Tbx3 can alter limb position along the rostrocaudal axis of the developing embryo

Vertebrate limbs develop as budding outgrowths from the lateral plate mesoderm (LPM) on either side of the main body axis. The forelimb and hindlimb fields are located at specific positions along the rostrocaudal axis of the embryo and this position is fixed across vertebrate species. The forelimb forms at the cervical-thoracic junction, while the hindlimb develops at the level of the lumbar-sacral junction (Burke et al., 1995). Hox genes are candidates to specify limb position. Many of these genes are expressed in nested patterns along the rostrocaudal axis of the embryo and may provide positional cues to cells of the LPM that will give rise to limb buds (Burke,2000; Burke et al.,1995; Cohn et al.,1995; Cohn et al.,1997). However, neither gene deletion nor gene misexpression experiments have provided direct evidence for a role of Hox genes in limb positioning.

Tbx3 belongs to the Tbx2/3/4/5 subfamily of T-box genes that originated from a single ancestral gene through gene tandem duplication and cluster dispersion (Agulnik et al.,1996; Minguillon and Logan,2003; Ruvinsky et al.,2000; Wilson and Conlon,2002). Tbx3 is expressed in the limb-forming territories prior to overt limb bud outgrowth. At later stages (st.24 chick, 11.5 dpc in the mouse), Tbx3 is expressed in two stripes in the anterior and posterior limb mesenchyme (Gibson-Brown et al., 1998; Logan et al.,1998; Tumpel et al.,2002). Tbx3 is required for normal limb development as mutations in human TBX3 are associated with Ulnar-Mammary Syndrome(UMS, OMIM #181450), a dominant disorder characterized by upper(fore) limb deficiencies (Bamshad et al.,1997). Posterior structures of the limb, e.g. the ulna and fifth digit, are predominantly affected. Tbx3 deletion studies in the mouse produce phenotypes consistent with the abnormalities observed in UMS(Davenport et al., 2003).

Experiments in the chick have shown that the posterior domain of Tbx3 expression in the limb is positively regulated by Shhsignalling, while the anterior expression domain is repressed by Shh,suggesting a potential role of Tbx3 in the anteroposterior patterning of the vertebrate limb (Tumpel et al.,2002). Furthermore, recent misexpression experiments have suggested that Tbx3 can alter the identity of posterior digits in the developing chick hindlimb (Suzuki et al.,2004).

Misexpression and gene deletion studies have implicated Tbx3 in limb patterning during limb bud stages. However, Tbx3 is expressed in the limb-forming region prior to overt limb bud outgrowth. To examine a potential early role of Tbx3 in normal limb development, we have misexpressed transcriptional repressor and activator forms of Tbx3 in the developing forelimb region using the avian retroviral system. We provide evidence for a new role for Tbx3 in the genetic network that positions the limb along the rostrocaudal axis of the vertebrate embryo.

Eggs (Needle's farm, Winter's farm, The Poultry Farm) were incubated at 37°C and staged according to Hamburger-Hamilton (HH)(Hamburger and Hamilton,1951).

Production of retroviral supernatants were carried out as described previously (Logan and Tabin,1998). Two full-length Tbx3 viruses were produced; one includes amino acid residues 1-732 the other amino acid 15-732 of the predicted protein (AF033669). Both forms produced identical results. Tbx3^EN contains amino acids 15-289 of Tbx3, which spans the N terminus and DNA-binding T-domain, fused to the engrailed repressor domain (Jaynes and O'Farrell,1991). Tbx3^VP16 contains the same residues fused to two VP16 activation domains(Ohashi et al., 1994). The Gli3ZnF-Vp16 construct contains amino acids 471-636 of the human GLI3(XP_004833) fused to two VP16 activation domains. The prospective forelimb territory on the right side of the embryo were infected between stages 8 and 10, as previously described (Logan and Tabin, 1998). The left limb served as a contralateral control. Each virus produced a limb shift phenotype in ∼30% of infected embryos. For embryos analyzed before a limb shift phenotype was morphologically obvious, batches of infected embryos were analyzed. For embryos analyzed at later stages, embryos with a phenotype were selected.

Whole mount in situ hybridizations were carried out essentially as described (Riddle et al.,1993). Probes used were Shh(Riddle et al., 1993), Fgf8 (Vogel et al.,1996), MyoD, Pax3(Pourquie et al., 1996), Hoxb8, Hoxb9, Hoxc5, Hoxd12 (Burke et al., 1995), Wnt26(Kawakami et al., 2001), Tbx5, Tbx2, Tbx3 (Logan et al.,1998), dHand (also known as Hand2)(Fernandez-Teran et al.,2000), Gli3(Schweitzer et al., 2000) and Bmp2 (Schlange et al.,2002). The Tbx15 probe was produced from a cDNA isolated from a limb cDNA library (M.P.O.L., unpublished).

Whole-mount immunohistochemistry was performed as previously described(Kardon, 1998). Axons were stained using the 3A10 monoclonal antibody (hybridoma supernatant diluted 1/100 from DSHB, Iowa, USA) and detected with a peroxidase-conjugated anti-mouse secondary (Jackson ImmunoResearch) diluted 1/250.

Luciferase assays were performed using COS1 cells. Transfections were performed using Superfect transfection reagent (Qiagen) following the manufacturer's protocol. For expression studies, full-length, Engrailed and VP16 fusion forms were cloned into pcDNA3.1(–) (Invitrogen). The reporter pGL3-promoter vector (Promega) containing a single Brachyurybinding site (Kispert et al.,1995) together with a basal SV40 promoter upstream of the Firefly Luciferase gene. Luciferase assays were carried out using the appropriate Reporter Assay System (Promega) according to the manufacturer's protocol. Normalization of the results was carried out usingβ-Galactosidase Reporter Assay (Promega) according to the manufacturer's protocol. As an internal control the reporter plasmid was co-transfected with the β-galactosidase reporter only. All experiments were performed in triplicate. Error bars represent the standard deviation over three experiments.

DiI crystals (Sigma) were diluted in 100% ethanol (5 mg/ml). A 10% working solution was prepared in 30% sucrose/PBS solution. Misexpression of retrovirus was performed at HH stage 8-10. Twenty-four hours after retrovirus infection(stage 14) DiI solution was injected into the embryos at several levels in the limb-forming region of the LPM and in the adjacent somites, to serve as axial reference. Equivalent DiI injections were performed in the injected and control side of the embryo.

To analyse the expression pattern of Tbx3 at stages prior to limb outgrowth, we performed whole-mount in situ hybridization between stages 12-16. Tbx3 expression is observed in the future forelimb and hindlimb domains at stages 12-13 (Fig. 1A), earlier than previously reported(Tumpel et al., 2002). Just prior to overt limb outgrowth (stage 16), Tbx3 is expressed in the presumptive forelimb and hindlimb areas(Fig. 1B) and expression appears more robust in the posterior region(Fig. 1B). In situ hybridization on sections of the forelimb region of stage 16 embryos(Fig. 1C,D) reveals that Tbx3 is present in the mesoderm of the presumptive wing region and in the lateral lip of dermomyotome. At stage 19, when a limb bud is clearly evident, expression is prominent in the posterior of the limb(Fig. 1E). At later stages(stage 25), expression is located in two stripes in the anterior and posterior limb mesenchyme (Fig. 1F).

In vitro data suggest that human TBX3 is a transcriptional repressor (Carlson et al.,2001). We generated three constructs: full-length Tbx3 (Tbx3); a construct that contains the N-terminus and T-domain of Tbx3 fused to the VP16 transcriptional activator domain (Ohashi et al., 1994) (Tbx3^VP16); and a construct with the same residues fused directly to the repressor domain of the Drosophila engrailed gene (Jaynes and O'Farrell,1991) (Tbx3^EN) (Fig. 1G, see Materials and methods). The Tbx3^VP16 construct is predicted to compete with endogenous Tbx3 protein for binding sites and activate target genes. The Tbx3 and Tbx3^EN forms are predicted to repress target genes. Engrailed repressor activity is Groucho dependent(Jimenez et al., 1997) and Groucho homologues are present in the limb mesenchyme(Rallis et al., 2003). To test the transcriptional properties of these three Tbx3 constructs, we performed in vitro luciferase assays using a reporter plasmid containing a single T-box binding site. Both Tbx3 and Tbx3^EN fusion constructs repress expression to an almost equal extent. By contrast, the Tbx3^VP16fusion construct activates expression (Fig. 1H). Co-transfection of Tbx3/Tbx3^VP16 generates luciferase activity at intermediate levels between those achieved with Tbx3 or Tbx3^VP16 alone. Together, the data demonstrate that Tbx3 functions as a transcriptional repressor in vitro.

We have investigated the role of Tbx3 in normal limb development using the chicken retroviral system. Following our targeting strategy(Materials and methods), we could detect virus broadly in the limb-forming region from stage 14 onwards (data not shown). Strikingly, misexpression of full-length Tbx3 shifted the axial position of the injected limb rostrally (embryos that show rostral limb shift phenotype: n=142;phenotype frequency 25-30%). Identical results were produced with Tbx3^EN (embryos with phenotype; n=82, phenotype frequency 25-30%), further suggesting that Tbx3 normally functions as a transcriptional repressor. Comparison with MyoD, which is expressed in the dermomyotomal compartment of each somite and serves as an axial reference, and Shh, which marks the zone of polarizing activity (ZPA)in the posterior limb, demonstrates the rostral shift in limb position following Tbx3 misexpression (limb on right), relative to contralateral control limbs (on left in all cases shown)(Fig. 2A, n=6/6,100%). The shift in axial position can extend over the distance of one to three somites; however, the limb itself is otherwise morphologically normal. Pax3 is expressed in the dermomyotome of the developing somites and the migrating myoblasts (Williams and Ordahl, 1994). Following misexpression of Tbx3 and mislocation of the limb to a more rostral position along the embryo axis,myoblasts that migrate into the shifted limb are derived from somites at a more rostral level. Myoblasts at more caudal levels that normally migrate into the limb (Fig. 2B, black arrow), no longer contribute to the limb musculature and remain within the dermomyotome (Fig. 2B, red arrow) (n=7/7, 100%). Both delamination and migration of the myoblasts into the limb depend on the Met receptor and its ligand HGF, also called scatter factor, produced by non-somitic mesoderm(Dietrich et al., 1999). Following shift of the limb from its normal position, the source of HGF is presumably also shifted, leading to migration of myoblasts from somites at the incorrect axial level.

The neurons that innervate the wing and form the brachial plexus originate from the 16th to 19th spinal ganglia(Lillie, 1927). Transplantation of a limb to the interlimb region results in the migration of neuron axons from the spine into the ectopic limb from axial levels that normally do not contribute to limb innervation(Hamburger, 1939). Consistent with transplantation studies the shifted limb is innervated by neurons from the 15th to 18th spinal ganglia (Fig. 2C, n=6/6). Therefore, following rostral shift of the limb, neuron axons that normally would not participate in innervation of the wing, are recruited into the limb (Fig. 2C, red arrows) and axons that would normally enter the limb mesenchyme no longer do so (Fig. 2C, black arrows). Therefore, following the mislocation of the limb to a more rostral position, several cell-types undergo changes in their developmental program and contribute to different regions of the body.

In a previous study, misexpression of Tbx3 and Tbx2 in the developing hindlimb, results in changes in anteroposterior patterning of digits, suggesting these genes have a role in specifying digit identity(Suzuki et al., 2004). To determine if patterning of the shifted limb is altered following our misexpression strategy, we analyzed the expression of genes normally regionally restricted within the limb. Fgf8 is expressed in the AER of the injected wing in a pattern indistinguishable from the control limb(Fig. 2D; n=5/5,100%). Bmp2, which is expressed in the AER and in the posterior of the limb as a response to Shh signalling(Francis et al., 1994), is expressed in an identical pattern in injected and uninjected limbs(Fig. 2E; n=5/5, 100%)in contrast to other reports (Suzuki et al., 2004). This is consistent with the normal distribution of Shh in Tbx3-injected limbs(Fig. 2A). Hoxc5,which is normally expressed in proximal regions of the limb mesenchyme(Burke et al., 1995), is unaffected in the injected limb (Fig. 2F; n=6/6, 100%). Expression of other T-box genes is also unaffected within the shifted limb despite the rostral mislocation: Tbx5 is expressed throughout the limb mesenchyme(Fig. 2G; n=7/7,100%); Tbx15 is expressed in medial regions of the limb(Fig. 2H; n=6/6,100%); and Tbx2 is expressed in anterior and posterior stripes, in a similar pattern to Tbx3 (Fig. 2I; n=6/6, 100% compare with Fig. 1H).

We also analyzed skeletal preparations of embryos with a shift phenotype that were allowed to develop to later stages (stage 27). In these examples,the vertebral column of the embryo is normal. However, the limb skeletal elements, including the scapula are mislocated rostrally, although there is no alteration in their morphology (Fig. 2J, n=10/10, 100%). Moreover, the morphology of the digits in the shifted wing (Fig. 2K) is normal compared with those in the contralateral control limb (Fig. 2L). No alterations in digit identity were observed. The same results were obtained following misexpression of Tbx3 in the developing hindlimbs(Fig. 2M; n=10/10,100%). The axial position of the injected limb is shifted rostrally along the rostrocaudal axis, but hindlimb digit morphology(Fig. 2N) is indistinguishable from that in the contralateral control leg(Fig. 2O). In Tbx3-injected embryos with a shifted wing (n=8) or leg(n=7) analyzed at stage 27, Sox9 expression, which marks the skeletal progenitors, is normal in both forelimbs and hindlimbs, confirming the results obtained from skeletal preparations (data not shown).

To understand the mechanism underlying the rostral shift in limb position,we investigated the effects of Tbx3 misexpression at stages prior to morphological limb shifts (stage 13-16). Following our retroviral targeting strategy, by stage 13, transcripts for Tbx3 are detected at normal levels on the injected side (right) (Fig. 3A; 15/15, 100%). By stage 16, transcripts are present throughout the limb-forming territory on the injected side, while on the uninjected side they are restricted to the posterior (Fig. 3B; 2/12, 17%). Although there is no direct evidence that axial Hox gene expression controls the position of the limb primordia, the axial position at which the limbs develop correlates with the expression of several Hox genes in the LPM (Burke,2000; Burke et al.,1995; Cohn et al.,1995; Cohn et al.,1997; Rancourt et al.,1995). Axial Hox gene expression is not altered in the Tbx3-injected forelimb area. The rostral expression boundary of Hoxb8 is not changed (Fig. 3C; n=30/30, 100%). In addition, the expression boundary of Hoxb9, which is caudal to the region of the LPM that will give rise to the posterior forelimb mesenchyme(Burke et al., 1995; Cohn et al., 1997), is at the same level on the rostrocaudal axis of the embryo in both the control and injected side (Fig. 3D; n=27/27, 100%). Further analysis of the expression of several other Hox genes (Hoxb4, Hoxc4, Hoxb5, Hoxc6, Hoxa9, Hoxc9 and Hoxd9) produced identical results (data not shown). These results indicate that the mechanism that shifts limb position in Tbx3-injected embryos lies downstream of any axial Hox code that may act to position the limbs.

Experiments in the mouse, chick and zebrafish have established that Tbx5 is required for forelimb initiation(Agarwal et al., 2003; Ahn et al., 2002; Ng et al., 2002; Rallis et al., 2003; Takeuchi et al., 2003). The secreted factor Wnt2b has also been implicated in limb initiation(Kawakami et al., 2001). Experiments in zebrafish have suggested that wnt2b acts upstream of tbx5 during forelimb initiation(Ng et al., 2002). Following misexpression of repressor forms of Tbx3, the domain of Wnt2bexpression is unaffected (Fig. 3E; n=27/27, 100%); however, the domain of Tbx5expression is expanded to more rostral regions of the LPM, whereas the more caudal expression domain, which would normally give rise to the posterior limb mesenchyme, is initially unaffected (Fig. 3F; n=6/19, 32%). These results would suggest that the effects of Tbx3 are mediated downstream or in parallel with Wnt2b. Expression of Hoxc5, which is also expressed in the early limb bud mesenchyme, is also expanded rostrally(Fig. 3G; n=7/20,35%). These data demonstrate that the rostral shift in the limb is preceded by a rostral expansion of the limb mesenchyme, while their caudal domain is unchanged. At later stages, however, the expression domains of Hoxc5and Tbx5 within the limbs are shifted rostrally(Fig. 2F,G). At this stage, Shh expression is not expanded but is present in a more rostral domain (Fig. 2A). In conclusion, after misexpression of Tbx3, there is an initial rostral expansion of the limb-forming region that is followed by a subsequent shift in axial limb position.

To investigate whether cell movement accounts for the phenotype, we performed DiI labelling experiments to follow the fates of cells of the prospective limb-forming region. Following misexpression of Tbx3,cells that normally become part of the posterior limb mesenchyme are instead incorporated into the interlimb flank (Fig. 3H; n=5/5, 100%). Cells in more rostral locations that would not normally contribute to the limb, are recruited to form (anterior)limb. Therefore, the shift in limb position cannot be attributed to migration of cells in the LPM.

A shift in limb position is demonstrated by the expression of Shh,in the posterior of the limb bud, at an inappropriate axial level(Fig. 2A). We therefore examined the expression of dHand and Gli3, genes which are involved in pre-patterning the anteroposterior axis of the limb and establishing the position of Shh expression in the ZPA(Charite et al., 2000; Fernandez-Teran et al., 2000; te Welscher et al., 2002). During limb induction stages, dHand (also known as Hand2), a bHLH transcription factor, is expressed throughout the limb-forming region(Fig. 4A)(Charite et al., 2000; Fernandez-Teran et al., 2000). Subsequently, Gli3, a zinc-finger transcription factor, is expressed throughout almost the entire limb mesenchyme in an anterior-to-posterior graded fashion (Schweitzer et al.,2000). Genetic antagonism between Gli3 and dHandresults in downregulation of dHand expression in the anterior limb mesenchyme (Fig. 4B). At later stages, dHand and Gli3 are expressed in the anterior and posterior limb mesenchyme, respectively, with an overlapping domain of co-expression in the medial limb (Fig. 4C). The interactions between dHand and Gli3ultimately position the ZPA prior to Shh signalling(te Welscher et al., 2002; Zuniga and Zeller, 1999). Following misexpression of Tbx3, dHand is expressed throughout the limb-forming region, while in the control side dHand is restricted to the posterior limb mesenchyme (Fig. 4D; n=7/22, 32%). In addition, there is a downregulation of Gli3 throughout the injected limb and the caudal border of its graded expression domain shifts rostrally(Fig. 4G; n=8/22,36%). At slightly later stages (stage 19), when the shift phenotype is already apparent, the domain of dHand expression is shifted to a more rostral location and is no longer expressed throughout the limb mesenchyme but is restricted to the posterior (Fig. 4E; n=7/7, 100%). Similarly, at stage 19, Gli3is expressed in more rostral locations, in an anterior-posterior gradient in the limb mesenchyme of the shifted limb(Fig. 4H; n=8/8,100%). At later stages (stage 21) when the shift phenotype is obvious, dHand is expressed normally in the posterior mesenchyme of the shifted limb (Fig. 4F; n=7/7, 100%) and Gli3 has a normal distribution in the anterior mesenchyme (Fig. 4I; n=8/8, 100%). To more accurately define the time course of the limb shift phenotype, we also analyzed expression of Shh and a downstream target of Shh, Hoxd12, in Tbx3-injected limbs at stage 19. Expression domains of both Shh(Fig. 4J, n=2/2, 100%)and Hoxd12 (Fig. 4L; n=2/2, 100%) were shifted to more rostral positions. At later stages(stage 21), Shh (Fig. 4K, n=2/2, 100%) and Hoxd12(Fig. 4M; n=2/2, 100%)were expressed in the normal, posterior domains within the shifted limb. These data show that following misexpression of Tbx3, the normal restriction of dHand expression to the posterior limb mesenchyme and Gli3 to anterior is initially disrupted at early limb-forming stages prior to Shh expression. However, as soon as Shh is detectable, its expression domain is shifted to more rostral locations. The establishment of this altered expression domain serves as the first molecular evidence of a shift in the rostrocaudal location of the limb. After limb position has shifted, dHand and Gli3 expression is normal within the limb mesenchyme.

As a complementary approach to misexpressing Tbx3 and Tbx3^EN, we misexpressed Tbx3^VP16 in the presumptive forelimb area. In contrast to Tbx3 and Tbx3^EN, which act as repressors, Tbx3^VP16 behaves as a transcriptional activator(Fig. 1F). Moreover,Tbx3^VP16 has the ability to interfere with the repressor activity of Tbx3 in our in vitro luciferase system(Fig. 1F). We predict that Tbx3^VP16 is able to compete with endogenous Tbx3 for binding sites,and targets will be activated instead of repressed. Following misexpression of Tbx3^VP16, the injected limb is displaced caudally in axial position(embryos that show caudal limb shift phenotype n=42; frequency of phenotype 30%). In situ hybridization for MyoD and Shhdemonstrates the caudal shift in axial position(Fig. 5A, n=7/7,100%). The limb displacement can extend over the distance of one to three somites but, as with rostral limb shifts, the limb itself is otherwise normal. We examined the expression of dHand and Gli3 at pre-limb bud stages, following misexpression of Tbx3^VP16. Although in the control, dHand expression is restricted to posterior mesenchyme of the limb, following misexpression of Tbx3^VP16, there is a caudal displacement of the expression domain of dHand(Fig. 5B, n=9/30,33%). In addition, there is an upregulation of Gli3 expression(Fig. 5C, n=8/25,32%). These results show that by misexpressing Tbx3^VP16,we are able to generate a limb shift in the opposite direction to that obtained using Tbx3 and Tbx3^EN. The data also suggest that Tbx3 acts as a transcriptional repressor in the limb bud.

Following misexpression of Tbx3, Gli3 mRNA levels are decreased within the limb mesenchyme and the caudal border of its expression is shifted rostrally (Fig. 4G). Gli3 is a bifunctional zinc-finger transcription factor that undergoes default proteolysis to a truncated form (Gli3^R) that represses expression of Shh target genes. In the presence of Shh signalling, processing to produce the Gli3^R form is blocked and a full-length Gli3 protein is formed. Contradictory data, generated both in vitro and in vivo, suggest that the full-length Gli3 molecule is either an activator or a repressor(Bai et al., 2004; Litingtung et al., 2002; Sasaki et al., 1997; Wang et al., 2000). The effect of Tbx3 misexpression on Gli3 is observed at stages prior to Shh expression when Gli3 is acting as a repressor. To investigate whether Gli3 can directly alter limb positioning, we misexpressed a form of Gli3 that contains the DNA-binding domain of the protein fused with two VP16 activation domains (Gli3ZnF-VP16, Fig. 6A). Gli3ZnF-VP16 has the ability to bind Gli3 binding sites and activate transcription (D. Stamataki,F. Ulloa and J. Briscoe, personal communication). Misexpression of this form of Gli3 is predicted to compete with the endogenous Gli3 repressor(Gli^R) for binding sites and to activate, rather than repress,target genes and thereby lower the repressor activity of endogenous Gli3. Misexpression of Gli3ZnF-VP16 generated a phenotypically similar result to that obtained following misexpression of Tbx3 repressor forms (embryos that show the rostral limb shift phenotype n=48; phenotype frequency 35-40%). A rostral shift in axial limb position is evident by comparing MyoD and Shh expression domains(Fig. 6B, n=7/7, 100%)and can extend from one to three somites. Following misexpression of Gli3ZnF-VP16, the Tbx3 expression domain is expanded rostrally in the injected side compared with the contralateral control side(Fig. 6C, n=6/16,38%). A similar expansion is observed in dHand expression(Fig. 6D, n=5/14,36%). By contrast, the endogenous expression of Gli3 is decreased following misexpression of Gli3ZnF-VP16(Fig. 6E, n=5/13,38%). These results suggest that Gli3 is a candidate to play a role in positioning the limb field and may do so through a genetic interaction between dHand and Tbx3.

Our results, and those of others, demonstrate that Tbx3 can act as a transcriptional repressor in vitro (Fig. 1) (Carlson et al.,2001). Consistent with these observations, misexpression of Tbx3 and Tbx3^EN in the forelimb region produces identical phenotypes, while misexpression of Tbx3^VP16generates the opposite phenotype. Data from in vitro and in vivo experiments suggest that Tbx3 acts as a transcriptional repressor in the developing limb bud.

Following misexpression of Tbx3, the expression domains of genes expressed in the limb are initially expanded rostrally and this is followed by a shift in limb position. Cells of the flank, rostral to the limb, which would not normally contribute to the limb, now become incorporated into the limb. Furthermore, cells that normally form posterior limb mesenchyme now no longer contribute to the limb. Strikingly, these cells had presumably initially expressed Tbx5, a gene required and apparently sufficient for limb initiation (Agarwal et al.,2003; Ahn et al.,2002; Ng et al.,2002; Rallis et al.,2003; Takeuchi et al.,2003). Fate mapping shows that the altered contribution to the limbs is not simply explained by migration of limb bud precursors. Our data suggest that once a ZPA is established in a more rostral position, the positive feedback loop between ZPA and AER now operates at altered axial levels and the limb develops in an ectopic site. Noticeably, the size of the ectopic limb is the same as that of the normal limb, suggesting that a mechanism, yet to be determined, is functioning to regulate limb size.

Other members of the T-box gene family can influence the extent of the limb. Tbx18 is expressed in lateral plate mesoderm as the limb buds form and the anterior limit of Tbx18 expression coincides with the anterior border of the limbs. Following misexpression of Tbx18 in the presumptive wing bud region, the anterior extent of the limb is expanded(Tanaka and Tickle, 2004). The wing bud is extended rather than shifted in position and this extension is only transient. At later stages, the limb appears to regulate its size and develops normally. Brachyury is expressed in the LPM at the onset of limb formation and at later stages of limb development, in the distal limb mesenchyme that lies underneath the AER. Misexpression of Brachyuryin the chick wing results in anterior expansion of the AER and produces limb phenotypes consistent with augmented AER extent and function, including anterior digit duplications and, in rare cases, thickening of the anterior-most metatarsal (Liu et al.,2003).

Tbx3 misexpression does not affect the pattern of axial Hox gene expression. Hox genes have been implicated in limb position(Cohn et al., 1997) and a role for Hox genes in limb positioning is supported by Hoxb5 knockout mice, which exhibit a unilateral or bilateral rostral shift in axial forelimb position (Rancourt et al.,1995). However, the phenotype in Hoxb5^–/– mice differs in several aspects from that obtained with Tbx3 or Tbx3^EN misexpression. In Hoxb5^–/– mice, homeotic transformations of the cervico-thoracic vertebrae from C6-T1 are observed. The clavicle retains a medial articulation with its normal target, the sternum, resulting in a V-shaped shoulder girdle. The alteration in limb position in the Hoxb5^–/– mice is therefore associated with a transformation of the entire axial skeleton. The absence of an effect of Tbx3 on the expression of Hox genes indicates that Tbx3mediates its effects on limb position independently of any axial Hox code.

Roles for Tbx3 and the related gene Tbx2 in specifying posterior digit identity via Shh and Bmp signalling has been suggested (Suzuki et al.,2004; Tumpel et al.,2002). Our data implicate Tbx3 in positioning the nascent limb at pre-bud stages, prior to Shh expression. However, we also analyzed the shifted limb at later stages. Following misexpression of full-length Tbx3 and Tbx3^EN, the expression pattern of Bmp2 is unaffected, not expanded, as seen by Suzuki et al., and ultimately digit identity is unchanged. We also performed injections of Tbx3 in the hindlimb-forming region and obtained the same result as seen in forelimbs: the injected hindlimb is shifted rostrally, while the digit array is unaffected, in contrast to previous results. Our results do not support a role for Tbx3 in specifying digit identity. One difference between our experiments and those of Suzuki et al. is the timing of the viral injections: while we performed our injections at stage 8-10, Suzuki et al. injected at stage 11-12. Our earlier stage misexpression protocol may account for our ability to generate limb shifts that were not reported by Suzuki et al. However, it is not clear why different effects on limb patterning were observed in these two sets of experiments, as our injection strategy does lead to broad ectopic expression of Tbx3 at limb bud stages(Fig. 3B).

At limb development stages, prior to Shh expression, genetic antagonism between dHand and Gli3 establishes an anteroposterior pre-pattern of the limb that is ultimately responsible for establishing the position of the ZPA in the posterior limb. At these stages, Gli3 is acting as a repressor(Wang et al., 2000), while dHand is shown to be a transcriptional activator(Dai and Cserjesi, 2002; Dai et al., 2002). Misexpression of Tbx3 leads to an expansion, or failure of repression, of dHand, potentially through an interaction between Tbx3 and Gli3 (Fig. 7). We predict that Tbx3 is acting to repress Gli3 expression in the future posterior limb mesenchyme. Repression by Tbx3 could be responsible for generating the anterior-to-posterior graded expression of Gli3 in the developing limb primordium. This model is consistent with the observation that in Tbx3 mutant mice, dHand is downregulated in the forelimbs and absent in the hindlimbs,which subsequently leads to a disruption in Shh expression. Our model would predict that downregulation of dHand is due to high Gli3 expression that, in the absence of Tbx3, is no longer restricted to the anterior limb and expands to the posterior.

Misexpression of Tbx3^VP16 produces the opposite phenotype to that obtained with Tbx3 and Tbx3^EN;the limb shifts to a more caudal position than normal. Gli3expression is upregulated and expanded caudally in the Tbx3^VP16-injected limbs, lending further support to a model in which Tbx3 normally functions to restrict Gli3 to the anterior mesenchyme. As a consequence of this Gli3 expansion, the dHand expression domain is shifted caudally. By contrast, in Tbx3- and Tbx3^EN-injected embryos, disruption of the normal repression of dHand by Gli3 resets the rostrocaudal position of the ZPA to a more rostral position and this ultimately results in the limb shift phenotype. In Tbx3^VP16-injected embryos, de-repression of Gli3shifts the expression domain of dHand and this ultimately leads to the ZPA forming more caudally. Our data demonstrate that the signals required for pre-patterning the AP axis of the limb and setting the position of the ZPA can influence the position of the limb along the rostrocaudal axis of the embryo.

We predict that the effects of Gli3ZnF-VP16 on dHand and Tbx3 are caused by a disruption of endogenous Gli^Ractivity. Misexpression of Gli3ZnF-VP16 leads to the expansion of Tbx3 and dHand expression domains, rather than the induction of ectopic patches of expression. This may indicate a non-cell autonomous action of Gli3 on dHand and presumably also on Tbx3. A regulatory relationship between Gli3 and Tbx3 has been demonstrated during lung organogenesis(Li et al., 2004). Tbx3 is normally expressed in the mouse lung in the presence of Shh. In this environment, Gli3 acts as an activator of Tbx3 transcription. In Shh^–/–animals, where Gli3 acts as a repressor, Tbx3 transcripts are significantly reduced. However, in Shh^–/–/Gli3^–/– animals,de-repression of Tbx3 is observed and Tbx3 expression is, at least partially, restored (Li et al.,2004). These results, in combination with our own, suggest that regulatory relationships between Tbx3 and Gli3 may exist broadly during embryogenesis.

Our data implicate Gli3 in a genetic network that can influence limb position, yet mice mutant for Gli3 (Extra toes,Xt^J), are not reported to exhibit any shift in axial limb position (Buscher et al., 1997; Buscher et al., 1998; Hui and Joyner, 1993; Litingtung et al., 2002). In Xt^J/Xt^J mice (that lack all Gli3 activity), dHand is expressed throughout the limb and subsequently Shhis expressed ectopically in the anterior limb mesenchyme. Misexpression of dHand alone, at limb bud stages, is capable of inducing Shhexpression, but only in the anterior limb mesenchyme rather than medial locations (Charite et al.,2000; Fernandez-Teran et al.,2000). These results suggest that anterior and posterior limb mesenchyme may express a `licensing factor' required together with dHand for the induction of Shh and that this factor is absent from the medial limb mesenchyme. Tbx3 is expressed in two stripes in the anterior and posterior limb mesenchyme at later stages of limb development (Fig. 1D) and may normally act as such a factor.

A requirement for Tbx3 to establish or re-set the domain of Shh-expressing cells in the ZPA may also explain why no limb shift phenotype is observed in Xt^J/Xt^J mice. Tbx3 expression is not expanded in Xt^J/Xt^J mice(Tumpel et al., 2002). Without the presence of both Tbx3 and dHand, the position of the limb is not altered. In Tbx3 misexpression experiments, however, Tbx3 can repress Gli3 and this leads to an expansion of dHand. The co-expression of Tbx3 and dHand can, in turn, `license' ectopic Shh expression, which, in turn, alters the position of the limb (Fig. 7).

Mice in which Tbx3 is inactivated and humans with UMS who are haploinsufficient for TBX3 are not reported to exhibit any shift in axial limb position. This is consistent with the model we propose for the role of Tbx3 in the early limb bud. Although a limb shift phenotype is not observed in Tbx3^–/– mice, the expression domains of dHand and Shh are downregulated or even eliminated (Davenport et al.,2003), consistent with Tbx3 being required for their normal expression. This finding supports our conclusions that, although not strictly required to fix limb position, Tbx3 is an important component of the signals establishing the position of the domain of Shh-expressing cells that comprise the ZPA in the posterior limb.

We thank D. Stamataki, F. Ulloa and J. Briscoe for providing the Gli3ZnF-VP16 construct. The 3A10 monoclonal antibody was obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biological Sciences,Iowa City, IA 52242. We thank Kate Storey for providing the DiI labelling protocol. We thank members of our laboratory and T. Heanue for their comments on the manuscript. M.P.O.L. is an EMBO young investigator.