Finger or toe: the molecular basis of limb identity

The mature limb is a complex structure comppsed of many different tissues types. To produce the interconnected array of bones, muscles and tendons that make up the limb, processes such as cell fate specification, proliferation,apoptosis and migration must be tightly controlled. Many of the genes required for initiating limb development, and for subsequent patterning of the limb bud, are now known. In most cases, these genes are expressed in identical patterns in both the developing forelimb and hindlimb, and are believed to play identical roles in generating homologous limb elements; for example,fingers in the hand and toes in the foot. This raises the important and general question in developmental biology of how two different structures are produced by common signalling cascades. Recent work has begun to explain how genes that specify limb-type identity interact with signalling cascades common to both forelimb and hindlimb. Limb-type specification therefore provides an ideal model for studying the way in which the embryo produces serially homologous structures by modulating the outcome of conserved gene regulatory networks.

The mechanisms that control the development of both limb-types have been reviewed comprehensively elsewhere(Capdevila and Izpisua Belmonte,2001; Tickle,2003). The aim of this review is to cover recent developments that have advanced our understanding of how the differences between forelimbs and hindlimbs are established in vertebrates. I will discuss the roles of genes that are expressed exclusively in one limb-type and not the other during embryonic development, and outline the current understanding of how these genes interact with regulatory networks common to both limb-types to produce structures unique to the forelimb and hindlimb.

The forelimb and hindlimb buds are derived from regions of the lateral plate mesoderm (LPM). The LPM comprises two strips of tissue that run along the length of the main body axis, lateral to the somites(Fig. 1A). The territories of the LPM that give rise to limbs are located at defined rostrocaudal levels along the main body axis (Fig. 1A). Limb bud outgrowth is initiated when cells within the prospective limb-forming regions respond to cues from more medial tissues(reviewed by Capdevila and Izpisua Belmonte, 2001). Initially, the limb buds are morphologically uniform collections of cells, and these subsequently develop into structurally distinct limb elements. The limb musculature is derived from myotomal cells that originate in the somites medial to the limb bud and then migrate into the limbs.

The limb has three primary axes. The anteroposterior axis runs from the thumb or big toe down to the little finger or toe, respectively, the proximodistal axis starts at the shoulder or hip and runs down to the tips of the digits, whereas the dorsoventral axis passes from the back to the front of the hand or foot (Fig. 1B). Patterning of each limb axis is controlled by key signalling centres within the limb bud. The determination of the anteroposterior axis is regulated by cells in a posterior domain of the limb bud called the zone of polarizing activity (ZPA), which express the secreted protein sonic hedgehog (Shh)(reviewed by Tickle, 2003). Proximodistal patterning is controlled by cells of the apical ectodermal ridge(AER), a specialised ectodermal structure that runs along the length of the distal tip of the limb bud and that expresses proteins belonging to the fibroblast growth factor (Fgf) family of secreted proteins (reviewed by Martin, 1998). In the proximal limb bud, Meis genes, induced by retinoic acid (RA), appear to control the development of proximal structures, indicating that patterning of the proximodistal axis is controlled by a combination of opposing RA and Fgf signals (reviewed by Capdevila and Izpisua Belmonte, 2001). Dorsoventral patterning of the limb is controlled, at least in part, by signalling from the dorsal ectoderm to the underlying mesoderm, possibly by Wnt7a (reviewed by Chen and Johnson, 1999). In the developing forelimb bud, cells coordinate the cues from all three signalling centres to produce characteristic forelimb elements. Cells in the hindlimb respond to the same signals but produce structures characteristic of the hindlimb.

The first morphological sign of limb development is a budding of tissue from the LPM. The first transplant experiments to explore the determination of limb-type identity were carried out in the newt. Harrison demonstrated that cells of the limb-forming region were able to give rise to fully developed limbs when transplanted to ectopic sites along the flank of the embryo(Harrison, 1918). Similar experiments using chick embryos demonstrated that limb-type identity is determined by the origin of the graft(Hamburger, 1938): tissue taken from the wing-forming region produces an ectopic wing and cells transplanted from the leg-forming region always produce a new leg. Tissue grafts from the prospective limb-forming tissue therefore behave as classically defined developmental fields in that they have the properties of self-determination and self-organisation(Jacobson and Sater, 1988). Building upon these original observations, more extensive transplantation experiments in the chick have mapped the temporal and spatial sequence of the limb-forming capacity of the LPM (Stephens et al., 1989). Transplanted LPM tissue taken from prospective limb-forming regions forms a limb when placed in an ectopic location at stages just prior to overt limb outgrowth. Similar transplants of LPM tissue from the neck or inter-limb flank at the same stage will not form a limb. The limb-type potential of the grafts is fixed by this stage; prospective wing and leg forming regions will only develop into wing or leg, respectively.

If heterotopic grafts in the chick are performed after a limb bud has formed, a chimaeric limb develops that contains elements of both limb-types(Saunders, 1948; Kieny, 1964). Furthermore,when prospective thigh mesoderm is grafted beneath the AER of the developing chick wing (Saunders et al.,1957; Saunders et al.,1959) it is re-specified in its proximodistal fate, but its limb-type identity remains fixed. These grafts form hindlimb toes and not wing digits, demonstrating that the identity of the developing limb bud is stably determined by this stage and remains unaltered if grafted to a new position within the embryo.

Tissue recombination experiments in the chick have also demonstrated that limb-type identity is determined by the mesodermal, rather than the ectodermal, component of the limb(Zwilling, 1956a; Zwilling, 1956b). If wing-bud mesenchyme is combined with leg-bud ectoderm and then transplanted to the flank, a normally patterned wing is produced. The reciprocal recombination,leg-bud mesenchyme combined with wing-bud ectoderm, produces a normally patterned leg. Furthermore, when surface ectoderm from one limb-type is grafted to a host bud of another, the ectoderm assumes the identity of the host mesenchyme (Saunders and Gasseling,1968). Limb-type identity is therefore a property of the mesoderm,which provides instructions to the overlying ectoderm.

Classical embryological experiments have therefore established the criteria that candidate specifiers of limb-type identity need to meet. Such candidates are expected to be expressed prior to overt limb-bud formation and to show maintained expression during subsequent limb bud outgrowth. Their expression is also expected to be restricted to the limb mesenchyme.

Most factors thought to play important roles in vertebrate limb patterning have common expression patterns in both forelimb and hindlimb, but there are exceptions to this general rule. The first genes found to be expressed exclusively in either forelimb or hindlimb were members of the Hox gene family(Nelson et al., 1996; Oliver et al., 1988; Simon and Tabin, 1993; Tabin, 1989). However, none of these genes satisfy the necessary criteria to qualify as candidates for specifying limb-type identity; they are expressed at relatively late stages of limb development and are not expressed throughout the limb mesenchyme. Instead, they may be downstream targets, and perhaps downstream effectors, of factors that specify limb-type identity. Other data indicate that Hox genes also act upstream of factors that control limb-type identity. Hox9 paralogues and other genes of the four main Hox clusters are believed, from their nested expression patterns along the main body axis, to create a combinatorial code that may specify the molecular identity of the prospective forelimb and hindlimb territories (Cohn et al.,1997; Rancourt et al.,1995). However, gene deletion analysis in the mouse has not been informative regarding the potential roles for these genes in positioning the limb-forming regions.

Studies on a range of vertebrate species have identified three genes with temporal and spatial expression patterns consistent with their having roles in specifying limb-type identity. Two T-box family transcription factors, Tbx5 and Tbx4, are expressed in the forelimb and hindlimb buds, respectively (Gibson-Brown et al.,1996; Ruvinsky et al.,2000; Simon et al.,1997; Takabatake et al.,2000) (Fig. 2A,B). A third gene, the paired-type homeodomain factor Pitx1, is expressed in the developing hindlimb but not in the forelimb(Lamonerie et al., 1996; Szeto et al., 1996)(Fig. 2C). Each of these three genes are expressed throughout the limb mesenchyme and not in the limb ectoderm (Gibson-Brown et al.,1998; Isaac et al.,1998; Logan et al.,1998; Ohuchi et al.,1998). Furthermore, these studies show that the limb-type restricted expression pattern of each gene is retained in grafts of wing tissue into leg bud, or leg tissue into wing bud in chick embryos.

Tbx5, Tbx4 and Pitx1 are also expressed in ectopic limbs that have been induced by Fgf. Ectopic limbs form following the application of Fgfs to the cells of the interlimb flank of the chick(Cohn et al., 1995; Ohuchi et al., 1995). The type of limb that forms (wing or leg) depends upon the rostrocaudal level at which the Fgf is applied. A source of Fgf applied adjacent to the wing-forming region (opposite somite 21/22, see Fig. 1A) induces an ectopic wing, whereas similar application adjacent to the leg-forming region (opposite somite 25) generates an ectopic leg. Remarkably, application of Fgf in the middle of the interlimb region produces mosaic limb elements, the anterior portion with wing-like character and the posterior portion possessing leg-like elements(Isaac et al., 1998; Logan et al., 1998; Ohuchi et al., 1998). The expression patterns of Tbx5, Tbx4 and Pitx1 in the ectopic limb correlate with the position of the ectopic bud, and are consistent with the wing or leg-like structures that are ultimately produced. The tight correlation of gene to limb-type is also observed in mosaic buds that give rise to limbs comprising both wing-like and leg-like elements. The anterior portion expresses Tbx5, whereas the posterior expresses Tbx4and Pitx1. Significantly, however, the boundary between domains of ectopic Tbx5 or Tbx4 and Pitx1 expression is not fixed at a particular somite level, suggesting that there is no definitive rostrocaudal level that divides the ectopic wing and leg territories(Ohuchi et al., 1998).

Direct evidence that Tbx5, Tbx4 and Pitx1 have roles in specifying limb-type identity has been obtained from misexpression experiments in the chick. Ectopic expression of Tbx5 in the developing chick hindlimb bud can, at least partially, transform the identity of the leg to a more wing-like morphology(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). In the limb bud, ectopically expressed Tbx5 can induce the expression of forelimb markers, such as Hoxd9, and repress the expression of hindlimb markers, for example Hoxc9. If allowed to develop to later stages, the resulting limbs produce elements with morphological characteristics of a wing, such as wing digits and feather buds. In the converse experiment, hindlimb-specific genes Tbx4 or Pitx1 misexpressed in the forelimb can transform the wing to a more hindlimb character (Logan and Tabin,1999; Takeuchi et al.,1999). Similarly, Tbx4 can transform the identity of Fgf-induced ectopic limbs induced in the interlimb region of the LPM(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). Again, genes that are normally restricted to the hindlimb,such as Hoxc10 and Hoxc11, are ectopically induced in the targeted forelimb or ectopic limb and, if these limbs are allowed to develop further, they produce characteristic leg-like elements such as clawed toes and scales. In each case, the transformed limb contains elements that are derived from both the mesenchyme, such as muscle and skeletal elements, and the ectoderm, such as feather buds and scales. This indicates that ectopically expressed Tbx5 and Tbx4 alter cell fate by acting directly on the mesoderm, and indirectly on the ectoderm.

Although the limb phenotypes obtained in chicks following the ectopic expression of Tbx5, Tbx4 or Pitx1 are dramatic, they are not completely penetrant, and the resulting limbs are made up of constituent forelimb, hindlimb or intermediate limb-type characteristics. This can be explained in part by the observation that, following the misexpression of Tbx4 or Pitx1, the endogenous expression of Tbx5 in the forelimb is unchanged (Logan and Tabin, 1999; Rodriguez-Esteban et al., 1999; Takeuchi et al.,1999). This observation indicates that in these misexpression experiments, cells in the limb are presented with two competing programmes; in some cases one programme may dominate, whereas in others the cells may assume a mixed or intermediate fate.

The requirement for Tbx5 during vertebrate forelimb initiation has been confirmed by knockdown experiments in the chick and by targeted gene inactivation in the mouse. Mouse embryos mutant for Tbx5 have no detectable forelimb bud outgrowth and no morphologically distinct AER(Agarwal et al., 2003; Rallis et al., 2003). Tbx5 is required for normal heart development and therefore Tbx5 mutant mice do not survive beyond early limb bud stages(Agarwal et al., 2003). To overcome this mid-gestation lethal phenotype, a conditional approach has been taken to delete Tbx5 function in the developing limbs while leaving the gene intact in the heart; this generates embryos that survive until birth(Rallis et al., 2003). In these embryos, all the elements of the limb, including the shoulder girdle,are absent, demonstrating that Tbx5 is required for the formation of all skeletal elements of the forelimb.

Misexpression experiments in the chick have indicated that Tbx5specifies forelimb identity, in part, by repressing expression of the hindlimb genes Tbx4 and Pitx1(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). However, in mouse embryos mutant for Tbx5, the hindlimb genes Tbx4 and Pitx1 were not induced in the presumptive forelimb-forming region. These genetic studies indicate that, in the mouse, Tbx5 does not normally repress Tbx4 in the forelimb (Agarwal et al.,2003).

Targeted inactivation of Tbx5 in the mouse, and knockdown of Tbx5 in zebrafish and chick, have recently provided new insights into the function of this gene during limb development. To understand the role of Tbx5 in early limb formation, it is necessary to review our current understanding of the molecules involved in limb initiation. A number of experimental approaches in a range of model organisms indicate that Tbx5 is essential for forelimb bud formation, and that it interacts with the Fgf and Wnt signalling pathways to initiate limb bud outgrowth (Agarwal et al.,2003; Ahn et al.,2002; Garrity et al.,2002; Ng et al.,2002; Rallis et al.,2003; Takeuchi et al.,2003). In the following sections, the current understanding of the interactions between Tbx5 and Fgf and Wnt family proteins, which initiate and maintain limb bud formation, is outlined.

Members of the Fgf family of secreted proteins are thought to play essential roles in both forelimb and hindlimb bud initiation. As outlined earlier, several Fgfs are able to induce the formation of ectopic limbs from the flank of the embryo (reviewed by Martin, 1998). In addition,deletion of Fgf10 in the mouse results in the absence of limbs(Min et al., 1998; Sekine et al., 1999). Fgf10 signalling in the limb mesenchyme is required to induce and maintain normal Fgf8 expression in the AER(Min et al., 1998; Ohuchi et al., 1997; Sekine et al., 1999). A positive-feedback loop is then established in which Fgf8 expression in the AER maintains expression of Fgf10 in the underlying mesenchyme(Fig. 3).

Members of the Wnt family of secreted proteins are also thought to have roles during both forelimb and hindlimb initiation in the chick. Two Wnts are differentially expressed in the LPM just prior to limb outgrowth: Wnt2b is expressed in the prospective forelimb region whereas Wnt8c is expressed in the prospective hindlimb region. Wnt2band Wnt8c can induce Fgf10, and, if misexpressed in the chick interlimb flank, ultimately the growth of an ectopic limb(Kawakami et al., 2001). However, although Wnt2b and Wnt8c can induce limb initiation, and are differentially expressed in either the forelimb or hindlimb forming regions, they themselves do not appear to influence the identity of the limb. The type of limb they induce depends on where along the rostrocaudal axis they are misexpressed, rather than on the identity of the individual Wnt molecule.

In the zebrafish, initiation of the pectoral fin bud, which is analogous to the forelimb of the chick and mouse, is regulated by both tbx5 and wnt2b. Functional knockdown of tbx5 in the zebrafish using antisense morpholino oligonucleotides results in a failure to initiate pectoral fin bud formation, and the absence of fgf10 and fgf8 expression in the fin bud mesenchyme and AER, respectively(Ahn et al., 2002; Garrity et al., 2002; Ng et al., 2002). A mutation in the zebrafish tbx5 gene heartstrings, induced by the chemical mutagen ENU, produces an identical phenotype to the tbx5morpholino-knockdown phenotype (Garrity et al., 2002). Morpholino knockdown of wnt2b produces phenotypes similar to those obtained following knockdown of tbx5, and tbx5 expression is downregulated in embryos in which wnt2bactivity is disrupted (Ng et al.,2002). Furthermore, although tbx5 partially rescues the phenotype of wnt2b knockdown, injection of wnt2b RNA could not rescue the phenotype resulting from tbx5 knockdown. Together, the results suggest that pectoral fin initiation is co-regulated by wnt2band tbx5, and that tbx5 lies downstream of wnt2b(Ng et al., 2002).

As yet, no Wnt family member has been described in the mouse that is expressed in a similar manner to that of Wnt2b and Wnt8c in the chick, or wnt2b in the zebrafish. Furthermore, in mouse embryos mutant for two nuclear components of Wnt signalling, Lef1 and Tcf1, limb bud outgrowth is initiated normally(Galceran et al., 1999). This suggests that a comparable role for Wnt signalling in limb bud initiation may not have been conserved in the mouse. Alternatively, other Wnt genes in the mouse might be expressed in a comparable pattern to Wnt2b and Wnt8c in the chick, and other Tcf genes may compensate for the loss of Tcf1 in Lef1^-/-/Tcf1^-/-embryos (Ng et al., 2002). Further experiments to determine the signalling hierarchy between Tbx5 and Fgf and Wnt proteins are discussed later in the review.

To examine the genetic hierarchy of Tbx5 and Fgfs in limb initiation, the expression patterns of these genes have been analysed in different mutant mouse backgrounds. In the forelimbs of embryos mutant for Fgf10, Tbx5 is initially activated but expression is not maintained,suggesting that Tbx5 initially acts upstream of Fgf10(Sekine et al., 1999)(Fig. 3). Consistent with this model, deletion of Tbx5 in the mouse results in the loss of Fgf10 expression in the limb bud mesenchyme(Agarwal et al., 2003; Rallis et al., 2003). Similarly, following knockdown of tbx5 function in zebrafish(Ahn et al., 2002; Ng et al., 2002), or misexpression of dominant-negative forms of Tbx5 in the chick(Ng et al., 2002; Takeuchi et al., 2003), Fgf10 expression is either absent or downregulated in the developing limb. As a consequence of failure of normal Fgf10 signalling in Tbx5 mutants, Fgf8 expression is also disrupted in the limb(Agarwal et al., 2003; Ng et al., 2002; Rallis et al., 2003; Takeuchi et al., 2003).

A number of observations suggest that activation of Fgf10 by Tbx5 expression is direct, as T-box binding sites have been identified in the Fgf10 promoter in human, mouse and zebrafish(Agarwal et al., 2003; Ng et al., 2002). The human and mouse Fgf10 promoter also contains consensus binding sites of the nuclear mediators of Wnt signalling Lef1 and Tcf1,suggesting additional regulation by Wnt family members(Agarwal et al., 2003; Ng et al., 2002). In in vitro transactivation studies, Tbx5 can activate a Fgf10 promoter construct. This promoter can also be activated by a constitutively-active form of β-catenin, which is an essential intracellular component of the canonical Wnt signalling pathway. Furthermore, Tbx5 andβ -catenin, in combination, have an additive effect on activation of the Fgf10 promoter, indicating that, in mouse, Tbx5 and Wnt signalling may act together to regulate Fgf10(Agarwal et al., 2003). Direct positive-feedback loops between T-box genes and Fgfs have been established in other tissues during development. In the chick, a regulatory loop between Tbx4 and Fgf10 has been implicated in lung formation(Sakiyama et al., 2003),whereas in Xenopus a positive regulatory loop between eFgfand Brachyury operates during mesoderm formation(Schulte-Merker and Smith,1995).

Several experiments have been carried out to determine the hierachical relationship between Tbx5 and Wnt and Fgf signalling. Lef1 is a transcription factor that is part of the transcriptional read-out of Wnt signalling during limb initiation. In in vitro co-transfection experiments, mutation of a conserved Lef1 binding site in the 5′ promoter of both human and mouse Tbx5 leads to a decrease in the activation of expression constructs containing this promoter. This indicates that Wnt signalling may normally activate the Tbx5 promoter. Consistent with this model, in chick embryos, misexpression of axin, a negative regulator of the Wnt pathway, leads to the downregulation of Tbx5(Ng et al., 2002).

However, other experiments carried out in mouse and chick show that Tbx5 acts upstream of Wnt signalling during limb initiation. The expression of Tbx5 and Fgf10 was examined in Lef1^-/-/Tcf1^-/- mouse embryos(Galceran et al., 1999). In these embryos, expression of Tbx5 was unaffected, whereas Fgf10 expression was detected only at low levels(Agarwal et al., 2003). These results indicate that, in the mouse, Tbx5 acts upstream of Wnt signalling in the forelimb field, and that Wnt signalling, viaβ -catenin, Lef1 and Tcf1, is required to maintain normal levels of Fgf10 expression. In support of these conclusions,in the chick, dominant-negative forms of Tbx5 misexpressed in the limb led to the absence of Wnt2b and Fgf10 expression in the prospective wing region (Takeuchi et al.,2003). In addition, misexpression of a dominant-negative form of Lef1 initially had no effect on Tbx5 expression(Takeuchi et al., 2003). These observations suggest that Tbx5 acts upstream of Wnt signalling(Agarwal et al., 2003; Takeuchi et al., 2003),although contradictory data in chick (Ng et al., 2002) indicates that this may have to be confirmed. Moreover, the evidence from both mouse and chick contrasts with the results from zebrafish, described earlier (Ng et al., 2002), which indicate that in zebrafish Tbx5 and Wnt2b co-regulate limb initiation.

To understand the relationship between Tbx5 and Fgf signalling during limb outgrowth, mouse embryos with deleted Fgfr2 have also been also examined. Two different isoforms of Fgfr2 exist; these isoforms act as receptors for Fgf8 (isoform Fgfr2c) or Fgf10(isoform Fgfr2b), and both isoforms are required for Fgf signalling during limb bud formation (reviewed by Martin,1998; Xu et al.,1998). In the presumptive forelimb buds of Fgfr2 mutant mouse embryos, Tbx5 expression was initiated but was not maintained,indicating that the initial expression of Tbx5 is independent of both Fgf8 and Fgf10, but might be maintained by Fgf signalling. Furthermore,application of a potent inhibitor of Fgfr, SU5402, led to the downregulation of Tbx5 expression in the limb(Ng et al., 2002), consistent with a role of Fgf signalling in the maintenance of Tbx5 expression. Together, these results indicate that Tbx5 acts directly upstream of Fgf10 to initiate limb outgrowth. Subsequently, Tbx5 and Wnt signalling act cooperatively to maintain expression of Fgf10. It appears that a positive-feedback loop is established in which Fgf10is required to maintain expression of Tbx5.

Although in the mouse, Wnt genes do not appear to be involved in forelimb initiation, Wnt signalling is required for proper establishment and maintenance of the AER. In Lef1^-/-/Tcf1^-/- double-mutant embryos, Tbx5 expression is unaffected and limb bud outgrowth is initiated,but the AER does not form and outgrowth is not maintained(Galceran et al., 1999). This is consistent with the proposed roles of Wnt3 in the mouse(Barrow et al., 2003) and Wnt3a in the chick (Kengaku et al., 1998; Kengaku et al.,1997) in AER formation.

Following morpholino knockdown of tbx5 in the zebrafish, tbx5-expressing cells in the dorsoanterior LPM fail to move into the pectoral fin-bud producing region, which is situated in a more ventroposterior position. This indicates that tbx5 is required for the proper migration of cells that will form the pectoral fin(Ahn et al., 2002). It is unclear whether this observation has any significance in higher vertebrates because, in chick and mouse, Tbx5 is first detected in the region of the LPM that gives rise to the forelimb buds(Gibson-Brown et al., 1996; Gibson-Brown et al., 1998),and these cells do not undergo a similar migration. Zebrafish ikarus(ika) mutants have mutations in fgf24, which functions in the genetic cascade responsible for fin initiation and is implicated in the migration of tbx5-expressing mesenchymal cells into the fin bud(Fischer et al., 2003). ika mutants survive to adulthood but lack pectoral fins, although they have normal pelvic fins and no other obvious defects. tbx5-expressing cells fail to migrate normally in ikamutants, indicating that fgf24 is required for their correct migration to the pectoral fin primordium. As the fgf24 mutation only affects the pectoral fin, other molecules must operate during pelvic fin initiation.

Experiments in which Tbx5 function has been disrupted are of clinical significance because TBX5 mutations in humans are associated with Holt-Oram Syndrome [HOS; Online Mendelian Inheritance in Man (OMIM)database number 142900 http://www.ncbi.nlm.nih.gov/omim/],a dominant disorder that is characterised by upper(fore)limb abnormalities and heart defects (Basson et al.,1997; Li et al.,1997). Haploinsufficiency of TBX5 causes a range of limb abnormalities, the mildest phenotype being triphalangeal (three-jointed) thumb but which most commonly results in the failure of limb elements to form. A consistent feature of HOS is that abnormalities affect the anterior structures of the limb, such as the thumb or radius. Consistent with observations in humans, the heterozygous Tbx5 knockout mouse has defects in digit I(Bruneau et al., 2001). The knockdown of Tbx5 function in zebrafish and chick also produces phenotypes that are strikingly similar to those observed in HOS patients. Morpholino knockdown of tbx5 in the zebrafish produces a range of phenotypes. The most severe cases are identical to those observed in mouse null mutants: complete absence of the entire pectoral fin. However, by decreasing the amount of morpholino injected, which is expected to decrease the effective knockdown of the gene product, progressively milder phenotypes are produced, such as shortened pectoral fins and a reduced number of rays(Ng et al., 2002). In the chick, a consistent feature of misexpression of dominant-negative forms of Tbx5 is the disruption of the anterior limb mesenchyme and, as a consequence, the structures that are ultimately derived from this region of the limb bud (Ng et al., 2002; Rallis et al., 2003; Takeuchi et al., 2003). The most common phenotype is the absence of the radius and anterior digits, and this is associated with an observed downregulation of Fgf10 and Fgf8 in the anterior of the limb, as well as of molecular markers of the anterior mesenchyme. The analysis of Tbx5 mutants in a range of vertebrate model organisms from fish to mouse will provide molecular insights into how the defects observed in HOS arise.

Gene deletion, knockdown and misexpression experiments have demonstrated that Tbx5 is the primary and direct initiator of forelimb bud outgrowth. In a search for candidate genes with similar roles in the hindlimb,the simplest model proposes that the closely-related gene Tbx4 plays an analogous role in the hindlimb. Experiments in the chick have demonstrated that this might be the case; misexpression of Tbx4 in the flank was sufficient to induce the formation of an ectopic limb(Ng et al., 2002; Takeuchi et al., 2003). However, deletion of Tbx4 in the mouse indicates that the action of Tbx4 in the hindlimb may not simply parallel that of Tbx5 in the forelimb. In mouse embryos mutant for Tbx4, hindlimb bud induction and initial patterning occurs normally. However, at later stages of development Fgf10 expression is not maintained in the mesenchyme and mutant limbs fail to develop beyond early limb bud stages(Naiche and Papaioannou,2003). In addition to displaying hindlimb defects, Tbx4mutant embryos fail to undergo chorioallantoic fusion. The failure of this fusion is thought to cause embryonic lethality by embryonic day (E) 10.5. Unfortunately, this lethality prevents a more detailed analysis of the hindlimb defects of Tbx4 null mice at later stages of development. Nevertheless, two aspects of the Tbx4 mouse mutant phenotype raise particularly intriguing questions. First, what are the mechanisms that allow hindlimb bud initiation to proceed normally in the absence of Tbx4when forelimb bud initiation does not occur in Tbx5 mutants? Second,why do Tbx4 mutant limbs fail to develop beyond early limb bud stages?

To understand the events surrounding hindlimb bud initiation, markers of the AER and limb mesenchyme have been analysed in Tbx4 null-mutant embryos. In these embryos, at early stages of hindlimb initiation, Fgf8 is expressed normally in the AER. Msx1, a gene normally expressed in the underlying mesenchyme in response to Fgf signalling from the AER, is also unaffected (Naiche and Papaioannou, 2003). The establishment of normal Fgf signalling in mutant limb mesenchyme was confirmed by the distribution of doubly phophorylated Erk protein (dpErk), which detects activation of the Map kinase cascade by the Fgf receptor and serves as a read-out of Fgf signalling. Normal Fgf signalling in mutant limb mesenchyme may explain why limb bud initiation proceeds normally in Tbx4 null embryos. Pitx1 expression is also maintained in mutant embryos, indicating that its expression is not dependent on Tbx4 (Naiche and Papaioannou, 2003). This may also explain how normal hindlimb initiation occurs in the absence of Tbx4; the loss of Tbx4in the hindlimbs may be compensated for by a Pitx gene. Compensation for the loss of Tbx4 by Tbx5 can be ruled out because Tbx5expression remains restricted to the forelimb, and is not ectopically induced in the mutant hindlimb. Nevertheless, despite the apparently normal induction of the hindlimb, and the establishment of the Fgf-feedback loop between the mesenchyme and ectoderm, Fgf10 expression is not maintained in the hindlimb of Tbx4 null-mutant embryos. The mechanism of this failure is not understood but it may explain why Tbx4 mutant limbs fail to develop beyond early limb bud stages.

A role for Pitx1 in the specification of hindlimb identity has been supported by both gene-misexpression and gene-inactivation experiments. Misexpression of Pitx1 in the developing chick wing can, at least partially, transform both muscle and skeletal elements of the forelimb and give them a more hindlimb-like character(Logan and Tabin, 1999; Takeuchi et al., 1999). In Pitx1 null mice, the hindlimb does form, but features of hindlimb identity are lost and the limb has a more forelimb-like character(Lanctot et al., 1999; Szeto et al., 1999). For example, the diameters of the tibia and the fibula in the Pitx1mutant limb are similar to those of the homologous elements in the forelimb,the radius and the ulna, respectively. In addition, the secondary cartilage of the knee joint, comprising the patella and fabella, fails to form, resulting in an articulation that is more reminiscent of the elbow joint. Levels of Tbx4 transcripts are lower in Pitx1^-/- hindlimbs than in wild type, but are nevertheless detectable(Lanctot et al., 1999). This is consistent with the notion that Pitx1 contributes to Tbx4expression in the hindlimb, but demonstrates that Pitx1 is not absolutely required for the induction of Tbx4.

A surprising observation to emerge from of the analysis of the Pitx1^-/- null mouse embryos was an occasional left-right(L-R) asymmetry in the severity of the limb phenotype(Lanctot et al., 1999). The right hindlimb was often more severely affected than the left hindlimb,possibly due to a degree of redundancy between Pitx1 and the Pitx1-related homeobox factor Pitx2. During embryogeneis,the Pitx2 gene is asymmetrically expressed at many sites on the left side of the embryo and is an effector of L-R asymmetry (reviewed by Capdevila et al., 2000). Remarkably, analyses of the Pitx1^-/-;Pitx2^-/- double knockout and Pitx1^-/-;Pitx2^-/+ mouse embryos have confirmed this, and have established that Pitx1 and Pitx2 can co-operate during hindlimb formation(Marcil et al., 2003). Pitx1 and Pitx2 are co-expressed in the tailbud region,which includes a region that is destined to become hindlimb, although this co-expression is not maintained at later stages of development. Hindlimb buds in Pitx1^-/- embryos are smaller than normal limb buds. The smaller limb size relative to normal hindlimbs is more pronounced in Pitx1^-/-;Pitx2^-/+ mouse embryos than in Pitx1^-/-;Pitx2^-/- embryos(Marcil et al., 2003). The reduction in limb size is more pronounced in the anterior than the posterior of the limb bud, and the skeletal defects that ultimately arise are consistent with the reduction in anterior mesenchyme observed at earlier stages. Although no dramatic reduction in Fgf10 or Fgf8 expression is observed in mutant limbs, their onset of expression is delayed and their expression levels are slightly reduced, which may account, in part, for the phenotype of Pitx1^-/-;Pitx2^-/+ double mutant embryos. Accordingly, the phenotype of Pitx1^-/-;Pitx2^-/- mice is similar to that observed following conditional deletion of Fgf8 in the AER(Sun et al., 2002), and the phenotypes that result from disruption of Tbx5 and Tbx4 in the forelimb and hindlimb, respectively. However, the higher levels of Pitx1 protein compared with Pitx2, and the absence of any obvious limb phenotype in Pitx2^-/- embryos, is consistent with Pitx1 being the primary determinant of hindlimb identity and its subsequent development. Only in the absence of Pitx1 does the contribution of Pitx2 become evident.

The observation that the hindlimbs of Pitx1 null mice possess forelimb-type characteristics raises some interesting points of interpretation: does the phenotype correspond to a loss of hindlimb character and the `acquisition' of forelimb identity, or does it represent a loss of hindlimb character that results in the reversion to an ancestral limb `ground state'? Tbx5 expression remains forelimb-specific in Pitx1^-/- embryos, and is not ectopically induced in the hindlimb. By this criteria at least, the mutant hindlimb has not `acquired'forelimb identity as such, but rather has lost hindlimb identity. By extension, forelimb morphology may be most similar to the ancestral ground state and hindlimb identity the more derived character, perhaps due to the acquisition of Pitx1 expression in the hindlimb-forming region(Marcil et al., 2003). Indeed, Pitx1 may play a pivotal role in defining the difference between forelimb and hindlimb identity, the two distinct limb-forming territories being distinguished by whether they express Pitx1 or not. To date no forelimb specific equivalent to Pitx1 has been identified. A third member of the Pitx subfamily, Pitx3(Semina et al., 1997), is not expressed in developing limbs (Marcil et al., 2003). Further experiments will be required to clarify these issues.

Gene deletion and misexpression studies have demonstrated that Tbx5,Tbx4 and Pitx1 have fundamental roles in limb-type specification(Table 1). Furthermore, gene deletion experiments have shown that all three genes are required for normal limb development and that in their absence limb characteristics are lost. Tbx5 is required for initiation of the forelimb, and interacts with the Fgf and Wnt signalling pathways. Tbx4 is required for hindlimb development, but hindlimb bud formation is initiated in the absence of Tbx4. In Pitx1 mutant embryos, the defects in hindlimb development are more subtle than those in Tbx4 mutant embryos. The limbs are smaller overall and hindlimb characteristics are lost. Furthermore,analysis of Pitx1^-/-;Pitx2^-/- and Pitx1^-/-;Pitx2^-/+double-mutant mice indicates that Pitx genes are required for normal limb outgrowth.

Tbx5 and Tbx4 are closely related T-box genes that share a high degree of sequence conservation. Both genes have fundamental roles in controlling limb initiation and continued limb outgrowth but, over time, may have acquired functional differences that contribute to the determination of limb identity. Understanding the nature of these differences is an important area of future study. As Tbx5 and Tbx4 have diverged relatively recently in evolutionary terms, it is likely that they still share some targets, but they may also have acquired some unique ones. In two urodele amphibians, newts and axolotls, the tight limb-type restricted regulation of Tbx4 and Tbx5 differs from other vertebrates. Both mRNA and protein of Tbx5 and Tbx4 are initially co-expressed in the developing forelimb and hindlimbs (Khan et al., 2002). However, during regeneration, Tbx5 is upregulated exclusively in the forelimbs while Tbx4 is upregulated only in the hindlimb. The co-expression of Tbx5 and Tbx4during urodele limb development argues that additional factors are involved in the control of limb-type identity. Indeed co-factors could play important roles in altering the regulatory specificities of Tbx5 and Tbx4 (Khan et al.,2002). This is not without precedence, as residues N-terminal to the DNA binding, T-domain have been shown to interact with a NK class of nuclear protein, Nkx2.5, expressed in the heart(Bruneau et al., 2001; Hiroi et al., 2001). Furthermore, putative dominant-negative Tbx5 constructs that lack these N-terminal residues produce deletion deformities when misexpressed in the limb (Ng et al., 2002). The identity of potential cofactors of Tbx5 and Tbx4 in the limb remains to be determined, but will explain some important aspects of how these factors may modulate limb-type identity.

Identifing the targets of Tbx5, Tbx4 and Pitx1 will help elucidate the mechanisms by which common regulatory pathways are modulated to produce distinct structures in the forelimbs and hindlimbs. Towards this goal,serial analysis of gene expression (SAGE) of forelimb and hindlimb transcripts has identified candidate target genes(Margulies et al., 2001). For example, Hox genes were identified in the SAGE screen, and may be targets of these limb identity determinants that act later during the morphogenesis of forelimb and hindlimb elements. Further work to understand the targets of Tbx5, Tbx4 and Pitx1, and the molecular basis by which limb-type morphologies are produced, is certain to enhance our understanding of the broader issue of how organisms have reused regulatory signalling cascades in diverse developmental contexts to produce novel structures.