Antagonism of Nodal signaling by BMP/Smad5 prevents ectopic primitive streak formation in the mouse amnion

Cell fate is determined by intrinsic programs and external cues, such as growth factors, cell-cell contacts and cell-matrix contacts. In the early embryo, Nodal and BMPs are morphogens that play pivotal roles in gastrulation and tissue patterning. These ligands belong to the transforming growth factor (TGF) type β family. Nodal and BMPs present frequently opposing roles during fate determination, as demonstrated in several cellular and developmental contexts, such as in differentiation of mouse embryonic stem cells (Fei and Chen, 2010; Galvin et al., 2010) or in patterning of the nascent mesoderm in the primitive streak during the gastrula stage of mouse development (Tam and Loebel, 2007; Willems and Leyns, 2008). In the primitive streak, epiblast cells ingress while undergoing an epithelial-to-mesenchymal transition (EMT), and form mesoderm and definitive endoderm. While BMP signaling is required for specification of derivatives formed at the posterior part of the primitive streak (Bachiller et al., 2000; Kalantry et al., 2001; Lawson et al., 1999; Winnier et al., 1995), increasingly higher Nodal signaling is required for specification of derivatives formed at the anterior part of the primitive streak (Norris et al., 2002; Vincent et al., 2003).

Nodal and BMP also function cooperatively during, for example, formation of the primitive streak. Robust Nodal expression (and signaling) is required for primitive streak formation and its correct positioning is triggered by the Nodal autoregulatory loop and amplified via the slow positive-feedback loop, which requires Bmp4 and Wnt3 expression and signaling (Ben-Haim et al., 2006; Shen, 2007). High levels of Nodal signaling are required for specification and allocation of the anterior visceral endoderm (AVE) (Mesnard et al., 2006; Norris et al., 2002; Yamamoto et al., 2001a). The latter tissue secretes antagonists of Nodal, BMP and Wnt signaling, namely Cerberus-like (Cer1), Dickkopf-related protein 1 (Dkk1) and Lefty1, which restrict the mesoderm-inducing signals to the prospective posterior side of the embryo where the primitive streak is formed (Kimura et al., 2000; Perea-Gomez et al., 2002).

The binding of Nodal ligand to the Alk4 and ActRIIA/ActRIIB receptor complex activates the downstream effectors, or receptor-regulated Smads (R-Smads), Smad2 and Smad3 (Sakuma et al., 2002). Binding of BMPs to receptor complexes activates another subset of R-Smads, namely Smad1, Smad5 and Smad8 (Kawai et al., 2000; Meersseman et al., 1997). Unique to the Nodal system is the crucial requirement of a co-receptor from the EGF-CFC family – Cripto and Cryptic (Ding et al., 1998). Activated R-Smads form heteromeric complexes with the common Smad (Smad4), translocate into the nucleus and regulate, in combination with other transcription factors and co-factors, the transcription of target genes. In the case of Nodal signaling, activated Smad complexes interact with transcription factors such as Mixl1 and Foxh1 (Shi and Massague, 2003). The Nodal-Smad2/4-Foxh1 cascade induces Nodal expression itself (the Nodal autoregulatory loop), and also establishes a negative regulatory loop by the direct induction of the Nodal antagonist Lefty2 (Hoodless et al., 2001; Yamamoto et al., 2001a). The Lefty proteins are TGFβ family ligands that compete extracellularly with Nodal and its (co-)receptors, and hence interfere via different modes with the activation of the Nodal signaling cascade (Chen and Shen, 2004; Cheng et al., 2004; Sakuma et al., 2002; Tanegashima et al., 2004).

Crosstalk and genetic interaction between Nodal and BMP signaling pathways has been demonstrated in different developmental contexts where these ligands co-signal in zebrafish, Xenopus and mice (Chocron et al., 2007; Yamamoto et al., 2001b). BMP signaling attenuates Nodal signaling and expression during left-right (LR) axis determination in mouse embryos (Chang et al., 2000; Furtado et al., 2008; Mine et al., 2008). Smad1-mediated BMP signaling sets a repressive threshold for Nodal signaling in the left lateral plate mesoderm (LPM) by limiting the availability of Smad4. This attenuates Nodal signaling and establishes a robust, but not exceedingly high, level of signaling during LR axis determination (Furtado et al., 2008). Mice lacking Smad5 express Nodal bilaterally in the LPM instead of unilaterally in left LPM, and the expression of the Nodal target genes Pitx2, Lefty1 and Lefty2 was affected (Chang et al., 2000). Even though Smad1 and Smad5 mutant mice have similar LR asymmetry defects associated with altered Nodal signaling, it remains unclear whether and how Smad5 signaling antagonizes Nodal signaling.

Smad1 and Smad5 share striking sequence homology and their mRNA is fairly ubiquitously expressed in the embryo proper. Smad1 (Lechleider et al., 2001; Tremblay et al., 2001) or Smad5 (Chang et al., 1999; Yang et al., 1999) loss-of-function mouse mutants or compound heterozygosity of Smad1 and Smad5 (Arnold et al., 2006) are all embryonic lethal. The conventional Smad5 knockouts die around mid-gestation from several embryonic and extra-embryonic defects that can be attributed to reduced BMP signaling. Amnion closure is delayed in the Smad5 mutant, which resembles a milder phenotype of the Bmp2 mutant mice (Zhang and Bradley, 1996). The amnion is normally a thin and avascular extra-embryonic tissue that forms from a single amniochorionic fold during gastrulation and ultimately surrounds and cushions the fetus (Pereira et al., 2011). After amnion closure, the Smad5 mutant amnion develops specific defects that cannot be attributed to reduced BMP signaling (Bosman et al., 2006). The cells of the amniotic ectoderm are cuboidal in the anterior part of the Smad5 mutant amnion, and the anterior delineation between neural, surface and amnion ectoderm is lost. At early somite stages, mutant amniotic ectoderm and mesoderm thicken anteriorly, resulting in a clump of cells. The amniotic clump displays ectopic hematopoiesis and vasculogenesis, and develops de novo PGC-like cells. These different mesodermal cells normally form at the posterior part of the primitive streak. These cells are stereotypically organized within the clump of cells in the Smad5 mutant amnion. Increased or ectopic Bmp and target gene expression in the mutant amnion suggest that lack of Smad5 results in a gain-of-function of BMP signaling in the amnion (Bosman et al., 2006). Given that the thickened amnion contains mesoderm cell types normally specified in the posterior part of the primitive streak, we investigated whether an ectopic primitive streak is induced. Knowing that Nodal signaling is crucial for establishment of the primitive streak and that this function is highly conserved (Shen, 2007; Tian and Meng, 2006), we questioned whether Smad5-mediated signaling is required in the mouse embryo to antagonize Nodal signaling and to prevent ectopic primitive streak formation in amnion.

Here, we find that the formation of a primitive streak-like structure in Smad5 mutant amnion is accompanied by ectopic Nodal expression. Biochemical studies in cells demonstrate that Smad5-mediated BMP signaling interferes with the Nodal autoregulatory loop. Comparative quantitative gene expression analysis is compatible with ectopic Nodal in the Smad5-deficient amnion being induced via the Nodal autoregulatory loop and amplified via the slow positive-feedback loop. We propose that Smad5-mediated antagonism of Nodal signaling in the amnion is an essential mechanism to prevent primitive streak formation in this extra-embryonic tissue.

Smad5^m1/m1 is a conventional Smad5 knockout mouse (Chang et al., 1999). In Nodal^lacZ/+ mice, one Nodal allele is replaced by cDNA encoding β-galactosidase (Varlet et al., 1997). BAT-gal mice express lacZ under control of canonical Lef/Tcf-dependent Wnt signaling (Maretto et al., 2003). BRE-lacZ mice express lacZ under control of the BMP responsive element (BRE) from the Id1 promoter region (Monteiro et al., 2004). All experiments were approved by the ethical commission from KU Leuven (097/2008).

Whole-mount in situ hybridization was performed using standard procedures (Rosen and Beddington, 1993) with antisense probes for the following genes: T (Herrmann et al., 1990), Cer1 (Belo et al., 1997), Dlx5 (Yang et al., 1998), Dkk1 (J. Debruyn, KU Leuven, Belgium), Eomes (Ciruna and Rossant, 1999), Fgf8 (Crossley and Martin, 1995), Gsc (Blum et al., 1992), Hesx1 (Thomas and Beddington, 1996), Lefty1 and Lefty2 (Meno et al., 1997), Nodal (Perea-Gomez et al., 2004) and Wnt3 (Liu et al., 1999).

β-Galactosidase staining was carried out according to a standard protocol using 0.5 mg/ml X-gal (Fermentas). Embryos were embedded in Technovit 8100 (Heraeus Kulzer), sectioned and stained with 0.05% Neutral Red solution.

Immunohistochemistry with anti-E-cadherin (BD Transduction) and anti-Snail (Abcam) antibodies was performed using an automated platform (Ventana Discovery, Ventana Medical Systems) on 5 μm paraffin sections of 4% paraformaldehyde-fixed embryos.

HEK293T cells were cultured in 96-well plates and transfected using Lipofectamine 2000 (Invitrogen). The following plasmids were used: Myc-tagged Foxh1, Flag-tagged Cripto, Nodal and A3-luc (Iratni et al., 2002); HA-tagged Smad4 (de Winter et al., 1997); Flag-tagged Smad1 and Flag-tagged Smad5 (Imamura et al., 1997); and 3xFlag/Strep-tagged Smad5 and 3xFlag/Strep-tagged Δ-Smad5 (F.C. and A.Z., unpublished). The latter is a Smad5 deletion mutant lacking the C-terminal SSXS domain. Recombinant human rhBMP4 50 ng/ml (R&D Systems) was used to stimulate cells. Cells were harvested after 24 hours for luciferase determination (Promega). Normalization was to β-galactosidase activity from co-transfected pCMVβ-gal plasmid (Clontech) assayed using Galscreen Tropix (Applied Biosystems). At least two independent experiments were performed in sextuplicates. Silencing endogenous Smad5 was performed by co-transfecting ON-TARGET plus SMARTpool Smad5 siRNA (Dharmacon) or non-targeting control siRNA at a final concentration of 10 nM. Blockade of protein synthesis was with cycloheximide 50 μg/ml (Merck) pretreatment of cells for 1 hour, followed by 5 hours stimulation with 50 ng/ml rhBMP4 in the presence of cycloheximide. RNA was extracted and purified (RNeasy RNA purification columns, Qiagen). Reverse transcription (RT) was with MuMLV reverse transcriptase (Revert-aid, Fermentas), oligo-dT₁₈ and random primers (Invitrogen). Luciferase expression was analyzed by quantitative (q)RT-PCR, and normalized to β-galactosidase expression.

Western blot analysis was carried out according to standard procedures, using monoclonal M2 α-Flag (Sigma), α-Myc 9E10 (Santa Cruz) and α-HA (CoVance), and polyclonal α-phosphoSmad1/5/8 (Cell Signaling), α-phosphoSmad5 or α-Smad5 (Epitomics), α-phosphoSmad2 (a gift from P. ten Dijke, LUMC, The Netherlands), α-Foxh1 (R&D), α-tubulin (ProBio) primary antibodies and HRP-conjugated secondary antibodies (Jackson ImmunoResearch). Chemiluminescence [ECL reagent (Perkin Elmer) or ECL advance (GE Healthcare Biosciences)] was detected with a LAS-3000 Imager (Fujifilm) and quantified with AIDA Image Analyzer V.4.2.2-software (Raytest).

HEK293T cells were cultured in 9 cm dishes and transfected as indicated using polyethyleneimine (PEI; Sigma), stimulated the next day with 25 ng/ml rhBMP4 for 8 hours, and lyzed using 170 mM NaCl, 10 mM EDTA, 50 mM NaF, 50 mM Tris-HCl (pH 7.4) and 0.5% Nonidet-P40 supplemented with 0.8 mM Na₄P₂O₇.10H₂O, 4 mM Na₃VO₄ and complete protease inhibitor (Roche). Lysates were incubated with the α-HA antibody overnight at 4°C, followed by precipitation of complexes with protein G Sepharose beads (GE Healthcare Biosciences) for 3 hours at 4°C. Beads were washed with lysis buffer and protein solubilization was in 2× SDS-containing sample buffer. Immunoprecipitates were analyzed by western blotting.

Endogenous Smad2-Smad5 interactions were demonstrated in human embryonic stem cells (H9) and MCF-7 cells. H9 cells were induced to undergo initial steps of mesendoderm/liver differentiation as described previously (Roelandt et al., 2010). At day 4 of differentiation, cells were stimulated with 50 ng/ml activin (R&D) or 50 ng/ml activin and 50 ng/ml rhBMP4 for 90 minutes. Serum-starved MCF-7 cells (DMEM 0.1% FCS) were transfected with ON-TARGET plus SMARTpools for Smad5 and Smad1 siRNA (Dharmacon) or non-targeting control siRNA (final concentration of 10 nM), and the following day they were ligand stimulated for 60 minutes.

Lysis of H9 or MCF-7 cells was with 100 mM KH₂PO₄, 0.2% Triton-X100, 1 mM DTT, 0.8 mM Na₄P₂O₇.10H₂O, 4 mM Na₃VO₄ and complete protease inhibitor (Roche). Sepharose G pre-cleared lysates were overnight incubated with anti-Smad2/3 antibody (R&D, AF3797) at 4°C, followed by precipitation of complexes as described for HEK293T cells.

Serum-starved MCF-7 cells cultured on coverslips were stimulated with 100 ng/ml activin and 50 ng/ml rhBMP4 for 30 or 60 minutes and subjected to in situ PLA using Duolink Detection kit II (Olink Bioscience) using primary antibodies against Smad2 (Abcam, AB71109) and C-terminally phophorylated Smad1/5/8 (Cell Signaling, 9511S). Heteromers were counted in three fields per coverslip and three wells were analyzed per condition (Moya et al., 2012). Quantifications are given as mean±s.e.m. Statistical evaluation was carried out using Student’s t-test (P<0.05 was considered significant; ^***P<0.001).

Amnion samples from wild-type and Smad5 knockout littermates were micro-dissected in ice-cold PBS at neural plate to headfold stages, followed by snap-freezing. Reverse transcription was performed in the sample lysate using SuperScript III Reverse Transcriptase (Invitrogen) and poly-dT universal primer 1. After 3′ poly-A tailing with terminal deoxynucleotidyl transferase (Invitrogen), second-strand cDNA was produced using poly-dT universal primer 2, followed by amplification consisting of 18 and 12 PCR cycles using poly-dT universal primers 1 and 2, and Ex Taq HS polymerase (Takara Bio) and column purification (QIAquick, QIAGEN). This procedure was adapted from double-stranded cDNA synthesis protocol from a single cell (Applied Biosystems demonstrated protocol). qPCR was performed in technical duplicates in 96-well plates using an ABI7000 instrument, each well containing 25 ng target cDNA, specific primer pair and Platinum SYBR Green qPCR SuperMix-UDG with ROX (Invitrogen). Primers (supplementary material Table S1) were in silico validated using RTPrimerDataBase (www.rtprimerdb.org). Relative RNA levels were calculated using the ΔΔCt method with Gapdh and Ubc as reference genes.

The development of an ectopic agglomerate of cells consistently located in the anterior midline quadrant of the amnion is a phenotype unique to Smad5^m1/m1 mutant mouse embryos. We hypothesized that the amniotic clump in Smad5 mutants is an ectopic primitive streak-like structure. Analysis of the expression of several markers normally expressed in emerging mesoderm at the early primitive streak – brachyury (T), eomesodermin (Eomes), Fgf8 and goosecoid (Gsc) – demonstrated that these markers were ectopically expressed in the clump in embryonic day (E) 8.5 Smad5 mutants (Fig. 1A-D). Furthermore, analysis of mutant amnions at E7.5 by qRT-PCR demonstrated not only ectopic T and Gsc expression in most mutant amnions analyzed, but also ectopic expression of other primitive streak markers such as Mixl1 and Foxa2 (supplementary material Fig. S1A). This suggests that the clump of cells in the amnion of Smad5 mutant embryos is an ectopic primitive streak-like structure.

In the primitive streak, epiblast cells ingress while undergoing EMT, a multi-pathway process associated with loss of E-cadherin and development of Snail-positive mesoderm (Tam and Loebel, 2007). To evaluate presence of ectopic EMT, we compared the localization of E-cadherin and Snail in amnion of Smad5 mutant and control littermates. In wild-type amnion, E-cadherin localizes basolaterally in amniotic ectoderm shortly after amnion closure, whereas E-cadherin relocalizes mainly to the basal side of amniotic ectoderm when the ectoderm acquires its squamous architecture (supplementary material Fig. S1B,D). In contrast to the primitive streak region, Snail was not only detected in amniotic mesoderm but also in ectoderm of control amnion (supplementary material Fig. S1C,E, arrowheads). This observation concealed the analysis of (stereotype) EMT in the Smad5 mutant amnion. Nonetheless, the staining for Snail appeared less uniform in the amniotic ectodermal side of the amniotic clump in Smad5 mutant embryos (supplementary material Fig. S1D,F, outlined area). Such areas matched with regions of high E-cadherin in adjacent sections of the amniotic clump (supplementary material Fig. S1B,D). The amniotic mesodermal side of the clump facing the exocoelom was devoid of E-cadherin. Overall, the E-cadherin distribution seemed compatible with the amniotic clump in Smad5 mutant embryos, resembling an ectopic primitive streak-like structure.

Expansion of the primitive streak or ectopic primitive streak formation is often associated with increased/ectopic Nodal expression and signaling. This has, for example, been demonstrated in compound mutants of the Nodal antagonists Cer1 and Lefty1, which are normally secreted by the AVE (Perea-Gomez et al., 2002). Similarly, mice deficient in Lefty2, which is normally expressed in LPM at gastrula stages, develop an expanded primitive streak associated with increased Nodal signaling (Meno et al., 1999).

Nodal expression is undetectable in control amnion but Nodal mRNA becomes ectopically expressed in the amnion of Smad5^m1/m1 embryos from the neural plate (NP, E7.5) stage onwards, well before the first visible thickening of the Smad5 mutant amnion (Fig. 2A). At the one-somite stage (1S, E8.0), ectopic Nodal was consistently observed in the anterior midline quadrant of the not yet thickened mutant amnion near the junction with anterior ectoderm (Fig. 2B), coinciding with where the amniotic clump would form. This suggests that ectopic Nodal presages the formation of the amniotic clump. Later, at the >5S stage (E8.5) when the amnion thickening is clearly discernible, ectopic Nodal remained detectable in the amniotic clump, in the left and right side of the cardiac crescent, and frequently in the distal tip of the allantois (5/7 embryos, Fig. 2C, supplementary material Fig. S2). These observations were confirmed by staining for β-galactosidase in Smad5-knockout embryos obtained after crossing Smad5^m1/+ heterozygotes with Smad5^m1/+;Nodal^lacZ/+ reporter mice (Fig. 2D,E). Reducing the allele dose of Nodal in these Smad5^m1/m1;Nodal^lacZ/+ embryos did not rescue the amnion defect (Fig. 2D,E). The three ectopic expression domains of Nodal in E7.5-E8.5 Smad5 mutant embryos (Fig. 2A-C and supplementary material Fig. S2) overlapped with reported Bmp2 (Ishimura et al., 2008; Madabhushi and Lacy, 2011) and Bmp4 (Lawson et al., 1999) expression domains, and intense BMP/Smad activation in a BRE-lacZ reporter mouse (Monteiro et al., 2004) that monitors BMP/Smad1-5 signaling (Fig. 2F). These observations show that Nodal expression is regulated by BMP signaling in the amnion, but probably also in other regions of the embryo.

Chimeric embryos resulting from aggregation of Smad5^m1/m1 ES cells and diploid morulae still developed amniotic clumps despite the presence of wild-type visceral endoderm/AVE (Bosman et al., 2006). This demonstrates that the formation of a primitive streak-like structure in the amnion is not the indirect consequence of the absence of Smad5 in AVE. Nonetheless, we analyzed the expression of mRNAs encoding the AVE secreted antagonists and also Lefty2 in Smad5 mutant embryos to determine whether the mechanisms that normally prevent Nodal expression and signaling in the anterior side of the embryo proper were affected. Analysis of the expression of Cer1, Lefty1, Dkk1 and Lefty2 revealed that their domains were similar in Smad5 mutant and control littermates (Fig. 3A-D). Furthermore, comparable levels of expression of markers for most rostral ectoderm (Dlx5) and early forebrain (Hesx1) in Smad5 mutant and control embryos, suggest that patterning of the anterior part of the embryo proper is preserved at this stage (Fig. 3E,F). This indicates that the known mechanisms of Nodal antagonism in the anterior side of the embryo are not affected in the Smad5 mutants. Together, these data suggest the existence of a BMP/Smad5-dependent mechanism of regulation of Nodal expression in the amnion that prevents the formation of an ectopic primitive streak-like structure in this extra-embryonic tissue.

Smad1-mediated BMP signaling was shown to affect Nodal expression in LPM by limiting the availability of Smad4 and a concomitant interference with the Nodal-Smad2/4-Foxh1-dependent autoregulatory loop (Furtado et al., 2008). To determine whether Smad5-mediated BMP signaling can antagonize the Nodal autoregulatory loop, we made use of a Nodal/Activin-responsive luciferase reporter, which contains Foxh1-and Smad2/4-binding sites from the Mix2 promoter (Watanabe and Whitman, 1999). The effect of gain-and loss-of-function of Smad5-dependent BMP signaling was tested in these reporter assays. Notably, we used ligands to activate the signaling pathways, whereas others investigated the role of BMP/Smad1 signaling by using overexpression of constitutively active type I receptor/Alk variants (Furtado et al., 2008).

Using transfected HEK293T cells, co-production of Foxh1 and Cripto resulted in only mild activation of the reporter (Fig. 4A, lane 2), whereas co-synthesis of Nodal resulted in an ∼12-fold increase in luciferase activity (Fig. 4A, lane 3). Stimulation with BMP4 for 5 or 20 hours progressively reduced the response to Nodal (Fig. 4A, lanes 4 and 5), an effect that has previously been reported (Osada et al., 2000). Repression of the reporter activity by BMP4 was further potentiated when Smad5 was co-produced (Fig. 4A, lanes 6 to 8). Transfection of a vector encoding a Smad5 mutant protein (Δ-Smad5) that can no longer be phosphorylated in its C-terminal SSXS domain, did not repress reporter activity (Fig. 4A, lanes 9 to 11). Furthermore, it tempered the phosphorylation of the endogenous Smad5, indicating that Δ-Smad5 probably functions as a dominant-negative Smad5 variant that interferes with the receptor-mediated activation of endogenous Smad5 (Fig. 4A). Similarly, silencing of endogenous Smad5 by small interfering RNA (siRNA) potentiated Nodal signaling (Fig. 4B, lane 6). Synthesis of exogenous mouse Smad5 in siRNA-treated cells restored the inhibitory effect of Smad5 on Nodal signaling (Fig. 4B, lanes 7 and 8). These results demonstrated that activated Smad5 signaling tempers Nodal signaling, and that C-terminal activation of Smad5 is essential to repress Nodal signaling.

Interestingly, activation by the liganded Nodal receptor complex seemed undisturbed by BMP signaling because high levels of phospho-Smad2 were maintained under all conditions where Nodal was present, regardless of stimulation with BMP (Fig. 4A). This excludes the possibility that BMP/Smad5 signaling is upstream of the expression of secreted Nodal antagonists or interferes with activation of the Nodal receptor complex. Activated Smad5 may limit the availability of Smad4 to activated Smad2, as proposed previously for BMP/Smad1 signaling during establishment of LR asymmetry [see above (Furtado et al., 2008)]. However, in our experimental setup, stimulation with ligand at increasing doses of co-transfected Smad4 did not restore the BMP4-mediated repression on the response to Nodal (Fig. 4C). This indicated that, in this case, Smad5 does not limit the availability of Smad4 to Smad2 upon activation of the latter by Nodal.

Next, we examined the effect of increasing quantities of Smad5 or Smad1 in the luciferase assay in the presence of exogenous Smad4. Activated Smad1 and Smad5 inhibited Nodal signaling in a dose-dependent manner (Fig. 4D). The lower decrease in Nodal signaling with Smad1, when compared with Smad5, correlated with the lower phosphorylation status of Smad1 (Fig. 4D, considering the ratio of phospho-Smad1/5/8 over tubulin in the α-phospho-Smad1/5/8 and α-tubulin blots). This suggests that activated Smad1 and Smad5 have similar potential to antagonize Nodal signaling in vitro.

Blocking protein translation with cycloheximide prior to stimulation with BMP4 (supplementary material Fig. S3) did not affect repression of Nodal-dependent luciferase expression by activated BMP/Smad5 (Fig. 5A, lane 6). In our experimental setup, Nodal signaling is therefore directly regulated by BMP/Smad5 signaling.

The persistence of high phospho-Smad2 levels in the reporter assays (Fig. 4A) suggested that Smad5 interferes downstream of Nodal receptor and Smad2 activation. Furthermore, we found that Smad5 did not influence Nodal signaling by limiting the availability of Smad4. We therefore investigated whether activated Smad5 co-immunoprecipitated with Smad2 and whether this interaction would affect co-precipitation of Foxh1 with Smad2. HA-tagged Smad2 was immunoprecipitated from whole-cell extracts of HEK293T cells producing HA-tagged Smad2, Myc-tagged Foxh1 and 3xFlag/Strep-tagged Smad5 or 3xFlag/Strep-tagged Δ-Smad5 in the presence or absence of BMP. Foxh1 co-immunoprecipitated with Smad2 when Nodal signaling was active (Fig. 5B, lanes 3 and 7, panel HA-Smad2 IP). Smad5 co-precipitated with Smad2; however, upon stimulation with BMP4, co-immunoprecipitation of Smad5 – but not Δ-Smad5 – with Smad2 was enhanced (Fig. 5B, lanes 4 and 8, respectively). This is consistent with the reporter assays, where phosphorylation of Smad5 was found to be required to decrease Nodal signaling. Furthermore, BMP4/Smad5 signaling negatively impacted on the co-immunoprecipitation of Foxh1 with Smad2 (Fig. 5B, lane 4). The condition with BMP4/Δ-Smad5 seems also to reduce co-immunoprecipitation of Foxh1 with Smad2 (Fig. 5B, lane 8), which can be explained by remaining activation of endogenous Smad1/5 in HEK293T cells. These data indicate that activation of the BMP pathway resulted in interaction of Smad5 with Smad2, which in turn interfered with Smad2-Foxh1 complex formation.

Subsequently, endogenous Smad5-Smad2 interactions were detected in human H9 embryonic stem cells (ESCs) and breast cancer MCF-7 cells. To obtain H9 cells that were fully BMP4 and activin responsive, cells were differentiated towards mesendoderm using a liver differentiation protocol (Roelandt et al., 2010). On day 4 of the differentiation, cells were co-stimulated with activin and BMP4, or with activin alone for 90 minutes. Smad2/3 was immunoprecipitated from whole-cell extracts. Co-immunoprecipitation of (phospho)Smad5 with Smad2/3 was observed when cells were co-stimulated with BMP4 and activin (Fig. 5C).

Endogenous ligand-induced Smad5-Smad2 interactions were also demonstrated in MCF-7 cells (Yum et al., 2009), and the reduced presence of such complexes in MCF-7 cells upon siRNA mediated knock-down of Smad5 and Smad1 further demonstrated the specificity of the Smad5-Smad2 complexes (Fig. 5D). The impact of Smad5-Smad2 complex formation on Smad2-Foxh1 heteromerization was – in contrast to the observations in overexpression (Fig. 5B) – marginal in H9 and MCF7 cells. In addition to the endogenous co-immunoprecipitation experiments, Smad5 and Smad2 interactions were also visualized and quantified by in situ proximity ligation assay (PLA) (Fig. 5E). The co-stimulation with activin and BMP4 resulted in a threefold increase in the number of Smad5-Smad2 heteromers when compared with basal levels (Fig. 5E).

The lack of effective tools to detect endogenous Smad5-Smad2 interactions in the mouse embryo prohibited analyzing whether Nodal and BMP/Smad5-mediated signaling results in direct Smad5-Smad2 interaction in the amnion, which would support the BMP/Smad5-mediated antagonism of the Nodal autoregulatory loop in vivo. Therefore, we performed a comparative gene expression analysis on RNA isolated from individual amnions of Smad5 mutant and control littermates from neural plate until headfold stages, well before morphological appearance of the cell clump in mutant amnion, and tested the expression of Nodal and Lefty2 as target genes of the Nodal autoregulatory loop. In addition, the expression of Nodal, Bmp4 and Wnt3 was used as read-out of activation of the slow positive feedback loop.

Ectopic Nodal was detected in most of the Smad5 mutant amnions analyzed (20/26) (Fig. 6A). Ectopic Lefty2 was also detected in the majority of the mutant amnions but in fewer embryos than Nodal (15/26). In situ hybridization analysis demonstrated ectopic Lefty2 in the amniotic clump, in allantois and both sides of the cardiac crescent at E8.5 (Fig. 6C), which correlated with ectopic Nodal at a similar stage (Fig. 2C, supplementary material Fig. S1). Altogether, these results support the involvement of Smad2/4-Foxh1 signaling in the mutant amnion and are compatible with Nodal autoregulatory loop activity.

None of the amnions analyzed here showed differential Bmp4 expression (Fig. 6A), which contrasted to increased Bmp4 in Smad5 mutant amnion at E7.5 that was detected by radioactive in situ hybridization (Bosman et al., 2006). Unlike the unchanged Bmp4 levels, Wnt3 expression is ectopically expressed in nearly half of the Smad5 mutant amnions analyzed by RT-PCR (12/26) (Fig. 6A). Indeed, in situ hybridization analysis confirmed ectopic Wnt3 in the mutant amnion at headfold stages and demonstrated that it persisted in the amnion thickening at E8.5 (Fig. 6D,E). Furthermore, crossing of the Smad5^m1/+ with the Lef/Tcf-dependent Wnt signaling reporter mouse (Smad5^m1/+;BAT-gal^+/–) showed that ectopic canonical Wnt signaling occurred in Smad5-deficient amnion (Fig. 6F-H). Taking into consideration that canonical Wnt3 signaling induces Nodal via an identified proximal epiblast enhancer (PEE) in its promoter containing Lef/Tcf-binding sites (Ben-Haim et al., 2006; Granier et al., 2011), our data are also in agreement with the presence of a functional slow positive-feedback loop in Smad5 mutant amnion.

The higher prevalence of amnions expressing ectopically target genes from the Nodal-Smad2/4-Foxh1-dependent feedback loop, namely Nodal and Lefty2, rather than Wnt3 (slow positive-feedback loop component), is compatible with induction of ectopic Nodal by the fast autoregulatory loop in the Smad5 mutant amnion. This became more evident when Smad5 mutant amnions were divided into classes according to their ectopic expression profile (Fig. 6B). Three of the Smad5^m1/m1 amnions had no ectopic expression of any of these genes, suggesting that the defect had not been initiated in these embryos at the time of amnion isolation. Three mutant amnions were found to express only Wnt3 ectopically, or both Wnt3 and Lefty2 ectopic expression but not Nodal. During dissection, our efforts to prevent contamination of amnion samples with non-amnion tissue may have resulted in 3/26 samples in loss of the ectopic Nodal expression domain, which is typically restricted to the anterior most part of the amnion. From all the Smad5 mutant amnions with ectopic Nodal expression, 11 expressed only Nodal-Smad2/4-Foxh1-dependent target genes (Nodal ± Lefty2), suggesting that ectopic Nodal in the Smad5 mutant amnion is induced via the Smad2/Foxh1-dependent autoregulatory loop. The remaining nine mutant amnions co-expressed Wnt3 (Fig. 6B), suggesting that the slow positive-feedback loop contributes to the amplification of ectopic Nodal in the mutant amnion. This observation, however, does not exclude the possibility that both loops initiate simultaneously in the Smad5 mutant amnion, but as the slow positive-feedback loop requires the expression and signaling of the effector proteins, it probably takes longer to have an effect on Nodal expression. In conclusion, these data support that, in the absence of Smad5, the ectopic Nodal in the mutant amnion is induced by the Nodal autoregulatory loop and is amplified via the slow positive-feedback loop.

The lack of Smad5 in mouse embryos results in ectopic Nodal expression and signaling in the anterior part of the amnion, leading ultimately to the formation of an ectopic primitive streak-like structure therein. Before appearance of the clump in the Smad5 mutant amnion, ectopic Nodal localizes in cuboidal amniotic ectoderm at the anterior midline quadrant of the amnion, whereas the remainder of the mutant amniotic ectoderm acquires the characteristic squamous amniotic epithelium architecture. The (1) altered cellular architecture of the amniotic ectoderm in the anterior midline quadrant of the mutant amnion, (2) the ectopic presence of mesoderm inducers herein, (3) the subsequent presence of several mesoderm markers, including Eomes (which is required in the primitive streak region to downregulate E-cadherin and allow EMT) (Arnold et al., 2008), and (4) the altered regionalized localization of E-cadherin and Snail are compatible with the amniotic clump in the Smad5 mutant embryo being an ectopic primitive streak-like structure.

Development of (multiple) ectopic primitive streak(s) in the epiblast has been described in Cer-l;Lefty1 compound mutants (Perea-Gomez et al., 2002). Analysis of AVE secreted antagonists and neural plate/early forebrain markers (Thomas and Beddington, 1996) demonstrated that the Smad5 mutant AVE is well established, maintained and functional. These data complement our former analysis of chimeric embryos that developed amniotic clumps even when AVE/visceral endoderm lineages were derived from wild-type recipient embryos (Bosman et al., 2006), suggesting that ectopic Nodal is not associated with an impaired AVE.

Ectopic expression of T has been shown in amnion of Lefty2-deficient embryos (Meno et al., 1999) and, hence, we reasoned that Lefty2 may diffuse from the nascent mesoderm and antagonize Nodal signaling in amnion. However, the expression of Lefty2 in LPM appeared unaffected in Smad5 mutants at late NP stage. Collectively, these data suggest that BMP/Smad5-mediated signaling antagonizes Nodal signaling in the amnion, preventing the propagation of Nodal-regulating loops that may ultimately lead to formation of ectopic primitive streak-like structures.

Removal of one allele of Nodal in a Smad5 mutant background (Smad5^m1/m1;Nodal^lacZ/+) did not rescue the amnion phenotype or the different ectopic Nodal expression domains. This contrasts with the partial rescues reported when crossing this same Nodal^lacZ mouse strain with various other loss-of-function mice with increased Nodal and concomitant enforced mesoderm formation, including mutants of the AVE-secreted Nodal antagonists, and of Drap1, Lefty2 and Cyp26 (Iratni et al., 2002; Meno et al., 1999; Perea-Gomez et al., 2002; Uehara et al., 2009). The rescue of the Lefty2 mutants by Nodal haploinsufficiency is, however, only modest. The lateral expansion of the expression domain of mesoderm markers expressed early during primitive streak formation was maintained in Lefty2 mutants, but ectopic expression of these same marker genes in more anterior regions or even the amnion was lost (Meno et al., 1999). So, reducing the Nodal signaling levels by (about) half in the amnion seems sufficient to block defects caused by the absence of Lefty2. These findings are consistent with the existence of a BMP/Smad5-mediated mechanism that contributes normally to neutralize Nodal signaling in the amnion at E7.5.

Our data show that BMP-dependent antagonism of the Nodal autoregulatory loop is not restricted to Smad1 (Furtado et al., 2008). Indeed, BMP/Smad5 attenuates Smad2/4-Foxh1-mediated Nodal signaling in cultured cells; interfering with BMP/Smad5 activity, either by siRNA-mediated silencing of endogenous Smad5 or by co-synthesis of a dominant-negative Smad5 variant, results, respectively, in increased or sustained high levels of Nodal signaling. Conversely, stimulation with BMP and activation of endogeneous Smad5 results in decreased Nodal signaling. Activated Smad5 directly regulates Nodal signaling downstream of Smad2 phosphorylation. Mixed activated Smad complexes between the BMP-Smads and TGFβ/Activin/Nodal-Smads in response to stimulation of epithelial cells with TGFβ have previously been reported (Daly et al., 2008; Gronroos et al., 2012). According to the present co-immunoprecipitation data on overexpression, co-activation of BMP and Nodal/Activin pathways induces the formation of Smad5-Smad2 complexes; in addition, such complexes are also demonstrated to occur endogenously in two different cell lines by two independent methods. We demonstrate that these complexes result in reduced Smad2-Foxh1 interaction and a concomitant reduced signaling via the Nodal Smad2/4-Foxh1-dependent autoregulatory loop in overexpression conditions.

Considering that we are technically unable to determine in the mouse embryo whether antagonism of the Nodal autoregulatory loop by BMP/Smad5 in the amnion requires Smad2-Smad5 heteromerization, the possibility remains that BMP/Smad5 affects Nodal signaling by competing for Smad4, as has been proposed for BMP/Smad1 (Furtado et al., 2008). Alternatively, BMP/Smad5 may compete for other shared transcription factors or adapter proteins and/or antagonize Nodal signaling indirectly [e.g. Jmjd3 (Kdm6b – Mouse Genome Informatics) (Dahle et al., 2010), Cited2 (Lopes Floro et al., 2011) and Xmsx-1 (Yamamoto et al., 2001b)]. Regardless of the mechanism, our data indicate that BMP/Smad5 signaling is required, in addition to Lefty2 (Meno et al., 1999), to antagonize Nodal signaling in amnion (Fig. 7A). Lefty2 that is induced in LPM by Nodal signaling has higher diffusion properties than Nodal ligand itself (Sakuma et al., 2002) and may therefore restrict Nodal signaling efficiently in the majority of the epiblast at E7.5, including in the amnion. As a consequence, Nodal expression becomes confined to the node region and the amnion acquires its typical squamous architecture (Fig. 7B). In the Smad5 mutants, minute availability of Nodal signaling in anterior Smad5-deficient amnion can be expected to lead to ectopic Nodal induced via the Nodal-Smad2/Foxh1-dependent autoregulatory loop and is amplified by the slow positive-feedback loop (Fig. 7C,D). The signaling by the major primitive streak inducers in Smad5 mutant amnion leads to ectopic primitive streak-like formation. This mechanism is probably more important in the anterior midline quadrant of the amnion, which is the most distant from the source of Lefty2, where the primitive streak-like structure forms in the Smad5 mutant. Alternatively, the focal localization of this structure may reflect the different properties and origin of the (mutant) anterior amniotic ectoderm cells that are localized at the anterior delineation between neural, surface and amnion ectoderm at the extra-embryonic/embryonic junction. Furthermore, Bmp2 expression has recently been demonstrated in visceral endoderm close to the anterior extra-embryonic/embryonic junction, coordinating morphogenetic cell behaviors in multiple epiblast-derived lineages (Madabhushi and Lacy, 2011). The Bmp2- and Smad5-null mice both display anterior amnion morphogenesis defects (Chang et al., 1999; Zhang and Bradley, 1996) and hence it is plausible that impaired BMP2/Smad5-mediated signaling underlies the focal localization of the primitive streak-like structure in Smad5 mutant amnion.

Interestingly, Nodal is also ectopically expressed in the Smad5 mutant distal allantois and cardiac crescent, which are also regions undergoing active BMP/Smad signaling, as revealed in BRE-lacZ embryos. This suggests that an interplay between Nodal and BMP/Smad(5) signaling may also occur in other sites where Nodal and BMP co-signal. The distal allantois and cardiac crescent may, however, have lost competence to primitive streak-inducing signals at E8.0.

Detection of ectopic Lefty2 by qRT-PCR shortly after the onset of ectopic Nodal in the Smad5-deficient amnion indicates that the Smad2-Foxh1-dependent negative-feedback loop is also activated in the mutant amnion. This may explain why mutant amniotic mesoderm becomes progressively patterned into more BMP-dependent posterior primitive streak derivatives at E8.5 (Bosman et al., 2006). Previously, we reported that gain-of-function of BMP signaling by injection rhBMP4 into the amniotic cavity of wild-type embryos could partially phenocopy the initial phase of amniotic clump formation (Bosman et al., 2006). Small rudimentary amnion thickenings were observed in a fraction of the injected embryos. Although gene expression analysis was not performed, histomorphological analysis was not indicative of ectopic primitive streak formation and mesoderm patterning. Injected BMP4 alone is probably insufficient to induce formation of ectopic primitive streak-like structures in amnion.

It is remarkable that Smad5 mutant amnion is still competent for primitive streak induction at E7.5. This may reflect the fairly naïve state of the amnion at that stage, which could even persist until later stages. Indeed, human and rodent term amnion were recently described to harbor resident stem cell-like cells (Dobreva et al., 2010); whether such cells are functional stem cells retaining the potential to contribute to amnion homeostasis is unclear. Here, we provide evidence that BMP/Smad5 signaling is a pivotal regulator of early amnion fate through insulation of the amnion from Nodal signaling.

We thank all Zwijsen/Huylebroeck team-members: K. Beets, S. Thysen, P. Hembrechts and L. Ordovas for assistance; M. Missoul for mouse husbandry; S. Piccolo, J. Collignon, R. Monteiro and C. Mummery for reporter mice; J. Debruyn and M. Chen for probe/plasmids; P. ten Dijke for anti-phospho-Smad2 antiserum; and K. A. Lawson for instructive discussions.