BMPR-IA signaling is required for the formation of the apical ectodermal ridge and dorsal-ventral patterning of the limb

Classical transplantation experiments, and more recently molecular studies, demonstrate that the vertebrate limb is regulated along three cardinal axes; proximal-distal (PD), dorsal-ventral (DV) and anterior-posterior (AP) (Johnson and Tabin, 1997; Martin, 1998). Each of these cardinal axes is regulated by an organizing center within the developing limb. The PD axis is dependent on signals emanating from the apical ectodermal ridge (AER). The AP axis is regulated by the zone of polarizing activity (ZPA) on the posterior margin of the limb. The DV axis is regulated by non-ridge ectoderm. In each case, recent advances have elucidated the molecular mechanisms that mediate the function of these organizing centers. However, much less is known about the mechanisms that establish the formation of these organizing centers. We will address the molecular mechanisms required to establish the AER and DV patterning.

Limbs grow and develop with proximal structures forming first, and then progressively distal structures forming with time. The AER forms as a specialized epithelial structure at the distal DV border of the developing limb bud (Tickle and Altabef, 1999). Classical transplantation experiments demonstrate a crucial role for the AER in limb outgrowth and this, in turn, leads to the proper formation of structures along the PD axis (Saunders, 1948). Removal of the AER very early in limb development drastically stunts the growth of the limb bud. If the AER is removed at progressively later stages of development, then the more distal structures are formed (Saunders, 1948). Fibroblast growth factors (FGFs) mediate the function of the AER. Several FGFs are expressed normally in the AER, including Fgf2, Fgf4, Fgf8, Fgf9 and Fgf17 (Cohn et al., 1995; Niswander and Martin, 1993; Sun et al., 2000). Beads soaked in FGFs can largely replace the function of the AER (Niswander et al., 1993), and induce ectopic limbs in ovo (Cohn et al., 1995). Previous transplantation experiments have demonstrated that limb mesoderm induces the AER (Carrington and Fallon, 1984; Saunders and Reuss, 1974). However, the molecular mechanisms that mediate this induction are not well understood.

Complex interactions between ectoderm and mesoderm regulate DV patterning during limb development (Chen and Johnson, 1999). Before limb bud formation, inductive signals from the mesoderm are required to establish DV pattern in the overlying ectoderm of the limb field (Geduspan and MacCabe, 1987; Geduspan and MacCabe, 1989; Michaud et al., 1997). Using transplantation analyses in chick/quail chimeras, Le Douarin and colleagues demonstrated that signals that ventralize limb ectoderm emanate from the lateral mesoderm, whereas signals that dorsalize limb ectoderm are derived from more medial somitic mesoderm (Michaud et al., 1997). However, the molecular mechanisms that mediate these mesoderm derived signals are unknown.

Once the presumptive limb ectoderm is induced by the mesoderm, it then plays a primary role in specifying DV pattern as the limb bud forms (Carrington and Fallon, 1984; Michaud et al., 1997; Saunders and Reuss, 1974). The molecular mechanisms that regulate DV patterning by the limb ectoderm have been characterized by molecular genetic analyses, which are summarized in Fig. 8A,B. The homeodomain gene, Engrailed 1 (En1), is expressed in the ventral ectoderm of the limb, where it specifies ventral limb identity (Davis et al., 1991; Gardner and Barald, 1992; Loomis et al., 1996). Null mutations in the En1 gene dorsalize the ventral limb ectoderm and subsequently the distal region of the limb (Cygan et al., 1997; Loomis et al., 1996; Loomis et al., 1998). These genetic analyses suggest a model in which En1 is the first step in specifying DV patterning in the ectoderm of the limb (Fig. 8B). Knockout analyses demonstrate that En1 specifies ventral limb identity, at least in part, by suppressing the expression of the growth factor, Wnt7a (Cygan et al., 1997; Loomis et al., 1998). Wnt7a expression is normally restricted to the dorsal ectoderm (Dealy et al., 1993; Parr et al., 1993), and null mutations in the Wnt7a gene give a double ventral phenotype, which includes the formation of ventral foot pads on the dorsal limb instead of claws and ventralization of dorsal muscles and tendons (Parr and McMahon, 1995). Therefore, DV patterning in the limb ectoderm is specified by domains of En1 expression ventrally, and Wnt7a expression dorsally.

Classical transplantation experiments demonstrate that once DV patterning is established in the limb ectoderm, this ectodermal DV pattern is subsequently conferred upon the underlying mesoderm, at least in the distal limb region (Chen and Johnson, 1999). Molecular analyses suggest the model that ectodermal induction of the underlying limb mesoderm is mediated by Wnt7a, which has been shown to induce the LIM-domain factor, Lmx1b, in the dorsal mesoderm of the limb (Riddle et al., 1995; Vogel et al., 1995). Null mutations in the Lmx1b gene lead to loss of dorsal limb phenotype in mouse and humans (Chen et al., 1998; Dreyer et al., 1998). Therefore, the molecular mechanisms regulating DV patterning by the ectoderm during later stages of limb development are well characterized. However, little is known about the initial phase of limb development in which the mesoderm establishes DV patterning in the overlying pre-limb ectoderm.

We now describe our analyses of a conditional knockout of the most widely expressed type I BMP receptor, BMPR-IA. We specifically abrogated expression of BMPR-IA in the pre-hindlimb ectoderm to determine the role of BMP signaling during limb ectoderm development. BMP receptors are serine/threonine kinases that require both a type I and type II receptor subunit to function efficiently (Massague, 1998). Previous knockout analyses have demonstrated that BMPR-IA, encoded by the Bmpr gene (Bmpr1a – Mouse Genome Informatics), plays a crucial role in BMP signaling. A null mutation of the Bmpr gene results in early embryonic lethality around the time of gastrulation (Mishina et al., 1995). Therefore, genetic analyses of BMPR-IA function during limb development requires a conditional knockout of the gene. To overcome the early embryonic lethal phenotype of the Bmpr knockout (Mishina et al., 1995), we used the loxP/Cre approach to induce tissue-specific mutations in the Bmpr gene (Nagy, 2000).

Using the loxP/Cre mutagenesis approach, we demonstrate that BMP signaling is required for the formation of the organizing centers that regulate PD and DV patterning in the limb. Abrogation of BMPR-IA signaling interferes with the formation of the AER, and abolishes the expression of FGFs that mediate AER function, as well as other AER molecular markers. Surprisingly, the Bmpr conditional knockout results in the formation of double dorsal hindlimbs. This alteration dramatically affects the overall expression of DV regulatory genes; En1 gene expression is lost in ventral limb ectoderm, and ectopic expression of Wnt7a and Lmx1b is observed. Finally, we examine the expression profile of the BMPR-IA ligands, BMP4 and BMP7, which are detected early in the lateral mesoderm at a time in which the mesoderm induces DV pattern in the overlying ectoderm. These data demonstrate that BMP signaling is required to establish DV patterning in the limb ectoderm. Furthermore, this is the first demonstration that BMP signaling is required for the formation of critical organizing centers during limb development.

The null mutation of the Bmpr gene was published previously (Mishina et al., 1995). Alleles of the Bmpr gene (floxP and floxP-neo) containing loxP sites flanking exon 2 (floxed alleles) were generated as described in Mishina et al. (Y. M., M. C. H. and R. R. B., unpublished). Briefly, a PGK-neo cassette flanked by loxP sites was introduced into intron 2 of Bmpr and a third loxP site marked with a NheI site was introduced into intron 1. The initial targeting event created the floxP-neo allele (Fig. 1D). Mice heterozygous for floxP-neo were mated with CMV-Cre mice (Arango et al., 1999) and resulting progeny were screened by Southern blot for those containing the floxP allele (see Fig. 1D and Southern blot analyses described below).

The Brn4-Cre was generated using a 5.6 kb SalI-BamHI fragment encompassing the 5′ flanking regions of the Brn4 gene, whose 3′ end falls 14 bp 3′ of the initiator methionine. This promoter region was fused to the Cre recombinase gene and polyA sequences from pOG231 (a generous gift of S. O’Gorman, Salk Institute). The transgenic construct was introduced into CD-1 mice using standard transgenic techniques (Hogan et al., 1994). Three viable transgenic founders were generated. Two of these demonstrated tissue-specific recombination, including the Brn4-Cre pedigree, designated bcre-32, used in all studies described in this paper. The third pedigree, designated bcre-23, appears to be expressed during early embryogenesis, and has been used as a ‘deletor’ strain in other studies (Cho et al., 2001). We have not detected expression of the endogenous Brn4 gene (Phippard et al., 1999; Phippard et al., 1998) or transgenes containing the ∼6 kb Brn4 flanking region (Heydemann et al., 2001) in the limb. The mating scheme used to generate mutant and normal littermates used in this study is shown in Fig. 1B. Animals that were hemizygous for the Brn4-Cre transgene (Fig. 1B; Parent#1) were also used in this study. The ROSA reporter pedigree was a generous gift from P. Soriano (Hutchinson Cancer Center, Seattle, WA).

Genomic DNA was isolated from the tissues indicated in Fig. 1C from an 18.5 dpc embryo, digested with NheI/SacI, and probed with the H23 probe, which encompasses the third exon (Fig. 1D; box above exon 3 of modified alleles). Each of the Bmpr alleles and the Cre-mediated rearrangements of the alleles can be distinguished by Southern analyses of genomic DNA using this strategy.

Staining for lacZ expression was accomplished as described previously (Phippard et al., 1999). Histological analyses of neonatal and adult limbs were accomplished by Hematoxylin and Eosin staining of paraffin sections. When necessary, the limbs were decalcified using Cal ExII (Fisher Scientific). In situ hybridization was accomplished by a modification (T. A. Sanders and C. W. Ragsdale, unpublished) of previously described techniques (Wilkinson, 1992). The mouse Wnt7a and Shh probes were a kind gift from Andrew McMahon (Parr et al., 1993; Echelard et al., 1993); the mouse Lmx1b probe was a kind gift from Randy Johnson (Chen et al., 1998); and the mouse En1 probe was kind gift from Alexandra L. Joyner (Logan et al., 1992). The Fgf8, Fgf4, Bmp2, Bmp4 and Bmp7 probes were generated by PCR amplification of ∼0.5 kb regions of the cDNAs that do not cross hybridize.

Immunohistochemical analyses were carried out on paraffin sectioned material (7 μm) that had been fixed overnight in 4% paraformaldehyde/phosphate-buffered saline at 4°C prior to embedding in paraffin. Sections were processed by immunoperoxidase labeling using the Vectastain ABC Kit (Vector Labs). The immunoperoxidase signal was amplified with the TSA Indirect Tyramide Signal Amplification Kit (Perkin Elmer Life Science) according to the manufacturer’s instructions with the following modifications. For antigen unmasking, sections on slides were heated in a microwave oven in 10 mM sodium citrate pH 6.0 for 4 minutes. Microwave power settings were adjusted to maintain temperature just below boiling during heating. The sections were then treated to quench endogenous peroxidase activity with 2% H₂O₂. The specimens were blocked in 5% normal goat serum for 1 hour at room temperature and incubated overnight at 4°C with the Phospho-Smad1 antibody (Cell Signalling Technology) diluted 1:3000 in 5% normal goat serum. The specimens were then incubated in 0.05% blocking reagent supplied in the tyramide kit for 30 minutes at room temperature to further eliminate nonspecific binding. All further incubations were done in the 0.05% blocking reagent as per manufacturer’s instructions.

A transgenic mouse pedigree was generated that contained loxP sites in the first and second introns of the Bmpr gene (Fig. 1). This modification generated no apparent phenotype in the resulting animals. However, the Bmpr gene was now susceptible to conditional inactivation in transgenic animals expressing the Cre recombinase gene. Cre-mediated recombination between the loxP sites removes the second exon of the gene, which encodes roughly a third of the extracellular domain of the receptor within the ligand-binding domain. RNA splicing between the first and third exons results in a frame-shift, therefore further ensuring that Cre-mediated excision will result in a null allele. Mice homozygous for the ‘exon2-excised’ deletion die without gastrulating, which is identical to the original Bmpr null mutant phenotype (Mishina et al., 1995).

Tissue-specific expression of the Cre recombinase gene was driven by the Brn4/Pou3f4 promoter region in a Brn4-Cre transgenic pedigree (Fig. 1A). Previous analyses demonstrated that the Brn4 proximal 5′ flanking region directs expression to the neural tube (Heydemann et al., 2001). The Brn4-Cre pedigree used in this study efficiently induced Cre-mediated rearrangement of the Bmpr gene in tissues derived from the neural tube, such as spinal cord, hindbrain and forebrain (Fig. 1C-D). In addition, Southern analyses and the mutant limb phenotype demonstrated that this Brn4-Cre pedigree was ectopically expressed in the limbs (Fig. 1C-D). We have not detected expression of the endogenous Brn4 gene (Phippard et al., 1999; Phippard et al., 1998) or transgenes containing the ∼6 kb Brn4 flanking region (Heydemann et al., 2001) in the limb. Therefore, the ectopic expression of the Brn4-Cre transgene in the ventrolateral ectoderm encompassing the limb field is probably a consequence of the site of transgene integration. This transgene integration site does not appear to have interrupted a gene necessary for limb development, because the transgene can be homozygosed without any detectable limb phenotype. Furthermore, molecular markers of limb development are not affected in non-mutant animals containing the Brn4-Cre transgene.

To determine the spatial and temporal expression of the Brn4-Cre pedigree in the limb, we intercrossed the Brn4-Cre pedigree with the ROSA reporter strain, which activates the expression of the lacZ gene upon Cre-mediated recombination (Soriano, 1999). As shown in Fig. 2A, Brn4-Cre-mediated expression of lacZ is initially detected at 9.75 dpc in the ventrolateral ectoderm in a region encompassing both of the embryonic limb anlage. At this stage of embryogenesis, the forelimb bud has begun to form, but the hindlimb bud, which typically forms about a half a day later than the forelimb, has not formed yet (Fig. 2A). The expression of the Brn4-Cre transgene is restricted to the ectoderm of the limb with the highest degree of lacZ reporter expression detected in the AER. Abundant lacZ reporter expression was also detected in the ventral limb ectoderm, and some expression was detected in the dorsal limb ectoderm (Fig. 2B). Significant Brn4-Cre-mediated induction of reporter transgene expression was detected at the earliest stages of hindlimb bud formation (Fig. 2C). The induction of reporter gene expression increased in the limb ectoderm, particularly in the pre-AER region, as the limb develops (Fig. 2C-E). The overall pattern of Cre-mediated lacZ expression was identical in the forelimb at this stage, but the abundance of reporter expression was reduced in comparison to the hindlimb (Fig. 2G,H). These data demonstrate that the Brn4-Cre transgene expression is activated simultaneously throughout the ventrolateral ectoderm encompassing the limb fields/bud. However, the forelimb bud begins to form before Cre-mediated expression, whereas the hindlimb forms after Cre-mediated lacZ expression is induced.

To assess directly the loss of BMP signaling, we examined the degree of phosphorylation of SMAD1 (phospho-SMAD1), which is phosphorylated by BMPR-IA signaling. At 9.75 dpc, phospho-SMAD1 immunopositive cells can be detected in normal embryos throughout the ventral mesoderm of the embryo, and is particularly high in the most ventral regions of the embryo and in the ventral wall of the coelom (Fig. 3A). Higher magnification view (Fig. 3B) demonstrates the phospho-SMAD1 signal is detected at high levels in both the ectoderm and the mesoderm. The specificity of the immunostaining is demonstrated by the detection of phospho-SMAD1 immunopositive cells in regions that have previously shown to express high levels of BMP growth factors, such as the dorsal neural tube (Fig. 3C), ventral mesoderm (Fig. 3A), the neural retina (Fig. 3M) and the dorsal hindbrain (Fig. 3N). Furthermore, absence of the primary antibody results in complete loss of immunolabeling (data not shown). In 9.75 dpc Bmpr mutants, the pattern of phospho-SMAD1 immunostaining looks similar to that found in normal embryos (Fig. 3D,E). Most ectodermal cells demonstrate robust immunolabeling. However, unlike in normal embryos, a rare ectodermal cell can be identified that has little or no immunostaining (Fig. 3E). Although the Brn4-Cre transgenic pedigree eliminates BMPR-IA in the neural tube, phospho-SMAD1 staining is still detected in the dorsal neural tube. This result is consistent with the observation that we do not detect phenotypic changes in the dorsal neural tube of these mice (data not shown), and suggest that functional redundancy of BMP receptor function occurs in the neural tube, but not the limb ectoderm.

In the normal embryos at 10.0 dpc, nuclear phospho-SMADl immunolabeling is detected at high levels in both the mesoderm and the ventral ectoderm of the limb (Fig. 3G,H), but not in the dorsal limb ectoderm (Fig. 3G, inset). However, in the 10.0 dpc mutant animals, phospho-SMAD1 labeling is not detectable in the vast majority of ectodermal cells, although robust staining is still detected in adjacent mesodermal cells (Fig. 3J,K). The loss of phospho-SMAD1 immunolabeling is specific for limb ectoderm, because we do not detect changes in the dorsal neural tube (Fig. 3L), the lateral mesoderm (Fig. 3J), dorsal hindbrain or neural retina (data not shown).

To corroborate that BMP signaling is abrogated at 10.0 dpc, we examined the expression of Msx2, whose expression is often induced by BMP signaling (Hogan, 1996). Msx2 expression is not detected in mutant limb ectoderm at 10.0 dpc, although its expression is still detected in the dorsal neural tube (Fig. 3O,P). No differences in Msx2 expression were detected at 9.75 dpc (data not shown). Attempts to detect loss of BMPR-IA expression directly by in situ hybridization or by immunohistochemistry did not yield reliable results. However, three different criteria, namely, ROSA reporter expression, phosphorylation of SMAD1 and expression of Msx2, demonstrate that the Bmpr gene is mutated resulting in loss of BMP signaling between 9.75 dpc and 10.0 dpc.

Conditional inactivation of the Bmpr gene using the Brn4-Cre transgenic pedigree resulted in a severe limb phenotype (Fig. 4). Although the phenotype was variable, the most severely affected animals (8/42 hindlimbs) demonstrated complete agenesis of the hindlimb (Fig. 4A). The forelimbs typically demonstrated subtle malformations, which occasionally resulted in an ectopic distal phalange (Fig. 4B,D). However, the hindlimbs of the mutant animals were more severely affected, presumably because the Brn4-Cre transgene was expressed before limb bud formation in the hindlimb, but after initial forelimb bud formation. The mutant hindlimbs were grossly malformed (Fig. 4C,F,G,I). A common feature of the hindlimb malformations was polysyndactyly (Fig. 4C,F,G,I). This phenotype included fusion of digits (syndactyly), as well as partial duplication of distal segments of the digits (Fig. 4F,G,I). The typical hindlimb had fewer digits (Fig. 4C,F). However, the mutant limbs rarely demonstrated supernumerary (polydactyly) digits (Fig. 4G,I). In addition, hematoma at the distal tips of the digits was commonly observed (Fig. 4C). Transverse sections through the hindlimbs of moderately affected mutants demonstrated a loss of ventral structures (Fig. 4H,I,L,M). This was most clearly illustrated by the loss/transformation of the prominent ventral flexor digitorum profundus tendon, and an overall mirror image symmetry of the mesenchyme in the dorsal/ventral plane (Fig. 4H,I,L,M). Skeletal preparations demonstrated a loss of the sesamoid process, which is a ventral bone structure (Fig. 4J,K). Additionally, malformed hair follicles were found on both the ventral and dorsal surfaces of the foot, whereas follicles were normally restricted to the dorsal limb (data not shown). Finally, eccrine glands, which were normally found on the ventral side of the limb, were drastically reduced in number on the ventral side of the mutant hindlimb (data not shown). These data demonstrate that the morphology of the hindlimb displayed a double dorsal phenotype.

To determine the molecular genetic basis for the observed phenotype, we queried each of the cardinal limb axes using molecular markers. Fgf8 is expressed in the AER and pre-AER, where it mediates, in part, the organizing function of the AER along the proximal/distal axis (Crossley et al., 1996; Mahmood et al., 1995; Sun et al., 2000; Vogel et al., 1996). Fgf8 expression was initially detected in the hindlimb field of normal littermates at approximately 10.0 dpc (Fig. 5A). We detected little, if any, Fgf8 expression in the hindlimb of mutant embryos at this stage of embryogenesis (Fig. 5B). Later in embryogenesis (10.5-11.5 dpc), we detected Fgf8 expression in mutant hindlimbs, but the expression varied considerably from virtually complete loss of expression (Fig. 5D) to largely intact expression along the distal rim of the limb with short breaks (Fig. 5F). In Fig. 5E, we depict a mid-range phenotype with large stretches of Fgf8 expression along the distal rim interspersed with regions that do not express Fgf8.

The variability of Fgf8 expression in the mutant hindlimb buds reflected the variability that we observed in the hindlimb phenotype from complete agenesis of the hindlimb to a largely intact hindlimb. Table 1 indicates the extent of variability in the domain of Fgf8 expression in embryos of ages 10.25-11.5 dpc. Embryos were scored by quartiles between ‘0’ and ‘4’, according to the length of Fgf8 expression along the distal rim of the hindlimb. None of the mutant embryos scored in Category ‘4’ which corresponded to normal Fgf8 pattern (as depicted in Fig. 5C). The majority of embryos scored in Categories ‘1’ and ‘2’, corresponding to approximately a quarter to a half of the length of Fgf8 expression detected in normal embryos. Overall, a total of 8/35 embryos scored in Category ‘0’, corresponding to no expression or only a focal patch of expression (Fig. 5D). The number of embryos in Category ‘0’ corresponded approximately with the number of animals that demonstrated agenesis of the limbs (8/35 versus 8/42). Therefore, the variability in hindlimb phenotype correlated with the degree to which the AER formed in the mutants.

Additionally, we observed a similar variability in the expression of other AER markers, including Bmp4 (Fig. 5G,H) and Fgf4 (data not shown). In the mutant forelimbs, Fgf8 expression was largely intact, although the borders of Fgf8 expression were typically less distinct than the sharp borders detected in normal limbs (data not shown). Rarely, we detected segments of the presumptive AER that failed to express Fgf8 (data not shown) in the forelimbs. This low incidence of interruptions in Fgf8 expression may explain the occasional forelimb malformation detected in the mutant animals.

To examine the status of the anterior/posterior limb organizer, the zone of polarizing activity (ZPA) (Tickle, 1981; Wolpert, 1969), we analyzed the expression of sonic hedgehog (Shh) (Krauss et al., 1993; Riddle et al., 1993) in the Bmpr conditional mutants. We observed a variable expression of Shh, which reflected the variability observed with Fgf8 expression and AER formation (see below). Because Shh expression requires the expression of the Fgfs to maintain appropriate expression levels (Laufer et al., 1994; Sun et al., 2000), we hypothesized that the variability of Shh expression is dependent on FGF expression in these mutants. To test this directly, we double-labeled embryos with probes against Fgf8 and Shh (Fig. 5C-F). Because the distribution of Fgf8 expression in the mutants was largely stochastic, the domain of Shh expression correlated with the amount of Fgf8 expression (Fig. 5D-F). However, occasionally a modest domain of Fgf8 expression would lie directly above the posterior margin of the hindlimb resulting in a robust patch of Shh expression (data not shown). These observations are consistent with previous data suggesting that the maintenance of Shh expression is dependent on the AER.

To assess dorsal/ventral patterning of the limb, we examined the expression of the transcription factor gene, Lmx1b. This gene is normally expressed in the dorsal mesenchyme of both the hindlimb bud (Fig. 6A, Fig. 8) and the forelimb bud (Fig. 6C, Fig. 8). In the Bmpr conditional mutant, Lmx1b was expressed in both the dorsal and ventral mesoderm of the hindlimb, consistent with the double dorsal phenotype observed by histological analyses (Fig. 6B). Expression of Lmx1b was restricted to dorsal mesoderm in the mutant forelimb (Fig. 6D) indicating that dorsal/ventral patterning was not disrupted in the forelimb.

Lmx1b expression in the dorsal mesoderm is induced by Wnt7a (Cygan et al., 1997; Loomis et al., 1998; Riddle et al., 1995; Vogel et al., 1995), which is expressed in the dorsal ectoderm (Parr and McMahon, 1995) of normal mouse limbs (Fig. 6E, Fig. 8). In Bmpr mutant embryos, Wnt7a gene expression was expanded ventrally in the hindlimb field (Fig. 6F).

Because previous studies demonstrated that Wnt7a expression is repressed in the ventral limb ectoderm by En1 (Cygan et al., 1997; Loomis et al., 1996; Loomis et al., 1998), we examined En1 gene expression in the Bmpr conditional mutants. As shown in Fig. 6G,H, En1 expression was virtually lost in mutant embryos. We did not detect substantial En1 expression at any stage of embryogenesis from 10.0-11.5 dpc, although En1 expression in the forelimb was normal in the same mutant embryos (data not shown).

Unlike the AER phenotype, the double dorsal phenotype, as assessed by molecular markers, was completely penetrant. All 58 embryos examined with molecular markers of dorsal/ventral patterning demonstrated a double dorsal phenotype in the hindlimb (En1, Lmx1b and Wnt7a probes on embryos 10.25-11.5 dpc). Therefore, these data indicate that BMPR-IA signaling during early limb development is required for the expression of En1 and therefore, the establishment of ventral limb patterning (Fig. 8).

To assess the role of previously characterized ligands of BMPR-IA during limb development, we examined the expression pattern of Bmp2, Bmp4 and Bmp7 in the hindlimb at timepoints just before and during the period in which the molecular phenotypes of the Bmpr mutants first appeared (Fig. 7). We have observed an extremely dynamic pattern of Bmp gene expression during this period. Although Bmp2 is expressed in the AER at later stages of development (Lyons et al., 1990), we did not detect appreciable levels of expression in the hindlimb during the 9.75-10.0 dpc timeframe (data not shown). At 9.75 dpc, before hindlimb bud formation, Bmp4 is highly expressed throughout the ventral-lateral part of the embryo, including expression in the lateral mesoderm (Fig. 7A). By 10.0 dpc, the hindlimb bud has begun to form, and Bmp4 expression has been downregulated in the limb mesoderm (Fig. 7D). Expression levels remain high in the pre-AER region in a pattern that correlates well with the expression of the AER marker, Fgf8 (Fig. 7D,E). Bmp7 demonstrates a more restricted pattern of expression than Bmp4 during the 9.75-10.0 dpc time period. At 9.75 dpc, Bmp7 expression is detected in the ventral region of the embryo, but is restricted to the outermost region of the embryo within or in close apposition to the ectoderm (Fig. 7B). Interestingly, this expression domain correlates well with the expression domain of Msx2 at this stage of development (Fig. 7C). By 10.0 dpc, the expression of Bmp7 has been largely downregulated (Fig. 7F). However, Msx2 expression retains an expression pattern roughly comparable with that observed at 9.75 dpc (Fig. 7G). These data demonstrate that Bmp4 and Bmp7 are expressed in a temporal and spatial sequence that is consistent with the specification of early ventral limb ectoderm, and then rapidly downregulated after limb bud formation is initiated.

We have demonstrated that BMPR-IA signaling is required during critical steps in the specification of limb ectoderm. Conditional knockout of BMPR-IA function in the ectoderm of the limb prior to limb bud formation disrupts the specification of ventral limb identity and AER formation. It is likely that the DV patterning phenotype is primarily due to the loss of En1 gene expression, because En1 has previously been shown to specify ventral pattern in the limb ectoderm. To further assess the role of BMP signaling during early limb development, we examined the expression pattern of the genes encoding the known BMPR-IA ligands, BMP4 and BMP7. These genes demonstrate a dynamic pattern of expression, and are expressed with the correct temporal and spatial sequence to play a role in the specification of ventral limb phenotype and AER induction.

The mutant phenotype is much more profound in the hindlimb than in the forelimb of the Bmpr mutants. Although we cannot rule out the possibility that En1 expression and DV patterning is differentially regulated in the fore- and hindlimbs, the difference in phenotype between the fore- and hindlimbs most likely results from the difference in the timing and extent of Bmpr gene inactivation in the limbs (see Fig. 2A). We hypothesize that DV patterning is established in the forelimb prior to extensive Bmpr inactivation, whereas gene inactivation occurs before DV patterning during hindlimb formation. These data would suggest that DV patterning is established in a narrow temporal window just prior to the initial outgrowth of the limb bud. In addition, our observations are consistent with those of Niswander and colleagues, who have shown that BMP signaling is required for DV patterning and AER formation in both the fore- and hindlimbs of chickens (Pizette et al., 2001).

Classical transplantation experiments have demonstrated that mesoderm dictates the DV patterning before limb bud formation, whereas DV patterning is regulated by the ectoderm after the limb bud forms (Chen and Johnson, 1999). In the chicken, the transition of control over DV patterning from the lateral mesoderm to the somatic ectoderm occurs between stages HH14 and HH16 (Geduspan and MacCabe, 1987; Geduspan and MacCabe, 1989). The first indication of limb bud formation occurs with the condensation of lateral mesoderm opposite somites 15-16 at stages HH16 (Hamburger and Hamilton, 1951). The transition from mesodermal to ectodermal control of DV patterning is demonstrated by the fact that a 180° rotation of prospective limb ectoderm at stage HH14 results in a limb with normal DV polarity. However, ectodermal inversion at HH16 results in formation of a limb with reversed DV polarity (Geduspan and MacCabe, 1987; Geduspan and MacCabe, 1989). The acquisition of DV polarity by the ectoderm is accompanied by the expression of genetic regulatory genes, En1 and Wnt7a, at HH15-16 (Davis et al., 1991; Dealy et al., 1993; Gardner and Barald, 1992; Riddle et al., 1995). Mutational analyses in mice have clearly demonstrated that En1 and Wnt7a play complementary roles in specifying ventral and dorsal limb ectoderm, respectively (Fig. 8) (Cygan et al., 1997; Loomis et al., 1996; Loomis et al., 1998; Parr and McMahon, 1995). Therefore, one would predict that candidate factors which regulate the initial induction of DV patterning would be expressed in the lateral mesoderm before limb bud formation and would be required for the induction of factors that specify DV patterning in the overlying ectoderm. Our data demonstrate that ligands of BMPR-IA fulfill these criterion, because Bmp4 and Bmp7 are expressed in lateral mesoderm and the overlying ectoderm before limb bud formation, and BMPR-IA signaling is required for the induction of En1.

Although classical transplantation analyses have been undertaken in chickens and amphibians, analyses of the cellular and molecular events regulating limb morphogenesis suggest that mechanisms of DV patterning and AER formation are conserved between vertebrates (Johnson and Tabin, 1997; Martin, 1998; Vogt and Duboule, 1999). Specifically, regarding the role of BMP signaling, observations from Niswander and her colleagues (Pizette et al., 2001) are similar, if not identical, to those we have observed in mice. Therefore, our model for BMP signaling will incorporate embryological and molecular observations from all vertebrate model systems.

Although the role of ectoderm in specifying DV patterning after limb bud formation has been well characterized, the precise role of mesoderm before limb bud formation has been less clear, and the molecular mechanisms are completely unknown. In the 1970s, transplantation experiments characterizing the ability of lateral mesoderm to specify early DV patterning were conflicting (Michaud et al., 1997). However, more recent transplantation and cell marking experiments suggest that paraxial mesoderm plays a role in specifying dorsal limb ectoderm, whereas lateral mesoderm specifies ventral limb ectoderm (Altabef et al., 1997; Michaud et al., 1997). The conclusions from these studies is incorporated into the model schematized in Fig. 8C,D. For example, Michaud et al. demonstrated that an 180° inversion of lateral mesoderm alone at pre-limb bud stages (HH12-13) (dark blue in Fig. 8C) does not result in an inversion of limb polarity (Michaud et al., 1997). However, inversions of both lateral and somitic mesoderm (dark blue and red/orange, respectively, in Fig. 8C) does invert DV polarity of the resulting limbs. In addition, transplantation of an ectopic somite into lateral mesoderm, such that the lateral mesoderm was flanked by the endogenous somite medially and the ectopic somite laterally, resulted in the formation of a bidorsal limb. These data and other data suggested that a dorsalizing signal was emanating from the somitic mesoderm, and a ventralizing signal emanated from the lateral mesoderm.

If the dorsalizing signal emanates from the somitic mesoderm, how then does this give rise to dorsal limb ectoderm? Both transplantation (Michaud et al., 1997) and cell labeling experiments (Altabef et al., 1997), demonstrate that ectoderm overlying the somites are fated to become dorsal limb ectoderm (depicted as purple ectoderm in Fig. 8C,D). Tickle and colleagues have undertaken cell fate mapping experiments that demonstrate a large fraction of dorsal limb ectoderm originates from the ectoderm overlying the somites (Altabef et al., 1997). These data suggest that the ectoderm overlying the somites is fated to become dorsal limb ectoderm, and then morphogenetic movements result in the translocation of this ectoderm such that it covers the dorsal limb (as schematized in Fig. 8D).

Our model hypothesizes that BMP signaling pathway mediates the initial induction of DV pattern in the presumptive limb ectoderm before limb bud formation. Bmp4 and Bmp7 are expressed in the lateral mesoderm and the overlying ectoderm just prior to the induction of limb bud formation, and therefore are correctly positioned in both time and space to induce ventral limb ectoderm (Fig. 7, Fig. 8). Furthermore, we have demonstrated that BMPR-IA signaling is required for the induction of En1, and subsequently the specification of ventral limb identity. Finally, the dorsalizing effect of somites, we hypothesize, is mediated by the expression of the BMP antagonist, noggin, in the myotomal compartment of the somite (Capdevila and Johnson, 1998; Hirsinger et al., 1997; Marcelle et al., 1997; McMahon et al., 1998; Reshef et al., 1998; Tonegawa and Takahashi, 1998). Noggin expression in the somites could explain the ability of somites to dorsalize the ectoderm in transplantation experiments because of their ability to inhibit BMP signals from the lateral mesoderm. Noggin expressed in the somites could neutralize the effects of BMPs in the ectoderm that is in close apposition to the somites (Fig. 8C). Alternatively, the somitic mesoderm could induce the expression of an unidentified BMP antagonist in dorsal ectoderm, which suppresses the BMP signaling within the dorsal ectoderm itself.

Finally, the mesoderm-derived inductive signal would have to be downregulated as the lateral mesoderm loses its ability to induce the overlying ectoderm and as the dorsally fated ectoderm moves over the limb mesenchyme. Our analyses demonstrate that the expression of Bmp4 and Bmp7 are rapidly downregulated in most of the lateral mesenchyme as the limb bud is formed (Fig. 7D,F), although Bmp4 expression is maintained in the distal limb where BMP signaling is required for AER formation.

Our data demonstrate a crucial role for BMPR-IA signaling in the formation of the AER. It is conceivable that the AER defects are secondary to DV patterning defects. However, this seems unlikely because the penetrance of the DV patterning defect is complete, whereas the AER defect is quite variable. This difference in penetrance argues that AER formation and DV patterning are two independent processes. Our argument is further bolstered by analyses of the eudiplopodia chick mutant that suggest that AER formation is not strictly dependent on establishment of a DV border at the distal tip of the limb (Laufer et al., 1997).

Classical studies have shown that lateral mesoderm induces the AER (Carrington and Fallon, 1984; Saunders and Reuss, 1974). As Bmp4 and Bmp7 are expressed in lateral mesoderm, they are candidates for the initial inductive event required for AER formation.

There are mutants, such as limbless, in which both AER formation and DV patterning are affected. One parsimonious explanation is that the limbless gene product may be epistatic to BMP signaling and therefore affects both independent processes because they are mediated by one signaling pathway. An additional mutant that affects both DV patterning and AER formation is the En1 knockout. In this case, there are important differences between the Bmpr mutant and the En1 mutant. Most importantly, the En1 knockout does not abrogate AER formation, but does affect the positioning of the AER. En1 mutants demonstrate a ventral extension of the AER (Loomis et al., 1996; Loomis et al., 1998). The AER, where it is formed in the Bmpr mutant, does not demonstrate the ventral extension seen in En1 mutants (Fig. 5; data not shown). Michaud et al. also did not observe a ventral extension of the AER in double dorsal phenotypes generated by transplantation experiments in chick (Michaud et al., 1997). These data, in conjunction with our data, suggest that the ventral AER extension is not a consistent feature of the double dorsal phenotype.

In summary, the conditional knockout of the gene for the most widely expressed type I BMP receptor, BMPR-IA, results in limb malformations that are due to the disruption of AER formation and loss of DV patterning. This is the first demonstration that BMPR-IA signaling is essential for these early events in limb morphogenesis.

We thank C. W. Ragsdale and T. A. Sanders for providing us their whole-mount in situ hybridization protocol prior to publication; L. Niswander for providing results prior to publication; S. O’Gorman for the Cre expression plasmid; A. Joyner, A. McMahon, D. Phippard, R. Johnson, R. Riddle for probes; and J. Golden for help with vibratome sectioning. We would also like to thank R. Riddle for advice on the limb analyses and for reading the manuscript; Randy Johnson for comments on the manuscript. E. B. C.’s and R. R. B.’s laboratories are supported by the NIH.