Multiple roles of mesenchymal β-catenin during murine limb patterning

The complex structure of the vertebrate limb requires the correct formation of early signaling centers that define the primary axes of the early limb bud and the subsequent patterning of internal structures such as skeletal elements(Tickle, 2003). Members of the Wingless/Int (Wnt) family of secreted signaling molecules have been shown to play important roles in both of these processes (Church et al., 2002; Yang, 2003). In chick, Wnt2b and Wnt8c have been implicated in Fgf10activation via the canonical Wnt/β-catenin pathway, thereby inducing limb bud outgrowth (Kawakami et al.,2001; Ng et al.,2002). Fgf10 in the mesenchyme is required for induction of the apical ectodermal ridge (AER) marker Fgf8 in a strip of flank ectoderm before the formation of a morphological AER(Martin, 1998). The AER is a strip of stratified epithelia at the distal limb bud margin that is essential for limb outgrowth and for proximodistal patterning(Saunders, 1948). Induction of Fgf8 by mesenchymal Fgf10 is thought to be mediated by ectodermal Wnt signals, such as Wnt3a in chick and Wnt3 in mouse, which activate and maintain the expression of Fgf8 in the AER through the canonical Wnt/β-catenin pathway(Barrow et al., 2003; Kengaku et al., 1998; Soshnikova et al., 2003). At least two of the β-catenin co-factor genes, lymphoid enhancer binding factor 1 (Lef1) and T-cell factor 1 (Tcf1; Tcf7 -Mouse Genome Informatics), are required for Fgf8 expression in mice(Galceran et al., 1999). However, in Lef1/Tcf1 double-mutant mice, Fgf10 expression is initiated normally, which might be due to residual canonical Wnt signaling,as two other members of the Tcf family, Tcf3 and Tcf4(Tcf7l2 - Mouse Genome Informatics), are still present(Agarwal et al., 2003). Signals from the AER are involved in two important feedback loops, one with the distal mesenchyme required for AER maintenance and further outgrowth of the limb; the other one with the zone of polarizing activity (ZPA), which is important for AER regression (Niswander,2003; Tickle,2003). In the latter, Fgf4, probably together with other Fgfs, maintains the expression of Shh in the ZPA(Chiang et al., 1996; Laufer et al., 1994). Shh in turn downregulates Bmp activity by directly or indirectly inducing the Bmp-antagonist gremlin (Grem1 - Mouse Genome Informatics) (Khokha et al.,2003). Without gremlin, the AER regresses prematurely as a result of the downregulation of Fgf4 (and probably Fgf9 and Fgf17) by Bmps in the posterior AER(Guha et al., 2002; Khokha et al., 2003; Pizette and Niswander, 1999; Zuniga et al., 1999; Zuniga et al., 2004). During normal limb development, this feedback loop stops, leading to regression of the AER when regions of the limb mesenchyme can no longer respond to Shh and, thus, gremlin becomes downregulated(Scherz et al., 2004).

In addition, Wnt signaling is required for dorsoventral patterning of the limb mesenchyme. Wnt7a is expressed only in the dorsal ectoderm because of the activity of the ventral ectodermally expressed engrailed 1(En1) (Loomis et al.,1996; Parr and McMahon,1995). Wnt7a regulates expression of the transcription factor Lmx1b, known as Lmx1 in chick, presumably through a non-canonical signaling pathway (Cygan et al., 1997; Kengaku et al.,1998; Riddle et al.,1995; Vogel et al.,1995). Lmx1b, expressed exclusively in the dorsal mesenchyme, is required for the specification of dorsal limb identity in mouse and chick (Chen et al., 1998; Dreyer et al., 1998; Riddle et al., 1995; Vogel et al., 1995).

Previous loss-of-function studies in mouse addressing the role of canonical Wnt-signaling in limb patterning either solely focused on its role in the ectoderm, or did not allow the differences between the mesenchymal and the ectodermal contribution to be distinguished(Barrow et al., 2003; Galceran et al., 1999; Soshnikova et al., 2003). Therefore, we specifically addressed the contribution of mesenchymalβ-catenin during early limb development using conditional loss- and gain-of-function alleles and the limb mesenchyme-specific Prx1-Cre line. Here,we show that canonical Wnt signaling in the mesenchyme is required for maintaining Fgf10 and Fgf8 expression, which suggests a specific role for mesenchymal β-catenin in AER maintenance, in addition to the one proposed in the ectoderm (Barrow et al., 2003; Soshnikova et al., 2003). Interestingly, we also identified a novel role for active Wnt signaling in AER regression. This effect is concomitant with an upregulation of Bmps and Bmp target gene expression, suggesting that canonical Wnt signaling lies upstream of Bmps in the limb mesenchyme.

Unexpectedly, we found that mesenchymal β-catenin is involved in regulating the dorsal determinant Lmx1b. Loss of β-catenin activity results in a reduced and partial loss of Lmx1b expression,whereas expression of a dominant-active form results in its expansion;however, Wnt7a and En1 remain correctly localized in the dorsal and ventral ectoderm, respectively.

Finally, we found that, unlike many other mutants that affect limb bud outgrowth, loss of mesenchymal β-catenin results in agenesis of the scapula, despite the presence of a humeral-like element. This phenotype is probably primarily due to the loss of expression of the homeobox gene Emx2 in β-catenin mutant limbs, as Emx2 mutants also display severe scapula agenesis(Pellegrini et al., 2001). Emx2 is upregulated in the gain-of-function limbs and is therefore likely to be under the direct control of canonical Wnt signaling.

β-catenin loss-of-function embryos were generated by crossing Prx1Cre/+;β-cat^+/- males, heterozygous for the β-catenin lacZ knock-in allele (-)(Huelsken et al., 2000), to females homozygous for a floxed allele (fl)(Huelsken et al., 2001), to generate Prx1Cre/+;β-cat^fl/- embryos[hereafter called β-cat^ΔPrx1/- or loss of function (lof)]. Gain-of-function embryos were generated by mating female mice homozygous for the β-catenin exon3 floxed allele (ex3fl)(Harada et al., 1999) to Prx1Cre heterozygous males to produce Prx1Cre/+;β-cat^ex3fl/+ embryos [hereafter called β-cat^ex3Prx1/+ or gain-of-function (gof)]. Prx1-Cre line activity was analyzed by crossing males to Z/AP reporter females (Lobe et al.,1999). Mouse embryos were collected and staged either by counting somites or according to Theiler (Theiler,1989). Genotypes of the embryos were determined by PCR and by morphology. Embryos were fixed in 4% paraformaldehyde/PBS at 4°C overnight, washed, and dehydrated into either 100% methanol for whole-mount in situ hybridization or 100% ethanol for paraffin wax embedding. Paraffin sections were cut at 5 μm. Whole-mount alkaline phosphatase staining was performed as described previously (Lobe et al., 1999).

Embryos were skinned, eviscerated and fixed in 95% ethanol. Skeletons were stained with Alizarin Red and Alcian Blue, or just with Alcian Blue(McLeod, 1980). For SEM analysis embryos were fixed overnight in 2% glutaraldehyde/0.1 M sodium cacodylate buffer (SCB), pH 7.2, washed in SCB, postfixed for 90 minutes in 1%osmium tetroxide, washed in SCB, and dehydrated into 100% ethanol and critical-point dried.

Whole-mount and section in situ hybridisation was performed as described previously (Murtaugh et al.,1999; Riddle et al.,1993). For sectioning of whole-mount, stained embryos, embryos were embedded in gelatin and vibratome sectioned at 20 μm. Double-fluorescent in situ hybridisations on paraffin sections were performed using biotin- and DIG-labelled probes. After hybridisation, slides were washed, quenched and blocked. Probes were detected by incubation with streptavidine-HRP (Perkin Elmer, diluted 1 in 100) and anti-DIG-HRP (Roche,diluted 1 in 50), followed by Cy3- or Cy5-tyramide-labelled fluorescent dyes(according to instructions of the TSA Plus Fluorescent Systems kit, Perkin Elmer). All probes have been described previously. The minimum number of specimens analyzed for each marker was three for all stages, except for AER measurements (see legend to Fig. 3), and two for SEM, unless otherwise stated. All comparative images are at the same magnification, unless otherwise noted. AER measurements were performed on Fgf8-stained limbs by measuring the pixel number for each limb using Metamorph.

Immunohistochemistry was performed on paraffin sections using monoclonal mouse anti-β-catenin (BD Transduction Laboratories, diluted 1 in 200)after heat-induced antigene retrieval. The signal was detected using a biotinylated secondary antibody (diluted 1 in 250; Vector Laboratories) in combination with the ABC Kit (Vector Laboratories) and DAB substrate (Sigma). TUNEL assays were performed using the Flourescein In Situ Cell Death Kit(Roche).

Micromass cultures were performed as described previously(Hill et al., 2005). Primer sequences for RT-PCR are available upon request. Total RNA (1 μg), isolated using TRIZOL from cultures 12 hours and 14 hours after plating and infection,was used for reverse transcription. Semi-quantitative RT-PCR was performed using fivefold dilutions of the cDNA template, starting at a 1 in 10 dilution.

In order to address the role of canonical Wnt signaling in the mesenchyme during limb development, we analysed conditional β-catenin loss- and gain-of-function embryos using the mesoderm-specific Prx1Cre line(Logan et al., 2002). Loss-of-function β-cat^ΔPrx1/- embryos(Fig. 1B) died shortly after birth and had severely shortened fore- and hindlimbs, with respect to wild type (Fig. 1A). Notably,hindlimbs developed further than forelimbs, as can be seen in skeletal preparations at E13.5 and E14.5 (see Fig. S1 in the supplementary material). Gain-of-function β-cat^Δex3Prx1/+ embryos also had shortened limbs (Fig. 1C). Similar to β-cat^ΔPrx1/- embryos, forelimbs were more severely affected than hindlimbs (Fig. 1C). From E13.5 onwards, formation of haematomes in distal regions could be observed, primarily but not exclusively, in the hindlimbs of both mutants.

Deletion of β-catenin in the limb ectoderm has been shown to result in patterning defects (Barrow et al.,2003; Soshnikova et al.,2003). Therefore, we confirmed, by in situ hybridization using an antisense probe specific to the floxed exons(mβ-catenin del), that β-catenin was still expressed in the ectoderm and the AER inβ -cat^ΔPrx1/- limbs(Fig. 1E-G). Although wild-type limbs displayed uniform β-catenin expression at E10.5(Fig. 1D),β -cat^ΔPrx1/- forelimbs showed a high degree of deletion in the limb and flank mesenchyme at E10.5(Fig. 1E, red and black arrow,respectively), and almost complete deletion at E11.5(Fig. 1G). By contrast, only patchy deletion was detected in the hindlimb mesenchyme at E10.5(Fig. 1F, red arrow). As has been previously published, by E11.5 deletion had occurred in most mesenchymal cells of the hindlimb (Hill et al.,2005). ZAP-reporter analysis of the Prx1-Cre line revealed that deletion in the forelimb occurred in a mosaic pattern as early as somite stage 15 (see Fig. S2B,B′ in the supplementary material). The different kinetics of deletion observed between fore- and hindlimb(Fig. 1E-G) (see also Logan et al., 2002), are probably responsible for the differences in fore- and hindlimb development of the mutants.

Loss of β-catenin in the mesenchyme resulted in severely shortened fore- and hindlimbs. Scanning electron microscopy (SEM) analyses of the forelimbs revealed that the β-cat^ΔPrx1/- limb bud was slightly smaller along the anteroposterior (AP) axis at E9.5(Fig. 2A,A′). At E10.5,β -cat^ΔPrx1/- limbs were growth retarded and showed notches in the AER (Fig. 2B,B′). Although we did not observe any gross morphological difference at E9.5, it is possible that molecular changes contributing to the later phenotype have occurred. We therefore analyzed the expression of genes involved in limb outgrowth. Fgf8, one of the earliest known ectodermal markers for limb outgrowth, was expressed in its normal stripe-like pattern at 17 somites, before any apparent limb outgrowth had occurred, in wild-type and β-cat^ΔPrx1/- limbs(Fig. 2C,C′). At 20 somites, Fgf8 expression was still comparable inβ -cat^ΔPrx1/- and wild-type limbs(Fig. 2D,D′). However, at 25 somites, before the morphological appearance of an AER, Fgf8expression was reduced in a patchy manner inβ -cat^ΔPrx1/- limbs (see arrows in Fig. 2E,E′). Shortly after the formation of the stratified AER at stage E10.5, Fgf8expression was restricted to patches along the distal margin inβ -cat^ΔPrx1/- limbs(Fig. 2F,F′). The patchy downregulation of Fgf8 in the AER and the subsequent loss of the AER suggest a defect in AER maintenance upon loss of β-catenin activity in the mesenchyme.

β-catenin had been implicated in the regulation of Tbx5 in chick (Kawakami et al., 2001)and Fgf10 in mouse (Agarwal et al., 2003) during limb bud initiation. We therefore examined their expression in β-cat^ΔPrx1/- limbs. Around 19-20 somites, the expression of Tbx5 and Fgf10 inβ -cat^ΔPrx1/- forelimbs was not altered(Fig. 2G,G′,I,I′),suggesting either that these genes are not directly controlled byβ-catenin or that the Prx1-Cre line does not delete early and broadly enough to see an effect. However, at E10.75, Fgf10 expression was downregulated (Fig. 2H,H′), although Tbx5 expression was still maintained (Fig. 2J,J′,see also Fig. S3A-A″ in the supplementary material).

Because Fgf8 expression and functional AER activity was significantly affected in forelimbs lacking β-catenin, we analysed whether a gain of mesenchymal β-catenin activity would result in the ectopic expression of AER markers. Fgf8 was expressed ectopically in distal patches in β-cat^Δex3Prx1/+ forelimbs at E10.5 (data not shown) and E11.5 (Fig. 3A′, red arrow). However, this was not observed for other AER markers, such as Fgf4 (Fig. 3G′, E10.5), Fgf9(Fig. 3H′, E11.5) and Wnt3 (data not shown). Fgf8 section in situ hybridizations revealed that the ectopic expression was mesenchymal (see Fig. S3B′ in the supplementary material).

In addition to being distally ectopically expressed, Fgf8 was prematurely lost in β-cat^Δex3Prx1/+ forelimbs from the posterior AER at E11.5 (Fig. 3A′, white arrow). At E12.5, in comparison to wild-type limbs, Fgf8 expression was either completely absent, or only maintained in the most anterior AER (Fig. 3B″, data not shown),(Fig. 3B). By contrast,although no ectopic Fgf8 expression was detected inβ -cat^Δex3Prx1/+ hindlimbs at E11.5(Fig. 3A′) or E12.5(Fig. 3B′), Fgf8expression was also lost from the posterior AER from E12.5 onwards(Fig. 3B′). In order to quantify the extent of AER loss, we measured the Fgf8 expressing region in the fore- and hindlimbs of wild-type andβ -cat^Δex3Prx1/+ embryos at E10.5 and E11.5(Fig. 3C). No significant differences in AER length were detected at E10.5 between wild type andβ -cat^Δex3Prx1/+. However, at E11.5 there was a significant reduction in both fore- (45.4%) and hindlimbs (7.9%) ofβ -cat^Δex3Prx1/+ embryos(Fig. 3C). SEM analyses were performed at E10.75 and E11.5. These revealed slight morphological differences between wild-type (Fig. 3D,E)and β-cat^Δex3Prx1/+(Fig. 3D′,E′)forelimbs at E10.75, such as a flattened posterior AER (bracket) and differences in the shape (arrowhead). At E11.5, the AER was only visible in the anterior region of β-cat^Δex3Prx1/+forelimbs (Fig. 3F′,between arrowheads).

As other FGF molecules have been implicated in AER maintenance, we analyzed Fgf4, Fgf9, Fgf17 and Fgf10 expression inβ -cat^Δex3Prx1/+ limbs. Fgf4, expressed in the posterior AER of wild-type forelimbs up to around E11.5, was severely reduced in the posterior AER of β-cat^Δex3Prx1/+forelimbs at E10.5 (Fig. 3G′). Similarly, Fgf9 expression was also prematurely lost in β-cat^Δex3Prx1/+ forelimbs at E11.5 (Fig. 3H,H′). Fgf17 was already downregulated in the AER of both wild-type andβ -cat^Δex3Prx1/+ forelimbs around E10.5 (data not shown). Mesenchymal expression of Fgf10 was not altered at early stages, but was reduced in β-cat^Δex3Prx1/+forelimbs from E11 onwards (Fig. 3I,I′; and data not shown). The premature loss of AER markers and Fgf10 in gain-of-function limbs suggests premature regression of the AER.

AER regression has been linked to a cross-talk between ZPA and AER. Therefore, we analysed the expression of Shh, the key ZPA morphogen. Shh expression was downregulated inβ -cat^Δex3Prx1/+ forelimbs starting at E11.0(data not shown) and was nearly completely lost at E11.5(Fig. 3J,J′). Similarly, Ptc1, which serves a molecular readout for Shh signaling was highly reduced in E11.5 β-cat^Δex3Prx1/+ limbs(Fig. 3K,K′).

We analyzed whether the levels of Bmps and their target genes were changed in β-cat^Δex3Prx1/+ limbs, as Bmp activity has been implicated in AER regression. At E11.5, Bmp4 and Bmp7were upregulated throughout β-cat^Δex3Prx1/+fore- and hindlimbs (Fig. 4A′,B′,H, and data not shown), whereas Bmp2was only upregulated distally (Fig. 4C′). All three Bmp genes were highly expressed in the most posterior region, including the AER (Fig. 4A′,B′,C′). In addition, the Bmp antagonist gremlin was upregulated (Fig. 4D′), as were the Bmp target genes Msx1 and Msx2 in the distal limb (Fig. 4E′,F′). Vibratome sections of whole-mount stained embryos (Fig. 4G-G″), and in situ hybridization on sections (Fig. 4H, and data not shown) demonstrated that the ectopic expression was mesenchymal. Upregulation of Bmp genes and gremlin occurred earlier(E10.5-E10.75) and more extensively in fore- than in hindlimbs (data not shown), implying that the upregulation is probably dependent on the level of activated β-catenin, as Cre recombination occurred earlier in fore- than in hindlimbs. Conversely, levels of Bmp2, Bmp4, Bmp7, gremlin, Msx1 and Msx2 were reduced inβ -cat^ΔPrx1/- limbs at E11.5 (data not shown). Our data suggest that the limb truncations inβ -cat^Δex3Prx1/+ embryos are due to a premature regression of the AER, caused by an upregulation of Bmp gene expression and activation of Bmp signaling.

The AER defects observed in β-catenin lof and gof limbs could also be due to excessive cell death in the distal mesenchyme or AER. We therefore performed TUNEL assays on transverse sections throughβ -cat^ΔPrx1/- andβ -cat^Δex3Prx1/+ limbs at different developmental stages. At 25 somites, corresponding to the onset of downregulation of Fgf8 expression, no significant difference in apoptosis was observed (Fig. 5A′). However, more TUNEL-positive cells were detected in the proximal mesenchyme of β-cat^ΔPrx1/-forelimbs at E10 (Fig. 5B′). At E11.5, β-cat^ΔPrx1/-hindlimbs showed an extensive area of cell death in the central core of the limb bud (data not shown). In gof forelimbs, an increased number of TUNEL-positive cells, predominately scattered throughout the proximal mesenchyme, was detected at E11 (Fig. 5C′). At E11.5, apoptosis had further increased and was more widespread, including distal regions (data not shown). The amount of cell death in the AER was comparable inβ -cat^Δex3Prx1/+ and wild-type limbs(Fig. 5C,C′). The temporal occurrence of apoptosis increase in the mesenchyme of gof forelimbs corresponds to Bmp upregulation, suggesting that they may play a role. However, gof hindlimbs, in which Bmps were also expressed ectopically, showed no increase in TUNEL-positive cells at E11.5 (data not shown).

Dorsoventral limb patterning is determined by the correctly localized expression of three genes: Wnt7a and En1, expressed in dorsal and ventral ectoderm, respectively, and Lmx1b, which is regulated by Wnt7a and expressed in the dorsal mesenchyme. At E10.5,the ectodermal regulators of Lmx1b, Wnt7a(Fig. 6A′,A″) and En1 (Fig. 6B′,B″) were expressed in both β-cateninlof and gof limbs similar to wild type(Fig. 6A,B). Loss ofβ-catenin activity resulted in reduced and partial loss of Lmx1bexpression in the forelimb already at E9.5(Fig. 6C′), when compared with wild type (Fig. 6C). Conversely, in β-catenin gof limbs, Lmx1b expression was expanded into the ventral mesenchyme at E9.5(Fig. 6D′, black arrow)compared with wild type (Fig. 6D). At E11.5, Lmx1b expression was maintained inβ -cat^ΔPrx1/- forelimbs, primarily in regions adjacent to the remnants of a functional AER(Fig. 6E′). Section in situ hybridizations on E10.5 β-cat^ΔPrx1/-forelimbs revealed an overall downregulation of Lmx1b and an even more striking downregulation in the dorsal-most regions, primarily in the distal half (Fig. 6F′). The expansion of Lmx1b into the ventral mesenchyme was clearly visible on E10.5 β-cat^Δex3Prx1/+ forelimb sections. However, a few Lmx1b-negative patches were observed within the dorsal and ventral mesenchyme (Fig. 6F″). These areas were not apoptotic, as confirmed by TUNEL staining (data not shown). Notably, no ectopic Lmx1b expression directly adjacent to the ventral ectoderm(Fig. 6F″, black arrow)was observed in any of the examinedβ -cat^Δex3Prx1/+ forelimb sections(n=3). In the hindlimbs, small patches of cells expressing Lmx1b adjoining the ventral ectoderm were occasionally detected(n=1/3; data not shown).

The correct localisation of Wnt7a and En1 in the ectoderm of β-cat^Δex3Prx1/+ embryos suggested a cell-autonomous effect on Lmx1b. To further investigate this, the distribution of stabilized β-catenin protein was compared with Lmx1b,Lef1 and Tcf1 expression inβ -cat^Δex3Prx1/+ hindlimbs on adjacent sections at E11.5. This revealed high levels of β-catenin protein throughout the entire mesenchyme (black arrow in Fig. 6G′,G″; compared with the trunk, white arrow),intermingled with areas with lower levels(Fig. 6G′,G″;arrowhead). The latter were probably the result of incomplete Cre-recombination. Lmx1b expression always correlated with the presence of stabilised β-catenin (Fig. 6G′,H′). However, in the ventral-proximal region, Lmx1b was not upregulated, despite the presence of high levels of stabilized β-catenin (Fig. 6G′,H′ black arrows). The two known target genes and transcriptional co-regulators of β-catenin Lef1 and Tcf1 (Hovanes et al.,2001; Kengaku et al.,1998) were also upregulated in areas with high levels ofβ-catenin (Fig. 6I′-J′). In contrast to Lmx1b, they were upregulated in ventral-proximal regions (black arrows in Fig. 6I′,J′). Like Lmx1b, Lef1 and Tcf1 expression were downregulated inβ-catenin lof limbs (see Fig. S3D″,E″ in the supplementary material). Double-fluorescent section in situ hybridization revealed that Lmx1b-positive cells also upregulated Lef1 and Tcf1(Fig. 6H″,I″,K, and data not shown). Our results suggest that Lmx1b is probably regulated, in part, by the canonical Wnt signaling pathway.

Early limb patterning defects are often reflected in alterations of internal structures, such as the skeletal elements. Whole-mount Alizarin Red/Alcian Blue skeletal staining of E18.5 embryos revealed the presence of a hypoplastic scapula in the β-cat^ΔPrx1/-forelimb. This consisted only of a small remnant of, presumably, the scapula blade (Fig. 7A′,arrowhead) and of a small, round element attached to the proximal end of an element, which is likely to result from incomplete separation of the humerus/radius/ulna anlage (Fig. 7A′, arrow). Section in situ hybridizations on E12.5 forelimbs using probes for collagen 2α1 and Gdf5 showed that part of the scapula anlage (Fig. 7B, arrowhead) was absent inβ -cat^ΔPrx1/- embryos(Fig. 7B′, arrowhead). The Gdf5 staining showed that a separate element articulated with the`humeral' end (Fig. 7D′,green arrow). At E14.5, this element (Fig. 7C, black arrow) appeared to be partially fused to the `humerus'(Fig. 7C,E). This element probably corresponds to the glenoid fossa that articulates with the humerus.

As the skeletal elements of loss- and gain-of β-catenin embryos were either severely malformed or almost completely missing, respectively, we analyzed various proximodistal markers, such as Hoxc6, Hoxa9, Hoxa10,Meis1 and Meis2, at E10.5(Fig. 7F-I″), to see whether limb patterning was still normal(Capdevila et al., 1999; Favier et al., 1996; Fromental-Ramain et al., 1996; Mercader et al., 1999; Nelson et al., 1996; Wahba et al., 2001). The expression of Hoxc6, normally expressed in the proximo-anterior region (Fig. 7F), shifted distally in β-cat^ΔPrx1/-(Fig. 7F′), but was not altered in β-cat^Δex3Prx1/+ limbs(Fig. 7F″). The Hox genes Hoxa9 and Hoxa10, expressed in the distal regions of the limb (Fig. 7G,H), remained distally expressed in lof (Fig. 7G′,H′) and gof(Fig. 7G″,H″)limbs. The proximal markers Meis1 and Meis2(Fig. 7I, and data not shown)remained proximal in lof (Fig. 7I′) and gof (Fig. 7I″) limbs.

Because the skeletal preparations suggested abnormal scapula development,we analyzed Pax1 and Emx2, two genes that have been shown to be important for scapula formation. Genetic inactivation of Pax1results in a specific loss of the acromion, whereas Emx2 null mice maintain only the glenoid fossa and a small piece of the mediovertebral border of the scapula (Pellegrini et al.,2001; Timmons et al.,1994; Wilm et al.,1998). In wild type, Pax1 and Emx2 were normally expressed in the proximo-anterior forelimb mesenchyme at E10.5(Fig. 7J,K), whereas their expression was highly reduced in β-cat^ΔPrx1/-forelimbs (Fig. 7J′,K′). Section in situ hybridization revealed that Emx2 was still expressed in a small patch of cells in the anterior-proximal region of β-cat^ΔPrx1/- limbs(data not shown).

In β-cat^Δex3Prx1/+ limbs, the Pax1domain was slightly reduced (Fig. 7J″), whereas Emx2 expression was upregulated throughout the proximal limb at E11 (Fig. 7K″). These data suggest that Emx2, but not Pax1, could be a direct target of canonical Wnt signaling in the developing limb. Furthermore, agenesis of the scapula in lof β-catenin mutants is probably due to the downregulation of Emx2 and Pax1 in β-cat^ΔPrx1/- mutant embryos.

Using an in vitro system, we assayed for expression changes of various genes, including Bmp2, Bmp4, Bmp7, Lmx1b, Lef1, Tcf1, Emx2 and Twist, which were upregulated in theβ -cat^Δex3Prx1/+ limbs (see Figs 4, 6, 7, and Fig. S3C′,D′,E′ in the supplementary material). Semi-quantitative RT-PCR analysis was performed on RNA from high-density micromass cultures from E11 ex3fl/ex3fl embryos infected with an Adeno-Cre virus, resulting in the accumulation of significant amounts of the N-terminal truncated stabilized form of β-catenin 8-10 hours after transfection(Hill et al., 2005). Control cultures were infected with an Adeno-Gfp virus. Bmp4, Bmp7, Lef1,Tcf1 and Emx2 (Fig. 7L, see also Fig. S4 in the supplementary material) were upregulated in this assay, similar to the in vivo observations. Surprisingly, Lmx1b, and also Twist and Bmp2, were not upregulated in vitro, unlike in vivo (Fig. 7L, Figs S3, S4 in the supplementary material). Like in vivo,expression of Tbx5 and Fgf10 was not altered (see Fig. S4 in the supplementary material). Sox9 was downregulated, as previously shown (Fig. S4 in the supplementary material)(Hill et al., 2005). The in vitro results suggest that Tcf1, Lef1, Bmp4, Bmp7 and Emx2could be direct targets of active β-catenin signaling in the limb. However, Lmx1b and also Twist, which cannot be upregulated by β-catenin in vitro, might be under the co-regulatory control of aβ-catenin-mediated Wnt signal and another signal emanating from the ectoderm in vivo.

Based on studies in chick, it has been proposed that canonical Wnt signaling, together with Tbx5, is essential for forelimb initiation and outgrowth (Kawakami et al.,2001; Ng et al.,2002). So far, no Wnt mutant in mouse has been described where limb initiation failed. Similarly, limb initiation still occurs in Lef1/Tcf1 double mutants; however, this might be due to the presence of Tcf3 and Tcf4 (Agarwal et al., 2003). By contrast, inactivation of Tbx5 activity results in a complete absence of forelimbs or pectoral fins(Agarwal et al., 2003; Ahn et al., 2002; Garrity et al., 2002; Ng et al., 2002; Rallis et al., 2003). Despite the fact that we used the same Cre-deleter line (Prx1-Cre) as in a previous study demonstrating that Tbx5 is essential for limb initiation(Rallis et al., 2003), we observed limb outgrowth and initially normal expression of Tbx5,Fgf10 and Fgf8. This would be in agreement with the previously suggested model that Tbx5 acts upstream of Wnt/Fgf signaling(Agarwal et al., 2003; Takeuchi et al., 2003). However, the Prx1-Cre line is probably not suitable to address the question of whether canonical Wnt/β-catenin is required for Tbx5 initiation,as Tbx5 expression is observed at 8 somites (E8.5)(Agarwal et al., 2003), whereas Prx1-Cre was active only from 15 somites onwards (see Fig. S2 in the supplementary material).

Contrasting with a previous in vitro observation showing thatβ-catenin can activate a Fgf10 promoter construct(Agarwal et al., 2003), our results suggest that Fgf10 is not regulated by Wnt/β-catenin signaling. The observed Fgf10 downregulation is probably due to a disruption in the Fgf8/Fgf10-feedback loop, as Fgf8expression in the AER is clearly reduced prior to Fgf10expression.

Our data show that mesenchymal β-catenin, together with ectodermal Wnt/β-catenin signaling, is required for Fgf8 expression and AER maintenance (Barrow et al.,2003; Soshnikova et al.,2003); however, the underlying molecular mechanism needs to be further investigated, as Fgf10 does not seem to be the direct target. Possible targets could be brachyury or Twist, which are regulated by canonical Wnt/β-catenin signaling in other tissues(Galceran et al., 2001; Howe et al., 2003; Yamaguchi et al., 1999b). They are expressed in the distal mesenchyme and have previously been implicated in AER maintenance (Liu et al.,2003; Zuniga et al.,2002). Twist is the most likely candidate, as it was upregulated in gain-, and downregulated in loss-of-function β-catenin limbs (see Fig. 3C′,C″).

The proximodistal shortening ofβ -cat^Δex3Prx1/+ limbs is due to premature regression of the AER starting posteriorly, whereby loss of Fgf4 and Fgf9 preceded that of Fgf8. Downregulation of Shhcoincided with the loss of Fgf4 and Fgf9, suggesting that the regression is caused by a disruption in the ZPA/AER-feedback loop(Sun et al., 2000). The molecular mechanism underlying the observed AER regression is probably theβ-catenin-mediated upregulation of Bmp genes, as BMPs can induce AER regression (Guha et al., 2002; Khokha et al., 2003; Pizette and Niswander, 1999; Zuniga et al., 1999). This activity is normally antagonized by gremlin(Khokha et al., 2003). Gremlin was also upregulated upon activation of β-catenin; however, we think that this is a secondary effect, because concurrent expansion of the Bmp target genes Msx1 and Msx2 suggests that active Bmp signaling was occurring in β-cat^Δex3Prx1/+ limbs, and because Bmps can induce gremlin expression(Pereira et al., 2000).

It is interesting to note that Wnt5a, which is expressed in the distal limb mesenchyme, has recently been shown to downregulate β-catenin levels (Topol et al., 2003). Wnt5a^-/- embryos show a suppression of Sox9expression and chondrogenesis in the distal structures, resembling the phenotype of β-cat^Δex3Prx1/+(Hill et al., 2005; Topol et al., 2003). However,no AER defects have been reported and the expression of Bmp genes was not analyzed in these limbs (Yamaguchi et al.,1999a). These discrepancies could be due to differences in the level or timing of β-catenin stabilization betweenβ -cat^Δex3Prx1/+ and Wnt5a^-/- embryos. It would certainly be interesting to see which aspects of the Wnt5a limb phenotypes could be rescued by removing β-catenin. Proper limb development probably requires the fine-tuning of β-catenin levels in the distal mesenchyme through the Wnt5a destabilizing activity and the stabilizing activity of canonical Wnts from the AER.

Lmx1b expression has been shown to be regulated by Wnt7ain chicken and mouse (Cygan et al.,1997; Loomis et al.,1998; Riddle et al.,1995; Vogel et al.,1995); however, it is probably not mediated by the canonicalβ-catenin pathway (Kengaku et al.,1998). Surprisingly, we found that the loss of β-catenin activity in mice resulted in a downregulation of Lmx1b as early as E9.5 in the dorsal mesenchyme. Concomitantly, overactivation of β-catenin resulted in a cell-autonomous upregulation of Lmx1b; however, not all regions of the limb were capable of responding. This, together with the observation that Lmx1b expression could not be upregulated in vitro,unlike a number of other genes that were also upregulated in vivo, suggests that Lmx1b regulation by β-catenin requires an additional signaling input.

Despite the recent data showing that Wnt7a signaling in the presence of Lrp6 stabilizes β-catenin (Adamska et al., 2005; Caricasole et al.,2003), we do not think that Wnt7a uses the canonical pathway to regulate Lmx1b. Especially because Lmx1bexpression is only downregulated in the distal mesenchyme of Wnt7aand Lrp6 knockout mice at E11.5 and E12.5, respectively(Adamska et al., 2005; Cygan et al., 1997), whereas in β-cat^ΔPrx1/- limbs it is reduced as early as E9.5. Lmx1b is also upregulated inβ -cat^Δex3Prx1/+ limbs around this stage. By contrast, ectopic Wnt7a expression in En1^-/-limbs from E9.5 onwards leads to ventral expansion of Lmx1b only at E11.0 (Cygan et al., 1997; Loomis et al., 1998). Therefore, we favor the hypothesis that, in addition to Wnt7a, one or more Wnt(s) are involved in Lmx1b regulation, and that these other Wnt(s) signal through the canonical pathway. Possible candidates would be Wnt3, Wnt4, Wnt6 and Wnt7b, which are expressed in the ectoderm (Bennett et al.,2002). However, these Wnts are expressed throughout the limb bud ectoderm and therefore should activate canonical Wnt/β-catenin signaling in both dorsal and ventral mesenchyme adjacent to the ectoderm. This would correspond to the observed high levels of β-catenin protein, and the expression of Tcf1 and Lef1 in the subectodermal mesenchyme(Fig. 6). Therefore, we propose that β-catenin-dependent Lmx1b expression requires probably an additional factor, emanating from the dorsal ectoderm or restricted to the dorsal mesenchyme. This is in agreement with the observation that the in vitro stabilization of β-catenin was insufficient to upregulate Lmx1b. Alternatively, a secreted repressor of Lmx1b may be present in the ventral ectoderm, which could explain why ectopic Lmx1b expression was never observed in the ventral-most mesenchyme. Based on our results, we favor a combination of both mechanisms in order to restrict Lmx1b to the dorsal mesenchyme. Hence, the ventral ectoderm might play a dual role in preventing Lmx1b activation in the ventral mesenchyme: by expressing En1, which inhibits Wnt7a, and by the expression of a putative secreted inhibitor interfering specifically withβ-catenin-mediated activation of Lmx1b. Interestingly, loss ofβ-catenin in the ectoderm also affects dorsoventral patterning, but this is due to loss of En1 (Barrow et al., 2003; Soshnikova et al.,2003).

β-cat^ΔPrx1/- embryos display severe agenesis of the scapula. Although the process of skeletogenesis is closely connected to that of limb patterning, scapula agenesis in mice lacking β-catenin in the limb mesenchyme is probably not caused by the changes observed in skeletal differentiation (Hill et al.,2005). Lack of β-catenin does not inhibit chondrocyte differentiation, on the contrary it stimulates it(Day et al., 2005; Hill et al., 2005). However,in β-cat^ΔPrx1/- limbs the cartilaginous scapula anlage is not formed (Fig. 7B′), resembling the Emx2^-/- phenotype(Pellegrini et al., 2001). Concomitantly, Emx2, but also the transcription factor Pax1,which is important for scapula formation as well, are severely downregulated(Pellegrini et al., 2001; Timmons et al., 1994; Wilm et al., 1998). The loss of these two transcription factors during early limb bud development is probably the cause for scapula agenesis. β-catenin seems to be necessary and sufficient for Emx2 expression, making Emx2 a potential direct target of canonical Wnt/β-catenin signaling in the limb mesenchyme, as has been previously shown in the telencephalon(Backman et al., 2005; Theil et al., 2002). The observed increase in cell death in the proximal region ofβ -cat^ΔPrx1/- limbs at E10 could in principal contribute to scapula agenesis; however, a similar pattern of cell death is observed in Fgf4/Fgf8 double mutants, and even more extensive cell death is found in Shh and gremlin mutant limbs, but those develop a normal scapula (Boulet et al., 2004; Michos et al.,2004; Sun et al.,2002; te Welscher et al.,2002).

The loss of the scapula, along with the other appendicular skeletal elements, in β-cat^Δex3Prx1/+ forelimbs is due to the fact that stabilization of β-catenin results in a complete block of skeletogenesis, as has been shown previously(Hill et al., 2005), and is likely to be independent of the changes in Emx2 and Pax1expression.

Our results provide strong genetic evidence that, in the limb mesenchyme,β-catenin activity is required for AER maintenance, and also, eventually,for its regression, via the control of Bmps. Furthermore, mesenchymalβ-catenin signaling is required for dorsal mesenchyme identity through regulation of the dorsal selector gene Lmx1b, and, in addition, is required for scapulagenesis, principally via the regulation of Emx2.

We thank A. Neubüser, M. Torres, N. G. Copeland, and C. J. Tabin for probes; M. Taschner for technical advice on fluorescent in situ hybridisation;M. Logan for the Prx1-Cre mice; H. Halbritter for SEM assistance; and M. Busslinger and E. Wagner for critical comments. This research was supported by Boehringer Ingelheim.