Dual role of Ovol2 on the germ cell lineage segregation during gastrulation in mouse embryogenesis

In mice, primordial germ cells (PGCs) are specified in the posterior epiblasts at around embryonic day (E) 6.0 in response to BMP4 and WNT3 (Lawson et al., 1999; Liu et al., 1999; Ohinata et al., 2009). During this period, PGCs start to express a specific set of transcription factors that repress the somatic cell program, re-acquire latent pluripotency, trigger epigenetic reprogramming and thereby determine the PGC fate. The mechanisms underlying the PGC-specification process, including key signaling pathways, such as BMP and WNT, and transcription factors, such as BLIMP1 (PRDM1) and TFAP2C, are largely conserved among mammalian species, including, mice, rabbits, pigs and primates (Ohinata et al., 2005; Kobayashi et al., 2017, 2021; Kojima et al., 2017). Notably and without exception, PGCs form at the posterior part of the epiblast in these animals, as a set of cytokines initiates another key event in embryogenesis: gastrulation. Therefore, PGC specification essentially entails gastrulation in mammalian embryogenesis.

During gastrulation, epiblast cells undergo epithelial-to-mesenchymal transition (EMT) along the primitive streak followed by bilateral ingression underneath the epiblast layer. During this process, epiblast cells swiftly lose expression of pluripotency-associated genes and then differentiate into the primary germ layers (Nakamura et al., 2016; Hamidi et al., 2020). Considering re-acquisition of latent pluripotency with constant expression of pluripotency-associated genes in PGCs, mutually exclusive programs are executed in a group of cells in the posterior epiblast under the influence of BMP4 and WNT3. However, how these cells are properly sorted into PGCs or somatic cells remains an enigma. Recent findings have shown that both PGC and somatic cell precursors express genes involved in nascent mesoderm differentiation. For example, T (brachyury) is expressed in the posterior epiblast prior to gastrulation and is required for the formation of both PGCs and mesoderm (Aramaki et al., 2013, 2021). These findings indicate that PGC specification and gastrulation are initiated by an identical transcription network, and then regulated by the balance of transcription factors.

A plausible approach to the mechanism of the fate determination would be to focus on whether or not epiblast cells undergo EMT, as EMT-related genes such as Snai1 and Vim are repressed in PGCs (Kurimoto et al., 2008); vice versa, Cdh1 (E-cadherin), a key molecule maintaining the epithelium, is required for PGC differentiation (Okamura et al., 2003). Based on previous studies, genes such as Snai1, Snai2, Zeb1, Zeb2, Twist1 and Twist2, have been well characterized as key transcription factors promoting EMT in various cellular contexts (Peinado et al., 2007; De Craene and Berx, 2013; Dongre and Weinberg, 2019). However, genes such as Elf5, Gata3, Grhl2, Pou5f1, Klf4 and Ovol2 have been identified as transcription factors that counteract EMT (Li et al., 2010; Chakrabarti et al., 2012; Cieply et al., 2012; Watanabe et al., 2014; Takaku et al., 2016; Jägle et al., 2017). Among them, Ovol2 might have an important role in PGC specification, as evidenced by the fact that Ovol2-deficient embryos exhibit a reduced number of PGCs (Hayashi et al., 2017).

Ovo genes encode C2H2 zinc-finger transcription factors and are widely conserved among flies, nematodes, mice and humans (Mevel-Ninio et al., 1991; Chidambaram et al., 1997; Dai et al., 1998; Lü et al., 1998). The Drosophila ovo gene plays pivotal roles in the development and sex determination of germ cells (Oliver et al., 1987, 1990, 1994; Mevel-Ninio et al., 1991; Lü et al., 1998; Andrews et al., 2000). In mice, the Ovol gene family comprises Ovol1, Ovol2 and Ovol3, and, furthermore, the Ovol2 gene locus generates at least three splicing variants: Ovol2a, Ovol2b and Ovol2c. Based on the findings of the Drosophila ovo gene, the splicing variants have a unique role in germ cell and epidermal cell development in a highly cell context-dependent manner (Payre et al., 1999; Andrews et al., 2000). These findings imply that mouse Ovol genes would have a unique role in PGC specification. Thus, in this study, we investigate the functional involvement of mouse Ovol genes in the lineage segregation between PGCs and somatic cells during gastrulation.

To clarify mechanistic insights into Ovol2, we first investigated the number of cells expressing nascent PGC markers, such as Blimp1 (also known as Prdm1) and Tfap2c, in Ovol2 mutant embryos (Ovol2^−/−) during gastrulation. We crossed Ovol2^+/− males harboring the Blimp1-mVenus (BV) reporter gene (Ohinata et al., 2008) with Ovol2^+/− females. Immunofluorescence analysis showed that the differentiation of BV-positive PGCs was severely disturbed in Ovol2^−/− embryos at E7.5 (Fig. 1A). Compared with wild-type embryos holding a cluster of PGCs expressing both BV and TFAP2C, Ovol2^−/− embryos showed sparse formation of PGCs with weak levels of BV and TFAP2C expression (Fig. 1B). The number of BV-positive cells was reduced in Ovol2^−/− embryos around gastrulation (Fig. 1C), a far earlier stage than the somite stage reported previously (Hayashi et al., 2017). These results demonstrated that PGC specification is hampered in the Ovol2^−/− embryos at the gastrulation stage.

As the number of nascent PGCs is reduced in Ovol2^−/− gastrulating embryos, we employed an in vitro differentiation system, in which PGC specification is faithfully reproduced in a series of differentiation stages from embryonic stem cells (ES cells) to PGC-like cells (PGCLCs) via epiblast-like cells (EpiLCs) (Hayashi et al., 2011). Taking advantage of this culture system, we assessed the individual involvement of splicing variants of Ovol2 (Ovol2a, Ovol2b and Ovol2c) and other Ovol family genes (Ovol1 and Ovol3) in PGCLC differentiation. Quantitative PCR (Q-PCR) analysis revealed that Ovol1 expression was continuously increased from ES cells to PGCLCs, and Ovol2a and Ovol2b expression reached a peak in EpiLCs and PGCLCs at 3 days of culture, respectively; Ovol2c expression was barely detectable and Ovol3 expression was constant throughout the differentiation process (Fig. 1D). We interrogated expression patterns of Ovol family genes in epiblast cells from E4.5 to E6.5 in the single-cell RNA sequence (RNA-seq) dataset provided by SC3-seq (Fig. S1A) (Nakamura et al., 2015, 2016). Considering that ES cells and EpiLCs correspond to E4.5 and E5.5 epiblast cells, respectively, and E6.5 epiblast contains early PGCs, Ovol1 and Ovol2 expression was largely consistent between in vitro and their in vivo counterparts. However, Ovol3 expression was not detectable in the dataset, possibly owing to the 3′UTR of Ovol3 being indistinguishable from that of Polr2i (91 bp are completely matched).

Next, we induced PGCLCs from individual knockout (KO) (Ovol1^−/−, Ovol2^−/− or Ovol3^−/−) and triple KO (TKO) ES cells harboring the BV reporter (Fig. S1B). At day 2 of PGCLC induction, BV expression in Ovol2^−/− and TKO aggregates was clearly weaker than that in wild type, whereas BV expression in Ovol1^−/− and Ovol3^−/− appeared to be comparable with that in wild type (Fig. 1E). Quantification of the fluorescence intensity showed that the levels of the BV intensities in Ovol2^−/− and TKO aggregates were significantly reduced (Fig. 1F). Q-PCR analysis of Ovol1^−/−, Ovol2^−/−, Ovol3^−/− and TKO showed that endogenous Blimp1 expression was downregulated not only in Ovol2^−/− and TKO but also in Ovol3^−/− (Fig. 1G). Of note, we observed a distinctive expression pattern for Tfap2c, a functional marker of PGC specification (Weber et al., 2010), between Ovol2^−/− or TKO and Ovol3^−/−. To further characterize BV-positive cells in the KO aggregates, we examined expression of BV and TFAP2C by immunofluorescence analysis. The number of cells expressing both BV and TFAP2C at high levels was significantly decreased in Ovol2^−/−, Ovol3^−/− and TKO (Fig. 1H,I). In contrast, Ovol3^−/− aggregates contained larger percentages of TFAP2C-high- and BV-low-expressing cells (Fig. S1C), which might be responsible for the abundance of transcripts of Tfap2c in Ovol3^−/− detected by Q-PCR (Fig. 1G). This unique phenotype is not consistent with the PGC differentiation process in vivo, in which the expression of BV and TFAP2C are correlated (Fig. 1B), indicating a distinctive role for Ovol3 in PGC specification. At day 4 of PGCLC induction, the number of BV-positive cells in Ovol2^−/− and TKO aggregates decreased and the number in Ovol3^−/− was slightly restored (Fig. S1D,E). The number of BV and TFAP2C double-positive cells decreased in Ovol2^−/− and TKO aggregates (Fig. S1F,G). Importantly, the extent of the decrease in the number of BV and TFAP2C double-positive cells in Ovol2^−/− aggregates was similar to that in Ovol2^−/− embryos (Fig. 1B, Fig. S1G) (Hayashi et al., 2017). These results demonstrate that Ovol2 has a dominant role in PGC specification among the Ovol family genes.

To identify the molecular pathway downstream of each Ovol gene, we further investigated the transcriptomes of Ovol1^−/−, Ovol2^−/−, Ovol3^−/− and TKO. Principal component analysis (PCA) showed that the effect of the gene disruption was subtle in ES cells and EpiLCs, but it appeared clearly after PGCLC induction (Fig. 2A). Furthermore, PCA using only PGCLCs showed that these transcriptomes could be divided into three groups: wild type and Ovol1^−/−, Ovol2^−/− and TKO, and Ovol3^−/− (Fig. 2B). Analysis of genes reflecting on each group showed that genes expressed in nascent mesoderm but downregulated in PGCs, such as Hand1, Vim and Bmp4 (Kurimoto et al., 2008), were enriched in the direction to the group of Ovol2^−/− and TKO (Fig. 2C). In contrast, genes involved in both PGC and nascent mesoderm differentiation, such as T, Eomes and Wnt3 (Aramaki et al., 2013; Senft et al., 2019), were enriched in the direction of wild type and Ovol1^−/−. In addition, unexpectedly, pluripotency-associated genes, such as Nanog, Dppa3 and Zfp42, were enriched in the direction of Ovol3^−/−. It is noteworthy that EMT-related genes, such as Zeb1, Snai1 and Snai2, were enriched in the group of Ovol2^−/− and TKO, whereas genes representing the maintenance of cellular polarity, such as Cdh1, Cldn6 and Cldn7, were clearly enriched in the opposite direction. These expression profiles suggested that Ovol2^−/− and TKO accelerate mesoderm differentiation through the progression of EMT.

Consistent with the PCA results, differentially expressed genes (DEGs) compared with wild type were largely overlapped between Ovol2^−/− and TKO (Fig. 2D), whereas they were scarcely overlapped between TKO and either Ovol1^−/− or Ovol3^−/− (Fig. S2A). Gene ontology (GO) analysis of 198 commonly upregulated DEGs in both Ovol2^−/− and TKO exhibited terms such as ‘cell differentiation’ and ‘positive regulation of EMT’. On the other hand, 173 commonly downregulated DEGs belonged to ‘stem cell population maintenance’ and ‘cell adhesion’. Considering the involvement of OVOL2 in the inhibition of EMT (Lee et al., 2014; Watanabe et al., 2014; Wu et al., 2017), these results indicate that the defective PGCLC differentiation in Ovol2^−/− and TKO was attributable to aberrant regulation of EMT and/or maintenance of pluripotency. This idea was supported by RNA-seq analyses showing downregulation of pluripotency-associated genes and epithelial adhesion genes, and upregulation of EMT-related genes (Fig. 2E, Fig. S2B). The downregulation of Cdh1 in Ovol2^−/− and TKO was confirmed by immunofluorescence analysis (Fig. S2C). These results indicated that a key molecular pathway involving Ovol2 during PGC specification is maintenance of cell adhesion, which is lost during EMT entailing mesoderm induction.

Next, we further characterized the function of Ovol2 variants in PGC specification by enforced expression of Ovol2a and/or Ovol2b in TKO (Fig. S2D). With enforced expression of the variants, BV signals were restored in all Ovol2a-transgenic (Tg), Ovol2b-Tg and Ovol2a/2b-Tg aggregates, compared with the parental TKO aggregates (Fig. 2F). FACS analyses showed that the level of BV expression in Ovol2a/2b-Tg aggregates was higher than that in Ovol2a- or Ovol2b-Tg aggregates, suggesting a synergistic effect of these variants. RNA-seq analysis followed by PCA revealed that aggregates of Ovol2a-, Ovol2b- and Ovol2a/2b-Tg at day 2 of culture became closer, albeit not identical, to wild type (Fig. 2G). Of note, the similarity of gene expression profiles of Ovol2a-Tg aggregates and wild type was greater than that between the profiles of Ovol2b-Tg aggregates and wild type. In Ovol2a-, Ovol2b- and Ovol2a/2b-Tg aggregates at day 2 of culture, the expression of Tfap2c was significantly restored and so were pluripotency-associated genes, such as Pou5f1, Nanog and Sox2 (Fig. 2H). Expression of some EMT-related genes, such as Snai1 and Zeb1, was partially repressed in these transgenic aggregates, and the extent of this repression was slightly larger in Ovol2a-Tg than in Ovol2b-Tg. Consistent with the repression of the EMT-related gene expression, expression of the epithelial adhesion genes Cdh1 and Epcam was partially but significantly restored in these transgenic aggregates (Fig. 2H). These results indicate that Ovol2a and Ovol2b promote PGC specification by preventing EMT and promoting pluripotency.

CDH1-mediated cell-cell interaction is essential for PGC formation at around E7 (Okamura et al., 2003). Consistently, the level of CDH1 was correlated with BV protein expression in aggregates at day 2 of PGCLC induction (Fig. S2C). Thus, we examined whether a scarcity of CDH1 in Ovol2^−/− and TKO is the main cause of the defective PGCLC differentiation by enforced expression of Cdh1 in TKO ES cells (Cdh1-Tg) followed by PGCLC induction (Fig. S3A). In Cdh1-Tg aggregates at day 2 of PGCLC induction, BV expression was restored to a level comparable with that in the wild-type aggregates (Fig. 3A). However, despite BV expression, Cdh1-Tg aggregates had a closely similar transcriptome to TKO aggregates: the difference in number of DEGs between Cdh1-Tg and TKO (134 genes) was much smaller than that between Cdh1-Tg and wild type (587 genes) (Fig. 3B). PGC-related genes, such as Blimp1 and Tfap2c, were restored in Cdh1-Tg, whereas pluripotency-associated genes and EMT-related genes were not nominated as DEGs (log₂ fold change >1, FDR <0.001) (Fig. 3C). Indeed, compared with the PGC-related genes, the changes in expression of pluripotency-associated genes (Pou5f1, Nanog and Sox2) and EMT-related genes (Snai1, Snai2 and Zeb1) were subtle (Fig. 3D). PCA confirmed that genes whose expression was altered in Cdh1-Tg were Cdh1 itself and genes that are irrelevant to PGC specification (Fig. 3E, Fig. S3B). These results suggest that enforced expression of Cdh1 partially restores PGCLC induction in TKO through upregulation of the PGC-related genes. On the other hand, transcriptional regulation of pluripotency-associated genes and EMT-related genes was independent of Cdh1 expression, therefore indicating that the Ovol2-mediated gene regulatory network is required for PGC(LC) induction in a Cdh1-independent manner.

To explore the genome-wide targets for OVOL2A and OVOL2B, we performed chromatin immunoprecipitation sequencing (ChIP-seq). For this purpose, we used Ovol2a-Tg and Ovol2b-Tg aggregates at day 2 of PGCLC induction and antibodies against FLAG-tag fused to exogenous OVOL2A and OVOL2B. ChIP-seq analyses using biologically duplicated samples detected 1215 and 5157 peaks as candidates for OVOL2A- and OVOL2B-binding sites, respectively (Fig. 4A). Nearly all of the peaks of OVOL2A-binding sites were overlapped with those for OVOL2B, except in the case of five genes specific to OVOL2A (Fig. 4A,B). De novo motif-finding analysis identified a binding consensus sequence (CCGYTA) of both OVOL2A and OVOL2B (Fig. 4C), which is consistent with the fact that these variants harbor the identical zinc-finger domain. These consensus sequences were almost identical to a known OVOL1/2 binding sequence (CCGTTA) (Nair et al., 2007; Watanabe et al., 2014).

We tested whether these binding peaks are associated with gene expression dynamics. By referring to the published datasets (Kurimoto et al., 2015), we analyzed histone marks, including trimethylated histone H3 lysine 4 (H3K4me3), trimethylated histone H3 lysine 27 (H3K27me3) and acetylated histone H3 lysine 27 (H3K27ac) around these peaks for OVOL2A and OVOL2B in day 2 aggregates. Genomic regions highly mapped as OVOL2A- and OVOL2B-binding sites were correlated with enrichment of H3K4me3, and H3K27ac was slightly enriched in these regions (Fig. 4D). However, H3K27me3 was enriched in genomic regions flanking OVOL2A- and OVOL2B-binding sites that were moderately mapped. Next, we assigned genes harboring a OVOL2A- or OVOL2B-binding peak within a 50 kb region flanking the longest transcripts detected, and then interrogated their expression dynamics upon enforced expression of Ovol2a or Ovol2b. Unsupervised hierarchical clustering (UHC) of transcriptionally altered OVOL2A- or OVOL2B-binding genes revealed that these genes were classified into four (cluster 1A-4A) or five (cluster 1B-5B) clusters in the comparison between wild type and either Ovol2a-Tg or Ovol2b-Tg, respectively (Fig. 4E). Genes consistently up- or downregulated in both the wild type and Ovol2a-Tg were enriched in cluster 2A or cluster 4A, respectively; cluster 2A included WNT-related genes, such as Wnt3 and Sp5, and cluster 4A included EMT-related genes, such as Zeb1 and Zeb2 (Fig. 4E). Genes consistently up- or downregulated in both wild type and Ovol2b-Tg were enriched in cluster 1B or cluster 4B, respectively. Of note, cluster 1B included exclusively PGC-related and pluripotency-associated genes, such as Blimp1, T and Nanog. The Zeb1 locus showed a peak of OVOL2A and OVOL2B with bivalent histone modification at the transcription start site (TSS), and Blimp1, Nanog and T loci showed peaks of OVOL2B with H3K27ac at the region upstream of each TSS (Fig. 4F, Fig. S4A). Among genes harboring a OVOL2A- or OVOL2B-binding peak within the 0.5 kb region flanking the TSS, 236 OVOL2A-binding genes and 656 OVOL2B-binding genes were associated with both H3K4me3 and H3K27me3, so-called bivalent histone modification (Fig. S4B). Of the 236 and 656 genes, 119 and 303 genes, respectively, overlapped with the genes with the bivalent histone modification in PGCLCs at day 2 of induction (Fig. S4B) (Kurimoto et al., 2015), indicating that OVOL2A and/or OVOL2B play a role in the establishment of the PGC-specific epigenetic landscape.

To confirm the functional involvement of OVOL2A and OVOL2B in transactivation of these genes, we performed luciferase reporter analyses using genomic fragments upstream of each gene. Transcriptional activity of the luciferase gene with the ∼1000 bp upstream of Zeb1 was clearly repressed by expression of OVOL2A in HEK293T cells, whereas it was not repressed by expression of OVOL2B (Fig. 4G, Fig. S4C). These results demonstrate that OVOL2A but not OVOL2B directly represses Zeb1. Next, we examined the transcriptional effects of genomic fragments upstream of Blimp1, T and Nanog. To stringently assess the transcriptional activity in an appropriate cell context, each luciferase reporter construct was integrated into the genome of Ovol2b-Tg (Fig. S4C). Using these Ovol2b-Tg lines that harbor a similar copy number of the reporter construct (Fig. S4D), luciferase activities were determined at day 1 of PGCLC induction with or without doxycycline (Dox), which transactivates the exogenous Ovol2b gene. In Ovol2b-Tg with Blimp1-luc, luciferase activity was elevated in a presence of Dox, whereas other genomic constructs did not enhance transactivation (Fig. 4H). The upregulation of Blimp1-luc was OVOL2 dependent, as deletion of the OVOL2B-binding element (OBE) nullified the effect of transactivation (Fig. S4E). In contrast to the similar and higher levels of endogenous expression of Nanog and T, respectively, compared with that of Blimp1, in TKO (Figs 1G and 2E), the basal levels of luciferase activity of T-luc and Nanog-luc without Dox were much lower than that of Blimp1-luc. This could be due to the genomic region in the construct being insufficient for the transactivation, raising the possibility that these genes are regulated by OVOL2B. Nevertheless, this analysis demonstrates that transcription of Blimp1 is promoted by OVOL2B through direct binding to its enhancer region.

As Zeb1 was identified as a direct target of OVOL2A during PGC specification, we investigated the functional consequence of repression of Zeb1. For this purpose, we deleted Zeb1 in TKO ES cells and then induced PGCLCs (Fig. S5A). Surprisingly, BV expression was almost restored in Zeb1^−/− aggregates (Fig. 5A). In sharp contrast, deletion of other EMT-related genes, such as Snai1 and Snai2, did not restore BV expression in TKO (Fig. 5A, Fig. S5A), demonstrating that Zeb1 plays a unique role in counteracting PGC specification. Q-PCR analysis revealed that not only BV expression but also the expression of PGC-related genes (Blimp1, Prdm14 and Tfap2c) and pluripotency-associated genes (Pou5f1, Nanog and Sox2) were restored in Zeb1^−/− cells to levels comparable with those in wild type (Fig. 5B). In addition, a substantial restoration of Cdh1 expression was observed in Zeb1^−/− but not in Snai1^−/− and Snai2^−/− cells (Fig. S5B). Compared with the PGC-related genes, expression of pluripotency-associated genes was partially restored in Snai1^−/− and Snai2^−/− cells, suggesting that suppression of EMT promotes the maintenance of pluripotency, as previously reported (Li et al., 2010; Samavarchi-Tehrani et al., 2010). The unique potential of Zeb1 in PGC specification was confirmed by enforced expression of Zeb1, Snai1 and Snai2 in wild-type ES cells, followed by PGCLC induction. Among these ES cell lines, BV expression at day 2 of PGCLC induction was severely disturbed in Zeb1-Tg line but not in Snai1-Tg and Snai2-Tg lines (Fig. 5C, Fig. S5C). We noticed that Zeb1-Tg aggregates were ruffled and fragile seemingly because of a deficit in their cell adhesion. These results demonstrated that Zeb1-mediated progression of EMT hampered PGCs through downregulation of pluripotency-associated genes and PGC-related genes.

Finally, we verified whether the defective PGC specification in Ovol2^−/− embryos was attributed to advanced EMT progression, as observed in the in vitro system. As a marker of EMT, we examined HMGA2 protein, a representative factor promoting EMT (Dong et al., 2017), because the commercially available ZEB1 antibodies failed to detect endogenous ZEB1 in gastrulating embryos. Hmga2 was upregulated in Ovol2^−/− and TKO aggregates in the same manner as Zeb1, Snai1 and Snai2 (Fig. 2E, Fig. S5D). Immunofluorescence analysis revealed that BV-positive PGCs in E7.5 Ovol2^+/− embryos showed a negligible level of HMGA2 expression and a specific level of CDH1 (Fig. 6A); the expression level of CDH1 was comparable with that in visceral endoderm, as reported previously (Okamura et al., 2003). In sharp contrast, HMGA2 was clearly visible and CDH1 became faint in BV-positive cells of E7.5 Ovol2^−/− embryos (Fig. 6A). These results demonstrated that EMT occurred in parallel with PGC specification in BV-positive cells in the Ovol2^−/− embryos. The level of HMGA2 expression was also elevated in embryonic and extra-embryonic mesodermal cells, suggesting that Ovol2 modulates EMT in not only PGCs but also surrounding somatic cells.

Based on this series of results, we propose a role for Ovol2 in the fate determination of PGCs and surrounding somatic cells during gastrulation as follows (Fig. 6B). As gastrulation occurs, OVOL2A directly binds to the TSS of Zeb1 gene and then represses the gene expression, thereby maintaining an epithelium state with expression of cell adhesion molecules, including Cdh1. A continuous epithelium state in part contributes to the maintenance of pluripotency in the epiblast. Simultaneously, OVOL2B directly binds to the promoter/enhancer of Blimp1 gene and then activates gene expression, which elicits the downstream gene expression program for PGC specification. This double-edged function ensures allocation of the PGC population during gastrulation.

Here, we revealed that Ovol2 balances segregation through the repression of EMT-related genes and the activation of PGC-related genes. Using an in vitro culture system, we have found that Ovol2, but not Ovol1 or Ovol3, plays a crucial role in PGC specification through the inhibition of EMT-related genes and activation of Blimp1. There are distinct role(s) in PGC specification for different gene family member and also for splice variants, as enforced expression of Ovol2a or Ovol2b in TKO resulted in distinct downstream gene expression (Fig. 2H). Of note, it was revealed that OVOL2B promotes Blimp1 expression through direct binding to an enhancer region enriched with H3K27ac (Fig. 4F,H). The unique role of OVOL2B may be due to its domain structure, which has only a transactivation domain, as opposed to OVOL2A, which has both transactivation and repression domains. These structural and functional differences resemble the Drosophila ovo gene that encodes at least two variants, OVO-A and OVO-B, which have both transactivation and repression domains, and only a transcription activation domain, respectively. These splicing variants have distinctive, nearly opposite, roles in Drosophila oogenesis (Andrews et al., 2000). As in the case of Drosophila OVO-A and OVO-B, OVOL2A and OVOL2B possess an identical zinc-finger domain, and indeed almost all OVOL2A-binding peaks correspond to OVOL2B-binding peaks (Fig. 4A,B). What makes the difference in the accessibility of each variant to the binding sites in the genome is currently unclear. However, different functions of the variants could be regulated by the distinctive expression dynamics of Ovol2a and Ovol2b, in which the expression peak of Ovol2a is earlier than that of Ovol2b (Fig. 1D). Considering the dynamics, it is possible that repression of EMT precedes activation of Blimp1 during mouse PGC specification.

One of the major outcomes of Ovol2-mediated EMT inhibition is that Cdh1 expression in the epithelium is sustained. Indeed, Cdh1 expression sharply dropped in Ovol2^−/− and TKO cells (Fig. 2E, Fig. S2C), and was partially restored by enforced expression of Ovol2a and/or Ovol2b (Fig. 2H). On the other hand, our results revealed that enforced expression of Cdh1 in TKO rescued only PGC-related gene expression but not pluripotency-associated gene or EMT-related gene expression (Fig. 3D). This suggests that functional requirement of Cdh1 in PGC specification is to promote the expression of the PGC-related genes, but not the pluripotency-associated genes. Interestingly, deletion of EMT-related genes rather enforced expression of Cdh1 restored pluripotency-associated gene expression in TKO cells (Figs 3D and 5B), suggesting that EMT-related genes play a dominant role in disrupting pluripotency during gastrulation. Among the EMT-related genes, Zeb1 was the crucial factor for limiting PGC differentiation, as evidenced by the observation that PGCLCs were induced from TKO ES cells by disruption of Zeb1 and that enforced expression of Zeb1 nullified PGCLC induction from wild-type ES cells (Fig. 5A,C). These characteristics were not observed for other EMT-related genes, such as Snai1 and Snai2, emphasizing the unique role of Zeb1. It is generally accepted that EMT is not simply a binary switch between the epithelium and mesenchyme, but a gradual transition with multiple intermediate states exhibiting, for example, co-expression of epithelial and mesenchymal markers (Bakir et al., 2020; Hamidi et al., 2020). A system biology approach has revealed that the balance between Ovol2 and Zeb1 governs the intermediate states in the MCF10A cell line (Hong et al., 2015). According to the general concept of EMT, there should be epiblast cells in the intermediate states during gastrulation. Considering the opposing roles of Ovol2 and Zeb1 in PGC specification, it is plausible that the balance of these genes determines the intermediate state, thereby generating heterogeneity in PGC competence in the embryonic region. Supporting this idea, earlier single cell analysis indicated that Ovol2 and Zeb1 are highly heterogenous in epiblast cells at E6.5 (Nakamura et al., 2016).

Given that Ovol2 and Zeb1 determine the balance between PGCs and somatic cells, it is still puzzling why enforced expression of Ovol2 did not induce the entire cell population into PGCLCs. This may be due to the functional threshold of Ovol2, as previous studies showed that overexpression of Ovol2, by as much as 12,000-fold, did not completely nullify the expression of EMT-associated genes (Roca et al., 2013; Kitazawa et al., 2016; Ye et al., 2016). Viewed from the other side, we should also consider why deletion of Ovol2 did not result in a complete loss of PGC(LC)s. A clue could be served from our result that Zeb1^−/− in TKO restored not only the PGC-related genes but also pluripotency-associated genes (Fig. 5B). Considering that EMT is a multistep process with intermediate states, it is possible that a small proportion of PGCs is specified without active repression of EMT by Ovol2. In this regard, Ovol2 safeguards a specific population of PGCs from EMT during gastrulation.

All animal experiments were performed in accordance with the guidelines established by Kyushu University (A20-26-3 and 1-15). Ovol2^+/− mice (RBRC02891) were provided by RIKEN BioResource Research Center (Unezaki et al., 2007; Hayashi et al., 2017). BVSC R26rtTA [reverse tetracycline transactivator (rtTA) under the Rosa26 locus] ES cells were provided by Prof. Saitou (Kyoto University, Japan) (Nakaki et al., 2013) and BVSC H18 ES cells (Hikabe et al., 2016) were used in this study. These ES cell lines were maintained under a 2i plus LIF condition without feeders (Ying et al., 2008). HEK293T cells were from American Type Culture Collection (ATCC; CRL-11268).

The CRISPR/Cas9 constructs were generated using pX330 vectors (Addgene 42230) expressing hCas9 and gRNAs against Ovol family genes. Guide RNAs for Ovol genes were designed to delete exon 1 of Ovol1, exon 2 of Ovol2 and exon 3 of Ovol3. Guide RNAs for EMT-related genes were designed to delete exons 1 and 2 of Snail1, exons 2 and 3 of Snail2, and exon 6 of Zeb1. Oligos were inserted into BbsI-digested pX330 vector. The Ovol2 variants and Cdh1-coding sequences for forced expression were amplified by PCR flanked with SfiI/NheI and NotI/EcoRI sites from cDNAs derived from ES cells, respectively. To construct these plasmids, cDNAs encoding Ovol2 variants and Cdh1 were cloned into PB-TET and PB-CAG destination vectors. To construct pX459-GFP, a GFP fragment from pCAG-Cre:GFP (Addgene 13776) amplified by PCR flanked with an EcoRI site was inserted into EcoRI-digested pX459 (Addgene 62988) vector. Two or four guide RNAs were designed for each gene. Oligos were inserted into BbsI-digested pX459-GFP vector.

Genomic regions containing regulatory elements of Zeb1, Blimp1, Nanog and T were amplified from mouse C57BL/6J genomic DNA. These regions were cloned into a pGL4.26-based (Promega, E8441) or a PL-sin-C(3+)A-based (Addgene 21313) luciferase reporter plasmid upstream of a minimal promoter. The primers used in this study are listed in Table S1.

For transfection, the pX330 vectors were transfected into BVSC R26rtTA ES cells with Lipofectamine 3000 (Invitrogen) together with pPB-CAG-rtTA-IRES-Hygro vectors (Addgene 102423) on feeders. The total amount of vector DNA was 2.5 µg. Transfectants were selected with hygromycin B for 2 days (150 µg/ml). Three days after the transfection, the cells were sorted for DsRed2 expression and seeded as a single cell. Single colonies were picked up and cultured on mitomycin C-treated mouse embryonic fibroblasts.

The PBTET-Ovol2a or -Ovol2b was transfected into TKO ES cells together with PBase vectors and pGG131 vectors using 4D-Nucleofector (LONZA) in a 60 mm dish under a 2i plus LIF condition. The total amount of DNA was 2 µg. Transfectants were selected with hygromycin B for 4 days. PBCAG-Cdh1, PBCAGDD-Snai1, -Snai2 or -Zeb1-IRESneo were transfected into TKO and wild-type ES cells together with PBase vectors using Lipofectamine 2000 (Invitrogen) in 12-well plates. The total amount of DNA was 4 µg. Transfectants were selected with G418 for 4 days. After drug selection, cells were seeded singly and single colonies were picked up.

HEK293T cells were seeded in one well of a 12-well plate. On the next day, PL-sin-C(3+)A-Blimp1, HPV275, P633, HPV17 and pHCMV-VSV-G plasmids were transfected into HEK293T cells using Lipofectamine 2000. After 24 h, the medium was replaced with 2i plus LIF medium. The next day, Ovol2b-Tg ES cells were seeded in one well of a 24-well plate with virus-containing supernatants from the HEK293T cultures for 24 h. The infected cells were selected with zeocin for 4 days (4 µg/ml).

5×10⁴ ES cells were cultured in one well of a 24-well plate coated with human plasma fibronectin (Merck Millipore) (16.7 µg/ml) in N2B27 medium containing activin A (20 ng/ml; Preprotech), bFGF (12 ng/ml; Wako) and KSR (1%). The medium was changed 24 h later. The EpiLCs were then cultured under a floating condition by plating 2×10³ cells in one well of a low-cell-binding U-bottom 96-well plate (Greiner) in GK15+BMP4 (500 ng/ml; R&D Systems), LIF (1000 u/ml; Nacalai), EGF (50 ng/ml; R&D Systems) and SCF (100 ng/ml; R&D Systems). For activation of Ovol2a- or Ovol2b-Tg, Dox (1.5 µg/ml) was added to the medium.

Total RNAs from ES cells, EpiLCs, aggregates and sorted BV-positive cells were extracted and purified using an RNeasy Micro Kit (QIAGEN), and reverse transcribed by PrimeScript (Takara). The first-strand cDNAs were used for Q-PCR analysis with Power SYBR Green (ABI).

Total RNAs were extracted and purified using an RNeasy Micro Kit, and mRNAs were isolated with the NEBNext poly(A) mRNA magnetic isolation module (NEB). Biologically duplicated samples were prepared at each stage. Purified RNAs were subjected to library construction using a NEBNext Ultra Directional RNA Library Prep Kit for Illumina (NEB). Adaptor-ligated cDNA libraries were amplified by 12-cycle PCR. Sequencing of the libraries was performed with Hiseq 2500 and Nextseq 550 (Illumina). Obtained reads were mapped to the mouse GRCm38/mm10 genome using Hisat2. Mapped reads were counted by featureCounts. Principal component analysis (PCA) was performed using R software with FactoMineR. For DEG analysis, the false discovery rate (FDR) and log₂ fold-change were calculated using edgeR (Robinson et al., 2010). The DAVID database was used for gene ontology (GO) analysis (Huang et al., 2009).

Whole aggregates at day 2 (equivalent to 2×10⁶ cells) were trypsinized, washed and collected by centrifugation at 270 g for 5 min. For crosslinking, the pellets were resuspended in PBS containing 1% formaldehyde, incubated for 10 min and quenched with 125 mM glycine. The fixed cells were resuspended in 1 ml of LB1 [50 mM HEPES-KOH (pH 7.5), 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% TritonX-100] and pelleted by centrifugation at 1500 g for 5 min. The pellets were resuspended in 1 ml of LB2 [20 mM Tris (pH 7.5), 200 mM NaCl, 1 mM EDTA and 0.5 mM EGTA] and pelleted by centrifugation at 1500 g for 5 min. The pellets were resuspended in 1 ml of LB3 [20 mM Tris (pH 7.5), 150 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% TritonX-100, 0.1% sodium deoxycholate and 0.1% SDS] and pelleted by centrifugation at 1500 g for 5 min. The nuclei were lysed in 200 µl of SDS buffer [20 mM Tris (pH 7.5), 150 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% sodium-deoxycholate, 1% SDS and protease inhibitor cocktail]. The lysed nuclei were sonicated using a sonicator (Branson) for 10 cycles. Protein-DNA complexes were immunoprecipitated at 4°C overnight using 2 µg of antibodies bound to 50 µl of Dynabeads Protein G (Invitrogen). Immunoprecipitates were washed with 1 ml of LB3 twice, high-salt buffer [20 mM Tris (pH 7.5), 500 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% TritonX-100, 0.1% sodium-deoxycholate and 0.1%SDS], RIPA buffer [50 mM HEPES-KOH (pH 7.4), 0.25 M LiCl, 1 mM EDTA, 0.5% sodium-deoxycholate and 1% NP-40] and TE buffer [50 mM Tris (pH 8.0) and 10 mM EDTA]. For collection of the protein-DNA complexes, beads were resuspended with 100 µl of elution buffer [50 mM Tris (pH 8.0), 10 mM EDTA and 1% SDS]. The immunoprecipitated and input DNA were reverse crosslinked by incubating at 65°C overnight. The mixtures were supplemented with 20 µg of RNaseA and incubated at 37°C for 1 h. After Proteinase K digestion, the DNA was purified using a PCR purification kit (Fastgene, FG-91302) and dissolved with distilled water.

The ChIP and input DNAs were sheared to an average size of ∼150 bp by ultra-sonication (Covaris, S220). Sonicated DNA fragments were end-repaired, ligated to sequencing adapters and amplified according to the manufacturer's instructions (NEB, E7645). Libraries were sequenced using NextSeq 550 with single-end 75 nucleotide read lengths.

For data analysis, ChIP-seq reads were aligned to the mouse reference genome (GRCm38/mm10) using Bowtie v2.3.5 using default parameters (Langmead and Salzberg, 2012). Peaks were called with MACS version 2.2.6 (Zhang et al., 2008) with default settings and visualized using an Integrative Genomics Viewer (Thorvaldsdottir et al., 2013). The consensus sequences of OVOL2A and OVOL2B were identified by HOMER (Heinz et al., 2010). Genomic annotation of the peaks identified from the ChIP-seq data was performed using bedtools (Quinlan and Hall, 2010). Unsupervised hierarchical clustering (UHC) was performed using the hclust function with Pearson correlation distances and Ward's method (ward.D2). The normalized IP/input ratios were determined as peak density divided by input within 1 kb of the TSSs. Previously published H3K4me3, H3K27me3 and H3K27ac ChIP-seq datasets (Kurimoto et al., 2015) were aligned to the mouse reference genome with Bowtie v2.3.5 using the ‘-N 1 -3 5 –local’ options. The reads were processed according to the experimental application, as described above, except that the normalized IP/input ratios were determined as the peak density within 1 kb of the TSSs divided by input within 5 kb. Bivalent genes have been listed previously (Kurimoto et al., 2015).

HEK293T cells were used to test the regulatory elements for Zeb1 and were transfected with the reporter plasmids (80 ng/well) together with the Ovol2a or Ovol2b expression plasmids (or a mock plasmid) (120 ng/well). The transfections were performed with Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. Transfected cells were seeded at a density of 7.5×10⁴ cells in one well of a 96-well plate and were lysed 24 h after the transfection for analyses using the Dual-Glow Luciferase Assay System (E2920; Promega).

Whole aggregates derived from stable luciferase reporter Ovol2b-Tg ES cells at day 1±Dox (equivalent to 5×10⁴ cells) were trypsinized and collected by centrifugation. Luciferase assays were performed with a ONE-Glo Luciferase Assay system (Promega, E6110) using Ensight (PerkinElmer).

For whole-mount immunofluorescence analysis of aggregations, aggregates were fixed at day 2 in 4% paraformaldehyde (PFA) in PBS for 1 h at room temperature, washed with PBST (0.2% Tween20), soaked in blocking buffer (PBS containing 0.1% BSA and 0.3% Triton X-100) overnight at 4°C and incubated with primary antibodies diluted with blocking buffer for 2 days at 4°C. The samples were washed with washing buffer (PBS containing 0.3% Triton X-100) and then incubated with secondary antibodies and DAPI overnight at 4°C. Finally, the samples were washed and mounted in Fluoro-KEEPER antifade reagent (Nacalai Tesque, 12593-64). For whole-mount immunofluorescence analysis of embryos, isolated embryos were fixed in 4% PFA in PBS for 1 h at 4°C, washed with PBST (0.2% Tween20) and incubated in blocking solution 1 (PBS containing 1% FBS and 0.2% Tween20) overnight. Embryos were incubated with primary antibodies in blocking solution 1 for 3 days at 4°C, washed with PBST, incubated with secondary antibodies and DAPI for 2 days at 4°C, and then washed and mounted in Fluoro-KEEPER antifade. For immunofluorescence analysis of PGCLCs at day 4, aggregates were trypsinized and then spread onto MAS-coated glass slides (Matsunami, MAS-04). The slides were fixed in 4% PFA in PBS for 15 min at room temperature, washed with PBST and permeabilized with 0.2% Triton X-100 in PBS for 15 min at room temperature. Next, the slides were incubated in blocking solution 2 (PBS containing 5% FBS and 0.2% Tween20) for 1 h at room temperature followed by incubation with primary antibodies in blocking solution 2 overnight at 4°C. After washing with PBST, the slides were incubated with secondary antibodies and DAPI in blocking solution 2 for 1 h at room temperature, washed and mounted in Fluoro-KEEPER antifade. The antibodies used in this study are listed in Table S2.

We are grateful to Drs K. Nakashima, T. Matsuda, Y. Ohkawa, T. Ito, F. Miura, and S. Okada for technical support and to Dr M. Saitou for providing BVSC R26rtTA ES cells. We thank the Research Support Center, Kyushu University Graduate School of Medical Sciences for their technical assistance.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200319.