BMP signals control limb bud interdigital programmed cell death by regulating FGF signaling

It is becoming increasingly clear that cellular activities such as cell proliferation, apoptosis and migration that contribute to the generation of pattern during development are often regulated by coordinated interactions between two or more signaling cascades. The developing limb provides an excellent model system for studying these interactions, owing to the availability of a rich context of genetic and embryological data(Capdevila and Izpisua Belmonte,2001; Niswander,2002; Tickle,2006).

Vertebrate limbs begin as buds of the lateral plate mesoderm, jacketed in surface ectoderm. In the mouse, at about embryonic day (E) 10.5 and 11.0 in the forelimb and hindlimb bud, respectively, cells have finished moving from the ventral ectoderm of the early bud (the `preAER') to the bud apex to form a stratified columnar epithelial structure called the apical ectodermal ridge(AER) (Kimmel et al., 2000; Loomis et al., 1998; Guo et al., 2003). Although the AER itself is fated to undergo cell death, it is a source of secreted factors required for limb bud outgrowth. This is demonstrated by the surgical removal of the AER in chick embryos, which results in proximal-distal truncations (Dudley et al.,2002; Saunders,1948; Summerbell,1974).

The AER-specific activity required for limb bud outgrowth is encoded by members of the FGF family (reviewed by Martin, 1998). FGF protein can rescue limb development in chick embryos that have a surgically removed AER(Fallon et al., 1994; Niswander et al., 1993). Four Fgf genes (Fgf4, 8, 9 and 17, referred to here collectively as AER-FGFs) are expressed in the AER(Martin, 1998; Sun et al., 2000). The Fgf8 expression domain is the largest spatially, being strongly expressed throughout the anterior-posterior extent of the AER, and is also the longest temporally, occurring earliest, in the preAER(Lewandoski et al., 2000), and ending during AER regression (E12.5-13.5)(Crossley et al., 1996; Mahmood et al., 1995; Salas-Vidal et al., 2001). By contrast, the other three AER-FGF genes display more restricted expression domains within the posterior two-thirds of the AER, with expression initiating later and ceasing earlier than Fgf8(Sun et al., 2000). In the mouse, single-gene inactivation studies reveal that, of the four AER-FGFs,only Fgf8 is required for normal limb development, although most limb skeletal elements still form in limb buds lacking Fgf8(Colvin et al., 2001; Lewandoski et al., 2000; Moon et al., 2000; Moon and Capecchi, 2000; Sun et al., 2000; Xu et al., 2000). However,when Fgf4 and 8 are simultaneously inactivated throughout the preAER, mice are born without limbs, indicating that these two Fgf genes are genetically redundant (Sun et al.,2002; Boulet et al.,2004). Thus, Fgf4 and 8 may be considered the principal AER-FGFs.

Another set of molecules playing a role in limb development are the members of the bone morphogenetic protein (BMP) family, which, in contrast to FGFs,have been implicated in a variety of diverse aspects of limb development including digit identity, anterior-posterior and dorsal-ventral patterning of the limb and AER regulation (Niswander,2002; Tickle,2006). Early in limb bud development, BMP signaling is required for normal formation of the AER (Ahn et al., 2001; Pizette et al.,2001). As the AER develops, regulation of BMP activity by the antagonist gremlin is required to activate Fgf4, 9 and 17(Capdevila et al., 1999; Khokha et al., 2003; Michos et al., 2004; Zuniga et al., 1999; Zuniga et al., 2004).

Later in limb development, BMPs also play a role in programmed cell death(PCD) of the interdigit region, which is required to separate digits and prevent soft tissue syndactyly (webbing) (reviewed by Chen and Zhao, 1998; Zuzarte-Luis and Hurle, 2002; Zuzarte-Luis and Hurle, 2005). Prior to cell death, Bmp2, 4 and 7 are upregulated within the interdigit mesenchyme (Chen and Zhao,1998; Hogan, 1996)and inhibition of BMP signaling in chick embryos, via the retroviral expression of transgenes encoding dominant-negative BMP receptors, inhibits interdigit PCD (Yokouchi et al.,1996; Zou and Niswander,1996). Furthermore, in the mouse, syndactylous limbs are caused by inhibiting BMP signaling through the ectopic expression of the BMP antagonist noggin (Guha et al., 2002; Wang et al., 2004).

These data are generally construed to signify that BMPs act as direct effectors of PCD within the interdigit region(Chen and Zhao, 1998; Zou and Niswander, 1996; Zuzarte-Luis and Hurle, 2002; Zuzarte-Luis and Hurle, 2005). However, FGF-soaked beads can prevent interdigital PCD in chick embryos(Buckland et al., 1998; Ganan et al., 1996; Macias et al., 1996; Ngo-Muller and Muneoka, 2000),suggesting that interdigital PCD might be induced by the loss of AER-FGF signaling that occurs during AER regression(Salas-Vidal et al., 2001; Yokouchi et al., 1996; Zou and Niswander, 1996). On the other hand, application of FGF-soaked beads can also increase interdigital apoptosis, consistent with a model of cooperation between FGFs and BMPs acting as cell death triggers (Ganan et al.,1998; Montero et al.,2001). Thus, the relationship between BMP and FGF signaling during interdigital PCD is unclear.

Here, we define a relationship between BMP and FGF signaling during interdigital PCD. First, we inactivate the gene encoding the BMP receptor, Bmpr1a, specifically in the limb bud preAER or AER via Cre-mediated DNA recombination. Our data reveal a dynamic role for BMPR1A signaling: early BMP activity is required for AER formation and later activity is required for cessation of AER-FGF expression, which is correlated with interdigital PCD. Then, by analyzing mice carrying different combinations of double and triple AER-specific gene inactivations of Fgf4, 8 and Bmpr1a, we provide genetic evidence that BMPs regulate interdigit cell death through the regulation of AER-FGF signals.

All mutant alleles, the Msx2-Cre transgene and PCR conditions for genotyping these markers have been described previously(Lewandoski et al., 2000; Mishina et al., 2002; Mishina et al., 1995; Sun et al., 2000). Mutants were generated on a mixed CD1/129SvEv/C57BL/6 background. Msx2-Cre;Bmpr1a^flox/null mice were produced as described in Fig. 1. Msx2-Cre;Bmpr1a^flox/null mutants that also carry mutated Fgf4(or Fgf8) alleles were generated by mating BmpR1a^flox/flox; Fgf4^flox/flox (or BmpR1a^flox/flox;Fgf8^flox/flox)females with Msx2-Cre; BmpR1a^null/wt;Fgf4^null/wt (or with Msx2-Cre;Fgf8^null/wt) males. Likewise, to generate triple mutants, BmpR1a^flox/flox; Fgf4^flox/flox,Fgf8^flox/flox females were mated with Msx2-Cre;BmpR1a^null/wt; Fgf4^null/wt, Fgf8^null/wt)males. Noon of the day on which the vaginal plug was detected was considered E0.5. More precise age of E9.5-10.5 embryos was determined by counting the somites posterior to the forelimb, scoring the first as somite 13.

RNA localization by whole-mount in situ hybridization was performed as described (Pizard, 2004), with a minor change: detection of AER transcripts required no proteinase K treatment. β-gal and Lysotracker Red (Molecular Probes) staining was performed as previously described (Chi et al., 2003; Zucker et al.,1999).

To study the role of BMP signaling in the limb bud AER, we sought to produce Msx2-Cre; Bmpr1a^flox/null (or `mutant') mice so that Msx2-Cre activity in the preAER (Sun et al., 2000) will inactivate the Bmpr1a^floxallele. In the course of performing the required genetic crosses, we discovered that the Msx2-Cre transgene is integrated on chromosome 14, very closely linked to the Bmpr1a locus. Therefore, in order to generate mutant mice, we needed to first produce mice carrying both the Msx2-Cre transgene and the Bmpr1a^null allele in cis on chromosome 14. To achieve this, we screened approximately 1300 progeny of Msx2-Cre;Bmpr1a^null/+ mice crossed to wild-type mice before identifying one offspring with a recombinant chromosome 14 containing both genetic markers. This mouse and its progeny served as sires in the genetic cross indicated in Fig. 1A, in which 50% of the offspring are mutant and 50% (Bmpr1a^flox/+) are controls.

We and others (Barrow et al.,2003; Sun et al.,2000) have previously shown that Msx2-Cre is active earlier in the prospective AER cells of the hindlimb region where expression is widespread by the 24-somite stage (ss), as compared with the forelimb where extensive expression occurs by 28ss. To determine whether the kinetics of Bmpr1a inactivation followed this pattern, we performed RNA in situ hybridization (ISH) on mutant embryos using a riboprobe derived from those sequences in the Bmpr1a^flox allele deleted by Cre. Although we could detect Bmpr1a expression at E10.5 in the control forelimb (including the AER) and demonstrate Bmpr1a inactivation in the mutant AER (Fig. 1A,inset), we could not reproducibly detect Bmpr1a expression at limb bud stages prior to AER formation. However, using a full-length Bmpr1a riboprobe, we were able to detect expression in the preAER(data not shown), although this riboprobe cannot discriminate between mRNA transcribed from the unrecombined floxed allele and the recombined derivative. Therefore, we conclude that Bmpr1a is expressed in the limb bud surface ectoderm/preAER, but below detectable levels when using the floxed Bmpr1a sequences as an ISH riboprobe, and we confirm a previous report that these sequences function inadequately as a riboprobe in ISH analysis (Liu et al.,2005).

As an alternative measure of whether the kinetics of Msx2-Cre activity occurred as expected, we stained for the Cre-dependent β-gal activity of the R26R reporter allele(Soriano, 1999) in Msx2-Cre; Bmpr1a^flox/null; Rosa26^R26R/+embryos. As expected, ectoderm in the mutant hindlimb territory was extensively recombined by 24ss, whereas the mutant forelimb preAER was extensively recombined by 28ss (Fig. 1B,C).

Because Msx2-Cre is active earlier in the prospective hindlimb ectoderm relative to forelimb Cre activity, we expected a more severe hindlimb mutant phenotype. We found that in all skeletal preparations examined, mutant hindlimbs never formed (n=16) and the pelvic bones were reduced to two rudimentary elements, each resembling a fused ischium and ilium(Fig. 1D-G). All forelimb skeletal elements were present (n=15), with excess material in the digit tips (n=15) (Fig. 1D,E,H,I). Furthermore, 20% (3 of 15) contained a forelimb preaxial extra digit and 33% (5 of 15) displayed a split preaxial digit (data not shown). Surprisingly, these mutant phenotypes are not similar defects that vary in severity, but rather the hindlimb phenotype (no outgrowth) and the forelimb phenotype (extra outgrowth) are opposite in nature, suggesting that BMP signaling plays a dynamic role during limb development.

The AER is derived from the preAER, a broad domain of thickened ventral limb bud ectoderm (Kimmel et al.,2000). To determine whether the lack of hindlimbs in mutants is due to a failure in preAER/AER formation, we examined histological sections of mutant and control hindlimb buds from E9.5 (31ss) through to E10.5. The morphological preAER and AER were both absent in all mutant hindlimb buds(Fig. 2A-D). We used ISH to assay for expression of Fgf8, the earliest expressed AER-FGF,normally present in the hindlimb preAER by 27ss(Lewandoski et al., 2000). At this stage, Fgf8 expression was not detected in mutant hindlimb bud preAER (Fig. 2E,F). Furthermore, we never detected expression of Fgf8 (or other AER markers: Fgf4, Msx2, Jag2, Bmp2, 4 and 7) at later stages of hindlimb development through to E12.5 (data not shown).

Signals from the AER are required to maintain Fgf10 expression in the limb bud mesenchyme. Mice lacking Fgf10 expression do not form limbs (Min et al., 1998; Sekine et al., 1999). Fgf10 expression was normally induced in mutant hindlimb buds at 29ss(Fig. 2G,H), but maintenance was lost by 33ss (Fig. 2I,J). Thus, we conclude that early Bmpr1a inactivation in the hindlimb prevents both the induction of Fgf8 in the preAER and the formation of a morphological AER. Owing to the lack of AER-FGFs, mesenchymal Fgf10 expression is not maintained.

The adult forelimbs of Msx2-Cre; Bmpr1a^flox/null mice display soft tissue syndactyly (webbing between the digits)(Fig. 7A,B). In the mouse,between E12.5 and E13.5, PCD eliminates cells in the interdigit region,leading to the separation of individual digits(Fernandez-Teran et al., 2006; Zuzarte-Luis and Hurle, 2002; Zuzarte-Luis and Hurle, 2005). To determine whether mutant webbing was due to a PCD defect, we examined E12.5-13.0 control and mutant embryos by staining with the fluorophore Lysotracker Red, and observed a significant reduction in the degree of interdigital cell death in mutant forelimbs(Fig. 3A,B).

BMP signals to the interdigit region have been implicated as direct effectors or triggers of PCD (Ganan et al., 1996; Macias et al.,1997; Merino et al.,1999a; Yokouchi et al.,1996; Zou and Niswander,1996). Therefore, we examined the expression of genes related to mesenchymal BMP signaling to determine whether mutant forelimb webbing might be explained by Bmpr1a inactivation in the AER indirectly causing a diminution of interdigital BMP signaling. Bmp2, 4 and 7 are normally upregulated at E12.5-13.5 in the interdigit region, suggesting that they are candidate ligands for activating BMP signaling in this tissue(Hogan, 1996; Salas-Vidal et al., 2001). All three Bmp genes were induced normally in the interdigit region in both control and mutant E12.5 forelimbs (Fig. 3C-H), with enhanced Bmp4 expression in mutant subapical mesenchyme (Fig. 3F). Gremlin antagonizes mesenchymal BMP activity(Zuniga et al., 1999) and its enhanced expression in the interdigit region of ducks correlates with the decrease in interdigital PCD occurring in these birds, as compared with chickens (Merino et al.,1999b). However, we found no significant difference between mutants and controls in interdigital gremlin expression(Fig. 3I,J).

The closely related homeobox genes, Msx1 and Msx2, are downstream targets of BMP signaling and are expressed in the PCD zones of the developing limbs, including the interdigit regions(Houzelstein et al., 1997). Furthermore, misexpression of Msx2 induces cell death in the chick limb bud (Ferrari et al.,1998; Merino et al.,1999b) and Msx1^-/-; Msx2^-/- embryos retain their interdigit mesenchyme(Lallemand et al., 2005). We detected no decrease in the expression of Msx1 or 2 in mutant forelimbs at E12.5 (which would have been indicative of a reduction of BMP signaling), but did note an expanded expression domain that is likely to occur as a consequence of the retention of interdigit tissue(Fig. 3K-N).

AER-specific Msx2 expression(Coelho et al., 1991; Ferrari et al., 1998) was however absent in Msx2-Cre; Bmpr1a^flox/null forelimbs from E10.5-12.5 (Fig. 4A,B and data not shown), demonstrating that Msx2 is downstream of BMPR1A signaling in the AER. We also detected a reduction of AER-specific expression of Bambi (Fig. 4C-F),which is also a BMP target and encodes a natural dominant-negative BMP receptor (Onichtchouk et al.,1999). Together, these data suggest that only AER-specific BMP targets are affected in Msx2-Cre; Bmpr1a^flox/nullforelimbs, and that mesenchymal BMP-signaling is relatively normal.

To analyze AER function in mutant forelimbs, we examined the expression of the four AER-FGFs (Fgf4, 8, 9 and 17). We found no significant changes in Fgf9 expression, but did note a slight increase in Fgf17 expression (data not shown). At 22ss, we detected no change in Fgf8 expression between normal and mutant forelimbs(Fig. 5A,B). However, by E10.5(approximately 24 hours after forelimb Msx2-Cre activation), Fgf8expression in the mutant forelimb was less constricted dorsal-ventrally, but normally extended anterior-posteriorly(Fig. 5C,D). At E11.5, this expanded Fgf8 pattern was more pronounced in the anterior AER of mutant forelimbs (Fig. 5E,F)relative to the posterior AER. By E12.5, normal AER-specific Fgf8expression was ceasing over the interdigit region, but remained over the developing digits (Fig. 5G),consistent with published reports(Crossley and Martin, 1995; Salas-Vidal et al., 2001). However, in Msx2-Cre; Bmpr1a mutants, this cessation of Fgf8expression failed to occur, although we did note some infrequent small discontinuities in Fgf8 expression at the anterior and posterior end of the AER that did not correlate with the digit/interdigit pattern(Fig. 5H). By E13.5, normal Fgf8 expression was present only over the digits(Fig. 5I,K), whereas punctate Fgf8 expression occurred throughout the mutant AER(Fig. 5J,L), with regions of expression still present in the interdigit region associated with convex outgrowths of the distal autopod (see Fig. 5K,L). Normal AER Fgf8 expression completely ceased by E14.5, although mutant expression could still be detected at E15.5 (data not shown).

As with Fgf8 expression, mutant AER-specific Fgf4expression is upregulated. Normally, Fgf4 expression is first activated at about 28-29ss in a few cells in the posterior AER, and then by E10.5 is restricted to about two-thirds of the posterior AER(Fig. 5M,O)(Niswander and Martin, 1992; Sun et al., 2000). In mutants, Fgf4 activation occurred more extensively in the anterior-posterior axis and in several distinct patches (Fig. 5N), and by E10.5 extended more anteriorly with some anterior discontinuities in expression (Fig. 5P). By E11.5, Fg4 expression had ceased in normal forelimbs, whereas in mutants expression remained robust in the posterior region of the AER with weaker expression in the anterior AER(Fig. 5Q,R). Punctate AER-specific expression of Fgf4 remained through E12.5 and E13.5(Fig. 5S-X).

To determine whether the expression of other AER markers was temporally extended, we examined Bmp2, 4 and 7 expression in the mutant AER at E12.5-13.5. Bmp4 expression was not upregulated in the AER,although expression was increased in the subapical region(Fig. 3F and data not shown). At these stages, mutant AER-specific expression of Bmp2 and 7 was either absent or upregulated in a few posterior cells(Bmp2, 3 of 13 limbs; Bmp7, 5 of 7 limbs) (data not shown). Thus, considering the extent of their upregulation, Fgf4 and 8 were uniquely affected, owing to the lack of BMPR1A-mediated signaling to the AER.

The upregulation of the principal AER-FGFs (Fgf4 and 8)in mutant forelimbs raises the possibility that webbing might be due to prolonged expression of these FGFs acting as survival factors for the underlying interdigital mesenchyme. To test this, we examined webbing in mutant animals in which we genetically lowered the AER-FGF levels in the forelimb AER by generating Msx2-Cre; Bmpr1a^flox/nullanimals that also carry null or floxed alleles of Fgf4 and/or 8. However, we could not assess the complete contribution of both Fgf4 and 8 to the mutant webbing phenotype (i.e. in animals of the genotype Msx2-Cre; Bmpr1a^flox/null;Fg4^flox/null; Fgf8^flox/null) because Msx2-Cre inactivation of both Fgf genes causes a severe disruption of limb development,preventing analysis of interdigital PCD(Sun et al., 2002). Nevertheless, we could evaluate the effect of less extreme genotypes (i.e. with at least one copy of Fgf4 or Fgf8 intact), which reduce AER FGF4/8 activity to a level above zero, on the webbing phenotype of Msx2-Cre; Bmpr1a^flox/null progeny.

Webbing was essentially unchanged in mutant mice with Msx2-Cre-mediated deletion of Fgf4 in the AER (Msx2-Cre;Bmpr1a^flox/null;Fgf4^flox/null) (data not shown). Similarly, mutant mice that lacked Fgf8 expression in the AER(Msx2-Cre; Bmpr1a^flox/null;Fgf8^flox/null) possessed webbed limbs(Fig. 7B,C). However, we noted that the forelimbs of these animals contained all five digits (n=16 limbs) (Fig. 6C; Fig. 7C), which was surprising because Msx2-Cre-mediated deletion of Fgf8(Msx2-Cre;Fgf8^flox/null) causes a hypodactyly phenotype in which at least one and occasionally two digits fail to form(Fig. 6A,B)(Lewandoski et al., 2000). Thus, this hypodactyly phenotype was rescued in Msx2-Cre;Bmpr1a^flox/null;Fgf8^flox/null animals. We speculated that the levels of Fgf4 expression in Msx2-Cre;Bmpr1a^flox/null;Fgf8^flox/null animals might be above a threshold level required to rescue digit formation and maintain lower interdigit PCD. We also reasoned that Fgf4 expression levels in Msx2-Cre;Fgf8^flox/null animals, although elevated(Lewandoski et al., 2000),might be below this threshold as hypodactyly but no webbing occurs in these mutants. Indeed, semi-quantitative ISH revealed that Fgf4 expression in E11.5 and E12.5 forelimbs was higher in Msx2-Cre;Bmpr1a^flox/null;Fgf8^flox/null than in forelimbs in which only Fgf8 had been deleted by Msx2-Cre(Fig. 6D-I).

Next we examined the phenotype of mutant animals in which only one copy of Fgf8 was deleted (i.e. of the genotype Bmpr1a^flox/null;Fgf8^flox/+, or Bmpr1a^flox/null; Fgf8^null/+; n=26 limbs). We reasoned that webbing might be reduced in these mice because we do not expect a compound increase in Fgf4 expression as Fgf8heterozygotes do not affect AER Fgf4 expression(Lewandoski et al., 2000). Webbing was partially reduced to a variable extent in this genotype(Fig. 7D). Lastly, we found that reducing FGF4 levels by generating Msx2-Cre;Bmpr1a^flox/null;Fgf4^null/+;Fgf8^flox/nullanimals resulted in forelimbs in which interdigit PCD was rescued as these limbs displayed little or no webbing (n=14 limbs)(Fig. 7E,F). Furthermore, the hypodactyly phenotype due to loss of Fgf8(Lewandoski et al., 2000) was usually restored (10 of 14 forelimbs had four digits and two of 14 had three digits, and in each of six skeletal preparations generated, at least one digit lacked a phalanx, see Fig. 7Finset), demonstrating that elevated AER Fgf4 expression rescues digit formation and causes webbing in the absence of Fgf8 and Bmpr1a. Taken together, these data provide genetic evidence that normal interdigit PCD is regulated by the BMP-mediated downregulation of AER-FGFs, which act as cell survival factors of the interdigit mesenchyme.

By specifically inactivating the BMP receptor Bmpr1a in the AER of the limb bud, we uncover three aspects of BMP signaling during limb development. First, we demonstrate that BMPR1A-mediated signals are required early to induce the morphological preAER, and then later to extinguish normal AER-FGF expression; these experiments confirm and extend previous studies(Ahn et al., 2001; Pizette et al., 2001; Soshnikova et al., 2003; Zou and Niswander, 1996). Second, we demonstrate that a secreted signal, downstream of Bmpr1a,emanating from the early limb bud ectoderm to the underlying lateral plate mesenchyme, is required for normal pelvic bone development. Lastly, we define a relationship between BMP and FGF signaling and the roles that these pathways play during interdigital PCD. We provide genetic evidence that BMP signaling to the AER is required for the patterned downregulation of Fgf gene expression, which in turn encodes cell survival activity for the interdigit tissue.

Msx2-Cre is active before formation of the hindlimb preAER but after forelimb preAER induction (Sun et al.,2000). In other studies, in which Msx2-Cre has been used to study preAER/AER gene function, this pattern resulted in a phenotype that affected the hindlimb more than the forelimb (Barrow et al., 2003; Lewandoski et al., 2000; Pan et al.,2005; Sun et al.,2000; Sun et al.,2002). However, in the case of Msx2-Cre-mediated Bmpr1ainactivation, opposite phenotypes are observed: forelimb AER function is longer-lived owing to prolonged AER-FGF expression, whereas the hindlimb preAER/AER never forms. We have performed confocal imaging of forelimb interdigit mutant and normal ectoderm at E12.5-E13.5 and could not demonstrate a longer-lived morphological AER (data not shown). Moreover, we examined cell proliferation (by immunostaining for phosphorylated histone H3) and cell death(by Lysotracker Red staining) in the normal and mutant forelimb AER, and found no qualitative differences between the two (data not shown). Therefore, we do not suggest that BMPR1A signaling is required for the normal regression of the AER that occurs at about E12.5, but rather that BMPR1A signals specifically regulate the cessation of AER-FGF expression.

It is formally possible that the different Msx2-Cre;Bmpr1a^flox/null AER phenotypes reflect inherent differences in BMP signaling between forelimb and hindlimb development. However, we reject this idea on the basis of work undertaken in chick embryos, in which the results of manipulations of BMP signaling in both wing and leg bud development at early and late embryological time points are more consistent with a dynamic role for BMP signaling within each limb bud(Pizette et al., 2001; Pizette and Niswander, 1999). Thus, we interpret our data as supporting a hypothesis in which, within each limb bud, BMP signals to the early ectoderm are required to induce the preAER,and later signals to the AER are required to cease AER-FGF expression. This is consistent with the data of Ahn et al. (Ahn et al., 2001), in which a variable degree of hindlimb AER disruption occurred due to the AER-specific inactivation of Bmpr1a.The Cre line used by Ahn et al. is active in the hindlimb ectoderm at about E9.75, which is later than Msx2-Cre activation (at 24ss, ∼E9.5). Hence we propose that E9.75 is in the temporal window when BMP signaling is beginning to cease functioning in hindlimb AER induction.

The molecular basis for this switch in BMP function might be an intrinsic property of preAER/AER cells, i.e. over time the competence of these cells changes such that at early stages they respond to BMPR1A signals by forming an AER, and at later stages by shutting down AER-FGF expression. Alternatively,the integration of BMP with other signals might translate to different cellular behaviors. In support of this, mesenchymal FGF10 signals and ectodermal WNT3 signals are both required for AER formation(Barrow et al., 2003; Min et al., 1998; Sekine et al., 1999).

Surgical removal of chick ectoderm prior to limb bud formation prevents normal pelvic bone development, suggesting that a preAER factor signals to the lateral flank mesoderm for pelvic girdle development(Malashichev et al., 2005). Also, certain mouse mutations that prevent hindlimb AER induction also affect pelvic bone development, including defects in p63(Trp63-Mouse Genome Informatics), Wnt3 orβ -catenin (Barrow et al.,2003; Mills et al.,1999; Yang et al.,1999). Our observation that the pelvic bones in Msx2-Cre;Bmpr1a^flox/null mice are reduced to two elements resembling rudimentary ischia/ilia, indicates that the preAER factor is downstream from BMPR1A and secreted from the Msx2-Cre expression domain.

Pelvic bones are also diminished in Fgf10^-/- animals(Min et al., 1998; Sekine et al., 1999). Given that mesenchymal Fgf10 expression is not maintained in Msx2-Cre;Bmpr1a^flox/null hindlimb buds (owing to a lack of AER-FGF signals), one might consider that loss of Fgf10 expression causes the Msx2-Cre; Bmpr1a^flox/null pelvic girdle phenotype. However, we speculate that this is not the case, because Fgf10expression is also not maintained in Msx2-Cre; Fgf4^flox/null;Fgf8^flox/null mice and yet there is no similar pelvic bone phenotype in these mutants (Sun et al.,2002). [A very subtle pelvic bone phenotype occurs, but with all pelvic bones present in Msx2-Cre; Fgf4^flox/null;Fgf8^flox/null mutants (B. Rosenman and C. O. Lovejoy, personal communication)]. Indeed, our observations suggest the novel interpretation that the Fgf10^-/- pelvic girdle phenotype might be non-cell-autonomous in that it could be a secondary effect resulting from the lack of a preAER signal.

Consideration of the relatively normal pelvic bones in Msx2-Cre;Fg4^flox/null; Fgf8^flox/null mutants also indicates that neither of the principal AER-FGFs, FGF4 and FGF8, is the secreted factor downstream of BMPR1A required for pelvic development. WNT signaling might be involved as BMPR1A has been shown to control WNT-mediated limb bud outgrowth(Soshnikova et al., 2003) and Msx2-Cre inactivation of Wnt3 similarly affects pelvic bone development (Barrow et al.,2003).

In Msx2-Cre; Bmpr1a^flox/null mutants, a lack of normal interdigital PCD causes forelimb cutaneous syndactyly to occur between all forelimb digits. During normal development, elevated BMP levels within the interdigit region have been proposed to be a direct effector of cell death within this region (Zuzarte-Luis and Hurle, 2002; Zuzarte-Luis and Hurle, 2005). However, no change indicative of a reduction of BMP signaling occurred in the mesenchymal gene expression levels of BMP ligands(Bmp2, 4 and 7), the antagonist gremlin, or the downstream targets Msx1 and 2, indicating that this proposed BMP role does not cause mutant webbing. Instead, a prolonged period of AER Fgf4 and Fgf8 expression occurred in mutants. By examining Msx2-Cre; Bmpr1a^flox/null mutants in which we reduced AER-FGF activity by also inactivating Fgf4 and/or 8, we observed a partial rescue of normal interdigit PCD with one copy of Fgf8 inactivated, and a more complete rescue with both copies of Fgf8 and one copy of Fgf4 inactivated. Thus, we provide genetic evidence that one mechanism by which BMPs can regulate interdigit PCD is through the regulation of AER-FGFs acting as cell survival factors.

In the course of these experiments, we found that inactivation of both copies of Fgf8 (with both Fgf4 copies intact) in Msx2-Cre; Bmpr1a^flox/null mutants did not rescue normal interdigit development because such rescue was prevented by elevated Fgf4 expression levels that were higher than the elevation due to loss of Fgf8 alone. This elevated Fgf4 expression rescued the hypodactyly that occurs in Msx2-Cre; Fgf8^flox/nullmutants, and hypodactyly was restored when FGF4 levels were reduced (in Msx2-Cre; Bmpr1a^flox/null;Fgf8^flox/null;Fgf4^flox/+ mutants). These data demonstrate that FGF4 can replace FGF8 function in both normal digit development and abnormal repression of interdigit cell death. Consistent with this, Msx2-Cre-mediated ectopic expression of Fgf4 has been shown to rescue limb development in Msx2-Cre;Fgf8^flox/null mutants as well as cause cutaneous syndactyly (Lu et al., 2006). However, given that during normal development, AER Fgf4 expression ceases by E11.5 (Lewandoski et al.,2000) and that Fgf8 expression ceases at E12.5 first over the interdigit regions preceding interdigital cell death(Salas-Vidal et al., 2001), we postulate that FGF8, not FGF4, is the principal AER-FGF that maintains interdigit cell survival, until it is downregulated through BMPR1A-mediated signaling (Fig. 8).

The insight that FGF8 controls interdigital cell death is reminiscent of studies in which Fgf8 inactivation causes an abnormal increase in cell death in the first branchial arch(Trumpp et al., 1999), the brain (Chi et al., 2003; Storm et al., 2003), the early limb bud (Moon and Capecchi,2000; Sun et al.,2002) and the kidney(Grieshammer et al., 2005; Perantoni et al., 2005). However, the current study is the first report demonstrating that the embryo regulates Fgf8 expression to control normal cell death during development.

Our genetic experiments support speculations, based on gene expression patterns (Salas-Vidal et al.,2001) and the effects of FGF-bearing bead insertion into chick limb buds (Montero et al.,2001; Macias et al.,1996), that AER FGF8 activity might act as a cell survival activity for the interdigit region. Genetic proof for this idea was not obtainable by a simple Msx2-Cre-mediated inactivation of Fgf8 itself(in which one would expect premature interdigit cell death), because the overall development of the autopod is retarded(Lewandoski et al., 2000). However, this model is supported by the observation that cutaneous syndactyly in the classical human craniosynostosis syndrome, Pfeiffer's, is caused by mutations causing constitutive signaling of the FGFR2c isoform, which is active in limb bud mesenchyme and is thought to act as an FGF8 receptor. Also,we propose that excess AER-FGF activity might be the mechanism by which syndactyly occurs in a number of mouse mutants in which an expanded AER is evident, which include mouse lines carrying mutations affecting the NOTCH and WNT pathways (Jiang et al.,1998; Mukhopadhyay et al.,2001; Pan et al.,2005; Sidow et al.,1997). This proposal can be tested directly by specifically inactivating Fgf4 and 8 in the AER of these mutants as we have done here.

A supernumerary pre-axial digit occurs in 25% of Msx2-Cre;Bmpr1a^flox/null forelimbs, as well as in other mouse mutants with enhanced AER-FGF signaling (Adamska et al., 2003; Lu et al.,2006; MacDonald et al.,2004; Wang et al.,2004). However, the prolonged AER-FGF expression in mutant forelimbs was not associated with the formation of extra phalanges. This contrasts with what has been observed in chick embryos, in which prolongation of AER Fgf8 expression can cause an additional penultimate phalange within a digit, supporting a model in which the duration of AER FGF8 signaling controls elongation of digit primordia and the normal decrease in AER FGF8 signaling generates digit tip formation(Sanz-Ezquerro and Tickle,2003). The discrepancy between these studies may be explained by inherent differences in AER FGF8 function between mouse and chick limb buds. Alternatively, extra phalange development might also require enhanced SHH as well as FGF8 signaling, considering that AER Fgf8 expression was prolonged in the chick studies by the insertion of SHH-soaked beads in the underlying mesenchyme (Sanz-Ezquerro and Tickle, 2003). However, consistent with these chick studies we did observe thickened digit tips in all Msx2-Cre;Bmpr1a^flox/null forelimbs.

In our model (Fig. 8), we propose that FGF8 acts as a cell survival factor for the interdigit region;its normal downregulation via BMPR1A-mediated signals allows interdigit PCD to occur. In our mutant model, Bmpr1a inactivation results in prolonged FGF4 and FGF8 signaling, leading to abnormal cell survival and webbed limbs. What is the source of these interdigital BMP signals during normal development? We have examined Bmp gene expression in the AER prior to and during the initiation of interdigital PCD (E12.5-13.5), and found that that Bmp2 and 7 are not expressed in the AER, whereas Bmp4 is specifically expressed in the AER/subapical region overlying the digits. However, because Bmp2, 4 and 7 are upregulated in the interdigit mesenchyme prior to PCD(Hogan, 1996; Salas-Vidal et al., 2001), we speculate in our model that this tissue is the BMP source. It is tempting to further speculate that elevated BMP activity during development of the digits,as chondrogenesis proceeds (Yoon and Lyons, 2004), might cause the cessation of AER Fgf8expression overlying the digits. Also, earlier studies that described the inhibition of interdigit PCD in chick embryos through the blocking of BMP signaling with retroviral expression of transgenes encoding dominant-negative BMP receptors (Yokouchi et al.,1996; Zou and Niswander,1996), may be reinterpreted in the context of our model as these retroviral vectors might have inhibited BMP signaling to the AER, elevating Fgf gene expression. Alternatively, our demonstration that BMPs regulate interdigital cell death through the regulation of AER-FGF expression does not exclude a direct role for BMPs as effectors of cell death. Evaluating such a role requires the inactivation of Bmp receptor genes specifically within the interdigit region without affecting AER morphology or gene expression.

Lastly, these data have significant implications for the evolution of limb morphology. Considering the well-established role that the AER plays in the outgrowth of skeletal limb elements, it has been proposed that evolutionary changes in AER-FGF activity might reduce or extend the different cartilaginous and bony limb structures that occur in different species(Cohn and Tickle, 1999; Freitas et al., 2006; Richardson and Oelschlager,2002; Thewissen et al.,2006). The data presented here suggest that evolutionary alterations in AER-FGF activity might also contribute to the shaping of soft-tissue structure in vertebrates, resulting in normal free digits or webbed limbs. For example, a recent study provides evidence that FGF8-mediated signaling is pivotal in maintaining the interdigit webs in the bat wing(Weatherbee et al., 2006). Future studies into the targets of AER-FGF activity and the interaction of FGF signaling with other pathways are needed to understand how AER-FGF signaling differentially affects the patterning of bony structures and soft tissue.

This research was supported by the Intramural Research Program of the NIH,National Cancer Institute, Center for Cancer Research. We thank L. Grotewold for the Bambi probe; J. Matta for expert technical assistance; B. Rosenman and O. Lovejoy for helpful discussion on pelvic bone development; S. Mackem, N. Moore, L. Niswander, A. Perantoni, X. Sun and T. Yamaguchi for manuscript comments; and L. Niswander for making data available prior to publication.