Increasing Fgf4 expression in the mouse limb bud causes polysyndactyly and rescues the skeletal defects that result from loss of Fgf8 function

The apical ectodermal ridge (AER), a specialized epithelium that rims the distal tip of the limb bud, is a major source of signals required for vertebrate limb development (Martin,1998; Capdevila and Izpisua Belmonte, 2001). In the mouse, AER formation begins at the stage when the limb bud can first be discerned. It involves the initiation of gene expression that later marks the mature AER, movement distally of cells within the epithelium (particularly on the ventral side of the limb bud) and changes in cell shape. After the AER matures into a narrow band of pseudostratified columnar cells, it persists for 3-4 days while the limb bud enlarges and cells in the underlying mesenchyme become determined to form the progenitors of the skeletal elements and some of the other mesodermal cell types in the mature limb. At the end of this period, cells in the AER die and the AER regresses(Guo et al., 2003).

The importance of the AER has been well-established by experiments in the chicken embryo showing that AER removal causes limb truncations(Saunders, 1948; Summerbell, 1974; Rowe and Fallon, 1982), and that this is due primarily to death of the underlying limb bud mesenchyme(Rowe et al., 1982; Dudley et al., 2002). The fact that such limb truncations can be rescued by implanting beads soaked in FGF proteins (Niswander et al.,1993; Fallon et al.,1994) indicates that an important function of the AER is to produce FGFs. In addition to sustaining cell survival in limb bud mesenchyme(Sun et al., 2002; Boulet et al., 2004), FGF signaling from the AER is thought to be an essential component of a positive gene regulatory feedback loop that maintains the expression of sonic hedgehog(Shh) and its downstream targets in the mesenchyme(Laufer et al., 1994; Niswander et al., 1994), and also to regulate the expression of other genes vital for limb development(Martin, 1998; Capdevila and Izpisua Belmonte,2001).

Of the 22 mouse genes that encode FGF ligands(Itoh and Ornitz, 2004), four(Fgf4, Fgf8, Fgf9 and Fgf17) are expressed in the AER but not in other cells in the limb bud. Fgf8 is unique among these`AER-FGFs' in several respects. Fgf8 expression is first detected in AER progenitors in the ventral ectoderm, and subsequently along the entire anteroposterior (AP) length of the AER until it regresses(Crossley and Martin, 1995; Mahmood et al., 1995). By contrast, Fgf4, Fgf9 and Fgf17 expression commences only after the AER has fully formed, is restricted to the posterior half to two-thirds of the AER, and ceases at least 1 day before AER regression(Sun et al., 2000). Furthermore, gene knockout studies have shown that when each of the AER-FGFs is individually inactivated, only loss of Fgf8 function causes limb abnormalities (Lewandoski et al.,2000; Moon and Capecchi,2000); no defects have been detected in Fgf4(Moon et al., 2000; Sun et al., 2000), Fgf9 (Colvin et al.,2001) or Fgf17 (Xu et al., 2000) null limbs. However, Fgf4 is essential in the absence of Fgf8, because if both Fgf4 and Fgf8 are never expressed in the limb bud, the limb does not form(Sun et al., 2002; Boulet et al., 2004).

One model to explain these observations is based on the premises that a certain amount of AER-FGF is required at each stage of limb development, and that the contributions made by FGF4 and FGF8 (and perhaps the other AER-FGFs)to the ligand pool are functionally similar. According to this model, there are two reasons why inactivation of Fgf8, but not of other individual AER-FGF genes, causes abnormalities in limb development. First, at some stages(e.g. in the early and late limb bud) Fgf8 is the only AER-FGF gene that is expressed and thus is contributing to the ligand pool. Second, at stages when all four AER-FGF genes are co-expressed in the posterior limb bud, Fgf8 is expressed more abundantly and more broadly than the others,and therefore makes a more substantial contribution to the ligand pool. In either case, when any other AER-FGF gene is individually is inactivated, the total amount of ligand in the pool does not fall below the level needed for normal limb development. However, when both Fgf4 and Fgf8are completely inactivated, there is insufficient ligand to sustain limb development. An alternative model is that FGF8 has unique functions in limb skeletal patterning that FGF4 (and presumably the other AER-FGFs) cannot perform.

To distinguish between these models and to determine the consequences of increasing Fgf4 expression in the limb bud, we employed a conditional gain-of-function (GOF) approach. We found that activating an Fgf4^GOF transgene in wild-type limb buds, in a domain that largely resembles that of Fgf8 in the normal limb bud, caused several abnormalities, including formation of a supernumerary posterior digit(postaxial polydactyly) and inhibition of programmed cell death in the interdigital mesenchyme, resulting in retention of tissue between the digits(cutaneous syndactyly); together these phenotypes are termed polysyndactyly. Activation of the Fgf4^GOF transgene in an Fgf8-null limb bud still caused polysyndactyly, but it also rescued all the skeletal defects that result from loss of Fgf8 function. Together, these data support the hypothesis that AER-FGF signaling plays a role in regulating digit number and cell death in the interdigital mesenchyme,and that FGF4 and FGF8 perform similar functions in patterning the limb skeleton.

To construct the Fgf4^GOF transgene, we replaced the human placental alkaline phosphatase sequence in the Z/AP vector(Lobe et al., 1999) (kindly provided by Dr Andras Nagy) with a fragment of a mouse cDNA containing the entire Fgf4 open reading frame. The Fgf4 and loxPsegments of the resulting construct (Fig. 1A) were sequenced to confirm that no mutations had arisen during vector construction.

The construct was electroporated into a subclone of E14Tg2a ES cells(Hooper et al., 1987) isolated by A. Smith (kindly provided to us by W. Skarnes). A total of 384 G418-resistant clones were isolated, and an aliquot of each clone was assayed for β-galactosidase (β-gal) activity(Lobe et al., 1999). Of these,the 36 clones with the strongest and most uniform staining were selected for further study. An aliquot of each was frozen, and another aliquot was expanded and cultured under conditions appropriate for embryoid body formation(Martin and Evans, 1975). To obtain differentiation to a wide variety of cell types, the embryoid bodies from each clone were allowed to attach to a tissue culture surface. After∼8-10 days of culture, when differentiated cells including rhythmically contracting myocardial cells were observed, the cells were fixed and stained for β-gal activity. Nine of the clones were chosen for further analysis because their differentiated derivatives almost all displayed strongβ-gal activity; in the remaining clones, β-gal activity was either weak or not detected in their differentiated derivatives. Chromosome analysis was performed on six out of these nine clones, and more than 80% of the cells examined for each clone were found to contain 40 chromosomes. When assayed by Southern blotting using a variety of restriction enzymes, the transgene in all of these six clones appeared to have inserted into a single site.

Two of these ES cell clones were injected into C57BL/6 blastocysts by the Stanford University transgenic facility, and germ-line transmission of the Fgf4^GOF transgene was obtained from male chimeras produced from one of these cell lines. The mouse line thus established transmitted the transgene at Mendelian frequency. When crossed inter se, viable homozygotes were obtained, but they bred poorly. The Fgf4^GOF transgene was maintained on a mixed genetic background.

The genotypes of embryos and adult mice were determined by PCR assays using DNA extracted from embryonic yolk sacs or tails as a template. The presence of the Fgf4^GOF allele was detected using primers that amplify the lacZ sequence in the transgene(5′-GTCTCGTTGCTGCATAAACC-3′ and 5′-TCGTCTGCTCATCCATGACC-3′). Genotyping for Fgf8 alleles and for the presence of the Msx2-cre transgene was performed as previously described (Sun et al.,2002).

Embryos were collected in cold PBS, fixed in 4% paraformaldehyde and stored in 70% methanol at -20°C. Noon of the day when a vaginal plug was detected was considered approximately embryonic day (E) 0.5. To stage embryos more precisely, the somites posterior to the forelimb bud were counted and the total number of somites was determined by scoring the first one counted as somite 13. Standard protocols were employed for RNA in situ hybridizations and skeletal preparations. Assays for cell death were performed in whole mount, by staining with LysoTracker (Molecular Probes L-7528)(Grieshammer et al.,2005).

Limb buds were dissected from 5 Msx2-cre;Fgf4^GOF and two control(Msx2-cre hemizygous) embryos at E10.5 and preserved in RNAlater(Ambion, 7020). Each sample contained either the two forelimb buds or the two hindlimb buds from an individual embryo. Total RNA was isolated using an RNeasy Kit (Qiagen 74104). First-strand cDNA synthesis was performed by the Biomolecular Resource Center at UCSF using a Bio-Rad kit (170-8891). Quantitative PCR was performed using an ABI Prism 7900 HT Sequence Detection System. The primers amplified a 57 bp sequence spanning sequences at the 3′ end of exon 2 to sequences at the 5′ end of exon 3 of the Fgf4 gene: forward primer 5′-GAAAGGCACACCGAAGAGCTT-3′ and reverse primer 5′-GCCAGCCGGTTCTTCGT-3′. The sequence of the Fgf4 TaqMan probe was 5′-CCTGCTGCTCATAGC-3′. Gapdh sequences were also amplified and the amount of product was used to normalize each sample.

To determine the effect of increasing Fgf4 expression in the limb bud, we produced a mouse line carrying the Fgf4^GOFtransgene illustrated in Fig. 1A (see Materials and methods). Assays for β-gal activity at various stages of development demonstrated that prior to Cre-mediated recombination, the CAGG promoter in this transgene, which contains CMV enhancer/chicken β-actin sequences(Niwa et al., 1991), drives expression of a floxed (flanked by loxP sites) β-Georeporter gene in most or all cells of the embryo(Fig. 1B, and data not shown). After Cre-mediated recombination, which excises the intervening floxedβ-Geo sequences, the promoter drives expression of an Fgf4 cDNA (Fig. 1A) in the cells in which recombination occurred and their descendants, presumably for as long as they survive. Fgf4^GOF mice were crossed to animals carrying Msx2-cre, a transgene that expresses cre in early limb bud ectoderm, as well as in other tissues of the embryo. We previously used Msx2-cre to inactivate Fgf4(Sun et al., 2000), Fgf8 (Lewandoski et al.,2000), and both Fgf4 and Fgf8(Sun et al., 2002) during limb development. At birth, Msx2-cre;Fgf4^GOF offspring were present at Mendelian frequency and were viable, but generally grew more slowly than did their wild-type littermates. All Msx2-cre;Fgf4^GOFanimals displayed limb abnormalities, as described below, as well as hyperplasia of the eyelids, malformations of the external genitalia and other defects (not shown). These animals carrying the activated transgene (A-Tg) are described here in comparison with their `normal' littermates, which were wild-type or carried either Msx2-cre or Fgf4^GOF,but not both transgenes.

During normal limb development, Fgf4 is expressed only after the AER has fully formed, and therefore Fgf4 RNA is not detected until the 29- to 30-somite stage (ss) (∼E10.0) in forelimb (FL) and 33-34 ss(∼E10.25) in hindlimb (HL) buds (Sun et al., 2000) (Fig. 1D,F,H,J). At these stages and later, Fgf4 expression is normally restricted to the posterior two-thirds of the AER(Niswander and Martin, 1992)(Fig. 1H,L,M), and ceases by∼E11.5 in FL and ∼E12.5 in HL buds(Lewandoski et al., 2000)(Fig. 1N-Q). By contrast, in Msx2-cre;Fgf4^GOF embryos, as a result of Msx2-cre activity in the early limb bud(Sun et al., 2000), Fgf4 RNA was already detected at 28 ss in FL and 30 ss in HL buds, in a domain that was abnormally extended anteriorly(Fig. 1E,G,I,K, and data not shown). At ∼E10.5, Fgf4 expression was detected throughout the entire AP length of the AER (Fig. 1L,M). Moreover, Fgf4 expression persisted for at least 1 day longer than normal, until the AER itself regressed(Fig. 1N-S). Together, these data show that in Msx2-cre;Fgf4^GOF limb buds, Fgf4 expression is initiated earlier, extends throughout a longer domain and persists for longer than in normal limb buds. In these respects,the expression of Fgf4 in Msx2-cre;Fgf4^GOF limb buds was more similar to that of Fgf8 than Fgf4 in normal limb buds.

We detected one significant difference between Fgf4 expression in Msx2-cre;Fgf4^GOF limb buds and the normal expression patterns of both Fgf4 and Fgf8. In the normal embryo, Fgf4 RNA is never detected in ventral limb bud ectoderm, and Fgf8 RNA is detected in that domain only transiently, presumably in AER progenitors before they move distally and become incorporated into the nascent AER (Guo et al.,2003). By contrast, because Msx2-cre is active in a few cells that remain in the ventral ectoderm(Barrow et al., 2003), Fgf4 RNA was detected at a low level in the ventral ectoderm of Msx2-cre;Fgf4^GOF limb buds from early stages until at least E10.5 and presumably longer(Fig. 1I; data not shown).

To quantify the increase in Fgf4 expression in Msx2-cre;Fgf4^GOF limb buds at E10.5, we performed a quantitative RT-PCR assay (see Materials and methods). The level of Fgf4 RNA was found to be 3.5-fold higher in FL and twofold higher in HL buds from Msx2-cre;Fgf4^GOF embryos than from control (Msx2-cre hemizygous) embryos(Fig. 1C). From our in situ hybridization data, it is clear that at this stage, some of the Fgf4expressed by the activated Msx2-cre;Fgf4^GOF transgene is localized in two ectopic domains, i.e. regions in which Fgf4 is never expressed in the normal limb bud: one in the anterior AER, where Fgf8is normally expressed, and the other in the ventral ectoderm, where AER-FGF gene expression is not normally detected. Because Msx2-cre is known to function throughout the AER (Sun et al., 2000), the remainder of the Fgf4 produced by the activated Fgf4^GOF transgene is presumably localized within the normal Fgf4 expression domain in the posterior AER. However,expression of Fgf4 in this domain occurs both earlier and for longer than normal in Msx2-cre;Fgf4^GOF limb buds.

Before assessing the effects of the excess/ectopic Fgf4 expression on limb morphogenesis, we examined Msx2-cre;Fgf4^GOF limb buds to determine whether the AER itself or the expression of other AER-FGF genes was perturbed. Fgf8 expression appeared normal in Msx2-cre;Fgf4^GOF limb buds at E10.5 and E11.5(Fig. 2A-B′, and data not shown), indicating that the AER, both with respect to its length along the AP axis and its width along the dorsoventral (D-V) axis, was normal. The expression of the other AER-FGF genes, Fgf9 and Fgf17, also appeared similar in Msx2-cre;Fgf4^GOF and normal limb buds(data not shown). Significantly, we found that at E13.5, Fgf8expression was detected in both normal and Msx2-cre;Fgf4^GOF limb buds in a thin domain at the distal tip of each of the developing digits, but not in the regions overlying the interdigital mesenchyme (Fig. 2C,D, and data not shown). These observations indicate that the AER regresses at the same stage in Msx2-cre;Fgf4^GOF and normal limb buds. One consistent difference we observed was that by E11.5 the limb buds of the Msx2-cre;Fgf4^GOF embryos were slightly larger than those of their normal littermates (Fig. 1N-Q).

As signaling by AER-FGFs is thought to influence outgrowth and patterning of the limb bud mesenchyme, at least in part via its contribution to a positive feedback loop involving Sonic hedgehog (Shh) and its downstream target Gremlin (Grem1)(Laufer et al., 1994; Niswander et al., 1994; Zuniga et al., 1999; Khokha et al., 2003; Michos et al., 2004; Scherz et al., 2004), we next investigated whether Shh or Grem1 expression was affected in Msx2-cre;Fgf4^GOF limb buds. We found that both Shh and Grem1 expression patterns appeared normal at∼E10.5 (Fig. 2E-H). However, at ∼E11.5 we detected a significant abnormal expansion of the Shh expression domain both ventrally and especially proximally, with the effect being much more pronounced in HL than in FL buds(Fig. 2I,J). Grem1expression was similarly affected, but to a lesser extent(Fig. 2K,L). Analysis of sections through the limb buds demonstrated that these ectopic Shhand Grem1 expression domains were localized exclusively in the mesenchyme (data not shown). Such ectopic mesenchyme gene expression in the Msx2-cre;Fgf4^GOF limb buds might be due to either excess FGF4 produced in the AER, or more likely to ectopic Fgf4 expression in the ventral ectoderm (Fig. 1I; data not shown), because the expansion of the Shh and Grem1 expression domains was detected in ventral but not dorsal limb bud mesenchyme, and was not observed in the distal-most region near the AER.

We next determined the consequences for skeletal patterning of the excess/ectopic Fgf4 expression in Msx2-cre;Fgf4^GOF limb buds. Analysis of eight Msx2-cre;Fgf4^GOF postnatal skeletons revealed that the stylopod and zeugopod were normal, but the autopod was not(Fig. 3A-F; data not shown). In all 32 limbs examined, there was a supernumerary digit on the posterior side,containing a partial or, more frequently, a complete metacarpal/metatarsal and one or two phalanges (Fig. 3B,E). The metacarpal/metatarsal was generally fused to the posterior side of the metacarpal/metatarsal of digit V. In a few cases, there were additional phalanx-like elements between digit V and the supernumerary digit (Fig. 3C). In addition,in eight out of 16 hindlimbs, but in none of the forelimbs, there was a very thin digit-like structure ventral to digit I, which was detected when the limbs were viewed from the anterior side. It consisted of a partial or more complete metatarsal, and one or two phalanges, the most distal of which was sometimes attached to the distal phalanx of digit I(Fig. 3F; data not shown). We also found that in all Msx2-cre;Fgf4^GOF hindlimbs, the proximal calcaneus was significantly enlarged(Fig. 3E). This largest of the eight elements that comprise the tarsus (ankle) is the most proximal tarsal element on the posterior side. Other carpal and tarsal elements appeared normal, although occasionally fusions of two elements were observed. Together,these data show that the effects on skeletal development of expressing Fgf4 throughout the AER as well as in the ventral ectoderm are most pronounced on the posterior side of the limb.

An in situ hybridization assay for Sox9, the earliest known marker of skeletal element progenitors (Wright et al., 1995; Akiyama et al.,2002), showed that as early as E12.5, a supernumerary skeletal condensation was present posterior to the condensation representing digit V(Fig. 3G-J). At this stage,there were no obvious supernumerary condensations on the anterior ventral side of Msx2-cre;Fgf4^GOF limb buds. These data indicate that formation of the supernumerary postaxial digit occurred during limb bud stages(prior to the regression of the AER).

Another abnormality observed in Msx2-cre;Fgf4^GOFautopods was the presence of soft tissue between the digits(Fig. 4A-D), a phenotype presumably caused by a lack of the interdigital programmed cell death (IPCD)that normally occurs in the mouse limb at ∼E12.5-13.5. To confirm this, we assayed Msx2-cre;Fgf4^GOF embryos at ∼E13.5 by staining with LysoTracker, which labels acidic compartments (lysosomes) within apoptotic cells themselves as well as in healthy cells that are engulfing apoptotic debris (Zucker et al.,1999; Schaefer et al.,2004). In hindlimbs of normal embryos, intense LysoTracker staining was observed in the interdigital mesenchyme(Fig. 4E,E′); by contrast, we saw little or no LysoTracker staining in the hindlimbs of their Msx2-cre;Fgf4^GOF littermates(Fig. 4F,F′). These data suggest that abnormally prolonged expression of Fgf4 in the AER (see Fig. 1N-Q), or possibly a low level of ectopic expression in the ventral ectoderm, prevents cell death in the interdigital mesenchyme of Msx2-cre;Fgf4^GOFembryos.

There is an extensive literature indicating that IPCD in the chicken embryo is mediated by BMP signaling and showing that IPCD can be prevented by ectopic expression of BMP antagonists(Zuzarte-Luis and Hurle,2005). To determine whether the observed lack of IPCD in Msx2-cre;Fgf4^GOF limb buds was correlated with changes in the level of expression of genes involved in BMP signaling, we assayed for Bmp2 and Bmp7, two members of the BMP family known to be expressed in the interdigital mesenchyme(Hogan, 1996), for Msx2, which is thought to be a downstream target of BMP signaling in the limb (Pizette et al.,2001), and for Grem1, which encodes an antagonist of BMP signaling in the limb (Hsu et al.,1998; Khokha et al.,2003; Michos et al.,2004). In situ hybridization assays showed no reproducible difference in the expression of these genes in Msx2-cre;Fgf4^GOF when compared with normal limb buds at E12.5 (Fig. 4G-J),approximately 1 day earlier than the stage at which IPCD was detected in normal but not in Msx2-cre;Fgf4^GOF hindlimb buds. These data suggest that excess/ectopic Fgf4 expression prevents IPCD via a mechanism that does not involve changes in the expression of genes involved in BMP signaling.

Finally, we sought to determine whether the Fgf4 expressed by the activated Fgf4^GOF transgene can functionally replace Fgf8 expression in the limb bud. We crossed male mice carrying both Msx2-cre and an Fgf8 null allele, Fgf8^Δ, to females carrying an Fgf8conditional allele, Fgf8^flox, and Fgf4^GOF, to generate Msx2-cre;Fgf8^Δ/flox;Fgf4^GOFmice. In these animals, Msx2-cre-mediated recombination inactivates Fgf8 by converting Fgf8^flox to a null allele and concomitantly activates Fgf4 expression from the Fgf4^GOF transgene. Msx2-cre;Fgf8^Δ/flox littermates from this cross displayed an Fgf8 loss-of-function phenotype similar to, but somewhat less severe than that previously described(Lewandoski et al., 2000): the FL stylopod was usually mildly hypoplastic and missing the deltoid tuberosity(Fig. 5A,E), whereas the HL stylopod was often severely reduced (Fig. 5B,F); the zeugopod elements were mildly hypoplastic(Fig. 5A,B,E,F); and the autopod was frequently missing a digit: digit II or III in the forelimb(Fig. 5C,G, and data not shown)and digit I in the hindlimb (Fig. 5D,H). In addition, FL digits I and V and HL digits II and V were frequently missing a phalanx (Fig. 5C,D,G,H; data not shown).

Significantly, the defects that result from loss of Fgf8 function were completely rescued in all Msx2-cre;Fgf8^Δ/flox;Fgf4^GOF skeletons. Thus, the deltoid tuberosity was always present(Fig. 5I), the femur appeared normal (Fig. 5J), and digits I-V were all present in both the FL and HL autopod and contained the normal number of phalanges (Fig. 5K,L). However, all Msx2-cre;Fgf8^Δ/flox;Fgf4^GOF limbs examined (n=7) still displayed the abnormalities observed when the Fgf4^GOF transgene was activated by Msx2-cre in embryos that were wild type for Fgf8(Fig. 3), including enlargement of the calcaneus and a supernumerary posterior digit(Fig. 5I-L), as well as cutaneous syndactyly (data not shown). It is not clear precisely which aspects of the Msx2-cre-activated Fgf4^GOF expression pattern are responsible for each of these specific abnormalities (see Discussion). What is most important about these data is that they clearly demonstrate that Fgf4 is capable of performing the functions of Fgf8 that are essential for normal limb skeletal development.

The importance of FGF signaling from the AER to the limb bud mesenchyme is firmly established, but the functional significance of the expression of multiple FGF family members in the AER is not well understood. Using a Cre-mediated conditional gain-of-function strategy, we have shown that the excess/ectopic expression of Fgf4 that results from activation of an Fgf4^GOF transgene by Msx2-cre function causes postaxial polydactyly and other skeletal abnormalities, as well as cutaneous syndactyly between all the digits. These data underscore the importance for normal limb morphogenesis of precisely controlling the amount and spatiotemporal aspects of AER-FGF gene expression. Furthermore, using a novel approach for assessing the functional equivalence of genes in the mouse, we have demonstrated that FGF4 can perform the function(s) of FGF8 in patterning the limb skeleton.

Based on the results of AER-FGF gene inactivation studies, it has been proposed that a key role of AER-FGF signaling is to establish the number of chondrocyte progenitors available to form limb skeletal elements(Sun et al., 2002), as a result of an anti-apoptotic effect and/or stimulation of cell proliferation. Our data, showing that the excess/ectopic Fgf4 expression in Msx2-cre;Fgf4^GOF limb buds results in the formation of supernumerary skeletal elements, are consistent with this hypothesis. However,because the normal pattern of Fgf4 expression is altered in several ways in Msx2-cre;Fgf4^GOF limb buds (including increased and prolonged expression in the posterior AER, and ectopic expression in the anterior AER and ventral ectoderm), we can only speculate on the source of the FGF4 that causes the observed defects in the autopod skeleton.

For example, the ventral location of the thin anterior digit-like structure found in many Msx2-cre;Fgf4^GOF hindlimbs suggests that it forms in response to ectopic Fgf4 expression in either the ventral ectoderm or the anterior AER. By contrast, it seems likely that excess Fgf4 expression in the posterior AER is responsible for the formation a supernumerary digit on the posterior side of Msx2-cre;Fgf4^GOF limbs, based on evidence from several mouse mutants. For example, mice with mutations in Dkk1(Adamska et al., 2003; MacDonald et al., 2004) or in which the BMP antagonist Noggin is ectopically expressed in the AER(Wang et al., 2004) display an increase in AER-FGF expression and have a supernumerary digit on the posterior side. However, if the supernumerary posterior digit in Msx2-cre;Fgf4^GOF limbs is indeed due to excess FGF expression in the AER, then why does an extra digit still form in Msx2-cre;Fgf8^Δ/flox;Fgf4^GOFlimbs, which lack Fgf8 expression? A possible explanation comes from the observation that in the absence of Fgf8 function in the AER,endogenous Fgf4 expression is significantly upregulated(Lewandoski et al., 2000; Moon and Capecchi, 2000),although obviously not to the level needed for formation of the normal complement of skeletal elements. Thus, we suggest that the AER-FGF produced by the expression of the activated Fgf4^GOF transgene in combination with the normal levels of endogenous Fgf4 and Fgf8 expression in Msx2-cre;Fgf4^GOF limbs or in combination with upregulated endogenous Fgf4 expression in Msx2-cre;Fgf8^Δ/flox;Fgf4^GOF limb buds, is sufficient to produce six digits.

We were surprised to find that Msx2-cre;Fgf4^GOF limbs did not contain one or more supernumerary digits on the anterior side of the limb, because there are numerous mouse mutants in which Fgf4expression in the anterior AER is correlated with such preaxial polydactyly(Masuya et al., 1995; Qu et al., 1997; Hayes et al., 1998; Yang et al., 1998; Yada et al., 2002; Zhang et al., 2003). However,in those mutants the anterior expansion of the Fgf4 expression domain appears to be a secondary response to ectopic Hedgehog pathway activity in the anterior mesenchyme (Talamillo et al.,2005; Liu et al.,2005). As yet, there are no data addressing the issue of whether such ectopic Fgf4 expression in the anterior AER is necessary for the formation of the supernumerary preaxial digit(s).

Together, the data suggest a model in which excess AER-FGF signaling per se promotes the formation of a supernumerary digit on the posterior side,possibly by increasing SHH signaling or by expanding the number of descendants of Shh-expressing cells, which are known to contribute to posterior digits (Ahn and Joyner, 2004; Harfe et al., 2004). By contrast, changes in the expression of other genes, such as those that cause ectopic Hedgehog pathway activity in anterior mesenchyme, are required for supernumerary digit formation on the anterior side. According to this model,the postaxial and preaxial polydactyly reported in mouse hindlimbs in which Bmp4 has been conditionally inactivated in the mesenchyme(Selever et al., 2004) can be explained by two separate effects: the supernumerary posterior digit forms in response to an observed increase in AER-FGF gene expression resulting from expansion of the AER along the DV axis, and the additional supernumerary anterior digit forms as a consequence of other changes in gene expression in the anterior mesenchyme. In this context it is interesting that in embryos null for both Msx1 and Msx2, which are thought to be required for BMP signaling, preaxial polydactyly seems to correlate with a very small ectopic anterior domain of Shh expression(Lallemand et al., 2005). However, it is important to note that anterior mesenchyme is capable of forming supernumerary digits in response to FGF signaling, as for example in transgenic mice expressing a hypermorphic allele of Fgfr1(Hajihosseini et al., 2004),and in chicken limb buds implanted with cells producing FGF2(Riley et al., 1993). Curiously, in the latter experiments, posterior mesenchyme showed no response to FGF2-producing grafts. Further studies will be needed to elucidate the mechanism(s) that regulate digit number and determine the location(s) at which supernumerary digits form.

In addition to abnormalities in skeletal pattern, Msx2-cre;Fgf4^GOF limbs have cutaneous syndactyly between all the digits. As in many mouse mutants, such syndactyly correlates with a lack of interdigital programmed cell death (IPCD)(Zuzarte-Luis and Hurle,2005). A potential link between AER-FGF signaling and IPCD is suggested by the hypothesis that AER-FGFs provide proliferation/survival factors for interdigital mesenchyme cells(Macias et al., 1996), and that under normal circumstances IPCD is triggered by the loss of these factors when the AER regresses. Thus, a delay in AER regression should delay or block IPCD and result in persistence of interdigital tissue. Support for this model comes from gain-of-function experiments in both chicken limb buds and transgenic mice, showing that ectopic expression of the BMP antagonist Noggin results in a delay in AER regression and cutaneous syndactyly(Pizette and Niswander, 1999; Wang et al., 2004). Based on those studies, it was proposed that during normal limb development, one role of BMP signaling is to promote AER regression. In support of this hypothesis,mice that lack both Msx1 and Msx2 function(Lallemand et al., 2005), or in which Bmpr1a has been inactivated in the AER (M. Lewandoski,personal communication), display abnormal AER regression and persistence of interdigital tissue.

However, AER regression does not appear to be delayed in Msx2-cre;Fgf4^GOF limb buds, and other effects besides delay of AER regression are observed in mouse limb buds lacking Msx1/Msx2 function, lacking Bmpr1a function in the AER, or with ectopic Noggin expression in the AER. In particular, the AER is expanded along its DV axis (Wang et al.,2004; Lallemand et al.,2005) (M. Lewandoski, personal communication), raising the possibility that excess/ectopic AER-FGF expression from the expanded AER,rather than a delay in AER regression per se, accounts for the observed persistence of interdigital tissue in those limb buds. If so, then the cutaneous syndactyly we observed in Msx2-cre;Fgf4^GOF limbs might likewise be explained by excess/ectopic Fgf4 expression in the AER, or perhaps in the ventral ectoderm. Such FGF signaling might directly oppose the pro-apoptotic effects of BMP signaling(Yokouchi et al., 1996; Zou and Niswander, 1996; Macias et al., 1997), as suggested by studies showing that implanting BMP4-soaked beads into interdigital mesenchyme causes cell death and this is prevented by co-implanting FGF2-soaked beads (Ganan et al., 1996). A more complex relationship between FGF and BMP signaling in the control of IPCD has also been suggested(Montero et al., 2001). Alternatively, excess FGF signaling might inhibit BMP activity via effects on genes involved in BMP signaling. However, we saw no effect on the expression of Bmp2, Bmp4, Msx2 or Grem1 in Msx2-cre;Fgf4^GOF limb buds just prior to the stage when IPCD normally occurs. Further studies will be needed to unravel the potentially complex interactions between FGF and BMP signaling in the regulation of IPCD.

Although FGF4 and FGF8 belong to different subgroups of the FGF family(Itoh and Ornitz, 2004) and share only ∼32% identity in the most conserved core domain(Coulier et al., 1997), they have been found to have similar activities in various developmental settings. For example, in the chicken embryo, beads soaked in recombinant FGF4 or FGF8 can induce the formation of an ectopic limb(Martin, 1998) and induce cells in the diencephalon to change fate and develop into a midbrain(Crossley et al., 1996). However, during gastrulation in the chicken embryo, primitive streak cells are attracted to a source of FGF4 and repelled by a source of FGF8(Yang et al., 2002),indicating that these two FGF family members do not necessarily perform the same function in all developmental contexts. Genetic analyses in the mouse have left open the issue of whether FGF4 and FGF8 have similar or different functions in limb development.

To resolve this issue, we employed a novel approach to assess whether one gene can functionally replace another in the mouse, rather than the conventional knock-in strategy (Hanks et al., 1995), in which the sequence of one gene is inserted into the genomic locus of a second, thereby inactivating the second gene and expressing the first in its place. Instead, we replaced the expression of one gene(Fgf8) with that of another (Fgf4) by using a single cre transgene to concomitantly inactivate and activate conditional Fgf8 loss- and Fgf4 gain-of-function alleles, respectively,in the same cells. The knock-in strategy has the advantage that the gene that has been inserted should faithfully recapitulate the expression of the gene it is meant to replace, although this is sometimes difficult to achieve if regulatory elements that control the expression of the gene being replaced are perturbed. Using our strategy, the expression of the gain-of-function target gene may ultimately be broader than the gene expression it replaces, because it continues in all descendants of the cells in which it was initially activated. That was indeed the case in our study, as Fgf4 expression,which is driven by the promoter in the transgene, persisted in the limb bud ventral ectoderm, whereas Fgf8 expression, although initially detected in that domain, normally becomes restricted to the AER. As discussed above, such discrepancies between the expression patterns of the two genes can make interpretation of the data difficult. However, our strategy has some distinct advantages. First, an individual conditional gain-of-function transgene can be readily used in combination with various conditional loss-of-function alleles to study the functional relationships between different members of a multigene family. Second, the conditional gain-of-function transgene can be activated without inactivating any other gene, allowing analysis of the consequences of excess or ectopic expression of the gain-of-function target gene, in our case Fgf4 in the limb bud. In future, Fgf4^GOF could be activated in a variety of developmental settings using appropriate cre transgenes to gain further insights into the function of FGF signaling.

Here we made use of the Fgf4^GOF transgene to show that all of the limb skeletal defects caused by loss of Fgf8 function can be rescued by increasing Fgf4 expression in the limb bud in a pattern that largely resembles that of Fgf8 in normal limb buds. This demonstrates the utility of our strategy, and more importantly, provides conclusive evidence that FGF4 can functionally replace FGF8 in limb skeletal patterning. Furthermore, our data emphasize the importance of mechanisms that control FGF gene expression for obtaining normal digit number and appropriate separation of the digits.

We thank R. Harland, J. Hébert, N. Itoh, V. Lefebvre, D. Ornitz and J. Wozney for providing plasmids from which the probes for in situ hybridization were made. We also thank Deborah Kurrasch for expert advice on quantitative RT-PCR assays. We are grateful to Guijuan Qiao, Christina Ahn,Prajakta Ghatpande, Lisa Matsubara and Ariana Nemati for technical assistance,and to our colleagues in the Martin laboratory for critical comments on the manuscript. Contributors to this work were supported by: NIH training grant HL07731 (P.L.) and a KO8 award from the NIH (CA60538; G.M.). This work was supported by a grant RO1 HD34380 (to G.R.M.).