The role of FGF/Erk signaling in pluripotent cells

Fibroblast growth factors (FGFs) and their receptor tyrosine kinases control a multitude of developmental processes, including proliferation, survival, migration and differentiation. In this review, we discuss recently emerging evidence that FGF signaling plays an important role in regulating pluripotency and lineage segregation in both the early mouse embryo and in pluripotent mammalian stem cells. Although mouse and human embryonic stem (ES) cells are both derived from the pluripotent inner cell mass (ICM) cells of the pre-implantation blastocyst, they exhibit fundamental differences, whereby mouse ES cells appear to represent a developmentally more `naïve' pre-implantation state and human ES cells a `primed' post-implantation state (Nichols and Smith, 2009), similar to mouse epiblast-derived stem cells (EpiSCs) (Brons et al., 2007; Tesar et al., 2007). Intriguing recent findings suggest an important role for FGF/extracellular signal-regulated kinase (Erk) signaling in promoting the transition from a naïve to a primed state and in preventing primed cells from reverting back to a naïve state, in effect stabilizing the primed cell state (Hanna et al., 2009; Li et al., 2009; Greber et al., 2010; Hanna et al., 2010). Modulating FGF/Erk signaling might therefore facilitate the conversion of conventional human ES cells into a more naïve and mouse-like state. The existence of multiple and possibly interconvertible pluripotent stem cell states controlled by the extrinsic signaling environment could have significant implications not only for early developmental biology, but also for the field of reprogramming and induced pluripotency.

FGF signaling is activated by a ligand-receptor interaction that results in the autophosphorylation of tyrosine residues in the intracellular region of an FGF receptor (FGFR). The signal is further relayed through four distinct pathways: the Janus kinase/signal transducer and activator of transcription (Jak/Stat), phosphoinositide phospholipase C (PLCγ), phosphatidylinositol 3-kinase (PI3K) and Erk pathways (Dailey et al., 2005) (Fig. 1). The formation of a complex between FGFR, fibroblast growth factor receptor substrate 2 (Frs2α), Src homology region 2 domain-containing phosphatase 2 (Shp2; also known as Ptpn11) and growth factor receptor-bound protein 2 (Grb2) facilitates son of sevenless homology (Sos)-Ras-Erk activation, whereas FGFR-Frs2α-Grb2 complex formation drives PI3K activation via Grb2-associated binding protein 1 (Gab1) (see Fig. 1). FGF signaling is tightly regulated and modulated at multiple levels, both intracellularly and extracellularly. For example, the bioavailability of secreted FGF ligands is regulated by their binding to extracellular heparan sulfate proteoglycans (HSPGs). HSPGs also act as important co-receptors for FGF ligands by facilitating the assembly of activated ligand-receptor complexes (Ornitz and Itoh, 2001). Intracellularly, the FGF signaling cascade is further regulated by negative-feedback mechanisms at multiple levels by dual specificity phosphatases (DUSPs), similar expression to FGF (Sef), Spred and Sprouty proteins (Dailey et al., 2005). Several of these inhibitors are themselves downstream transcriptional targets of the FGF pathway.

Vertebrates have 22 FGF ligands and five receptors (Fgfr1-5), each with specific expression patterns throughout development. Alternative splicing of the receptors provides some degree of ligand specificity (Itoh and Ornitz, 2004). The importance of FGF signaling in the early mouse embryo is well established, as mutations in Fgf4, Fgfr2, Frs2a, Grb2 and Erk2 (Mapk1) all result in peri-implantation lethality (Feldman et al., 1995; Arman et al., 1998; Cheng et al., 1998; Hadari et al., 2001; Hatano et al., 2003; Saba-El-Leil et al., 2003; Chazaud et al., 2006) (summarized in Table 1). Fgf4 is the predominantly expressed ligand in the early embryo. It is expressed in the 8- to 16-cell morula but becomes restricted to the epiblast cells of the ICM in the expanded blastocyst, and has been shown to be under the direct regulation of the pluripotency factors Oct4 (Pou5f1) and Sox2 (Yuan et al., 1995). Fgfr2 appears to be the predominantly expressed receptor in the blastocyst, although it is restricted to the extra-embryonic, trophectoderm (TE) and primitive endoderm (PE) lineages. This differential receptor-ligand expression suggests that epiblast-produced Fgf4 could be important for the establishment or maintenance of the two extra-embryonic tissues. Although loss of Fgf5 does not result in any early phenotype, onset of its expression is often used to mark the transition from pre- to post-implantation epiblast (Hebert et al., 1991; Hebert et al., 1994). Fgf8 and Fgfr1 are also expressed in the blastocyst, but they appear to function slightly later as loss of their function results in post-implantation epiblast phenotypes with impaired axis formation and mesoderm specification (Deng et al., 1994; Yamaguchi and Rossant, 1995; Meyers et al., 1998).

Four distinct stem cell types can be derived from the early mouse embryo, representing the first cell types of mammalian development: trophectoderm stem (TS) cells from the TE; extra-embryonic endoderm stem (XEN) cells from the PE; ES cells and EpiSCs from the pre- and post-implantation epiblast, respectively (Box 1) (Evans and Kaufman, 1981; Martin, 1981; Tanaka et al., 1998; Kunath et al., 2005; Brons et al., 2007; Tesar et al., 2007). Reflecting the differential FGF ligand-receptor expression in the early embryo, these cell lines show differential dependency on exogenous FGF. TS cells and EpiSCs require exogenous FGF for their derivation and maintenance. Although FGF is routinely added to XEN cell cultures, it has not formally been tested whether this is crucial for their derivation. By contrast, mouse ES cells are derived and maintained without exogenous FGF. Although Fgf4 is produced by mouse ES cells and can activate FGF/Erk signaling in an autocrine manner, it is dispensable for the maintenance of the undifferentiated state of mouse ES cells. Indeed, Fgf4-null ES cells do not suffer from reduced proliferation, unlike the Fgf4-null ICM (Wilder et al., 1997). However, it was recently demonstrated that FGF/Erk signaling is required to instruct ES cells to exit from the self-renewal program and to begin differentiation (Kunath et al., 2007); inhibition of this signaling pathway actually increases the efficiency of ES cell

derivation (Kunath et al., 2007; Ying et al., 2008; Nichols et al., 2009a). As we discuss below, this is in stark contrast to human ES cells, which are strictly dependent upon exogenous FGFs for their self-renewal.

Until recently, the predominant view of mouse ES cells has been that their pluripotent state must be actively maintained by the addition of extrinsic signaling molecules. The factor that is best known for promoting the self-renewal of ES cells is leukemia inhibitory factor (LIF), which acts through the Jak/Stat3 signaling pathway (Smith et al., 1988; Williams et al., 1988; Ying et al., 2003). ES cells can be maintained in serum-free conditions when LIF is combined with bone morphogenetic protein 4 (Bmp4). Bmp4 induces the inhibitor-of-differentiation (Id) proteins, which are DNA-binding proteins that inhibit the default differentiation of ES cells into neuronal lineages (Tropepe et al., 2001; Ying et al., 2003). However, the view that differentiation is the default pathway for ES cells was challenged when two back-to-back papers showed that autocrine FGF-induced Erk1/2 signaling (Erk1 is also known as Mapk3) occurs in ES cells and is required to initiate their differentiation (Kunath et al., 2007; Stavridis et al., 2007).

Stavridis et al. set out to examine the role of FGF signaling in neural specification using mouse ES cells and the chick embryo (Stavridis et al., 2007). They first showed that FGF signaling activates downstream Erk1/2 and not PI3K or PLCγ during neuronal differentiation. Inhibition of either FGFR or Erk1/2 abolished neuronal induction. They could detect induction of the post-implantation epiblast marker Fgf5 but no reduction in Nanog, and concluded, therefore, that FGF signaling is required for ES cells to differentiate beyond the early epiblast state. Kunath et al. showed that although Fgf4-null ES cells behave normally in the undifferentiated state in the presence of LIF, they are unable to make the transition from pluripotency to neural lineage commitment when LIF is removed (Kunath et al., 2007). The addition of exogenous Fgf4 to Fgf4-null ES cells rescued their differentiation into not only neural, but also mesendoderm, lineages. By contrast, the addition of Fgf5 failed to rescue differentiation, suggesting that Fgf4 has a ligand-specific signaling potential. Similar results were observed in Erk2^–/– mouse ES cells (Kunath et al., 2007). These authors therefore concluded that autocrine Fgf4/Erk signaling is needed for ES cells to exit from self-renewal and initiate differentiation.

Thus, the role of LIF and BMP in promoting ES self-renewal may be to counterbalance the pro-differentiation autocrine Fgf4/Erk signal. Ying et al. hypothesized that the requirement for exogenous factors might be eliminated if autocrine Fgf4/Erk signaling were blocked (Ying et al., 2008). They found that addition of the FGFR inhibitor SU5402 or the Erk signaling inhibitor PD184352 could maintain mouse ES cells in an undifferentiated state in the presence of LIF. These inhibitors could also block differentiation in the absence of LIF, but the cultures progressively degenerated. 6-bromoindirubin-3′-oxime (BIO), a glycogen synthase kinase 3 (GSK3) inhibitor, had previously been shown to promote the maintenance in culture of both mouse and human ES cells (Sato et al., 2004). Using a GSK3 inhibitor (CHIR99021) in combination with FGFR/Erk inhibition, Ying et al. could eliminate the need for LIF while maintaining the pluripotency and clonogenicity of ES cells (Ying et al., 2008). In vivo, GSK3 is inactivated by the canonical Wnt signaling pathway, resulting in the nuclear accumulation of β-catenin. One of the effects of adding a GSK3 inhibitor would therefore be to mimic canonical Wnt signaling. Inhibition with CHIR99021 did reduce β-catenin phosphorylation and induced a T-cell factor/lymphocyte enhancer factor (TCF/Lef)-responsive promoter, both indicating that the activation of downstream canonical Wnt signaling had occurred. However, addition of recombinant Wnt3 together with FGF/Erk inhibitors failed to fully recapitulate the effect of the GSK3 inhibitor. Furthermore, ES cells that stably express ΔNhLef1, a dominant-negative form of lymphoid enhancer binding factor 1 (Lef1; which mediates a transcriptional response to canonical Wnt signaling together with TCF1), which should attenuate canonical Wnt signaling, still responded to GSK3 inhibition. As a result, Ying et al. suggested that GSK3 inhibition might have a Wnt-independent function in promoting the overall viability and proliferation of mouse ES cells. Together, these data show that mouse ES cells can be cultured independently of LIF if the cells are shielded from autocrine FGF/Erk signaling and subjected to GSK3 inhibition.

The derivation of truly pluripotent ES cells from mammalian species other than mice has been very difficult to achieve. This has been attributed to the inability of most other early embryos to respond to LIF. Ying and colleagues subsequently showed that germline-competent rat ES cells could be successfully generated through the combined use of LIF with FGF/Erk and GSK3 inhibition (Buehr et al., 2008; Li et al., 2008). This approach has also been used to derive germline-competent ES cells from non-permissive mouse strains, such as the non-obese-diabetic (NOD) strain (Hanna et al., 2009; Nichols et al., 2009b). Although LIF itself failed to sustain derived rat ES cells, it did induce Stat3 activation and improved colony formation from single cells when combined with Erk and GSK3 inhibitors (Buehr et al., 2008; Li et al., 2008). This finding suggests that the previous failure to derive ES cells from species other than mice might not solely be attributed to the absence of LIF responsiveness but to differences in endogenous FGF/Erk signaling.

In parallel with these recent findings regarding the important role of FGF/Erk signaling in balancing pluripotency versus differentiation in ES cells, several papers have presented convincing data that pluripotent ES cultures exhibit highly dynamic heterogeneous expression of pluripotency markers, such as Nanog (Chambers et al., 2007; Singh et al., 2007), platelet/endothelial cell adhesion molecule (Pecam1 or CD31) (Furusawa et al., 2004; Furusawa et al., 2006), zinc-finger protein 42 (Zfp42 or Rex1) (Toyooka et al., 2008) and developmental pluripotency-associated 3 (Dppa3 or Stella) (Payer et al., 2006; Hayashi et al., 2008), even though all cells are positive for Oct4 and stage-specific embryonic antigen 1 (SSEA1; Fut4). Furthermore, genes associated with differentiation, such as brachyury (Suzuki et al., 2006), Gata6 (Singh et al., 2007), Hex (Hhex) (Canham et al., 2010) and Sox17 (Niakan et al., 2010), are also expressed in a subset of cells in undifferentiated ES cell cultures.

Nanog is expressed in the mouse blastocyst ICM and in ES and germ cells. The Nanog null mutation causes embryonic lethality owing to an inability to form or maintain an epiblast in pre-implantation blastocysts. Conversely, forced expression of Nanog in ES cells maintains cells in an undifferentiated state even in the absence of LIF (Chambers et al., 2003; Mitsui et al., 2003). Chambers et al. reported that mouse ES cells could be maintained in the absence of Nanog, although they are more prone to differentiation (Chambers et al., 2007). They further showed that Nanog is heterogeneously expressed in normal Oct4⁺ ES cell cultures. A knock-in GFP reporter line revealed that ES cells fluctuate between a Nanog-high and Nanog-low state. Nanog-low cells appeared to be poised or `primed' for differentiation, similar to the Nanog-null ES cells. However, Nanog-low cells could still revert back to a Nanog-high state, and Nanog-high cells could regenerate a Nanog-low population.

Heterogeneous and fluctuating expression has also been reported for two additional transcription factors, Rex1 and Stella, which are specifically expressed in pluripotent ICM cells but not in post-implantation epiblast. Toyooka et al. reported that the Rex1-positive population in cultured ES cells is enriched for markers that are expressed in the ICM, such as Nanog, Tbx3, Klf4 and Tcl1, whereas the Rex1-negative population is enriched in post-implantation epiblast markers, such as Fgf5 and brachyury (Toyooka et al., 2008). Hayashi et al. reported that Stella expression undergoes a similar transition and that mouse ES cells cultured on feeder cells maintain more cells in a Stella-GFP-positive state, indicating that the balance between the different populations is influenced by external stimuli. In the case of all three markers, Rex, Stella and Nanog, the positive and negative population of cells, when isolated, can regenerate the heterogeneous expression of the starting population. This suggests that the ES cell population is truly in a metastable state in which a dynamic continuum exists between multiple interconvertible states. It is likely that Nanog-low, Rex1-low, Stella-low expression mark the same population of cells, although this has not been formally proven at the single-cell level. This suggests that ES cultures maintained in LIF undergo a form of lineage priming (Enver et al., 2009), constantly generating a population of cells poised to become post-implantation epiblast (Fig. 2).

The epiblast-primed population might not be the only primed population among ES cells. Singh et al. reported that a subpopulation of ES cells is Nanog-low and positive for Gata6, a PE marker (Singh et al., 2007). When the two populations were sorted by flow cytometry, each population could regenerate the initial heterogeneity. However, when the Nanog-low cells were differentiated, they only induced the expression of extra-embryonic endoderm markers, such as Gata6 and Sparc, whereas Nanog-high cells remained capable of also inducing mesodermal and endodermal transcripts. This suggests that the Nanog-low, Gata6-high cells were primed for extra-embryonic endodermal differentiation. Supporting the existence of a subpopulation that is primed to become PE, Canham et al. identified a subset of ES cells that expresses Hex transcripts and is reversibly primed for extra-embryonic endodermal differentiation (Canham et al., 2010). Furthermore, a Sox17-expressing subpopulation has been described in ES cell culture that appears to be irreversibly poised for differentiation to the extra-embryonic endodermal lineage (Niakan et al., 2010) (Fig. 2).

It is still not clear whether all ES cell lines under all culture conditions display the same range of lineage-primed states. Most studies have focused on one particular ES cell line and one lineage marker, making it difficult to compare results from different studies. However, it is clear that ES cultures can undergo fluctuations in gene expression that prime cells for at least two kinds of differentiation: PE and post-implantation epiblast.

It has been suggested that these fluctuating patterns of gene expression could result from cell-autonomous stochastic variations in the expression of pluripotency-associated genes that are driven by transcriptional noise (Enver et al., 2009; Kalmar et al., 2009). However, external signaling feedback loops could also explain the cyclical fluctuations in the expression of pluripotency genes that have been observed. Niwa et al. recently suggested an elegant model in which heterogeneous expression is maintained by LIF through opposing downstream signaling pathways (Niwa et al., 2009). The authors showed that the LIF-Jak/Stat3 pathway induces Klf4 expression, which preferentially activates Sox2 expression, whereas the LIF-PI3K pathway induces Tbx3, which they suggest activates Nanog expression. They also reported that LIF activates the Erk signaling cascade, which counteracts the PI3K signal and reduces Tbx3 expression, thereby also reducing Nanog levels.

FGF signaling also acts through the Erk pathway and, given that it is required for ES cells to exit from the pluripotent state, it could play a role in modulating the metastable states of these cells. Indeed, it has been shown that manipulating FGF signaling levels in ES cultures can bias the proportions of the different states. Canham et al. showed that the Hex-expressing, PE lineage-primed population can be reduced by blocking Erk signaling or increased by stimulating phosphorylation with sodium vanadate (Canham et al., 2010).

Extracellular HSPGs are key co-activators of FGF receptors. ES cells with targeted mutations in genes involved in the HSPG biosynthesis pathway have been shown to act similarly to Fgf4^–/– ES cells in being unable to exit from the pluripotent state and initiate differentiation (Johnson et al., 2007; Kraushaar et al., 2010; Lanner et al., 2010). Ndst1/2^–/– cells lack the enzymes (N-deacetylase/N-sulfotransferase 1) that sulfate HSPGs, whereas exostoses 1 and 2 (Ext1/2)-null cells fail to synthesize proteoglycan chains at all. Such HSPG-deficient ES cells are also enriched for the Nanog⁺ Klf4⁺ ES population versus the Nanog^– Fgf5⁺ lineage-primed epiblast-like ES cells. Treatment of wild-type cells with sodium chlorate, which blocks sulfation of proteoglycans, had the same effect as inhibition of Erk signaling in terms of enriching for the Nanog⁺ population (Lanner et al., 2010). Together, these data implicate the Fgf4-mediated activation of the Erk pathway in triggering the exit of ES cells from a self-renewing into a primed, but reversible, ES cell state. Microarray comparisons of wild-type ES cells with HSPG-deficient, sodium chlorate-treated or with FGFR inhibitor-treated ES cells have revealed the downregulation of Dusp4 and Dusp6, together with the Ras/Erk signaling inhibitor Spred1 (Lanner et al., 2010), when FGF signaling is inhibited. Active FGF signaling thus both promotes a primed state and induces the expression of inhibitors of FGF signaling to generate a negative-feedback loop that would facilitate reversion back to the original unprimed state. Thus, extracellular signaling feedback controls could drive the heterogeneous and potentially oscillating gene expression patterns that are observed in ES cells. In this way, a complex balance between multiple signaling pathways and their feedback mechanisms may underlie the heterogeneous nature of ES cell cultures (Fig. 2).

The complex involvement of FGF signaling in regulating the pluripotent versus differentiated state in ES cells has some parallels in the mouse embryo itself. As the blastocyst develops, the enclosed ICM cells all express the pluripotency marker Oct4, but as the blastocyst expands ICM cells show mosaic and progressively complementary expression of Nanog versus PE markers, such as Gata6 and Pdgfrα (Chazaud et al., 2006; Plusa et al., 2008). Using Pdgfra^H2B-GFP reporter mice, which express nuclear histone H2B-GFP, Plusa et al. went further and showed that ICM cells initially express both markers but randomly segregate into the two populations by the expanded blastocyst stage. Over the course of the next day, the Pdgfrα-expressing cells end up on the blastocoelic surface and form PE, whereas the Nanog-high cells form the epiblast progenitors. The correct location of cells involves both the migration of cells to the appropriate position and the death of cells that are in the wrong place (Plusa et al., 2008).

By embryonic day (E) 4.5, the two lineages are irreversibly committed to their future fate, as assessed by chimera analysis (Gardner and Rossant, 1979), but the earlier, apparently stochastic variation in Nanog expression in cells that are all Oct4 positive is very reminiscent of the heterogeneous Nanog expression observed in ES cells. Transcriptional microarray analysis of individual ICM cells has demonstrated that cells can be clustered into two groups that are enriched for many epiblast or PE markers, and not just for Nanog and Gata6 (Kurimoto et al., 2006). A recent study using single-cell quantitative PCR analysis of mouse embryos up to the 64-cell blastocyst stage elegantly provides further support to the mosaic segregation model, as it reported the existence of a temporally progressive inverse correlation between epiblast and PE marker expression (Guo et al., 2010).

Mice carrying mutations in components of the FGF/Erk pathway, including Fgf4, Fgfr2 and Grb2 (Feldman et al., 1995; Arman et al., 1998; Cheng et al., 1998; Chazaud et al., 2006), all have peri-implantation lethality and lack PE formation, strongly suggesting a crucial role for this pathway in lineage specification. Pharmacological inhibition of FGF signaling in mouse embryos cultured from the 8-cell to expanded blastocyst stage can completely eliminate the Gata6-positive PE cells (Yamanaka et al., 2010). Conversely, the addition of exogenous Fgf4 to the cultures shifts all ICM cells to adopt the PE fate. The number of ICM cells does not significantly change when embryos are exposed to FGF inhibitors or to exogenous FGF, indicating that there is no selective apoptosis driving this apparent lineage shift. Using the Pdgfra^H2B-GFP reporter mice, we observed the conversion of individual cells from either a GFP-positive or GFP-negative state in response to the inhibition or activation of the FGF/Erk signaling pathway (Yamanaka et al., 2010). Importantly, the fate of the cells could still be modulated after the mutually exclusive Nanog and Gata6 expression had been established at E3.5. The observed plasticity was, however, progressively lost by E4.0 (Fig. 3). Nichols et al. have reported similar results showing that the PE lineage is irreversibly lost when culturing embryos with FGF/Erk inhibition from E2.5-4.5 (Nichols et al., 2009a). Guo et al. also observed the elimination of PE transcripts (Gata4, Sox17) and the induction of epiblast transcripts (Nanog, Esrrb) when embryos were cultured in the presence of FGFR inhibitors (Guo et al., 2010).

Fgf4 is suggested to be produced by ICM cells under the direct transcriptional control of Oct4 and Sox2 (Yuan et al., 1995). Interestingly, Guo et al. detected a stronger and earlier inverse correlation between Fgf4 and Fgfr2 transcripts in ICM cells than between Nanog and Gata6, supporting a model in which FGF/Erk signaling variation is upstream of Nanog and Gata6 (Fig. 3). What is still unclear is how the initial variation in the levels of Fgf4 and Fgfr2 expression is established. This initial variation could be stochastic, as suggested for ES cells. However, unlike the ongoing heterogeneity in ES cells, there is only a brief window of time during which expression of lineage-specific genes such as Nanog and Gata6 is still labile and responsive to levels of FGF signaling, before progressive feedback loops lock in cell fate. It is also possible that the initial variation in Fgf4 and Fgfr2 expression among ICM cells could reflect the lineage of these cells during earlier cleavage.

Inside cells are generated through two distinct rounds of asymmetric cell divisions as the embryo divides from the 8- to 16-cell and from the 16- to 32-cell stages (Chisholm and Houliston, 1987; Chazaud et al., 2006). It is thus an attractive model to suggest that these two sets of inside cells could have intrinsic differences that bias them towards either the epiblast or PE fate. Two recent papers attempted to test this by lineage tracing the progeny of the primary and secondary inner cells. Using a transgenic mouse line expressing a membrane-linked GFP, Morris et al. traced cell fate from the 8-cell to the expanded blastocyst stage, when individual cells were scored as being epiblast or PE by morphology (Morris et al., 2010). They reported that primary inner cells are biased towards epiblast, although 50% still give rise to both epiblast and PE, whereas secondary inner cells are heavily biased towards PE. Using a different lineage-tracing method that allowed cell fate to be followed through to the established epiblast and PE lineages post-implantation, Yamanaka et al. could not detect a clear bias towards either lineage (Yamanaka et al., 2010). Although these two papers show apparently conflicting data, in neither case was there an absolute relationship between lineage and fate, indicating that additional mechanisms still influence the final lineage allocation.

The derivation of a new pluripotent stem cell type (the EpiSC) directly from the early post-implantation epiblast (Brons et al., 2007; Tesar et al., 2007) has shed new light on the evolving roles of FGF signaling in the transition from one pluripotent state to another. EpiSCs express the pluripotency markers Oct4 and Sox2 and can differentiate into derivatives of all germ layers in vitro. However, they cannot contribute to normal development in chimeras and they differ transcriptionally and epigenetically from ES cells. Importantly, they express markers of the post-implantation epiblast, such as Fgf5, and female cells are X-inactivated, unlike ES cells. Maintenance of these cells in the undifferentiated state in culture requires exogenous FGF and activin A, clearly distinguishing them from mouse ES cells. In the context of EpiSCs, FGF signaling has been demonstrated to act by inhibiting both forward differentiation to the neural lineage and reversion to a pre-implantation mouse ES cell-like state (Greber et al., 2010) (Fig. 3). Both these findings are surprising as neuronal differentiation is generally thought to be promoted by FGF signaling, and EpiSCs were proposed to have irreversibly lost the potential to revert back to an ES cell-like state. Results from Stavridis et al. support a neuronal suppression function of FGF in EpiSCs (Stavridis et al., 2010). They have shown that although FGF signaling is crucial for making the transition from ES cell to a post-implantation epiblast state, the subsequent transition to neurectoderm is inhibited by FGF signaling.

This dual effect ensures that the EpiSC state is more stable than that of the epiblast-like cells seen in normal metastable ES cultures. However, it is possible to convert EpiSCs back to an ES-like state. Guo et al. and Hall et al. reported that this reversion from EpiSCs to the ES-like state does not occur spontaneously but depends upon the forced expression of Klf2 or Klf4 (Guo et al., 2009; Hall et al., 2009). However, using the distal Oct4 promoter that is preferentially expressed in ES cells and not in post-implantation epiblast cells to drive GFP, Bao et al. reported a modest but spontaneous reversion of EpiSCs back to an ES cell-like state following their extended culture in media containing serum and LIF (Bao et al., 2009). These cells showed several hallmarks of true conversion, including X reactivation and the potential to contribute to somatic tissues and germ cells in chimeras. Greber et al. showed that this spontaneous conversion is greatly enhanced by combined FGF/Erk and GSK3β inhibition in LIF-containing media (Greber et al., 2010). Mechanistically, they showed that the switch in culture conditions induces Klf2 but not Klf4. Since Klfs appear to have overlapping functions, they suggest that FGF suppression of Klf2 normally prevents EpiSCs from spontaneously reverting to an ES cell-like state, which is in line with the data presented by Guo et al. (Guo et al., 2009) (Fig. 3). Hanna et al. also reported that EpiSCs can be converted to an ES cell phenotype by application of Erk and GSK3 inhibition, and demonstrated that this reverted state is stable in permissive mouse strains, such as 129, but not in non-permissive strains, such as NOD (Hanna et al., 2009). Removal of the inhibitors leads to a conversion back to an EpiSC state for NOD lines but not for 129 lines. Thus, the two pluripotent states – ES-like and EpiSC-like – may have an intrinsic potential for interconversion, but the relative stability of the two states may vary according to genetic background. In all cases, stable interconversion is still a relatively rare event, which suggests that there is still more to discover about the control of the different stem cell states. However, the modulation of FGF signaling is a key element in all cases.

Although human and mouse ES cells are derived from the ICM of the blastocyst and rely on similar transcriptional regulators, such as Nanog, Oct4 and Sox2, they require completely different culture conditions for their self-renewal in vitro. Indeed, although human ES (hES) cells show increased STAT3 phosphorylation in response to exogenous LIF, LIF addition does not seem to enhance self-renewal in culture (Thomson et al., 1998; Reubinoff et al., 2000; Sumi et al., 2004). Numerous protocols for hES cell culture have been reported, utilizing a variety of growth factors, such as FGF2, activin A, transforming growth factor β1 (TGFβ1) and Wnt. The common denominator in all these protocols is the use of exogenous FGF2. Indeed, the combination of FGF and activin signaling, which maintains mouse EpiSC self-renewal, is the most common and effective culture condition for the maintenance of hES cells as well. Comparison of gene expression profiles among hES cells, mouse ES cells and mouse EpiSCs showed that hES cells were more similar to mouse EpiSCs than to mouse ES cells (Brons et al., 2007; Tesar et al., 2007), which has led to the suggestion that the pluripotent cells captured from human blastocysts are actually more equivalent to the post-implantation-derived EpiSC than they are to the mouse ES cell (Brons et al., 2007; Tesar et al., 2007; Rossant, 2008).

The mode of action of FGF signaling in promoting hES self-renewal seems to be different from that by which FGF signaling blocks both differentiation and reversion to an ES-like state in mouse EpiSCs (Greber et al., 2007; Greber et al., 2010) (Fig. 4). In contrast to mouse ES cells, hES cells and EpiSCs are reported to exhibit a high degree of spontaneous differentiation into both extra-embryonic lineages (Xu et al., 2002; Brons et al., 2007; Tesar et al., 2007). It has been suggested that FGF is crucial for blocking such extra-embryonic differentiation. Inhibition of the FGF receptors or of downstream Erk signaling in hES cells induces the TE markers CDX1/2 and gonadotrophin-α/β and the PE marker SOX7 (D'Amour et al., 2005; Li et al., 2007). Furthermore, activated HRas robustly induces the extra-embryonic differentiation of ES cells (Lu et al., 2008). FGF has also been shown to cooperate with activin signaling through SMAD2/3 to activate NANOG expression in hES cells (Vallier et al., 2005; Li et al., 2007; Greber et al., 2007).

Some of the activity of FGF may be indirect via its action on the supporting feeder cells in hES cultures. Examining the transcriptional response of feeder cells to FGF2 identified the induction of TGFβ1 and activin A in these cells, which can subsequently promote the undifferentiated state of hES cells (Greber et al., 2007). hES cells can be cultured in the absence of feeders or of feeder-conditioned media, but in these conditions the ES cells may generate their own `niche' cells, which are flattened cells that respond to exogenous FGF2 in a similar manner to feeder cells (Greber et al., 2007; Bendall et al., 2007). Bendall et al. reported that the predominant FGF receptor, FGFR1, is expressed exclusively in these endogenous feeder-like cells and that they secrete TGFβ1 and insulin-like growth factor 2 (IGF2) in response to exogenous FGF2 (Bendall et al., 2007). They then showed that the function of exogenous FGF in promoting ES self-renewal could be replaced by addition of IGF2 alone, and suggested that the role of FGF is in fact an indirect one, via its effect on the niche in which hES cells grow. However, this extreme model has been challenged in subsequent publications. Wang et al. reported that exogenous IGF2 alone was insufficient to maintain undifferentiated growth in defined (TeSR1) culture conditions (Wang et al., 2009) and suggested that FGF has a direct role in blocking caspase-activated apoptosis through anoikis in ES cells. An anti-apoptotic action of FGF is also supported by data from Eiselleova et al., who were able to detect all four FGF receptors and weak FGF2 binding to undifferentiated hES colonies (Eiselleova et al., 2009). However, after a few days of differentiation, they did detect increased binding of FGF2 to peripheral OCT4-negative cells, as shown by Bendall et al. (Bendall et al., 2007). It seems likely that FGF2 acts both directly and indirectly on hES cells to maintain the undifferentiated state, promoting expression of pluripotency markers, such as OCT4 and the DNA methyltransferase DNMT3B, while reducing expression of TE markers and preventing cell death and apoptosis.

Given this complex interplay, it is not clear how to relate the effects of FGF signaling in hES cells with its known roles in modulating the metastable pluripotent states in mouse ES cells, EpiSCs and embryos. If hES cells are really like mouse EpiSCs, it might be possible to drive them towards a state more similar to the mouse ES cell by blocking FGF/Erk signaling and GSK3, as observed in the mouse. However, it has been clearly shown that blocking Erk signaling alone causes hES cells to lose their self-renewal capacity (Li et al., 2007). No successful direct derivation of hES cells through FGF/Erk signaling inhibition has been reported so far. However, human induced pluripotent stem (iPS) cells with morphological and transcriptional characteristics more similar to those of mouse ES cells have been established through genetic reprogramming in the presence of Erk and GSK3β inhibition (Li et al., 2009; Hanna et al., 2010). Li et al. reported that supplementing the inhibitor cocktail with activin A signaling inhibitor is crucial for preventing differentiation. Hanna et al. showed that induced pre-implantation hES cells require at least the transient expression of OCT4 and KLF4 and degenerate upon removal of the transgenes unless the inhibition cocktail was supplemented with Forskolin (Fig. 4). This suggests that a pre-implantation ES cell state can also be stabilized in human cultures. It has also recently been reported that it is possible to achieve conversion of hES cells to a mouse-like ES cell phenotype through the combined use of LIF with Erk and p38 (Mapk14) inhibitors (Xu et al., 2010). It will be interesting to find out whether fluctuations and interconversions between these different pluripotent states occur as in the mouse, and whether they can be modulated by the same or different external signaling pathways.

FGF signaling clearly plays multiple and sometimes contrasting roles in the early embryo and pluripotent stem cells. In one setting, FGF/Erk maintains pluripotency and in another it drives lineage differentiation. It is still unclear how such diversity is generated from seemingly identical FGF signaling responses. As ligands and receptors show restricted expression patterns throughout early development, it is possible that different ligand-receptor combinations could facilitate different signaling outputs. Much of the recent work on FGF signaling in pluripotent stem cells has focused on the FGF/Erk signaling axis, whereas the potential roles of FGF-activated PI3K, PLCγ and Jak/Stat signaling have been largely overlooked. Niwa et al. have shown that parallel LIF signaling through Jak/Stat and PI3K pathways maintains the pluripotent ES cell state, whereas Erk signaling appears to destabilize it (Niwa et al., 2009). This finding suggests that FGF-activated pathways might also exhibit such differential and opposing functions. Eisellova et al. also showed that the FGF signal is transduced via PI3K-Akt, as well as through the Erk pathway in hES cells (Eiselleova et al., 2009). This raises the possibility of differential functions of FGF in hES cells that are dependent on the particular downstream signaling pathway activated. Fine-tuning the balance between multiple and opposing signals downstream of FGF receptors could therefore generate contrasting functional outcomes, either maintaining pluripotency or instructing lineage differentiation. This fine-tuning might be further influenced by additional extracellular signals, such as LIF, BMP and Wnt, with downstream signaling responses that may interconnect with the FGF pathways. Recent data using global phosphoproteomics found that several pluripotency-associated proteins exhibit regulated phosphorylation in hES cells, suggesting that there might be direct post-transcriptional regulation of the pluripotent state (Brill et al., 2009; Van Hoof et al., 2009). The role of FGF and other signaling pathways in activating downstream pathways in ES cells at the post-transcriptional and transcriptional levels remains to be elucidated. A fuller understanding of the extrinsic and intrinsic pathways that regulate interchangeable and metastable pluripotent stem cell states will have a significant impact on the field of reprogramming and iPS cells.

We thank Brian Cox and Peter Rugg-Gunn for critical reading of the manuscript and useful discussions. F.L. was supported by the Swedish Research Council.