Tbx4 is not required for hindlimb identity or post-bud hindlimb outgrowth

Vertebrate limbs consist of paired appendages that emerge from lateral plate mesoderm during embryonic development. The outgrowth of the limb along the proximodistal axis is largely governed by reciprocal fibroblast growth factor (FGF) signaling between the limb mesenchyme and the apical ectodermal ridge (AER), a ridge of specialized epithelium extending along the dorsoventral boundary of the limb apex(Capdevila and Izpisua-Belmonte,2001; Niswander,2003). Limb budding is initiated by the expression of Fgf10 in the limb field mesenchyme, and mutation of Fgf10ablates limb initiation (Min et al.,1998; Ohuchi et al.,1997; Sekine et al.,1999). In response to mesenchymal Fgf10 signaling, the AER upregulates multiple partially redundant FGF genes, Fgf8 and Fgf4 being of central importance. Disruption of this mesenchymal-epithelial feedback loop, either by genetic manipulation or physical removal of the AER, results in loss of most limb-specific markers and drastic truncation of the limb. During the process of distal outgrowth, the limb is patterned along the anteroposterior axis by a mesenchymal posterior signaling center called the zone of polarizing activity (ZPA). Shhhas been shown to be both necessary and sufficient for ZPA activity. The AER and ZPA function similarly in both forelimb and hindlimb and little is known about the regulatory differences that produce the morphological differences between, for instance, the wings and legs of a chicken or the arms and legs of a human. Candidate transcription factors that could coordinate morphogenetic differences throughout limb development have been sought, and attention has centered on the T-box genes Tbx4 and Tbx5.

The T-box family is an evolutionarily ancient family of transcription factors characterized by a shared DNA-binding domain. Several T-box genes are expressed in the limb (Naiche et al.,2005), and heterozygous mutations in TBX3, TBX4 and TBX5 cause limb defects in humans(Bamshad et al., 1997; Basson et al., 1997; Bongers et al., 2004). In mouse, all of the Tbx2 subfamily (Tbx2, Tbx3, Tbx4 and Tbx5)have been shown to play roles in limb development(Agarwal et al., 2003; Davenport et al., 2003; Harrelson et al., 2004; Naiche and Papaioannou, 2003). Tbx5 is expressed specifically in the forelimb, while Tbx4,a closely related gene, is expressed in the hindlimb. Both genes are expressed in their respective limb fields well before the morphological appearance of the limb bud and continue to be expressed throughout the limb mesenchyme through late gestation (Gibson-Brown et al., 1996). Due to this differential and early expression, it has been proposed that Tbx4 and Tbx5 play central roles in creating the differences between forelimbs and hindlimbs, and several experiments in chick have suggested that ectopic expression of these genes can transform limb fates (Rodriguez-Esteban et al., 1999; Takeuchi et al.,1999; Takeuchi et al.,2003). However, these experiments were complicated by the expression of endogenous Tbx4 and Tbx5. More recent work in which Tbx5 was replaced with Tbx4 has suggested that Tbx4 and Tbx5 can substitute for each other and serve little or no role in establishing the differences between limbs(Minguillon et al., 2005).

Tbx5 plays a crucial role in the initiation of the forelimb bud. Null alleles of Tbx5 in mouse and zebrafish result in embryos that show no forelimb bud formation and do not express Fgf10 in the forelimb field (Agarwal et al.,2003; Garrity et al.,2002; Ng et al.,2002; Rallis et al.,2003). Several lines of evidence have been used to show that Tbx5 mediates limb outgrowth through direct transcriptional regulation of Fgf10 (Agarwal et al., 2003; Ng et al.,2002), and experiments with a presumed dominant negative allele of Tbx5 suggest that loss of this gene during limb development truncates limb outgrowth (Rallis et al.,2003). The high conservation, similar expression pattern and functional redundancy between Tbx4 and Tbx5 strongly suggest that Tbx4 would operate similarly in the hindlimb. However, our previous work has shown that null mutations in Tbx4 produce a slightly different limb phenotype than Tbx5 null mutations, in that Fgf10 expression is initiated in the hindlimb field of Tbx4mutant embryos and a morphological bud is formed, but the hindlimb bud does not maintain Fgf10 expression or grow in explant cultures(Naiche and Papaioannou,2003). However, embryonic lethality of the Tbx4 null mutation due to failure of chorioallantoic fusion prevented our examining limb outgrowth in vivo.

In this study we exploit a conditional allele of Tbx4 to circumvent allantois failure and extend our data on the phenotype of Tbx4 null hindlimbs. As Tbx4 and Tbx5 have subtly different roles in limb bud initiation, we investigated whether loss of Tbx4 during later limb development also results in limb truncation,similar to the dominant-negative Tbx5, by ablating Tbx4 gene function at multiple stages of limb development. Our findings indicate that early expression of Tbx4 is required for maintaining proximal and medial limb tissue, but does not solely regulate Fgf10 or limb outgrowth. Examination of hindlimbs that had lost Tbx4 function revealed no role for Tbx4 in hindlimb identity.

Mice carrying the Tbx4^tm1Pa allele containing three loxP sites (Naiche and Papaioannou, 2003) were mated with EIIa-cre transgenic mice to generate germline mosaic males with all possible recombination products (Xu et al., 2001),which were outcrossed to isolate the conditional allele. Routine genotyping was performed using primer A: GAGGATGTTCCCCAGCTAC and primer B:CAGTCTGAGAGGGTCAGACTC (Fig. 1A).

The Rosa-ERcre^T2 (de Luca et al., 2005) and Prx-cre(Logan et al., 2002) lines were the kind gifts of Thomas Ludwig and Cliff Tabin, respectively. The EIIa-cre line was obtained from Jackson Labs (strain #003314). All mice were kept on outbred backgrounds. The dark period was 19.00 to 05.00 h and noon on the day a mating plug was observed was identified as embryonic day(E) 0.5. Yolk sacs were used for genotyping.

Tamoxifen (Sigma #T5648, 20 mg/ml in sunflower oil) was administered via intraperitoneal injection. At E7.5, 6 mg (approximately 0.18 mg/g body weight)was administered between 17.30 and 23.30 h. At E9.5 and later time points, 7 mg was administered between 17.30 and 19.30 h.

Embryos from Tbx4^cond/cond; Rosa-ERcre matings were examined for the recombination of Tbx4^cond into Tbx4^tm1.1Pa (the null allele) using primers A and B(above) in combination with the excision-specific primer C:TCATCTAGGCTTCACAGCC. For Prx-cre crosses, all embryos were genotyped for wild type, Tbx4^tm1.1Pa, and Tbx4^cond alleles using primers A, B and C as well as for the presence of Prx-cre (using primers CGATGCAACGAGTGATGAGG and GCATTGCTGTCACTTGGTCGT). Quantitation of alleles in Fig. 1 was done using primers A, B and C with a graded series of mixed DNA of known composition prepared identically to sample DNA.

Whole-mount in situ hybridization was performed according to standard protocols (Wilkinson and Nieto,1993). Two to five embryos were used for each marker and stage. Alcian Blue and Alcian Blue/Alizarin Red skeletal preparations were performed according to standard protocols (Nagy et al., 2003), with the modification that Alcian Blue stain in the latter protocol was prepared at 150 mg/l in 80% ethanol, 20% acetic acid.

Limb widths were determined by photographing the dorsal aspect of each limb and measuring across the widest part of the limb perpendicular to the proximodistal axis. Cell counts were obtained by dissecting off limbs,dissociating the tissue with 2 μg/ml collagenase for 1 hour at 37°C and counting cells using a hemacytometer. The mitotic index of the progress zone was computed by staining sections with anti-phosphohistone H3 (Upstate),counterstaining with Nuclear Fast Red and counting nuclei in the area 150μm subjacent to the AER in the center of the limb. Cell death was assessed by incubating live embryos in 5 μmol/l Lysotracker Red (Invitrogen L7528)in Hank's balanced salt solution for 30 minutes at 37°C.

A Tbx4 conditional allele, Tbx4^tm1.2Pa, was generated by excision of the selection cassette from the Tbx4^tm1Pa allele(Naiche and Papaioannou,2003), leaving intronic loxP sites flanking exon 5(Fig. 1A). Our previous work has demonstrated that removal of exon 5 produces a null allele(Naiche and Papaioannou,2003). Animals homozygous for the conditional allele, hereafter referred to as Tbx4^cond, were viable and fertile and showed no phenotype. Homozygous Tbx4^cond females were mated to homozygous Tbx4^cond males that also carried an optimized tamoxifen-inducible cre gene expressed ubiquitously under the Rosa26 locus, Rosa-ERcre^T2(de Luca et al., 2005). Pregnant females were administered a single injection of tamoxifen at various stages between E6.5 and 11.5. When females were injected at E7.5, embryonic excision of Tbx4^cond was complete within 24 hours(Fig. 1B), and a probe directed against the deleted region showed no intact Tbx4 transcript in the hindlimb field (Fig. 1C). When females were injected at E9.5, embryos showed 70-95% excision of Tbx4^cond within 24 hours, greater than 98% excision after 48 hours (Fig. 1D), and complete excision at later time points (not shown). Despite incomplete excision at the DNA level, Tbx4 transcripts were undetectable by in situ hybridization after 24 hours (Fig. 1E).

In experiments with the Rosa-ERcre transgene, we compared Tbx4^cond/cond embryos (control) to Tbx4^cond/cond; Rosa-ERcre/+ embryos (hereafter referred to as ERcre embryos) to ensure that the effects we saw were not due to either the Tbx4^cond allele or to tamoxifen administration. During these experiments, we noted that Cre activity from the Rosa-ERcre caused apoptosis and fetal lethality, which will be described elsewhere, so we also examined Tbx4^cond/+; Rosa-ERcre/+ embryos to ensure that the observed phenotypes were due to the loss of Tbx4 rather than the effects of Cre activity. To first verify that excision of Tbx4^cond produced the expected phenotype, we administered tamoxifen at E6.5, approximately 24 hours before expression of Tbx4 is seen in any tissue. ERcre embryos(n=19) were indistinguishable from Tbx4 null embryos and died at E10.5 due to failure of chorioallantoic fusion (data not shown).

We then administered tamoxifen at E7.5 to excise Tbx4 after the initial formation of the allantois, but before expression appeared in the hindlimb field. Tamoxifen injections at E7.5 resulted in complete excision of Tbx4 within 24 hours (Fig. 1B), well before hindlimb expression appeared at E9.5. Approximately half of the ERcre embryos recapitulated the null phenotype, while the remaining ERcre embryos underwent allantoic fusion and survived past E10.5. In these embryos, a visible limb bud was formed but was degenerating by E11.5 (Fig. 2A,B). Early limb markers in ERcre embryos injected with tamoxifen at E7.5 were identical to those of Tbx4 null embryos: at E10.0 Fgf10 was weakly expressed in the hindlimb field(Fig. 2C,D) and sporadic Fgf8 was seen in the AER (Fig. 2E,F), but Fgf10 was lost from the hindlimb bud by E10.5(Fig. 2G,H).

Survival of some ERcre embryos given tamoxifen at E7.5 allowed us to examine patterning of Tbx4-ablated hindlimbs at later stages. The posterior determinant dHand (Hand2 - Mouse Genome Informatics), which is upstream of Shh in the very early limb bud, is normally expressed in the posterior of the hindlimb bud at E10.5, but was diffusely expressed across the ventral margin of the entire ERcrehindlimb (Fig. 2I,J). Conversely, the expression domain of Alx4, which normally marks the anterior of the limb bud, was also expanded across the entire hindlimb bud in ERcre embryos at E10.5 (Fig. 2K,L). In more advanced E10.5 embryos, Shh appeared in the posterior hindlimb, but it was not observed in the ERcrehindlimbs (Fig. 2M,N). By E11.5 all limb markers examined, including Fgf8, Shh, dHand, Fgf10 and Alx4, had disappeared from the ERcre hindlimb(Fig. 2O-T and data not shown). This confirms that Tbx4 expression in the hindlimb field is required for hindlimb outgrowth in vivo as well as in vitro. This also demonstrates that the Tbx4^cond allele in combination with ERcre can successfully be used to bypass early lethality caused by the effect on the allantois while still producing null phenotypes in later tissues.

Tbx4 is expressed broadly throughout E9.5-12.5 hindlimbs and then progressively distally restricted(Gibson-Brown et al., 1996),in a similar fashion to its putative target Fgf10. To test the requirement for Tbx4 in post-bud hindlimb development and outgrowth,we ablated the gene during stages at which it is widely expressed. Tamoxifen injection at E9.5, 10.5 and 11.5 did not produce limb truncations in E14.5 embryos, although the hindlimbs of ERcre embryos injected at E9.5 and 10.5 were slightly shorter (Fig. 3A-D). Anterior digit fusion was seen in some ERcreembryos from E10.5 injections (42%, n=24)(Fig. 3C). Anterior digits were abnormal in all ERcre embryos injected at E9.5, with 67% having four symmetrical digits and 33% showing fusion of anterior digits (n=88)(Fig. 3D-G).

Cartilage stains were done on E14.5 embryos from tamoxifen injections at all of the above stages (Fig. 3H-J). Abnormal pelvic connections were seen in ERcreembryos from E10.5 and 9.5 tamoxifen injections(Fig. 3I,J). No defects were observed in the forelimbs of any embryos(Fig. 3K,L) or in hindlimbs of embryos injected at E11.5 (data not shown). Cre-expressing Tbx4heterozygous embryos given tamoxifen at E9.5 developed hindlimbs identical to control embryos, demonstrating that Cre activity produced no hindlimb defects when an intact copy of Tbx4 was present(Fig. 3M,N). However, hindlimbs from ERcre embryos injected at E10.5 and 9.5 had hypoplastic pelvises and fibulas, aplastic or severely hypoplastic femurs, which did not articulate with the pelvis, and abnormal anterior digits(Fig. 3O-T). In some ERcre embryos from E10.5 injections digit formation was normal, but in some of these embryos the hindlimb digits I and II were thinner than controls and the metatarsal of the first digit appeared to originate near the middle of the metatarsal of digit II instead of near the tarsal bones(Fig. 3Q). The autopod of some ERcre embryos injected at E9.5 had thin and partially fused digits I and II (Fig. 3R), similar to E10.5 injections, while others had four-digit autopods, representing either a complete fusion between digits I and II or the loss of digit II(Fig. 3S). In some embryos,soft-tissue fusion appeared to have occurred between digits II and III(Fig. 3G and data not shown),but no fusion between the skeletal elements of digits II and III were ever observed. Despite the obvious abnormalities, all skeletal elements formed in ERcre embryos injected at E9.5, including the most distal phalanges,revealing no outgrowth requirement for Tbx4.

Our previous work with the null allele showed that Tbx4 is not required for the initial expression of hindlimb identity markers at E10.5, but we could not evaluate the maintenance of hindlimb identity in the absence of Tbx4 due to the failure of hindlimb outgrowth. Using Tbx4^cond we examined hindlimb identity in ERcreembryos injected with tamoxifen at E9.5.

In normal E14.5 embryos, all forelimb skeletal elements reside in the same plane (Fig. 3K), whereas hindlimb skeletal elements rotate such that the femur were nearly perpendicular to the plane of the tibia and footplate(Fig. 3O). In ERcreembryos, the relative positions of the pelvis and distal limb elements clearly indicated a hindlimb-like orientation (Fig. 3S,T). Additionally, the carpal bones of the normal forelimb are short (Fig. 3K), while the homologous tarsals in the normal hindlimb form two noticeably longer bones,the talus and the calcaneus. ERcre hindlimbs showed evident formation of the talus and calcaneus (black arrowheads in Fig. 3O,S).

Tbx4 itself is a marker of hindlimb identity, so the expression of this gene was examined using a probe 3′ of the deleted region (3′probe, Fig. 1A). Tbx4was expressed robustly throughout the hindlimbs of ERcre embryos at all stages examined (Fig. 4A-Dand data not shown). Tbx5 is a marker of forelimb identity, and some evidence suggests that Tbx4 and Tbx5 function antagonistically to exclude each other from their respective limbs(Takeuchi et al., 1999). However, Tbx5 was maintained exclusively in the forelimb of ERcre embryos at least 3 days after Tbx4 function had been lost (Fig. 4E-H). Recent evidence suggests that Pitx1 is a major determinant of hindlimb identity (Logan and Tabin,1999; Minguillon et al.,2005), and this gene was maintained in ERcre hindlimbs at all stages examined (Fig. 4I-L). A more downstream reporter of hindlimb fate, Hoxc9, was also maintained in embryos that had lost Tbx4gene function (Fig. 4M-P). By both molecular and morphological markers, we found no evidence for a role for Tbx4 in determining hindlimb identity.

Our data show that Tbx4 plays a major role in formation of the skeletal elements of the limb, and both anterior and posterior skeletal elements (digits and fibula, respectively) were missing from E14.5 ERcre embryos injected with tamoxifen at E9.5. We therefore investigated earlier limb patterning in ERcre embryos injected at E9.5 to determine the source of these defects.

The timing of the appearance and growth of the hindlimb buds at E10.5 did not vary between control and ERcre embryos (data not shown). To determine the presence of the ZPA, we first examined the expression of Shh in the limb. Shh was expressed in posterior mesenchyme of E10.5 and 11.5 hindlimbs and was not appreciably different in ERcre embryos (Fig. 5A-B′). To observe the effects of Shh signaling, we examined the expression of Ptc (Ptch1 - Mouse Genome Informatics), a Shh response gene. Ptc appeared in its normal domain in both E10.5 and 11.5 ERcre hindlimbs, but by E11.5 this domain encompassed a larger proportion of the hindlimb(Fig. 5C-D′). We also examined dHand, which is both required for and dependent on Shh signaling, and found a similar result, with normal expression in E10.5 ERcre embryos and a normal-sized expression domain in E11.5 ERcre embryos, but with that domain encompassing more of the hindlimb(Fig. 5E-F′). The expression domains of Ptc and dHand at E12.5 were also normally sized but comprised a greater proportion of the hindlimb in ERcre embryos than in controls (data not shown).

Having observed an apparent expansion of the posterior limb domain, we also examined anterior patterning. Alx4 marks the anterior half of the E10.5 hindlimb bud and is restricted to an anteroproximal domain by E11.5(Fig. 5G,H). In ERcreembryos, there is a dramatic expansion of Alx4 expression, comprising nearly the entire hindlimb in E10.5 embryos(Fig. 5G′) and a broad expanse of the proximal limb bud in E11.5 embryos(Fig. 5H′). Previous work has suggested that Alx4 represses Shh expression and is repressed by dHand (Qu et al.,1997; te Welscher et al.,2002), but both these genes showed expression that overlapped that of Alx4. A second anterior marker, Pax1, was examined and found to be normal in E10.5 ERcre hindlimbs (data not shown).

As both anterior and posterior limb markers are present in ERcrehindlimbs, we examined the limb margins using Tbx2 and Tbx3. Both these genes were expressed normally in E10.5 ERcre embryos (data not shown). At E11.5, Tbx2 and Tbx3 were expressed in both the anterior and posterior margin of the limb mesenchyme of control embryos(Fig. 5I,K). In ERcrehindlimbs, these expression domains were present and normally sized, but the space between them was obviously narrower, suggesting that there is less tissue in the medial core of the limb (Fig. 5J,L). This was confirmed by cell counts, which showed that the hindlimbs of ERcre embryos contained significantly fewer cells(2.8±1.0×10⁶, n=12) than control hindlimbs(4.3±1.1×10⁶, n=13) with a confidence interval of P<0.015, while forelimbs showed no significant difference (4.2±1.0×10⁶ and 5.2±1.2×10⁶ cells, respectively). In addition,forelimb width was similar between E11.5 control and ERcre embryos,but the ratio of hindlimb to forelimb width was significantly lower in ERcre embryos, with a confidence interval of P<0.0001(Fig. 5S). This loss of limb core tissue explains the apparent expansion of Shh responsive genes across the limb, as posterior signaling will reach proportionately farther across a thinner limb.

Tbx15, the loss of which has been shown to decrease proliferation in the limb core (Singh et al.,2005), was normal in ERcre embryos, indicating that the loss of limb core in Tbx4 mutants occurs along a different pathway(Fig. 5N,O). Bone morphogenetic proteins (Bmps) have multiple roles in limb development, including AER maintenance, skeletal formation and apoptosis of interdigital regions(Capdevila and Izpisua-Belmonte,2001), and Bmps are known to interact with T-box genes in several systems (Papaioannou and Goldin,2003). Because post-bud Tbx4 ablation causes aberrant hindlimb digit formation, we looked at both early and late Bmp expression. The Bmp2 and Bmp4 expression domains were normal in ERcre embryos at E10.5 and 11.5, although they also suffered from a loss of medial non-expressing tissue (Fig. 5O,P and data not shown). In E13.5 ERcre embryos,interdigital expression of Bmp2 was distally restricted in the anterior of the hindlimb compared with control embryos, but remained present between each digit, suggesting that the digital fusions observed are due to a change in Bmp regulation rather than the absence of Bmp2 expression(Fig. 5Q,R).

To explain the loss of limb core tissue observed with these various assays,we examined the mitotic index in control and ERcre hindlimbs progress zones, but found no significant differences (3.00±0.75 in controls,2.86±0.73 in ERcre hindlimbs). Cell death was difficult to assess due to the heightened background of cell death caused by Rosa-ERcre activity.

Our previous work showed a requirement for Tbx4 in both outgrowth and maintenance of Fgf10 expression shortly after limb initiation. Since our current evidence showed no loss of limb outgrowth after Tbx4 ablation at a later stage, we examined ERcre embryos injected with tamoxifen at E9.5 for genes known to be involved in the FGF feedback loop. Despite previous indications that Fgf10 is a direct transcriptional target of Tbx4, we found substantially normal expression of Fgf10 in ERcre embryos at E10.5, 11.5, 12.5 and 13.5 (Fig. 6A-H). The only perturbation of Fgf10 was associated with digit fusion at E13.5(Fig. 6H).

Fgf8 is the first, and only non-redundant, FGF ectodermal response to Fgf10 limb signaling and is normally expressed along the entire AER. Fgf8 expression was present in the AER of ERcre embryos at E10.5 and 11.5 (Fig. 6I-L),but close examination revealed an anterior truncation of the Fgf8domain (Fig. 6K′-L′). This anterior loss is consistent with the anterior bias of digit loss observed in E14.5 embryos.

FGF receptor FgfR1 is known to be key in the limb FGF signaling loop (Ciruna et al., 1997; Verheyden et al., 2005), and loss of its expression immediately after hindlimb bud formation leads to a similar phenotype to that of Tbx4(Li et al., 2005). Expression of FgfR1 was nonetheless robust throughout the hindlimbs of ERcre embryos (Fig. 6M,N). Spry1 is a mesenchymal response to Fgf8signaling and is thought to act as a negative regulator of FGF signaling. Spry1 was observed in its normal domain immediately underlying the AER in ERcre hindlimbs, although that domain was anteriorly truncated, mirroring the truncated expression domain of Fgf8(Fig. 6O,P). Wnt signaling is thought to be part of the FGF limb feedback loop and regulated by Tbx4 and Tbx5 in limbs, but the precise Wnt genes involved are not known in mouse (Agarwal et al.,2003; Kawakami et al.,2001; Takeuchi et al.,2003). As a proxy, Lef1 expression, which is activated in response to Wnt signaling, was examined. Lef1 expression was present throughout the distal limb in both control and ERcre embryos. Thus,all elements of the FGF limb feedback loop examined in ERcre embryos injected at E9.5 were either expressed normally or with minor perturbations in the anterior hindlimb.

Prx-cre drives cre expression in forelimbs and hindlimbs(Logan et al., 2002) and in combination with the conditional Tbx5 allele results in pups wholly lacking forelimbs (Rallis et al.,2003). To bypass the fetal lethality caused by the Rosa-ERcre allele, we produced a limb-specific deletion of Tbx4 by mating Tbx4^cond/cond females with Tbx4^tm1.1Pa/+(null allele); Prx-cre/+ males to generate Tbx4^cond/tm1.1Pa; Prx-cre/+ embryos(prx-cre) and Tbx4^cond/+; +/+ embryos (control). We dissected early limb bud embryos to observe the kinetics of Tbx4excision in the presence of Prx-cre. Intact Tbx4, as measured by the deletion-specific probe(Fig. 1A), was significantly downregulated but not entirely lost in advanced E10.5 prx-cre embryos(Fig. 7A-D), indicating that prx-cre hindlimbs express Tbx4 for longer than ERcre hindlimbs administered tamoxifen at E9.5. Cartilage staining of E15.5 embryos revealed that the hindlimb in prx-cre embryos had a hypoplastic pelvis and fibula, severely hypoplastic femurs and mild or nonexistent anterior digit fusions, consistent with the phenotype of ERcre embryos given tamoxifen at E10.5(Fig. 7E,F). The ilium and ischium were identifiable, but the pubic rami were missing. At E11.5, we did not observe any major difference in apoptotic cells in the hindlimb between prx-cre embryos and controls (data not shown). prx-cre pups were viable but had abnormally turned hindlimbs and abnormal pelvic regions(Fig. 7G,H). Skeletal preparations showed normal skeletal development in prx-cre pups(Fig. 7I,J), but the hindlimbs were turned nearly backwards and not articulated with the pelvis(Fig. 7I-L). It is unclear whether the abnormal turning of the hindlimb in prx-cre embryos is a primary defect due to the loss of Tbx4 or secondary to the loss of the proximal tissues as leverage points.

Compared with control hindlimbs, neonatal prx-cre hindlimbs showed relatively normal tibia growth and ossification, while only scraps of cartilage remained of the fibula and the femur. The hypoplastic pelvis had ossified, but the absence of a central cartilaginous region suggests that this represents the fusion of the remaining ilium and ischium(Fig. 7M-O). Foot development was only mildly affected in prx-cre embryos, which had partially fused tarsals and a reduced second digit(Fig. 7P,Q). These defects all appeared to be the direct outcome of defects observed at E14.5 in both prx-cre and ERcre embryos injected at E10.5, suggesting that Tbx4 has no further role in limb development between E14.5 and birth. Comparison of the prx-cre and ERcre phenotypes also demonstrates that all of the defects seen in the ERcre hindlimbs are specific to the loss of Tbx4 from hindlimb mesenchyme and that hypothetical interacting or migrating tissues play no role in producing the observed defects.

We have demonstrated that, despite widespread hindlimb-specific expression, Tbx4 is only required in a limited developmental window from its earliest time of expression, at E9.5, through to early limb formation, at E11.5. Loss of Tbx4 in the early part of this window halted limb outgrowth during limb bud formation, while loss of Tbx4 during bud formation or early limb formation did not affect limb outgrowth, but did result in a dramatic loss of proximal skeletal elements and a more modest loss of distal skeletal elements. During early limb development these limbs showed a consistent loss of core tissue, leading to the relative expansion of Shh responsive genes across the limb. Because these differences were already clear at E11.5 and ablation of Tbx4 after E11.5 led to no additional phenotype, the early loss of limb core tissue was almost certainly the origin of the skeletal malformations observed at E14.5. This suggests that the limb malformations seen in human heterozygous mutations of TBX4(Bongers et al., 2004), which cause mild malformation of pelvis, patella size and toe placement (Small Pateela Syndrome, OMIM 147891), are determined in early development rather than during later limb outgrowth.

Gene ablation and lineage tracing has shown that digit I is Shh-independent, and digit II is induced by neighboring Shhsignaling, while digits III, IV and V arise from tissue that once expressed Shh (Harfe et al.,2004). In the post-initiation Tbx4-ablated hindlimb, the reduction of limb core tissue and the relatively greater area of Shhexpression reduced the tissue available to form digits I and II, resulting in thin digits, or partial transformation of all remaining tissue to a digit II-like fate.

Our previous work showed no role for Tbx4 in the initial establishment of hindlimb identity (Naiche and Papaioannou, 2003), and our current work shows no requirement for Tbx4 in maintenance of hindlimb identity during later stages of limb development. This confirms that other transcription factors, such as Pitx1 (Minguillon et al.,2005), are responsible for coordinating hindlimb-specific transcriptional regulation and morphological formation.

Limb outgrowth is controlled by reciprocal signaling between Fgf10in the mesenchyme and Fgf8, Fgf4 and other Fgf genes in the AER(Niswander, 2003). Previous work has suggested that Tbx5, and by implication Tbx4, is a direct regulator of Fgf10 in the limbs(Agarwal et al., 2003; Ng et al., 2002). Although hindlimbs that never express Tbx4 initiate low levels of Fgf10 expression, this expression fails by early E10.5. However, as shown by our experiments with the conditional allele, the loss of Tbx4 at approximately E10.5 has no apparent effect on Fgf10expression. Thus, our work conclusively shows that Tbx4 is not required for Fgf10 expression in limb development after E10.5.

There nevertheless appear to be problems with FGF signaling in the absence of Tbx4. Several genetic manipulations that produce a partial loss of FGF signaling also produce limb phenotypes remarkably similar to that observed in post-bud loss of Tbx4. Loss of FgfR1 immediately after the initiation of hindlimb budding, approximately the same stage at which we have ablated Tbx4, causes anterior digit fusions producing a symmetrical four digit autopod (Li et al.,2005). Likewise, the loss of Fgf8 from the AER produces hindlimbs with hypoplastic femurs and fibulas and loss of anterior digits(Lewandoski et al., 2000). Other FGF perturbations have phenotypes that are more or less severe,reflecting different degrees and timing of FGF pathway disruption(Li et al., 2005; Moon and Capecchi, 2000; Sun et al., 2002; Verheyden et al., 2005), or that produce similar phenotypes in the forelimb(Moon and Capecchi, 2000). Partial loss of FGF signaling can also produce a similar spectrum of gene expression changes to loss of Tbx4, including loss of anterior Fgf8 expression and expansion of Alx4(Li et al., 2005; Sun et al., 2002; Verheyden et al., 2005). Narrowing of the non-expressing domain of Tbx2 and Tbx3 is also evident in figures (Li et al.,2005). It is probable that the reduction we have observed in Fgf8 is indicative of a partial loss of FGF signaling in Tbx4-ablated hindlimbs, a difference probably so slight that it is not observable by in situ hybridization for FGF pathway components.

Shh is thought to be repressed in the normal limb by Alx4(Qu et al., 1997). Shh also induces dHand expression, which in turn is thought to repress Alx4 (te Welscher et al., 2002), forming mutually exclusive domains of anterior and posterior limb signaling. However, ablation of Tbx4 either before limb development or during early limb development resulted in overlapping areas of Alx4, dHand and Shh expression. As noted above,several other disruptions of the FGF pathway cause expansion of the Alx4 domain, suggesting that this may be a general feature of loss of FGF signaling. It is possible that Alx4-Shh interactions are dependent on FGF, and consequently Tbx4, function.

The relatively minor effect of post-initiation Tbx4 loss on FGF signaling is inconsistent with a report using a dominant-negative allele of Tbx5, showing that post-initiation loss of Tbx5 halts FGF signaling and limb growth in the forelimb(Rallis et al., 2003). Recent results from the same lab show that post-initiation ablation of Tbx5function using a conditional allele does not result in limb truncations or loss of Fgf10 (Hasson et al.,2007). Instead, reduction of the forelimb along the anteroposterior axis is seen early in forelimb development, analogous to the thinner hindlimbs seen in our corresponding Tbx4 study. This suggests that Tbx4 and Tbx5 behave similarly in the limb, but that dominant-negative alleles produce a different phenotype than conditional ablation.

In order to explain this difference, we propose that regulation of Fgf10 in the limb is regulated by (at least) two modules, one (TBE)that is responsive to Tbx4 and Tbx5 and one (XBE) that is responsive to an unidentified transcription factor(Fig. 8). In the early limb field, this hypothetical transcription factor is either absent and unnecessary(in the forelimb) or insufficient to drive threshold levels of Fgf10expression (in the hindlimb). Thus both limbs are dependent on Tbx4or Tbx5 for establishment of the FGF feedback loop. Once the FGF signaling feedback loop has been successfully set up, Tbx4 or Tbx5 and the hypothetical transcription factor have additive effects on total FGF signaling, so the loss of Tbx4 or Tbx5 produces relatively mild FGF hypomorphic phenotypes. A dominant negative, where Tbx4 or Tbx5 is fused to a transcriptional repression domain, is capable of reducing Fgf10 transcription below threshold levels, halting limb outgrowth. This model also agrees with in vitro data according to which Tbx5 alone is capable of driving expression of an Fgf10 reporter (Agarwal et al.,2003; Ng et al.,2002). The differential requirement for T-box genes between initiation and maintenance of FGF signaling also explains the disparity in Tbx4 and Tbx5 conditional phenotypes when combined with the Prx-cre allele, as this transgene has been observed to start expression relatively later in the hindlimb than in the forelimb(Kmita et al., 2005; Logan et al., 2002), and does not seem to ablate Tbx4 gene function until after an FGF feedback loop has formed.

There are several candidates for this proposed Fgf10 regulator. Sall4 can drive Fgf10 limb expression and activates an Fgf10 reporter synergistically with either Tbx4 or Tbx5, but Sall4 expression is dependent on T-box regulation and is a poor candidate for an independent regulator of Fgf10(Harvey and Logan, 2006; Koshiba-Takeuchi et al.,2006). Pitx transcription factors are known to interact directly with T-box genes (Lamolet et al.,2001), and double mutation of Pitx1 and Pitx2produces a phenotype very similar to the post-initiation loss of Tbx4(Marcil et al., 2003), but loss of these genes also dramatically downregulates Tbx4, suggesting that the observed phenocopy is due to the downstream loss of Tbx4. A better candidate is Snai1, which is expressed in the hindlimb at a relatively earlier stage than in the forelimb, as predicted by our model, and appears to be upstream of Fgf10(Isaac et al., 2000). Also as predicted, Snai1 expression is maintained in the Tbx4 null hindlimb field (data not shown) and could therefore drive Fgf10 in the absence of Tbx4. Another excellent candidate is Lef1,which can directly regulate Fgf10 in vitro(Agarwal et al., 2003), and is maintained in the hindlimb after Tbx4 has been ablated.

While previous hypotheses (ours included) proposed that Tbx4 was a`master switch' that dictated hindlimb outgrowth and identity, it now appears that Tbx4 plays a more cooperative role in regulating these functions. Tbx4 probably coordinates with numerous other transcription factors to guide limb formation, but its importance should not be underplayed. Not only is Tbx4 crucial for starting hindlimb outgrowth and for the formation of hindlimb skeletal elements, but also it has been conserved as a hindlimb-specific transcription factor since the evolution of cartilaginous fish (Tanaka et al.,2002). Organisms that have subsequently lost hindlimbs have also lost Tbx4 expression (Cole et al.,2005; Shapiro et al.,2004; Tanaka et al.,2005). This suggests that there are still roles to be discovered for Tbx4, possibly with regards to regulation of elements of limb development other than the skeleton.

We thank Lee Ramsey, Andrew Goldsmith, Elana Ernstoff and Elinor Pisano for technical assistance, and members of the Papaioannou laboratory for critical discussion. We thank Malcolm Logan and Peleg Hanson for discussions and for sharing unpublished data. All animals were treated in accordance with local laws and regulations. This work was supported by N.I.H. grant #HD33082,V.E.P.