Klf5 maintains the balance of primitive endoderm versus epiblast specification during mouse embryonic development by suppression of Fgf4

Mammalian preimplantation embryo development segregates into three fundamental cell lineages. The first lineage segregation event separates an epithelial cell layer called the trophectoderm (TE) on the surface of the embryo, which gives rise to trophoblast tissues of the placenta, and the inner cell mass (ICM), which gives rise to the embryo proper and extra-embryonic mesoderm. The second lineage segregation event further differentiates the ICM into epiblast (EPI) cells and the primitive endoderm (PrE). EPI cells generate most of the embryo proper and are a source of pluripotent embryonic stem cells (ESCs), whereas the PrE generates visceral and parietal endoderm tissues, and these become the visceral and parietal yolk sacs (Rossant and Tam, 2009). After compaction, inner cells generated in the first wave of cell division (8- to 16-cell stage) mainly contribute to EPI cells, whereas inner cells generated in the second wave (16- to 32-cell stage) are biased towards PrE cells (Morris et al., 2010). However, a previous study did not observe such predetermination in EPI and PrE specification (Yamanaka et al., 2010).

Single-cell analysis revealed that inner cells at embryonic day (E) 3.25 randomly coexpress EPI and PrE markers and eventually acquire either an EPI or a PrE identity (Guo et al., 2010; Kurimoto et al., 2006; Ohnishi et al., 2014). The mechanism involved in the emergence and specification of EPI and PrE precursor cells is not fully understood (Bedzhov and Zernicka-Goetz, 2015; Hermitte and Chazaud, 2014). The cell fate of EPI and PrE precursors is still plastic at E3.75 and is fully committed to EPI and PrE cells at ∼E4.0-4.25 (Grabarek et al., 2012; Nichols et al., 2009; Yamanaka et al., 2010).

Nanog and Gata6 mRNAs are detectable as early as the 2-cell stage (Guo et al., 2010), and both Nanog and Gata6 proteins are coexpressed in all inner cells at E3.25 (∼32 cells) (Dietrich and Hiiragi, 2007; Plusa et al., 2008). Variability in the initial Nanog expression level shows no correlation with that of Gata6 in individual inner cells at E3.25 (Ohnishi et al., 2014). As embryos develop, the salt-and-pepper distributions of Nanog and Gata6 are evident in ICM cells until E3.5-4.0 (Chazaud et al., 2006; Plusa et al., 2008). The establishment of this salt-and-pepper distribution of Nanog and Gata6 is poorly understood; however, fibroblast growth factor (FGF)–extracellular signal-regulated kinase (ERK) signalling is postulated to play key roles (Chazaud et al., 2006; Kang et al., 2013; Krawchuk et al., 2013; Nichols et al., 2009; Ohnishi et al., 2014; Yamanaka et al., 2010). Fgf4-Fgfr-ERK signalling determines the balance of the EPI and PrE cell lineages: overactivation of FGF signalling caused by exogenous FGF4 converts all ICM cells to PrE cells (Yamanaka et al., 2010), whereas all ICM cells acquire EPI identity when FGF signalling is blocked by small chemical inhibitors of Fgfr and ERK (Nichols et al., 2009), by gene knockout (KO) of Fgf4 (Kang et al., 2013; Krawchuk et al., 2013), and by KO of the expression of the adapter molecule Grb2 (Chazaud et al., 2006). Fgf4 KO embryos show normal Nanog and Gata6 expression levels at E3.25; thus, the initial coexpression of Nanog and Gata6 is independent of Fgf4 (Kang et al., 2013; Krawchuk et al., 2013; Ohnishi et al., 2014). Subsequently, the segregation of EPI and PrE precursor cells is believed to be mediated by reciprocal repression between Nanog and Gata6 (Singh et al., 2007). In agreement with this model, the loss of Gata6 results in a complete shift into the EPI lineage, while there is no effect on the Nanog expression level at E3.0-3.25; thus, the initial expression of Nanog is independent of Gata6 at E3.0-3.25 (Bessonnard et al., 2014; Schrode et al., 2014). Importantly, single-cell analysis showed that bimodal Fgf4 expression precedes asymmetric Nanog and Gata6 expression and is the first sign of the segregation of the EPI and PrE lineages (Guo et al., 2010; Ohnishi et al., 2014). Currently, what regulates Fgf4 at this developmental stage is unknown (Artus and Chazaud, 2014; Chazaud and Yamanaka, 2016).

Klf5, a member of the Krüppel-like factor (Klf) family of transcription factors, functions in the maintenance of pluripotency and in somatic cell reprogramming (Takahashi and Yamanaka, 2006). Klf5 marks a naïve state of human pluripotent stem cells (Chan et al., 2013; Theunissen et al., 2014). Target gene inactivation of Klf5 causes failure of TE and ICM development (Ema et al., 2008; Lin et al., 2010), but the molecular mechanism underlying Klf5-regulated ICM development is not well understood.

Here, we show that the inner cells of Klf5 KO embryos adopt a PrE lineage fate at the expense of EPI cells, while Klf5-overexpressing (OE) embryos show incomplete lineage segregation as indicated by the persistence of Nanog⁺/Gata6⁺ double-positive cells. We show that Fgf4 expression is upregulated in Klf5 KO embryos at E3.0, whereas Fgf4 is repressed in Klf5 OE blastocysts. Importantly, single-cell analysis clearly demonstrates that Fgf4 is derepressed in a subset of Fgf4-high inner cells of Klf5 KO embryos. Chromatin immunoprecipitation (ChIP) assays indicate that the Fgf4 locus is occupied by Klf5, suggesting direct regulation of Fgf4 by Klf5. In terms of the emergence of Nanog⁺ pluripotent EPI cells, the phenotypes of Klf5 KO embryos can be reversed by either Fgfr or ERK inhibition. Taken together, these results provide new insights into the interplay between Klf5 and the Fgf4-Fgfr-ERK pathway crucial for the proper lineage segregation.

Although Klf5 is indispensable for blastocyst development, the mechanistic functions of Klf5 in ICM development and early lineage segregation remain elusive (Ema et al., 2008; Lin et al., 2010). In a previous study, the Klf5-lacZ allele (Ema et al., 2008) was generated by inserting a lacZ cassette into the second exon of Klf5, leaving the rest of the Klf5 locus with the potential to generate a C-terminally truncated protein of ∼164 amino acids in length (Fig. S1A). To delete almost the entire open reading frame of Klf5, we generated a new KO mouse for Klf5 that removes the two major coding exons (designated Klf5Δ2nd3rd exon) (Fig. S1B). Similar to Klf5-lacZ mice (Fig. S1A) (Ema et al., 2008), no homozygous pups were obtained from heterozygous intercrosses of the Klf5Δ2nd3rd exon mice, and homozygous Klf5^{Δ2nd3rd exon/Δ2nd3rd exon} embryos showed implantation defects (Fig. S1C). When Klf5 KO morulae (Klf5^lacZ/lacZ or Klf5^{Δ2nd3rd exon/Δ2nd3rd exon}) at E2.5 were stained with antibodies against lineage markers such as Oct3/4 (Pou5f1) and Cdx2, both types of Klf5 KO embryos showed decreased levels of Cdx2 and normal levels of Oct3/4 protein expression (Fig. S1D,E). Hereafter, the Klf5Δ2nd3rd exon allele was used as the null allele for our study.

To obtain insight into the role of Klf5 in early embryogenesis, we established a new ESC line with Cre-mediated overexpression of the FLAG/HA-Klf5 cassette (Fig. S1F). We used the ESC line, which expresses GFP prior to Cre-mediated excision, to generate the conditional Klf5 OE mice (Fig. S1G). Upon crossing to the Ayu1-Cre driver line, we confirmed that Klf5 protein is overexpressed 1.5-fold in Klf5 OE blastocysts compared with wild type (WT) (Fig. S1H).

First, we collected embryos carefully timed every 6 h from E3.25 onwards and found that Klf5 KO embryos had fewer cells than their WT counterparts (Fig. S1I). The total cell number per Klf5 KO embryo never exceeded 64 until E4.25 (Fig. S1I; data not shown). Given that bromodeoxyuridine (BrdU) incorporation was severely affected, it is likely that cell cycle progression had been disturbed (Fig. S1J,K). Co-staining embryos from E3.25 to E3.5 for BrdU incorporation and terminal deoxynucleotidyl transferase-mediated dUTP nick-end labelling (TUNEL) showed that the cells defective in BrdU incorporation were TUNEL⁺, suggesting that defective cell cycle progression was promoting apoptosis (Fig. S1L).

Examination of Nanog and Gata6 expression levels in Klf5 KO embryos revealed that the initial Nanog and Gata6 expression at E3.25 was, overall, similar to that of WT embryos (Fig. 1A). Analyses of Klf5 KO blastocysts at E3.5 and E3.75 showed that most, if not all, ICM cells were Gata6⁺ and that few cells were Nanog⁺ (Fig. 1A,C, Fig. S2A,C). Consistent with this result, Nanog protein expression levels decreased in Klf5 KO embryos during development from E3.25 to E4.0, whereas Gata6 protein levels increased (Fig. 1B). Since it was reported that Nanog⁺/Gata6⁺ double-positive (DP) common precursors differentiate progressively into a Nanog⁺/Gata6⁻ (Nanog⁺) EPI or Nanog⁻/Gata6⁺ (Gata6⁺) PrE fate in an asynchronous manner (Saiz et al., 2016), we evaluated the percentage of DP cells, Nanog⁺ cells and Gata6⁺ cells from E3.25 to E4.0. The percentage of DP cells in Klf5 KO embryos decreased rapidly and, in turn, the percentage of Gata6⁺ cells increased, indicating that bipotential DP cells in Klf5 KO embryos prefer to differentiate into Gata6⁺ PrE cells (Fig. 1C). At the E4.0 late blastocyst stage, most cells in Klf5 KO embryos acquired the Gata6⁺ PrE fate (Fig. 1A,C).

Our finding that only Gata6-expressing cells in Klf5 KO embryos remain at E4.0 could be the consequence of apoptosis of Nanog⁺ EPI cells rather than of more cells differentiating into PrE. To gain insight into the role of apoptosis in the EPI lineage, we analysed the number of TUNEL⁺ cells in Klf5 KO and WT embryos at E3.5. We found no significant changes in the percentage of Nanog⁺, Gata6⁺, DP, or double-negative (DN) cells undergoing apoptosis between WT and Klf5 KO embryos (Fig. S1M-O). This indicated that the increase in the percentage of Gata6⁺ cells was not caused by the death of any specific cell lineage, including EPI cells. Because the percentage of DP cells in Klf5 KO embryos was decreased (Fig. 1C), DP cells were likely to have differentiated into PrE. Because the Klf5 KO embryos showed a cell proliferation defect (Fig. S1I) we categorised embryos by similar total cell numbers (<32 or 32-64) from various days of development and reached the same conclusion (Fig. S2B-D).

When we investigated Nanog and Gata6 expression in Klf5 OE embryos at E3.25, Klf5 OE blastocysts showed overall coexpression of Nanog and Gata6 (Fig. 1A). The numbers of DP cells per embryo at E3.5 to E4.0 were significantly increased in Klf5 OE embryos (Fig. 1C, Fig. S2C,D). Whereas there were Gata6⁺ endoderm layers and Nanog⁺ EPI cells in WT blastocysts at E4.5, there were still significant numbers of DP cells centrally located towards presumably uncommitted cells in Klf5 OE blastocysts at E4.5 (Fig. S2A,A′).

Specification of the PrE lineage involves Gata6, followed by sequential activation of Pdgfra, Sox17, Gata4 and Sox7 (Artus et al., 2010, 2011; Kang et al., 2013; Plusa et al., 2008). Sox17 is activated between the 32-cell and 64-cell stages (Artus et al., 2011; Morris et al., 2010; Niakan et al., 2010); Gata4 expression marks the onset of the mutual exclusion of Gata6 and Nanog and is activated at the 64-cell stage (Artus et al., 2011; Grabarek et al., 2012). Because we observed a reduction in the percentage of Nanog⁺ cells and an increase in Gata6⁺ cells in Klf5 KO embryos at E3.75, we tested whether inner cells at E3.0 already exhibit signs of accelerated PrE lineage specification. Immunohistochemistry did not show any detectable Pdgfra or Sox17 protein expression in Klf5 KO morulae at E3.0 (data not shown; Fig. S3A). However, at E3.25, immunohistochemistry showed strong Pdgfra expression in most of the inner cells of Klf5 KO embryos but not WT embryos (Fig. 2A). Quantitation of Pdgfra⁺ cells indicated that over 80% of the inner cells of Klf5 KO embryos were Pdgfra⁺, whereas fewer than 20% of the inner cells of WT embryos were Pdgfra⁺ (Fig. 2B). It is of note that a small increase in Pdgfra⁺ cells was observed in the outer cells of Klf5 KO embryos (Fig. 2B). Taken together, the data revealed that the inner cells of Klf5 KO embryos exhibit accelerated PrE lineage specification as early as E3.25 (Fig. 2C).

We also found an increase in the percentage of Sox17⁺ cells in Klf5 KO blastocysts at E3.5 (Fig. S3B-D), which was consistent with a previous report (Lin et al., 2010). Furthermore, we found an increase in the numbers of Gata4⁺ cells in Klf5 KO blastocysts at E3.75, which was in sharp contrast with the small number of cells found in Klf5 OE embryos (Fig. 2D,E). Gata4 protein expression levels were increased in Klf5 KO embryos and decreased in Klf5 OE embryos (Fig. 2F). Collectively, these data indicated that Klf5 KO embryos show an accelerated PrE lineage specification in the ICM.

During maturation of EPI cells at E4.25-4.5, it has been demonstrated that downregulation of Nanog expression is a hallmark of the naïve-to-primed transition of EPI cells (Kang et al., 2017; Saiz et al., 2016; Smith, 2017). Elevated Nanog expression in Klf5 OE embryos from E3.25 to E4.25 suggests that these cells might remain in a naïve state and fail to differentiate into mature EPI cells. To examine the consequence of ubiquitous overexpression of Klf5, Klf5 OE embryos at E5.5 (egg cylinder stage) were dissected and subjected to immunohistochemistry. In WT embryos, Klf5 protein was expressed in extra-embryonic ectoderm cells but not in EPI cells; however, Klf5 OE embryos still expressed Klf5 protein in all cell lineages, including EPI cells (Fig. S4). Thus, the elevated Nanog expression caused by Klf5 OE did not block differentiation into EPI cells. Yamanaka and colleagues indicated that even though Fgf4 heterozygous embryos exhibit a reduction in the number of PrE cells, the embryos eventually develop normally (Krawchuk et al., 2013). Thus, although lower activity of the Fgf4-Fgfr-ERK pathway affects PrE maturation it is restored during development. We presume that the PrE maturation of Klf5 OE embryos is restored, as seen in Fgf4 heterozygous embryos. Klf5 OE embryos developed normally until E8.0 but then died at E11.5 for unknown reasons, while Klf5 KO embryos showed reduced Cdx2 expression and failed to promote blastocoel expansion, indicating a defect in TE development. Taken together, these results indicated that loss or overexpression of Klf5 results in skewed cell fate specification in the EPI/PrE lineages during preimplantation development.

To elucidate the molecular mechanism involved in the accelerated PrE lineage specification of Klf5 KO embryos, microarray analyses were performed using amplified cDNAs from WT embryos and Klf5 KO embryos at E3.0 (Fig. S5A), which at this stage showed no apparent defects and had normal expression levels of Oct3/4, Nanog, Sox2 and Cdx2 mRNAs (Fig. S5B). Bioinformatic analysis indicated that Fgf4 expression was upregulated in Klf5 KO embryos, whereas Spry4, a negative regulator of FGF-induced ERK activation, was downregulated (Fig. 3A). Pdgfra and Sox17, markers for the PrE lineage, were upregulated in Klf5 KO embryos (Fig. 3A). In agreement with these observations, quantitative reverse-transcription PCR (RT-qPCR) analysis confirmed that Fgf4, Pdgfra and Sox17 were significantly upregulated in Klf5 KO embryos at E3.0 (Fig. 3B). Quantification of Sox17 immunostaining in Klf5 KO embryos showed increased numbers of Sox17⁺ cells per embryo, as well as increased staining for Sox17 on a per-cell level at E3.5 (Fig. S3C,D). Therefore, the upregulation of PrE genes in the microarray analysis appears to reflect a combination of increased PrE cells per embryo and, in some cases such as Sox17, an increased level of PrE gene expression per cell.

To examine FGF4 protein expression in Klf5 KO embryos, we validated an anti-FGF4 antibody by staining mouse Fgf4 KO and WT ESCs and found that it could exclusively recognise endogenous FGF4 protein expressed in these cells (Fig. S5C). We further validated the anti-FGF4 antibody by staining the Nanog⁺ EPI cells of WT blastocysts at E3.75 (Fig. S5D) and found that the staining pattern was consistent with a previous report by Frankenberg et al. (2011). Immunohistochemical analysis with this antibody confirmed that FGF4 was abundantly expressed in Klf5 KO embryos at E3.0 and E3.25 (Fig. 3C-F). By contrast, FGF4 expression was significantly reduced in Klf5 OE embryos at E3.25, indicating that Klf5 suppresses Fgf4 (Fig. 3E,F). Since Fgf4 encodes a secreted protein, it is difficult to identify Fgf4-expressing cells. To resolve this issue, we performed single-cell RT-qPCR analysis with amplified cDNA prepared from individual blastomeres of inner cells of WT, Klf5 KO and Klf5 OE embryos at E3.25 using the single-cell mRNA 3-prime end sequencing (SC3-seq) method (Nakamura et al., 2015) (Fig. 4A). There were two populations: Fgf4-high inner cells and Fgf4-low/negative inner cells. Given that Pdgfra, Fgfr2 and Sox17 are expressed in Fgf4-low/negative inner cells but not in Fgf4-high inner cells, these populations might represent PrE and EPI cells, respectively (Fig. 4A). There was no significant difference in the expression pattern of major lineage markers, such as Nanog, Gata6, Oct3/4 and Sox2, between WT, Klf5 KO and Klf5 OE embryos (Fig. 4A). Importantly, the proportion of Fgf4-high inner cells to Fgf4-low/negative inner cells was significantly increased in Klf5 KO embryos at E3.25 (Fig. 4B,C). Interestingly, Fgf4 mRNA was significantly upregulated in Fgf4-high inner cells of Klf5 KO embryos but downregulated in Fgf4-high inner cells of Klf5 OE embryos, as compared with those of WT embryos (Fig. 4B,D). These results clearly demonstrated that Klf5 suppresses Fgf4 in Fgf4-high inner cells at E3.0-3.25.

To investigate whether Klf5 directly regulates Fgf4, we surveyed the genomic binding sites of Klf5 by examining public ChIP-seq data and found that three candidate regions in Fgf4 loci were occupied by Klf5 in mouse ESCs (Fig. 4E). To verify this result, we established ESC lines that overexpressed epitope-tagged Klf5 and confirmed that the tagged protein binds to the three regions of Fgf4 (Fig. 4F) and to the promoter and enhancer regions of Nanog (Jeon et al., 2016). These data suggest that Klf5 represses Fgf4 through direct regulation. Although inner cells at E3.25 express Fgfr2 homogeneously, Fgf4 was observed to be expressed at two distinct levels, namely high and low, in populations of cells (Guo et al., 2010; Kurimoto et al., 2006; Ohnishi et al., 2014). However, the regulatory mechanism of Fgf4 at early stages, such as in the morula, is unknown (Chazaud and Yamanaka, 2016). To the best of our knowledge, we are the first to identify Klf5 as a crucial regulator of Fgf4 at E3.0-3.25.

Of note, although cis-regulatory regions in the Fgf4 promoter and enhancers are occupied by Klf5 (Fig. 4E,F), a lack of Klf5 expression did not significantly alter Fgf4 expression in mouse ESCs (Fig. S5E). This finding suggests a minor role for Klf5 in the transcription of Fgf4 in mouse ESCs, in contrast to its strongly suppressive role on Fgf4 at E3.0 or E3.25. Previous work showed that Klf5 is a context-dependent transcription factor and, depending on the co-factor or nuclear environment, behaves as a transcriptional repressor or activator on the same set of genes (Oishi et al., 2008). Therefore, the different outcomes of Fgf4 transcription promoted by the absence of Klf5 might result in a different cell type, i.e. inner cells at E3.0-3.25 versus mouse ESCs (equivalent to EPI cells at E4.25) (Boroviak et al., 2014).

Our data indicate that Klf5 KO blastocysts show cell proliferation defects and accelerated PrE specification (Fig. S1I, Figs 1 and 2), yet it is not clear whether the overactivation of Fgf4-Fgfr-ERK signalling is responsible for the cell proliferation defects and accelerated PrE lineage specification in Klf5 KO embryos. In fact, Lin et al. (2010) reported no difference in the phosphorylated ERK (pERK) signal between WT and Klf5 KO blastocysts. However, it is of note that careful and extensive immunohistochemical analyses showed that a strong background signal hampers the detection of pERK in preimplantation embryos (Frankenberg et al., 2011), whereas pERK signals can be observed reproducibly in whole embryos after E5.5 (Corson et al., 2003).

To examine whether Fgf4-Fgfr-ERK signalling is responsible for the phenotype of Klf5 KO blastocysts, we used chemical inhibitors to block the kinase activities of Fgfr1/2 (SU5402) and MEK (PD0325901) (Fig. 5A). Morulae at E2.75 were collected and cultured for 24 h in the presence or absence of these inhibitors, and we found that Klf5 KO embryos treated with either SU5402 or PD0325901 showed marked phenotypic rescue in terms of a normal morphology with an expanded blastocoel, indistinguishable from WT blastocysts (Fig. 5B). The number of cells per embryo was also increased but still significantly different from that of WT embryos, suggesting that Klf5 regulates cell proliferation in part through an Fgf4-Fgfr-ERK-independent mechanism (Fig. 5C). The number of TUNEL⁺ cells was reduced dramatically in the inhibitor-treated Klf5 KO embryos and was similar to that of WT embryos (Fig. 5C). Notably, inhibitors of the JNK and p38 MAPK (Mapk14) pathways did not significantly rescue the cell cycle defects of Klf5 KO blastocysts (Fig. S6), indicating that JNK and p38 MAPK are not involved in this process.

We also attempted to test whether excess FGF4 activity is sufficient to cause defective cell proliferation in WT embryos and found that culturing in the presence of FGF4 slightly increased the number of TUNEL⁺ cells but did not significantly change total cell number (Fig. 5D,E). When Klf5 OE embryos were cultured in the presence of saturated levels of FGF4 (1 μg/ml) from E2.5 to E3.75 (Fig. S7A), this did not change total cell number (Fig. S7B-D); however, Gata6 protein expression was significantly upregulated in FGF4-stimulated Klf5 OE embryos (Fig. S7E).

To investigate whether the precocious activation of Fgf4-Fgfr-ERK signalling was responsible for the altered lineage specification of Klf5 KO blastocysts, WT and Klf5 KO morulae at E2.5 were collected and cultured for 24 or 48 h in the presence or absence of PD0325901 and were then subjected to immunohistochemistry (Fig. 6A). Klf5 KO blastocysts cultured in vitro for 24 h from E2.5 showed reduced Nanog and increased Gata6 expression levels, as did the Klf5 KO blastocysts at E3.5 (Fig. 6B,C). MEK inhibitor treatment dramatically reversed the alterations in Nanog and Gata6 expression levels (Fig. 6B,C). When Klf5 KO morulae were treated with PD0325901 from E2.5 for 24 h, most of the ICM cells were Nanog⁺ EPI-biased cells (Fig. 6B,C). By contrast, when Klf5 KO morulae were treated with vehicle alone for 24 h from E2.5, most of the ICM cells were Gata6⁺ PrE cells. Importantly, the percentage of Nanog⁺ cells among the ICM cells of Klf5 KO blastocysts cultured in the presence of PD0325901 was increased compared with WT, but was still less than that of PD0325901-treated WT embryos (Fig. 6C). When Klf5 KO morulae were treated with PD0325901 from E2.5 for 48 h (corresponding to E4.5), all the cells in the ICM were Nanog⁺ EPI cells (Fig. 6B′).

Because our previous results indicated that ESC lines could not be established from Klf5 KO blastocysts and that Klf5 was indispensable for ESC derivation from the ICM (Ema et al., 2008), we attempted to derive ESC lines from PD0325901-treated Klf5 KO blastocysts. Sixty ESC lines were established and genotyped. Seven Klf5 KO ESC lines were obtained (data not shown), demonstrating that treatment with this MEK inhibitor rescued the emergence of pluripotent EPI cells in Klf5 KO blastocysts. By contrast, treatment with the MEK or Fgfr inhibitors did not affect the reduced expression of Cdx2 (Fig. S8). This suggested that the dysregulation of Cdx2 in Klf5 KO blastocysts was not caused by increased signalling through the Fgf4-Fgfr-ERK pathway but was nonetheless controlled by Klf5.

Taken together, our studies showed that the loss of Klf5 results in the induction of Fgf4 in morula at E3.0, followed by the rapid upregulation of Pdgfra in the inner cells at E3.25 and a decrease in Nanog⁺ cells and DP cells at E3.5, which ultimately led to Gata6-expressing cells only. This skewed EPI/PrE phenotype with Gata6-only ICM cells was reversed by MEK inhibitors. However, overexpression of Klf5 resulted in the reduction of Fgf4 in blastocysts at E3.5 and an increase in Nanog⁺ cells and DP cells at E3.5 or later, which ultimately led to the presence of DP ʻuncommitted' cells by E4.0. Therefore, our model is that Klf5 represses Fgf4-Fgfr-ERK signalling to suppress its precocious activation of the PrE specification programme, thus ensuring the emergence of Nanog⁺ naïve pluripotent cells during development (Fig. 7).

Previous experiments showed that Nanog KO caused a severe reduction in Fgf4 expression in blastocysts at E3.5, demonstrating that Nanog activates Fgf4 expression in EPI precursor cells (Frankenberg et al., 2011; Messerschmidt and Kemler, 2010). Moreover, Oct3/4 KO led to reduced Fgf4 expression in blastocysts at E3.5 (Frum et al., 2013; Le Bin et al., 2014). Furthermore, mutant embryos lacking Sox2, which cooperates with Oct3/4 in the maintenance of ESC pluripotency, showed reduced Fgf4 expression (Wicklow et al., 2014). Because Oct3/4-Sox2 complexes can directly induce Fgf4 expression in vitro (Ambrosetti et al., 2000), there is an interplay between Nanog, Oct3/4 and Sox2 to regulate Fgf4 for proper lineage segregation (Chazaud and Yamanaka, 2016). Thus, these studies showed that Nanog and Oct3/4 activate Fgf4 expression in EPI precursor cells at the E3.5 blastocyst stage; however, these studies did not address the regulatory mechanism of Fgf4 at an earlier stage, such as the morula. The mechanism involved in the induction of Fgf4 expression in a subset of inner cells at this stage is still unknown (Chazaud and Yamanaka, 2016). Our study demonstrated for the first time that Klf5 is a crucial regulator for Fgf4 in the morula at E3.0, before the blastocyst stage.

Bimodal Fgf4 expression levels precede the exclusive production of Nanog and Gata6 (Guo et al., 2010; Ohnishi et al., 2014) and is the first sign of the segregation of the EPI and PrE lineages. However, what regulates Fgf4 is unknown (Artus and Chazaud, 2014; Chazaud and Yamanaka, 2016). Yamanaka and colleagues clearly demonstrated that all ICM cells acquire a PrE fate when cultured in the presence of saturated levels of FGF4, and they proposed that the local concentration of FGF4 in the inner cells is important for the establishment of its salt-and-pepper distribution at the blastocyst stage (Yamanaka et al., 2010). At E3.0-3.25, DP inner cells exist as a common precursor pool and have the potential to commit to either fate asynchronously (Saiz et al., 2016). Activity of the Fgf4-Fgfr-ERK pathway regulates the balance of EPI and PrE differentiation from common precursors. We demonstrated that Fgf4 expression is induced in a subset of Fgf4-high inner cells of Klf5 KO embryos at E3.0-3.25. At that time, the number of DP cells was reduced in Klf5 KO embryos. Furthermore, the skewed cell fate of Klf5 KO embryos was markedly reversed by inhibitors of MEK and Fgfr. These results demonstrated that Klf5 is involved in the segregation of the EPI and PrE lineages by suppressing precocious activation of the Fgf4-Fgfr-Fgfr-ERK pathway. Further examination of the mechanisms involved in the transcription of Klf5 and the transcriptional activity of Klf5 protein could reveal how the bimodal expression of Fgf4 is generated.

Given an elevation in Gata6 protein and Sox17 mRNA expression in Klf5 KO embryos at E3.0 or E3.25 (Fig. 1B, Fig. 3B, Fig. 4D), this might suggest a cell-autonomous role for Klf5 in the PrE lineage. However, the mRNA expression level of Sox17 at E3.0 was much lower than that of other transcription factors such as Nanog, Sox2 and Gata6 (Sox17/Nanog ratio=5.6×10⁻⁶, Sox17/Sox2=5.28×10⁻⁵, Sox17/Gata6=0.00125), suggesting that Sox17 mRNA expression is very low. Furthermore, we detected no Sox17 protein expression (Fig. S3A). This is consistent with previous reports that Sox17 is activated between the 32- and 64-cell stages in mouse embryos (Artus et al., 2011; Morris et al., 2010; Niakan et al., 2010). Thus, the elevation in Sox17 mRNA expression in Klf5 KO ICM is still too low to cause PrE differentiation. The Gata6 protein expression level also showed a slight increase (29%) at E3.25 (Fig. 1B), but the mRNA expression level of Gata6 was not changed significantly between WT and Klf5 KO embryos at E3.0 (Fig. S5B) and E3.25 (Fig. 4D), indicating that Gata6 is not regulated by Klf5 at the transcriptional level at ∼E3.0-3.25. Taken together, these data suggest that Fgf4 regulates the induction of PrE differentiation, rather than there being a Sox17- or Gata6-mediated cell-autonomous mechanism.

Previous studies have demonstrated that Klf5 regulates lineage specification in TE and ICM (Ema et al., 2008; Lin et al., 2010). While Klf5 directly regulates Sox17 (Lin et al., 2010), it is still unclear how Klf5 regulates the balance of EPI and PrE. Our results indicate that Klf5 is required to suppress Fgf4 in morula at E3.0, but it is not clear whether it continues to suppress Fgf4 later in development because Klf5 KO ESCs expressed normal levels of Fgf4 (Fig. S5E). Fgf4 and Klf5 are expressed abundantly in such cells and are important for their differentiation and for self-renewal, respectively (Ema et al., 2008; Kunath et al., 2007). Cis-regulatory regions in the Fgf4 promoter and enhancers are occupied by Klf5 in mouse ESCs (Fig. 4F). Nevertheless, the lack of Klf5 expression did not significantly alter Fgf4 expression (Fig. S5E). Therefore, we speculate that the transcriptional repression of Fgf4 by Klf5 occurs within a brief developmental window, such that the induction of Fgf4 does not hamper the normal segregation of EPI and PrE lineages, and later Klf5 expression does not repress Fgf4 by the time the mature EPI cells arise. This is consistent with the observation that aggregation of Klf5 KO ESCs with WT tetraploid embryos generates Klf5 KO embryos that appear normal at E8.5, indicating that Klf5 is not required for normal development once the EPI is established (Ema et al., 2008).

Activation of the Fgf4-Fgfr-ERK pathway destabilises a naïve pluripotent state in mouse ESCs and promotes a primed state, whereas a reduction in ERK activity strongly promotes naïve pluripotency (Hamilton et al., 2013; Kunath et al., 2007; Ying et al., 2008). However, the precise molecular mechanisms of self-renewal promoted by ERK inhibition remain elusive. Although ERK inhibition contributes to self-renewal in part through the stabilisation of Klf2 and Klf4 proteins, which are subject to proteasome-dependent degradation of ERK-phosphorylated forms in mouse ESCs (Kim et al., 2012; Yeo et al., 2014), it is interesting to note that Klf5 modulates the level of pERK in mouse ESCs (T.A., unpublished observations). ERK inhibition also facilitates the emergence of naïve pluripotent cells in the blastocyst during murine development (Nichols et al., 2009). It remains unclear whether there is a specific physiological mechanism that mediates ERK inhibition to promote pluripotency in vivo, but Klf5 could be a key genetic component in this regard. Further investigation of the functions of Klf5 might allow us to understand how the symmetric expression of Fgf4 and Fgfr2 is altered during early development.

Klf5^lacZ/lacZ mice were generated as previously described (Ema et al., 2008). The lacZ cassette was inserted into the second exon of the Klf5 gene. Conditional KO and OE alleles for Klf5 were generated as described in the supplementary Materials and Methods. Primers used for the genotyping of the conditional KO and OE alleles for Klf5 mice are described in Table S1. Mouse embryos were recovered at noon of the day on which the vaginal plug was discovered (considered E0.5).

Mouse Klf5^+/+ (WT)::Oct3/4-ireszeocinR ESCs, Klf5^lacZ/lacZ (KO)::Oct3/4-ireszeocinR ESCs and Klf5^+/+ (WT)::Oct3/4-ireszeocinR::Klf5 OE ESCs were generated as previously described (Ema et al., 2008) and were cultured in DMEM+15% Knockout Serum Replacement (KSR; Invitrogen). Details are provided in the supplementary Materials and Methods.

Embryo manipulations were performed according to Nagy et al. (2003). For immunosurgery followed by ESC derivation, blastocysts were incubated with rabbit anti-mouse red blood cell antibody (Inter Cell Technologies, A3840; 1:8) for 10 min. After the blastocysts were briefly washed twice in M2, they were incubated with guinea pig serum (Calbiochem) for 15 min. After removal of the zona pellucidae with acidic Tyrode's solution (Sigma-Aldrich), the cells were cultured in ESC medium for 2 weeks on a gelatine-coated dish. For in vitro culture of early mouse embryos, embryos at the 2-cell stage or later were incubated in KSOM (Millipore) in the presence or absence of PD0325901 (1 µM; Wako), SU5402 (2 µM; Wako), CHIR99021 (Chiron; 3 µM; Wako), JNK inhibitor II (5 µM; Calbiochem), SB203508 (10 µM; Calbiochem) or LY924002 (10 µM; Calbiochem). To activate the Fgf4-Fgfr-ERK pathway, recombinant human FGF4 (R&D Systems) was added at a saturated concentration (1 μg/ml) prepared in KSOM. 1 µg/ml heparin (Sigma) was added together with FGF4 or control. BrdU was added (10 µM; BD Pharmigen) for 2 h, and BrdU Flow Kits (BD Pharmingen) were used for detection.

Embryos were fixed in 4% paraformaldehyde (PFA) in PBS for 15 min, permeabilised in 0.5% Triton X-100 for 15 min and incubated in blocking reagent (PBS, 10% donkey serum, 0.1% BSA, 0.01% Tween 20) for 1 h. Embryos were incubated at 4°C overnight with primary antibodies prepared in blocking reagent. After the embryos were washed with PBS+0.5% Triton X-100, they were incubated with secondary antibodies in the blocking reagent for 3 h at 4°C. Nuclei were stained with Hoechst 33342 (10 µg/ml; Molecular Probes). Antibodies used for the immunohistochemistry are described in Table S2. Immunohistochemistry with anti-FGF4 antibody was performed as described in the supplementary Materials and Methods.

Embryos were mounted in drops of 30% glycerol on glass-bottom dishes. Confocal images were acquired using a Leica TCS SP5 or SP8 camera. Fluorescence was excited with a 405 nm UV laser for Hoechst 33342, a 638 nm laser for Cy5 or Alexa Fluor 633, a 552 nm laser for Cy3, and a 488 nm laser for Alexa Fluor 488. Images were acquired using an HC PL APO CS2 40×/1.30 oil-immersion objective lens (Leica), with optical sections of 2-2.5 µm. A hybrid detector system (Leica) was used for the acquisition of raw images, which were processed using Leica software or Photoshop CS6 (Adobe). Cell nuclei were counted manually using ImageJ image analysis software (National Institutes of Health). Protein expression levels were analysed as described (Dietrich and Hiiragi, 2007; Kang et al., 2013). Briefly, mean fluorescence intensities inside regions of interest (for example, nuclei) were measured, and subtracted from background signals, which were defined as the average of the mean fluorescence intensities of randomly chosen cytoplasmic signals, and were then normalised against the mean fluorescence intensity in the Hoechst channel using ImageJ (Dietrich and Hiiragi, 2007). We defined a cell as positive if it showed a higher fluorescence signal than the background. To quantify Nanog and Gata6 protein expression levels, we measured Nanog expression in both Nanog⁺/Gata6⁻ and Nanog⁺/Gata6⁺ (DP) cells, and Gata6 expression in both Gata6⁺/Nanog⁻ and DP cells.

For the quantification of FGF4 expression in preimplantation embryos, background signals were subtracted from mean fluorescence intensities inside regions of interest in the cytoplasm of FGF4⁺ cells. Background signals were defined as the average of the mean fluorescence intensities of randomly chosen cytoplasmic regions in without-antibody negative controls. FGF4 expression values were then normalised against the mean fluorescence intensity in the Hoechst channel using ImageJ. Individual cells were distinguished using intercellular gaps seen in differential interference contrast (DIC) images from the same focal plane as the corresponding confocal microscopy image.

Apoptotic cells were identified using the DeadEnd Fluorometric TUNEL System (Promega). For counting TUNEL⁺ cells, individual cells were identified using the intercellular gaps seen in DIC images merged with images of nuclear staining and TUNEL staining. Fragmented or pyknotic nuclei bounded by the same intercellular gap were counted as one cell.

ChIP assays were performed as previously described (Ito et al., 2013). Cells were fixed with 1% formaldehyde and sonicated. The samples were incubated with normal mouse IgG and anti-FLAG-M2 antibody. For details, see the supplementary Materials and Methods. Primers used for ChIP-qPCR are listed in Table S3.

To locate Fgf4 loci occupied by Klf5 in mouse ESCs, ChIP-seq data of Aksoy et al. (2014) were analysed as described in the supplementary Materials and Methods.

cDNAs were synthesised from individual WT and Klf5 KO embryos at E3.0 as previously described (Kurimoto et al., 2007; Nakamura et al., 2015). The cDNAs were amplified further in a linear fashion and labelled with a Cy3- or Cy5-conjugated nucleotide. Hybridisation procedures were performed by TaKaRa Bio. Raw data were analysed using GeneSpring software version 13.0 (Agilent Technologies). The raw probe intensities were background subtracted. Signal values were set to threshold level 10 and log2 transformed. Normalization was performed using a 75th percentile shift algorithm. Further normalization using baseline to median of all samples algorithm was performed. Transcripts were filtered with 2-fold expression change compared with the median intensity in all samples.

Outer cells of E3.25 embryos were removed by immunosurgery. Inner cells were incubated in a 1:1 mixture of Accutase (Nakalai Tesque) and 0.25% trypsin-EDTA (Invitrogen) for ∼5 min at 37°C and then dissociated into single cells by pipetting. cDNAs were synthesised from the isolated single cell using the single-cell mRNA 3-prime sequencing (SC3-seq) method as previously described (Nakamura et al., 2015).

First-strand cDNA was synthesised from total RNA using the QuantiTect Reverse Transcription Kit (Qiagen). Real-time PCR was performed with SYBR Premix Ex Taq II (TaKaRa) and analysed on a Thermal Cycler Dice Real-Time System (TP850; TaKaRa). The amount of target RNA was estimated using an appropriate standard curve and divided by the estimated amount of β-actin for normalisation. Primers used for RT-qPCR are listed in Table S3.

Statistical analyses were performed using the nonparametric Mann–Whitney U-test or Student's t-test. Data are expressed as the mean and s.e. Differences were considered significant at P<0.05. Statistical analyses were performed using Prism6 (GraphPad) for the nonparametric Mann–Whitney U-test and Excel (Microsoft) for Student's t-test. Single-cell qPCR data analysis was performed using R software with gplots (ver. 3.0.1) and ggplot2 (ver. 2.2.1) and Excel (Microsoft).

We thank Drs Hitoshi Niwa, Yojiro Yamanaka and Tomoyuki Tsukiyama for helpful discussions and reagents. M.E. thanks Dr Vincent Kelly for critical reading of the manuscript.