LIF supports primitive endoderm expansion during pre-implantation development

Embryonic stem cells (ESCs) are karyotypically normal self-renewing cell lines derived from the mammalian embryo during pre-implantation development. They are typically considered to be pluripotent, able to give rise to all lineages of the future embryo when reintroduced into an embryo or differentiated in vitro (Evans and Kaufman, 1981; Martin, 1981; Smith et al., 1988; Williams et al., 1988; Beddington and Robertson, 1989). ESCs are derived from the mammalian blastocyst at the stage when the inner cell mass (ICM) contains a heterogeneous mix of progenitors of two lineages, the epiblast (Epi) that will give rise to the embryo proper, and the primitive endoderm (PrE) that will give rise to the extraembryonic visceral and parietal endoderm (Chazaud et al., 2006). Like the ICM, ESC cultures are heterogeneous and contain populations primed towards both Epi and PrE (Chambers et al., 2007; Singh et al., 2007; Hayashi et al., 2008; Toyooka et al., 2008; Kobayashi et al., 2009; Canham et al., 2010). However, in ESC cultures these progenitors are in a dynamic equilibrium (Canham et al., 2010; Morgani et al., 2013; Morgani and Brickman, 2014; Posfai et al., 2014).

The in vitro culture of ESCs can be supported by a number of factors, but typically require Leukaemia inhibitory factor (LIF) (Smith et al., 1988; Williams et al., 1988; Ying et al., 2003, 2008). In defined culture conditions, containing small molecule inhibitors of GSK3 and MAPK (2i) (Ying et al., 2008), LIF is required for efficient clonal expansion. However, when LIF is removed from 2i, ESCs do not differentiate. Hence, it is a suitable cell culture system to study the immediate early effects of LIF in the absence of differentiation-related gene expression changes. In ESCs, LIF binds to a heterodimeric receptor complex composed of the LIF receptor (LIFR) and Glycoprotein 130 (GP130). Binding to the receptor complex results in activation of JAK tyrosine kinases that, in turn, phosphorylate the transcription factor STAT3. This results in STAT3 dimerization and translocation to the nucleus where it can activate target genes (Hirai et al., 2011). Whereas there are other signalling pathways downstream of LIF (Burdon et al., 2002; Liu et al., 2007; Annerén, 2008), activation of STAT3 is necessary and sufficient to support ESC self-renewal (Niwa et al., 1998; Matsuda et al., 1999). A second cytokine that acts through GP130 to activate JAK/STAT, IL-6, has also been implicated in pre-implantation development (Do et al., 2013) and can replace LIF as a means to sustain ESC self-renewal (Yoshida et al., 1994).

Although LIF supports ESC self-renewal, under standard conditions embryonic mutations of LIF pathway components do not show pre-implantation phenotypes (Li et al., 1995; Ware et al., 1995; Takeda et al., 1997; Nichols et al., 2001; Do et al., 2013). However, certain mammals have the ability to arrest development reversibly at the blastocyst stage as a result of suboptimal conditions, such as nutrient scarcity, a phenomena known as diapause. Development can subsequently resume under more favourable conditions. Diapause can therefore be seen as a developmental mechanism to support sustained pluripotency, and the induction of diapause can be used to improve ESC derivation (Nichols et al., 2001). When diapause is induced, gp130^−/− embryos exhibit a complete loss of the pluripotent Epi population (Nichols et al., 2001), indicating that LIF might play a role in vivo in maintaining pluripotency, but that it is not required during normal embryonic development. However, it is also possible that JAK/STAT signalling is activated via a GP130-independent maternal mechanism during early development that masks a pre-implantation phenotype. Consistent with this, inhibiting downstream signalling, either through pharmacological inhibition of JAK or in embryos in which both maternal and zygotic Stat3 have been removed, suggested that ICM maintenance requires active JAK/STAT signalling (Do et al., 2013).

Here, we show that LIF acts through a two-pronged mechanism to support ICM expansion, both in vivo and in vitro, in ESCs. LIF acts to suppress Epi differentiation and also to support PrE expansion. We had previously demonstrated that LIF promotes the expansion of an extraembryonic-primed population of cells within ESC cultures (Morgani et al., 2013), consistent with its well-characterised role in extraembryonic development and implantation (Stewart et al., 1992; Takahashi et al., 2003; Poehlmann et al., 2005; Prakash et al., 2011). We show that LIF supports this population via JAK/STAT signalling, the same pathway that promotes ESC self-renewal (Niwa et al., 1998; Matsuda et al., 1999). We find that STAT3 phosphorylation correlates with a PrE rather than Epi identity in vivo, and that LIF stimulates PrE survival and proliferation, whereas inhibition of JAK activity leads to enhanced apoptosis in the PrE. We hypothesise that LIF maintains ESC self-renewal and ICM expansion by enhancing the survival of an extraembryonic-primed population that provides paracrine factors for Epi expansion.

We have previously shown that LIF promotes a population of ESCs primed towards an extraembryonic fate (Morgani et al., 2013), marked by the expression of a highly sensitive fluorescent reporter for the endoderm marker Hhex [Hhex-Venus (HV)] (Canham et al., 2010). We tested whether this was a general property of signalling through GP130, as other members of the same cytokine family, such as IL-6, have also been associated with self-renewal (Conover et al., 1993; Rose et al., 1994; Wolf et al., 1994; Yoshida et al., 1994; Pennica et al., 1995). Fig. 1A shows that not only can IL-6 induce HV expression but also, at a high dose, it induces elevated levels of this transgene compared with LIF treatment. HV ESCs were cultured in 2i, to prevent differentiation in the absence of LIF, with increasing doses of IL-6. Expression of the HV reporter was assessed by flow cytometry (Fig. 1A). As previously reported (Morgani et al., 2013), we observed HV induction in 2i/LIF and obtained similar levels of HV expression with a low dose of IL-6 (500 ng/ml, Fig. 1A). At 1000 units, the effect of LIF on the HV population becomes saturated (Morgani et al., 2013), but HV expression can be further increased with high doses of IL-6 (1000 ng/ml, Fig. 1A).

LIF was previously shown to alter the relative number of actively proliferating cells within the embryonic and extraembryonic-primed ESC populations (Morgani et al., 2013). Consistent with this, when LIF was added to ESCs cultured in 2i, HV expression gradually increased over several days (Fig. S1A). This observation was confirmed by qRT-PCR, with PrE genes exhibiting a delayed response to LIF, including HHex, Hnf4a, Dab2 and Sox17, contrasting with the rapid response (within 1 h) of the immediate early target of LIF signalling, Socs3, or the sharp threshold responses of LIF transcriptional components, Stat3 and Klf4, at 8-24 h (Fig. 1B). Although Gata4 demonstrated an early response to LIF, as with other PrE markers, the expression change was gradual (Fig. 1B). Previous microarray data showed that LIF also increased the expression of extracellular matrix (ECM) genes associated with PrE development (Morgani et al., 2013). Here, we observed that these genes also increased in expression at 8-24 h (Fig. 1B; Fig. S1B). Interestingly, we observed that mesoderm and ectoderm markers, including Lhx1, Cdx2, Wnt3, Otx2 and Fgf5, were rapidly downregulated between 1 and 8 h following LIF addition, prior to changes in PrE gene expression (Fig. 1B; Fig. S1B).

As LIF acts on multiple downstream pathways (Burdon et al., 2002; Liu et al., 2007; Annerén, 2008), we tested which pathway was required for HV induction. We found that JAK/STAT signalling, also crucial for ESC self-renewal, supported the expansion of the HV PrE-primed population. HV ESCs were cultured in 2i or 2i/LIF medium in the presence of small molecule inhibitors of the PI3K (LY) and JAK/STAT (JAKi) pathways. After 3 days of culture in these conditions, HV ESCs were analysed by flow cytometry. Induction of HV was blocked by inhibition of JAK/STAT, but not PI3K (Fig. 1C). Although GP130 can also signal through activation of MAPK, all cells were cultured in 2i, containing a block to MEK, and hence MAPK signalling was not required for the induction of HV expression in response to LIF.

We found that LIFR was expressed heterogeneously within ESC cultures (Fig. 1D) and at higher levels in the HV-positive (HV⁺) compared with the HV-negative (HV⁻) population (Fig. 1E). This suggests that the pathway is differentially activated in PrE and Epi progenitors. Antibody staining of ESCs for the LIF targets (Bourillot et al., 2009; Niwa et al., 2009; van Oosten et al., 2012) phosphorylated STAT3 (pSTAT3) and KLF4 suggested that, although there is some overlap with expression of the Epi marker NANOG, this could not account for the all of the KLF4⁺ or pSTAT3⁺ cells. KLF4 expression was heterogeneous and showed some degree of co-expression with both NANOG and HV (Fig. 1F). Owing to a lack of appropriate antibody combinations, we could not co-stain for pSTAT3 and HV. However, although there was some co-expression of pSTAT3 and NANOG, there were also cells that expressed high levels of NANOG but low levels of pSTAT3 and, vice versa, a fraction of cells that only expressed pSTAT3 (Fig. 1G), probably corresponding to the HV population.

We assessed the localisation of pSTAT3 and its target KLF4, throughout pre-implantation development and also quantified their expression within individual cells using CellProfiler (Carpenter et al., 2006; Kamentsky et al., 2011) (Figs 2 and 3). We validated this method by quantifying immunostaining of the three well-characterised lineage markers NANOG (Epi), GATA6 (PrE) and CDX2 (trophoblast) (Fig. S2). Although there are low levels of NANOG expression at the 2-cell stage (Figs 2 and 3; Fig. S2), approximately tenfold lower than in the late blastocyst, no obvious nuclear staining was observed, suggesting that this is likely to be a non-specific signal. Both inbred C57BL/6 and outbred CD1 mouse lines were analysed but, as results were comparable between different genetic backgrounds, data were combined (see Materials and Methods). The analysis in Fig. S2B shows the progressive segregation of these three lineages and suggests some new correlations; for example, at the 8-cell stage all NANOG⁺ cells are expressing CDX2, but not the reverse.

Strikingly, although there was an early correlation between NANOG and pSTAT3, this was lost by the late-blastocyst stage (Fig. 2) after the PrE and Epi had segregated, prior to Epi differentiation. This contrasts with the overlap of NANOG with pSTAT3 that we observed in ESCs (Fig. 1G). As in ESCs, there was a correlation between KLF4 and NANOG, although KLF4 was also expressed at low levels within the PrE (Fig. 3).

pSTAT3 is first observed in some, but not all, 2-cell embryos, but was present in all cells of the 8-cell morula (Fig. 2A,B). pSTAT3 was also observed in all cells until the early blastocyst stage [embryonic day (E) 3.5], when it became heterogeneous within the ICM and in some trophoblast cells (Fig. 2A). At this stage, pSTAT3 showed a strong correlation with both GATA6 and NANOG expression (Fig. 2A,B). By the late blastocyst stage (E4.5), pSTAT3 and GATA6 remained correlated, and there was now a significant negative correlation between NANOG and pSTAT3 (Fig. 2B; *P=0.04). pSTAT3 was expressed almost exclusively in the extraembryonic PrE and trophoblast, and only at low levels within the Epi (Fig. 2A,B).

KLF4 was also expressed from the 2-cell stage and continued to be expressed in all cells until the early blastocyst stage (E3.5) (Fig. 3A,B). KLF4, as in ESCs, was then heterogeneously expressed, overlapping with both the NANOG⁺ and GATA6⁺ populations (Fig. 3A,B). In the late blastocyst (E4.5), KLF4 was expressed both in the Epi and PrE, although at a lower level in the PrE (Fig. 3A,B). As it is highly expressed in the Epi, KLF4 appeared correlated with NANOG expression at this late blastocyst stage. We also quantified the correlation between pSTAT3 and KLF4 expression and observed that, although at the early blastocyst stage there is a correlation between these markers, this decreases by the late blastocyst (Fig. S3A,B).

As we observed that the JAK/STAT pathway promoted an extraembryonic fate in ESCs, and that pSTAT3 was associated with the extraembryonic lineages in vivo, we tested whether LIF could regulate lineage choice in vivo. Embryos were cultured from E2.5 for 3 days, until the late blastocyst (equivalent of E4.5 in vivo), the time period that the decision between Epi and PrE fates is being made, in the presence or absence of LIF. As shown in Fig. 4, LIF produces an apparent increase in the fraction of GATA6⁺ PrE, but has little effect on NANOG.

Although the observation shown in Fig. 4 is robust, there is a range of phenotypes, and we aimed to quantify the impact of LIF on PrE versus Epi specification. Based on counting cells in control ICMs, i.e. those cultured in only KSOM, we observed a fraction of embryos that had a significantly higher than average proportion of either GATA6⁺ or NANOG⁺ cells within the ICM (Fig. S4A,B; P=0.0025 and 0.0003, respectively). As we observed a variation in the number of GATA6⁺ and NANOG⁺ cells within ICMs, we assigned embryos as ‘GATA6^Hi’ and ‘NANOG^Hi’ if the average proportion of GATA6⁺ or NANOG⁺ ICM cells was outside of one standard deviation (s.d.) of control embryos (Fig. S4A,B and Fig. S5A). Conversely, ‘normal’ embryos were within 1 s.d. of the proportion of GATA6⁺ or NANOG⁺ ICM cells in controls (Fig. S4A,B and Fig. S5A). There was no significant difference in the total number of cells in embryos from each of these categories, indicating that this variation was not due to disparity of developmental stage (Fig. S4C). When embryos were cultured in the presence of LIF from E2.5 until the late blastocyst, we observed a corresponding increase in the level of pSTAT3 and in both the PrE and Epi (Fig. S5B). However, pSTAT3 continued to be present at levels approximately threefold higher in the PrE than in the Epi (Fig. S5B). KLF4 also showed an increase in expression in both Epi and PrE, and was now expressed at similar levels in both cell types (Fig. S5B). We also found that the fraction of GATA6^Hi embryos more than doubled, from 15% to ∼40% (Fig. 4A,B; Fig. S5C-E). Similar observations were made when scoring based on GATA4 staining (Fig. S5F,G). Additionally, at higher doses of LIF, we observed a fraction of embryos that had only GATA6⁺ cells in the ICM (‘GATA6 only’, Fig. 4B; Fig. S5E). Treatment with LIF had no effect on the absolute number of trophoblast cells (Fig. S5H). We observed a similar increase in the proportion of GATA6⁺ cells when embryos were cultured from E1.5 or E3.5 until E4.5 (data not shown). In addition, we analysed LIF-treated embryos at earlier stages to determine how this phenotype developed over time. Following 1 day of culture in LIF, when embryos had reached the late morula stage, we observed no difference in the total number of cells in embryos (Fig. S5I). However, after 2 days of culture in LIF, at the early blastocyst stage, we observed a significant increase in both the total number of cells in the embryo (Fig. S5J) and the absolute number of GATA6⁺ PrE cells per embryo, but not in the number of NANOG⁺ Epi cells (Fig. S5K). Thus, the increase in cell number was a result of an increase in the number of GATA6⁺ PrE cells in response to LIF, and this was also represented by an increase in the proportion of GATA6^Hi embryos (Fig. 4C). Treatment with LIF also appeared to reduce the number of cells that had yet to commit to either lineage, as the population of cells co-expressing GATA6 and NANOG at these stages is reduced (Fig. 4C).

As we did not observe a decrease in the number of NANOG⁺ ICM cells (Fig. S5C), it seemed unlikely that LIF was directly acting on cell fate to switch ICM cells from Epi to PrE. Additionally, we had previously observed in ESCs (Morgani et al., 2013) that LIF supported expansion of the PrE-primed population by a selective increase in proliferation. We therefore tested whether LIF was regulating the balance of cell types in the embryo by altering proliferation or apoptosis in either of these lineages. Embryos were cultured from E2.5 to E4.5 in the presence or absence of LIF and were immunostained for NANOG, GATA6 and CDX2, as well as for a marker of dividing cells, phospho-Ser10 HISTONE H3 (pHH3). In the presence of LIF, a slight but significant increase in proliferation was observed in GATA6^Hi embryos (Fig. 4D; P=0.0492). To assess the level of apoptosis, we immunostained E4.5 embryos for the three lineage markers as well as for CLEAVED CASPASE-3 (CC3). In control, KSOM-only conditions, we observed a significant increase in the number of apoptotic cells in GATA6^Hi embryos (P=0.0081), which was decreased upon the addition of LIF (Fig. 4E,F). As the nuclear membrane is generally broken down in CC3⁺ cells, transcription factor expression is lost and it was not possible to assign lineages to apoptotic cells. However, when scored on physical location, we observed a clear and significant decrease in apoptotic cells or debris positioned adjacent to or inside the cavity (potentially originating from PrE cells) which is also reflected in the relative increase in the proportion of apoptotic cells in the inside of the ICM (Fig. 4F-H; P<0.0001). There was no significant affect on proliferation or apoptosis in NANOG^Hi embryos in the presence of LIF (Fig. 4D,E).

As we had observed that the ability of LIF to promote a PrE-like population in ESC cultures was mediated by JAK/STAT signalling (Fig. 1C), we used a pharmacological inhibitor of JAK activity (JAKi) to assess the consequences of blocking this pathway for lineage segregation in the ICM. It was recently shown that culturing pre-implantation embryos in JAKi caused a generic defect in ICM maintenance (Do et al., 2013), and we aimed to understand how this defect relates to PrE expansion. We therefore cultured embryos from E2.5 to E4.5 in control conditions (KSOM) or in KSOM with JAKi. We titrated the inhibitor in ESCs and observed that, at doses previously reported (5 μM) (Do et al., 2013), we observed widespread cell death after 2 days (Fig. S6A,B). However, we found that doses as low as 500 nM JAKi were sufficient to suppress STAT3 phosphorylation. In order to minimise toxicity to embryos, we cultured embryos at 500 nM and 100 nM. At 500 nM JAKi, we observed no pSTAT3 in embryos and a reduction in KLF4 expression (Fig. S6C,D). At doses of 100 and 500 nM JAKi, we observed a decrease in the number of NANOG⁺ cells and a slight decrease in the number of GATA6⁺ ICM cells (Fig. 5A). However, we observed obvious phenotypes in ∼50% of embryos, including embryos cavitated without an ICM, embryos with a cavity where the Epi should be located but still maintaining a GATA6⁺ PrE, and embryos that had ICM cells that expressed neither NANOG, GATA6 nor CDX2 (Fig. 5B,C; Fig. S6E,F). Although these cells no longer expressed NANOG, they maintained expression of OCT4 (Fig. S6G), suggesting that they represent differentiated Epi. In addition, we saw an increase in the proportion of embryos that had only GATA6 or only NANOG within the ICM (Fig. 5B; Fig. S6F). However, in embryos where there were only NANOG⁺ ICM cells, the number of ICM cells was also significantly decreased, indicating Epi defects (Fig. S6F). Little effect was observed on the number of trophoblast cells (Fig. S6H) and, although there was a slight decrease in total cell number in embryos without an ICM or with cavities, this was not significant (Fig. S6I). To assess the point when these phenotypes first become apparent, we examined embryos that had been cultured in JAKi for one or two days. We observed no change in the total numbers of cells at the late morula or early blastocyst stages (Fig. S6J,K), but observed an equal increase in the number of GATA6^Hi and NANOG^Hi embryos in the early blastocyst (Fig. S6L). As with the LIF-treated embryos, JAKi appeared to accelerate the resolution of Epi and PrE lineages, reducing the number of GATA6- and NANOG-co-expressing cells at this earlier stage.

Apoptosis in GATA6^Hi embryos was still detected and slightly increased as a result of JAK inhibition (Fig. 5D,E). Additionally, we observed a significant increase in the fraction of apoptotic cells adjacent to or inside the cavity (presumably originated from PrE cells) (Fig. 5F; P<0.05).

In this paper, we have shown that STAT3 phosphorylation, while initially throughout the pre-implantation embryo, becomes localized to the PrE in the late blastocyst. This suggests a role of JAK/STAT in PrE specification that is consistent with our observations that LIF supports extraembryonic gene expression in ESCs. As the induction of extraembryonic gene expression in vitro was a delayed response, it suggests that JAK/STAT acts to support a PrE population rather than directly regulating PrE gene expression. Thus, we observed that the activation of STAT3 could enhance the percentage of ESCs that experienced PrE priming and increase the proportion of PrE cells at blastocyst stages of development.

In addition to its action on PrE priming, we observed that LIF caused the rapid downregulation of mesoderm and ectoderm markers prior to the upregulation of PrE markers, consistent with the fact that the mesoderm marker Lhx1 and numerous Wnt pathway components are direct targets of STAT3 (Kidder et al., 2008). This is in keeping with previous observations that ESCs can differentiate to PrE and form basement membrane in the presence of LIF, but that ectoderm differentiation is blocked (Murray and Edgar, 2001). It has been previously suggested that MEK and GSK3 inhibitors in 2i/LIF lead to a downregulation of mesoderm and ectoderm markers, but not of PrE markers, compared with standard serum culture conditions (Marks et al., 2012). However, our findings suggest that this effect is not mediated by 2i alone but also by LIF. If LIF is able to suppress ectoderm and mesoderm differentiation, this might explain the ability of its primary target KLF4 to drive Epi stem cells back to a naïve mESC cell state (Guo et al., 2009). Thus, cells induced by JAKi in vivo that express neither GATA6 nor NANOG, but continue to express OCT4, might represent cells that have differentiated to later Epi derivatives. Taken together, our findings suggest that LIF acts in pre-implantation development to block Epi differentiation and support a PrE progenitor population.

We had previously shown that LIF enhances the number of actively proliferating PrE cells in vitro (Morgani et al., 2013), and have shown here that it does this through the JAK/STAT pathway. In vivo LIF had a modest effect on proliferation, which might translate into a more robust PrE phenotype over time. However, our exploration of the role of LIF in vivo also suggests that it acts to suppress apoptosis in a class of embryos with an elevated proportion of GATA6 cells. Similarly, LIF has been shown to suppress context-dependent apoptosis in ESCs. ESCs have the capacity to eliminate cells based on genotype as they embark on differentiation, a process mediated by competitive apoptosis that is suppressed via LIF (Sancho et al., 2013). In ESCs, the exit from self-renewal into differentiation triggers cell competition and associated apoptosis (de Beco et al., 2012; Levayer and Moreno, 2013; Wagstaff et al., 2013). Hence, selective apoptosis is perhaps triggered as a result of PrE differentiation, and this is suppressed by limiting concentrations of LIF.

The trophoblast expresses Lif during embryonic development, and Lifr and Gp130 are expressed within the ICM cells (Nichols et al., 1996). Recent single-cell microarray data also show that Lifr is expressed at higher levels within PrE cells than in Epi cells at E3.5 and E4.5 (Ohnishi et al., 2014), consistent with our data in HV ES cells. Moreover, these data reveal that Il-6 is expressed in the Epi, although data on Lif expression are not clear (Ohnishi et al., 2014). In addition, it has been shown that LIF receptors are enriched within extraembryonic endoderm (XEN) stem cell cultures (Artus et al., 2012). This suggests that the number of LIF/IL-6 producing cells, outside of the PrE, calibrate the size of the PrE. When there is insufficient PrE, excess LIF prevents apoptosis and stimulates proliferation. However, if the PrE expands beyond a certain point, such that the available amount of LIF is insufficient to support the size of the PrE, apoptosis ensues and the embryo recalibrates.

We also observed a requirement for JAK/STAT signalling in ICM formation. Similar observations have been made by Solter et al. (Do et al., 2013). Although we observed an effect of LIF on Epi differentiation in ESCs, we detected little change in pluripotent gene expression (Morgani et al., 2013). How, then, does JAK/STAT signalling support self-renewal of the ICM and ESCs? One of the principal GO terms to come out of our previous analysis of LIF stimulation (Morgani et al., 2013) was ‘ECM components commonly produced by the PrE’. Perhaps one mechanism by which LIF supports the expansion of the ICM is through the production of basement membrane required for its survival. It has been shown in ESCs that basement membrane is important to maintain ESCs in an undifferentiated state (Przybyla and Voldman, 2012). Additionally, the only way in which individual Epi cells can be shown to generate ESC lines in the presence of a pharmacological block to PrE specification (inhibition of MEK) is by the provision of specific ECM components (Boroviak et al., 2014). We also rarely observed embryos (3/98) in which the number of NANOG⁺ cells exceeded the number of GATA6⁺ cells, suggesting that a critical mass of PrE is required to support Epi expansion. Thus, LIF could have two roles in supporting Epi expansion, a block to later differentiation as well as a paracrine support network that depends on the PrE.

As we found that LIF blocks later Epi differentiation, it is not surprising that the correlation between pSTAT3 and NANOG is lost at the late blastocyst stage, prior to Epi maturation. In vivo, JAK/STAT signalling might only be necessary to block Epi differentiation for a limited time window when the PrE is actively being specified in response to the production of FGF4 by NANOG⁺ Epi cells (Messerschmidt and Kemler, 2010; Frankenberg et al., 2011). This differs from the activity of LIF in ESCs in vitro, in which we observed pSTAT3 in both NANOG⁺ Epi and HV⁺ PrE-primed ESCs. However, ESCs exist in a state of perpetual expansion in which these early developmental differentiation decisions are blocked. As a result, the requirement for pSTAT3 in both populations is prolonged indefinitely.

Taken together, our observations suggest that development of the three primary lineages is intricately and dynamically linked. The level of LIF secreted by the trophoblast and Epi (Nichols et al., 1996; Ohnishi et al., 2014) might specify the number of PrE cells within the ICM. High levels of PrE:Epi leads to apoptosis in the PrE, whereas too little PrE would lead to a failure in Epi expansion as a result of reduced basement membrane. We observed that accelerating PrE expansion through the manipulation of JAK/STAT signalling results in a loss of NANOG⁺GATA6⁺ cells in the ICM of early blastocysts. This suggests that these sorts of paracrine loops might have a selective role in the specification of both lineages from the point that differentiation begins in the first few cells. The combination of these two mechanisms would therefore be a means by which the embryo measures the relative levels of PrE and Epi, and ensures that both lineages are present and correctly specified throughout early embryonic expansion.

E14 and HV ESCs (129/Ola background) were used in this study. ESCs were maintained in serum/LIF or 2i medium (Morgani et al., 2013). Stat3^−/− ESCs (a kind gift from J. Nichols, Wellcome Trust-Medical Research Council Stem Cell Institute, University of Cambridge) were maintained in 2i or 2iLIF. ESCs were also cultured in the presence of the small molecule inhibitors JAKi (5 µM, Calbiochem), LY294002 (5 µM, Promega) and PD0325901 (1 µM, Sigma).

Cells were collected by trypsinisation and stained against a marker of undifferentiated ESCs, either Pecam-1 (BD Biosciences, APC-conjugated, 551262; 1:400) or SSEA-1 (DSHB, mc480; 1:1000), and DAPI (Molecular Probes, D1306, 100 ng/ml), to exclude dead cells (Canham et al., 2010; Morgani et al., 2013). The anti-LIFR antibody (R&D, MAB5990) was used at 2.5 μg/10⁶ cells, and ESCs were stained for 30 min at room temperature before being washed and resuspended in FACs buffer with DAPI as described for other antibodies. Flow cytometry analysis was carried out using a BD LSR Fortessa. Data analysis was carried out using FlowJo software (Tree Star) by gating on forward and side scatter to identify a cell population and eliminate debris, then gating DAPI⁻, viable cells before assessing the level of Venus or APC.

ESCs were cultured in 6-well plates on 25-mm glass coverslips coated with gelatin or in μ-slides (Ibidi). ESC immunostaining was carried out as previously described (Canham et al., 2010). For pStat3 staining, ESCs were permeabilised in methanol at −20°C for 5 min. The following antibodies were used (all at 1:200): anti-Nanog (eBioscience, 14-5761), anti-GFP Alexa 488-conjugated (Molecular Probes, A21311), anti-Klf4 (R&D, AF3158) and anti-pStat3 XP (Cell Signaling, 9145). Coverslips were mounted onto glass slides using Vectashield mounting medium (Vector Labs) and imaged by confocal microscopy using a Leica TCS SP8.

Both wild-type inbred C57/BL6 and outbred CD1 mouse strains were used for these experiments. Animal work was carried in accordance with European legislation and was authorized by and carried out under Project License 2012-15-2934-00743 issued by the Danish Regulatory Authority. Mice were maintained in a 12-h light/dark cycle in the designated facilities at the University of Copenhagen, Denmark. Natural matings were set up in the evening. Mice were checked for copulation plugs the following morning. Embryos were flushed in PB1 medium from oviducts between E1.5 and E2.5, and from the uterus at later stages. If culturing was needed, embryos were cultured in KSOM medium (Millipore), to which different concentrations of LIF or JAKi inhibitor (Calbiochem) were added. The majority of embryos were flushed at E2.5. These embryos were then cultured for 3 days to reach the equivalent of an E4.5 in vivo embryo. The resulting embryos were labelled as E4.5. Embryos were cultured in distinct microdrops for each condition, overlaid with embryo culture mineral oil (Sigma). Embryos were culture at 37°C, 5% CO₂ and 90% relative humidity.

Immunostaining of embryos was carried out as previously described (Nichols et al., 2009). When using antibodies against phospho-proteins, embryos were permeabilised in methanol at −20°C for 5 min. The following primary antibodies were used: anti-Nanog (eBioscience; 1:200), anti-Cdx2 (Biogenex, MU392A-UC; 1:200), anti-Gata6 XP (Cell Signaling, 5851; 1:200), anti-Gata6 (R&D, AF1700; 1:100), anti-pStat3 XP (Cell Signaling; 1:200), anti-Klf4 (R&D; 1:200), anti-phospho-Histone H3 (Ser10) (Abcam, ab14955; 1:200), anti-Cleaved Caspase-3 (Cell Signaling, 9664; 1:200) and anti-Gata4 (Santa Cruz, sc1237; 1:100). Embryos were imaged in an Attofluor chamber (Invitrogen) on a 25-mm coverslip using 20× magnification on a Leica TCS SP8 confocal microscope.

Confocal immunostaining of embryos was quantified using open access software, CellProfiler (developed at MIT and the Broad Institute, USA, www.cellprofiler.org). Individual cells stained by DAPI were identified using a manual selection tool. To determine pSTAT3 and KLF4 levels after treatments in PrE and Epi, GATA6 and NANOG staining, respectively, were used to identify nuclei. Entire E1.5 and E2.5 embryos were quantified, whereas two distinct confocal z-planes were quantified per blastocyst stage embryo. The mean pixel intensity (measured in arbitrary units) was measured in each selected nucleus for all fluorescence channels, allowing a measure of co-localisation of different markers. Results were plotted as scatter plots in GraphPad Prism software, and correlation statistics and linear regression analysis were carried out to generate a line of best fit. Both C57BL/6 and CD1 embryos were utilised for quantification. As there was no obvious difference in the results between these strains, data were combined. Interestingly, we observed that NANOG expression turns on in only a subset of C57BL/6 blastomeres at the 8-cell stage, whereas NANOG is present in all blastomeres of CD1 mice at the 8-cell stage.

The number of GATA6⁺ and NANOG⁺ ICM cells was counted in each embryo. Control (KSOM-cultured) and LIF-treated (KSOM+LIF-cultured) embryos were categorised as: 1, normal; 2, GATA6^Hi; 3, GATA6 alone; 4, NANOG^Hi; 5, NANOG alone. Embryos were categorised as ‘normal’ if the ratio of GATA6⁺:NANOG⁺ cells was within the average of control embryos ± one s.d. Embryos were scored as GATA6^Hi if the proportion of GATA6⁺ ICM cells was above this range, and as NANOG^Hi if the proportion of NANOG⁺ ICM cells was above this range. Embryos were scored as GATA6 or NANOG alone if the ICM consisted only of GATA6⁺ or NANOG⁺ cells. JAKi-treated embryos also exhibited distinct phenotypes, and were hence additionally categorised as 6. No ICM, 7; cavity (if there is an empty space where the Epi should be, still surrounded by PrE cells), 8. A question mark denotes cells if there are unidentified cells within the ICM expressing neither GATA6, NANOG nor CDX2. Both C57BL/6 and CD1 embryos were utilised for quantification. As there was no obvious difference in the results between these strains, data were combined.

Statistical analysis was performed using the QuickCalc GraphPad website (http://www.graphpad.com/quickcalcs/index.cfm). For non-parametric data, two-tailed chi-square tests were performed. For parametric data, unpaired Student's t-tests were used to determine significance. A P-value (P) is shown wherever the difference between compared groups was significant. As no apparent difference was observed, in some experiments with different LIF and JAKi concentrations, data were pooled for statistical analysis.

We thank the entire Brickman lab, Kim Sneppen and Ala Trusina for critical discussion of this manuscript; Javier Martin Gonzalez, Kasper Bonderup and Gelo de la Cruz for technical assistance; and Jennifer Nichols for reagents.