Involvement of T-box genes Tbx2-Tbx5 in vertebrate limb specification and development

One of the defining features of jawed vertebrates (gnathostomes) is the development of two sets of paired appendages at specific levels along the primary body axis: the pectoral and pelvic fins of fishes, and their derived homologs, the forelimbs and hindlimbs of tetrapods (Carroll, 1988). Classical embryological studies demonstrate that limb identity and early anteroposterior limb polarity are first established in flank (lateral plate) mesoderm (Hamburger, 1938; Zwilling, 1955). Signals emanating from the mesoderm induce the formation of an apical ectodermal ridge (AER) in surface ectoderm at specific axial levels for limb formation. The AER in turn provides signals that maintain high proliferation rates in the adjacent mesoderm, forming a progress zone that drives limb outgrowth along the proximodistal axis (Saunders, 1948; Summerbell et al., 1973). Signals emanating from the AER induce the formation of an organizer, the zone of polarizing activity (ZPA), at the posterior margin of the bud (Saunders and Gasseling, 1968; Rowe and Fallon, 1981). The ZPA then takes over the role of providing positional information along the anteroposterior axis, and also provides signals required to maintain the AER (Saunders and Gasseling, 1968; Tickle et al., 1975). Dorsoventral axis specification is sequentially controlled by signals from the somites, somatopleure and non-ridge ectoderm (MacCabe et al., 1974; Pautou, 1977; Michaud et al., 1997).

A great deal is now known about the molecular genetic pathways involved in the outgrowth and patterning of developing appendages (reviewed by Tickle, 1995, 1996; Niswander, 1997; Shubin et al., 1997). However, far less is known about the genes responsible for initiating these pathways, how positional information is provided, so that appendages do not develop in inappropriate places, or how morphological differences between the forelimb and hindlimb are specified. Hox genes are clearly key regulators involved in determining the axial levels at which limbs and fins develop (Charité et al., 1994; Rancourt et al., 1995; Cohn et al., 1997) and Fibroblast Growth Factor-8 (FGF8) is known to be an important effector mediating the early inductive interactions between surface ectoderm and lateral plate mesoderm required to initiate limb outgrowth (Mahmood et al., 1995; Crossley et al., 1996; Vogel et al., 1996). However, Hox genes are transcription factors expressed in lateral plate mesoderm, and FGF8 is a secreted protein expressed in surface ectoderm only after the limb territories have been specified. It is therefore apparent that a signal transduction mechanism must exist to interpret the relevant Hox code and signal the adjacent ectoderm to express FGF8. FGF10 is a good candidate for a mesoderm-secreted factor that could signal the ectoderm to express FGF8 (Ohuchi et al., 1997), however transcription factors responsible for activating Fgf-10 expression have not been identified.

We have recently identified a family of putative transcription factors, the T-box genes, shown to regulate a variety of inductive interactions during embryogenesis (reviewed by Papaioannou and Silver, 1998). Four of these genes, Tbx2, Tbx3, Tbx4 and Tbx5, are represented in the mouse genome as two cognate, linked gene pairs. These have evolved from a single ancestral locus by a two-step process in which an initial tandem-duplication event was followed by duplication of the derived gene pair (Agulnik et al., 1996; Ruvinsky and Silver, 1997), possibly during one of the complete genome duplications thought to have occurred near the root of the vertebrate lineage (Holland et al., 1994). We have found that all of these genes are expressed in developing mouse limbs, either at the time of limb field specification, during bud outgrowth, or both (Chapman et al., 1996; Gibson-Brown et al., 1996). Interestingly, Tbx2 and Tbx3 are expressed in similar spatiotemporal patterns in both limbs, whereas Tbx4 and Tbx5 expression is primarily restricted to the developing hindlimb and forelimb buds, respectively. Since Tbx4 and Tbx5 are expressed in lateral plate mesoderm at the time of limb field specification, well before bud formation, we have proposed that the limb-specific expression of these genes may play a role in the specification of limb identity (fore versus hind) and limb axis determination (Gibson-Brown et al., 1996). In this investigation, we examine the role of all four of these T-box genes in chick limb development, as the chick model system makes it possible to perform many different embryological manipulations.

It has recently been reported that provision of an exogenous source of FGF to medial somatopleure can induce the formation of additional, ectopic limb buds in the flank of chick embryos (Cohn et al., 1995, 1997; Ohuchi et al., 1995, 1997; Crossley et al., 1996; Vogel et al., 1996), and that the identity of the limb that develops (wing, leg, or mosaic limb) is dependent on the specific axial level at which the FGF source is provided: FGF-soaked beads placed adjacent to somites 21-22 generally induce ectopic wings, beads next to somite 25 induce ectopic legs, and those placed in between induce either wings, legs or mosaic limbs, at approximately equal frequency (Cohn et al., 1995, 1997). We have cloned the chick orthologs of Tbx4 and Tbx5 to take advantage of this novel experimental system, and to investigate whether the identity of the ectopic limb that develops is related to the identity of the limb-specific T-box gene, or genes, being expressed. Since it has been shown previously, using heterotopic grafts of limb mesenchyme, that limb identity is a stable property of the mesenchyme (Cairns and Saunders, 1954; Saunders et al., 1957, 1959), we also examined Tbx4 and Tbx5 expression after grafting blocks of isolated leg mesenchyme beneath the AER of wing buds and blocks of wing mesenchyme beneath the AER of leg buds. In addition, we also obtained the chick orthologs of Tbx2 and Tbx3, to investigate the regulation of these genes during limb development.

Three further experimental manipulations were employed in these studies. The AER was surgically removed from either wing or leg buds to investigate whether maintenance of T-box gene expression is dependent on the continued provision of AER/ZPA-derived signals. Sonic hedgehog (SHH)-expressing cells were grafted under the anterior AER, thereby generating an ectopic ZPA (Riddle et al., 1993; Chang et al., 1994), to investigate whether Tbx2 expression in the posterior part of the wing autopod is dependent on, and therefore downstream of, SHH expression. Finally, SHH-expressing cells were grafted to the anterior margin of wing buds from which the AER had been removed, to investigate whether Tbx2 expression in the autopod is directly dependent on SHH, or indirectly regulated through a dialogue between the ZPA and the AER.

A stage 20-24 chick limb cDNA library (provided by Dr Juan Carlos Izpisúa-Belmonte), a stage 17-18 chick leg library and a stage 18-24 chick limb library (both provided by Dr Susan Mackem; Ranson et al., 1995), were screened with 32P-labeled probes from the T-box portion of the mouse Tbx4 and Tbx5 genes (Agulnik et al., 1996). Low-stringency filter hybridization was performed at 50°C (Church and Gilbert, 1984). Filters were washed either at 65°C or 55°C in 2× SSC, 0.1% SDS, and positive clones recovered through secondary and tertiary screens. Following excision of the pBluescript phagemids from the λZAP vectors, positively hybridizing clones were sequenced by standard methods. These screens led to the identification of multiple cDNA clones containing portions of the chick Tbx3, Tbx4 and Tbx5 genes. A 197 bp PCR product from the T-box portion of chick Tbx2 was generously provided by Dr Juan Carlos Izpisúa-Belmonte. Sequence comparisons were performed by FASTA and PILEUP routines in the Wisconsin GCG sequence analysis package (Genetics Computer Group, 1989).

Whole-mount in situ hybridization was performed as described by Wilkinson (1992), except that both hybridization and post-hybridization washes were carried out at 62°C (Tbx2) or 70°C (Tbx3-Tbx5) in the following solution: 50% formamide, 1.3× SSC, 5mM EDTA, 50 μg/ml yeast RNA, 0.002% Tween-20, 0.005% CHAPS, 100 μg/ml heparin, 2% Blocking Reagent (Boehringer Mannheim). Sense and antisense RNA probes including all or part of the T-box coding region were transcribed from linearized plasmids using T3 or T7 RNA polymerase (Promega Biotech) in the presence of digoxigenin-UTP (Boehringer Mannheim) according to the protocols suggested by the manufacturers. The Tbx2 plasmid consisted of a 197 bp portion of the T-box inserted at the SmaI site of pCR-Script; Tbx3 and Tbx5 plasmids consisted of 1.4 kb and 1.3 kb of the respective cDNAs for the genes inserted at the BstXI site of pBluescript II KS (−); the Tbx4 plasmid consisted of a 1.0 kb portion of the gene inserted at the EcoRI site of pBluescript II KS (−).

Fertilized White Leghorn chicken eggs (SPAFAS, CT) were incubated at 38°C in a humidified atmosphere, with rocking. Embryos were operated on between stages 13 and 21 (Hamburger and Hamilton, 1951, 1992), the eggs resealed with duct tape and returned to the incubator.

Induction of ectopic limbs was achieved by modifying the technique described by Cohn et al. (1995). Embryos were operated on between stages 13 and 16. The vitelline membrane was torn open with the tip of an 18-gauge hypodermic needle to expose the embryo. A small transverse slit was made through the medial somatopleure with a sterile microscalpel at the level of somites 19-20. A single heparin-acrylic bead (H-5263, Sigma), 200 μm in diameter, which had been soaked in 1 mg/ml FGF-2 (133-FB-025, R&D Systems) in PBS for at least 1 hour at room temperature, was inserted through the slit into the coelom and pushed caudally to the desired axial level using the closed tips of a pair of fine forceps. Operated embryos were reincubated for a further 2 days before fixing for whole-mount in situ hybridization, or for a further 7 to 9 days before dissection for whole-mount skeletal preparation at stages 35-37.

Donor and host embryos were operated on at stages 19-20. Wing and leg buds were dissected from donor embryos, rinsed in PBS, and placed in 2% trypsin in PBS for 10 minutes at room temperature. After washing once in PBS, the buds were placed in complete chick embryo fibroblast culture medium (Yang and Niswander, 1995) on ice, and the surface ectoderm teased away using fine forceps and discarded. Donor mesenchyme from the distal tip was cut into small blocks (100-200μm in diameter, 1 to 3 blocks from each bud) and a single block heterotopically grafted under the AER of a recipient limb bud (wing mesenchyme from a donor being grafted to the leg bud of a host recipient, and vice versa) as described previously (Saunders et al., 1957, 1959). Recipient embryos were reincubated for a further 2 days before being dissected and fixed for whole-mount in situ hybridization.

The AER of stage 20-21 wing or leg buds was teased away from the underlying mesenchyme using fine tungsten needles and then removed with fine forceps. Operated embryos were reincubated for 24 to 48 hours before dissection and fixation for whole-mount in situ hybridization.

Pellets of chick embryo fibroblasts expressing the SHH protein were prepared as described previously (Yang and Niswander, 1995). The pellets were cut into small pieces (100-200 μm diameter) and either grafted under the anterior AER of stage 19-21 wing buds, or pinned with fine platinum wire to the anterior margin of wing buds from which the anterior AER had been surgically removed. Operated embryos were reincubated for a further 24 to 42 hours before dissection and fixation for whole-mount in situ hybridization, or for 7 or 8 days before dissection for whole-mount skeletal preparation.

To visualize skeletal patterns and determine the identity of ectopic limbs, embryos were dissected at stages 35-37, stained with Alcian blue and cleared as described by Tickle (1993). These preparations were photographed and then ectodermal structures were visualized by placing the embryos in absolute alcohol overnight.

Comparison of the T-box regions of the chicken genes, Tbx2, Tbx3, Tbx4 and Tbx5, with their corresponding mouse orthologs reveals a considerable degree of sequence conservation between the chicken and mouse genes. Orthologous genes are defined as direct decendants of a single ancestral gene that was present in the genome of the common ancestor of the two species under analysis. At the amino acid level the chicken and mouse Tbx2-Tbx5 orthologs exhibit 98.5%, 95.6%, 99.4% and 99.4% identity, respectively, within the T-box domain (Fig. 1A). PILEUP comparison of these sequences against all other T-box genes for which putative orthologs have been identified in both mouse and chick clearly demonstrates the close degree of relatedness between the newly cloned chick genes and their corresponding mouse orthologs (Fig. 1B).

We examined the expression patterns of all four of the newly cloned chick T-box genes during embryogenesis and limb development to compare these with the previously reported expression patterns of their mouse orthologs (Chapman et al., 1996; Gibson-Brown et al., 1996). Here we report expression in the limbs throughout limb specification and development. Expression in other parts of the embryo is reported elsewhere (Gibson-Brown et al., 1998).

Tbx2 expression was detectable in flank mesoderm at stage 14 adjacent to somites 6-20 (Fig. 2A), a region encompassing, and extending beyond, the prospective wing field. Expression was also observed in lateral plate mesoderm adjacent to the caudal end of the segmental plate, corresponding to the region fated to form the anterior part of the leg bud. By stage 15 the entire flank between somites 4-24 was positive, with strongest expression in the wing and leg fields. At stages 17 and 18 expression was decreased in the medial part of the wing and leg buds, respectively (Fig. 2B). This pattern continued through later stages, such that only the anterior and posterior margins of the limbs expressed Tbx2 (Fig. 2B-D). Beginning at stage 20-21, a marked asymmetry in expression became apparent in the anterior and posterior expression domains at the wing margins. In the posterior domain, expression extended from the point of limb articulation with the body wall to the distal tip of the bud whereas, in the anterior domain, expression did not extend so far distally (Fig. 2C,D). This asymmetry of expression was less apparent in the leg (Fig. 2D). Between stages 20 and 25, the AER of the wing, but not the leg, also expressed Tbx2 (Fig. 2C). None of the other T-box genes examined were expressed in the AER of either limb at any stage.

Strong expression of Tbx3 was detected in lateral plate mesoderm adjacent to somites 10-23 and the caudal end of the segmental plate at stage 14. At stage 15 expression in the lateral plate of the wing field had shifted caudally, now lying adjacent to somites 11-24 (Fig. 2E), and expression in the tail region lay adjacent to the caudal half of the segmental plate. By stage 19 limb expression was confined to mesenchyme in the posterior part of all limb buds (Fig. 2F), but by stage 21 was also detected in the anterior mesenchyme of all limb buds (Fig. 2G). Expression in the anterior and posterior limb margins persisted through stage 27, with the posterior domain extending further distally than the anterior domain (Fig. 2H). At stage 25 expression was also detected in the interdigital regions between wing digits 3 and 4, and between leg digits 2/3 and 3/4 (Fig. 2H).

Limb-associated Tbx4 expression was confined to the hindlimb at all stages (Fig. 2I-L). It was first observed at stage 14 in medial somatopleure adjacent to the caudal end of the segmental plate (Fig. 2I). By stage 15 the anterior boundary of this expression lay adjacent to the border between somites 24/25. At stage 16 a posterior boundary appeared adjacent to the segmental plate at around the level from which somite 32 is destined to bud off; leg buds form between somites 26-32.

By stage 19 expression was detected throughout the leg bud (Fig. 2J), persisting through stage 26 (Fig. 2K,L), with strong expression in the leg interdigits at stage 26.

In contrast to Tbx4, Tbx5 was expressed only in the forelimb (Fig. 2M-P). It was detected in lateral plate mesoderm adjacent to somites 10-23 in the prospective wing field at stage 14. By stage 15, this expression had shifted caudally to lie adjacent to somites 11-24 (Fig. 2M). Tbx5 was expressed throughout the wing bud from stage 17, after the bud has become morphologically defined, to stage 23, when expression becomes regionalized (Fig. 2N,O). Beginning at stage 23 expression was strongest at the anterior and posterior wing margins (Fig. 2O,P), and also transiently extended into the flank between the limb buds (Fig. 2O).

Ectopic limb buds were induced in the flank of 69 chick embryos by implanting FGF2 beads at different axial levels between somites 21 and 26. Half of these were probed for Tbx4 expression by whole-mount in situ hybridization, and half were probed for Tbx5. In rostally positioned buds, developing close to the normal wing, Tbx4 was expressed only at the posterior margin of the bud, whereas Tbx5 was expressed throughout all or most of the bud (Table 1; Fig. 3A,B). Buds developing in the middle of the flank expressed Tbx4 mainly in the posterior half of the bud, and Tbx5 in the anterior half (Table 1; Fig. 3E,F). In caudally positioned buds, developing close to the normal leg, Tbx4 was expressed throughout all or most of the bud, whereas Tbx5 was only expressed in the anterior part of the bud (Table 1; Fig. 3I,J). In buds developing at all axial levels, there was a region of overlapping gene expression, such that the posterior boundary of Tbx5 expression extended further caudally than the anterior boundary of Tbx4 expression. This was confirmed by two-probe hybridization to an additional set of embryos in which ectopic buds had been induced at various axial levels in the flank (n=7). In all of these embryos, expression was detected throughout the ectopic buds, with no evidence of a gap between the expression domains of the two genes (data not shown). Thus, depending on the rostrocaudal position at which the bud developed, the region of overlapping expression shifted along the primary body axis: in rostral buds, the region of overlap lay adjacent to somite 22, in mid-flank buds, adjacent to somites 23-24, and in caudal buds, adjacent to somite 25.

To determine how the patterns of gene expression in ectopically induced limb buds are related to the identity of the limb that subsequently develops (wing, leg or mosaic limb), ectopic limbs were examined at stage 35-37. By this stage, it is possible to distinguish between the fingers and toes of distally complete limbs; the number of phalanges in a digit is characteristic of the identity of that digit within a hand or a foot, and wing digits develop feather buds along their margins, whereas toes develop scales. Out of 51embryos which developed an ectopic limb, all but one developed an overtly mosaic limb (Table 2). The direction of limb mosaicism (wing-like versus leg-like) was directly related to the axial level of bead implantation, as reported previously (Cohn et al., 1995, 1997). Thus while anteriorly placed beads induced mosaic limbs, most of whose elements consisted of wing-like structures (Table 2; Fig. 3C,D), the more posterior the bead, the more leg-like the limb (Table 2; Fig. 3G,H,K,L). The single exception was an ectopic limb that was morphologically indistinguishable from a normal leg (4 toes, each containing the appropriate number of phalanges, and no feather buds). This leg developed in an embryo that had received a bead adjacent to somite 26, at the caudal end of the region shown previously to be capable of forming ectopic limbs (Cohn et al., 1995).

2 days after blocks of leg mesenchyme were grafted beneath the AERs of recipient wing buds (n=15), Tbx4 expression was clearly observed in a characteristic wedge-shaped inclusion in all of the host wings (Fig. 4A,B). Likewise, 2 days after blocks of wing mesenchyme were grafted beneath the AERs of recipient leg buds (n=12), Tbx5 expression was clearly observed in a similar inclusion in all of the host legs (Fig. 4C,D). The stability of this limb-specific T-box gene expression in the grafted tissue is in accordance with the stability of graft identity demonstrated previously (grafted wing mesenchyme autonomously differentiates into fingers, grafted leg mesenchyme autonomously differentiates into toes; Cairns and Saunders, 1954; Saunders et al., 1957, 1959).

1 or 2 days after wing or leg AER removal, Tbx2 expression was not reduced in either of the wing (n=6) or leg (n=7) margins (Fig. 5A,B). Likewise, Tbx3 expression was not reduced in these regions following wing (n=6) or leg (n=7) AER removal (Fig. 5C,D). Similarly, expression of Tbx4 was maintained all the way to the distal tip of the truncated leg bud (n=9, Fig. 5E), and Tbx5 expression was maintained to the distal tip of the truncated wing bud (n=9, Fig. 5F) at least 48 hours after AER removal. Thus maintenance of T-box gene expression in the developing limbs is not dependent on the continued provision of AER/ZPA-derived signals.

Tbx2 expression is not present in the anterior autopod but extends to the distal tip of the posterior wing margin and may therefore be regulated by signals emanating from the ZPA. We generated an ectopic ZPA, by grafting a SHH-expressing cell pellet under the wing anterior AER, to examine whether Tbx2 expression is regulated, directly or indirectly, by SHH. One (n=12) or 2 (n=11) days after grafting a SHH-expressing cell pellet, expression of Tbx2 could be detected in the mesenchyme of the anterior autopod close to the implant immediately underlying the AER (data not shown). In control embryos, which had also received a SHH-expressing implant under the anterior AER but were allowed to develop to stages 35-36, digit duplications ranged from partial (3-2-2-3-4) to full (4-3-2-3-4) in all cases (n=5; data not shown), indicating successful generation of an ectopic ZPA.

Since SHH-induced expression of Tbx2 was detected only in mesenchymal cells close to the implant, SHH-expressing implants were grafted to the anterior margin of wing buds from which the anterior AER had been removed to determine whether gene expression is directly induced by SHH, or indirectly induced via interactions between the ZPA and AER. No Tbx2 expression could be detected distal to the implant either 1 or 2 days after grafting a SHH-expressing cell pellet onto the anterior side of a wing bud from which the AER had been removed, (n=14). However, in all cases where the cell pellet could clearly be seen in the limb margin after incubation, cells in direct contact with the pellet did express Tbx2 (Fig. 6A). Wing buds that received control pellets of chick embryo fibroblasts that did not express the SHH protein (n=16) did not exhibit any signal (Fig. 6B). The results of these two experiments suggest that expression of Tbx2 in the posterior autopod lies downstream of SHH, and that Tbx2 may be a direct, short-range target of SHH.

Expression of the chick Tbx2-Tbx5 genes is generally very similar, both spatially and temporally, to the patterns previously reported for their mouse orthologs (Gibson-Brown et al., 1996, this report). However, mouse Tbx3 is expressed in the AER of both the forelimb and hindlimb buds, whereas, in the chick, only Tbx2 is expressed in the AER and, in this case, expression is transient and confined to the forelimb (wing) only. The functional significance of this observation is not yet clear, but it might prove instructive to examine expression of these genes in the fin folds of fish (the ectodermal structures homologous to the AER of tetrapods) to determine which of these patterns is ancestral and which is derived.

Tbx4 and Tbx5 are first expressed in lateral plate mesoderm within clearly defined territories at the time the prospective limb fields are being specified by Hox genes (Gibson-Brown et al., 1996, this study; Cohn et al., 1997). Hox genes may therefore be responsible for regulating expression of these T-box genes within the limb fields. Fgf-10 expression is also initiated in lateral plate mesoderm around this time, and FGF10 is a good candidate for the mesodermal factor that initiates limb outgrowth and signals the adjacent ectoderm to express FGF8 (Ohuchi et al., 1997; Xu et al., 1998). This makes Tbx4 and Tbx5 prime candidates to encode transcription factors, directly or indirectly regulated by Hox genes, required for the initiation of bud outgrowth and makes Fgf-10 a possible downstream target in the mesoderm (Fig. 7).

In addition to being implicated in determining the specific axial levels at which limbs develop, the results of our experiments with ectopic limbs and limb mesenchyme grafts indicate that Tbx4 and Tbx5 are also involved in the patterning of skeletal elements and determination of the surface structures that form during later stages of development. The identity of limb elements developing in ectopic limbs was directly related to the position and rostrocaudal range of the limb-specific T-box genes being expressed. Rostral buds expressed Tbx4 minimally and Tbx5 extensively, and developed into wing-like mosaic limbs. Mid-flank buds expressed Tbx4 posteriorly and Tbx5 anteriorly, and developed into highly mosaic limbs. Caudal buds expressed Tbx4 extensively and Tbx5 minimally, and developed into leg-like mosaic limbs (Tables 1,2; Fig. 8). Similar findings have recently been reported by Ohuchi et al. (1998). Thus bud regions expressing Tbx4 develop toes and scales, whereas regions expressing Tbx5 develop wing digits and feather buds. Patterning of skeletal elements is controlled by Hox genes differentially expressed during different phases of limb outgrowth (Nelson et al., 1996), probably by the differential regulation of local cell proliferation rates and cell adhesion properties (Dollé et al., 1993; Yokouchi et al., 1995; Goff and Tabin, 1997). Tbx4 and Tbx5 could therefore be interacting with Hox genes during these later phases to regulate the precise position and pattern of the bones that develop. Determination of ectodermal structures (feather buds versus scales) is also controlled by the underlying mesenchyme (Cairns and Saunders, 1954; Saunders et al., 1957, 1959; Saunders and Gasseling, 1968). As in the limb fields, Tbx4 and Tbx5 may therefore regulate the expression of secreted molecules needed to instruct the adjacent ectoderm to develop the appropriate structures.

These observations raise the question of functional equivalence between the Tbx4 and Tbx5 gene products. Wings and legs are serially homologous structures that use the same basic set of signalling molecules to control their outgrowth and patterning (Tickle, 1996; Shubin et al., 1997). It seems likely, therefore, that initiation of wing and leg outgrowth is also effected by common signalling molecule(s). The mutually exclusive, limb-specific expression of Tbx4 and Tbx5 provides an exception to this general paradigm, which could be explained if the two gene products are biochemically equivalent, regulating expression of the same downstream targets (Fig. 7). In contrast to limb bud initiation, however, the nature of the skeletal and ectodermal structures that form at later stages of development appears to be directly related to the specific identity of the T-box gene being expressed, implying non-equivalence of the Tbx4 and Tbx5 gene products. When surface ectoderm from either the wing or the leg is heterotopically transplanted to a host bud from which the ectoderm has been removed, the ectoderm assumes the identity of the host mesenchyme (Saunders and Gasseling, 1968). This shows that the ectoderm is competent to respond appropriately to signals emanating from the mesenchyme, and that these signals must be different in each limb. If T-box genes are involved in sending signals to the ectoderm that determine the identity of ectodermal structures, their downstream targets in the mesenchyme should be different in each limb, suggesting that Tbx4 and Tbx5 have acquired unique downstream targets and lost biochemical equivalency. Alternatively, since genes of the HoxC cluster are also known to be differentially expressed in the two limbs (reviewed by Nelson et al., 1996), the different context within which the two T-box genes are operating could provide a basis for the differential regulation of downstream targets. Clearly these issues will require both gain- and loss-of-function experiments before being resolved.

Tbx2 and Tbx3 are expressed in the posterior part of both limbs prior to the onset of Shh expression at stage 17 (Riddle et al., 1993). In the forelimb, this corresponds to the region expressing Hoxb-8 (Charité et al., 1994; Chan et al., 1995), which has been shown to be important for development of the ZPA (Charité et al., 1994). Tbx2 and/or Tbx3 may therefore be involved in establishment of the ZPA and the induction of Shh. After formation of the ZPA, however, both of these genes continue to be expressed in the posterior limb margin and, at least in the case of Tbx2, we have shown that this expression can lie downstream of SHH. This observation is consistent with the recent observation that omb, the Drosophila ortholog of Tbx2, may be a direct target of hedgehog, the Drosophila ortholog of vertebrate Shh, during patterning of the abdominal segments in the adult fly (Kopp and Duncan, 1997). It therefore appears that, depending on the phase of limb development, expression of these two genes can lie either upstream or downstream of SHH. This observation is similar to previous findings that Hox gene expression is differentially regulated during different phases of limb development (Nelson et al.,1996).

In contrast to the effect on all other genes previously implicated in limb outgrowth and patterning, removal of the AER does not lead to loss of expression of Tbx2-Tbx5. The functional significance of this observation is not yet clear, however it does imply that, once established, some sort of autoregulatory loop might be maintaining T-box gene expression in the absence of external inductive signals. The autoregulatory loop demonstrated for T (Brachyury) (Schulte-Merker and Smith, 1995) provides a precedent for this type of control in the T-box gene family.

Since Tbx2-Tbx5 are all expressed in lateral plate mesoderm before the appearence of a morphologically defined bud, beginning at the time of anteroposterior axis specification, we have proposed that these genes could all be involved in establishment of the limb axes (Gibson-Brown et al., 1996). Recently it has been found that mutations in the human orthologs of two of these genes (TBX3 and TBX5) are responsible for two dysmorphic developmental syndromes. TBX3 mutations are responsible for causing ulnar-mammary syndrome, in which postaxial limb defects, including loss of digit 5 and reduction or loss of the ulna, are found (Bamshad et al., 1997). TBX5 mutations are responsible for causing Holt-Oram syndrome, in which preaxial limb defects, including loss of the thumb, reduction of the radius and phocomelia, are found (Basson et al., 1997; Li et al., 1997). These heterozygous phenotypes support our hypothesis that both of these genes are involved in establishment of the anteroposterior limb axis. Interestingly, both of these syndromes result from haploinsufficiency of the respective gene products, and because no nullizygous patients have been reported, it is possible that a more severe, possibly embryonic-lethal, phenotype might result from complete loss-of-function.

Holt-Oram syndrome patients exhibit forelimb defects only, which is not surprising as limb-associated Tbx5 expression is confined to the forelimb (Gibson-Brown et al., 1996; Li et al., 1997). Ulnar-mammary syndrome patients usually exhibit forelimb defects only, although considerable variation in expressivity is found, with several cases reported in which there are also subtle postaxial defects in the hindlimb (Schinzel et al., 1987). Tbx3 is expressed in both forelimbs and hindlimbs, and the general absence of a hindlimb phenotype in ulnar-mammary syndrome suggests at least partial functional redundancy for TBX3 in the hindlimb. Since Tbx2 and Tbx3 are expressed in very similar patterns in both limbs, it is possible that TBX2 can compensate for the loss of TBX3 function in the hindlimb in most ulnar-mammary syndrome patients. So far, no human developmental syndromes have been identified as candidates for TBX2 or TBX4 mutations. It will be interesting to see whether mutations in TBX2 and TBX4 have similar phenotypic consequences in the hindlimb as mutations in TBX3 and TBX5 exhibit in the forelimb.

The first vertebrates to develop paired appendages, the osteostracan fishes, appeared in Devonian seas around 400 million years ago (Carroll, 1988; Maisey, 1996). However, these jawless fishes only possessed paired pectoral appendages; no evidence of pelvic fins has ever been discovered in a fossil or extant agnathan. Tbx2 and Tbx3 are expressed in very similar patterns in both the forelimb and hindlimb (Gibson-Brown et al., 1996, this study), and are both derived from a common ancestral locus (the Tbx2/3 locus) that was duplicated at a very early point along the vertebrate lineage (Agulnik et al., 1996; Ruvinsky and Silver, 1997). Tbx2 and Tbx3 are linked, respectively, to Tbx4 and Tbx5, which were also derived from a common ancestral locus (the Tbx4/5 locus) in the same duplication event (Agulnik et al., 1996; Ruvinsky and Silver, 1997). This raises the possibility that the ancestral Tbx2/3, Tbx4/5 gene pair was involved in development of the paired pectoral fins of ancient agnathans, and that evolution of paired pelvic fins may only have been possible following duplication of these genes and establishment of the two cognate gene pairs (Tbx2/Tbx3 and Tbx4/Tbx5). According to this model, Tbx4/5 gene function was conserved by Tbx5 for specification and development of the pectoral appendages, whereas Tbx4 was then available to be recruited (co-opted) into serving an analogous role in specifying novel structures, the paired pelvic fins, at a different level along the primary body axis. Elaboration of the Tbx2-Tbx5 subfamily may therefore have been an important element in the evolution of gnathostome appendages.

This work was supported in part by NIH grants HD 20275 (L. M. S.) and HD 33082 (V. E. P.), by the Raymond and Beverly Sackler Foundation (V. E. P., J. J. G. B.), by an MSKCC support grant, and by the Pew Scholars Program (L. N.). We would like to thank Dr Juan Carlos Izpisùa-Belmonte and Dr Susan Mackem for providing the chick cDNA libraries, and Dr Ispisùa-Belmonte for generously providing the chick Tbx2 plasmid. We would also like to thank the members of our laboratories for excellent technical support and constructive criticism of the manuscript, and Dr Claudio Stern and the members of his laboratory for stimulating discussions and advice. Particular thanks are due to Drs Debbie Chapman and Jeff Yoder and to Naiche Adler.