BMP signalling inhibits premature neural differentiation in the mouse embryo

Bone morphogenetic proteins (BMPs) are a family of secreted growth factors belonging to the Tgfβ superfamily. They bind to a heterotetrameric receptor complex composed of type I and type II receptors to activate an intracellular pathway that has Smad1/5/8 (Smad8 is also known as Smad9 - Mouse Genome Informatics) and Smad4 as crucial effector molecules(Shi and Massague, 2003). In mouse, although a large number of BMPs have been identified, only Bmp2 and Bmp4 have been shown to have roles during early embryonic development. Analysis of null mutations for Bmp2 and Bmp4 has revealed multiple roles for these proteins during the early post-implantation period, for example in primordial germ cell (PGC) induction(Lawson et al., 1999),mesoderm and node formation, primitive streak morphogenesis(Fujiwara et al., 2001; Winnier et al., 1995),extra-embryonic mesoderm development(Zhang and Bradley, 1996) and left-right patterning (Fujiwara et al.,2001). The fact that embryos carrying mutations in either type I or type II BMP receptors display phenotypes that are more severe than those of Bmp2 or Bmp4, suggests that functional redundancy exits between BMP proteins(Mishina, 2003).

In Xenopus, chick and zebrafish, BMP signalling has been shown to have a central role in neural induction(Harland, 2000). Although there is much debate regarding the precise mechanism for this process, two main models have been proposed. The first, based primarily on experiments performed in Xenopus, argues that BMP signalling prevents a `default'neural differentiation of the naïve ectoderm. Central to this model is the idea that secretion of BMP antagonists protects the presumptive neural plate from the action of BMPs and allows neural differentiation to occur(Vonica and Brivanlou, 2006). Experiments in both chick and Xenopus embryos have challenged this idea and have led to the proposal that inhibition of BMP signalling alone is not sufficient for neural induction and that other signals are required(Stern, 2005; Wilson and Edlund, 2001).

Recent reports have indicated that fibroblast growth factors (FGFs) may constitute one of these signals, as inhibition of FGF signalling blocks neural induction in chick (Streit et al.,2000; Wilson et al.,2000) and Xenopus(Delaune et al., 2005). The emerging picture of how FGFs promote neural differentiation seems to be complex. FGF signalling can repress BMP mRNA expression(Wilson et al., 2000) and downregulate BMP signalling intracellularly(Pera et al., 2003), but, in addition to this, FGF signalling is required for neural induction independently of BMPs (Aubin et al.,2004; Delaune et al.,2005; Linker and Stern,2004). In zebrafish, it has been proposed that both BMP inhibition and FGF signalling can act as neural inducers, with BMP antagonism inducing anterior and FGFs inducing posterior neural fates(Kudoh et al., 2004).

In the mouse embryo, a role for BMPs in neural induction has not been shown. Single mutations in BMPs or BMP antagonist do not affect the initial size of the neural plate and overlapping activities have been proposed as an explanation for this lack of neural induction phenotype. More surprisingly,double mutants for the BMP antagonists chordin and noggin are able to form a neural plate although fail to maintain forebrain markers at later stages(Bachiller et al., 2000). In mouse embryonic stem (ES) cells, BMP proteins have been implicated in maintaining pluripotency. In ES cells, BMPs are required to induce the expression of the inhibitor of differentiation (Id) genes, an essential step for ES self-renewal (Ying et al.,2003a). Bmp4 has also been shown to be required for ES cell self-renewal by inhibiting mitogen-activated kinase pathways(Qi et al., 2004). The induction of Id genes in ES cells specifically blocks neural differentiation. The role of BMPs in maintaining ES cell self-renewal while blocking neural differentiation suggests that in the embryo, BMP-mediated inhibition of neural fate might be part of a mechanism to maintain epiblast pluripotency. In the mouse embryo, anterior neural markers are not expressed until 7.25 days post-coitum (dpc) (Yang and Klingensmith,2006) and the epiblast remains pluripotent until this stage(Beddington, 1982; Beddington, 1983; Diwan and Stevens, 1976; Lawson et al., 1991). However,little is known about the mechanisms that maintain this pluripotency.

We have analysed the role of BMP signalling in patterning the epiblast of the mouse embryo. Three type I receptors have been shown to act downstream of BMPs: Acvr1, Bmpr1a and Bmpr1b (also known as Alk2,Alk3 and Alk6). Acvr1 expression is restricted to the visceral endoderm at 6.5 dpc. At 7.5 dpc, Acvr1 is expressed in the extra-embryonic mesoderm, extra-embryonic ectoderm, in the mesoderm and in the visceral endoderm (de Sousa Lopes et al.,2004; Gu et al.,1999; Kishigami et al.,2004; Roelen et al.,1994). Expression of Bmpr1b has not been detected prior to 9.5 dpc (Dewulf et al.,1995). The ubiquitous expression of Bmpr1a after implantation makes it the only type I BMP receptor expressed in the epiblast(Mishina et al., 1995; Roelen et al., 1994). For this reason, to address the role of BMP signalling in the epiblast we have used a null mutation and an epiblast-specific deletion of Bmpr1a.

Bmpr1a^+/- (Mishina et al., 1995) and Bmpr1a^fx/fx mice(Mishina et al., 2002) were maintained on a 129SvCC genetic background. The Bmpr1a^+/-;Sox2Cre^+/- mice were kept on a mixed 129SvCC/CBA/J/C57BL6 background. Embryos were genotyped after whole-mount in situ hybridisation or alkaline phosphatase staining as described(Martinez Barbera et al.,2000) using primers that are specific for the Bmpr1a-null mutation (Mishina et al.,1995), the Bmpr1a^fx/fx mutation(Mishina et al., 2002) and Sox2Cre transgene (Hayashi et al., 2002).

Staining for pSmad1/5/8 (a gift from Ed Laufer, Columbia University, New York, NY) was performed as described(Persson et al., 1998; Yang and Klingensmith, 2006). WISH was carried out following standard procedures(Thomas and Beddington,1996).

Western blots were carried out following manufacturers conditions for the pSmad1/5/8 antibody (Cell Signaling)

For inhibition of FGF signalling, embryos from Bmpr1a^+/- inter-cross litters or from Bmpr1a^fx/fx×Bmpr1a^+/-;Sox2Cre^+/- crosses were dissected in M2 medium and cultured for 24 or 48 hours in 1:1 DMEM:rat serum plus 80 μM SU5402 (Sigma) or plus DMSO alone at 37°C, 5% CO₂, fixed in 4% paraformaldehyde (PFA),dehydrated through a methanol series and processed for WISH.

Blastocysts were harvested from Bmpr1a^+/- intercrosses and cultured in gelatin-coated 48-well plates in ES cell medium containing 20%serum, or in serum-free medium with Bmp4 alone (ESGRO Clonal Grade, Chemicon)or supplemented with 200 ng/ml Bmp7 (R&D Systems), 20 ng/ml activin(Sigma), or 200 ng/ml activin for either 5 or 10 days. Blastocysts were then stained for alkaline phosphatase activity according to manufacturer's instructions (Sigma) or trypsinised and used to derive Bmpr1a^-/- ES cells following standard feeder-free conditions (Nichols et al.,1990). Given that we were not aware that it was possible to derive Bmpr1a^-/- ES cells in the absence of MAP kinase inhibition prior to our outgrowth experiments, Bmpr1a^-/- lacZ ES cells were generated from parental ES cells homozygous for the floxed Bmpr1a allele and heterozygous for the ROSA26lacZ allele by infection of Cre-expressing adenovirus in the presence of the SB203580 MAP kinase inhibitor as previously described(Qi et al., 2004).

Chimeras were generated by blastocyst injection as described(Bradley and Robertson, 1986)and stained for lacZ expression following standard procedures(Hogan et al., 1994).

We first wanted to determine to what extent was BMP signalling affected in Bmpr1a mutant embryos. A critical step in the BMP signalling pathway is the activation by phosphorylation of Smad1/5/8 by the type I BMP receptors. The phosphorylated form of Smad1/5/8 (pSmad1/5/8) associates with Smad4 and translocates to the nucleus. In wild-type embryos at 5.5 dpc, pSmad1/5/8 expression was observed in the majority of visceral endoderm cells and in a mosaic fashion in the epiblast (Fig. 1A). By 6.5 dpc, pSmad1/5/8 expression had become restricted to the proximal epiblast and proximal visceral endoderm(Fig. 1C) and remained expressed in these cell populations until 7.5 dpc(Fig. 1F). By contrast, we found no expression of pSmad1/5/8 in Bmpr1a^-/- embryos at 5.5 dpc (Fig. 1B). At 6.5 dpc,pSmad1/5/8 expression could be found in the proximal visceral endoderm of Bmpr1a^-/- embryos, but was completely absent from the epiblasts of 25% of the mutant embryos(Fig. 1D) and present in a weak and mosaic manner in the epiblast of the rest of the Bmpr1a^-/- embryos (Fig. 1E). By 7.5 dpc, no pSmad1/5/8 expression could be observed in Bmpr1a^-/- embryos (Fig. 1G). These results indicate that Bmpr1a is the type I receptor required for BMP signalling in the epiblast during the pre- and early gastrulation stages.

A previous characterisation of the Bmpr1a mutant phenotype found that mutant embryos fail to gastrulate and do not express the mesoderm marker brachyury (T), suggesting that severe patterning defects are present(Mishina et al., 1995). In ES cells, BMP signalling has been implicated in the maintenance of pluripotency(Qi et al., 2004; Ying et al., 2003a). We therefore studied the state of differentiation of the epiblast of Bmpr1a mutants. For this we analysed the expression of Oct4(also known as Pou5f1 - Mouse Genome Informatics), Nanog and Fgf5. Oct4 is expressed throughout the epiblast at 5.5 and 6.5 dpc(Rosner et al., 1990). At 5.5 dpc, we observed no difference in the level of Oct4 expression in Bmpr1a mutant embryos when compared with controls(Fig. 2A,B). However, at 6.5 dpc, half of the Bmpr1a mutant embryos displayed a clear downregulation of Oct4 expression(Fig. 2C,D; n=4/8). Nanog is required for ES and inner cell mass pluripotency(Chambers et al., 2003; Mitsui et al., 2003). At 6.5 dpc, Nanog expression was observed in the proximal-posterior epiblast(Fig. 2E). In all Bmpr1a^-/- embryos analysed at 6.5 dpc, Nanogexpression was completely lost from the epiblast(Fig. 2F). Fgf5 is a primitive-ectoderm marker and is expressed throughout the epiblast at 6.5 dpc(Fig. 2G)(Kaji et al., 2006; Sun et al., 1999). When we examined Fgf5 expression in Bmpr1a mutant embryos we observed a severe downregulation or loss of expression of this gene(Fig. 2H). The downregulation or loss of Oct4, Nanog and Fgf5 expression indicates that BMP signalling is required to maintain epiblast pluripotency.

The downregulation of pluripotency markers suggests that a premature differentiation of the epiblast is occurring in Bmpr1a mutant embryos. In other vertebrates, inhibition of BMP signalling has been proposed as a central step in neural induction(Harland, 2000; Vonica and Brivanlou, 2006). For this reason we investigated the possibility that the epiblast of Bmpr1a^-/- embryos is undergoing neural differentiation. At 7.5 dpc, markers of the anterior neural ectoderm are expressed in a discrete patch of the anterior epiblast (Yang and Klingensmith, 2006). We analysed the expression of the anterior neural markers Hesx1 and Six3 and of the pan-neural marker Sox1 and found that at 7.5 dpc, these genes are expressed throughout the epiblast of Bmpr1a mutant embryos(Fig. 2J,L,N; n=7/8,4/6 and 7/8), indicating that ectopic neural differentiation of the epiblast is occurring in these embryos. Foxg1 is a forebrain marker that is first expressed at the three-somite stage(Xuan et al., 1995). At 7.5 dpc, a distinct patch of Foxg1 expression could be observed in Bmpr1a^-/- embryos at the boundary between the embryonic and extra-embryonic regions (Fig. 2P; n=4/6), indicating that, at least to some degree, the ectopic neural tissue is regionalised in Bmpr1a mutants. We next analysed the expression of midbrain marker engrailed 1(Davis and Joyner, 1988). In Bmpr1a^-/- embryos, no engrailed 1 expression was observed at 7.5 dpc (data not shown, n=7), suggesting that there is only premature expression of anterior neural genes in these embryos.

Given that at 6.5 dpc we had found downregulation of pluripotency markers,we wanted to test whether neural differentiation of the epiblast was evident as early as 6.5 dpc. Interestingly, we found Hesx1 to be expressed throughout the epiblast of 3/10 Bmpr1a^-/- embryos, and Six3 throughout the epiblast of 3/5 mutant embryos analysed at 6.5 dpc (Fig. 2R,L); this is a similar proportion to that of mutant embryos showing no pSmad/1/5/8 at this stage. The precocious expression of anterior neural gene markers coupled with the downregulation of pluripotent epiblast markers observed in Bmpr1a^-/- embryos suggests that BMP signalling is required to prevent premature neural differentiation of the epiblast prior to gastrulation.

A necessary step for neural induction to occur is the suppression of mesoderm and mesoderm-inducing signals in the prospective neural plate region. To test whether the loss of pluripotency in the epiblasts of Bmpr1a^-/- embryos leads to any mesoderm differentiation we analysed the expression of Fgf8, T and eomesodermin. At 6.5 dpc, Fgf8 is expressed in the posterior epiblast and visceral endoderm(Fig. 3A)(Crossley and Martin, 1995)and both eomesodermin and T are expressed in the posterior epiblast and extra-embryonic ectoderm (Fig. 3C,E) (Perea-Gomez et al.,2004; Russ et al.,2000). In Bmpr1a mutant embryos, Fgf8 expression was lost from the epiblast, but could still be observed in the visceral endoderm (Fig. 3B) and expression of both eomesodermin and T was lost from the epiblast and strongly downregulated in the extra-embryonic ectoderm(Fig. 3D,F).

Two signalling molecules that have been shown to be essential for mesoderm specification are Nodal (Brennan et al.,2001) and Wnt3 (Liu et al.,1999). At 6.5 dpc, Nodal and the Nodalco-receptor Cripto (also known as Tdgf1 - Mouse Genome Informatics) (Gritsman et al.,1999) were found to be expressed in the posterior epiblast(Fig. 3G,I), but expression of both genes was lost in Bmpr1a mutants(Fig. 3H,J). Wnt3 is normally expressed in the posterior epiblast and adjacent visceral endoderm(Fig. 3K), but in embryos lacking Bmpr1a, expression of Wnt3 was strongly downregulated or lost from both these tissues(Fig. 3L). The lack of Fgf8, eomesodermin, T, Nodal, Cripto and Wnt3expression in Bmpr1a ^-/- embryos indicates that the ectopic neural differentiation observed in these embryos is accompanied by a suppression of mesoderm and mesoderm-inducing signals.

Bmpr1a is expressed in all three tissues of the early post-implantation embryo: the epiblast, the visceral endoderm and the extra-embryonic ectoderm (Mishina et al.,1995; Roelen et al.,1994). To test whether the defects observed in Bmpr1a^-/- embryos are due to requirements of BMP signalling in the epiblast, we deleted Bmpr1a specifically from this tissue. We used Bmpr1a^fx/fx mice that carry a conditional mutation in Bmpr1a (Mishina et al., 2002), and the epiblast-specific Cre-driver mouse line Sox2Cre^+/- (Hayashi et al., 2002).

We first analysed how BMP signalling is affected in Bmpr1a^-/fx; Sox2Cre^+/- embryos by looking at pSmad1/5/8 expression. At 5.5 dpc, we found pSmad1/5/8 to be slightly reduced in the epiblast of Bmpr1a^-/fx; Sox2Cre^+/-embryos when compared with controls (Fig. 4B; n=4/7). This expression in the epiblast of Bmpr1a^-/fx; Sox2Cre^+/- embryos was drastically reduced at 6.5 dpc in the majority of embryos analysed(Fig. 4D; n=4/7) and almost absent by 7.5 dpc in all embryos studied(Fig. 4E; n=4/4). Therefore, in Bmpr1a^-/fx; Sox2Cre^+/- embryos there is a loss of BMP signalling; however, this loss occurs later than in Bmpr1a-null mutants.

We next studied mesoderm specification and posterior epiblast patterning in Bmpr1a^-/fx; Sox2Cre^+/- embryos by analysing the expression of Fgf8, Nodal and Cripto. At 6.5 dpc, we observed either a complete loss of expression or a downregulation of these markers in the majority of Bmpr1a^-/fx;Sox2Cre^+/- embryos analysed (for Fgf8 n=4/6, Fig. 4H; Cripto n=3/9, Fig. 4J; Nodal n=3/4, Fig. 4L). To determine whether ectopic neural differentiation is also occurring in Bmpr1a^-/fx;Sox2Cre^+/- embryos, we analysed the expression of Hesx1,Six3 and Sox1. At 6.5 dpc, no expression of either Hesx1 (n=10, data not shown) or Six3(n=13, data not shown) was detected, but by 7.5 dpc a large expansion in the domain of expression of both these markers could be seen in Bmpr1a^-/fx; Sox2Cre^+/- embryos (Hesx1 n=4/6, Fig. 4N and Six3 n=8/10 Fig. 4P). Similarly, Sox1 was also expanded at 7.5 dpc in Bmpr1a^-/fx; Sox2Cre^+/- embryos(Fig. 4R; n=2/2). These observations indicate that BMP signalling is required in the epiblast to inhibit neural differentiation. Furthermore, given that BMP signalling is lost later in Bmpr1a^-/fx; Sox2Cre^+/- embryos than in Bmpr1a^-/- embryos, and that this later loss is correlated with a later onset of expression of ectopic neural markers, one possibility is that BMP inhibition of neural differentiation is occurring continually in the epiblast rather that at one specific time-point. Alternatively, it is also possible that BMP signalling might regulate the expression in the visceral endoderm of other secreted factors that are also required to maintain pluripotency.

We next asked which BMPs signal via Bmpr1a to maintain pluripotency. In Drosophila, the Bmpr1a homologue thickveins preferentially transduces dpp signals (reviewed by Podos and Ferguson, 1999),and in vertebrates the Dpp homologues Bmp2 and Bmp4 bind with high affinity to Bmpr1a (ten Dijke et al.,1994). We therefore tested whether Bmpr1a is required for Bmp2/4 signalling in the early embryo. We first analysed the ability of Bmp4 to stimulate the phosphorylation of Smad1/5/8 in vitro. We found that in wild-type ES cells, Bmp4 can stimulate phosphorylation of Smad1/5/8. By contrast, in Bmpr1a mutant ES cells, Bmp4 was unable to simulate Smad1/5/8 phosphorylation (Fig. 5A). We further tested the ability of Bmp2/4 to signal via Bmpr1a by using an in vitro blastocyst outgrowth assay. Wild-type and Bmpr1a^-/- blastocysts were cultured on gelatin-coated dishes in serum-free conditions in the presence of Bmp4 and assayed for alkaline phosphatase activity, a characteristic marker of pluripotent cells. When grown for 10 days in these conditions, wild-type and heterozygous blastocysts had formed large outgrowths that displayed strong levels of alkaline phosphatase activity (Fig. 5G and Fig. 6I). By contrast, Bmpr1a^-/- blastocysts showed hardly any proliferation during the 10 days in culture and exhibited variable levels of alkaline phosphatase activity (Fig. 5H and Fig. 6J),indicating that Bmp4 was unable to sustain self-renewal of the inner cell mass in these embryos. The loss of pSmad1/5/8 expression observed in Bmpr1a^-/- mutant embryos at 5.5 dpc, coupled with the inability of Bmp4 to signal in Bmpr1a mutant embryos and ES cells,indicates that Bmp2/4 signalling via Bmpr1a is the BMP pathway that maintains pluripotency prior to gastrulation.

Are Bmp2/4 the only BMPs required to maintain pluripotency? As a first approach to address this question, we analysed in chimeras the developmental potential of Bmpr1a^-/- ES cells that are unable to respond to Bmp2/4 signalling. We found that although high-contribution chimeras generated with Bmpr1a^-/- ES cells recapitulated the phenotype of the epiblast-specific deletion of Bmpr1a mutant embryos(data not shown), in low-contribution chimaeras Bmpr1a^-/-cells were able to contribute efficiently to cell types, except the allantois,formed from all three germ layers (Fig. 5C-F). This suggests that BMP signalling via other type I receptors is also required to maintain pluripotency.

Bmp7 is a BMP of the 60A subgroup that is first expressed in mouse at the start of gastrulation (Solloway and Robertson, 1999). In both Drosophila and vertebrates,BMPs of the 60A subgroup have been shown to signal preferentially via the type I BMP receptor Acvr1 (Podos and Ferguson, 1999; ten Dijke et al., 1994). For this reason, we first tested whether Bmp7 could signal in Bmpr1a^-/- cells. We found that in contrast to Bmp4, Bmp7 was capable of stimulating phosphorylation of Smad1/5/8 in Bmpr1a^-/- ES cells(Fig. 5A). Interestingly, the ability of Bmp7 to signal in Bmpr1a mutant ES cells correlated with an upregulation of Acvr1 (Fig. 5B), but not of Bmpr1b, in these cells (data not shown),suggesting that Bmp7 was signalling preferentially via this type I receptor.

Given the ability of Bmp7 to signal in Bmpr1a mutant ES cells, we analysed whether Bmp7 could rescue the defects of Bmpr1a^-/- cells in vivo. For this we carried out blastocyst outgrowth assays. When Bmpr1a^-/- blastocysts were cultured in the presence of Bmp7 they formed outgrowths that were equivalent both in size and alkaline phosphatase activity to control outgrowths (Fig. 5I,J). Together, these results suggest that prior to gastrulation, Bmp2/4 signalling via Bmpr1a is the main source of BMP signalling that maintains epiblast pluripotency, but as gastrulation commences, signalling by BMPs of the 60A subgroup via Acvr1 cooperates with Bmp2/4 signalling to inhibit neural differentiation of the epiblast.

Nodal signalling has been shown to be required to maintain pluripotency and inhibit neural differentiation in the mouse embryo(Camus et al., 2006; Mesnard et al., 2006). To determine whether at least part of the defects of Bmpr1a^-/- embryos are due to an absence of Nodalsignalling, we analysed how early Nodal expression is lost in Bmpr1a mutant embryos. At 4.5 dpc, Nodal is expressed in the epiblast and primitive endoderm (Takaoka et al., 2006) and at 5.5 dpc in the epiblast and overlying visceral endoderm (Varlet et al.,1997). In Bmpr1a mutant embryos, we found no difference in the expression of Nodal at 4.5 dpc(Fig. 6A,B), but a clear downregulation relative to wild-type embryos at 5.5 dpc(Fig. 6C,D). We next analysed the expression of the Nodal co-receptor Cripto and of the Tgfβ Gdf3 at 5.5 dpc. At this stage, both these markers are normally expressed throughout the epiblast(Chen et al., 2006; Ding et al., 1998). Interestingly, although expression of Gdf3 was normal, Cripto expression was downregulated in Bmpr1a^-/-embryos (Fig. 6F,H). These observations indicate that epiblast patterning defects are present in Bmpr1a mutants as early as 5.5 dpc, and that BMP signalling is required to maintain Nodal signalling in the early embryo.

We next tested whether Nodal signalling was capable of rescuing the pluripotency defects of Bmpr1a^-/- embryos. For this,we cultured Bmpr1a^-/- blastocysts in serum-free conditions in the presence of 20 ng/ml or 200 ng/ml activin (also known as inhibin) for 10 days. Interestingly, we found that although 200 ng/ml activin did not rescue the proliferation or pluripotency defects of Bmpr1a^-/- embryos (n=0/6; data not shown), 20 ng/ml activin did rescue both these defects (n=5/7; Fig. 6L). Therefore, Nodal signalling can, at least in part, rescue the pluripotency defects of Bmpr1a^-/- embryos.

FGFs have been proposed as inducers of neural fate in both chick(Streit et al., 2000; Wilson et al., 2000) and Xenopus embryos (Delaune et al.,2005), and in mouse have been shown to act antagonistically to BMP signalling during mesoderm development(Miura et al., 2006). We tested wether FGF signalling is responsible for the precocious and ectopic neural induction observed in Bmpr1a mutant embryos. In agreement with previous studies (Corson et al.,2003), we found that 80 μM SU5402 could efficiently inhibit FGF signalling within 1 hour in cultured 6.0 dpc embryos (data not shown). Given that Bmpr1a^-/fx; Sox2Cre^+/- embryos do not show precocious neural marker expression at 6.5 dpc, we used these embryos to test whether inhibition of FGF signalling during gastrulation could block the expression of the anterior neural markers. 6.5 dpc control and Bmpr1a^-/fx; Sox2Cre^+/- embryos were cultured for 24 hours with DMSO or 80 μM SU5402 and assayed for Hesx1expression. After 24 hours of culture, only a small proportion of DMSO-treated controls (2/30) or Bmpr1a^-/fx; Sox2Cre^+/- (2/7)embryos had initiated Hesx1 expression in the epiblast(Fig. 7A,C). Surprisingly,although SU5402 inhibition did not significantly change the proportion of control embryos expressing Hesx1 (5/26; Fig. 7B), it led to a drastic increase in the proportion of Bmpr1a^-/fx;Sox2Cre^+/- embryos displaying ectopic Hesx1expression (11/13; Fig. 7D). These results indicate that inhibition of FGF signalling during gastrulation stages does not inhibit neural specification in Bmpr1a^-/fx;Sox2Cre^+/- embryos and suggest that BMP and FGF signalling might act synergistically to inhibit neural differentiation at these stages.

To test whether FGF signalling may be acting prior to gastrulation to induce neural fate, we inhibited FGF signalling by culturing 5.5 dpc Bmpr1a^-/fx; Sox2Cre^+/- and Bmpr1a^-/- embryos for 48 hours in the presence of SU5402. No control (0/33) or Bmpr1a^-/fx; Sox2Cre^+/-(0/6) embryos and only 55% of Bmpr1a^-/- embryos (5/9)showed Hesx1 expression after 48 hours culture in DMSO(Fig. 7E,G,I). By contrast, 29%of control (15/52), 80% of Bmpr1a^-/fx;Sox2Cre^+/- (4/5) and 90% of Bmpr1a^-/-embryos (10/11) displayed ectopic Hesx1 expression after 48 hours culture with SU5402 (Fig. 7F,H,J). Interestingly, the extra-embryonic region of these SU5402-treated embryos was much reduced compared with controls, most probably owing to the effects of FGF inhibition on the trophoblast stem cells(Tanaka et al., 1998). These observations indicate that the ectopic neural induction seen in Bmpr1a mutants is not mediated by FGF signalling and also suggest that FGFs are not acting as direct neural inducers in the early post-implantation mouse embryo.

The role of BMPs in patterning the early vertebrate embryo and inducing differentiation of a number of cell types is well established. For example, in the early mouse embryo Bmp4 plays a crucial role in PGC induction, mesoderm and node formation and left-right asymmetry. In contrast to these findings, we have identified a role for BMP signalling in maintaining epiblast pluripotency and have shown that loss of BMP signalling leads to ectopic neural induction in the mouse embryo.

Strong evidence exists that right up to 7.5 dpc, a proportion of cells from the epiblast remain pluripotent(Beddington, 1982; Beddington, 1983; Diwan and Stevens, 1976; Lawson et al., 1991). The maintenance of this progenitor population is crucial for the allocation of cells to the three germ layers: mesoderm, endoderm and ectoderm. A limited number of factors have been identified as required to maintain epiblast pluripotency, including Oct4(Nichols et al., 1998), Nanog (Chambers et al.,2003; Mitsui et al.,2003), Sox2 (Avilion et al., 2003), Foxd3(Hanna et al., 2002) and Fgf4 (Feldman et al.,1995). However, given that mutations in these factors cause lethality at peri-implantation stages, it has not been possible to determine whether these genes affect pluripotency after implantation.

A number of observations indicate that BMP signalling via Bmpr1aplays a crucial role in maintaining epiblast pluripotency. Bmpr1a is expressed in inner cell mass-derived ES cells(Ying et al., 2003a) and was highly enriched in a screen aimed at identifying stem cell markers(Sato et al., 2003). In Bmpr1a^-/- embryos, we find a downregulation or loss of expression of genes required for epiblast pluripotency, such as Oct4,Nanog, Nodal and Cripto. Furthermore, we show that Bmpr1a^-/- embryos exhibit self-renewal defects in blastocyst outgrowth assays. In ES cells, BMP signalling cooperates with the Lif/Stat pathway to drive self-renewal(Ying et al., 2003a). BMP signalling is therefore required both in vitro and in vivo to maintain pluripotency.

A key aspect of the role of BMPs in maintaining epiblast pluripotency is the inhibition of precocious neural differentiation. In ES cells, Bmp4 maintains pluripotency by specifically blocking the neural differentiation pathway (Ying et al., 2003a). Similarly, we find that Bmpr1a mutant embryos show precocious and ectopic expression of anterior neural markers coupled to the loss of pluripotency markers. Inhibition of BMP signalling has been shown to be a central step for neural induction in a number of vertebrate species(Harland, 2000). Mainly based on experiments in Xenopus, a model has been proposed in which BMPs act prior to gastrulation to inhibit a `default' neural differentiation of the ectoderm. According to this model, neural specification occurs when the onset of expression of BMP antagonists allows differentiation to take place in the prospective neural ectoderm (Vonica and Brivanlou, 2006). The precocious expression of neural markers observed in Bmpr1a^-/- embryos indicates that as in Xenopus, inhibition of BMP signalling in the mouse embryo is occurring prior to gastrulation; therefore, a crucial step in neural induction must be the relief of BMP inhibition.

Our results also suggest that BMP inhibition of neural fate is required continuously in the epiblast rather than at a single specific stage of development. Support for this comes from the comparison of Bmpr1a^-/- and Bmpr1a^-/fx;Sox2Cre^+/- mutant phenotypes. In Bmpr1a^-/-embryos, a complete loss of pSmad1/5/8 expression occurs in the epiblast as early as 5.5 dpc and this correlates with the precocious and ectopic expression of neural markers at 6.5 dpc. By contrast, a significant decrease in BMP signalling only occurs in Bmpr1a^-/fx;Sox2Cre^+/- embryos from 6.5 dpc and ectopic neural markers are not seen until 7.5 dpc. Therefore, the timing of the loss of BMP signalling correlates with the timing of the onset of neural differentiation in two separate mutations. This observation suggests that in the wild-type embryo,BMP signalling is required to maintain epiblast pluripotency from the time of implantation until 7.5 dpc, when neural differentiation takes place. It is likely that this neural differentiation occurs when BMPs are antagonised(Fig. 8), first by factors secreted by the anterior visceral endoderm (AVE) and later by the axial mesendoderm, the two signalling centres required for neural specification(Beddington and Robertson,1999).

Which BMPs can inhibit neural fate? In Xenopus embryos, ubiquitous neural induction throughout the ectoderm only occurs after knock-down of four BMP family proteins: BMP2, BMP4, BMP7 and ADMP(Reversade and De Robertis,2005). By contrast, we find in mouse that Bmp2/4 signalling via Bmpr1a is the main source of BMP signalling that inhibits neural fate prior to gastrulation. The loss of pSmad1/5/8 expression in Bmpr1a^-/- embryos coupled with the fact that Bmp2/4 cannot signal in Bmpr1a^-/- ES cells but other BMPs can, supports this view. Interestingly, in mouse it is only as gastrulation commences that signalling by BMPs of the 60A subgroup via Acvr1 cooperates with Bmp2/4 signalling to reinforce this inhibition of neural differentiation(Fig. 8). Two observations support this. First, members of the 60A subgroup only start to be expressed at the onset of gastrulation (Solloway and Robertson, 1999). Secondly, we find that Bmp7 can rescue the self-renewal defects of Bmpr1a^-/- embryos in blastocyst outgrowth assays. Therefore, in mouse, although Bmp2 and Bmp4 signalling are initially sufficient to inhibit neural fate, as development proceeds multiple BMPs signalling via different receptor complexes are required to block neural differentiation.

One candidate pathway for acting downstream of BMP signalling is the Nodal signalling pathway. BMPs have been shown to act upstream of Nodal during left-right patterning(Fujiwara et al., 2002; Piedra and Ros, 2002; Schlange et al., 2002) and to amplify Nodal expression in the epiblast(Ben-Haim et al., 2006). Nodal mutant embryos display similar pluripotency defects and precocious neural differentiation to those we observe in Bmpr1amutant embryos (Camus et al.,2006; Mesnard et al.,2006). In Bmpr1a mutant embryos, we find a loss of expression of both Nodal and its co-receptor Cripto just prior to the time when we observe the precocious expression of neural markers. Furthermore, we find that activin/Nodal signalling can rescue the pluripotency defects of Bmpr1a mutant embryos in blastocyst outgrowth assays. These lines of evidence suggest that BMP signalling might be regulating the levels of Nodal signalling in the early mouse embryo and that, at least in part, the defects in Bmpr1a mutant embryos are owing to the lack of Nodal signalling activity.

Recent experiments in both chick and Xenopus embryos have challenged the idea that inhibition of BMP signalling is sufficient for neural induction and have suggested that FGF signalling might be an inductive signal during this process. These studies found that inhibition of FGF signalling by the small molecule SU5402 blocks neural induction in both chick(Streit et al., 2000; Wilson et al., 2000) and Xenopus (Delaune et al.,2005) embryos and that this role of FGFs is, at least in part,independent of antagonising BMP signalling. However, in contrast to these observations, we find that FGF signalling is not responsible for the ectopic neural induction observed in Bmpr1a^-/- embryos. Inhibition of FGF signalling using SU5402 24 hours before gastrulation did not inhibit the expression of Hesx1 in the epiblast of either Bmpr1a^-/fx or Bmpr1a^-/fx;Sox2Cre^+/- embryos. Furthermore, inhibition of FGF signalling led to an expansion of anterior neural tissue in a proportion of wild-type embryos. Given that an expansion of neural tissue is also observed in Fgf8-null mutants (Sun et al.,1999) and in Fgfr1 chimeras(Ciruna et al., 1997) and that Fgf2 has been shown to inhibit neural fate in mouse ectoderm explants(Burdsal et al., 1998), it seems unlikely that the primary role of FGF signalling in the early post-implantation mouse embryo is to act as a direct neural inducer, but instead is required to promote mesoderm formation and migration. In ES cells,inhibiting FGF signalling using SU5402 has been shown to block neural differentiation (Ying et al.,2003b). It is therefore possible that FGFs are required during the pre-implantation period for a maturation step in the epiblast that makes this tissue competent for neural specification, but once this step has taken place it is inhibition of BMP signalling that is the main event required for the induction of neural marker expression. It is also possible that, as has been suggested in Xenopus (Pera et al., 2003), a role for FGF signalling during early post-implantation development could be to attenuate BMP signalling and drive neural specification in this way.

In conclusion, we have found that Bmp2/4 signalling via Bmpr1a is inhibiting neural differentiation of the epiblast prior to gastrulation and that these BMPs cooperate with members of the 60A subgroup during gastrulation. We have also shown that, in contrast to what seems to occur in chick, in the early mouse post-implantation embryo inhibition of FGF signalling does not block neural specification and that it is relief of BMP inhibition that is the critical step for neural induction to occur.

We thank Josh Brickman, Rita Lopes, Tilo Kunah and Vasso Episkopou for critical reading of the manuscript and great suggestions; Shigeto Miura, Masa Yamamoto, Hiroshi Hamada, Jerome Collignon and Anne Camus for sharing unpublished information and helpful suggestions; Dr Ucchii Takashi Uchimura for preparing the Bmpr1a^-/- ES cells; Ed Laufer and Dan Vasiliauskas, Susan Morton and Tom Jessell for the pSmad1/5/8 antibody;Michael Shen for the Gdf3 probe; and Zöe Webster for her advice. M.S. holds an FCT fellowship. T.A.R. holds a Lister Institute of Preventive Medicine fellowship. The MRC and the Intramural Research Program of the NIH,National Institute of Environmental Health Sciences, also supported this research.