Loss of Tbx4 blocks hindlimb development and affects vascularization and fusion of the allantois

T-box genes encode a family of transcription factors that contain a conserved DNA-binding motif known as the T-box domain(Papaioannou, 2001). Members of this family are expressed dynamically during development, and mutations in T-box genes have profound effects on development in organisms ranging from C. elegans to humans. In vertebrates, the Tbx2 subfamily of T-box genes consists of four members, Tbx2, Tbx3, Tbx4 and Tbx5 (Ruvinsky et al.,2000), and in vertebrate species where all members have been cloned (mouse, chick and human) they are highly conserved. The Tbx2subfamily consists of two pairs of linked genes, which are thought to be the product of an ancestral tandem duplication, followed by duplication and dispersion of the two-gene clusters(Agulnik et al., 1996). The genes of one cluster, Tbx3 and Tbx5, cause dominant developmental syndromes in humans, known as ulnar-mammary(Bamshad et al., 1997) and Holt-Oram (Basson et al., 1997; Li et al., 1997) syndromes,respectively. In mouse, mutations in Tbx3 and Tbx5 also cause developmental abnormalities (Bruneau et al., 2001; Davenport et al.,2003). No mutations in genes of the other cluster, Tbx2and Tbx4, have been reported.

The first expression of Tbx4 in the mouse embryo is observed at 7.5 days post coitus (dpc) in the developing allantois. This expression is maintained through at least 9.5 dpc(Chapman et al., 1996). The allantois is an extra-embryonic, mesodermal structure that forms a connection between the posterior embryo and the chorion early in development. This structure later develops into the umbilicus, and is crucial in mammals for nutrient, waste and gas exchange between mother and embryo. Molecular pathways involved in development of the allantois are not well characterized. However,several targeted mutations in mice have indicated that bone morphogenetic protein (BMP) signaling is crucial for allantois development. Embryos mutant for Bmp4 lack an allantois(Lawson et al., 1999), and chimeric embryos lacking Bmp4 specifically in the epiblast lineage form only a small allantois that fails to fuse to the chorion(Fujiwara et al., 2001). Although neither Bmp5- nor Bmp7-mutant mice have allantois defects, the allantois of embryos doubly homozygous for both mutations fails to undergo chorioallantoic fusion(Solloway and Robertson,1999). Bmp8b mutants also have shortened allantoises, and a reduced number of primordial germ cells(Ying et al., 2000). Furthermore, mutations in the downstream effectors of BMP signaling, Smad1 (Madh1 – Mouse Genome Informatics) and Smad5 (Madh5 – Mouse Genome Informatics), produce aberrant allantois morphology, although not as severe as that of the BMP mutations (Chang et al., 1999; Lechleider et al., 2001; Tremblay et al., 2001).

Several genes involved in the process of chorioallantoic fusion have been identified. The adhesion molecule VCAM1, which is expressed in the distal tip of the allantois, and its receptor, α4 integrin, which is expressed in the chorion, are both required for successful chorioallantoic fusion(Gurtner et al., 1995; Kwee et al., 1995; Yang et al., 1995). Additionally, several transcription factors that are required for proper chorioallantoic fusion have been identified by targeted mutagenesis, including the forkhead transcription factor Foxf1(Mahlapuu et al., 2001) and Suppressor of Hairless homolog RBP-J κ(Rbpsuh – Mouse Genome Informatics)(Oka et al., 1995). Connection(s) between these various signaling factors, adhesion molecules and transcription factors have not been elucidated and previously no role in this process has been identified for Tbx4.

During organogenesis, Tbx4 is expressed in a variety of tissues,including the hindlimb, proctodeum, mandibular mesenchyme, lung mesenchyme,atrium of the heart and the body wall(Chapman et al., 1996). Hindlimb expression of Tbx4 initiates at 9.5 dpc in the posterior flank region known as the hindlimb field, before the hindlimb bud is formed. Tbx5, another Tbx2 subfamily member, mirrors this expression pattern with forelimb-specific expression starting at 8.5 dpc, when the forelimb field is specified in the anterior flank. This reciprocal expression of Tbx4 and Tbx5 in the hindlimb and forelimb, respectively,is maintained throughout development(Chapman et al., 1996; Gibson-Brown et al.,1996).

Previous work has suggested that Tbx4 plays a role in determining hindlimb identity. Transplants of limb mesenchyme and induction of ectopic limbs in chick show correspondence between the proportions of Tbx4and Tbx5 expression, and the degree of forelimb versus hindlimb fate,respectively (Gibson-Brown et al.,1998; Isaac et al.,1998; Ohuchi et al.,1998). Ectopic expression of Tbx4 and Tbx5 in developing chick limbs results in limb abnormalities that may represent limb identity transformations(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). Supporting evidence is offered by experiments with the homeodomain protein Ptx1 (Pitx1 – Mouse Genome Informatics), which is expressed specifically in the hindlimb(Lanctot et al., 1997; Shang et al., 1997) and which has also been suggested to play a role in hindlimb specification. Targeted mutagenesis of Ptx1 produces mice with shortened hindlimbs that have characteristics of forelimbs (Lanctot et al., 1999; Szeto et al.,1999), and ectopic expression of Ptx1 in chick forelimbs leads to both the upregulation of Tbx4, and the induction of the hindlimb-specific Hox genes Hoxc10 and Hoxc11(Logan and Tabin, 1999).

Interpretation of the gain-of-function experiments with Tbx4, Tbx5and Ptx1 in chick is confounded by the presence of endogenous Tbx4 and Tbx5 in the experimental limbs, and by limb abnormalities induced by overexpression of Tbx4 and Tbx5,even in their endogenous domains(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). We have therefore used targeted mutagenesis to investigate the role of Tbx4 in the development of the hindlimb and other regions of Tbx4 expression, including the allantois.

A targeting vector was created using a total of 7.3 kb of genomic DNA from a 129Sv/J phage library (Stratagene, catalog number 946309). A HindIII/EcoRV fragment containing exons 4 and 5 was subcloned, and a loxP-flanked PGK-neo PGK-thymidine kinase dual selection cassette was inserted into the first SpeI site of intron 5 (Fig. 1A),giving homology 3 kb upstream and 5.3 kb downstream, respectively. Oligonucleotides containing a third loxP site, as well as KpnI, NheI and XhoI sites, were ligated into the XhoI site 1.5 kb 5′ of the selection cassette, in intron 4. Theβ-actin diptheria toxin gene(Maxwell et al., 1987) was inserted at the distal 5′ end of the targeting cassette for negative selection against random integration. This targeting construct was linearized and electroporated into R1 ES cells (Nagy et al., 1993). G418 resistant colonies were screened by Southern hybridization for homologous recombination events, using both 5′ and 3′ genomic probes (Fig. 1B,C). Approximately two-thirds of the clones containing the targeted neo-tkcassette also contained the 5′ loxP site.

Chimeras were generated by injection of targeted ES cells into C57BL/6J blastocysts. ES cell-derived progeny of the chimeras and subsequent generations were routinely genotyped by PCR using the primers (a)5′-GAGGATGTTCCCCAGCTAC-3′ and (b)5′-CAGTCTGAGAGGGTCAGACTC-3′(Fig. 1A,D), which detect a 500 bp endogenous band and a 550 bp mutant band containing the insertion of the engineered loxP site. Mice heterozygous for the Tbx4^tm1Pa allele were then mated to mice carrying the cre transgene under control of the β-actin promoter(Lewandoski and Martin, 1997),to generate the Tbx4^tm1.1Pa allele by excision of the selection cassette and a 1.5 kb genomic region(Fig. 1A). Confirmation of the cre-mediated excision of all loxP-flanked sequences and genotyping of mice with the Tbx4^tm1.1Pa allele was performed with the primers (a) and (b) above, and (c)5′-TCATCTAGGCTTCACAGCC-3′, which produces a 250 bp band(Fig. 1A,D).

Both the Tbx4^tm1Pa and the Tbx4^tm1.1Pa alleles were maintained on mixed genetic backgrounds. Mice heterozygous for each allele were intercrossed to generate homozygous mutant embryos. Noon on the day of a mating plug was considered 0.5 dpc. Embryos were dissected in phosphate buffered saline (PBS) with 0.2%bovine albumin (fraction V). All embryos were scored for chorioallantoic fusion prior to yolk sac removal. Yolk sacs were taken for PCR genotyping and embryos were fixed in 4% paraformaldehyde in PBS at 4°C for two hours or overnight, then dehydrated in methanol and stored.

Embryos dissected at 8.0 dpc were scored for somite number and for the extent of allantois elongation into the yolk sac cavity. Because the overall embryo size is quite variable at this point in development, allantois elongation was measured as a proportion of the distance between the posterior end of the embryo and the edge of the chorionic plate. Allantoises were scored as: `early' if they had covered two-thirds or less than this distance; `late'if they had covered more than two-thirds of the distance or if they had entered the dome formed by the chorionic plate; or `fused' if they had formed a connection to the chorion.

Whole-mount in situ hybridization was performed according to previously described protocols (Wilkinson and Nieto,1993). Immunohistochemistry was performed according to standard protocols (Davis, 1993),except for VCAM1 immunohistochemistry, where DMSO was omitted from the bleach,PBSMT washes were replaced with PBSBT (4% BSA, 0.1% Triton X-100 in PBS) and primary antibody was diluted 1:150. Embryos used for doubly phosphorylated ERK protein (dp-ERK; EPHB2 – Mouse Genome Informatics) immunostaining were cut through the limb buds prior to bleaching to facilitate antibody penetration. Primary antibodies used included: anti-VCAM1 (PharMingen, catalog number 55333O), anti-phospho-Histone H3 (Upstate Biotechnology, catalog number 06570), anti-PECAM (PharMingen, catalog number 01951D) and anti-phospho-p44/42(anti-dp-ERK; Cell Signaling, catalog number 9101). All secondary antibodies were peroxidase-conjugated goat IgG from Jackson Immunochemicals.

Embryos were dissected at 8.0 dpc, scored as above and genotyped. Posterior halves were fixed in 4% paraformaldehyde in PBS for 1 hour at 4°C, and dehydrated in 70% ethanol for 3-6 hours. Embryos were rehydrated in NTMT, then stained with NBT/BCIP color reaction for 3 minutes(Wilkinson and Nieto, 1993)and photographed.

Terminal deoxynucleotidyl transferase biotin-dUTP nick end labeling (TUNEL)staining was performed using a Cell Death Detection kit (Roche, catalog number 1684817). Embryos were incubated in 10μg/ml proteinase K for 8-10 minutes depending on stage, and refixed in 4% paraformaldehyde, 0.1% glutaraldehyde for 20 minutes at room temperature. Embryos were then incubated in TUNEL reaction mixture at 37°C for 1 hour, blocked with 10% sheep serum in PBT for 1 hour and incubated with the converter-POD mix (same kit) for 30 minutes at 37°. Chromogenic reaction was developed with 0.08% NiCl₂ and 250 μg/ml diaminobenzidine in PBT for approximately 30 seconds.

At 8.25 dpc and 9.5 dpc, where simultaneous histology and genotyping were not possible, embryos were classified as mutant or normal based on the appearance and fusion of the allantois. Three embryos were discarded from the analysis because they were grossly retarded compared to littermates. Embryos at 8.25 dpc and 9.5 dpc were removed intact in the uterine horns and fixed in Bouin's fix overnight. Embryos at 10.5 dpc were dissected out of the decidua,and the yolk sacs removed for genotyping before being similarly fixed. After dehydration in ethanol, embryos were embedded in paraffin wax, sectioned at 8μm thickness and stained with Hematoxylin and Eosin.

Limb buds were dissected at 10.5 dpc in DMEM supplemented with 10 mM HEPES and 5% FCS. Explants were taken from embryos only if the embryo was alive and had a robustly beating heart. The embryo was transected immediately rostral and caudal to each set of limb buds (see Fig. 4A,F). Tissue ventral of the limb buds, including the heart and allantois, was trimmed off and the explant was placed ventral side down on a Transwell Clear (Costar)membrane. Explants were cultured for 4-7 days in DMEM/F-12 (Gibco BRL) with 10% FCS (Hyclone), 5× MITO⁺ Serum Extender (Becton Dickinson,catalog number 355006) at 37°C in 5% CO₂ in air.

The Tbx4 gene was disrupted by the insertion, via homologous recombination in ES cells, of a loxP site and a loxP-flanked neomycin-thymidine kinase selection cassette into the introns surrounding exon 5 (Fig. 1A, Tbx4^tm1Pa allele). Correctly targeted ES cells lines were identified by Southern blotting (Fig. 1B,C). Targeted ES cells were injected into C57BL/6J blastocysts to create chimeras, which were mated to identify germline transmission of the mutant allele Tbx4^tm1Pa. Transmission of two ES cell lines of the Tbx4^tm1Pa allele produced by independent targeting events was obtained. Mice derived from both cell lines have the same phenotype and the results have been combined in this report. Mice heterozygous for the Tbx4^tm1Pa allele were mated to mice ubiquitously expressing a β-actin-cre transgene(Lewandoski and Martin, 1997)to excise both the selection cassette and the floxed exon 5, thereby creating the Tbx4^tm1.1Pa allele(Fig. 1A).

Heterozygous mice of both alleles were apparently normal and fertile. Heterozygotes were intercrossed to generate embryos homozygous for each allele. Both the Tbx4^tm1Pa allele and the Cre-recombined Tbx4^tm1.1Pa allele were inherited in Mendelian ratios, but all homozygous mutants dissected after 10.5 dpc were dead(Table 1). In litters that were allowed to go to term, no homozygous mutants were observed among weanlings(n=397 and n=108, for the Tbx4^tm1Pa and Tbx4^tm1.1Pa alleles, respectively). All analyses,including in situ hybridizations for limb marker genes, were performed on embryos homozygous for each allele. Results for each were the same, and so have been combined in this report.

Tbx4-homozygous-mutant embryos display no apparent morphological abnormality until 8.0 dpc, when an allantois defect becomes evident. At this age, normal embryos have well-formed allantoises that elongate from the posterior end of the embryo and fuse with the chorionic plate at the 6- to 8-somite stage (Fig. 2A). Tbx4-mutant embryos of the same age display allantoises extended only partway through the yolk sac cavity, and 6- to 8-somite stage mutant embryos have not undergone chorioallantoic fusion(Fig. 2B). Homozygous-mutant embryos turn at the same stage as their normal littermates. By 9.5 dpc, normal embryos have formed a thick, vascular umbilicus(Fig. 2C) connecting the embryo to the placenta, whereas mutant embryos remain loose in the yolk sac,unattached to the placenta (Fig. 2C). No other abnormalities were seen at 9.5 dpc.

At 10.5 dpc, Tbx4-mutant embryos are loose in the yolk sac: some display swollen pericardial sacs and/or hemorrhage(Fig. 2D), and many are dead,as judged by the absence of a heartbeat. All homozygous mutants dissected after 10.5 dpc were dead (Table 1). Instead of the umbilical vessels of the wild type(Fig. 2E), the allantois of 10.5 dpc mutant embryos forms a compact amorphous mass surrounding blood-filled vesicles (Fig. 2F). There is no obvious leakage of blood from the residual allantois into the yolk sac cavity. At the latest stage reached by the homozygous mutants prior to their death (∼29- 32 somites), mutant embryos have morphologically apparent hindlimb buds(Fig. 2D). Hindlimb buds in 27-to 29-somite mutant embryos are similar in size to somite-matched normal littermates, but are modestly reduced in size in 30- to 32-somite mutant embryos when compared with somite-matched controls.

Two 10.5 dpc Tbx4-mutant embryos (representing less than 1%) have been observed in which the developing allantois had made a connection with the chorion. In both embryos the allantois was small and irregular, consisting of multiple blood-filled chambers. In each case, the allantoic attachment to the chorion was at only a single point, and had not spread over the chorionic surface. No apparent continuous vessel had formed between the embryo and the placenta, and no flow of blood could be detected.

Histological examination of 8.25 dpc embryos revealed multiple abnormalities in the allantois of Tbx4-mutant embryos. The wild-type allantois at this stage (n=10) is a funnel shaped structure tightly opposed to the chorion at the wide end and tapering towards the posterior end of the embryo (Fig. 2G). The mesenchyme near the embryo is dense and uniform, whereas near the chorion it is loose and cavitated. In mutant embryos (n=2), the allantois was not opposed to the chorionic plate and there was no evidence of allantois-derived cells adherent to the chorion(Fig. 2H), although the chorion itself appeared normal. The mutant allantois shows dense, irregularly packed cells in the base of the allantois, with multiple double-walled vesicles(Fig. 2I) and numerous pyknotic nuclei, which is indicative of dying cells.

At 9.5 dpc, the allantois of wild-type embryos (n=4) consists of moderately dense mesenchyme with a more open structure near the chorion. Blood vessels running through the allantois are visible(Fig. 2K). By contrast, the allantois of a mutant at this stage (n=1) was small and dense, and had numerous condensed cells (Fig. 2J,L). At 10.5 dpc,sections through a wild-type embryo show a developing hindlimb bud and an umbilical vessel (Fig. 2M). Sections through a Tbx4 mutant show some edema, but an otherwise normal development in embryonic tissues, including lung buds, heart, fore- and hindlimb (Fig. 2N,O). However, the allantois at this stage has not developed further, and consists of irregular mesenchyme with multiple irregularly shaped, dense condensations of cells and occasional empty vesicles (Fig. 2O).

Embryos dissected at 8.0 dpc were scored for allantois phenotype and somite number (Fig. 3A). Differences between wild-type and mutant embryos become apparent as early as the 4- to 5-somite stage, when wild-type embryos are mostly at the late allantois stage or fused to the chorionic plate but most Tbx4-mutants are still at the early bud stage. Heterozygous embryos at the 4- to 5-somite stage exhibit a lag in allantois development when compared with somite-matched wild-type controls. At the 6- to 8-somite stage, nearly all wild-type and heterozygous embryos have undergone chorioallantoic fusion, whereas most mutant embryos are at the early allantois stage. All wild-type and heterozygous embryos past the 8-somite stage have completed chorioallantoic fusion.

To further investigate the allantois growth defect, we examined the rate of cell proliferation and cell death in this tissue. Cellular proliferation in mutant allantoises (Fig. 3C) is comparable to wild-type embryos of the same stage(Fig. 3B), and mitotic cells are seen at all levels of the allantois, including the distal tip. TUNEL staining shows little or no apoptotic cell death in the allantois of wild-type embryos (Fig. 3D), whereas homozygous-mutant embryos exhibited extensive cell death over the entire distal tip of the allantois (Fig. 3E).

The differentiative capacity of the allantois in the absence of Tbx4 was assessed by examining the expression of a number of genes with characteristic expression patterns during normal allantois development. Bmp4 expression, which is required for proper allantois development,is apparent in the base of both wild-type(Fig. 3F) and mutant(Fig. 3G) allantoises. Tbx2 is expressed in the wild-type allantois(Fig. 3H) from its earliest morphological appearance until early 9.5 dpc(Chang et al., 1999; Mahlapuu et al., 2001) (L.A.N. and V.E.P., unpublished). In Tbx4-homozygous mutants, Tbx2expression is absent or dramatically reduced in the allantois of the 2-somite embryo (Fig. 3I), and entirely absent in the allantois of later embryos (data not shown).

Contrary to previous work published by this lab(Chapman et al., 1996), we were unable to detect Tbx5 expression in the allantois by whole-mount in situ hybridization. No Tbx5 expression was seen in the allantois of either mutant or wild-type embryos from 7.5 dpc to 8.5 dpc, although expression in the heart crescent was visible from late 7.5 dpc onward as expected (data not shown). The brachyury gene (T), which marks the primitive streak, notochord and the base of the allantois, is expressed normally in the Tbx4-mutant allantoises (data not shown). VCAM1, a cell adhesion molecule required for allantoic fusion, is seen by antibody staining in the extreme distal end of the fused allantois in the 8-somite wild-type embryo (Fig. 3J), but is absent from the unfused distal tip of Tbx4-mutant allantoises(Fig. 3K).

The primordial germ cells (PGCs) form a group of alkaline phosphotase-positive cells at the base of the allantois at 8.0 dpc, shortly before they migrate along the dorsal mesentery to the future gonads. Their formation is impaired in some of the BMP pathway mutations that affect allantois development (Lawson et al.,1999; Tremblay et al.,2001; Ying et al.,2000). In wild-type (Fig. 3L) and Tbx4-mutant (Fig. 3M) embryos, PGCs appear to be similar in number and location.

To assess vascularization in the allantois, we examined the appearance of PECAM-1 (PECAM – Mouse Genome Informatics), a late marker of endothelial cell development. In wild-type embryos at the 0- to 2-somite stage,anti-PECAM-1 antibody marks small clumps of endothelial cells in the allantois(Fig. 3N). At the 3- to 4-somite stage, these PECAM-positive cells form multiple small vessels(Fig. 3O). By the 6- to 8-somite stage, these vessels have largely collected into a central vessel(s)at the base of the allantois, with unincorporated vessels still apparent distally (Fig. 3P). In Tbx4 mutants, expression of PECAM-1 initiates normally(Fig. 3L), but soon acquires a strikingly different appearance. Numerous small clumps of PECAM-positive cells appear along the length of the allantois, but these fail to elongate into vessels and no central vessel is formed(Fig. 3O,P).

To investigate the role of Tbx4 in hindlimb development beyond the time of death of Tbx4-mutant embryos, limb buds from 10.5 dpc embryos were grown in culture. Both forelimb and hindlimb buds were dissected from live 10.5 dpc mutant embryos and littermate controls, and explanted onto membranes suspended on growth media (Fig. 4B,C,G,H)to examine their developmental potential. After 3 days of growth, forelimb buds of both normal (Fig. 4D)and homozygous Tbx4-mutant embryos(Fig. 4E) exhibited an increase in the distance between the right and left distal limb margins, corresponding to outgrowth of the limbs from the midline. Cultured limb buds also assumed the paddle-shaped morphology of handplate stage limbs. After the same culture period, wild-type hindlimb buds had also expanded distally and developed a recognizable paddle-shaped handplate (Fig. 4I, n=23), but mutant hindlimb explants failed to develop any obvious limb structures (Fig. 4J, n=11). To verify the identity of these limb structures, in situ hybridizations were performed on cultured explants using a Tbx4 probe from outside of the deleted region as a hindlimb marker(Fig. 1A). The mutant transcript was expressed normally in homozygous mutants at all stages examined. Tbx4 was clearly expressed in the handplate paddles of wild-type embryos (Fig. 4K),but it was expressed only in a few scattered patches of cells, with no obvious limb morphology, in mutant embryos (Fig. 4L).

To explain the absence of growth of the Tbx4-mutant hindlimbs in culture despite the presence of a morphologically obvious hindlimb bud, we examined the expression of genes known to be involved in the initiation and maintenance of limb formation (Martin,2001) prior to the development of the mutant phenotype. Limb outgrowth is known to be maintained by a mesenchymal-ectodermal feedback loop. Mesenchyme from the limb field signals to the overlying ectoderm through Fgf10 to induce the apical ectodermal ridge (AER). Limb ectoderm then signals back to the mesenchyme through several molecules, including FGFs. This reciprocal signaling is necessary for the maintenance of many limb mesenchyme genes, including Fgf10.

Fgf8 is an early marker of the AER, and is expressed normally in a narrow strip along the dorsoventral margin of both wild-type and Tbx4-mutant hindlimb buds (Fig. 5A,B), indicating that successful initial induction of the AER has occurred. Msx1,which lies downstream of limb ectodermal signaling(Wang and Sassoon, 1995), is also expressed normally in Tbx4-mutant hindlimbs(Fig. 5C,D). The twist gene, which is required in limb mesenchyme for the FGF-mediated feedback loop and for proper FGF receptor expression in the limb mesenchyme(O'Rourke et al., 2002; Zuniga et al., 2002), is also expressed normally in Tbx4 mutants(Fig. 5E,F).

To confirm that Tbx4-mutant limb mesenchyme receives FGF signals from the ectoderm, we examined the distribution of doubly phosphorylated ERK protein (dp-ERK), which is present when activated FGF receptors activate the mitogen-activated kinase (MAPK) cascade. At this very early stage of limb induction, dp-ERK staining is found asymmetrically in the hindlimbs with staining first appearing in the right hindlimb(Fig. 5G), although it evens out in more developmentally advanced limbs (data not shown). dp-ERK is observed in Tbx4-mutant hindlimb buds in the same asymmetric pattern,demonstrating mesenchymal reception of ectodermal FGF signals(Fig. 5G,H). However,despite an apparently normal limb induction and normal FGF-feedback signaling, Fgf10 is not maintained in Tbx4-hindlimb mesenchyme and is only seen in the flank underlying the hindlimb bud(Fig. 5I,J). As Fgf10 is required for limb outgrowth, this defect could explain the absence of Tbx4-mutant hindlimb development in vitro.

As Tbx4 expression is specific to the hindlimb and has been postulated to have a role in specifying hindlimb identity, we examined other genes that have differential expression between fore- and hindlimbs. The domain of Tbx5, which is normally expressed specifically in the forelimb, is unaltered in Tbx4-mutant embryos(Fig. 6A,B), indicating that Tbx4 plays no role in maintaining this differential expression. Likewise, the Tbx4-expression domain (visualized with a probe from outside of the deleted region) in the hindlimb is retained in Tbx4mutants (data not shown). Ptx1, which is co-expressed in the hindlimb field with Tbx4, is also unaltered in the Tbx4-mutant hindlimb field (Fig. 6C,D). Unfortunately the death of Tbx4-mutant embryos in vivo, and the absence of hindlimb growth in vitro, preclude the study of later fore- and hindlimb specific genes, such as limb-specific Hoxc genes.

We also examined early markers of dorsoventral and anterior/posterior axis formation in the hindlimb. Msx2, which marks the ventral limb ectoderm, is expressed ventrally in Tbx4-mutant hindlimb buds(Fig. 6E,F). We were unable to observe the expression of Shh, the definitive marker of the posterior zone of polarizing activity (ZPA) function, as this gene is not expressed in the hindlimb until late 10.5 dpc, after Tbx4 mutants are dead (data not shown). However, we did examine dHand (Hand2– Mouse Genome Informatics), which lies upstream of Shh and is also normally restricted to the posterior edge of limb buds. In Tbx4-mutant hindlimb buds, dHand is expressed along the entire distal margin of the bud, indicating a defect in anterior/posterior patterning (Fig. 6G,H). dHand is normally excluded from the anterior limb bud by Gli3 expression in the anterior, which is normal in Tbx4mutants (data not shown). Tbx3, which is also required for posterior limb formation (Bamshad et al.,1997; Davenport et al.,2003), is correctly expressed in the posterior margin of Tbx4-mutant hindlimb buds (Fig. 6I,J).

Tbx2 is also normally expressed in the posterior hindlimb margin at this stage, although no function in limb development has yet been assigned to it. Tbx2 is absent from Tbx4-mutant hindlimb buds(Fig. 6K,L). This demonstrates that Tbx4 lies upstream of Tbx2 in at least two tissues, the allantois and the hindlimb.

Our work describes two mutant alleles of the Tbx4 gene that are phenotypically indistinguishable. The Tbx4^tm1.1Pa allele excises exon 5, which encodes most of the 5′ half of the T-box domain of Tbx4. Because exon 5 is a non-unit exon, this excision also creates a frameshift in subsequent exons. The Tbx4^tm1.1Patranscript, if translated, would produce only the N-terminal 138 amino acids of the protein. This hypothetical truncated protein would lack both the DNA-binding T-box domain and the large C-terminal domain, and is therefore predicted to be a functional null.

The Tbx4^tm1Pa allele consists of the insertion of a lox site and a floxed selection cassette into the introns surrounding exon 5,which creates no disruption of the Tbx4 exon sequences or splice sites. Nevertheless, the phenotype of this allele is indistinguishable from the null, which indicates that these intronic insertions result in profound gene disruption. This disruption is presumably due to the exogenous promoter and cryptic splice sites found in the neo selection cassette, both of which have been shown to disrupt gene function when inserted into non-coding regions (McDevitt et al.,1997; Meyers et al.,1998; Nagy et al.,1998). These reports have shown only partial gene disruption;however, they involve only the neo cassette, whereas we have used a neo-tk cassette. The addition of the tk gene results in an additional exogenous promoter and a polyadenylation site, which probably exacerbates interference by the neo-tk cassette. Alternatively, the Tbx4 locus could be especially susceptible to splicing disruption, or Tbx4 dosage sensitivity could be such that even partial disruption of Tbx4 passes below a minimum threshold limit.

Tbx4 mutants first exhibit a defect at 8.0 dpc, when they demonstrate a short allantois and failure of chorioallantoic fusion. At this age, homozygous-mutant, heterozygous and wild-type embryos display considerable overlap in phenotype (Fig. 3A). This stage marks the only appearance of a heterozygous effect of Tbx4, resulting in a delay in chorioallantoic fusion, although all heterozygous embryos eventually undergo chorioallantoic fusion and become indistinguishable from their wild-type littermates. However,homozygous-Tbx4 mutants are readily distinguished from littermates by the failure of chorioallantoic fusion.

We have attributed the death of Tbx4-homozygous mutants at 10.5 dpc to failure of chorioallantoic fusion. In the absence of an umbilicus, the embryo cannot exchange gasses, nutrients or waste with the maternal blood supply. Absence of the umbilicus also alters normal blood flow patterns, most probably accounting for the observed hemorrhage and pericardial edema. Several other mutations that result in failure of chorioallantoic fusion produce similar phenotypes (Gurtner et al.,1995; Kwee et al.,1995; Tremblay et al.,2001; Yang et al.,1995).

Extension of the allantois from the posterior axis of the embryo to the chorion results from a number of processes, including influx of cells from the primitive streak, cellular proliferation and cavitation of distal tissue(Downs and Bertler, 2000). Embryos mutant for Tbx4 have a normal primitive streak, as evidenced by normal expression of T, and proliferation in the allantois is also normal (Fig. 2B,C). However,cells near the tip of the Tbx4-mutant allantois undergo apoptosis(Fig. 2D,E). Factors required for cell survival in the allantois are not known. However, it is interesting to note that Tbx2 is downregulated in Tbx4-mutant allantoises, and previous work has implied a requirement for Tbx2 to prevent cell cycle arrest in rapidly proliferating tissues (Jacobs et al., 2000). Therefore, the role of Tbx4 in the allantois may be the maintenance of Tbx2-mediated suppression of apoptosis.

Also, in contrast to normal development, the allantois in Tbx4-mutant embryos shows no cavitation near the distal tip. Although the molecular nature of this defect is unknown, we find it relevant that T-box genes have been implicated in the regulation of adhesion molecules in several systems. In zebrafish and Xenopus, T-box genes have been shown to regulate paraxial protocadherin (Kim et al., 1998; Yamamoto et al.,1998), and, in mice, Tbr1 is a direct activator of reelin, an extracellular matrix glycoprotein. Mouse mutations in T produce primitive streak cells with a substrate-dependent migration defect (Hashimoto et al.,1987), and work with Fgfr1-mutant mice has revealed links between T and cadherin regulation(Ciruna and Rossant, 2001). This data suggests that the failure of allantois cavitation could be caused by misregulation of adhesion genes in Tbx4 mutants.

Despite the various defects inhibiting allantois growth, as many as 15% of Tbx4-mutant allantoises observed had extended into the dome of the chorion at the 4- to 5-somite stage (Fig. 1A; L.A.N. and V.E.P., unpublished). However, chorioallantoic fusion rarely occurs in Tbx4 mutants. Of more than 250 mutant embryos observed, only two had established any connection between the allantois and chorion. The lack of expression of Vcam1, a key adhesion molecule affecting this process (Gurtner et al.,1995; Kwee et al.,1995), in the Tbx4 mutants is clearly a contributing factor. However, Vcam1 homozygous-null embryos undergo chorioallantoic fusion in 20-50% of cases, so this defect alone is insufficient to explain the Tbx4 phenotype. It is possible that the combination of the lack of VCAM1 protein and the extensive apoptosis in the tip of the allantois produces a synergistic loss of chorioallantoic fusion potential.

In addition to the failure of chorioallantoic fusion in Tbx4mutants, there is also a block in vascular remodeling in the allantois. Endothelial cells, evidenced by positive staining for PECAM, successfully differentiate from allantoic mesenchyme, but they remain as discreet clumps of cells and fail to remodel into primary vessels. These clumps presumably correspond to the double-layered vesicles seen histologically, and are also presumably the progenitors of the blood-filled vesicles seen morphologically in the residual allantois at 10.5 dpc. It is unclear whether the failure of endothelial cells to coalesce into vessels in Tbx4 mutants is cell autonomous, or represents a defect in signaling or adhesion characteristics in surrounding cells of the allantois.

Bmp4 has previously been identified as a key regulator of allantois development. It is required in the extra-embryonic ectoderm for allantois induction, and later in the allantois itself for normal development and chorioallantoic fusion. The phenotype produced by the absence of Bmp4 in the allantois shares many points of similarity with the Tbx4 phenotype. Both exhibit a shortened, unfused allantois with no apparent proliferative defect, neither express VCAM1 at the tip of the allantois and both have defects in the vascularization of the allantois. Previous work has suggested that T-box genes can be regulated by BMP signaling(Koshiba-Takeuchi et al.,2000; Smith et al.,1991; Yamada et al.,2000), and the observation that Bmp4 expression in the allantois is unaffected by the loss of Tbx4 is consistent with an upstream role for Bmp4 with respect to Tbx4. However, the absence of Bmp4 also severely disrupts PGC placement and formation,whereas this process is unaffected by a lack of Tbx4. It may be that Tbx4 is a downstream effector of Bmp4 in the allantois,whereas other molecules act as intermediaries for Bmp4 signaling to the PGCs.

Much is known about the initiation of limb morphogenesis(Tickle and Munsterberg,2001). Anterior/posterior, dorsal/ventral and distal/proximal axes are all set up early in limb bud formation. Fgf10 signaling from the limb mesenchyme induces an apical ectodermal ridge (AER) in the overlying ectoderm. The AER, in turn, signals back to the mesenchyme through Fgf8 and Fgf4 to maintain Fgf10 in the region underlying the AER known as the progress zone. Fgf10 is required to maintain proliferation in the progress zone, which drives limb outgrowth. This represents an FGF-mediated positive feedback loop, which is required for limb development. A second feedback loop is set up between the AER and a region in the posterior mesenchyme known as the zone of polarizing activity (ZPA). The ZPA directs posterior patterning in the limb and is marked chiefly by the expression of the secreted signaling molecule Shh. Failure of either of these feedback loops results in the absence or dramatic reduction of limb growth and patterning.

At early 10.5 dpc, Tbx4-mutant embryos have morphologically obvious hindlimb buds that are similar to stage-matched wild-type controls. In many respects, the initiation of these hindlimb buds is normal. Hindlimb specificity in Tbx4 mutants appears unaltered, as indicated by the presence of Ptx1 and Tbx4 expression, and the lack of Tbx5 expression. Ventrally restricted expression of Msx2 and correct positioning of Fgf8 expression both suggest that dorsoventral patterning in unaffected by the absence of Tbx4(Pizette et al., 2001). Several genes known to be involved in limb outgrowth and/or FGF reciprocal signaling are correctly activated, including Msx1 and the gene encoding dp-ERK in the progress zone, Fgf8 in the presumptive AER and Twist throughout the hindlimb mesenchyme.

However, the ablation of Tbx4 does result in some defects in anterior/posterior patterning of the hindlimb. In normal limbs, dHandis restricted to the posterior limb bud by the action of Gli3 in the anterior limb bud (te Welscher et al.,2002). dHand induces Shh in the posterior limb bud, and Shh signaling is key for the activity of the ZPA. Tbx4-mutant embryos die prior to the expression of Shh, but,in mutant embryos, dHand is shifted from posterior-specific expression to expression throughout the mutant hindlimb bud despite normal expression of Gli3. Conversely, Tbx3, which is known to affect posterior limb development in mouse and human(Bamshad et al., 1997; Davenport et al., 2003), is correctly expressed in the posterior margin of the Tbx4-mutant hindlimb. Tbx2, which is also normally expressed in the posterior margin of the hindlimb bud, is absent in Tbx4-mutant hindlimbs. As no function has been identified for Tbx2 in limb development, this is notable chiefly because it shows that Tbx2 expression is ablated in Tbx4 mutants in both of the tissues where the two genes are co-expressed prior to the death of the mutant embryos, thus revealing a role for a T-box gene in regulating another T-box gene.

Despite successful induction of the hindlimb bud, and of many outgrowth and patterning genes therein, neither hindlimb outgrowth nor Fgf10expression is maintained in Tbx4-mutant hindlimbs. Initial FGF signaling is apparently normal in Tbx4-mutant hindlimbs, as visualized by the induction of Fgf8 in the ectoderm and phosphorylation of ERK in the mesenchyme. However, by 10.5 dpc, Fgf10is absent from the distal mutant hindlimb buds and only residual flank expression can be seen. As the absence of Fgf10 has been shown to ablate limb formation (Min et al.,1998; Sekine et al.,1999), it is unsurprising that Tbx4-mutant hindlimbs fail to progress when grown in culture.

Similar roles of Tbx4 in the forelimb and Tbx5 in the hindlimb have been proposed in previous studies(Gibson-Brown et al., 1998; Logan and Tabin, 1999; Ohuchi et al., 1998; Rodriguez-Esteban et al.,1999; Ruvinsky and Gibson-Brown, 2000; Saito et al., 2002; Tanaka et al.,2002). This study, and the recently published study of Tbx5 in the forelimb (Agarwal et al., 2003), reveals that the roles of Tbx4 and Tbx5 are subtly different. In each case, the specification of the relevant limb field proceeds despite the lack of Tbx4 or Tbx5. However, in the case of Tbx4 ablation, FGF signaling is initiated but not maintained, whereas Tbx5 ablation precludes any expression of Fgf10 or Fgf8.

By marker analysis, the Tbx4 mutation closely resembles genetic manipulations that disrupt reciprocal FGF signaling from the AER to the limb mesenchyme, such as mutations in Twist, an upstream regulator of Fgfr1 in limb mesenchyme(O'Rourke et al., 2002; Zuniga et al., 2002), and the AER-specific ablation of both Fgf8 and Fgf4(Sun et al., 2002). All of these disruptions exhibit successful induction of AER-specific FGFs,disruption of some but not all anterior/posterior axis markers, normal induction of FGF-independent limb markers, such as Msx1, and rapid downregulation of Fgf10. In particular, mutations in Twistresult in the expansion of Shh and Hoxd11 into the anterior of the limb bud, whereas in the absence of Fgf8 and Fgf4,the anterior expression of Alx4 is expanded posteriorly. The anterior expansion of Shh and Hoxd11 occurs despite the normal expression of Gli3 in the anterior of these limbs, demonstrating the importance of FGF signaling in the Gli3-dependent maintenance of the anterior/posterior axis of the limb bud.

Our data shows that FGF reciprocal signaling in Tbx4-mutant hindlimb mesenchyme is successful up to the point of ERK phosphorylation by the FGF-activated MAPK cascade. However, we observed a failure in the maintenance of Fgf10 and an expansion of the posterior limb domain,as evidenced by dHand expression, both of which are consistent with a failure in FGF signaling. This suggests that the role of Tbx4 in early limb development is to transduce the signal of the MAPK cascade to its final limb targets (Fig. 7),including both anterior dHand repression and Fgf10upregulation. Aside from Gli3, the transcriptional co-factors involved in dHand repression are not known, but it is possible that Tbx4 is an FGF-sensitive, direct repressor of this gene;alternatively, the derepression of dHand in Tbx4-mutant hindlimbs could be a secondary result of disrupted FGF signaling. Fgf10 is probably a direct target of Tbx4, as T-box binding sites have been observed in the Fgf10 promoter and Fgf10 is capable of T-box dependent reporter gene activation in vitro(Agarwal et al., 2003; Ng et al., 2002). It seems likely therefore that Tbx4 is a MAPK-sensitive, direct regulator of both dHand and Fgf10.

We thank members of the laboratory, especially Kat Hadjantonakis and Sarah Goldin for helpful discussions, G. Shackelford, A. Joyner, A. McMahon, and G. Martin for generously supplying probes, and Peter Mombaerts for the floxed PGK-neo PGK-thymidine kinase double selection cassette. This work was supported by a grant from the NIH (HD33082) (to V.E.P.) and a National Science Foundation pre-doctoral fellowship (to L.A.N.).