How the embryo makes a limb: determination, polarity and identity

Introduction

The development of a limb with its rich anatomy is an embryological tour-de-force. A human arm has more than 30 bones and over 50 muscles in addition to tendons and ligaments. It is also innervated and vascularized and covered with skin with characteristic appendages such as nails. This complex structure develops from a small bud of undifferentiated mesoderm cells encased in ectoderm (Fig.1). The buds arise on both sides of the body at the appropriate levels along the antero-posterior axis; the buds which will develop into the forelimbs arise in register with the cervical/thoracic transition in the vertebral column (i.e. opposite somite numbers 15–20 in chick but 8/9–13/14 in mouse), the buds which will develop into the hindlimbs in register with the lumbar–sacral transition (i.e. opposite somite numbers 25/26–31/32 in chick and 23/24–28/29 in mouse; Burke et al. 1995). The region in between the buds where no limbs develop is known as the interlimb region.

The mesoderm of the early limb bud comes from the lateral plate mesoderm – the lateral plate mesoderm cells give rise to the limb connective tissues – cartilage, bone, tendon and muscle connective tissue. The bud mesoderm is subsequently populated by cells that migrate from the somites and give rise to the myogenic cells of the muscles (reviewed by Buckingham et al. 2003). Classical transplantation experiments showed that regions of the lateral plate mesoderm are determined to form limb buds long before any buds are visible. In contrast, the limb bud ectoderm is not determined and ectoderm from other regions of the embryo can replace the ectoderm overlying a limb-forming region and support normal limb development (reviewed by Balinsky, 1965).

The key question is how lateral plate mesoderm in the appropriate positions in the embryo becomes determined to form a limb. The position will determine the identity of the mesoderm cells that make up the limb bud and this in turn will influence how positional information is subsequently interpreted to generate the final limb anatomy. Interpretation will also be influenced by the genotype of the mesoderm cells and this will lead to the different anatomies of, say, a human arm and a mouse forelimb.

It is crucial that the limb buds not only arise in the proper place in the body but also that they have the same polarity as the body, so that the future limbs are orientated correctly. This is accomplished by the polarity of the limb buds leading to the establishment of signalling regions in the bud. These regions produce signals which provide the mesoderm cells in the limb bud as it develops with positional information and it is the interpretation of this positional information that generates the limb anatomy.

The positional signals operating in limb buds are highly conserved between forelimbs and hindlimbs and also between limb buds of different vertebrates. The dorso-ventral polarity of the limb (in our hand, from the back of the hand to the palm) is governed by signals from the ectoderm covering the dorsal and ventral sides of the limb bud. The dorsal ectoderm produces the dorsalizing signal Wnt7a (Parr & McMahon, 1995) and the ventral ectoderm bone morphogenetic proteins (BMPs; Pizette et al. 2001). Antero-posterior polarity (from thumb to little finger) is governed by the signalling molecule, Sonic hedgehog (Shh; Riddle et al. 1993), produced by the polarizing region (zone of polarizing activity, ZPA), a region of mesoderm cells at the posterior margin of the limb bud (reviewed by Tickle & Barker, 2012). Outgrowth of the limb bud from the body and specification of positional information along the proximo-distal axis of the limb (from shoulder to finger tips) depends on fibroblast growth factors (FGFs; reviewed by Martin, 1998) produced by the apical ectodermal ridge, the thickened epithelium that arises at the boundary between dorsal and ventral ectoderm and then rims the bud tip (Fig.1).

The signalling regions in a limb bud mutually maintain each other’s activity, providing a robust system that integrates pattern formation and growth in three dimensions (reviewed by Zeller et al. 2009). Therefore once the signalling regions have been established, the limb bud can develop more or less autonomously; for example, a chick wing develops normally when the bud is grafted to an ectopic site in the embryo or when placed in the coelomic cavity (Stephens et al. 1993).

Outline of how the chick wing bud arises

To understand how a limb bud arises, we need to identify the cells that will give rise to the bud in the early embryo and find out when they become specified and determined to form a limb. This has been accomplished in chick embryos by a combination of fate-maps and transplantation experiments in which prospective limb-forming regions are isolated or transplanted to other locations in the embryo. The most complete information on determination comes from work on the chick wing bud and we will therefore use this as our main example (Fig.2).

Fate maps show that the mesoderm cells that will form the wing buds have completed their gastrulation movements by the two-somite stage and are located at a specific antero-posterior position in the primitive paraxial mesoderm on either side of the midline (Fig.2, 2-somite stage; Rudnick, 1946; Chaube, 1959; Rosenquist, 1971). Shortly after, at the seven-somite stage, the prospective wing bud cells in the primitive paraxial mesoderm are expressing certain combinations of 3′ Hox genes (Strachan & Gaunt, 1994) and transplantation experiments show that the identity of these cells as wing cells and the antero-posterior polarity of the future wing bud have been determined (Fig.2; Chaube, 1959; Saito et al. 2002). The identity of the leg-forming region is specified at the same time but is not irreversibly determined until later (Saito et al. 2002).

About 20 h later, dorso-ventral patterning of the primitive paraxial mesoderm in the wing-forming region will have resulted in determination of the lateral plate mesoderm ventrally and of the cells that contribute to the presomitic mesoderm dorsally (Tonegawa et al. 1997). Lateral plate mesoderm isolated from the wing-forming region at this stage and transplanted elsewhere in the embryo has been shown to be determined to form a wing bud (Fig.2, 13-somite stage; Kieny, 1971). Then, as the adjacent presomitic mesoderm begins to segment, expression of Tbx5, which encodes a transcription factor of the T-box family, is initiated in the ‘wing’ lateral plate mesoderm (Fig.2, 19/20-somite stage). Determination of leg identity also occurs this stage and Tbx4 expression is initiated in the ‘leg’ lateral plate mesoderm (Saito et al. 2002). Between the 10- and 20-somite stages, the coelomic cavity forms and partitions the lateral plate mesoderm into two layers all along the antero-posterior axis – the upper layer, the somatopleure associated with the ectoderm, and the lower layer, the splanchnopleure associated with the endoderm (Funayama et al. 1999).

At the 19/20-somite stage, fate maps show that the wing mesoderm comes from the medial (future dorsal) region of the somatopleure and that the ectoderm overlying this region gives rise to the apical ectodermal ridge. The dorsal ectoderm of the future wing bud comes from the ectoderm overlying the somites and the ventral ectoderm from the ectoderm overlying the lateral (future ventral) somatopleure (Fig.2; 19/20-somite stage; Altabef et al. 1997; Michaud et al. 1997). The chick leg bud ectoderm has similar origins (Altabef et al. 1997). The dorso-ventral polarity of the future wing bud is then determined at around stage 15 (Geduspan & Solursh, 1992; Michaud et al. 1997) by which time mesoderm cells have started to accumulate in the wing-forming region to produce a thickening which will form the bud. Wnt7a is now expressed in the dorsal ectoderm and Bmps in the ventral ectoderm.

At the early stages described above, the chick embryo resembles a bearskin rug, with the somites developing on either side of the midline of the trunk flanked by the lateral plate mesoderm. The embryo then undergoes lateral folding, with the somites coming to lie dorsally and the edges of the lateral plate mesoderm eventually fusing in the ventral midline of the body; this folding brings the thickened mesoderm, which will form the wing buds to the sides of the body (Fig.2, stage 16). By stage 16, Fgf8 is expressed in ectoderm cells of the wing bud which will form the apical ectodermal ridge (Mahmood et al. 1995; Crossley et al. 1996). Shh expression in the mesoderm at the posterior margin is detected a few hours later at stage 17 when the wing bud first becomes clearly visible (Fig.1; Riddle et al. 1993). Similarly, in developing leg buds, Fgf8 is expressed in the presumptive apical ectodermal ridge just prior to Shh being expressed in the polarizing region.

Determination of limb identity

The experimental studies of chick wing development outlined above revealed that wing and leg identity is established very early. At this time, the prospective limb cells have undergone gastrulation and are expressing Hox genes. It has been suspected for a long time that limb identity, like vertebral identity, is determined by Hox genes. Hox gene expression is activated in the primitive paraxial mesoderm cells under the influence of FGF signalling during gastrulation. The expression domains then also spread anteriorly so that mesoderm cells at different positions along the antero-posterior axis of the body express different combinations of Hox genes, with anterior cells expressing 3′ Hox genes and posterior cells expressing 5′ Hox genes (Strachan & Gaunt, 1994; Iimura & Pourquié, 2007). It should be noted in passing that it seems likely that interference with the establishment of Hox gene expression domains explains the duplications of the posterior part of the body (complete with hindlimbs) and additional pectoral fins in zebrafish induced when embryos at gastrula stages are treated with retinoic acid (Niederreither et al. 1996; Vandersea et al. 1998; Grandel & Brand, 2011).

Recent work in the mouse has now provided evidence that the products of Hox4 and Hox5 paralogs directly regulate Tbx5 expression in the lateral plate mesoderm in the forelimb-forming region (Fig.3; Minguillon et al. 2012). As we discussed earlier, in the chick, the initiation of Tbx5 expression occurs shortly after the determination of the lateral plate mesoderm to develop into a wing (Fig.2). It is unlikely, however, even though the Hox 4 and 5 paralogous genes may determine wing/forelimb identity that they themselves are involved in interpreting positional information in the developing limb. Hoxc4 and Hoxc5 are expressed specifically in wing buds in chick embryos but their expression is restricted to anterior and proximal regions (Nelson et al. 1996). Tbx5 also does not appear to be involved in interpreting wing/forelimb identity. In genetic experiments in mice in which Tbx4 is expressed in place of Tbx5, a forelimb still develops (Minguillon et al. 2005).

Given the preceding discussion, it might be predicted that a particular combination of more 5′ Hox genes would initiate Tbx4 expression in the leg-forming region of the lateral plate mesoderm. However, so far, there is no evidence that Hox genes are involved and Tbx4 expression in the leg/hindlimb-forming region is regulated by the homeodomain transcription factor Pitx1 (Logan & Tabin, 1999; Szeto et al. 1999), probably in co-operation with the closely related Pitx2 transcription factor (Marcil et al. 2003). Pitx1 is expressed in the lateral plate mesoderm in the posterior region of the body and is a major determinant of hindlimb identity. When Pitx1 is expressed in the forelimb region in transgenic mouse embryos, the forelimbs acquire hindlimb characteristics (Minguillon et al. 2005). The transcription factor Islet1 is also expressed in the same region of the lateral plate mesoderm and is required for Tbx4 expression and formation of the hindlimb bud (Kawakami et al. 2011; Narkis et al. 2012).

Determination of limb antero-posterior polarity

According to the classical embryological experiments, antero-posterior polarity of the limb buds is also determined very early at the time that the pattern of Hox gene expression along the antero-posterior axis of the body is being established in the primitive paraxial mesoderm. There is now evidence that Hox genes are involved in establishing antero-posterior polarity in the wing/forelimbs. Sophisticated genetic experiments in which mouse embryos were produced that lacked all Hox5 function showed that their activity is required for the development of the anterior region of the forelimb bud (Xu et al. 2013). Similar experiments in which the function of all the Hox9 paralogous genes was inactivated showed that Hox9 activity is required for the development of the posterior region of the forelimb bud (Xu & Wellik, 2011). It should be noted that, even though these Hox genes are expressed very early in the chick at the time when antero-polarity is determined (Fig.2, 7-somite stage), a polarizing region expressing Shh at the posterior margin of the wing bud does not develop until much later in development (Fig.1, stage 17; see later).

In contrast, there is no evidence that Hox genes are involved in determining antero-posterior polarity of the legs/hindlimbs. The transcription factor Islet1 involved in regulating Tbx4 expression also fulfills, in the hindlimb bud, the same function as Hox9 paralogs in the forelimb buds in determining antero-polarity (see later).

The primitive paraxial mesoderm gives rise to both the lateral halves of the somites and the lateral plate mesoderm. Therefore the particular combination of Hox genes expressed in the primitive paraxial mesoderm cells at any given antero-posterior level could determine the subsequent fate of both somites and lateral plate and lead to wings/forelimbs, for example, developing opposite the somites that form the vertebrae in the cervical/thoracic transition region (Burke et al. 1995). As predicted, deletions of the Hox5 paralogs not only affect forelimb development but also produce pattern changes in the vertebral column in the cervical/thoracic transition region (McIntyre et al. 2007). However, although deletions of all Hox10 or all Hox11 paralogous genes lead to pattern changes in the lumbar/sacral transition region of the vertebral column, the formation of the hindlimb buds is unaffected (Wellik & Capecchi, 2003).

Determination of lateral plate mesoderm to form a limb

The region of mesoderm that will give rise to the limbs must be determined with respect not only to the antero-posterior axis of the body but also to its dorso-ventral axis. The classical transplantation experiments in the chick showed that the lateral plate mesoderm of the wing-forming region is determined to form a wing just prior to the segmentation of the presomitic mesoderm (timeline in Fig.2; Kieny, 1971). Before this stage, development of a wing only results if medial (dorsal) presomitic mesoderm is included in the transplants from the wing-forming level (Kieny, 1971; Noro et al. 2011). This has been interpreted as being due to the presomitic mesoderm producing a limb-induction signal. It is now known that lateral plate mesoderm is specified and maintained by high levels of signalling by BMP4, a member of the transforming growth factor (TGF)-β superfamily, (Tonegawa et al. 1997), whereas presomitic mesoderm is specified and maintained by low levels of BMP4 signalling as a result of the production of the BMP antagonist, Noggin (Tonegawa & Takahashi, 1998). Fate maps show that it is the medial region of the lateral plate that gives rise to the wing (Fig.2; 19/20-somite stage) and determination of this further subdivision of the lateral plate mesoderm might also be specified by the level of BMP signalling. The interaction between presomitic mesoderm and lateral plate mesoderm could explain why presomitic mesoderm is required for development of a wing at early stages and suggests that Noggin is the inductive signal required for wing determination. It seems likely that the same mechanisms control dorso-ventral patterning of the mesoderm of the entire trunk and therefore also determine the region of mesoderm along the dorso-ventral axis that will develop into legs/hindlimbs.

It should be noted that the limbs arise from the somatopleural layer of the lateral plate mesoderm, not the splanchnopleural layer. The somatopleure is specified by ectodermal signals whose identity is unknown (Funayama et al. 1999). However, the splanchnopleure in the limb-forming regions from 19/20-somite stage 15 chick embryos can form a limb when transplanted under the ectoderm (Yonei-Tamura et al. 2005) with the splanchnopleure from the wing-forming level forming wings and the splanchnopleure from the leg-forming level, legs. This seems likely to be due to the mesoderm cells of both somatopleure and splanchopleure expressing the same combination of genes that determine limb identity.

The molecular basis of limb determination

As we have discussed, the determination of the lateral plate mesoderm to form a wing occurs around the time that Tbx5 expression is initiated (Fig.2). Similarly Tbx4 is expressed in the region of the lateral plate mesoderm at the time it is determined to form the leg (Saito et al. 2002). Tbx5 function is essential for the formation of the forelimb in mouse embryos and the pectoral fin in zebrafish, whereas Tbx4 function is essential for the formation of the hindlimb in mouse embryos (reviewed Duboc & Logan, 2011a). Furthermore, forced expression of Tbx5 or Tbx4 in cells in the interlimb of early chick embryos has been reported to induce the development of wing-like or leg-like structures, respectively (Takeuchi et al. 2003). It seems likely that other transcription factors are required for limbs to be induced; these could be expressed in the interlimb, which has the potential to form a limb (see later), or their expression could be induced by Tbx5 or Tbx4. The identity of the transcription factors that act in combination with Tbx5 or with Tbx4 in determining ‘limbness’ is at present unknown. The transcriptome of mouse limb buds at different stages in development has been analysed (Gyurján et al. 2011; Taher et al. 2011) but these studies did not include analysis of the regions of the lateral plate mesoderm at the time of limb determination.

Retinoic acid signalling and determination of wing/forelimb

The expression of Tbx5 in the lateral plate mesoderm downstream of Hox genes and the subsequent development of a forelimb bud requires retinoic acid signalling. In contrast, retinoic acid signalling does not appear to be required for formation of the hindlimbs.

Genes encoding retinoic acid-generating enzymes are expressed in the mesoderm which will form wings, forelimbs and pectoral fins of chick, mouse and zebrafish embryos, respectively, just before Tbx5 begins to be expressed and the adjacent somites start to form (Swindell et al. 1999; Grandel et al. 2002; Zhao et al. 2009). In Raldh2^−/− and Raldh3^−/− mouse embryos, Tbx5 expression is not initiated and forelimb buds fail to form (Niederreither et al. 2002; Zhao et al. 2009). Likewise, zebrafish with mutations in Raldh2 lack pectoral fins (Grandel et al. 2002).

Experiments in which forelimb development is rescued in Raldh2^−/− mouse embryos by treating the pregnant mice carrying them with retinoic acid show that signalling is required at the time that Tbx5 expression is initiated. Surprisingly, however, recent work suggests that retinoic acid signalling is involved in local interactions between the forelimb-forming region and neighbouring tissues along the antero-posterior axis. Retinoic acid produced by forelimb lateral plate mesoderm appears to signal in a paracrine fashion (Zhao et al. 2009), inhibiting Fgf8 expression in neighbouring tissues, thus allowing Tbx5 to be expressed in the forelimb-forming region (Fig.3; Cunningham et al. 2013). Once somites form, they also produce retinoic acid which could act in paracrine fashion to help to initiate Tbx5 expression (Gibert et al. 2006).

Retinoic acid signalling by the lateral plate mesoderm determined to become a wing/forelimb bud is only transient. Expression of Raldh2 in the wing lateral plate mesoderm in chick embryos (Swindell et al. 1999), for example, decreases as the buds start to form. Indeed, retinoic acid signalling is incompatible with subsequent limb bud formation. It has been shown to prevent maintenance of the apical ectodermal ridge in chick wing buds (Tickle et al. 1989), and persistent retinoic acid signalling in developing mouse limb buds leads to limb defects (Yashiro et al. 2004). The transience of retinoic acid signalling suggests that it might be involved in a timing mechanism. Retinoic acid could serve to synchronize the determination of the lateral plate mesoderm to form wing/forelimb, downstream of Hox gene activity, and the segmentation of the adjacent presomitic mesoderm which gives rise to the vertebrae in the cervical/thoracic transition region. Such a synchronization mechanism may be required because forelimbs in the mouse arise opposite somites 8/9–13/14, but much later in the chick, opposite somites 15–20. In contrast, the determination of the lateral plate to form a leg/hindlimb does not appear to be synchronized with the segmentation of the adjacent presomitic mesoderm and the leg/hindlimbs develop opposite the same numbered somites in mouse and chick. Retinoic acid has also been implicated in regulating a timing mechanism that controls the duration of Shh expression in developing chick limb buds (Chinnaiya et al. 2014), suggesting that a general function of retinoic acid signalling is to time developmental events.

Formation of the bud

It has been known for some time that a major function of Tbx5 and Tbx4 is to control bud formation and that they initiate essentially the same signalling cascade in the lateral plate mesoderm in both wing/forelimb and leg/hindlimb regions, respectively (Fig.4). FGF10 is produced that induces Wnt signalling in the overlying ectoderm, leading to formation of the apical ectodermal ridge expressing Fgf8, and FGF8, in turn, maintains Fgf10 expression in the mesoderm. This FGF8-FGF10 feedback loop controls bud outgrowth (Ohuchi et al. 1997; Kengaku et al. 1998; Barrow et al. 2003; for details see Fig.4). Different Wnt ligands are employed downstream of Tbx4 and Tbx5 in the mesoderm. In addition, although the Twist transcription factor is involved in regulating mesodermal signalling in both forelimbs and hindlimbs in mouse embryos (Krawchuk et al. 2010), the Sall4 transcription factor, whose gene expression is regulated by Tbx5, is involved in mouse forelimbs (Koshiba-Takeuchi et al. 2006) and in pectoral fin buds in zebrafish (Harvey & Logan, 2006).

It has recently been demonstrated that Tbx5 is also involved in controlling a morphogenetic process that recruits mesoderm cells to wing/forelimb bud (Gros & Tabin, 2014; Fig.4). Classical work on limb development in amphibian embryos described mesenchyme cells leaving the coelomic epithelium of the somatopleure and accumulating in the limb-forming regions (reviewed by Balinsky, 1965). Analysis in chick embryos has shown that, similarly, cells in the coelomic epithelium in the wing-forming somatopleure undergo an epithelial-mesenchymal transition, dependent on Tbx5 expression, with the mesenchyme cells then making a substantial contribution to the bud mesoderm. Snail1, which encodes a transcription factor known to regulate epithelial–mesenchymal transitions in many regions of the embryo (reviewed by Thiery et al. 2009), is expressed in the limb-forming regions in chick and mouse embryos (Isaac et al. 2000; Agarwal et al. 2003). In mouse embryos, Snail1 expression in the presumptive forelimb depends on Tbx5 function (Agarwal et al. 2003) and it seems likely that Snail1 regulates the epithelial–mesenchymal transition of the cells in the coelomic epithelium in the limb-forming regions. There is growing evidence of the importance of mesenchymal cells derived from the coelomic epithelium, in development of other organs, such as the heart and the gut (von Gise & Pu, 2012; Carmona et al. 2013).

Determination of limb dorso-ventral polarity and establishment of the dorsal and ventral ectoderm signalling regions

The dorso-ventral polarity of the future wing bud is determined after the somites have formed opposite the wing-forming region and has been shown to involve interactions with the neighbouring tissues along the dorso-ventral axis – the somites dorsally and the lateral plate (prospective ventral body wall) ventrally (Kieny, 1971; Michaud et al. 1997). The dorso-ventral polarity of the mesoderm is then imposed on the overlying ectoderm, thus ensuring that the dorso-ventral polarity of the wing bud matches the dorso-ventral polarity of the body. As at earlier stages when the dorso-ventral pattern of the primitive paraxial mesoderm is determined, BMP signalling is likely involved. High levels of BMP signalling in prospective ventral body wall mesoderm could act as the ventralizing signal inducing and/or maintaining Bmp expression in the ectoderm that will form the ventral ectoderm of the future wing bud (Pizette et al. 2001). Noggin produced by the somites could act as a dorsalizing signal attenuating BMP signalling in the overlying ectoderm, and thus permit Wnt7a expression in the dorsal ectoderm of the future wing bud.

High levels of BMP signalling induce expression of the gene for the transcription factor Engrailed1 not only in the ventral ectoderm of the future wing buds but also throughout the entire ventral ectoderm of the trunk, including the ventral ectoderm of the future leg buds (Logan et al. 1997). In the ventral ectoderm of the future chick wing bud, Engrailed1 has been shown to maintain its identity as ventral ectoderm, thus preventing it from expressing Wnt7a (Altabef et al. 2000).

There is a sharp boundary in En1 expression and this reflects the fact that the trunk ectoderm consists dorsal and ventral cell lineage-restricted compartments (Altabef et al. 1997). Similar dorso-ventral lineage-restricted compartments have also been found in the ectoderm in the mouse (Kimmel et al. 2000) This dorso-ventral compartmentalization of the trunk ectoderm is not only fundamental to the establishment of the dorsal and ventral ectoderm signalling regions of the future limb bud but also the apical ectodermal ridge. The apical ectodermal ridge arises at the dorso-ventral compartment boundary, thus ensuring that the limb buds will grow out of the sides of the body.

Other features of the body appear to be positioned along the dorso-ventral axis by a similar mechanism. There is evidence that both the dorso-ventral pattern of coat colour in ‘black and tan’ mice (Candille et al. 2004), which have the same markings as Doberman dogs, and the dorso-ventral position in which mammary glands develop, along the so-called mammary line, may be controlled by high levels of BMP signalling ventrally (Cho et al. 2006). It is not clear whether the same dorso-ventral boundary that positions the limbs, displaced over time, also determines the coat colour pattern and the development of the mammary glands or whether separate boundaries are involved (Cho et al. 2006).

Establishment of the apical ectodermal ridge

The apical ectodermal ridge expressing Fgf8 forms at the dorso-ventral compartment boundary in the limb-forming regions (Altabef et al. 1997) and later comes to rim the tip of the limb bud (Fig.1). The induction of Fgf8 expression in the pre-ridge cells involves both BMP signalling in the ectoderm and Wnt signalling induced by mesodermal FGF10 signalling (Ahn et al. 2001; Barrow et al. 2003). It is not clear, however, whether Wnt signalling is upstream or downstream of Bmp expression in the ventral ectoderm (see conflicting data in Barrow et al. 2003 and Soshnikova et al. 2003). More details have emerged recently of how the transcription factors Sp6 and Sp8 participate in the process of inducing Fgf8 expression and are also involved in the induction of En1 expression in ventral ectoderm (Haro et al. 2014).

Fate-mapping experiments revealed that the cells that give rise to the apical ectodermal ridge are initially distributed more widely in the ectoderm in a salt and pepper arrangement, in both dorsal and ventral compartments in the chick (Altabef et al. 1997) and in the ventral compartment in the mouse (Kimmel et al. 2000).They then assemble at the dorso-ventral boundary. The transcription factor Arid3b is necessary for the movement of pre-ridge cells (Casanova et al. 2011).

The mature apical ectodermal ridge consists of a pseudostratified epithelium of tightly packed elongated cells linked by extensive gap junctions. p63 is involved in its maintenance as in other pseudostratified epithelia in the embryo and the adult body (Mills et al. 1999; Yang et al. 1999). The morphology of the apical ectodermal ridge is essential for its crucial, but often overlooked, mechanical function of shaping the limb bud. In its absence, the bud becomes bulbous and is no longer flattened dorso-ventrally (Tickle et al. 1989). A second cell lineage-restricted compartment boundary has been detected at the dorsal margin of the apical ectodermal ridge in the mouse limb bud and this has been suggested to help compress the assembled pre-ridge cells, causing the increase in cell packing and cell elongation (Kimmel et al. 2000).

Establishment of the polarizing region

The antero-posterior polarity of the limb is controlled by the polarizing region expressing Shh which forms at the posterior margin of the bud (Fig.1). As already mentioned, studies of the genetic loss of function in the mouse have shown that the antero-posterior polarity of the buds that will form the mouse forelimbs is determined by the activity of Hox9 and Hox5 paralogs. In the absence of all Hox9 activity in mutant mice, Shh expression is not initiated in the posterior region of the forelimb buds (Xu & Wellik, 2011), whereas in the absence of all Hox5 activity, Shh expression is not restricted to their posterior margin (Xu et al. 2013). In Hox9 mutant mice, a key transcription factor, Hand2, that interacts with the DNA cis-regulatory sequence that controls Shh expression in the polarizing region (ZRS, zone of polarizing activity regulatory sequence; Lettice et al. 2002; reviewed Hill, 2007) is not expressed at the posterior of the forelimb buds, although it is expressed as normal in the posterior of the hindlimb buds, where it is regulated by Islet1 (Itou et al. 2012). The ZRS integrates both positive and negative inputs including inputs regulated by FGF signalling from the apical ectodermal ridge (Mao et al. 2009; Zhang et al. 2009; Lettice et al. 2012; see details Fig.5). This involvement of FGF signalling in regulating Shh expression in the polarizing region may account for the delay between determination of antero-posterior polarity and the development of the polarizing region, with the polarizing region not developing until after establishment of the apical ectodermal ridge.

A complex regulatory network has been discovered that ensures that not only is Hand2 expressed in the posterior region of limb buds but also its expression is actively repressed in the anterior region of the limb buds, initially due to the transcriptional repressor Gli3 (reviewed by Panman & Zeller, 2003; Osterwalder et al. 2014). In the anterior region of the forelimb bud, the products of the Hox5 paralogs co-operate with the transcription factor PLZF (promyelocytic leucaemia zinc finger protein; Barna et al. 2000) to provide another layer of regulation that maintains the antero-posterior polarity; this reinforces the restriction of Hand2 expression and hence Shh expression to the posterior margin (Fig.5). PLZF is not involved in maintaining antero-posterior polarity in the hindlimb buds.

There are indications from work carried out in the 1980’s and 90’s in chick embryos that retinoic acid signalling is involved in determining antero-posterior polarity of the limb buds and establishing a polarizing region expressing Shh. Application of retinoic acid to the anterior margin of a chick wing bud can mimic the effects of a polarizing region graft (Tickle et al. 1982) by inducing Shh expression in anterior wing bud cells (Riddle et al. 1993), and treatment of chick embryos with inhibitors of retinoic acid signalling prevents Shh expression (Stratford et al. 1996). In chick embryos, Raldh2 expression persists in the interlimb lateral plate mesoderm after it has been lost in the wing buds (Swindell et al. 1999) and retinoic acid could diffuse into the posterior region of the wing bud from the interlimb. One possibility is that retinoic acid signalling could help to time the activity of Hox genes and thus ensure that Shh is expressed as the forelimb buds start to form (Fig.5). It is not clear, however, whether these results reflect a role for endogenous retinoic acid signalling in establishing the polarizing region. Shh expression is not restricted posteriorly in rescued Raldh2^−/− forelimb buds, although the basis for this is not clear (Niederreither et al. 2002; Zhao et al. 2009). It should be noted that retinoic acid can also polarize chick leg buds (Wilde et al. 1987).

Limb identity and interpretation of positional information

There is good evidence that leg/hindlimb identity, related to the antero-posterior position in which the bud forms, is determined by the Pitx1 transcription factor (Logan & Tabin, 1999; Szeto et al. 1999; Minguillon et al. 2005; reviewed by Duboc & Logan, 2011a). Pitx1 continues to be expressed in the hindlimb buds as they develop and, importantly, has been shown in transgenic mice to influence the interpretation of positional information that realizes specific hindlimb anatomy (Duboc & Logan, 2011b). Pitx1 directly regulates Tbx4 expression and genome-wide analysis of binding sites in the DNA occupied by Pitx1 in cells in developing hindlimbs has identified other potential target genes (Infante et al. 2013). Many of these genes are expressed in forelimbs as well as hindlimbs. Indeed, transcriptome analyses and genome-wide chromatin profiling for gene activity have identified relatively few genes differentially expressed in either forelimbs or hindlimbs (Margulies et al. 2001; Taher et al. 2011; Cotney et al. 2012). Therefore the interpretation of positional information mainly seems to involve subtle differences in gene regulation governing the levels, precise spatial patterns and timing of expression of the same sets of genes in forelimbs and hindlimbs. Comparisons of global gene activity in the cells of limb buds of different vertebrates including human limb buds (Cotney et al. 2013) may provide a way of beginning to address how genotype influences interpretation.

Potential for limb formation

It has been known for a long time that more extensive regions of the embryo are competent to form a limb but normally do not do so. This widespread competence seems to be due to the main regions of the embryo being sketched out in rough at early stages, with the precise regions that will form the limbs being refined later by local interactions. Such interactions have already been noted, for example, in the determination of the lateral plate mesoderm to form a forelimb. There may also be inhibitory mechanisms that prevent limbs developing in inappropriate places.

The competence to form limbs extends all along the antero-posterior axis of the trunk, including the interlimb, and corresponds with the Wolffian ridges described by embryologists; the neck is not competent to form limbs (Lours & Dietrich, 2005). In the 1920s, Balinsky discovered that an additional limb can be induced by grafting a nasal rudiment to the interlimb (Fig.6A; reviewed by Balinsky, 1965) and, more recently, additional limbs have been induced from the interlimb in chick embryos by applying a single signalling molecule, either an FGF (Fig.6B; Cohn et al. 1995) or a Wnt ligand (Kawakami et al. 2001). As we have seen, FGFs and Wnts play central roles in formation of a limb bud. Snail1 expression is induced very rapidly in the interlimb by FGF (Isaac et al. 2000) and could regulate recruitment of cells from the coelomic epithelium in this region. Hox gene activity is also implicated in the formation of the additional limbs, as wings or legs arise according to antero-posterior position in the interlimb.

The additional limbs induced in the interlimb have a curious inverted antero-posterior polarity (Fig.6A,B). This polarity corresponds with the greater competence of the anterior region of the interlimb to form a polarizing region (Tanaka et al. 2000) and changes in Hox9 gene expression have been seen following FGF application to the interlimb (Cohn et al. 1997). It has also emerged that the products of caudal Hox genes from more than one paralogous group actively repress Tbx5 expression in the interlimb region (Nishimoto et al. 2014). A speculation is that the persistent high levels of retinoic acid signalling in the interlimb contribute to this inhibitory mechanism and that applying FGF antagonizes retinoic acid signalling, thereby allowing a limb to develop.

The dorso-ventral polarity of additional limbs induced from the interlimb is normal and an apical ectodermal ridge expressing Fgf8 forms at the boundary between the dorsal and ventral ectoderm compartments, in line with the apical ectodermal ridges of the normal limb buds. Surprisingly, when FGF is applied to the interlimb in chick embryos, Fgf8 expression is also induced in the ectoderm at the dorsal midline of the body. It has been suggested that the ability of the ectoderm in this region to form an apical ectodermal ridge reflects a mechanism that evolved in ancestral vertebrates to make a median dorsal fin bud (Yonei-Tamura et al. 1999). In present-day sharks and lampreys, the mesoderm of the median fin originates from somitic paraxial mesoderm (Freitas et al. 2006). Interestingly, additional limbs arising dorsal to the forelimbs have been observed on rare occasions in domesticated animals (Fig.6C; Alam et al. 2007).

Some recent experiments with Xenopus tadpoles have shown that the ventral lateral plate mesoderm which would normally form the body wall can form a limb when BMP signalling is transiently inhibited (Christen et al. 2012). The additional limb arises at the same antero-posterior position as the normal limb, but in a more ventral position, and has normal dorso-ventral polarity, whereas the limb forming in the normal position is ‘double-dorsal’. These intriguing observations can be explained by the dorso-ventral compartment boundary being shifted from its original position to a second, more ventral position by the reduction in BMP signalling, and an additional apical ectodermal ridge forming at the new boundary. The accompanying ventral shift in En1 expression would result in expansion of Wnt7a expression into the ventral ectoderm of the normal limb bud, thus exposing the underlying mesoderm to dorsalizing signals from both sides. Additional limbs have also been reported to develop ventrally in mouse embryos when the genes for Myostatin and GDF11, two highly related members of the TGF-β superfamily, are knocked-out together (McPherron et al. 2009).

Concluding remarks

Recent work has begun to provide a much clearer picture of how limbs arise in the embryo; in particular, the molecules involved in the determination, polarity and identity. Hox gene activity along the antero-posterior axis of the body axis can now be firmly linked to determining the region of the embryo that will form the wings/forelimbs and their antero-posterior polarity. Rather unexpectedly, the anterior and posterior regions of the wing/forelimb bud are determined by different Hox genes. This is consistent with the provocative suggestion that the vertebrate limb bud may be composed of anterior and posterior compartments analogous to the developing insect wing (Towers et al. 2011). Hox gene activity also determines the interlimb region by repressing the formation of forelimbs in this region. In contrast, there is no evidence that Hox genes are involved in determining the regions that will form the legs/hindlimbs. If Hox genes are not involved, what is the mechanism that ensures that they arise opposite the lumbar–sacral transition in the vertebral column? Hox genes also do not appear to determine the antero-polarity of the leg/hindlimb buds, and Islet1, which is not involved in axial patterning, seems to fulfill the same function as the products of the Hox9 paralogs in the forelimb. This raises the question of how the antero-polarity of the leg bud is aligned with the antero-posterior polarity of the body.

There are also outstanding questions about the molecular basis of limb identity which provides the context for the interpretation of positional information and leads to different limb anatomies. Although it is now well-established that Pitx1 is a major determinant of hindlimb identity, a transcription factor with the same function has not yet been identified in the wing/forelimb. It is encouraging that the studies on Pitx1 are now providing insights into the basis for the interpretation of positional information, which has long been a mystery. Similar studies may also help in tackling the even more challenging problem of how species differences in anatomy arise.

Finally, despite these advances and uncovering key signals involved in regulating determination, polarity and identity of the limb buds including FGFs, retinoic acid, BMPs, Wnts and Shh, the fundamental question of the molecular basis of ‘limbness’ is still not resolved. What are the molecules expressed in the lateral plate mesoderm at precise positions in the embryo that endow those cells with the ability to form a limb? Tbx4 and Tbx5 are clearly crucial molecules but what other molecules are involved? It is also increasingly apparent that the Tbx transcription factors have a major function in the morphogenesis of the lateral plate to form a limb bud.

In conclusion, several ke y questions remain to be answered as well as many details to be filled in. It could be argued that it is not necessary to know all the details to understand how the embryo makes a limb. But details may be needed for practical applications. For example, we might wish to make limb mesoderm cells for transplantation to repair defective or damaged limbs, or to devise treatments to stimulate limb tissues surrounding the defects to make good the damage. A more complete understanding of the molecular basis of limb determination, polarity and identity and the definition of the precise sequence of signals involved could help to achieve these goals. There is still a lot to do!