Analysis of Msx1; Msx2 double mutants reveals multiple roles for Msx genes in limb development

The vertebrate limb is a model of choice with which to address fundamental questions related to signal transduction and pattern formation. Through the combination of experimental embryology and molecular genetics, the mechanisms that control growth and patterning of the limb along its three axes are beginning to be understood (reviewed by Niswander, 2003; Tickle, 2003).

Proximodistal (PD) development depends on the apical ectodermal ridge(AER), a transient epithelial structure rimming the distal margin of the limb bud (Saunders, 1948; Summerbell, 1974) that provides molecular signals to sustain cell proliferation in the underlying mesenchyme. Fgf8 and Fgf4 (and to a lesser extent Fgf9 and Fgf17) are expressed in the AER and promote normal limb development by maintaining the expression of another Fgf family member, Fgf10, in the mesenchyme (Sun et al., 2002). Reciprocally, in Fgf10^-/- mutant embryos, the AER does not form, leading to a complete truncation of fore- and hindlimbs (Sekine et al.,1999). Thus, it appears that PD limb outgrowth is dependent upon a positive feedback between the AER- and apical mesoderm-specific Fgf activities.

The key organizer for the anteroposterior (AP) axis has been identified as a population of cells from the posterior mesenchyme, called the zone of polarizing activity (ZPA), that specifically secretes the sonic hedgehog (Shh)protein (reviewed by Pearse and Tabin,1998). Heterotopic grafting of the ZPA(Saunders and Gasseling, 1968)or of Shh-expressing cells(Riddle et al., 1993) to the anterior margin of the limb bud leads to a mirror duplication of the digits. Conversely, the Shh null mutation provokes a dramatic loss of the anterior skeleton of the autopod (handplate/footplate) where a single digit I remains (Chiang et al., 2001). However, although Shh is necessary for normal expansion of the AP axis, the limb field is prepatterned along the AP axis prior to Shh activation through mutual genetic antagonism between Gli3 and Hand2(te Welscher et al.,2002).

Dorsoventral (DV) asymmetry is regulated, after limb bud formation, by the ectoderm. The dorsal and ventral limb ectoderm domains express, respectively, Wnt7a and En1. Mutations in both these genes have revealed that En1 is necessary to repress Wnt7a expression in the ventral ectoderm and that Wnt7a is the key inducer of Lmx1b in the dorsal mesenchyme. Thus, the En1 mutation leads to the ectopic ventral expression of Wnt7a in the ectoderm and Lmx1b in the mesoderm, and the distal structures develop with bi-dorsal characteristics(Loomis et al., 1996; Cygan et al., 1997). Conversely, in the absence of Wnt7a, the dorsal pattern of the autopod is not established and the limb appears biventral(Parr and McMahon, 1995).

These three organizing centres do not work independently. Rather,patterning and growth of the limb in three dimensions is coordinated through complex interactions between the different limb organizing centres (reviewed by Niswander, 2002). For example, AER induction depends not only on lateral plate-produced Fgfs but also on DV ectoderm polarity. This is related to Bmp signalling which is required both for DV patterning and AER induction(Ahn et al., 2001; Pizette et al., 2001).

Msx genes encode homeodomain transcription factors and their expression is associated with epithelio-mesenchymal interactions at many sites in vertebrate embryos such as limb buds, craniofacial processes and tooth buds(Hill et al., 1989; Robert et al., 1989) (reviewed by Davidson, 1995). Knockout experiments have shown that Msx1-null mutations provoke defects in craniofacial development, cleft palate, inner ear malformations and tooth agenesis (Satokata and Maas,1994; Houzelstein et al.,1997), whereas Msx2 null mutations lead to abnormal tooth development, loss of fur and reduced and disorganized cerebellar lobules(Satokata et al., 2000) (Y.L.,M-A.N., A.B. and B.R., unpublished). Single Msx homozygous mutants(Msx1^-/- or Msx2^-/-) do not display gross limb abnormalities. However, several lines of evidence suggest that this gene family may play a crucial role in limb morphogenesis. First, both genes are prominently expressed in the limb field from the earliest stages of limb formation and, later on, in both the AER and the subjacent mesenchyme. Furthermore, they require functional interactions between ectoderm and mesoderm for their expression and have been proposed to mediate transduction of inductive signals at this site(Davidson et al., 1991; Robert et al., 1991). Second,misexpression of Msx1 in dorsal ectoderm of the chick limb bud can induce formation of ectopic AERs, similarly to misexpression of a constitutively activated Bmp receptor, suggesting that Msx1 acts downstream of Bmp signals in AER induction (Pizette et al., 2001).

The co-expression of Msx1 and Msx2 and their possible overlapping functions might explain the absence of a limb phenotype in either of the simple mutants. We have analysed Msx1^-/-;Msx2^-/- double mutant embryos. We observed that Msx genes play a role at several stages of limb development and that their mutation affects the three organizing centres. In the double mutant, dorsoventral patterning is disturbed at the anterior border of the limb bud and this precludes AER formation anteriorly. Msx genes are necessary for the regression of the AER at later stages. The mutation of the two genes leads to further malformations along the AP axis, initially by preventing the development of the anteriormost region of the autopod (distal segment) and zeugopod (intermediate segment)and, later on, by provoking the abnormal overgrowth of the remaining anterior part of the autopod mesenchyme. These defects can be explained in the context of a role for Msx genes in Bmp signal transduction.

Insertional mutation of the mouse Msx1 homeobox gene by an nlacZ reporter gene and genotyping of the animals have been described previously (Houzelstein et al.,1997). The null allele of Msx2 was produced following a similar strategy (see Fig. S1 in the supplementary material). In brief, the recombination vector contained 3.1 kb and 5.6 kb of Msx2 sequences from the 5′ and the 3′ regions of the gene, respectively. Msx2 coding and intron sequences were replaced by an nlacZ-neo cassette, such that the nuclearβ-galactosidase is produced from the Msx2 initiation ATG. This construct was electroporated into 129 derived HM1 ES cells(Magin et al., 1992) (a gift from Dr D. Melton) and homologous recombination events identified. Following germline transmission, the mutation was transferred to a C57BL/6 background. Subsequently, the combined Msx1; Msx2 mutant strain was maintained on an outbred (NMRI) background. Double heterozygous animals displayed no apparent phenotype. Double homozygous embryos were obtained by Msx1^+/-; Msx2^+/- intercrosses. Day of the plug was considered as 0.5 dpc.

Whole-mount in situ hybridization was performed as described previously(Houzelstein et al., 1997). RNA probes were generated from the following DNA fragments: Fgf8 was a complete cDNA sequence obtained by PCR (forward primer CGC TCG GGC TCT CAG TGC TCC; reverse primer: GAG CTG GGC GAG CGC CTA TCG). Alx4 was a gift from F. Meijling; Hoxd11 from P. Dollé; Fgf4from G. Martin; Fgf9 from R. Kelly; Shh from A. McMahon; En1 from K. Schughart; Lmx1b from R. Johnson; Bmp2,Bmp4 and Bmp7 from B. Hogan; gremlin from R. Zeller; and Pax9 from R. Balling.

Skeletal preparation was as described in Zhang et al.(Zhang et al., 1995). Histological sections were carried out as described in Kaufman(Kaufman, 1994). For whole-mount cell death detection, embryos were fixed overnight in 4%paraformaldehyde and then dehydrated in methanol for long-term storage at-20°C. TUNEL analysis was performed on rehydrated embryos using the ApopTag Peroxidase detection kit (Q.BIOgene) according to the manufacturer's protocol for tissue cryosections and cells.

Msx1^-/-; Msx2^-/- double homozygous mutants display numerous deleterious defects and do not survive, on the NMRI outbred background we used, beyond the 15th day of gestation. At 14.5 dpc, the three segments of the double-null mutant limbs (i.e. from proximal to distal:stylopod, zeugopod, autopod), as well as the scapula and the pelvic girdle,were present but shortened (about 3/4 of the normal size)(Fig. 1A,B). In the autopod of control embryos two phalanges were visible at this stage, whereas, in the double-null mutants, only one had developed. In addition the interdigital webbing was retained (Fig. 1C-F).

Abnormalities along the anteroposterior axis were also very severe. They were confined to the anterior elements of the skeleton and displayed apparently paradoxical features. The main feature was a general truncation of the anterior part of the limbs with an almost systematic loss of the anterior element of the zeugopod and a frequent oligodactyly(Fig. 1A,B)(Table 1). In oligodactylous animals, the missing digit was always the anteriormost one (digit I). Carpals and tarsals were also affected in the anterior region. The posterior elements were present, the medium elements severely reduced or barely visible, and the anteriormost ones absent (Fig. 1C,D). Connections were maintained with the remaining digits,which allowed confirmation of their identity. In addition, double-null embryos lacked the pubis bone (data not shown), which corresponds to the anterior part of the pelvis, based on phylogenetic and embryological considerations(Hinchliffe and Johnson, 1980; Knezevic et al., 1997).

However, almost one quarter of the double-null forelimbs (4/18) and two out of 12 hindlimbs presented with five digits(Table 1). Furthermore, two forelimbs with six digits were observed(Fig. 1E,F; Table 1). In these examples,the anteriormost digit was longer than a normal thumb whereas the digit placed just posterior to it was reduced in length, suggesting that the most anterior digit had a more posterior identity. The complex double-null phenotype thus varied from an oligodactylous to a polydactylous morphology.

The size of double-null limb buds was reduced by about one third compared with controls at 10.5 to 11.5 dpc, even in somite-matched embryos (compare,for example, Fig. 3E with 3F, Fig. 3G with 3H or Fig. 4G with 4H). This prompted us to investigate the activity of the AER. Fgf8 is expressed specifically in pre-AER and AER cells (Loomis et al.,1998). Fgf8 expression was detected as early as the 14-somite stage in the double-null embryo limb field, indicating that AER initiation took place at the normal stage(Fig. 2A,B). However, its subsequent maturation was impaired. During mouse limb development, the cells giving rise to the AER are recruited from the ventral part of the limb bud ectoderm to form a broad pre-AER domain that becomes progressively thinner and ends as a thin strip of cells at the tip of the bud(Loomis et al., 1998). In double-null embryos, at 10.5 and 11.5 dpc, the AER appeared shorter and more diffuse than normal (Fig. 2C-H)as shown by Fgf8 (Fig. 2C,D) or Bmp7 (Fig. 2E,F) expression profiles, and histological analysis(Fig. 2 G,H). In addition, Fgf4 (Fig. 2I,J) and Fgf9 (data not shown), two other Fgf genes normally expressed in the AER as early as 10.5 dpc, were barely detectable at this stage in double-null embryos, proving that the activity of the AER was also delayed.

At 11.5 dpc, a sharp discontinuity was visible at the anterior of the limbs that appeared truncated anteriorly (Fig. 3A,B; see also Fig. 7C,D). At this stage, Fgf8 is normally expressed over the entire AER, while Fgf4 expression is absent from its anterior third(Niswander and Martin, 1992; Crossley and Martin, 1995)(see also Fig. 6A). These markers were analysed together in double in situ hybridisation. At 11.5 dpc, Fgf4 was expressed at a normal level in double-null limb buds but its expression pattern was abnormal relative to that of Fgf8. As expected, the AER of control embryos displayed an anterior domain expressing Fgf8 (Fig. 3A, red staining) but devoid of Fgf4-specific signal(Fig. 3A, purple staining). In double-null mutant embryos, Fgf8 and Fgf4 expression domains abutted exactly to the same anterior limit in the AER, which corresponded to the anterior limit of the ridge and to the sharp angular anterior border of the misshaped limb buds (Fig. 3B). These observations suggested that the anterior third of the AER was missing at this stage. In addition, the Fgf4 expression domain was shifted anteriorly (Fig. 3B; see also Fig. 6B).

Cell death was unlikely to be the primary cause of AER anterior truncation. At 10.5 dpc, cell death was high in the anterior part of the AER and adjacent ventral ectoderm in control embryos, whereas, in the mutants, it was reduced and scattered over the whole ventral limb bud ectoderm(Fig. 3I,J).

The mesenchymal anterior domain was not maintained either. At 9.5 and 10.5 dpc, Alx4 is specifically expressed in the anterior third of the limb bud mesenchyme. Taking into account the smaller size of the double mutant limb buds, the Alx4 expression pattern appeared similar in double-null and normal embryos at both 9.5 (data not shown) and 10.5 dpc(Fig. 3C,D). Therefore, until 10.5 dpc the anterior mesenchyme of the double mutant limb bud was present and apparently correctly specified. At 11.5 dpc, Alx4 is completely switched off in the autopod. We thus relied on Hoxd11 expression,which is symmetrical at 10.5 dpc in the distal part of the limb mesenchyme but becomes progressively polarized along the anteroposterior axis and is totally excluded from the anteriormost, future digit I (thumb) region from 11.5 dpc onwards (Charité et al.,2000; Chiang et al.,2001). At 10.5 dpc, no difference was observed between control and double mutant embryos (data not shown). However, 6-8 hours later (10.75 dpc,37-39 somites), when Hoxd11 expression begins to be excluded from the future digit I region, we observed a poor polarization of this domain in the double-null mutants. The size of the Hoxd11 expression domain was approximately the same in double-null and control embryos, but the Hoxd11 negative domain was almost absent in the mutant whereas it represents normally about one third of the limb volume at this stage(Fig. 3E,F). At 11.5 dpc, in double-null embryos, we observed a complete absence of the Hoxd11expression-free anterior region (Fig. 3G,H). Loss of mesenchyme with an anterior identity was confirmed using markers such as Hoxd12, which is expressed similarly to Hoxd11 (data not shown) and Bmp4(Fig. 7C,D).

These results, together with the morphological aspect of the double-null limb buds, demonstrate that the region corresponding to the anteriormost digit disappears between 10.5 and 11.5 dpc.

To analyse the possible involvement of Msx genes in DV polarity, we resorted to two markers of limb DV asymmetry. The first one, En1, is specifically expressed in the ventral ectoderm and the ventral AER of the limb bud, as early as 9.5 dpc (Loomis et al.,1998). At 9.75 dpc (24 somites)(Fig. 4A,B) and 10.5 dpc (data not shown), En1 expression was reduced in double-null limbs. However,at 11.5 dpc, En1 signal was detected similarly in control and double-null embryos, with the exception of the anterior part of the AER, which is missing at this stage in the double mutant(Fig. 4C,D). The second one, Lmx1b, is a specific marker of the dorsal mesenchyme and a read-out of ectodermal Wnt7a activity (Riddle et al., 1995; Cygan et al.,1997). At 9.5 dpc, no obvious abnorma1ity in Lmx1bexpression could be detected (data not shown) but, as early as 10.5 dpc, an ectopic ventral expression domain was visible at the anterior aspect of the double-null limb bud (Fig. 4E,F). This result was confirmed at 11.5 dpc. At this stage, the anterior ectopic expression domain corresponded precisely to the part of the limb that is devoid of AER (Fig. 4G-I). Thus, anteriorly, the normal DV boundary is not established in the double-null embryo limb bud. Considering the relationship between DV boundary formation and AER induction(Zeller and Duboule, 1997; Pizette et al., 2001), this is the most likely explanation for the absence of AER anteriorly. As the AER is required to maintain the underlying mesenchyme(Dudley et al., 2002), this would lead, at later stages, to the loss of anterior mesenchyme and truncation of the anterior skeleton elements.

Regression of the AER is retarded in double-null embryos. At 12.5 dpc, the AER of control embryos, as revealed by the expression of Fgf8, was present only at the tips of the digit anlagen and had disappeared at the level of the interdigital webbing (Fig. 5A). By contrast, in the double-null mutants, the AER was still continuous along the whole apex of the limb(Fig. 5B) and appeared thicker than normal (Fig. 5C,D). At 14.5 dpc, Fgf8 expression was still detectable at the tip of the digits in double-null embryos (Fig. 5E,F). Fgf8 overexpression was often correlated with the presence of supernumerary digit primordium tips, especially in forelimb buds where up to seven digit tips were observed(Fig. 5H).

In conclusion, Msx genes act in two different ways on AER formation. In a first period (9.5 to 11.5 dpc), the Msx1; Msx2 double mutation leads to AER truncation and incomplete maturation but, later on (12.5 to 14.5 dpc), it provokes its persistence beyond the normal stage.

In addition to AER persistence, overgrowth of the anterior-distal mesenchyme of the handplate was observed from 11.75 dpc in a large proportion of double-null mutant limbs, which may underlie the polydactyly observed at later stages (compare, for example, Fig. 5G with 5H, Fig. 6C with 6D, Fig. 6G with 6H or Fig. 7G with 7H). This was mainly observed in the forelimb whereas the hindlimb generally displayed only a discrete anterior deformation (compare Fig 5E with 5F or Fig. 6E with 6F). In addition, the extent of the outgrowth was variable from one individual to another and sometimes from one side to the other in the same embryo.

The first sign of this abnormal limb polarisation was indicated by Fgf4 expression. In the normal mouse at 11.5 dpc, Fgf4expression is observed in the apicoposterior part of the AER, but excluded from the ridge overlying the Shh- expressing mesenchymal domain(Lewandoski et al., 2000). In the double-null mutant, the Fgf4 domain was slightly shifted anteriorly, correlatively with the loss of the anterior part of the AER and the anterior extension of the Shh expression domain(Fig. 6A,B,I,J). By 11.75 dpc,the remaining part of the AER of the double-null mutant appeared to extend anteriorly (data not shown). The expression of Fgf4, which normally disappears at this stage, persisted anteriorly in the double-null mutant AER and was still detectable at its anterior tip up to 12.5 dpc, whereas in control embryos, Fgf4 expression had been switched off for more than 12 hours. This persistent expression domain corresponded to the tip of the anterior deformation observed at this stage(Fig. 6C,D).

Correlatively, overgrowth of the mesenchyme was observed, mainly in forelimbs. As the anterior outgrowth appeared secondarily on truncated limb buds devoid of anterior mesenchyme, we investigated the identity of outgrowing cells using the Hoxd11 marker. At 12.5 dpc, similar to 11.5 dpc, a prominent expression was observed in the anterior outgrowth of the limb in double-null embryos (Fig. 6F,H). Therefore, the whole limb bud mesenchyme of the double mutants assumes a posterior identity. Our interpretation is that the anterior region, being devoid of AER, does not grow out and even regresses. Secondarily, persistent Fgf4 activity at the anterior end of the AER together with Fgf8 activity may lead to the overgrowth of the remaining anterior mesenchyme, resulting in the anomalous shape of the limb observed from 12.5 dpc onwards. Proliferation rate was not noticeably enhanced, as judged by expression of phosphorylated histone H3 or cyclin D1(data not shown). However, at this stage, all limb bud cells are actively dividing, and a slight increase in the proliferation rate anteriorly would not be detected by these methods.

In the limb, posterior identity is associated with the expression of Shh, such that its anterior misexpression leads to pre-axial polydactyly (reviewed by Hill et al.,2003). In the double-null, Shh was weakly expressed at 10.5 dpc (data not shown) but its expression level was similar to control at 11.5 dpc (Fig. 6I,J). At this stage, in accordance with the normal expression of other mesenchymal markers posteriorly, the domain of Shh expression was normal in most cases,with the exception of a slight anterior extension(Fig. 6I,J). However, in some embryos, a small anterior ectopic domain of Shh expression could be observed, exclusively in the forelimbs (four forelimbs out of 14 recovered at 11.5 dpc) (Fig. 6J). The extent of the anterior outgrowth at 12.5 dpc may correlate with the intensity of Shh ectopic expression at earlier stages, but no direct evidence can be provided at present to support this hypothesis. In hindlimbs, Shhexpression appeared as a single posterior domain but slightly displaced distally (not shown).

Many of the double-null limb defects are reminiscent of phenotypes observed when Bmp signalling is experimentally reduced during limb development (see Discussion). Bmp4 has been shown to interact genetically with Msx1 in complex and reciprocal induction processes(Chen et al., 1996). In the limb, Msx genes have been proposed to be downstream targets of the Bmp signal in AER induction (Pizette et al.,2001). In double-null embryos, Bmp4 was not downregulated at early stages of limb development. At 9.5 dpc, when expression of Bmp4 is confined to the ventral ectoderm, its expression domain appeared normal (data not shown). At 10.5 dpc (32-35 somites)(Fig. 7A) and 10.75 dpc (37-40 somites) (data not shown), Bmp4 is expressed strongly in the AER and at a lower level in the mesenchyme where its expression is restricted to the posterior (strong expression) and the anterior (weaker expression) aspects of the limb bud. At these stages, expression of Bmp4 was not significantly altered in the mesenchyme or in the remnant of the AER in the double-null mutant (Fig. 7B and data not shown). At 11.5 dpc, the posterior Bmp4 expression domain was still present in the mesenchyme and the truncated AER also expressed this gene at a normal level. The anterior Bmp4 expression domain was missing, in conformity with the general loss of mesoderm of anterior identity at this stage (Fig. 7C,D). Thus, up to 11.5 dpc, the analysis of Bmp4 expression confirmed the agenesis of the limb anterior domain in double-null embryos but did not reveal any obvious downregulation of this gene in more posterior regions.

At 12.5 dpc, Bmp4 expression, in the mesenchyme, is mainly concentrated in the interdigital spaces and digit anlagen. In the mutant, the Bmp4 transcript level was not altered, with the exception of a slight decrease in the interdigital webbing (data not shown). At 13.5 dpc, Bmp4 expression was present in the digit anlagen of both controls and mutants (Fig. 7E,F). It was also observed in the distal mesenchyme of control embryos(Fig. 7E, arrowhead), but not in the mutant (Fig. 7F),suggesting that Msx genes act upstream of Bmp4 at this site. In addition,expression was observed in the AER that remained in the double-null mutant(Fig. 7F).

Bmp2, contrary to Bmp4, is not normally expressed in the anterior part of the autopod. Similarly, Bmp7 is not expressed anteriorly from 13.5 dpc. In the double-null mutant, at 13.5 dpc, the two genes were also expressed in the anterior domain, in keeping with its posterior identity. Outside this domain, expression pattern was similar but more intense in the mutant when compared with control embryos(Fig. 7G-J).

Gremlin expression was not significantly modified in the double mutant,except that it persisted somewhat longer in the interdigital spaces. Thus,gremlin expression was undetectable at 13.5 dpc in both the fore and hindlimbs of control embryos (Fig. 7K and data not shown) but was still detectable, at a low level, in the interdigital spaces of the sole hindlimbs of the mutants(Fig. 7L and data not shown).

In conclusion, Bmp expression was not significantly modified by the Msx gene mutation. In the mutant, phosphorylation of the Smad1, 5 and 8 proteins could be detected at 10.5, 12.5 and 13.5 dpc, wherever one of the Bmp genes is expressed (data not shown), indicating that Msx proteins do not interfere with the first steps of Bmp signalling.

We report the phenotype of limb buds in the Msx1^-/-;Msx2^-/- double mutant. Three Msx genes have been identified in the mouse genome. Only Msx1 and Msx2 are expressed in the limb, while Msx3 is expressed exclusively in the neural tube(Shimeld et al., 1996; Wang et al., 1996) (reviewed by Bendall and Abate-Shen,2000). We further verified that Msx3 is not upregulated in the limbs of double mutants (data not shown). Therefore, the Msx1^-/-; Msx2^-/- double mutant does not produce any functional Msx protein in the limb. Expression of both genes is prominent in the limb bud, in both the AER and the distal mesenchyme which is under the influence of the AER (Robert et al.,1989; Coelho et al.,1991a; Robert et al.,1991). Although the mechanism of mesodermal cell specification at this site is still controversial (Dudley et al., 2002), the interaction between ectoderm and mesoderm is crucial for limb outgrowth and patterning(Saunders, 1948; Summerbell, 1974; Rowe et al., 1982). Msx gene expression in the limb bud requires functional ectoderm-mesoderm interactions,and it has been proposed that these genes may play a crucial role in limb patterning (Davidson et al.,1991; Robert et al.,1991). In spite of this, limb buds do form in the Msx1^-/-; Msx2^-/- double mutants. All three segments (stylopod, zeugopod, autopod) are present, as well as the scapula/pelvis, and they are only moderately reduced in size along the proximodistal axis. Skeletal elements are nevertheless severely affected. The anterior bone of the zeugopod (radius, tibia) is nearly always truncated and in many cases, missing in the double mutant. Carpal and tarsal bones are also missing anteriorly, and anterior preaxial oligodactyly is frequent. Anterior limb truncation is thus the most constant phenotype observed. However, at the level of the autopod, this truncation is often compensated by a secondary outgrowth of the handplate that extends anteriorly and may lead to polydactyly. Thus, two phases can be distinguished in the function of Msx genes in the limb: an early one (before 11.5 dpc), when the genes are required to specify the anterior limb region; a later one (after 11.5 dpc), when they contribute to restrict the activity of the AER and allow completion of limb patterning.

Relationships between Bmp and Msx proteins have been extensively documented. Msx genes are downstream targets of Bmp proteins at many sites in the embryo (reviewed by Davidson,1995; Bendall and Abate-Shen,2000). As such, they may be required for Bmp signal transduction. Reciprocally, they can act upstream of Bmp genes and are required in some cases for their expression. This double role is best exemplified in the tooth germ where ectodermal Bmp4 induces Msx1 expression in the underlying mesenchyme (Vainio et al.,1993). Subsequently, Msx1 is necessary for mesenchymal Bmp4 expression, as shown by the downregulation of this gene in the tooth germ mesenchyme of the Msx1^-/- mutant(Chen et al., 1996; Zhang et al., 2000). In the palatal mesenchyme, Msx1 is required for expression of both Bmp2 and Bmp4, and a Bmp4-expressing transgene rescues the cleft-palate lethal phenotype of Msx1^-/- homozygous mutants (Zhang et al.,2002).

In the limb, a number of data have shown that Msx genes are downstream targets of Bmp signalling. At early stages, Msx1 and Msx2are expressed in nearly identical patterns that overlap significantly with Bmp genes, namely Bmp2, Bmp4 and Bmp7(Ahn et al., 2001; Pizette et al., 2001). Enhanced Bmp signalling, which is observed in animals heterozygous for the Fused toes (Ft) mutation (Heymer and Rüther, 1999), or in mutants homozygous for a null allele of gremlin, a potent Bmp antagonist(Khokha et al., 2003), leads to an upregulation of both Msx1 and Msx2 expression. Reciprocally, blocking Bmp signalling in the limb ectoderm provokes a decrease of Msx2 expression (Wang et al.,2004). Our data show that, during all phases of limb development,the absence of Msx activity leads to phenotypes that in many respects mimic the absence or decrease in Bmp signalling: delay in both maturation and regression of the AER (Pizette and Niswander, 1999; Wang et al.,2004); anterior upregulation of Fgf4 expression in the AER (Zùniga et al.,1999); tendency to anterior polydactyly(Hofmann et al., 1996; Dunn et al., 1997; Katagiri et al., 1998); and absence of interdigital tissue regression(Guha et al., 2002; Wang et al., 2004). The absence of DV specification anteriorly and the loss of the corresponding part of the AER reproduce locally the effect of Bmp-receptor inhibition(Ahn et al., 2001; Pizette et al., 2001). In limb development, Msx genes seem to act downstream of Bmp signalling, since Bmp expression profiles are little affected in the double mutant, as is the case for Smads.

The AER is established at the boundary between dorsal and ventral ectoderm(Altabef et al., 1997; Michaud et al., 1997; Tanaka et al., 1998; Kimmel et al., 2000). At the molecular level, this boundary takes place between Bmp-expressing(ventrally) and non-expressing (dorsally) domains at the stage when limb buds form. In the chick, Pizette et al.(Pizette et al., 2001) have demonstrated that depletion of the ventral Bmp signal prevents En1expression in the ventral ectoderm and precludes AER formation. Reciprocally,activation of Bmp signalling over the whole limb ectoderm leads to induction of En1 also in the dorsal ectoderm, which adopts a ventral identity,and abrogates AER formation. In addition, activated Bmp receptor gene expression in the dorsal ectoderm induces patches of Fgf8 expression. Based on these and other results, these authors have proposed that Bmp signalling 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. Furthermore, they proposed that Msx genes are downstream effectors of the Bmp signal in AER formation. This is supported by two main lines of evidence: (1) Msx1 and Msx2 display overlapping expression patterns with Bmp4 and Bmp7 in ventral ectoderm at early stages of limb development; and (2) 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 non-expressing cells. In accordance with this model, in the mouse, conditional inactivation of the Bmpr1a gene in limb ectoderm also elicits both DV patterning defects, involving En1 downregulation, and the absence of AER formation (Ahn et al.,2001).

Our data confirm that Msx play an essential role in AER formation. Nevertheless, this role does not fit exactly with the model proposed by Pizette et al. (Pizette et al.,2001). First, in the absence of Msx proteins, AER is induced and maintained over the major part of the limb bud apex, implying that Msx function is required for AER formation only at the anterior margin. Second, in the anterior region where the AER does not form, this is correlated with the absence of a DV boundary, as, based on the expression patterns of Lmx1b, the bud is bi-dorsal in this domain. Therefore, the deficiency in AER formation does not appear to be independent of DV patterning.

In the more posterior part of the double-null limb, the AER forms, but its maturation is delayed and it remains wider than normal. This too may result from impairment in Bmp signalling. Misexpression of noggin from a transgene in the AER and over the limb ventral ectoderm, in the mouse, also leads to ventral extension of the AER (Wang et al.,2004). Our results suggest that in the absence of Msx proteins,Bmp signalling is affected along the whole apex of the limb. Anteriorly, this would result in loss of the AER, agenesis of mesoderm and lack of skeletal elements, but posteriorly only to incomplete AER maturation. Posteriorly, a deficit in Bmp signalling may be of limited effect or, conversely, other genes may fulfil the role of Msx.

From 12.5 dpc to 14.5 dpc, the double-null limb phenotype is characterized by the slow and incomplete regression of the AER and the persistence of the interdigital soft tissues, preventing correct individualization of the digits. Regression of the AER is under the control of Bmp signalling. In the chick,retroviral expression of the Bmp antagonist noggin inhibits AER regression. Accordingly, from stage 24, Msx2 is downregulated in the AER and underlying mesenchyme, whereas Msx1 is downregulated in the posterior part of the apical mesenchyme (Pizette and Niswander, 1999). In the mouse, expression of noggin, from a transgene, in the ectoderm leads to both persistence of the AER and downregulation of Msx2 in this structure but not in the mesenchyme,whereas Bmp4 expression is not modified(Guha et al., 2002; Wang et al., 2004). Thus,during AER regression, Msx genes are probably under the control of Bmp signalling. Our data further suggest that they are required to transduce Bmp activity in the AER because, in the Msx1^-/-;Msx2^-/- double-null mutants, AER regression is impaired from 12.5 dpc while Bmp4 expression is not modified along the apical border of the limb at this stage.

The involvement of Msx genes in interdigital tissue regression by apoptosis has been proposed by several authors (reviewed by Chen and Zhao, 1998). Both Msx1 and Msx2 are expressed in the interdigital tissue at the time when apoptosis takes place(Robert et al., 1989; Coelho et al., 1991b). Furthermore, this expression pattern correlates with the extent of interdigital tissue regression in different species(Gañan et al., 1998). Our results confirm this role because, in the double-null mutant, interdigital tissue webbing fails to regress. Bmp proteins have been identified as the trigger in this process (Yokouchi et al.,1996; Zou and Niswander,1996; Macias et al.,1997; Guha et al.,2002). Whether Msx gene expression is required upstream or downstream of Bmp signalling is controversial(Ferrari et al., 1998; Merino et al., 1999). Our results suggest that, in interdigital webbing regression, Msx would act downstream of Bmp signalling as expression of Bmp2 and Bmp7 is unchanged and that of Bmp4 only slightly decreased at the stage when apoptosis is active.

Between 11.5 and 12.5 dpc, the double-null handplate, particularly in forelimbs, starts undergoing an important anterior overgrowth. This is correlated with a prominent anterior extension of the AER and persistence of Fgf4 and Fgf8 expression up to the anteriormost limit of the elongated AER. Experimental evidence shows that whenever the AER is elongated,or Fgf4 signalling increased, by genetic manipulation, this leads to an overgrowth of the underlying mesenchyme and later on, to anterior polydactyly(e.g. Hofmann et al., 1996; Liu et al., 2003; Wang et al., 2004). This is probably due to the mitogenic properties of Fgf proteins on limb mesenchyme(reviewed by Martin, 1998) but also to the capacity of Fgf4 to recruit cells to the limb bud from adjacent territories (Tanaka et al.,2000). Noticeably, the digits duplicated anteriorly display, in some cases, a posterior identity (Hofmann et al., 1996; Liu et al.,2003).

Anterior outgrowth, leading to polydactyly, may also be due to impairment in the transduction of Bmp signal in the absence of Msx function. Anterior polydactyly is often observed in situations where Bmp signalling is globally diminished, such as in heterozygous Bmp4-null mutants(Dunn et al., 1997), in Bmp7 homozygous mutants (Luo et al., 1995; Hofmann et al.,1996) or in Bmp4^+/-; Bmp7^+/- double heterozygous mutants(Katagiri et al., 1998). This is not necessarily associated with ectopic expression of Shhanteriorly (Dunn et al.,1997). Nevertheless, Bmp4 has been shown to repress Shhin the tooth epithelium and to act downstream of Msx1 in this process. This property is conserved in the limb, although a role for Msx1 at this site was not investigated(Zhang et al., 2000). We observed only a discrete and inconsistent ectopic expression of Shhat the anterior of mutant limb buds. However, this was observed only in the forelimb bud, which tends to produce more frequent and more extensive polydactylies than the hindlimb. In addition, the proportion of limb buds with an ectopic Shh domain corresponds approximately to that of polydactylous limbs at later stages. Furthermore, Shh upregulation may be undetectable by in situ hybridization, but nonetheless sufficiently high to induce extra digits, as exemplified by the raz mutant(Krebs et al., 2003). Therefore, it remains possible that Msx genes are required to repress Shh anteriorly, either directly or by mediating the Bmp signal.

Conversely, Msx genes may play a crucial role in the mechanisms that pattern the anterior limb region. Recent data have shown that Shh is instrumental in patterning digits V to II, and ulna/fibula in the zeugopod(reviewed by Zeller, 2004). The determinants of more anterior structures remain to be identified, and these are precisely the structures affected in the Msx1; Msx2 double null mutant.

We are very grateful to Dr M. Buckingham for critical reading of the manuscript. We are deeply indebted to Dr Paul Barton for his support in the initiation of this project. We are indebted to Dr D. Melton for providing HM1 ES cells and to Drs R. Balling, P. Dollé, B. Hogan, R. Johnson, R. Kelly, G. Martin, A. McMahon, F. Meijling, K. Schughart and R. Zeller for providing probes. This work was supported by the Institut Pasteur, the Centre National de la Recherche Scientifique, and grants from the Association pour la Recherche contre le Cancer and the Association Française contre les Myopathies. A.B. was the recipient of fellowships from the Ministère de la Recherche and the Association pour la Recherche contre le Cancer. C.R. was funded by the European community, then by the Fondation pour la Recherche Médicale.