Differential requirements for Smad4 in TGFβ-dependent patterning of the early mouse embryo

Members of the transforming growth factor β (TGFβ) superfamily of signaling molecules have been implicated in a diverse array of biological processes, including cell growth, differentiation, apoptosis and carcinogenesis (reviewed by Massagué, 1998; Massagué et al., 2000). In the early vertebrate embryo they are indispensable for elaborating the basic body plan and at later stages play a variety of roles during organogenesis and tissue homeostasis. Comprising over 30 different ligands,the superfamily includes TGFβs, activins, nodal, bone morphogenetic proteins (Bmps), and growth and differentiation factors (GDFs). Despite the large number of ligands and their diverse activities, these signals are received and transduced by a relatively small number of cell-surface receptors and downstream effectors.

The receptor complex is composed of two distinct transmembrane serine-threonine kinases termed type I and type II receptors. Upon ligand binding, the activated receptor complex phosphorylates members of the Smad family. The receptor-associated Smads (R-Smads) function as both signal transduction mediators and transcription factors essential for TGFβsignaling and can be subdivided into two groups. While Smad1, Smad5 and Smad8 are phosphorylated in response to Bmp signals, Smad2 and Smad3 are phosphorylated in response to activation of the TGFβ, activin and Nodal pathways. Phosphorylation of the R-Smads allows their association with the common-mediator Smad, Smad4, resulting in nuclear translocation and formation of higher order transcriptional complexes. Current models emphasize the central requirement of Smad4 for the regulation of TGFβ target genes. By itself or in combination with R-Smads, Smad4 weakly binds DNA and only moderately activates transcription. For robust gene activation or suppression,heteromeric Smad4-containing complexes must associate with additional nuclear co-factors or tissue-specific transcription factors such as FoxH1 or OAZ(reviewed by Attisano and Wrana,2000; Moustakas et al.,2001).

During early vertebrate development, the Nodal signaling pathway plays a conserved role in the establishment of the anteroposterior (AP) axis,gastrulation and left-right patterning (reviewed by Schier, 2003; Whitman, 2001). In the early mouse embryo, Nodal signals from the epiblast are transduced in the overlying visceral endoderm (VE) to establish the early anterior organizer known as the anterior visceral endoderm (AVE) (Lu et al., 2001). The AVE acts prior to gastrulation to impart early anterior character in the adjacent epiblast. Nodal mutants fail to pattern the AVE and arrest prior to gastrulation(Brennan et al., 2001). Similarly, null mutations within the Nodal receptor complex, either the type I receptor ALK4 or the type II receptors ActRIIA and ActRIIB, result in embryos that do not gastrulate and overtly resemble Nodal mutants(Gu et al., 1998; Song et al., 1999). Continued Nodal signaling during gastrulation instructs epiblast cells passing through the anterior primitive streak to become definitive endoderm (DE), prechordal plate, node and notochord. Downstream of Nodal signaling, loss of Smad2 in the epiblast or reducing both Smad2 and Smad3 gene dose yield phenotypes indistinguishable from Nodal hypomorphic mutations (Dunn et al., 2004; Lowe et al.,2001; Norris et al.,2002; Vincent et al.,2003). Similarly, loss of Foxh1, which encodes a forkhead transcription factor that associates with Smad2/3 and Smad4, yields phenotypes that closely resemble those resulting from disruption of the Nodal signaling pathway (Schier, 2003). Taken together, these biochemical and genetic results underscore the significance of the Smad pathway in transducing Nodal signals during early embryogenesis.

Signaling via the Bmp pathway is also crucial in early embryonic development, but loss-of-function phenotypes are much more varied (reviewed by Zhao, 2003). Mutations in the type I receptor Bmpr1a/Alk3 and type II receptor Bmpr2 lead to a block prior to gastrulation. Inactivation of Bmp4 results in a reduction in extra-embryonic mesoderm characterized by a lack of an allantois and loss of primordial germ cells (PGCs)(Lawson et al., 1999). Loss of Bmp2 and Bmp8b also affect allantois development and PGC number (Loebel et al., 2003). Mutations in the Bmp effectors Smad1 and Smad5 result in allantois, PGC and yolk sac defects as well as impaired angiogenesis. In addition, genetic analysis has implicated a number of Bmps in heart development including Bmp2, Bmp4, Bmp5, Bmp6 and Bmp7. Compared with the early gastrulation defects associated with Bmp receptor mutations, the phenotypes resulting from loss of individual Bmp ligands or their corresponding downstream Smads are more moderate, perhaps reflecting functional redundancy among Bmp or Smad members.

Consistent with the abundant biochemical data that suggest a central role for Smad4 in the transduction of all TGFβ-related signals, mouse embryos homozygous for a Smad4 null allele arrest prior to gastrulation(Sirard et al., 1998; Yang et al., 1998). Mutant embryos show disorganization by embryonic day (E) 6.5 and do not form mesoderm, a phenotype more severe than that of any individual R-Smad mutation. Limited analysis of Smad4-null embryos indicates that the initial requirement for Smad4 activity resides in the extra-embryonic lineages:aggregation of Smad4-deficient embryonic stem (ES) cells with tetraploid wild-type embryos gives rise to chimeric embryos that express the mesodermal marker T and show limited formation of mesodermal derivatives such as somites (Sirard et al., 1998). However, the specific roles of Smad4 within the developing embryo proper have not been well characterized.

We have selectively manipulated Smad4 gene activity in the epiblast and its derivatives in order to address the combined roles of TGFβ-related signaling during gastrulation and specification of the definitive embryonic lineages. We generated a Smad4 conditional allele that undergoes efficient Cre-mediated DNA recombination within the epiblast in response to the robust Sox2Cre transgene. Loss of Smad4 does not prevent the establishment of the primary AP axis and gastrulation is initiated normally. However, we observe a highly focal defect in patterning of the primitive streak as gastrulation proceeds. Mutant embryos fail to form derivatives of the anterior primitive streak (APS), such as the node,notochord and definitive endoderm, and thus share many phenotypic similarities associated with downregulation of the Nodal/Smad2/3/Foxh1 pathway. Surprisingly, the transduction of extra-embryonic Bmp signals within the epiblast is only moderately impaired in the absence of Smad4. Posterior streak derivatives probably patterned in response to extra-embryonic Bmp signals are induced normally, and mutant embryos consistently form a rudimentary heart tube, yolk sac mesoderm and allantois. By contrast,specification of the germline is Smad4-dependent as PGC formation is dramatically diminished. Taken together, these studies indicate that Smad4 is an essential component of the Nodal signaling pathway that specifies the APS, but is required to transduce only select Bmp signals during early development.

The Smad4 conditional (Smad4^RobCA; referred to throughout as Smad4^CA) and null(Smad4^RobN; Smad4^N) alleles were generated by flanking the first coding exon of Smad4 with loxP sites(Fig. 1A). The targeting construct contains a pgk-dta selection cassette; a 4.2 kb(SalI-StuI) 5′ homology arm containing a loxP site; a loxP flanked pgk-hygromycin cassette; and a 7.0 Kb(StuI-XhoI) 3′ homology arm(Fig. 1A). Linearized vector was electroporated into CCE ES cells, and drug-resistant colonies were genotyped by Southern blot analysis using a 3′ external XhoI-BstZ17I probe. Approximately 6% of the clones were correctly targeted. Southern blot analysis was performed using an internal StuI-AvrII probe on XbaI-digested clones to screen for loss of the pgk-hygromycin cassette and generation of the conditional and exon1-deleted null alleles (Fig. 1C). Two independently targeted ES cell clones were used to generate germ line chimeric mice. In one clone, the pgk-hygromycin cassette was excised by transient expression of Cre-recombinase in vitro; in a second clone, Smad4^TA ES cells were used to generate germline chimeric mice. Smad4^TA/+ mice were bred for in vivo Cre-mediated excision (O'Gorman et al.,1997) of the selection cassette and generation of the Smad4^CA and Smad4^N alleles. The Smad4^CA/CA and Smad4^N/N phenotypes obtained from these two ES cell clones are identical.

Mice, ES cells and embryos were genotyped by PCR analysis of cells lysed in buffer containing non-ionic detergents and Proteinase K. Primers used to genotype Smad4 alleles were F1,5′-CTTTTATTTTCAGATTCAGGGGTTC-3′; F2,5′-AAAATGGGAAAACCAACGAG-3′; and R1,5′-TACAAGTGCTATGTCTTCAGCG-3′. The combination of the F1, F2 and R1 primers amplifies 512 bp, 625 bp and 675 bp fragments from the Smad4^N, wild-type and Smad4^CA alleles,respectively (Fig. 1A,D).

Smad4^CA/CA was maintained on a (129×C57BL/6)hybrid background. Smad4^N/+ and Smad4^CA/CA;ROSA26^R/R were maintained on a partially outbred CD1 background. Sox2Cre/+;Smad4^N/+ mice were generated from crosses between Smad4^N/+ and Sox2Cre/+ parents(Hayashi et al., 2002b) and maintained on a CD1 background.

Whole-mount in situ hybridization was performed according to standard procedures (Nagy et al.,2003), with the following commonly available probes Shh, T,Nodal, Foxa2, Gsc, Hex, Cer1, Otx2, Fgf8, Krox20, Bmp4, Eomes and Oct4 (Brennan et al.,2001; Vincent et al.,2003). Additional probes used for this study: Lim1(Barnes et al., 1994), Mml (Pearce and Evans,1999), and Nkx2.5 and cardiac alpha-actin(Norris et al., 2002). X-gal staining was performed as described by Nagy et al.(Nagy et al., 2003). For Hematoxylin and Eosin histology, embryos were processed, sectioned and stained using standard procedures. Histochemical staining of PGCs and mounting of specimens were performed essentially as described(Lawson et al., 1999).

ES cells were sonicated in 2× sample buffer containing protease inhibitor cocktail (Roche). Protein extract (40 μg per cell line) was separated on a 10% SDS-PAGE gel and electroblotted onto nitrocellulose(Schleicher & Schuell). The membrane was then hybridized overnight at 4°C with a mouse monoclonal antibody showing affinity for the linker and/or MH2 domain of Smad4 (B-8; Santa Cruz) diluted 1:1000, followed by HRP-conjugated sheep anti-mouse IgG and ECL (Amersham).

Embryoid bodies were generated from ES cell lines(Robertson, 1987), harvested at day 5 of differentiation and then every 2 to 3 days until day 12. RNA was isolated using Absolutely RNA Miniprep Kit (Stratagene). RNA (10 μg) was reverse transcribed using Superscript II RNase H-reverse transcriptase(Invitrogen). For all genes analyzed, cDNA reverse transcribed from 0.1 μg of RNA was used for PCR amplification. PCR amplification was carried out using 1 μCi [α-³²P] dCTP per reaction and performed for 22-26 cycles, maintaining linear amplification. The PCR products were separated on a 6% nondenaturing polyacrylamide gel, vacuum dried and subjected to autoradiography. Primer sequences used were as follows: G3PDHF,5′-ACCACAGTCCATGCCATCAC-3′; G3PDHR,5′-TCCACCACCCTGTTGCTGTA-3′; Msx1F,5′-AACCCCTTGCTACACACTTCCTCC-3′; Msx1R,5′-GGACCACGGATAAATCTCTTGGC-3′; Msx2F,5′-GGAGCACCGTGGATACAGGAG-3′; Msx2R,5′-GCACAGGTCTATGGAAGGGGTAG-3′; HnfF,5′-ATGCCTGCCTCAAAGCCATC-3′; HnfR 5′-CCACTCACACATCTGTCCATTGC-3; TransferrinF,5′-GCCATCCCATCACAACAAGGTATC-3′; TransferrinR,5′-CTGCTTCAGATTCTTAGCCCATTC-3′

In order to obtain Smad4 null murine embryonic fibroblasts (MEFs),we crossed the tamoxifen-inducible Cre transgene CAGGCre-ER™(Hayashi and McMahon, 2002)onto the Smad4^N/+ background. Litters of CAGGCre-ER™;Smad4^N/+×Smad4^CA/CAembryos were harvested at E13.5. Embryos were genotyped and fibroblasts were isolated using standard procedures (Nagy et al., 2003). Fibroblasts from CAGGCre-ER™;Smad4^CA/N embryos were pooled as were Smad4^CA/+ littermates. Cre recombinase expression was induced by addition of 1 μM 4-OH-Tamoxifen to the culture medium at passage 2. Cells were passaged twice in the presence of 4-OH-Tamoxifen, and seeded for transfection at passage 5. Excision of Smad4 was verified by PCR genotyping.

Cells were seeded at 3×10⁵ cells/well in 12-well dishes prior to transfection with Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. The following plasmids (0.75 μg) were used for the 3TP-lux assay: 3TP-lux (Wrana et al.,1992) and pCMV5B-MADR4(Macías-Silva et al.,1996). For the Msx2-lux assay, 0.5 μg of the following plasmids were used: Msx2-lux (Sirard et al.,2000), pCMV5B-ALK3/HAQD(Hoodless et al., 1996) and pCMV5B-MADR4. Each well was supplemented with 0.25 μg pRL-CMV (Promega) and an empty CMV expression vector was added to bring the total amount of DNA to 1.75 μg/well. Fifteen hours after transfection, growth medium was replaced with serum-free medium with or without 1 nM TGFβ1 (R&D systems). Cell lysates were harvested 30 hours after addition of ligand, and analyzed using the Dual-Luciferase Reporter Assay System (Promega). Mean values of each set of triplicates were plotted as fold induction compared with activation of the reporter in the absence of ligand or constitutively active receptor.

Smad4-deficient mouse embryos exhibit arrested growth, failure to express early mesoderm markers and severe disorganization by E7.5(Sirard et al., 1998; Yang et al., 1998). Tetraploid analysis indicates that these defects reflect a requirement for Smad4 in extra-embryonic lineages. To evaluate more precisely Smad4 functions during gastrulation, particularly in specification of the definitive embryonic germ layers, we generated a conditional Smad4 allele (designated Smad4^CA) using gene targeting to flank the first coding exon of Smad4 with loxP sites(Fig. 1A-D). ES cells carrying the initial Smad4 targeted allele (Smad4^TA) were identified (Fig. 1B), and in vitro expression of Cre recombinase was used to recover derivatives carrying either Smad4^CA or Smad4^N (null)alleles (Fig. 1C). Smad4^CA heterozygous animals were intercrossed to generate stocks of Smad4^CA homozygotes(Fig. 1D).

As alternate splicing plays a significant role in Smad4 processing and activity (Pierreux et al.,2000), it was important to confirm that deletion of exon 1 generates a null allele. We generated both Smad4^N/N ES cells by retargeting the remaining wild-type allele in Smad4^N/+ ES cells and Smad4^N/N MEFs using a tamoxifen-inducible Cre transgene(Hayashi and McMahon, 2002)and performed functional assays (see Fig. S1). Western blot analysis shows that the predominant wild-type 64 kDa Smad4 protein present in control ES and MEFs is absent in Smad4^N/N cells. Embryoid bodies (EBs) made from Smad4^N/N ES cells were found to have marked downregulation of the visceral endoderm markers Hnf4 and transferrin and the Bmp-responsive genes Msx1 and Msx2, as described for the previously characterized null Smad4 allele(Sirard et al., 1998; Sirard et al., 2000). Similarly, transfection assays confirmed that Smad4^N/NMEFs are unresponsive.

Finally, we examined the phenotype of embryos resulting from Smad4^N/+ intercrosses. Mutant embryos were grossly identifiable at E6.5 and characterized by a shortened proximodistal axis and thickened endoderm (Fig. 2A,B). Histological analysis showed disorganization of the epiblast and overlying visceral endoderm, with no evidence of mesoderm formation(Fig. 2C,D). This phenotype is indistinguishable from the independently generated Smad4^tm1Ari and Smad4^ex8 null mutations (Sirard et al.,1998; Yang et al.,1998). Moreover, aged Smad4^N/+ heterozygous mice developed gastric tumors, similar to those previously reported (data not shown) (Takaku et al.,1999).

Both Smad4^N/N and Nodal-deficient embryos form a truncated egg cylinder with thickened VE and absence of mesoderm formation. This overt similarity prompted us to characterize Smad4^N/Nembryos molecularly. In Smad4^N/N embryos, the pluripotential marker gene Oct4 is maintained in the epiblast, in contrast to Nodal and Smad2^-/-;Smad3^-/- double mutants, in which Oct4 expression is downregulated (Brennan et al.,2001; Dunn et al.,2004). Bmp4, a marker of extra-embryonic ectoderm, is expressed in mutants but displaced distally, suggesting disorganization of embryonic and extra-embryonic tissues (Fig. 2G,H). Similarly eomesodermin, which is expressed in extra-embryonic ectoderm and the early primitive streak, shows disorganization along the proximodistal axis (Fig. 2I,J). In Nodal mutants, both Bmp4 and eomesodermin show transient expression which is downregulated by E6.5. Overall, these experiments suggest that although Smad4^N/Nembryos are morphologically similar to Nodal mutants, they show a greater degree of axial identity.

The failure of Smad4-deficient embryos to gastrulate prevents an analysis of the later roles of Smad4 during embryonic development. To characterize further the requirement of Smad4 during early gastrulation, we used the Sox2Cre deleter strain to specifically inactivate Smad4 in the epiblast(Hayashi et al., 2002b; Vincent et al., 2003). The Sox2Cre transgene exhibits activation in blastocyst outgrowths and is strongly expressed in the early epiblast prior to E5.75 (data not shown). Sox2Cre was first introduced into the Smad4^N/+line of mice, and Sox2Cre/+;Smad4^N/+ and Smad4^CA/CA animals were subsequently intercrossed. PCR analysis of tissue fragments confirms that rearrangement of the Smad4conditional allele efficiently occurs in a Cre- and cell lineage-specific manner (Fig. 3P).

Mutant Sox2Cre/+;Smad4^CA/N embryos, hereafter designated as Smad4 mutants, were indistinguishable from control littermates prior to E6.5 (Fig. 3A,A′). At E7.0, mutant embryos have visible anteroposterior(AP) polarity, with formation of the primitive streak(Fig. 3B,B′). However, as gastrulation proceeds, the AP axis of the embryo is broadened, and by E7.5 and early head fold stages the embryo adopts an elongated and distorted`boot-shaped' morphology (Fig. 3C,C′). In contrast to control littermates, Smad4mutants show rudimentary anterior neural development with no visible headfolds and recognizable foregut invagination. At E8.5 mutants reproducibly exhibit a flattened ventral surface with no visible formation of fore- or hindgut, and caudally displaced, bulbous neural tissue(Fig. 3D,D′,E,E′). Somites are formed but are fused across the midline. Other mesodermal derivatives also form in Smad4 mutants, including allantois, a heart-like structure, blood islands, vasculature and lateral plate mesoderm. By E9.5, mutant embryos show significant degeneration, and none are recovered at E10.5 (data not shown). These results indicate that mesoderm induction in response to Nodal and Bmp pathways does not require Smad4.

To examine axis formation and patterning in more detail, Smad4mutants were examined histologically and by whole-mount in situ hybridization. At E7.5, the primitive streak is evident both in cross section and by T expression, but is somewhat broadened(Fig. 3J,K, Fig. 4G-J). In wild-type embryos, mesoderm emerges from the primitive streak and migrates anterolaterally between the epiblast and VE. The anterior primitive streak gives rise to a specialized single layer of cells termed the axial mesendoderm(AME) that occupies the ventral midline and marks the point where axial mesoderm and DE converge (Fig. 3H). In Smad4 mutants, migrating mesoderm traverses the midline without interruption, suggesting loss of AME(Fig. 3I). At E8.5, transverse sections show no evidence of notochord. The neural plate is also flattened and fails to form a tube at any point along the AP axis(Fig. 3M-O).

Our histological analysis reveals that the earliest defects are in the specification of future midline structures, which are derivatives of the APS. We therefore analyzed expression of anterior streak markers in Smad4mutants. At E6.5, Foxa2, Gsc and Lim1 (Lhx1 - Mouse Genome Informatics) (Fig. 4A-F)are expressed in the AVE of both wild-type and mutant embryos. This result suggests that specific loss of Smad4 in the epiblast does not impact the production of epiblast-derived signals that pattern the early anterior organizer. By contrast, Foxa2, Gsc and Lim1 expression confined to the APS of wild-type embryos is not observed in Smad4mutants.

We next evaluated whether failure to pattern the anterior primitive streak in Smad4 mutants results in complete loss of its derivatives, the AME, the node and the notochord. At E7.5 the patent node is readily identified by the expression of Nodal. In Smad4 mutant embryos, Nodal expression is not observed(Fig. 4M,N). Consistent with this result, T and Shh transcripts, which also identify the node and anteriorly extending AME, were not detected(Fig. 4I,J,O,P). At E8.5 T and Shh are additionally expressed in the notochord, but no evidence of a notochord structure is apparent in mutants, although scattered Shh-expressing cells are occasionally observed(Fig. 4K,L,Q,R). From these analyses, we conclude that Smad4 in the epiblast functions to specify the APS,and in its absence, midline structures are completely lost.

Cell lineage experiments have shown that some of the epiblast cells entering the APS during gastrulation exits and intercalates into the overlying visceral endoderm. These nascent definitive endoderm cells form a sheet that displaces the visceral endoderm proximally and, with the exception of a few residual VE cells, eventually gives rise to the entire embryonic gut(Dufort et al., 1998; Lawson et al., 1991). Chimera studies have uncovered an important requirement for Foxh1 and Smad2 in specification of DE progenitors(Hoodless et al., 2001; Tremblay et al., 2000). To investigate whether Smad4 collaborates with these known components of the Nodal signaling pathway in DE specification, we analyzed expression of Hex, which identifies the AVE as well as the earliest population of anterior definite endoderm (ADE) to emerge from the APS(Martinez Barbera et al.,2000; Thomas et al.,1998). At E6.5, the ADE domain of Hex expression is absent in Smad4 mutants (Fig. 5A,B). Consistent with result, at E7.5 Smad4 mutants fail to activate a Hex-lacZ transgene that identifies ADE adjacent to the midline (Fig. 5C-F)(Rodriguez et al., 2001).

The DAN family member Cer1 provides an additional marker for early DE. In Smad4 mutants, Cer1 expression in the endoderm is highly reduced, with the extreme anteroproximal domain probably representing residual AVE expression (Fig. 5G,H). In addition, by the late primitive streak stage Foxa2 extends from the node to the anterior end of the embryo to mark AME derivatives, including DE. In Smad4 mutants, however, this broad domain of expression is absent (Fig. 5I,J).

To distinguish whether loss of DE markers reflects loss of gene expression or a failure to produce the cells themselves, we performed an in vivo cell lineage assay. The ROSA26^R conditional allele(Soriano, 1999) was introduced into the Smad4^CA background. The Sox2Cretransgene robustly activates the ROSA26^R reporter allele in the epiblast. At early headfold stages, the three embryonic lineages -ectoderm, mesoderm and definitive endoderm - comprise exclusively lacZ-positive cells, while extra-embryonic tissues are lacZnegative (Fig. 5M,N). By contrast, in Smad4 mutants the superficial endoderm layer is completely devoid of lacZ activity, indicating that these cells are extra-embryonic in origin (Fig. 5O,P). This result conclusively shows that Smad4-deficient epiblast cells do not form DE, and consequently the overlying visceral endoderm layer is not displaced.

Anterior identity in the mouse is initially specified by the AVE but subsequently reinforced by APS derivatives such as the ADE and prechordal plate. The AVE is normally patterned in Smad4 mutants and the AP axis is correctly specified. However, mutant embryos fail to form classic headfolds at E7.5 and form a pair of bulbous structures suggestive of rostral CNS at E8.5. We therefore characterized the degree of neural pattern in Smad4 mutants. Otx2 is initially expressed in the epiblast and visceral endoderm, but by E7.5 transcripts are restricted to the anterior neurectoderm in both control and mutant embryos(Fig. 6A,B). At E8.5 Otx2 expression normally resolves to demarcate the forebrain and midbrain. By contrast, expression is confined to the most apical portion of the headfolds in Smad4 mutants(Fig. 6C,D). A similar reduction was seen in the expression of the forebrain marker Six3(Fig. 6E,F). To evaluate anterior CNS defects further, we analyzed Fgf8 expression, which identifies the anterior neural ridge and the midbrain-hindbrain junction (or isthmus). In Smad4 mutants, while Fgf8 is weakly expressed in a region that probably corresponds to the isthmus, the most rostral domain of expression is lost (Fig. 6G,H). Hindbrain patterning was examined using Krox20expression, which delineates rhombomeres 3 and 5. Mutant embryos variably exhibit one or two stripes of expression that frequently span the midline(Fig. 6I,J). Collectively,these results suggest that Smad4 mutants exhibit limited neural patterning, probably imparted by the early activity of the AVE. In the absence of secondary refining signals provided by the anterior AME, the growth and patterning of the neural plate is disrupted.

Bmp signaling pathways play evolutionarily conserved roles in heart development. In Drosophila, the Bmp homolog Decapentaplegic (Dpp)activates the transcription factor Tinman that specifies contractile cardiomyocytes. In mouse the Tinman homolog Nkx2.5 is one of the earliest markers of cardiogenesis, expressed anteriorly at E7.5 in a crescent-shaped pattern (Fig. 7A,A′). This expression domain is regulated by an enhancer element containing multiple Smad-binding elements(Lien et al., 2002). Intriguingly, Nkx2.5 expression is unaffected in Smad4mutants (Fig. 7B,B′). By E8.5, mutant embryos show consistent formation of a rudimentary heart in the anteriormost part of the embryo (Fig. 7C). Histological analysis reveals a structure with endocardium and cardiomyocyte differentiation, and differentiation of cardiac mesoderm is confirmed by cardiac alpha actin expression(Fig. 7D-G). Although the specification of the cardiac lineage occurs in the absence of Smad4,development of the heart is abnormal, and at best a primitive unlooped heart tube forms.

Recent genetic experiments reveal that null mutations in genes encoding the Bmp2, Bmp4 or Bmp8b ligands or their cognate intracellular effectors Smad1 and Smad5 quantitatively impact primordial germ cell (PGC) formation and variably affect allantois development (McLaren,2003). These ligands are expressed in either the extra-embryonic ectoderm or visceral endoderm and act prior to gastrulation to predispose a portion of proximal epiblast cells, which express Smad1, Smad5 and Smad8, to the PGC and allantois fates. We therefore evaluated the contribution of Smad4 to the transduction of extra-embryonic Bmp signals by investigating whether loss of Smad4 from the epiblast compromises PGC specification and allantois formation.

PGCs can first be identified by their high levels of endogenous alkaline phosphatase (AP) activity around E7.5 and localization in a cluster at the base of the incipient allantois (Fig. 7J). Smad4 mutants at this stage show disorganized extra-embryonic development, with the collapsing extra-embryonic ectoderm shifted toward the posterior visceral yolk sac. Rudimentary allantois formation is observed in all mutant embryos. However, in the majority of mutants (seven out of nine) no cells resembling PGCs with darkly stained cell membranes or a coarse spot are detected either at the base of the allantois or ectopically within the neighboring visceral yolk sac(Fig. 7K).

To address the possibility that PGC formation is delayed in the absence of Smad4, PGC number was assayed one day later at E8.5. In wild-type embryos at this stage, AP-positive PGCs (∼100) have dispersed from the base of the allantois and intercalated into the hindgut epithelium(Fig. 7L). All Smad4mutants (21 out of 21) show an allantois-like outgrowth that varies in size and that in the majority of cases (70%) associates or fuses with the stalk-like chorionic ectoderm (Fig. 3O, Fig. 7H,I). In the remaining embryos, the allantois terminates blindly within the exocoelomic cavity (Fig. 7F). Despite relatively normal formation of the allantois, the majority of E8.5 mutant embryos (16 out of 21) lack AP-positive PGCs in the posterior extra-embryonic region, including the base of the allantois and visceral yolk sac, or in the posterior embryo proper, which lacks a recognizable hindgut(Fig. 7M). The remaining mutants have fewer than ten PGCs widely scattered in the disorganized posterior region (data not shown). Taken together, these data suggest that Smad4 is required in the epiblast for specification of the PGC precursor population.

We have exploited a conditional allele of Smad4 to reveal requirements for this gene in the early epiblast. Remarkably, in spite of the well-documented requirements for TGFβ signaling pathways, loss of Smad4 has a minimal impact on early development. Formation of numerous mesodermal derivatives is unaffected. Rather, our results uncover restricted rules for Smad4 in the specification of the APS derivatives and in formation of the germline, functions mediated by the Nodal and Bmp pathways, respectively.

After the onset of gastrulation, inductive events occurring within the APS are responsible for formation of the node and its derivatives, as well as definitive endoderm. In Smad4 mutants, these midline structures fail to form, resulting in an embryo with fused somites and anterior defects. The midline defects reflect the inability of the embryo to correctly pattern the APS as seen by failure to activate Gsc, Lim1 and Foxa2. The promoter regions of Gsc and Lim1 have been shown to contain activin response elements occupied by a complex containing Smad4, Smad2 and Foxh1 (Labbé et al.,1998; Watanabe et al.,2002), and failure to activate Gsc and Lim1expression in the APS domain is thus consistent with a essential role for Smad4 in the regulation of these genes.

The definitive endoderm arises from cells within the epiblast that ingress through the APS, displacing the visceral endoderm layer. Our lineage studies and marker analysis indicate that DE is not specified in Smad4mutants. As chimeric analysis and conditional gene ablation have also identified a role for Smad2 and Foxh1 in endoderm specification, our finding suggests that Smad4 complexes with Smad2/Foxh1 to mediate formation of DE.

Overall, removal of Smad4 from the epiblast leads to a phenotype that closely resembles that of several other mutants. Embryos in which extra-embryonic lineages are wild-type while the epiblast lacks Foxa2fail to form a node and notochord, have fused somites, and establish rudimentary neural patterning (Dufort et al., 1998; Hallonet et al.,2002). Foxh1 mutant embryos also lack APS derivatives,and show somite fusion (Hoodless et al.,2001; Yamamoto et al.,2001). Removal of Smad2 alone or in the context of Smad3 deficiency also leads to similar phenotypes (see below). Collectively, these findings provide genetic evidence that Smad4,Smad2/3 and Foxh1 lie in the same pathway and support previously characterized biochemical interactions. These mutant contexts each result in downregulation of Foxa2, probably accounting for the convergent phenotype. Curiously, no Foxh1/Smad-binding sites have been documented in the Foxa2 gene, and regulation by Foxh1/Smad proteins may be indirect(Nishizaki et al., 2001).

The central role of Smad4 in transducing TGFβ signals was in part initially characterized by gain-of-function experiments in Xenopusembryos. Overexpression of Smad1 in animal caps simulated the ability of Bmp2 and Bmp4 to induce ventral mesoderm, while Smad2 induced dorsal mesoderm similar to activin stimulation. As expected, Smad4injection produced both dorsal and ventral mesoderm(Lagna et al., 1996; Zhang et al., 1997). These results suggest that Smad4 contributes essential activities in the events of gastrulation and mesoderm formation. That Smad4 homozygous null embryos fail to gastrulate was unsurprising, especially in light of similar phenotypes observed in activin/TGFβ and Bmp receptor mutants.

However, when wild-type extra-embryonic lineages are provided either by aggregation of Smad4-deficient ES cells with tetraploid wild-type embryos or selective loss of Smad4 gene activity with Sox2Cre, gastrulation commences and mesoderm forms. In Sox2Cre;Smad4^CA/N mutant embryos, we observe numerous mesodermal derivatives, including somites, heart, allantois and lateral plate mesoderm. We also find that teratomas made from Smad4^N/NES cells form mesoderm (G.C.C. and E.J.R., unpublished). These results suggest that the primary defect in Smad4 homozygous null embryos is restricted to the extra-embryonic lineages. When this requirement is bypassed, Smad4-independent gastrulation and mesoderm formation proceeds. Of particular interest, the Smad4 mutant shows a phenotype less severe than a Smad2;Smad3 doubly-deficient epiblast(Dunn et al., 2004). Stepwise removal of Smad2 and Smad3 from the embryo results in a progressive loss of primitive streak lineages, first impacting APS derivatives such as the AME, node and notochord followed by middle primitive streak lineages such as somites and LPM. The more extensive mesoderm formation seen in Smad4 mutants therefore suggests that Smad2 and Smad3 function in a Smad4-independent manner in eliciting Nodal-mediated induction of middle streak derivatives. Thus, in the mouse embryo, as in Xenopus, Smad4 appears to potentiate TGFβsignaling to allow the correct patterning of the mesoderm lineage, but Smad4 activities are unnecessary for mesoderm formation per se.

The role of Bmp-dependent signals in heart formation is evolutionarily conserved. In Drosophila, Dpp initiates signaling involving the R-Smad Mad, the Smad4 homolog Medea and Tinman, the homolog of Nkx2.5, to specify the cardioblast fate (Zaffran et al., 2002). In chick, a homologous pathway has been described in which Bmp2 or Bmp4 induces expression of Nkx2.5 in non-cardiogenic mesoderm and directs cardiac differentiation (Schlange et al.,2000; Schultheiss et al.,1997). Intriguingly, we found that removal of Smad4 in the epiblast affected neither early cardiac development nor correct induction of Nkx2.5. Previous promoter analysis identified multiple Smad binding elements in the AR2 enhancer of Nkx2.5 that governs expression within the cardiac crescent. Mutation of a specific Smad4 binding site leads to complete loss of transgene activity in cardiogenic mesoderm(Lien et al., 2002). Our results therefore suggest that Smad4 is not essential for activation of the AR2 enhancer and that R-Smad binding and association with other requisite cofactors are sufficient to promote early Nkx2.5expression.

The Bmp signaling pathway is essential for the normal specification of the mammalian germline (McLaren,2003). Bmp4 and Bmp8b secreted by the extra-embryonic ectoderm predispose a small number of progenitor cells within the extreme proximal epiblast towards the germ cell fate. Importantly, these precursor cells are multipotent in that they give rise not only to committed PGCs but also to descendants in the extra-embryonic mesoderm, particularly the allantois(Lawson and Hage, 1994). Bmp4 homozygous null embryos entirely lack PGCs and allantois,whereas heterozygotes show reduced PGC numbers and normal allantois formation(Lawson et al., 1999). Similar gene dose effects are observed in Bmp8b mutant embryos(Loebel et al., 2003). These observations suggest that allocation to the germ-cell lineage is more sensitive to the levels of Bmps than is allocation to the allantois. Furthermore, fate mapping studies reveal that the allantois is derived from a much broader region of the proximal epiblast, where Bmp concentrations are expected to be lower, than the narrow belt at the junction with the extra-embryonic ectoderm in which the common PGC/allantois progenitors reside and which is presumably exposed to the highest local Bmp concentration. The loss of PGCs with persistent formation of allantois in Smad4 mutant embryos is therefore consistent with the simple model in which Smad4 is required for maximal Bmp signaling within the proximal epiblast, with intermediate signaling reliant upon the available Bmp R-Smads Smad1/5/8 within the epiblast and sufficient for formation of allantois(Hayashi et al., 2002a).

The widely held conclusion that Smad4 occupies a central role in transduction of TGFβ signals comes from multiple lines of biochemical and genetic evidence (reviewed by Massagué, 1998). Smad4 participates in both activin and Bmp pathways as demonstrated in Xenopus overexpression assays and in dominant-negative experiments. In reconstitution experiments, cell lines that lack Smad4 fail to respond to TGFβ signals; transfection of wild-type Smad4 restores the signaling capabilities of these cells(de Caestecker et al., 1997; Lagna et al., 1996; Zhang et al., 1997). Moreover,in Drosophila, Medea/Smad4 is necessary for Dpp function,and elimination of maternal and zygotic Medea in the embryo results in dorsoventral patterning defects identical to that of null Dppmutants, indicating that Medea is required for Dpp-dependent signaling (Das et al., 1998; Hudson et al., 1998).

By contrast, our data suggest that in the early embryo, Smad4 is required for certain TGFβ/Bmp signaling pathways, but not obligate for others. A growing body of data corroborates our observations. For example, in Smad4-deficient MEFs as well as several Smad4-deficient human tumor cell lines, TGFβ addition still results in classical growth inhibition (reviewed by Derynck and Zhang,2003; Wakefield and Roberts,2002). In Drosophila, oogenesis is more severely disrupted by the loss of the R-Smad Mad than by Medea. Interestingly, in wing imaginal discs, loss of Medea most severely affects regions receiving low Dpp signal(Wisotzkey et al., 1998). One possible explanation is that Medea normally potentiates weak Dpp signals and that in the absence of Medea, R-Smads regulate target genes only when the Dpp signal strength is high. R-Smads may substitute for or bypass the requirement for Smad4, resulting in unpotentiated levels of downstream signal(Yeo et al., 1999). Additionally, alternate TGFβ pathways have been described, including the Ras/Mapk, Pp2a/S6 kinase, Rhoa and PI3K-Akt pathways, some of which are entirely Smad independent and others that involve crosstalk with Smad signals. Finally, an independently-derived Smad4 conditional allele was recently employed to eliminate Smad4 function during CNS and mammary gland development and the resulting phenotypes were surprisingly mild(Li et al., 2003; Zhou et al., 2003). These findings in combination with our genetic results emphasize the versatility in the intracellular transduction of TGFβ-related signals, and encourage a more careful consideration of the current canonical model of TGFβsignaling that places Smad4 as a central effector molecule.

We are grateful to Elizabeth Bikoff for insightful comments and to Cindy Lu for review of the manuscript. We thank Shigemi Hayashi and Andrew McMahon for providing the Cre-ER™ and Sox2Cre mice, Tristan Rodriguez for Hex-lacZ mice, Phil Soriano for ROSA26^R mice, and Makoto Taketo for Smad4 genomic constructs. We also thank Joan Massagué, Robert Maxson, Jonathan Pearce and Christian Sirard for plasmid reagents and probes. G.C.C. was supported by the Howard Hughes Medical Institute Postdoctoral Fellowship for Physicians,N.R.D. by a postdoctoral fellowship from the NICHD, and L.O. by a fellowship from the Wenner-Gren Foundations. This work was supported by a grant from the NIH to E.J.R.