Bone morphogenetic protein signaling in limb outgrowth and patterning




Bone morphogenetic proteins (BMPs) are multifunctional growth factors belonging to the transforming growth factor beta (TGFβ) multigene family. Current evidence indicates that they may play different and even antagonistic roles at different stages of limb development. Refined studies of their function in these processes have been impeded in the mouse due to the early lethality of null mutants for several BMP ligands and their receptors. Recently, however, these questions have benefited from the very powerful Cre-loxP technology. In this review, I intend to summarize what has been learned from this conditional mutagenesis approach in the mouse limb, focusing on Bmp2, Bmp4 and Bmp7 while restricting my analysis to the initial phases of limb formation and patterning. Two major aspects are discussed, the role of BMPs in dorsal-ventral polarization of the limb bud, together with their relation to apical ectodermal ridge (AER) induction, and their role in controlling digit number and identity. Particular attention is paid to the methodology, its power and its limits.


Bone morphogenetic proteins (BMPs) are multifunctional growth factors that belong to the transforming growth factor beta (TGFβ) multigene family (reviewed in Hogan 1996; Chen et al. 2004). In the vertebrate limb bud, the expression of several members of the Bmp gene family has been documented, namely Bmp2, Bmp4 and Bmp7 (Figs 1 and 2; Geetha-Loganathan et al. 2006). Current evidence indicates that Bmp2, Bmp4 and Bmp7 are the main source of BMP signaling in the developing limb. They are structurally related genes in the large BMP-encoding gene family. BMP2 and BMP4 are the closest homologs to Drosophila DPP. BMP7 belongs to another BMP subgroup (Hogan 1996; Ducy & Karsenty 2000), but it heterodimerizes with BMP2 and BMP4, and current evidence shows that heterodimers provide a stronger signal than homodimers in Xenopus and Zebrafish development (Nishimatsu & Thomsen 1998; Schmid et al. 2000). The three genes have redundant function in the Xenopus early embryo (Reversade et al. 2005) and play prominent roles in tetrapod limb development. They have been implicated in apical ectodermal ridge (AER) formation (Ahn et al. 2001; Pizette et al. 2001), AER regression (Pizette & Niswander 1999), cartilage and bone differentiation (reviewed in Karsenty & Wagner 2002; Tsumaki & Yoshikawa 2005) and interdigital webbing regression (reviewed in Zuzarte-Luis & Hurle 2005). The picture that emerges is that BMPs may play different and even antagonistic roles at different stages of limb development, and that a dedicated dissection of their function in the ectoderm and mesoderm is required, stage by stage, to decipher these multiple roles.

Figure 1.

Figure 1.

In the mouse as in the chick, Bmp genes are expressed in the ectoderm at the proper time for dorsal-ventral (DV) specification and apical ectodermal ridge (AER) induction. In the mouse, Bmp4 is expressed at E9.75 (A) in the ectoderm and the mesoderm at the level of the hindlimb bud, in a ventral domain that extends dorsally up to, but excluding the somites. At E10.0 (D), its expression domain is restricted to the presumptive AER where it overlaps that of Fgf8 (E). Bmp7 (B,F) and Msx2 (C,G) are expressed predominantly in the ventral domain of the ectoderm (reproduced from Ahn et al. [2001] with permission of the Company of Biologists). Similarly, in the chick, at stage 17 (32 somites), Bmp4 (H), Bmp7 (I), Msx1 (L) and Msx2 (M) are all expressed in the ventral ectoderm, identified by the expression of En1 (J). Their expression extends dorsally up to the level of the Fgf8-expressing AER (K) (reproduced from Pizette et al. [2001] with permission of the Company of Biologists). Noticeably, Bmp4 (A,H) and Msx1 (L) are expressed in the mesoderm in a proper pattern to transfer DV identity to the ectoderm.

Figure 2.

Figure 2.

Bmp2, Bmp4 and Bmp7 are expressed in the apical ectodermal ridge (AER) and in the mesenchyme of the mouse limb bud. In the mesoderm, at E11.5, Bmp2 is expressed in a posterior domain that overlaps with Shh (arrowhead) and to a lesser extent, along the AER posteriorly. Bmp4 is expressed along the whole AER, with two broader domains, anteriorly and posteriorly (arrowheads). Bmp7 is expressed ubiquitously in the limb mesenchyme. At E12.5, Bmp2 expression remains prominent posteriorly and in the interdigital webbing more anteriorly. Bmp4 is expressed along the whole AER, with higher levels at the posterior and anterior domains (arrowheads). Bmp7 expression is nearly ubiquitous. The anterior is to the left. (Original data from Y. Lallemand and B. Robert.)

In the mouse, the study of BMPs has been impeded because of the early lethality of null mutants for Bmp2 and Bmp4, as well as their receptors, Bmpr1 and Bmpr2 (Mishina et al. 1995; Winnier et al. 1995; Zhang & Bradley 1996; Beppu et al. 2000). In addition, these BMPs share functional similarities and may display functional redundancy. As a result, the role of one particular Bmp gene may be hidden by the activity of others. For example, although the mutation of Bmp7 by itself has little influence on limb development, it does enhance the Bmpr1b mutant phenotype (Yi et al. 2000).

With the advent of the Cre-loxP technology (reviewed in Nagy 2000), and the design of tamoxifen-inducible Cre proteins (reviewed in Metzger & Chambon 2001), it has become possible to mutate genes in a tissue- and stage-specific manner in the mouse. Thus, transgenes have been designed that express Cre specifically in the limb ectoderm (Brn4-Cre: Ahn et al. [2001]; Msx2-Cre: Sun et al. [2000]; RAR beta2-Cre: Moon & Capecchi [2000]; En1-Cre: Kimmel et al. [2000]) or mesoderm (Prx1-Cre: Logan et al. [2002]). In parallel, floxed alleles of BMP ligand and receptor genes have also been created (this review). These have been recently applied to the analysis of BMP signaling in limb formation, and more specifically of the role of Bmp2, Bmp4 and Bmp7 in this process. In this review, I intend to summarize what has been learned from this conditional mutagenesis approach in the mouse limb. I shall therefore focus on Bmp2, Bmp4 and Bmp7. However, it should be kept in mind that other members of the Bmp gene family are expressed in the limb bud and may have a role in limb patterning. Beside the Bmp2/4 and the Bmp5/6/7/8 subgroups, a third Bmp subgroup, composed of growth/differentiation factors (Gdf) 5/6/7, has been described (Ducy & Karsenty 2000). Gdf5 is expressed at sites of joint formation (Storm & Kingsley 1996). Gdf5 mutations lead to brachypodism in mice and brachydactyly in humans (Storm et al. 1994; Seemann et al. 2005). These mutations affect length and number of bones in the digits, wrist and ankle, due to defects in joint formation that lead to fusion between bones and, as a result, bone number reduction (Storm & Kingsley 1996). Gdf6 is also expressed in forming joints, but only at the level of the elbow/knee and wrist/ankle, with little expression in digits. Accordingly, the Gdf6 mutation in the mouse leads to joint fusions that are different from those seen in Gdf5 mutants (Settle et al. 2003). However, Gdf5 has been shown to be required for cartilage development in addition to joint formation (Storm & Kingsley 1999). In the double Gdf5 : Gdf6 mutants, a number of additional bones are lost, further suggesting that these genes are required for skeletal element formation and may therefore play a role in limb patterning (Settle et al. 2003). Interactions between BMP2, 4 and 7 and these other BMPs have been little studied. The functions of Bmp4 and Gdf5 in chondrogenesis were thoroughly compared, but no attempt was made to look for gene interactions (Hatakeyama et al. 2004). Furthermore, no conditional mutation for these Gdf genes or their receptors has been used yet to investigate their role in limb development. Therefore, we may have only a partial view of BMP signaling in limb development at present.

In this review, I shall restrict my analyses to the initial phases of limb formation and patterning. Two major aspects will be discussed: (i) the role of BMPs in dorsal-ventral (DV) polarization of the limb mesoderm and ectoderm, and its relation to AER induction, and (ii) their role in controlling digit number and identity. At later stages, BMP are known to be involved in other aspects of limb development, for which the reader is referred to previous reviews. For example, BMP signaling has been proposed to direct morphogenesis of digit primordia from the interdigital mesenchyme (Dahn & Fallon 2000; reviewed in Robert & Lallemand 2006). BMPs are also required for cartilage and bone differentiation (reviewed in Karsenty & Wagner 2002; Tsumaki & Yoshikawa 2005); and at the stage of interdigital webbing regression, BMPs have a crucial role in initiating apoptosis (reviewed in Zuzarte-Luis & Hurle 2005).

Before discussing specific results, I would like to introduce a number of preliminary remarks on the conditional mutation strategy. In analyzing the resulting phenotypes, it should be kept in mind that the extent of Cre-induced gene inactivation is critical and must be assessed. The stage at which inactivation takes place and the stage at which the phenotype is analyzed are also critical since expression and activity of BMPs are highly dynamic. In addition, the results may vary considerably between fore- and hindlimb, such that generalization to a generic limb is not always possible. Finally, the nature and combination of the alleles used may have an importance. For example Bmp4floxed/floxed : Prx1Cre will affect only the mesoderm; Bmp4null/floxed : Prx1Cre will similarly affect the mesoderm but will also halve the gene dosage in the ectoderm.

BMP signaling in ectoderm and AER formation

The apical ectodermal ridge (AER) is a major signaling center for limb outgrowth. It forms at the boundary between the dorsal and ventral domains of the lateral ectoderm, that constitute compartments (Altabef et al. 1997; Michaud et al. 1997). It can further be ectopically induced by abutting dorsal and ventral ectoderm grafts (Tanaka et al. 1997). This demonstrates the role of a dorsal-ventral (DV) boundary in the limb field ectoderm to induce and position the AER. What the molecular determinants of dorsal and ventral identity may be in the flank ectoderm has been a matter of debate over the years. Wnt7a and En1 are expressed specifically in the dorsal and ventral domains, respectively (Zeller & Duboule 1997; Chen & Johnson 2002). Furthermore, they play a critical role in the specification of dorsal and ventral identity in both ectoderm and mesoderm. Wnt7a mutants are biventral (Parr & McMahon 1995). In the En1 mutants, Wnt7a expression expands to the ventral ectoderm, and limb buds are bidorsal (Loomis et al. 1996). In the subjacent mesoderm, Wnt7a induces expression of Lmx1b, a gene required for dorsal identity, which explains the effect of Wnt7a and En1 mutations on both ectoderm and mesoderm DV identity (Chen et al. 1998). However, mutation of neither of these genes leads to AER loss, even though the AER in En1 mutants is abnormal (Loomis et al. 1998). Notch signaling has been proposed to play a prominent role in specifying the dorsal-ventral boundary, via the activation of radical fringe (Rfng) specifically in the dorsal part of the limb field ectoderm (Laufer et al. 1997; Rodriguez-Esteban et al. 1997). But mutations for Rfng or both Rfng and lunatic fringe in the mouse have not confirmed a role for Rfng in AER formation (Moran et al. 1999; Zhang et al. 2002).

More recently, the conjunction of experimental embryology in the chick and conditional mutation analysis in the mouse has established the cardinal role of BMP signaling for the specification of DV polarity in the limb ectoderm and for AER formation (Ahn et al. 2001; Pizette et al. 2001).

Functional BMP receptors are heterodimers of type I and type II receptor subunits, both of which are indispensable for signal transduction. BMP type I receptors are encoded by two genes, Bmpr1a and Bmpr1b (Chen et al. 2004). The Bmpr1a gene was selectively inactivated in the limb ectoderm using a floxed Bmpr1a allele in conjunction with a Brn4-Cre transgene that is specifically expressed in the limb ectoderm (Ahn et al. 2001). Expression was monitored using the R26R Rosa reporter allele that expresses lacZ upon Cre activation (Soriano 1999). This showed Brn4-Cre expression is prominent in the ventral limb ectoderm and in the AER, and less strong in the dorsal ectoderm of both fore- and hindlimbs from embryonic day 9.75 (E9.75), i.e. after the forelimb has begun to form, but before the hindlimb has formed. For this reason, the phenotype of the hindlimb was studied in more detail. BMP activity was further evaluated by monitoring the degree of phosphorylation of SMAD1, using a phospho-SMAD1 specific antibody. At E9.75, phospho-SMAD1 is detected both in the lateral mesoderm, where it extends up to the lower aspect of the somites, and in the overlying ectoderm. In the mutant, phosphorylation was not affected in the mesoderm but became undetectable in nearly all cells in the ectoderm. This indicates that BMPR1a is the main BMP receptor in the limb ectoderm, and that by E10, the ectoderm does not signal back on the mesoderm to maintain BMP activity, at least not via the BMP pathway.

The mutant hindlimb exhibited a large range of phenotypes, ranging from amelia (complete absence of limb structures) to polysyndactyly. This could be correlated with the activity of the AER as monitored by Fgf8 expression, which ranged from undetectable to nearly normal levels, and demonstrated a requirement for BMP in the ectoderm to form the AER. In contrast with this variability in AER phenotype, dorsal-ventral polarity of the limb bud was abrogated with complete penetrance and the limb structure was bidorsal at birth. Correlatively, En1 expression was undetectable at any stage of embryogenesis, whereas the Wnt7a domain extended to the ventral ectoderm while Lmx1b, a Wnt7a read-out, extended to the ventral mesenchyme. Interestingly, DV polarity was not affected in the forelimb. This suggests that DV patterning in the ectoderm is established in a narrow time window, just prior to the initial outgrowth of the limb bud. A few hours after the limb bud is formed, BMP signaling is no longer required to maintain DV identity, and indeed, in the forelimb of the Brn4-Cre : Bmpr1a conditional mutants, En1 and Wnt7a are expressed normally at E11.5.

The discrepancy between the complete penetrance of the DV defects in the mutant hindlimb and the high variability in AER induction suggests that BMPs are required for AER formation beyond their role in DV specification. Ahn et al. (2001) have investigated the expression pattern of Bmp4 and Bmp7 during limb development (Fig. 1). At the hindlimb level, Bmp4 is expressed mostly in the ventral mesoderm at E9.75, overlapping precisely with the domain of phosphorylation of SMAD1 reported above. Bmp7 is expressed in the ventral ectoderm up to the presumptive region of the AER, thus defining the boundary between ventral and dorsal limb ectoderm at this stage. At E10, Bmp7 mRNA is nearly undetectable whereas the Bmp4 expression domain is restricted to the AER region, where it overlaps with Fgf8. Bmp4 might therefore play a specific role in AER formation. The requirement for BMP activity to form the AER independently of DV patterning, however, is not compatible with the results in the chick, reported by Pizette et al. (2001) simultaneously with Ahn et al.'s results.

In the chick, the expression patterns of Bmp genes in the lateral ectoderm and mesoderm are very similar to those seen in the mouse (Fig. 1). Pizette et al. (2001) demonstrated that depletion of the ventral BMP signal, elicited by over-expressing noggin over the whole limb ectoderm, prevents En1 expression in the ventral ectoderm. This results in the formation of a bidorsal limb bud, as assessed by Wnt7a expression, which precludes AER formation. Reciprocally, activation of BMP signaling over the whole limb ectoderm, by expressing a constitutively active BMP receptor gene (BmpR), leads to induction of En1 also in the dorsal ectoderm, which adopts a ventral identity, and abrogates AER formation. In addition, expression of the constitutively active BmpR in the dorsal ectoderm, where it creates ectopic boundaries between Bmp-expressing and nonexpressing ectoderm, induces patches of Fgf8 expression. Furthermore, Pizette et al. (2001) observed that Msx1 and Msx2 display overlapping expression patterns with Bmp4 and Bmp7 in the ventral ectoderm at early stages of limb development (Fig. 1), and that ectopic expression of Msx1 in limb dorsal ectoderm, where Msx genes are normally not expressed at stages of AER induction, may result in the formation of ectopic ridges expressing Fgf8. These extra ridges would be induced by the formation of a new boundary between Msx-expressing and nonexpressing cells, suggesting that Msx genes are downstream effectors of the BMP signal in AER formation. Based on these and other results, these authors proposed that BMP signaling is governing both DV limb patterning and AER formation via the induction of En1 and Msx genes, respectively, and that these two pathways are independent. Our own results on the mouse Msx1 : Msx2 double null mutant confirm, on one hand, that these genes are required for the specification of DV polarity in the ectoderm and AER induction. However, on the other hand, they do not support a specific role for Msx genes in AER formation independent of DV polarization, since both processes are affected in the anterior part of the mutant limb (Lallemand et al. 2005).

The similarity of phenotypes for the AER, which is lost in any case when a boundary fails to form between BMP signaling (ventral) and non-signaling (dorsal) ectodermal domains, whether this is elicited by inducing ectopic BMP activity or abrogating endogenous activity over the whole limb ectoderm, is the most compelling argument indicating a link between DV patterning and AER formation, since the requirement for a specific BMP or a critical level of BMP signaling cannot be invoked.

Experimental embryology has long established the interplay between mesoderm and ectoderm in DV polarity acquisition in the limb bud. Once the limb bud is formed, DV polarity is dictated by the ectoderm to the mesoderm (Pautou & Kieny 1973; MacCabe et al. 1974). Inversion of the ectodermal cap results in inverted DV identity in both ectoderm and mesoderm. However, before the limb bud has formed, DV polarity is deposited in the mesoderm. The transfer of DV polarity control from mesoderm to ectoderm takes place before limbs grow out, around stage HH15 in the chick (Geduspan & MacCabe 1989). Further investigations have shown that dorsal identity in the ectoderm is specified by signals from the paraxial mesoderm, whereas lateral somatic mesoderm plays a role in specifying ventral ectoderm identity (Michaud et al. 1997). Based on these observations and the data on the manipulation of the BMP signal in mouse and chick (Ahn et al. 2001; Pizette et al. 2001), Ahn et al. proposed a molecular model for ecto–mesodermal interactions during early limb development. Before limbs form, Bmp4 and Bmp7 are expressed in the lateral mesoderm. Their expression domains therefore are properly positioned to induce ventral activity of BMPs in the ectoderm, via BMPRIA. Noticeably, the Bmp4 expression domain in the mesoderm extends more dorsally than the level of the presumptive AER (Fig. 1). However, BMP4 activity can be counteracted in the most dorsal region by noggin, a BMP antagonist that is expressed in the myotomal compartment of the somite. Once established, BMP signaling in the ectoderm induces En1 ventrally, and the latter restricts Wnt7a expression to the dorsal ectoderm. From there, Wnt7a induces Lmx1b in the mesoderm. After En1 and Wnt7a expression is activated, DV polarity in the ectoderm becomes independent of BMP signaling. Specification of the dorsal and ventral ectoderm domains would then be the trigger for AER formation.

BMP signaling in the mesoderm

Inactivation of Bmpr1a in the whole bud mesoderm

Most studies on limb mesoderm-specific gene inactivation to date have been carried out using a Prx1-Cre transgene (Logan et al. 2002). This transgene was shown to activate Cre expression in the forelimb from the earliest stages of limb bud formation (E9.5), and to provide complete deletion of a floxed reporter gene by E10.5 in both the fore- and hindlimb mesoderm. Furthermore, expression was never detected in the limb bud ectoderm.

Bmp2, Bmp4 and Bmp7 are all expressed in the limb mesoderm (Fig. 2). In the limb mesenchyme, BMPR1A, a BMP2-BMP4 specific receptor that binds other BMPs with less affinity (Yamaji et al. 1994), appears to be the major type I receptor. Whereas at E12.5 Bmpr1a is expressed over the whole mesenchyme, Bmpr1b is expressed only in a central mesenchymal domain at E10.5, then in the cartilage condensations between E11.5 and E13.5, and in the perichondrium at E15.5 (Dewulf et al. 1995; Yi et al. 2000). Correlatively, mutations in Bmpr1b affect limb cartilage differentiation, but not limb bud outgrowth or digit blastema patterning (Baur et al. 2000; Yi et al. 2000). However, BMP7, and to a lesser extent, BMP2, can bind to the activin receptor (Macias-Silva et al. 1998; Greenwald et al. 2003) and in Xenopus, this receptor has been shown to mediate BMP4 signaling in the early embryo (Armes & Smith 1997). Furthermore, the activin receptor 1a (Acvr1a) is expressed in the limb mesenchyme (Verschueren et al. 1995). Acvr1a mutants do not survive beyond E9.5 (Mishina et al. 1999). Acvr2a and Acvr2b, which encode ACVRII peptides, the obligatory partners of type I receptors, have also been mutated. Neither single mutation leads to limb pattern defects. The double null mutants die at gastrulation, whereas Acvr2a+/– : Acvr2b−/– mutants survive only to E9.5, precluding study of the developing limb (Song et al. 1999). Therefore, it should be kept in mind that, while inactivation of Bmpr1a should abrogate most BMP signaling in the limb mesoderm, some aspects of the phenotypes described below might be explained by compensation by the activin receptors.

Using Prx1-Cre and a floxed allele of Bmpr1a (Mishina et al. 2002), Ovchinnikov et al. (2006) induced a deletion of Bmpr1a specifically in the limb bud mesoderm. Southern blot analysis of limb mesoderm DNA confirmed that, at E11.75, the deletion was complete in the forelimb and ~90% in the hindlimb. Most of the conditional mutants were viable and exhibited both fore- and hindlimb formation but with severe defects at birth. In the forelimb, the humerus lacked the deltoid process, a phenotype reminiscent of the Hoxa9 : Hoxd9 double mutant phenotype (Fromental-Ramain et al. 1996). The radius and ulna were present in the zeugopod, although abnormally shaped. In contrast, the presence of the fibula was variable in the hindlimb. The autopod was the most affected segment. In the forelimb, the carpals and metacarpals were disorganized and no digit formed. In the hindlimb, tarsals and metatarsals were abnormal but some phalanges could be identified. The defects could be traced back to E12.5. Alcian blue staining showed that forelimbs do not form digital rays, whereas hindlimbs have up to two underdeveloped cartilaginous digital precursors that do not give rise to normal digit structures later on.

The skeleton phenotype is in sharp contrast with the size and shape of the limb paddle, which appeared normal at least up to E11.5, and even expanded anteriorly (Fig. 3). SHH is the primary regulator of anterior-posterior (AP) patterning, as well as distal outgrowth, of the limb bud (reviewed in Robert & Lallemand 2006). In the mutant, its expression was not significantly changed. However, Ptch1, a direct target of SHH, was expressed ectopically in the anterior mesenchyme of the footplate, consistent with the anterior expansion of this region. Accordingly, Hoxd13 was expressed normally at E11.5, apart for a slight anterior expansion. Hoxd11 showed an anteriorly expanded expression pattern in the presumptive zeugopod and a normal pattern in the presumptive digit domain, albeit with a lower expression level. Correlatively, the AER extended more anteriorly than normal in the mutant hindlimb, as judged from Fgf8 expression. These data suggest a limited posteriorization of the most anterior structures of the limb, restricted to a proximal region of the limb bud, i.e. the presumptive zeugopod.

Figure 3.

Figure 3.

Decreasing the level of bone morphogenetic protein (BMP) signaling results in expansion of the limb footplate that does not necessarily result in polydactyly. (A) Computer-assisted imaging of the contours of the hindlimb shows that it is enlarged anteriorly in the Prx1-Cre : Bmpr1a conditional mutant, although this mutant does not develop digits (reprinted from Ovchinnikov et al. [2006], with permission from Elsevier). The limb paddle is also enlarged in the Prx1-Cre : Bmp2 : Bmp4 conditional mutant (B), both anteriorly and posteriorly, but in the enlarged bud, only the three anterior digits form (D), (B–D reproduced from Bandyopadhyay et al. [2006]). In contrast, in the Prx-Cre : Bmp4 conditional mutant, initial expansion of the limb mesoderm leads to polydactyly (Fig. 3; Selever et al. 2004).

The absence of a dramatic phenotype at the early stages of limb development suggests that most defects observed at birth are more likely related to a deficiency in the formation of mesenchymal condensations that prefigure the digits and in cartilage differentiation, rather than to a patterning problem. This is substantiated by the analysis of apoptosis and cell proliferation, which was carried out between E11.5 and E12.5. Cell proliferation was significantly reduced at E11.5, and CyclinD1 expression was reduced accordingly. This may account for the limited development of the autopod. Apoptosis was not affected in the mutant limb at these stages. I would like to propose, nevertheless, that the anterior expansion of the mesenchyme observed at E11.5 may be related to a decrease in cell death at earlier stages. In the mouse limb bud, two zones of apoptosis have been described, a central one and an anterior one, which are conspicuous from E10.25 (Harfe et al. 2004). The position of the anterior apoptotic zone fits well with a role in limiting anterior development of the limb bud. The influence of cell death, in addition, is not easy to quantify, because usual methods measure the number of apoptotic cells at a given instant. A change in this rate, even modest, can have a significant outcome over time.

From the model proposed by Ahn et al. (2001), BMP signaling is required in the mesoderm to specify DV polarity of the ectoderm and AER induction. In the Bmpr1a conditional mutant, the AER did form and DV polarity was only affected in the mesoderm, where the Lmx1b expression domain expanded ventrally. In contrast, expression of Wnt7a in the ectoderm appeared normal. This paradoxical observation may be explained by the relatively late stage at which Prx1-Cre is activated, hence deletion of Bmpr1a induced, as discussed below.

Inactivation of Bmpr1a in a subpopulation of limb bud mesoderm cells

Isl1 is expressed, as early as E8.5, in cells that contribute to the heart field, and in the lateral plate mesoderm adjacent to the hindlimb presumptive territory (Yang et al. 2006). These authors further achieved a lineage analysis, by combining a Cre knockin at the Isl1 locus with the R26R reporter genes (Soriano 1999). This showed that Isl1-expressing cells at the level of the hindlimb constitute a population of progenitors that colonize roughly the posterior half of the hindlimb bud mesoderm, but do not contribute to the ectoderm. Using the same Isl1-Cre knockin together with a floxed Bmpr1a allele (Mishina et al. 2002), Yang et al. (2006) could inactivate the Bmpr1a gene in the precursor cells that express Isl1 and in all their descendants, i.e. the whole posterior hindlimb mesenchyme.

The hindlimb phenotype of the Isl1-Cre : Bmpr1a conditional mutant is complex and, at first sight, puzzling. At E11.5, ectopic outgrowths were observed on the ventral side of the limb bud, at a right angle to the endogenous limb bud plane. These gave rise later on to extra digits pointing downwards. In situ hybridization with Fgf8 labeled, in addition to the normal AER expression domain, a transverse row of cells across the ventral aspect of the limb ectoderm, which functionally constituted a second ridge orthogonal to the normal one. The inductive capacity of this ectopic ridge was further assessed by the ectopic expression of Shh, which was observed underneath the Fgf8-expressing ectopic cells. Ridge activity likely led to the formation of the ectopic outgrowths and digits ventrally.

This complex phenotype can be interpreted as resulting from the formation of two distinct compartments in the hindlimb bud ventral ectoderm, one (anteriorly) with ventral identity and the other one (posteriorly) with dorsal identity. This indeed is indicated by the expression pattern of DV characteristic markers: En1 expression was lost from the posterior part of the ventral ectoderm and, accordingly, Lmx1b expression domain extended ventrally in the posterior mesoderm. Thus, a new DV boundary formed that was orthogonal to the normal one. As a result, Fgf8 expression was induced along the new boundary, which behaved as a new signaling center similar to the AER. This led to production of a few ectopic digits that formed along the ectopic ectodermal ridge and pointed ventrally.

This phenotype, however, is not expected since BMP signaling in this conditional mutant is abrogated only in the mesoderm by Cre activation. It implies that BMP signaling in the mesoderm has a direct influence on the overlying ectoderm, by a process analogous to homeogenetic induction (meaning ‘like begets like’) (De Robertis et al. 1989). This is in keeping with the scheme proposed by Ahn et al. (2001). In this model, BMP is the critical signal for transfer of DV polarity from the mesoderm to the ectoderm. Initially, only the somatopleural mesoderm is a source of BMP activity. This results in the capacity of the overlying ventral ectoderm to activate the BMP signaling loop, thus leading to formation of dorsal and ventral domains with distinct identities, differential expression of En1 and Wnt7a, as well as AER induction.

The BMP ligand(s) involved in signal transfer from mesoderm to ectoderm is not identified by the Isl1-Cre : Bmpr1a conditional mutant analysis. It is noteworthy, however, that expression of Bmp4 is affected in the limb mesenchyme of this mutant, suggesting that Bmp4 induction is under the dependence of BMP signaling itself and might act as a relay between mesoderm and ectoderm. This is further substantiated by the analysis of a limb mesoderm-specific Bmp4 mutation (Selever et al. 2004; see infra).

Complete abolition of BMP signaling in the mesoderm should prevent ectoderm DV polarity specification, AER formation, and lead to a complete absence of limb bud, as is the case in the extreme phenotypes of Brn4-Cre : Bmpr1a conditional mutants (Ahn et al. 2001). Why is this not observed in the Prx1-Cre-driven Bmpr1a mutation (Ovchinnikov et al. 2006)? This is probably because Prx1-Cre is not expressed as early as Isl1-Cre. In the mouse, the AER is not conspicuous before E10.5. However, Fgf8 expression shows that AER activity can be detected at stage 1 of limb development (Wanek et al. 1989), i.e. around E9.5 (Crossley & Martin 1995; Loomis et al. 1998). This indicates that AER induction takes place early in the somatopleural mesoderm, at the onset of limb bud formation if not before, as suggested by Milaire (1974) on the basis of histochemical and morphological studies. The mesoderm in which the Bmpr1a conditional allele is inactivated by Isl1-Cre transgene should be already affected for BMP signaling at this stage, while this is unlikely when the deletion is induced by Prx1-Cre.

Notably, in the chick Eudiplopodia mutant, a whole array of ectopic digits forms dorsally, in parallel to the normal digit array, and the defect is known to reside in the ectoderm (Fraser & Abbott 1971a,b). This should be revisited in light of the results summarized here.

Inactivation of Bmp4 in the mesoderm

Conditional inactivation of Bmp genes from the mesoderm allows their individual roles to be elucidated. Furthermore, while the mesoderm becomes defective in production of the BMP ligand(s), it is still competent to respond to BMP signals from the ectoderm. Bmp4 was selectively ablated from mouse limb mesoderm using the Prx1-Cre transgene, in conjunction with a floxed Bmp4 allele (Selever et al. 2004). The extent of Cre activity was monitored using R26R (Soriano 1999). The absence of Bmp4 mRNA was further demonstrated by in situ hybridization and reverse transcription-polymerase chain reaction (RT–PCR) analyses.

An unexpected aspect of the Bmp4-deficient phenotype is that the forelimb is little affected, showing only a mild posterior polydactyly with incomplete penetrance; whereas the hindlimb exhibits a severe polydactyly phenotype. This is in contrast with the phenotype of the Bmpr1a conditional mutant, where the forelimb is more severely affected. In the hindlimb, polydactyly was observed both on the anterior and posterior sides, with variable severity. The authors thus defined four phenotypic classes with decreasing severity of polydactyly (Fig. 4B–E). For the anterior phenotype, my interpretation is that the mutation affected exclusively the thumb. It was, from the mildest to most severe class (i) only partially duplicated, the extra-thumb consisting of a single cartilaginous element (Fig. 4E); (ii) completely duplicated (Fig. 4D); (iii) duplicated with the most anterior digit transformed into a triphalangeal digit (Fig. 4C); and finally (iv) duplicated with transformation of both digits into triphalangeal structures (Fig. 4B). In the latter class, the rudiment of a third thumb even formed. This suggests a progressive posteriorization of the duplicated structure associated with stronger expression of polydactyly. On the posterior side, duplication of digit five was always observed (Fig. 4B–E), and even the rudiment of a third digit five in the most affected class (Fig. 4E).

Figure 4.

Figure 4.

 Limb mesoderm-specific knockout of Bmp4 leads to anterior and posterior polydactylies. Prx1-Cre : Bmp4nul/floxed hindlimbs exhibit a variable polydactylous phenotype that is classified from severe (B, two triphalangeal thumbs plus an anterior nubbin) to moderate (E, only one ectopic anterior nubbin). On the posterior side, the number of extra-digits increases from B to E. Note that anteriorly as well as posteriorly, extra-digits branch more proximally from E to B, giving rise to more independent digits. In all panels, anterior is to the left. (Reprinted from Selever et al. [2004], with permission from Elsevier.)

Whereas digit duplications could be readily correlated with an increase in cell proliferation at E10.5 and E11.5, digit identity is more difficult to assess, because classical markers of AP polarity were little affected in the mutant. Hoxd12 and Hoxd13 expression domains were unchanged, while they expand anteriorly in polydactylies with changes in anterior identity, such as is elicited by the Msx1 : Msx2 double null mutant (Lallemand et al. 2005). However, Alx4 was down-regulated, indicating that anterior gene expression may be affected. Similarly, Fgf4 expression domain, which is normally restricted to the posterior two thirds of the AER, extended to the anterior-most aspect of the AER, further assessing the posterior identity this region assumes in the Bmp4 mutant limb. Together with the morphology of extra-digits, this indicates that anterior identity is affected in the Bmp4 mutant limb.

On the posterior side, the Shh expression domain was expanded and this correlated with expanded expression domains of Shh targets, such as Ptch1 and Gli1. This may explain the posterior polydactyly, as the extent of polydactyly induced by Shh in the chick is dose-dependent (Yang et al. 1997). This confirms that one function of Bmp4 is to repress Shh expression, by interfering with the Fgf-Shh loop (Zuniga et al. 1999), a process finely regulated by the extracellular inhibitor of BMP4 activity Gremlin (Khokha et al. 2003; Michos et al. 2004).

In the Prx1Cre : Bmp4 conditional mutant, the DV polarity of the limb is affected. Expression of En1 is reduced, while Wnt7a domain expands ventrally in the ventral ectoderm as Lmx1b in the mesoderm. This is in keeping with the role of mesoderm in the DV specification of the ectoderm I have discussed previously, and further suggests that mesodermal BMP4 is the major source of the signal that instructs the ectoderm. This alteration of ectoderm DV polarity influences the AER: although it does form, it appears abnormally enlarged on its ventral side as it does in the En1 mutant. This is similar to the broadening of the AER observed by forcing expression of noggin in the AER (Wang et al. 2004), thus reinforcing the hypothesis of a tight interplay in BMP signaling between mesoderm and ectoderm. This hypothesis implies that BMP produced in the mesoderm can exert its effect in the ectoderm, although it is not thought to cross the basal membrane (Kelley & Fallon 1976; Wilcox & Kelley 1993), raising the possibility of a requirement for another signal between the two layers. Ectodermal Wnt3 signaling was demonstrated to lie upstream of the BMP signaling pathway in establishment of the AER and regulation of limb DV polarity, and may be part of this relay (Barrow et al. 2003).

At later stages, when the AER has regressed in normal mice and Fgf8 is extinct, Fgf8 was still expressed in the mutant over the anterior digits. Indeed, BMP4 has been proposed as a major signal to shut down AER activity (Pizette & Niswander 1999). Based on these observations, Selever et al. (2004) propose that anterior polydactyly in the conditional mutant does not primarily depend on the mesoderm, but on sustained Fgf activity in the ectoderm.

Combination of several Bmp deficiencies in the limb mesoderm

To dissect out the specific role of each Bmp gene and the degree of redundancy between them, Bandyopadhyay et al. (2006) undertook to inactivate Bmp2, Bmp4 and Bmp7 in the limb mesoderm and combine the different mutations. This was achieved using the same Bmp4 conditional allele as Selever et al. (2004), a null allele of Bmp7 (Dudley et al. 1995) and a new conditional allele of Bmp2. Limb mesoderm-specific mutation was achieved using the Prx1-Cre transgene (Logan et al. 2002). Neither any single mutation nor any two-by-two combination of mutations prevented limb formation, and none of these combinations led to a major patterning defect of the limb skeleton, with the exception of the combination of Bmp2 : Bmp4 conditional alleles. Bmp2 deficient limbs appeared normal, except for a characteristic defect in the scapula, and those lacking Bmp7 only showed an occasional anterior polydactyly, as described previously (Luo et al. 1995; Hofmann et al. 1996). Bmp4-deficient limbs exhibited a phenotype similar to that described by Selever et al. (2004) for the same mutation, (i.e. anterior and posterior polydactyly), except that the phenotype had reduced penetrance as one extra digit, at most, was observed on either side (cf. Fig. 4). This was attributed to Bmp4 gene dosage in the ectoderm: while the mutation in the mesoderm was generated by Selever et al. (2004) using one null and one conditional allele, leading to a single functional allele in the ectoderm, Bandyopadhyay et al. (2006) used two conditional alleles, thus keeping two functional alleles in the ectoderm.

The combination of mutant alleles of two out of the three genes led to cumulative rather than synergistic defects when compared to the effect of each mutation. However, Bmp2 : Bmp4 double null mutant limbs were more severely affected. They displayed a shortened and malformed stylopod, and one of the bones in the zeugopod – the identity of which could not be determined – was almost always missing. The most informative phenotype in terms of Bmp function in the limb was that of the forelimb autopod, in which the two posterior digits were missing (Fig. 3C–D). This correlates well with the preferential expression of Bmp2 posteriorly (Fig. 2). Interestingly, at E11.5, the limb paddle was broader than normal, which should have led to poly, not oligodactyly (Fig. 3B). At E12.5, Sox9 in situ hybridization revealed that only the three anterior-most digit primordia had formed in this enlarged paddle (Fig. 3D). Similarly, the limb paddle shows an anterior expansion at E11.5 in the Bmpr1a-deficient hindlimb (Fig. 3A; Ovchinnikov et al. 2006). There is also an expansion of limb bud mesoderm in the single Bmp4 conditional mutant, where this correlates with an increase in cell proliferation that leads to polydactyly at later stages (Selever et al. 2004).

Role of BMP signaling in specifying digit number and identity

Bmp4, Bmp2 : Bmp4 and Bmpr1a mutants constitute a graded series in terms of residual BMP signaling activity in the limb mesenchyme, with the Bmpr1a mutant being the most, and Bmp4 mutant the least depleted. Whereas the limb paddle is almost equally enlarged at E11.5 in these different mutants, loss of digits prevails when no BMP signaling remains (Ovchinnikov et al. 2006) while polydactyly is prevalent when the single Bmp4 gene is altered (Selever et al. 2004). As the digit defect is conspicuous as early as E12.5, I would like to propose that BMP signaling plays two different roles in digit formation. Up to E11.5, BMP activity would restrict cell proliferation, especially in the anterior and posterior domains of the mesoderm where Bmp4 is most expressed, preventing excessive growth of the limb paddle and polydactyly. This is coherent with the long-recognized BMP capacity to antagonize FGF signaling from the AER (Niswander & Martin 1993; reviewed in Dahn & Fallon 1999). BMP may also modulate cell death in the anterior apoptotic zone, as discussed previously. From E11.5 on, BMP would be required for the condensation of the mesenchymal digit anlage. Indeed, in the chick, when noggin is introduced in limb buds prior to skeletogenesis, mesenchymal condensation does not take place (Capdevila & Johnson 1998; Pizette & Niswander 2000).

In this model, the level of BMP signal required for the first phase is not linear. A discrete decrease in BMP, such as provided by the mutation of the single Bmp4 gene, is sufficient for expansion of the limb bud mesoderm and thus results in polydactyly (Selever et al. 2004). The second phase in contrast is sensitive to the dose of residual BMP signal. In the Bmp4 mutant, Bmp2 and Bmp7 provide sufficient signals, such that polydactyly prevails. In the double Bmp2 : Bmp4 mutant, although the first phase takes place and limb bud is enlarged, Bmp7 is not sufficient to allow normal digit development, and the precartilaginous condensations of the two posterior-most digits are never discernable (Bandyopadhyay et al. 2006). When BMP signaling is completely abrogated, as in the Bmpr1a conditional mutant, no digit condensation forms and at birth, the animals are devoid of any digit structure in the hindlimb (Ovchinnikov et al. 2006). Further investigations, taking advantage of tamoxifen-inducible Cre recombinases (Metzger & Chambon 2001) to inactivate Bmp and Bmpr genes in a stage-specific manner, are required to challenge the model I propose here.

BMPs further have a role in the identity of digits along the AP axis. This is best exemplified by the Bmp4 conditional mutant, which forms additional triphalangeal digits at the anterior side of the limb (Selever et al. 2004; Fig. 4). Shubin & Alberch (1986) very clearly addressed the question of whether digits have an identity of their own, encoded in some way in the vertebrate genome, or whether their morphology is the product of a small set of developmental rules acting on an integrated developmental limb field. They proposed that ‘the development of the limb is the product of a combination of global organizers . . . and local interactions that characterize the process of chondrogenesis’. Their proposal has received experimental support recently. Indeed, the limb field is prepatterned at an early stage of development (Chiang et al. 2001; reviewed in Robert & Lallemand 2006). This translates into the stereotypical expression profiles of a number of cardinal genes, such as Hoxd11 and Hoxd12 (Charite et al. 2000; Chiang et al. 2001). Cell proliferation, cell death and cell migration must be locally modulated within this prepatterned field, and depending on its properties, to give rise to structures with specific morphologies that have been selected over evolutionary time for their functional adaptation. Altering either of these processes may lead to very similar changes in digit morphology, but the mechanisms underlying these changes are very different. In this scheme, BMPs are likely to play a role in the second phase, i.e. local modulation of cellular properties. Indeed, the limb prepattern is little affected by interfering with BMP signaling, as indicated by the almost normal expression pattern of Hoxd11, Hoxd12 and Hoxd13 (e.g. Selever et al. 2004; Ovchinnikov et al. 2006). BMPs in part may modulate local cell properties by antagonizing the mitogenic activity of FGFs produced by the AER (Niswander & Martin 1993; Dahn & Fallon 1999). They may also be involved in the regulation of apoptosis in the limb bud at early stages. Indeed, BMP signaling is involved in programmed cell death in the limb field as at several other sites (Zuzarte-Luis & Hurle 2005). In the mouse limb bud, apoptosis is quite limited, and two domains of apoptosis have been identified, a central one and an anterior one (Harfe et al. 2004). Decreasing the level of BMP signaling very often results in an anterior enlargement of the limb paddle and an anterior polydactyly that fits well with an alteration of the anterior apoptotic domain (Luo et al. 1995; Hofmann et al. 1996; Dunn et al. 1997; Katagiri et al. 1998). There are little experimental data on the relationship between apoptosis and these anterior polydactylies, and this remains a very promising domain of investigation.

Relationship to Msx genes

Epistatic relationships between Bmp and Msx genes have been established best in tooth and craniofacial development. Msx1 is both upstream and downstream of Bmp4 in the developing tooth germ (Chen et al. 1996; Bei & Maas 1998) and Bmp4 expression in the dental mesenchyme of the Msx1 null mutant restores part of the deleterious phenotype (Bei et al. 2000). Consequently, Msx genes have been proposed to be direct targets of BMPs, and even of specific SMADs (Hussein et al. 2003; Brugger et al. 2004; Binato et al. 2006).

Conditional inactivation of BMP signaling in the limb bud indicates that Msx genes are targets of BMPs at this site too. They may even be the mediators for BMP activity, as in the ectoderm where Msx1 mimics the BMP signal to support AER induction (Pizette et al. 2001). Inactivation of Bmpr1a in the ectoderm leads to complete down-regulation of Msx2 in the AER (Ahn et al. 2001). When Bmpr1a is mutated in the mesoderm (Ovchinnikov et al. 2006), expression of Msx1 and Msx2 is lost from most of their expression domains at the apex of the limb bud, apart for some residual Msx1 expression in small patches in the forelimb, possibly sites prepatterned to form digit tips, which may express Bmpr1b (Yi et al. 2000). However, in the forelimb, Msx1 and Msx2 expression is maintained in the anterior-most part of their expression domain, which, according to the Msx1 : Msx2 double null mutant phenotype, is the most critical one (Lallemand et al. 2005). They must therefore depend on other cues in this domain, where their expression is essential. Similarly, in the Bmp2 : Bmp4 conditional mutant, Msx2 expression domain expands in response to diminished BMP signaling (Bandyopadhyay et al. 2006). Msx2 is normally expressed in the interdigital mesenchyme and not in the digital condensations, in compliance with a general tendency of Msx genes to be expressed in undifferentiated tissues. In the Bmp2 : Bmp4 mutant, it is expressed over the whole posterior limb where undifferentiated mesenchyme replaces digit primordia, to comply with the new fate of this region. Only Bmp7 remains to support this extended expression, raising the possibility that at this site too, other cues are involved. Analysis of a conditional Bmp2 : Bmp4 : Bmp7 triple mutant should help clarify this issue.

Conversely, not all the BMP signaling is mediated by MSXs, since the limb phenotype of Msx1 : Msx2 double mutants is less severe than that of the Bmpr1a conditional mutant (Ovchinnikov et al. 2006).


Data acquired from conditional mutagenesis in the mouse have considerably enhanced our understanding of the role of BMP signaling in a number of aspects of limb outgrowth and patterning. The involvement of BMPs in setting up DV polarity in the mesoderm and transferring it to the ectoderm now lies on solid ground. This aspect has substantially benefited from the use of transgenes expressing Cre at different stages and different cell populations. Comparative analysis of results obtained with Brn4-Cre, Prx1-Cre and Isl1-Cre, taking into account the specificities of each transgene, leads to a coherent picture that confirms the predominant role of homeogenetic induction between ectoderm and mesoderm. It further demonstrates the power of comparing data obtained using different Cre-expressing transgenes to inactivate the same gene. This strategy should be widely used in the future.

Similarly, the role of BMP signaling in digit formation and identity is becoming clear. It seems that BMPs have a limited function in establishing the primary pattern of the limb. On the contrary, BMP signaling plays a predominant role in modulating local cell physiology. The effect of BMPs on AER development is well documented, and permits further understanding of several aspects of limb shaping by BMPs. It is very likely that BMPs are also involved in the modulation of apoptosis at several stages of limb development. While this has been extensively studied for the regression of interdigital webbing and individualization of digits, apoptosis at early stages of limb development and the role of BMPs in this process still deserve a systematic investigation. BMPs also play critical roles at later stages in cartilage and bone differentiation, which will have a very important influence on the definitive shape of the limb. This is currently the object of intensive investigation, but is beyond the scope of this review.


This work was supported by grants from the Institut Pasteur (GPH #7) and from ANR (# 06 MRAR 027 02). I am very grateful to Drs Colin Crist and Yvan Lallemand for their critical reading of the manuscript, to Elsevier and the Company of Biologists for allowing me to reproduce published data, and to Plos Genetics for unrestricted access to the data they publish.