 1b). Importantly, congenital and/or de novo mutations in these domains have been identified in human congenital disorders as well as recurrently in specific cancers [Li et al., 2019], revealing that de novo DNA methyltransferases play a critical role in disease (discussed in detail below). DNMT3C, which arose as a consequence of duplication of the Dnmt3b gene [Molaro et al., 2020], retains the ADD domain but lacks a PWWP domain [Barau et al., 2016] (shown in Fig. 1b). Notably, DNMT3C plays an essential role in de novo DNAme of evolutionarily young retrotransposons in male germ cells, likely targeted by a specialized PIWI-interacting RNA (piRNA)-pathway [Watanabe et al., 2011] (discussed in detail below). Unlike these catalytically active DNA methyltransferases, DNMT3L lacks DNMT activity but stimulates the activity of both DNMT3A and DNMT3B during the widespread waves of de novo DNAme that occur in both female and male germ cells [Bourc’his et al., 2001] and the early embryo [Guenatri et al., 2013], respectively. In addition to the DNMT3 family proteins, several studies reveal that DNMT1 exhibits weak de novo DNMT activity in mouse embryonic stem cells (mESCs) [Lorincz et al., 2002; Yarychkivska et al., 2018], in particular at H3K9me3-marked regions [Wang et al., 2020]. While this activity is also directed toward H3K9me3-enriched transposable elements (TEs) in vivo [Haggerty et al., 2021], the role of DNMT1 in de novo DNAme of TEs in the germline remains to be determined.

Conversely, DNAme is lost either by active or passive DNA demethylation. The former is initiated at least in part by oxidation of 5mC [Wu and Zhang, 2010; Branco et al., 2011; Wu and Zhang, 2017], mediated by the ten-eleven translocation (TET) methylcytosine dioxygenases family proteins TET1, TET2, and TET3 (shown in Fig. 1c), which progressively oxidize 5mC to 5-hydroxymethylcytosine (5hmC), 5-formylcytosine, and 5-carboxylcytosine. These 5mC derivatives are then either removed by thymine DNA glycosylase followed by base excision repair or diluted following replication due to the low affinity of UHRF1, the binding partner of DNMT1, for oxidized cytosines [Hashimoto et al., 2012; Otani et al., 2013]. In contrast, passive demethylation involves breakdown of maintenance DNAme activity, leading to replication-coupled dilution of 5mC [Wu and Zhang, 2017]. The combination of these two mechanisms likely drives widespread DNA demethylation in both the early embryo [Guo et al., 2014; Shen et al., 2014] and developing germ cells [Seisenberger et al., 2012; Hackett et al., 2013; Kagiwada et al., 2013; Yamaguchi et al., 2013b]. Importantly, active demethylation of specific germline genes harboring CpG island (CGI) promoters, including many involved in meiosis or genome defense, clearly plays a role in their transcriptional activation during germ cell development [Borgel et al., 2010; Guibert et al., 2012; Yamaguchi et al., 2012; Hackett et al., 2013; Hargan-Calvopina et al., 2016; Dahlet et al., 2020, 2021; Mochizuki et al., 2021].

DNAme is highly dynamic during embryonic and germ cell development (shown in Fig. 2a, b). Whole-genome bisulfite sequencing (WGBS), in particular the amplification-free post-bisulfite adaptor-tagging method [Miura et al., 2012], enables the mapping of DNAme at a single base resolution from as few as 800 diploid cells [Kobayashi et al., 2012]. Pioneering “DNA methylome” studies by WGBS revealed that ∼70% of the genome is methylated in mouse epiblast cells, which are the precursors of somatic as well as germ cells [Seisenberger et al., 2012]. Subsequently, during specification and differentiation of the germ cell lineage, DNAme is progressively lost with only ∼30% and ∼3% of the genome methylated in primordial germ cells (PGCs) at E9.5 and E13.5, respectively [Seisenberger et al., 2012; Kobayashi et al., 2013]. Consistent with WGBS analysis, liquid chromatography coupled with mass spectrometry analysis of genomic DNA purified from developing PGCs revealed that the levels of 5mC are low at E9.5 compared to mESCs and exhibit a further decrease in E13.5 PGCs [Hill et al., 2018]. Such analysis also clearly shows that the total amount of 5hmC in the genome in PGCs at both early and late stages is significantly lower than that of 5mC, indicating that the global reduction of DNAme occurs predominantly independent of the conversion of 5mC to 5hmC [Hill et al., 2018]. However, oxidation of 5mC is prevalent at a subset of promoter regions that show an apparent delay in DNA demethylation in early PGCs, perhaps reflecting the importance of this active demethylation pathway at specific genomic regions (discussed in detail below).

Two phases of DNA demethylation have been observed in male and female PGCs, with the initial phase of DNA demethylation occurring between E6.5 and E9.5 in mice [Seisenberger et al., 2012; Kobayashi et al., 2013; Hill et al., 2018]. While the majority of PGCs are in the G2 phase and slowly proliferate [Seki et al., 2007], this widespread erasure of DNAme can be explained by repression of both de novo and maintenance DNAme activities in PGCs, coupled with DNA-replication [Kagiwada et al., 2013]. Consistent with a role for replication-coupled passive DNA demethylation, PGCs proliferate twice per day between E9.5 and E12.5 [Kagiwada et al., 2013]. A recent study exploiting an in vitro PGC-like cell culture system which mimics the dynamics of DNAme in in vivo PGC differentiation [Hayashi et al., 2011; Shirane et al., 2016] revealed that PRDM14, a key transcription regulator for PGC specification [Yamaji et al., 2008], is required for widespread reduction of DNAme in PGC-like cells, likely through transcriptional repression of Uhrf1, Dnmt3a, Dnmt3b, and Dnmt3l [Shirane et al., 2016]. As PRDM14 binds at Dnmt3b and Uhrf1 loci in mESCs [Magnusdottir et al., 2013], this factor may play a direct role in the regulation of DNAme homeostasis. In addition, PRDM14 recruits TET1/TET2 to the Stella (also known as Dppa3 or PGC7) locus, and in turn promotes DNA demethylation of the promoter region of this gene, in association with its transcriptional activation [Mulholland et al., 2020]. As DPPA3 displaces UHRF1 from chromatin, activation of this gene inhibits maintenance DNAme activity [Mulholland et al., 2020]. These observations suggest that PRDM14 likely drives genome-wide DNA demethylation both by transcriptional silencing of DNAme writers as well as indirectly via promoting TET-mediated activation of DPPA3 expression, which in turn disrupts maintenance DNAme.

While both sexes show widespread decreases in DNAme during this first wave of genome-wide DNA demethylation, specific genomic regions, including germline genes, imprinted loci, and young TEs, including members of the intracisternal A-particle (IAP) family, retain relatively higher levels of DNAme [Seisenberger et al., 2012; Kobayashi et al., 2013]. While the chromatin feature(s) responsible for the resistance of specific genomic regions to passive DNA demethylation remain to be experimentally confirmed, regions showing relatively high levels of DNAme are marked by H3K9me3 in E13.5 PGCs [Liu et al., 2014] as well as early embryonic cells, indicating that H3K9me3 may play a direct role in this process. Indeed, long-terminal repeat (LTR) retrotransposons showing the highest levels of DNAme in wildtype (WT) E13.5 PGCs, predominantly IAP elements, show decreased DNAme in Setdb1 germline conditional knockout (KO) PGCs at this stage [Liu et al., 2014]. These retroelements also show loss of H3K9me3 and are concomitantly derepressed. A similar reduction of DNAme was observed in Setdb1 KO growing oocytes and embryos derived from these SETDB1-deficient gametes [Zeng et al., 2019] and LTR retrotransposons are also derepressed in Setdb1 KO oocytes [Eymery et al., 2016; Kim et al., 2016]. In addition to the potential role of this chromatin mark in inhibiting TET activity [Blaschke et al., 2013], the resistance of H3K9me3-marked regions to DNA demethylation may also involve “preferential” maintenance DNAme. Consistent with this model, the tandem Tudor Domain and plant homeodomain of UHRF1 (shown in Fig. 1a) bind to H3K9me3 and the H3 tail, respectively [Karagianni et al., 2008; Bartke et al., 2010; Nady et al., 2011; Arita et al., 2012; Xie et al., 2012; Cheng et al., 2013; Rothbart et al., 2013; Tauber et al., 2020]. UHRF1 binding subsequently promotes DNMT1 recruitment and in turn maintenance DNAme [Rothbart et al., 2012; Bashtrykov et al., 2014; Vaughan et al., 2018]. Although UHRF1 is expressed at relatively low levels in PGCs and is localized primarily in the cytoplasm [Seisenberger et al., 2012; Kagiwada et al., 2013], "residual" levels of UHRF1 in the nucleus may be sufficient to stimulate maintenance DNAme specifically at H3K9me3-marked regions, which represent ∼8% of the genome at E13.5 [Liu et al., 2014]. In addition, a recent study shows that the interaction between H4K20me3 and the BAH1 domain of DNMT1 (shown in Fig. 1b) stimulates DNMT1 catalytic activity [Ren et al., 2021]. As H4K20me3 is generally co-localized with H3K9me3, this histone PTM may also promote maintenance DNAme at H3K9me3-marked regions in PGCs. While the apparent resistance to DNA demethylation may be a direct effect of the presence of H3K9me3, binding of readers of this mark, such as HP1 proteins or ATRX [Eustermann et al., 2011; Iwase et al., 2011; Magaraki et al., 2017] or the association of the linker histone H1 with H3K9me3 may also play a role [Healton et al., 2020; Sheban et al., 2022].

Over the last decade, the mechanisms by which specific genomic regions are targeted for H3K9me3 deposition have begun to emerge. For example, KRAB-ZFP proteins, acting in concert with KAP1/TRIM28, recruit the H3K9 KMTase SETDB1 to specific TEs [Matsui et al., 2010; Rowe et al., 2010; Elsasser et al., 2015; Sadic et al., 2015; Hoelper et al., 2017], while the HUSH complex recruits SETDB1 to young LINE1 elements that are enriched within transcriptionally permissive euchromatin [Liu et al., 2018; Robbez-Masson et al., 2018; Douse et al., 2020]. In contrast, heterodimers of the basic-helix-loop-helix-leucine zipper proteins MGA and MAX recruit SETDB1 to a subset of germline genes [Karimi et al., 2011; Mochizuki et al., 2021]. Whether H3K9me3 deposited in these contexts safeguards against DNA demethylation (discussed below) and/or promotes DNMT1-mediated de novo DNAme in PGCs remains to be determined. Regardless of the mechanism, maintenance of H3K9me3-marked germline genes in a silent state during early PGC development is apparently essential, as earlier DNA demethylation of meiotic as well as genome defense genes [Mochizuki et al., 2021] in Dnmt1 germline conditional KO PGCs leads to their precocious expression, in association with hypogonadism and infertility [Hargan-Calvopina et al., 2016].

While replication-coupled passive DNA demethylation dominates in early PGCs, the second wave of DNA demethylation that takes place in later PGCs occurs coincident with the conversion of 5mC into 5hmC, including in the promoter regions of a subset of germline genes and gDMRs that show delayed demethylation relative to the rest of the genome [Yamaguchi et al., 2012; Hackett et al., 2013]. Notably, the CGI promoter regions of a subset of these “late-demethylating” germline genes are clearly marked by H3K9me3 in the epiblast as well as in E9.5 PGCs, albeit at reduced levels in the latter [Mochizuki et al., 2021]. H3K9me3 is subsequently lost over these promoter regions by E13.5 [Liu et al., 2014], coincident with their upregulation [Mochizuki et al., 2021]. As mentioned above, it is possible that H3K9me3 or readers of this histone PTM protects marked regions against the initial wave of passive DNA demethylation. However, deletion of Tet1 clearly leads to the failure of 5mC removal at a specific subset of imprinted gDMRs [Yamaguchi et al., 2013b] as well as germline genes, including many involved in meiosis, implicating active demethylation of these regions at later stages of PGC development [Yamaguchi et al., 2012].

Consistent with this observation, maternal deficiency of vitamin C, which stimulates the activity of TET1, leads to incomplete DNA demethylation of specific meiotic genes, coincident with lower expression levels of these genes in the germ cells of developing embryos, indicating that vitamin C supplementation during gestation plays a critical role in DNAme homeostasis in the germline of the next generation [DiTroia et al., 2019]. Intriguingly, a recent study proposes that TET1 actually counteracts residual de novo DNAme activity during this second wave in PGCs, in association with transcriptional activation of germline genes showing TET1-dependent hypomethylation in their promoter regions [Hill et al., 2018]. Regardless, while Tet1 KO female PGCs show downregulation of germline genes and in turn hypogonadism, Tet1 KO males are fertile and their testes are morphologically indistinguishable from WT [Yamaguchi et al., 2012]. Thus, Tet1 deficiency differentially impacts the transcriptome in male and female germ cells. Notably, while male PGCs are arrested in G0 at ∼E13.5, female PGCs enter meiotic prophase around this time. The observed sexually dimorphic impact of Tet1 deletion may reflect the female-specific dependence of entry into the meiotic program on expression of specific transcription factors that require TET1 activity for their timely expression [Soh et al., 2015].

In addition to germline gene promoters, retrotransposons marked by H3K9me3 [Karimi et al., 2011] are also apparently resistant to the global wave of TET-dependent DNA demethylation that occurs in mESCs following the addition of vitamin C to the culture medium [Blaschke et al., 2013]. As 5hmC accumulates at H3K9me3-marked regions in mESCs cultured with vitamin C as well as at pericentromeric heterochromatin in developing germ cells in both male and female [Yamaguchi et al., 2013a], it is possible that readers of 5hmC protect these regions against further oxidation by TET proteins. For example, UHRF2, a closely related paralog of UHRF1 (shown in Fig. 1a), was shown to bind preferentially to 5hmC [Zhou et al., 2014], though the relevance of this interaction to DNAme maintenance at H3K9me3-enriched regions has not been addressed. On the other hand, a recent report revealed that in the absence of SETDB1, TET2 does not regulate DNAme at IAP retrotransposons directly, but rather promotes loss of the repressive mark H4R3me2 at these elements [Deniz et al., 2018]. Regardless, as pericentromeric heterochromatin is generally marked by H3­K9me3, persistence of 5hmC at these regions may reflect a role for H3K9me3 in the “protection” of 5hmC against further oxidation during the initial wave of widespread loss of DNAme in PGCs. Consistent with this model, deletion of the H3K9 KMTases Suv39H1 and Suv39H2 in mESCs leads to a decrease of DNAme at satellite repeats [Lehnertz et al., 2003]. However, as DNAme broadly increases over late-replicating regions in E8.5 Tet-TKO embryos [Cheng et al., 2022], which are generally enriched for H3K9me3, the inhibition of active demethylation at heterochromatic regions may be restricted to specific developmental stages.

While the global loss of DNAme occurs in PGCs of both sexes with similar kinetics, widespread de novo DNAme occurs in distinct developmental stages in the female versus male germline (shown in Fig. 2b). In the former, de novo DNAme initiates postnatally in growing oocytes (which are arrested in prophase 1 of meiosis) and is almost completed at the fully grown stage. In contrast, de novo DNAme in male germ cells initiates at ∼E15.5 in mitotically arrested PSG and is almost completed by birth [Sasaki and Matsui, 2008; Kurimoto and Saitou, 2019] (shown in Fig. 2b). As female and male germ cells are non-dividing during their respective waves of de novo DNAme, a role for replication-coupled dilution of DNAme can be formally excluded, making the germline an ideal lineage to study the mechanisms underlying de novo DNAme. In addition to the clear differences in the timing of de novo DNAme, levels of DNAme in mature gametes are strikingly dimorphic, with ∼40% of the genome methylated in mouse oocytes, versus ∼80% methylated in sperm [Kobayashi et al., 2012]. Integrative analysis of transcriptome and WGBS datasets revealed that essentially all of the genomic regions that are robustly de novo methylated in oocytes coincide with transcribed regions [Smallwood et al., 2011; Kobayashi et al., 2012; Veselovska et al., 2015], with a minor fraction of intergenic regions showing modest levels of de novo methylation [Yano et al., 2022]. Subsequently, chromatin immunoprecipitation (ChIP) analyses revealed that gene bodies of active genes are also marked by H3K36me3, indicating that this transcription-coupled histone PTM may promote de novo DNAme in oocytes [Stewart et al., 2015; Brind’Amour et al., 2018]. In mature male germ cells in contrast, most of the genome is densely methylated, with the exception of regions enriched for H3K4me3 [Brykczynska et al., 2010; Erkek et al., 2013], indicating that this mark, which is normally associated with promoter regions [Barski et al., 2007], may inhibit de novo DNAme in PSG. Regardless, these descriptive studies indicated that distinct chromatin features may guide de novo DNAme in male versus female germ cells, which in turn stimulated the generation and analysis of germ cell-specific KOs of relevant chromatin factors, and detailed analyses of their expression patterns, as discussed in detail below.

As discussed above, ChIP-seq analysis using low-input methods has revealed that transcribed regions are marked by H3K36me3 in mouse oocytes [Stewart et al., 2015; Brind’Amour et al., 2018; Xu et al., 2019] and essentially all regions enriched for this mark are de novo methylated in these cells [Xu et al., 2019; Shirane et al., 2020]. H3­K36me3 is deposited in mammalian cells exclusively by SETD2, one of several genes encoding H3K36 KMTases (shown in Fig. 3a) [Edmunds et al., 2008]. Strikingly, deletion of Setd2 in oocytes leads to a dramatic reduction of H3K36me3 in transcription units of active genes, concomitant with reduced DNAme in these regions, indicating that SETD2 plays an instructive role in directing de novo DNAme in mouse oocytes [Xu et al., 2019] (shown in Fig. 3b). Indeed, gDMRs methylated in WT oocytes also show reduced DNAme in Setd2-deficient oocytes (see below for discussion of genomic imprinting). These observations are consistent with in vitro evidence showing that the PWWP domain of DNMT3A can bind to H3K36me2/me3 [Dhayalan et al., 2010], and the disruption of this interaction by the substitution of aspartic acid at residue 329 within the PWWP domain of mouse DNMT3A to alanine (D329A) [Dhayalan et al., 2010; Sendzikaite et al., 2019].

Loss-of-function mutations in the PWWP domain of human DNMT3A were recently identified in the overgrowth disorder Tatton-Brown-Rahman syndrome (TBRS), in association with reduced DNAme in the peripheral blood of TBRS subjects [Tatton-Brown et al., 2014; Jeffries et al., 2019]. In contrast, gain of function mutations in the PWWP domain of DNMT3A were found in subjects with Heyn-Sproul-Jackson syndrome, which is characterized by microcephalic dwarfism and impaired intellectual development [Heyn et al., 2019]. In contrast to TBRS, Heyn-Sproul-Jackson syndrome subjects show gain of DNAme at regions that are normally unmethylated [Heyn et al., 2019]. As described above, these observations are consistent with in vitro data indicating that the PWWP domain of DNMT3A plays a critical role in establishing the appropriate DNAme landscape in humans [Tatton-Brown et al., 2014; Heyn et al., 2019; Jeffries et al., 2019; Sendzikaite et al., 2019]. Intriguingly, mice carrying the D329A mutation exhibit a dwarfism phenotype but show normal DNAme at H3K36me3 regions in liver and brain [Sendzikaite et al., 2019]. Rather, Polycomb-targeted regions including Hox cluster genes, show an aberrant gain of DNAme, presumably due to ectopic targeting of DNMT3A^D329A [Sendzikaite et al., 2019]. Detailed analyses of DNMT3A1 either lacking the PWWP domain or harboring mutants in this domain that ablate binding to H3K36me2/me3 reveal aberrant targeting of this isoform in mesenchymal stem-like cells (MSCs) specifically to H2AK119ub1 marked CGIs [Weinberg et al., 2021]. Such binding to this PRC1-deposited mark is dependent on the amino-terminal tail of DNMT3A1, which is unique to this isoform of DNMT3A [Weinberg et al., 2021]. Curiously, oocytes harboring the D329A mutation show minimal loss of DNAme at H3K36me3 marked regions but rather aberrant gain of DNAme at regions that are normally hypomethylated [Kibe et al., 2021], indicating that DNMT3A^D329A can still catalyze de novo DNAme at H3K36me3 marked regions. While it is tempting to speculate that DNMT3A^D329A may be recruited by H2AK119ub1 in oocytes, these cells express predominantly the DNMT3A2 isoform, which lacks the amino-terminal domain reported to orchestrate the interaction with H2AK119ub1 [Weinberg et al., 2021]. Intersection of the recently published H2AK119ub1 profile in oocytes [Chen et al., 2021; Mei et al., 2021] with regions that gain DNAme in the D329A mutant oocytes would answer whether DNMT3A^D329A is aberrantly recruited to PRC1-targeted regions in oocytes as observed in MSCs. Notably, as NSD proteins make extensive intermolecular contacts with the C-terminal region of H2A including residue 119 [Li et al., 2021] and H2AK119ub1 was shown to inhibit the activity of NSD1 in vitro [Yuan et al., 2013], this mark may actually inhibit de novo DNAme in WT cells by blocking H3K36me2 deposition and in turn DNMT3A2 recruitment. While the PWWP domain of DNMT3A is clearly important for the binding to H3K36me2 and H3K36me3, determining whether other chromatin features also play a role in the recruitment of DNMT3A is clearly worthy of further study.

Analysis of the distribution of H3K36me3 in male PGCs revealed that some regions that are de novo methylated by E16.5 are already marked by H3K36me3 at E13.5 [Morselli et al., 2015]. This correlation led the authors to propose that H3K36me3 guides de novo DNAme in male germ cells. However, a subsequent study by our group revealed that, while a subset of gene body regions is indeed marked by H3K36me3 and de novo methylated at E16.5 PSG, intragenic regions of silent genes and many intergenic regions that lack H3K36me3 are also de novo methylated [Shirane et al., 2020]. Furthermore, germ cell-specific deletion of Setd2 minimally impacts DNAme, definitively showing that SETD2 is dispensable for the global wave of de novo DNAme in the male germline [Shirane et al., 2020]. Rather, essentially all of the genomic regions that are de novo methylated in E16.5 PSG are marked by H3K36me2. As the distribution of H3K36me2 was not altered upon deletion of Setd2 in E16.5 PSG, we surmised that another H3K36 KMTase, such as NSD1, NSD2, NSD3, and ASH1L (shown in Fig. 3a), likely deposits this mark in PSG. Notably, biochemical analysis revealed that DNMT3A can also bind to H3K36me2 with a relatively higher affinity than H3K36me3 [Weinberg et al., 2019]. Consistent with this observation, deletion or loss-of-function mutations of Nsd1 lead to hypomethylation in mESCs and MSCs as well as in several human diseases, including squamous cell carcinomas [Papillon-Cavanagh et al., 2017; Weinberg et al., 2019] and Sotos Syndrome [Choufani et al., 2015]. Accordingly, we found that PSG isolated from mice harboring a germline-specific Nsd1 KO exhibit widespread failure of de novo DNAme at H3K36me2-depleted regions, with ∼33% of regions showing a significant reduction of DNAme [Shirane et al., 2020]. The gDMRs of all three paternally imprinted genes also show loss of DNAme to varying degrees (see below), revealing that NSD1 also plays a role in the establishment of DNAme at gDMRs in the male germline. In contrast, gene bodies of active genes are still efficiently de novo methylated in the absence of NSD1, indicating that SETD2 activity is not dependent upon NSD1 and that SETD2-dependent H3K36me3 deposition is sufficient to promote DNMT3A binding to these regions in the male germline. Intriguingly, regions retaining H3K36me2 in the Nsd1 KO PSG also show residual de novo DNAme, likely reflecting the activity of H3K36 KMTases other than NSD1 or SETD2 that deposit H3K36me2 and in turn promote de novo DNAme [Shirane et al., 2020]. Thus, while widespread de novo DNAme in male germ cells is dependent in part on NSD1-deposited H3K36me2, an as yet to be determined alternative H3K36 KMTase is likely responsible for deposition of the residual H3K36me2 observed in Nsd1 KO PSG, and in turn de novo DNAme of these regions. Furthermore, as mentioned below, de novo DNAme of young TEs, which make up a relatively small fraction of the genome, is dependent upon a specialized piRNA-dependent pathway that apparently functions independent of NSD1 [Shirane et al., 2020] and may operate independent of H3K36 methylation altogether.

Consistent with the sexually dimorphic pattern of DNAme and other chromatin marks, oocytes and E16.5 PSG show distinct patterns of H3K36me2 genome-wide. In contrast to PSG, H3K36me2 in oocytes is predominantly enriched in the 5′ and 3′ regions of gene bodies that are transcribed and enriched in H3K36me3 (shown in Fig. 3b). Furthermore, deletion of Setd2 in oocytes results in the loss of both H3K36me3 and H3K36me2 in gene bodies, indicating that intragenic H3K36me2 in oocytes is dependent on SETD2 rather than NSD1 or any other H3K36 KMTase [Shirane et al., 2020]. Importantly, the residual H3K36me2 in Setd2 KO oocytes represents only ∼6.6% of the oocyte genome, indicating that other H3K36 KMTases (shown in Fig. 3a) play a relatively minor role in depositing this mark in oocytes. Consistent with these observations, a recent proteomics study showed that the level of NSD1 protein in mouse oocytes is low [Israel et al., 2019]. This may be explained by the fact that the predominant mRNA expressed from the Nsd1 locus in oocytes, which initiates in a murine-specific LTR, encodes a truncated protein which lacks the amino-terminal PWWP domain. Indeed, a previous study showed that the amino-terminal PWWP domain of NSD2 is essential for the spreading of H3K36me2 in cultured human cells [Sankaran et al., 2016]. Thus, this noncanonical form of NSD1 may act in a dominant negative fashion, reducing the activity of canonical NSD1. Specific deletion of the LTR driving expression of this truncated isoform of NSD1 in oocytes would address this question. Notably, a recent study of mouse oocytes expressing the oncohistone H3.3K36M revealed that accumulation of H3K36me2 outside of gene body regions is likely dependent on the activity of NSD1 or NSD2 [Yano et al., 2022]. Intriguingly, this study also reported that H3K36me2 is generally enriched at a higher level on the X-chromosome relative to the autosomes and that expression of H3.3K36M results in a reduction of the intermediate level of DNAme observed at regions enriched for H3K36me2 on this sex chromosome. What role H3­K36me2 enrichment on the X-chromosome plays in oocyte development or postfertilization remains to be determined.

In summary, these observations clearly show that distinct H3K36 KMTases play an instructive role in the divergent patterns of DNAme observed in mouse oocytes and sperm. Whether differential expression of H3K36 KMTases in germ cells is a common feature in mammals remains to be formally established, but a recent comparative epigenome study revealed that H3K36me2 and DNAme extend beyond transcribed bodies of genes in pig and cow oocytes and that the patterns of DNAme in eggs and sperm are more similar to each other in these species than in rodents [Lu et al., 2021].

H3K4me3 generally marks the promoter regions of active or poised genes. However, H3K4me3 profiling utilizing low-input ChIP-seq methods revealed that germinal vesicle oocytes show a much broader distribution of H3K4me3 than the canonical profile observed in somatic cells [Dahl et al., 2016; Liu et al., 2016; Zhang et al., 2016]. Such bdH3K4me3 envelop ∼22% of the oocyte genome [Dahl et al., 2016], covering not only promoter regions but also intergenic regions. As H3K4me3 in oocytes from postnatal day 5 mice shows the canonical pattern, expansion of this mark must occur during oocyte growth [Hanna et al., 2018]. Such spreading of H3K4me3 is apparently dependent upon MLL2, which is broadly recruited to unmethylated CGI promoters in oocytes and clearly required for H3K4me3 at bivalent CGI promoters in somatic cells [Hu et al., 2013]. Notably, EZHIP, an inhibitor of PRC2 activity, is expressed in growing mouse oocytes and broadly restricts H3K27me3 deposition [Ragazzini et al., 2019]. This reduced level of H3K27me3 may potentiate spreading of H3K4me3 into flanking regions, leading to bdH3K4me3 formation. Regardless, consistent with the biochemical evidence for an antagonistic relationship between H3K4me3 and DNAme, bdH3K4me3 are devoid of DNAme in mouse oocytes (shown in Fig. 3b) [Hanna et al., 2018]. Deletion of Dnmt3a/3b in oocytes results in modest spreading of H3K4me3 in a subset of regions that lose DNAme, suggesting that DNAme may limit the spreading of bdH3K4me3 into specific genomic regions that are normally DNA methylated [Hanna et al., 2018], perhaps by inhibiting binding of the MLL2 CXXC motif to CpG-rich regions [Hu et al., 2017]. Alternatively, the low levels of H3K36me2 in oocytes may somehow potentiate H3K4me3 spreading. Regardless, many bdH3K4­me3 persist in preimplantation stage embryos and are associated with zygotic genome activation (ZGA) in the late 2-cell embryo [Dahl et al., 2016]. Though not as prevalent as in mouse oocytes, bdH3K4me3 are also observed in human somatic cells and are associated with increased transcription and enhancer activity of a subset of tumor suppressor genes [Chen et al., 2015]. However, bdH3K4­me3 are not present in human oocytes [Xia et al., 2019], revealing that they are dispensable for ZGA in human embryos.

H3K4me3 and H3K9me3 generally mark distinct genomic regions [Zhang et al., 2015], consistent with their putative roles in promoting or inhibiting transcription, respectively. Recent work demonstrated that depletion of the H3K9 demethylase KDM4A in oocytes leads to spreading of H3K9me3 into bdH3K4me3, in association with the failure of ZGA and developmental arrest in maternal KO embryos [Sankar et al., 2020]. Another example of potential antagonism between histone marks was observed in Setd2-deficient oocytes, where H3K4me3 or H3K27me3 spread into H3K36me3-depleted regions (see below) [Xu et al., 2019]. While it is tempting to speculate that spreading of H3K4me3 during oocyte growth prevents the encroachment of repressive histone PTMs into such domains, bdH3K4me3 is itself apparently a repressive mark in mature oocytes [Zhang et al., 2016]. The underlying mechanism of the apparent exclusion of other histone H3 PTMs in (bd)H3K4me3 domains warrants further investigation.

In contrast, H3K4me3 in male germ cells shows a pattern similar to that observed in somatic cells, with promoter regions enriched for this modification and little spreading beyond these regions [Erkek et al., 2013]. Interestingly, Kdm4a KO males exhibit no apparent phenotype [Sankar et al., 2017], revealing that restricting H3K9­me3 spreading may be more important in the female than the male germline. One obvious explanation for this difference is the presence of protamines rather than canonical histones at promoter-distal regions in mature sperm [Brykczynska et al., 2010], which may obviate the requirement for H3K9 demethylase activity in the male germline.

A complex interplay between H3K27me3 and DNAme has been reported in cultured cells as well as human primary cells, with these marks generally anticorrelated [Fang et al., 2016; King et al., 2016; Lu et al., 2016; Streubel et al., 2018]. H3K27me3 is also strongly anticorrelated with H3K36me3 in somatic cells [Yuan et al., 2011; Zheng et al., 2012] and with both DNAme and H3K36me3 in oocytes [Xu et al., 2019] (shown in Fig. 3b), indicating that either H3K36me3 or DNAme may inhibit the activity of EZH2/PRC2 or vice versa. While DNAme is globally reduced in the absence of Dnmt3l in oocytes, essentially no changes in H3K27me3 were detected [Xu et al., 2019], indicating that DNAme per se does not restrict the distribution of H3K27me3 in oocytes. In contrast, in the absence of Setd2, H3K27me3 encroaches into a subset of gene body regions that lose H3K36me3 [Xu et al., 2019]. Taken together with the studies mentioned above [Yuan et al., 2011; Zheng et al., 2012], as well as structural evidence [Jani et al., 2019; Finogenova et al., 2020], these data indicate that H3K36me3 plays a direct role in inhibiting PRC2 activity in vivo. Interestingly, not all genes show such encroachment of H3K27me3 into regions that are depleted of H3K36me3 in Setd2 KO oocytes [Xu et al., 2019]. While bdH3K4me3 is an obvious candidate, the chromatin features that restrict the invasion of H3K27­me3 into this specific set of gene bodies remain to be investigated.

While H3K27me3 is also excluded from H3K36me3 marked regions in E16.5 male germ cells, H3K27me3 does overlap with a subset of H3K36me2 marked regions at this stage (shown in Fig. 3b). Nevertheless, in the absence of Nsd1, H3K27me3 levels are increased specifically at H3K36me2-depleted regions, including intragenic regions of genes expressed at a relatively low level [Shirane et al., 2020]. Thus, H3K36me2 or DNAme may limit the accumulation of H3K27me3 in male germ cells. However, as in oocytes [Xu et al., 2019], the absence of Dnmt3l in male germ cells does not lead to redistribution of H3K27me3, revealing that NSD1/H3K36me2, rather than DNAme, is responsible for restricting either the recruitment of PRC2 to these loci, or its catalytic activity [Shirane et al., 2020]. Consistent with a role for NSD1 at H3K27me3 marked regions, biochemical analysis shows that this H3K36 KMTase associates with the PRC2 complex in mESCs [Streubel et al., 2018], though the subunit of PRC2 required for this interaction remains to be determined. Consistent with the apparent low level of expression of canonical NSD1, H3K36me2 and H3K27me3 are anticorrelated in oocytes, unlike in PSG. Taken together, these observations suggest that the distinct distribution of H3K27me3 observed in female and male germ cells may be orchestrated by the distinct distributions of H3K36me2 and H3K4me3 and that this interplay at least in part directs the divergent DNAme landscapes observed in these cells.

CGIs are genomic regions with high CpG-density and GC-content [Cooper et al., 1983; Bird et al., 1985]. While the vast majority of CGIs are marked with H3K4me3 and remain unmethylated, those devoid of this mark gain DNAme during the waves of de novo DNAme that occur in the germline and early embryo. As discussed above, the levels of DNAme are higher in sperm than oocytes (shown in Fig. 4a). Surprisingly, however, a greater number of annotated CGIs [Illingworth et al., 2010] are de novo methylated in oocytes (1,779) than sperm (461) (shown in Fig. 4b). While 13,318 out of 23,011 annotated CGIs are associated with promoter CGIs, only 439 and 31 are methylated in oocytes and sperm, respectively, indicating that promoter CGIs are particularly resistant to de novo DNAme in the male germline. In oocytes, methylated CGIs tend to be embedded in the bodies of actively transcribing genes [Smallwood et al., 2011; Kobayashi et al., 2012] or transcription units outside of annotated genes [Brind’Amour et al., 2018], including those emanating from LTRs, which are highly active in oocytes. This positive relationship between transcription and DNAme presumably reflects the timing of transcription, and in turn deposition of H3K36me3, and the onset of de novo DNAme within these transcription units, including of embedded CGIs. In contrast, as discussed above, the timing of de novo DNAme and transcription in male germ cells is not as well correlated as those in oocytes. Furthermore, transcripts emanating from noncanonical promoters are less abundant in the male than the female germline, perhaps due to the relatively high level of expression from solo LTR elements in the latter [Brind’Amour et al., 2018]. Alternatively, H3K4me3 at CGI promoters may be more robust in the male than the female germline, inhibiting de novo DNAme even within transcription units. Regardless, the greater propensity of CGIs to become de novo methylated in the female than the male germline may explain the paradox that gDMRs, which generally include CGIs, are more prevalent in oocytes than sperm (discussed in detail below), despite the fact the DNAme overall is much higher in the male than female germline.

Genomic imprinting is an epigenetic marking mechanism that maintains mono-allelic expression in somatic cells of specific genes from one or the other parental genome. DNAme, the classic mark at imprinted gDMRs, is established de novo in the gametes with distinct sets of genes imprinted in oocytes and sperm. Notably, imprinted gDMRs methylated in sperm are located at intergenic regions, whereas those methylated in oocytes are located within transcription units. Indeed, de novo DNAme of these regions in oocytes is tightly coupled with transcription. Evidence for a direct role for transcription in establishment of imprints in the female germline was reported by Chotalia et al. [2009] who showed that truncation of a transcript at the Gnas locus leads to loss of DNAme specifically at the gDMRs of this locus [Chotalia et al.,2009]. Similarly, deletion of the LTRs from which transcription-initiates upstream of the imprinted genes Impact and Slc38a4 results in the loss of such LTR-initiated transcripts (LITs) in oocytes and failure of the establishment of DNAme at these loci, including across gDMRs [Bogutz et al., 2019]. Furthermore, like other transcribed regions, maternally methylated imprinted gDMRs are enriched for H3K36me3 and DNAme is lost at these regions following deletion of Setd2 in oocytes [Xu et al., 2019]. Taken together, these observations indicate that SETD2 is broadly required for DNAme in the female germline, including at maternally imprinted gDMRs, most likely via RNA pol II-dependent SETD2 recruitment [Kizer et al., 2005].

The observation that de novo DNAme in oocytes is dependent on transcription raised the question of whether methylation of paternal gDMRs (H19, Dlk1-Gtl2, and Rasgrf1) in sperm is also dependent on transcription. In support of this model, previous studies revealed that low-level transcription at paternally methylated gDMRs (H19 and Dlk1-Gtl2) is correlated with DNAme at these regions [Henckel et al., 2012; Singh et al., 2013]. However, genetic ablation of SETD2 in male germ cells has essentially no impact on DNAme at these loci [Shirane et al., 2020]. In contrast, ablation of NSD1 expression in PGCs led to a reduction of DNAme at all three paternally methylated gDMRs, indicating that H3K36me2, rather than H3K36me3, directs de novo DNAme at imprinted gDMRs in the male germline. Notably, however, de novo DNAme at the 3′ flank of the Rasgrf1 gDMR is not dependent upon NSD1. Rather, this region requires the piRNA pathway and the recently discovered DNMT3 family member DNMT3C for the establishment of DNAme, as discussed in detail in the following section.

LTR retrotransposons, one of the two major groups of retrotransposons, represent ∼8% of the mouse genome [Mouse Genome Sequencing Consortium, 2002] and are silenced by DNAme and/or histone PTMs in many cell types [Senft and Macfarlan, 2021]. Intriguingly, a subset of these LTRs is expressed in the germline and shows sexually dimorphic regulation in mouse germ cells [Liu et al., 2014; Huang et al., 2021]. In oocytes, MaLR (most in mammalian apparent LTR retrotransposons) elements are the most highly expressed [Peaston et al., 2004]. Of these, members of the mouse transcript (MT) subfamily frequently act as alternative promoters for the expression of novel transcripts expressed exclusively in oocytes [Peaston et al., 2004; Brind’Amour et al., 2018]. While not all such LITs harbor open reading frames, some LITs indeed play an important role in driving expression of specific genes in oocytes. For example, an MTC-initiated transcript that initiates downstream from the canonical transcription start site of Dicer1 encodes a truncated DICER1 protein in oocytes that promotes small interference RNA production [Flemr et al., 2013]. In addition, the LITs which initiate upstream of the canonical transcription start sites of Impact and Slc38a4 are required for de novo DNAme of gDMRs in these genes [Bogutz et al., 2019] and in turn their imprinting in the offspring. Thus, insertion of a specific LTR into a locus or regions within or adjacent to regulatory elements may not only change the expression of neighboring genes in oocytes but also the DNAme pattern in cis, which can impact expression from the locus post-fertilization. While LITs have also recently been identified in human oocytes [Brind’Amour et al., 2018], their functional relevance in primates remains to be determined. Though LITs are also detected in meiotic spermatocytes in the mouse [Davis et al., 2017], they are not as prevalent as in oocytes and their function also remains unknown.

Importantly, many retrotransposons are transcribed in the early stages of male germline development, including following the onset of de novo DNAme. A subset of such transcripts is generated from piRNA clusters and in turn processed into piRNAs which guide de novo DNAme of young interspersed homologous retrotransposons and in turn their silencing. Several studies in male mice have revealed that de novo DNAme of evolutionarily young LINE1 elements is dependent on a specialized piRNA biogenesis and effector pathway that includes PIWIL4, MitoPLD, TDRD family proteins, as well as the germline-specific PIWI protein MIWI2 [Aravin et al., 2007; Carmell et al., 2007; Aravin et al., 2008; Kuramochi-Miyagawa et al., 2008] and associated factors [Ma et al., 2009; Xiol et al., 2012; Saxe et al., 2013; Wenda et al., 2017; Schopp et al., 2020; Yang et al., 2020]. More recently, an analysis of the interactome of MIWI2 in male germ cells indicated that DNMT3A and DNMT3L as well as DNMT3C may be recruited to regions targeted by the MIWI2-piRNA complex via an interaction with SPOCD1, an uncharacterized protein not previously implicated in piRNA biogenesis or function [Zoch et al., 2020]. Indeed, deletion of SPOCD1 in male mice leads to hypomethylation and derepression of evolutionarily young TEs [Zoch et al., 2020]. Further supporting a role for piRNA-directed de novo DNAme in male germ cells, a noncoding RNA spanning a young retrotransposon embedded in the gDMR of Rasgrf1, one of the three gDMRs methylated in male germ cells, is targeted by piRNAs produced from a piRNA cluster on another chromosome, leading to de novo DNAme of the 3′ flanking region of the Rasgrf1 gDMR [Watanabe et al., 2011]. While the role of histone PTMs in this process has not been studied in detail, LINE elements targeted for DNAme by the piRNA pathway are not marked by H3K9me3 in PSG at E16.5, but rather by H3K4me3, consistent with their transcriptional activity earlier in male germ cell development. Intriguingly, the interactome study mentioned above also identified SPINDLIN, an H3K4me3 binding protein [Wang et al., 2011], as another MIWI2 interacting factor [Zoch et al., 2020]. Though the relevance of this interaction was not addressed, given that DNMT3C lacks a PWWP domain, it is tempting to speculate that the combination of H3K4me3 and piRNA sequence complementarity guides pi­RNA/DNMT3C-dependent de novo DNAme in the male germline (shown in Fig. 5) [Dura et al., 2022]. If so, this would effectively act as a secondary repressive pathway for silencing of TEs that evade KRAB-ZFP/H3K9me3-dependent silencing.

While piRNAs are clearly produced in oocytes, whether they play any role in guiding de novo DNAme remains unknown. Regardless, unlike the case for males, female mice deficient in factors essential for piRNA biogenesis, including PIWI proteins, are fertile [Carmell et al., 2007; Watanabe et al., 2008; Shoji et al., 2009]. Furthermore, though Mili KO oocytes show derepression of a subset of LTRs [Kabayama et al., 2017], the levels and prevalence of such transcripts are relatively low compared to Mili KO male germ cells, suggesting that the role of piRNAs differ between female and male germ cells. The lower sensitivity of female germ cells to disruption of the piRNA pathway may be explained by the presence of alternative posttranscriptional silencing pathways, such as small interference RNAs, which are also known to play a role in suppressing retrotransposons in the female germline [Watanabe et al., 2008].

As discussed above, DNMT3L plays an important role in de novo DNAme in both female and male germ cells, consistent with structural evidence for a DNMT3A/DNMT3L heterotetramer [Jia et al., 2007]. However, while essentially all de novo DNAme is disrupted in Dnmt3l KO oocytes [Shirane et al., 2013], substantial levels of residual DNAme are detected in DNMT3L deficient male germ cells at P10 [Barau et al., 2016] (shown in Fig. 6a). It is possible that male germ cells express higher levels of DNMT3A than female germ cells and in turn that DNMT3A, potentially acting as a homotetramer, is sufficient to catalyze DNAme independent of DNMT3L. Alternatively, another cofactor that stimulates DNMT3A activity may be expressed exclusively in male germ cells. Intriguingly, recent biochemical analyses reveal that DNMT3B isoforms, including the catalytically inactive DNMT3B3, enhance the activity of DNMT3A in a manner similar to that of DNMT3L [Van Emburgh and Robertson, 2011; Gordon et al., 2013; Zeng et al., 2020]. Mining of published RNA-seq data [Veselovska et al., 2015; Sangrithi et al., 2017] reveals that Dnmt3b3 is expressed exclusively in male germ cells when de novo DNAme occurs (shown in Fig. 6b). Whether this DNMT3B isoform plays a role in stimulating the activity of DNMT3A during postnatal male germ cell development is worthy of further study. Regardless, such DNMT3L-independent activity is clearly not sufficient to rescue the infertility phenotype in Dnmt3l KO males [Bourc’his et al., 2001], indicating that persistent hypomethylation of the regions that are dependent on DNMT3L and DNMT3C for their de novo DNAme, likely young TEs, is responsible for the observed “meiotic catastrophe” [Molaro et al., 2020]. As DNMT3C is unique to rodents, the expression profiles of DNMT3 isoforms should clearly be taken into consideration when studying the roles of the de novo methyltransferases in other species.

Notably, DNAme established in the gametes at both paternally and maternally methylated imprinted gDMRs are resistant to the global wave of DNA demethylation in the early embryo and in turn transmitted to all the somatic cells in adults, thus maintaining the mono-allelic expression of associated genes [Bartolomei and Ferguson-Smith, 2011]. The persistence of DNAme at such regions is dependent upon the DNA binding factors ZFP57 and ZFP445, which bind to a specific sequence motif only when the CpG embedded in this motif is methylated [Tucci et al., 2019]. Subsequent recruitment of KAP1 and in turn SETDB1 leads to H3K9me3 deposition in these regions, likely promoting efficient maintenance DNAme [Quenneville et al., 2011, 2012] in the early embryo and somatic tissues via UHRF1 and/or DNMT1 binding [Karagianni et al., 2008; Bartke et al., 2010; Nady et al., 2011; Arita et al., 2012; Rothbart et al., 2012; Xie et al., 2012; Cheng et al., 2013; Rothbart et al., 2013; Tauber et al., 2020], as discussed above for retrotransposons in PGCs.

While male mice deficient in Dnmt3a, Dnmt3c, or Dnmt3l are infertile [Bourc’his et al., 2001], female mice deficient in Dnmt3a or Dnmt3l can produce mature oocytes that support early embryonic development [Bourc’his et al., 2001; Hata et al., 2002; Kaneda et al., 2004]. However, the embryos derived from maternal KO of Dnmt3a or Dnmt3l die by E10.5, due in part to the biallelic expression of imprinted genes. Thus, while DNAme established during early male germ cell development is essential for spermatogenesis, DNAme established in the female germline is essential only after fertilization, driven by the influence of this epigenetic mark on gene expression during early embryonic development.

What role might the distinct patterns of histone PTMs established in the gametes play on the transcriptome after fertilization? As discussed above, bdH3K4me3 observed in oocytes are transmitted to the 2-cell embryos and regulate ZGA [Dahl et al., 2016; Liu et al., 2016; Zhang et al., 2016]. While a large fraction of histones is replaced by protamines during spermatogenesis, a subset is clearly retained, in particular in promoter regions of developmental genes [Erkek et al., 2013; Yamaguchi et al., 2018; Yoshida et al., 2018]. While one study suggested that H3K4me3 at CpG-rich promoters in sperm is not transmitted to the embryo [Zhang et al., 2016], more recent studies indicate that H3K4me3 persists on the paternal allele through early cell divisions in the embryo [Lismer et al., 2020, 2021] and aberrant deposition of this mark in sperm was reported to impact gene expression in the embryo.

While H3K36me2/me3 clearly play an instructive role in de novo DNAme in germ cells, the role of inheritance of H3K36 methylation on transcriptional regulation following fertilization is less clear than that of DNAme at imprinted gDMRs. A previous study using immunostaining for H3K36me3 showed that while this mark is enriched in the zygote, specifically in the maternal pronucleus, it is lost in cleavage stage embryos [Boskovic et al., 2012]. Consistent with this observation, allelic profiling of H3K36me3 in early embryos by ChIP-seq revealed that oocyte-derived H3K36me3 is inherited in zygotes but lost by the 8-cell stage [Xu et al., 2019]. Establishment of H3K36me3 in a subset of gene bodies gradually becomes apparent over both alleles in 2-cell stage embryos, presumably reflecting nascent transcription from both parental genomes in association with ZGA [Xu et al., 2019]. Notably, Setd2-deficient oocytes show cytosolic defects, resulting in developmental arrest of the Setd2 maternally deficient embryo at the zygote stage [Xu et al., 2019]. Furthermore, overexpression of an amino-acid substitution mutant of H3.3 at K36 (H3.3K36M) in oocytes, which abrogates methylation of H3.3K36, leads to growth retardation of the resulting embryos [Aoshima et al., 2015], indicating that methylation of H3K36 in oocytes may regulate preimplantation development. Indeed, a recent study employing CRE-mediated expression of H3.3K36M during mouse oocyte growth revealed a global loss of H3K36me2 in oocytes and embryonic lethality at the implantation stage [Yano et al., 2022]. However, it remains unclear how H3K36me2 and H3K36me3 established in oocytes regulates embryogenesis.

Recent analysis of H3K27me3 in oocytes and preimplantation embryos revealed that this mark persists through preimplantation at a subset of regions that are depleted of DNAme [Inoue et al., 2017a]. This oocyte-inherited H3K27me3 regulates the mono-allelic expression of a subset of genes expressed exclusively early in embryonic development, which is distinct from “canonical” imprinted genes whose mono-allelic expression is maintained in all somatic tissues. Interestingly, while many of these “non-canonical” imprinted genes lose H3K27me3 in later embryonic lineages, this histone PTM is replaced by DNAme specifically in extraembryonic tissues and mono-allelic expression of associated genes is maintained exclusively in this lineage [Chen et al., 2019; Hanna, 2020; Hanna and Kelsey, 2021]. The establishment of DNAme at these loci apparently depends upon H3K27me3 deposition in oocytes [Chen et al., 2019]. A subset of these noncanonical imprinted genes are clearly associated with placental development [Itoh et al., 2000; Inoue et al., 2018; Prokopuk et al., 2018; Matoba et al., 2019; Inoue et al., 2020]. Similarly, the Xist locus is decorated by broad H3K27me3 and depleted of DNAme in oocytes. This broad H3K27me3 is maintained through fertilization and apparently regulates the expression of Xist RNA in preimplantation embryos, ensuring imprinted X-inactivation [Inoue et al., 2017b]. Genetic ablation of Ezh2 in growing oocytes or Eed in primary oocytes leads to growth retardation or growth arrest in the progeny, respectively [Erhardt et al., 2003; Inoue et al., 2018]. Strikingly, a recent study shows that the growth retarded phenotype of Eed maternal KO embryos can be restored by the combined deletion of both Eed and Xist [Matoba et al., 2022]. Furthermore, this study clearly shows that while Xist imprinting regulates embryonic development, autosomal noncanonical imprinting is required for normal placental development [Matoba et al., 2022]. Taken together, these observations indicate that the establishment of H3K27me3 in the oocyte plays an essential role in transcriptional regulation in both the embryo and in extraembryonic tissues.

H3K9me3, another heterochromatic mark, is apparently essential for preimplantation development, as deletion of Setdb1 in oocytes leads to a reduction of H3K9me3 and developmental arrests in preimplantation stages [Eymery et al., 2016], likely in association with aberrant upregulation of genes that impact the meiotic program, such as Cdc14b [Kim et al., 2016]. Thus, gamete-derived DNAme as well as histone PTMs, in particular those that are deposited on the maternally inherited genome, are essential for embryonic development.

In contrast to the maternal genome, the paternal genome is dramatically depleted of canonical histones during spermatogenesis. Mature mouse sperm retain only ∼1% of the level of histone H3 detected in somatic cells, with H3K4me3-enriched and bivalent CGI promoters showing preferential retention of nucleosomes [Erkek et al., 2013; Yamaguchi et al., 2018; Yoshida et al., 2018]. Furthermore, residual H3K27me3 and H3K36me3 are reduced or lost from the paternal genome shortly after fertilization in mouse zygotes [Zheng et al., 2016; Wang et al., 2018; Xu et al., 2019]. The role of paternal inheritance of histone PTMs outside of promoter regions in intergenerational inheritance of chromatin states is therefore likely modest relative to maternally inherited histone PTMs. While H3K9me3 is also reduced or lost upon fertilization from both parental genomes, the low levels of H3K9me3 that are retained at specific loci may play a role in transcriptional silencing, in particular of TEs [Xu et al., 2022; Yu et al., 2022]. Regardless, a large fraction of LTR elements regain this mark following ZGA in the 2-cell stage embryo [Wang et al., 2018], likely guided by KRAB-ZFPs, as in somatic tissues. Whether other histone PTMs in mature gametes are transmitted to preimplantation embryos, and their impact on transcription, remains to be determined.

Genome-wide mapping of histone PTMs and DNAme in developing germ cells and embryos of mice deficient in the enzymes that deposit them have revealed a complex interplay between these chromatin marks during mammalian development. Furthermore, allelic profiling using hybrid mouse strains have allowed for genome-wide tracing of the fate of these epigenetic marks in preimplantation embryos and their impact on transcription. However, these observations are generally derived from “snapshots” of chromatin states derived from analyses of cells analyzed in bulk. It will be particularly interesting to profile the epigenomes and transcriptomes of germ cells and in early embryos with higher resolution and spatiotemporal information [Takei et al., 2021] and/or single-cell multi-omics [Shema et al., 2019] technologies. Such analyses will provide further insight into the heterogeneity of cellular epigenomic states and their impact on the transcriptome. For example, a recent integrative analysis of DNAme, histone PTMs and single-cell transcriptome in the developing mouse testis identified distinct epigenetic patterning of spermatogonial stem cell populations undergoing self-renewal versus differentiation [Cheng et al., 2020].

KO strategies have been used to investigate the consequence of disruption of chromatin modifying enzymes on their target histone PTMs as well as other histone PTMs that may be impacted. Clearly, attention should be paid to potential secondary effects of disrupting these enzymes, such as the induction of alternative cell fates, that may not reflect direct positive or negative interactions between these marks. Perturbation of such interactions at a specific developmental time point or at specific loci using recently developed CRISPR/Cas9-based approaches [Nakamura et al., 2021], may provide further insights into the hierarchical relationships between histone PTMs, DNAme, and transcription.

Investigation of the roles of many chromatin modifying enzymes in germ cells has been facilitated by analyses of conventional or germline-specific KO mice, combined with low-input genome-wide profiling methods that have been developed in recent years. However, such in vivo studies are time consuming and cost-prohibitive for systematic analyses. The application of recently developed in vitro differentiation systems that mimic in vivo PGC specification/differentiation [Hayashi et al., 2011; Ohta et al., 2017], female-sex determination of germ cells [Miyauchi et al., 2017], oogenesis [Hayashi et al., 2012; Hikabe et al., 2016; Hamazaki et al., 2021], or spermatogenesis [Ishikura et al., 2016; Ishikura et al, 2021] has the potential to circumvent the inherent limitations of such in vivo analysis, allowing for systematic gene KO or overexpression analyses at specific stages in germline development as well as the application of genetic screens, such as those based on CRISPR-mediated deletion [Schuster et al., 2019]. Such studies will further enhance our understanding of the underlying mechanism and order of establishment of chromatin marks and the impact of such sexually dimorphic epigenetic patterning on transcription and other nuclear processes in germ cells, the early embryo and beyond.

We thank A. Bogutz, K. Mochizuki, and S. Janssen for critical reading of the manuscript.

The authors have no conflicts of interest to declare.

This work was supported by grants from a JSPS Grant-in-Aid for Early-Career Scientists (JP22K15125) and Takeda Science Foundation to K.S.; Canadian Institutes of Health Research Project Grant PJT-166170 to M.L.

The authors contributed equally to all aspects of the article.