Progressive interdigital cell death: regulation by the antagonistic interaction between fibroblast growth factor 8 and retinoic acid

Interdigital cell death (ICD) is a morphogenetic event associated with digit separation in chick and mouse. The current model supposes the formation of an interdigital tissue that subsequently degenerates and, as a result,digits individualize. Selective growth of digits could also contribute to digit separation. Under this model, ICD would restrict the growth of the interdigital tissue. Several molecules have been identified that participate in the regulation of ICD; however, very little is known about their mechanism of action and how they interact and contribute to trigger ICD.

Members of the bone morphogenetic protein (Bmp) subfamily of Tgfβgrowth factors have been shown to be potent cell death inducers(Buckland et al., 1998; Ganan et al., 1996; Macias et al., 1997; Macias et al., 1999; Yokouchi et al., 1996). Bmp2, Bmp4, Bmp5 and Bmp7 are expressed in the autopod of the developing limb, each with a different but overlapping expression pattern(Francis et al., 1994; Geetha-Loganathan et al.,2006; Lyons et al.,1990; Pajni-Underwood et al.,2007; Robert,2007; Salas-Vidal et al.,2001; Zuzarte-Luis et al.,2004). In the mouse, near the time of ICD initiation, Bmp2 and Bmp7 are preferentially expressed in the interdigital areas, whereas Bmp4 is mostly found in the distal ectoderm (Pajni-Underwood et al.,2007; Salas-Vidal et al.,2001). The ectopic application of Bmp2, Bmp4, Bmp5 or Bmp7 in the interdigital mesenchyme promotes apoptosis in the chick limb and accelerates interdigit regression (Macias et al.,1997; Rodriguez-Leon et al.,1999; Zuzarte-Luis et al.,2004). In agreement with a function for Bmp in ICD, extensive expression of a dominant-negative type I Bmp receptor early in chick limb development reduces ICD and produces soft tissue syndactyly(Zou and Niswander, 1996). Bmp antagonists, such as noggin or gremlin, produce the same effect when ectopically applied to the interdigital mesenchyme(Merino et al., 1999). In the mouse limb, noggin expression in the apical ectoderm also reduces cell death and causes syndactyly (Guha et al.,2002; Wang et al.,2004). More importantly, mice lacking the Bmp receptor Bmpr1a or both Bmp2 and Bmp4 specifically in the apical ectodermal ridge (AER) show syndactyly(Maatouk et al., 2009; Pajni-Underwood et al., 2007). Therefore, Bmp members could regulate cell death directly in the mesenchyme(e.g. Bmp7) or indirectly through their activity in the apical ectodermal ridge (e.g. Bmp4).

Retinoic acid (RA) is a broad-spectrum cell death inducer. In the mouse limb, an excess of RA causes distal limb truncations, possibly owing to cell death (Alles and Sulik, 1989; Kochhar et al., 1993) and,when applied just before the onset of ICD, accelerates interdigit regression in mouse and chick limbs (Lussier et al.,1993; Rodriguez-Leon et al.,1999). All RA receptor (Rar) genes are expressed in the autopod region of the limb, but only Rarβ expression is restricted to the interdigital area (Dollé et al.,1989). Mouse embryos lacking Rarβ are viable and show no abnormalities in the limb. However, supporting the role of RA in ICD, Rarβ^-/-; Rarγ^-/- double mutant embryos show interdigital webbing and an absence of interdigital Bmp7 expression(Dupé et al., 1999). In the chick limb, it has also been proposed that RA is a physiological regulator of ICD as an all-trans-RA receptor antagonist decreases ICD(Rodriguez-Leon et al.,1999).

Several members of the Fgf family play a role in limb development. In particular, limb growth is promoted by the positive-feedback loop between different Fgf members, with Fgf4, Fgf8 and Fgf10 being the most important(Niswander, 2003). Fgf members have survival activity in developing limbs as indicated by their ability to reduce ICD when applied ectopically(Macias et al., 1996). This Fgf survival activity could be responsible for the growth-promoting activity of the AER. Fgf8, the gene of which is expressed throughout the AER during limb growth, is likely to be the most important factor performing this function. That endogenous Fgf8 promotes a survival activity in the distal limb mesenchyme is suggested from the syndactyly phenotype and reduction in ICD associated with its persistent expression in several mutant and transgenic mice (Guha et al., 2002; Lallemand et al., 2005; Wang et al., 2004), as well as its restricted expression to the digit tip ectoderm at the time ICD begins(Salas-Vidal et al.,2001).

Complex interactions are likely to occur among the factors participating in the control of ICD. Presently, little information exists regarding the mechanism by which different factors control ICD. Despite the data described above, the expression patterns of Rarβ, Bmp2, Bmp4,Bmp7, and the homeobox genes Msx1 and Msx2, in the mouse do not correlate in time and/or space with the ICD pattern(Salas-Vidal et al., 2001). Furthermore, it is known that many of the genes that might participate in ICD have other functions early in limb development. Therefore, evidence supporting a role of these genes in ICD, in which the effect of the experimental treatment is evaluated after other processes have occurred, should be taken with caution. For example, in the mouse, gene-specific deletions causing syndactyly occur long before ICD begins, leaving the possibility that the phenotype is the result of a sum of alterations occurring before ICD(Dupé et al., 1999; Lallemand et al., 2005; Pajni-Underwood et al.,2007).

We hypothesized that Fgf8 downregulation in the distal interdigital ectoderm triggers ICD. To better define the direct contribution of RA, Bmp7 and Fgf8 in ICD activation, we studied their regulatory function in cell death in a short time window (of no longer than 8 hours). We found that Fgf8 and RA are crucial regulators of ICD, the former preventing, and the latter promoting, cell death. Bmp activity in the mesenchyme, by contrast,appears to be required for ICD in the chick but not for ICD in the mouse. We discuss the participation of these signals in the context of a progressive ICD model of digit individualization.

Mouse strain CD1 was used in this study. Pregnant females were sacrificed by cervical dislocation from 12.5 to 13.5 days postcoitus (dpc; day 0.5 of coitus was the day on which a vaginal plug was found). Embryos were selected according to the limb stage required. Fore- or hind-limbs were dissected without preference, but always considering the limb developmental stage. Chick embryos [stage 28-30 (Hamburger and Hamilton, 1992)] were taken from fertilized eggs incubated for 6 to 7 days. The eggs were windowed at the desired stage and the right or left limb was exposed.

The limb culture protocol used here was as previously described(Salas-Vidal et al., 1998). Limbs were staged according to Wanek's system(Wanek et al., 1989). The stage indicated in the figures is that at which the experiment was performed. Solutions of growth factors [1 mg/ml in phosphate buffered saline (PBS)] were used to soak heparin beads (Fgf8, Peprotech; Bmp7, R&D Systems) or Affi-gel beads (noggin, R&D Systems) for 1 hour at room temperature. In the experiments presented, beads with growth factor were always implanted in the D3-D4 interdigital area of mouse limbs. The control beads were incubated in PBS and implanted in the D2-D3 interdigital area. Retinoic acid (10 μM,Sigma), retinoic acid receptor antagonist AGN193109 (100 μM, Allergan),Fgfr antagonist SU5402 (50 μM, Calbiochem), Mapk inhibitor UO126 (50 μM,Calbiochem), phosphoinositide 3-kinase (Pi3k) inhibitor LY294002 (25-50 μM,Calbiochem) and dorsomorphin (15 μM, Calbiochem) were added directly to the culture medium for 6 or 8 hours. For each treatment, at least three independent experiments were performed in triplicate; reproducibility in each condition tested was generally higher than 90%.

As a common practice, all limbs used in this study were stained with Acridine Orange (AO) in order to ensure the limb stage and/or the effectiveness of the treatments carried out. Lysotracker (LT; Molecular Probes) staining was used to detect dying cells and specific gene expression in the same sample. The TUNEL Kit (Roche) assay was used to detect apoptotic cells in 7-10 μm sections of paraffin-embedded tissues. The complete series was analyzed by confocal fluorescence microscopy. To detect cell death in developing chick limbs, they were rinsed in a Neutral Red solution (1 mg/ml in PBS) for 10 minutes at room temperature, then washed in PBS for 5 minutes, and finally viewed by fluorescence microscopy.

Wholemount in situ hybridization was performed as previously described(Salas-Vidal et al., 2001). Riboprobes were synthesized using the DIG RNA Labeling Kit (Roche) according to the manufacturer's recommendations. Probes for the following genes were used: Fgf8, Bmp7, Mkp3 (also known as Dusp6), Msx2,Raldh2 (Aldh1a2-Mouse Genome Informatics), Cyp26b1, Bcl2,Bclx and Bax. Immunohistochemistry in wholemount was performed as previously described (Corson et al.,2003). Primary antibodies used were phospho-Erk1/2 (pErk1/2),phospho-Smad1/5/8 (pSmad), Smad1, cleaved caspase 3 (all from Cell Signaling Technology) and Fgf8 (Sigma). After incubation with a biotinylated anti-rabbit secondary antibody (Vector Laboratories), the signal was developed with the Vectastain HRP-Avidin-Biotin Complex (ABC) Reagent (Vector Laboratories).

Small drops (20 μl) of supernatant from cultures of cells producing Ad-EGFP or Ad-noggin-EGFP adenovirus were placed on parafilm within a humidified chamber. Stage 8 (S8) mouse limbs were dissected, placed in the small drops with the adenovirus and incubated for 2 hours in a 5%CO₂ incubator. After adenovirus infection, mouse limbs were cultured on 0.45 μm durapore filters (Millipore) floating on DMEM(Invitrogen) with 10% fetal bovine serum and without antibiotic supplementation for up to 48 hours. Three independent experiments performed in triplicate with the same infection efficiency were analyzed.

Mouse limbs before ICD initiation (S8-S9) were immersed in a 10 μmol/l BrdU/DMEM solution (Roche) and incubated at 37°C in a 5% CO₂incubator for 1 hour. Right limbs were fixed immediately in 4%paraformaldehyde overnight, whereas left limbs were first washed in 3 ml dissection medium (L15; Microlab) and then cultivated for 12 hours as described above, before fixing. After TUNEL labeling, BrdU incorporation was detected by first incubating samples in 4 M HCl for 20 minutes (for antigen retrieval), followed by neutralization with 0.1 M boric acid, three washes in PBS, and a 2-hour incubation with anti-BrdU Alexa-Fluor 594 (Molecular Probes)in PBS. After one further wash in PBS, the complete series was analyzed by confocal fluorescence microscopy.

We previously estimated the growth of digital and interdigital regions around the time ICD occurs by comparison of digitalized limb contours(Salas-Vidal et al., 2001);however, the actual origin of the dying interdigital tissue has not been determined. We inserted a bead into the distal mesenchyme of both chick and mouse limbs just before the start of ICD. As shown in Fig. 1A, a significant amount of mouse interdigital tissue developed above the bead, indicating that interdigits grow within a 12-hour time window. In the chick, a significant proportion of the interdigital tissue also originated by growth after ICD began (Fig. 1B).

These previous results suggest that all dying interdigital cells derive from the most distal proliferating cells. In order to test this, distal interdigital cells were labeled with BrdU and their fate was followed during the formation of the dying interdigital region. One hour after BrdU addition,labeled cells were found all along a distal margin defining a 100 μm region(Fig. 1C). Twelve hours later,most BrdU-labeled cells were dying (92.5±9%; n=4) and were located in a more proximal region (Fig. 1C). Therefore, distal cells initiate the death process as they progressively acquire a lateral and proximal position.

Fgf8 is expressed in the distal ectoderm around the time that ICD and digit individualization initiate(Salas-Vidal et al., 2001);thus, distal proliferating cells are likely to be under the influence of this growth factor. Consistent with previous work showing that Fgf2 and Fgf4 have survival activity on interdigital mesenchymal cells(Macias et al., 1996), we observed that Fgf8-soaked beads implanted in the interdigital regions of S8-S9 mouse limbs, stages just prior to, or coincident with, the initiation of ICD,decreased ICD as determined by AO staining or TUNEL(Fig. 2A). Interestingly,application of the Fgf8 bead after ICD initiation (S9-S10) only slightly affected cell death (data not shown).

If Fgf8 is relevant for the survival of distal mesenchymal cells,downregulation of its gene could be the direct cause of ICD. An analysis of Fgf8 expression and cell death in the same limb along several developmental stages around the initiation of ICD supports this possibility(Fig. 3). In S8 limbs, Fgf8 expression was detected in the entire distal ectoderm without evidence of ICD. In limbs at a later stage than S8 but before S9 (S8+), Fgf8 expression decreased in the interdigital ectoderm, and cell death was initiated in the underlying mesenchyme. Finally, in S9 limbs, Fgf8 expression was restricted to the digital apical ectoderm and cell death extended laterally and proximally into the interdigital mesenchyme regions. Fgf8 and activated caspase 3 (an indicator of dying cells) showed the same pattern (data not shown).

In order to determine whether Fgf signaling contributes to the survival of distal mesenchyme, we studied the consequences of inhibiting Fgf receptor(Fgfr) tyrosine kinase activity in S8 limbs. Fgfr inhibition promoted cell death in the distal mesenchyme (Fig. 2B). In S9-S10 limbs, Fgfr inhibition increased cell death at the tips of digits, a location coincident with the mesenchyme underneath the Fgf8-expressing ectoderm (data not shown). Taking these data together, Fgf8 appears to be a survival factor in the distal mesenchyme of limbs around the time that digits individualize, and its disappearance could be the direct cause of the onset of ICD.

In the limb bud, Fgf8 can signal using the classical Ras-Mapk pathway(Delfini et al., 2005; Eblaghie et al., 2003; Echevarria et al., 2005). Accordingly, phosphorylation of the Mapk member Erk1/2 is closely associated with the AER Fgf8 activity during limb bud growth(Corson et al., 2003). The expression of Mkp3 (encoding a Mapk phosphatase) is also associated with this activity, suggesting that a negative-feedback loop regulates Fgf8/Mapk signaling (Eblaghie et al.,2003; Echevarria et al.,2005; Li et al.,2007). Nevertheless, it has been shown that the Pi3k pathway mediates the Fgf8 survival effect and activation of Mkp3 expression during growth of chick limb buds (Kawakami et al., 2003). In concordance with the lack of participation of Pi3k in the Fgf8 survival activity, the Pi3k inhibitor LY294002 did not affect the Mkp3 expression pattern in the distal mesenchyme of S8 or S9 limbs (see Fig. S1A,B in the supplementary material). Furthermore, the survival effect of exogenous Fgf8 was clearly not altered by the Pi3k inhibitor, and under these circumstances Mkp3 expression was markedly upregulated compared with that of untreated limbs (see Fig. S1B in the supplementary material).

In contrast to the previous results, in the presence of U0126 [a Mapk (Mek)inhibitor], cell death in S8 limbs was expanded proximally(Fig. 4A) and the protective effect of Fgf8 on the interdigital cells of S9 limbs was suppressed(Fig. 4B). Consistent with an effect reducing survival activity, Mek inhibition increased the number of TUNEL-positive cells (Fig. 4B). In agreement with a marked reduction in Fgf8 activity, the Mek inhibitor completely blocked Mkp3 expression induced by both endogenous Fgf8 produced in the distal ectoderm of S8 limbs(Fig. 4A) and exogenous Fgf8 implanted in the interdigits of S9 limbs(Fig. 4B). As expected from the above results, Erk1/2 was found phosphorylated in the mesenchyme near to the Fgf8 source, either in a naturally occurring state (i.e. distal ectoderm; Fig. 4A) or exogenously added(Fig. 4B), and the Mek inhibitor blocked such phosphorylation(Fig. 4A,B). These data suggest that, at least around the time that ICD begins, the survival effect of Fgf8 on mesenchymal cells is mediated by the Mapk pathway.

Although evidence indicates that RA is a positive regulator of ICD(Dupé et al., 1999; Lussier et al., 1993; Rodriguez-Leon et al., 1999),it is unclear whether only interdigital cells are responsive to this molecule,or whether this is a broader property of limb mesenchymal cells. Mouse limbs cultured with RA showed signs of cell death after 3 hours of treatment (data not shown), although induction of cell death was more evident between 6 and 8 hours of treatment. RA induced cell death in the entire distal mesenchyme of S8 limbs (Fig. 5A; also see Fig. S2 in the supplementary material). In S9 mouse limbs, at which stage ICD had already begun, RA increased the ongoing cell death and also induced cell death at the tip of the digits (Fig. 5B; also see Fig. S2 in the supplementary material). Endogenous RA showed the same death-inducing activity, as ICD decreased when S9 limbs were cultured with a Rar antagonist (AGN193109; Fig. 5C; also see Fig. S2 in the supplementary material). Therefore, all distal mesenchymal cells are responsive to the death-inducing activity of RA, suggesting that, in the presence of RA, a survival activity is needed to allow growth. Furthermore,these data support the view that, as in the chick, RA is a direct physiological regulator of ICD in the mouse limb.

The pro-apoptotic Bcl2 family members Bax and Bak appear to play a role in the activation of ICD (Lindsten et al.,2000). Therefore, it is possible that RA controls the expression of their corresponding genes. A bioinformatical approach detected, within 1.5 kb upstream of the transcription initiation sites, three putative RA-responsive elements in Bax and one in Bak (see Table S1 in the supplementary material). Before ICD initiates, Bax is expressed along the whole distal mesenchyme(Fig. 5D). In agreement with a role of RA in the regulation of Bax expression in the developing limb, Bax mRNA was upregulated by RA and downregulated by a Rar antagonist (Fig. 5D,E).

Taking the previous data together, it is possible that Fgf8 provides the survival activity that counteracts the death-inducing activity of RA, and thus this antagonism could modulate ICD activation. Indeed, as shown in Fig. 6A, Fgf8 reduced RA-induced cell death. The protection level provided by Fgf8 was RA dose dependent (data not shown), suggesting that the balance between these two signals defines the fate of distal mesenchymal cells.

Fgf8 might prevent RA-induced cell death by decreasing the RA concentration. In vivo, the concentration of RA is regulated by a balance between its synthesis and inactivation. Raldh2 is responsible for most of the RA-synthesizing activity during early mouse embryogenesis(Niederreither et al., 1999). By contrast, members of the Cyp26 family specifically metabolize RA to an inactive form (Fujii et al.,1997). An Fgf8-soaked bead in the interdigital region caused downregulation of Raldh2 expression, which was evident after 8 hours of treatment (Fig. 6B). By contrast, the Fgf8-soaked bead upregulated the expression of Cyp26b1,which encodes the major enzyme located in the mesenchyme(Fig. 6C). Although RA alone upregulated Cyp26b1 expression, it was further increased by Fgf8 (see Fig. S2 in the supplementary material). These data indicate that Fgf8 protects from RA-induced cell death by reducing RA production and increasing RA inactivation.

We also evaluated whether RA induces cell death by regulating the expression of Fgf8 and Fgfr1. Fgfr1 is expressed primarily in the mesenchyme of the developing mouse limb and appears responsible for most Fgf8 activity (Li et al.,2005; Verheyden et al.,2005). Under conditions in which RA increased cell death, Fgf8 expression did not change(Fig. 6D), but the expression of Fgfr1 significantly decreased(Fig. 6E). Accordingly, RA exogenously added to cultures of S8 and S9 limbs reduced Erk1/2 phosphorylation caused by endogenous Fgf8(Fig. 6F). Therefore, RA could provoke cell death, at least partially, by regulating Fgf8 intracellular signaling in distal mesenchymal cells.

As RA upregulated Bax expression, we determined whether Fgf8 could counteract the death-inducing activity of RA by controlling the expression or activity of anti-apoptotic Bcl2 members. RA downregulated Bcl2expression, but the expression pattern did not suggest a function in ICD (see Fig. S3 in the supplementary material). By contrast, the anti-apoptotic BclxL (Bcl2l1 - Mouse Genome Informatics) was also expressed in the limb (see Fig. S3A in the supplementary material); however, we could not detect regulation by either RA or Fgf8 (data not shown).

At the time that ICD begins, the only Bmp expressed in the interdigital regions of mouse limbs is Bmp7, whereas Bmp4 expression is predominant in the distal ectoderm(Pajni-Underwood et al., 2007; Salas-Vidal et al., 2001). We found that Bmp7 expression in the mouse limb was regulated by RA and Fgf8, and Bmp7 induced cell death in the distal mesenchyme and negatively regulated Fgf8 expression in the ectoderm (see Fig. S4 in the supplementary material). Surprisingly, a noggin-soaked bead in the interdigital region did not decrease cell death after 12 hours of culture(starting at S8-S9) or in the presence of RA(Fig. 7A), although noggin did efficiently inhibit cell death induced by exogenous Bmp7 (data not shown). Furthermore, the same noggin-soaked beads in the chick limb decreased ICD after 8 hours of treatment, with an increasing effect up to 24 hours. Noggin inhibited cell death in both the distal(Fig. 8A) and proximal(Fig. 8B) mesenchyme. Interestingly, Fgf8 expression persisted in the AER when noggin prevented distal ICD (Fig. 8A),but not when it prevented proximal ICD(Fig. 8B). Therefore, Bmp7 (or another Bmp with the ability to bind noggin) in the mesenchyme participates in ICD activation in the chick but not in the mouse limb.

To confirm that mesenchymal Bmp members do not participate in ICD in the mouse limb, we inhibited Smad activity with dorsomorphin, a recently identified compound that prevents Smad phosphorylation(Yu et al., 2008). Dorsomorphin markedly reduced Smad phosphorylation in the interdigital areas of S9 limbs after 8 hours of treatment(Fig. 7B). Reduction was also observed in distal digital areas, although it was less apparent. This might be due to the high levels of Smad1 in digits (R.H.-M. and L.C., unpublished) or the less efficient inhibitory effect of dorsomorphin on Smad1. Despite the reduction in phosphorylated Smad (pSmad) in the presence of dorsomorphin, ICD was not affected (Fig. 7B).

Our previous data contrast with the results observed in transgenic mice that overexpress noggin, in which interdigital regression is reduced, thus suggesting a participation of Bmp members(Guha et al., 2002; Wang et al., 2004). However,in those studies, noggin was overexpressed in the ectoderm; therefore, we hypothesized that modulation of ICD by Bmp proteins is limited to the ectoderm where it regulates Fgf8 expression. To test this idea, we infected the ectoderm of S8 limbs with an adenovirus that carries noggin coding sequences and EGFP. Twenty-four hours after infection, we observed that Fgf8 expression in the interdigital ectoderm of infected areas was persistent and both cell death and indentation at the interdigits was reduced in comparison with control limbs infected with an adenovirus carrying only EGFP (Fig. 9A,C). Interdigital webbing generated by noggin-expressing adenovirus increased at 48 hours after infection (data not shown). Similar results were observed when dorsomorphin was applied to S8 limbs (Fig. 9B,C). The marked reduction in cell death observed in these experiments (Fig. 9B) contrasts with the above-described data showing that dorsomorphin, at stages later than S8, was incapable of reducing cell death, and suggest that Bmp acts upstream of Fgf8 downregulation.

The present work proposes a mechanism by which ICD is activated. Our findings suggest that downregulation of Fgf8 expression is the trigger of ICD and that RA is the essential cell death inducer. Interestingly,Bmp members in the mouse interdigital mesenchyme are not essential for ICD activation, and only an ectodermal Bmp, probably Bmp2 and/or Bmp4(Maatouk et al., 2009),appears to be required. Our data suggest that an antagonistic interaction between Fgf8 and RA is the major mechanism that controls ICD in the mouse and also contributes to ICD regulation in the chick(Fig. 10A). This mechanism,together with the origin of the interdigital tissue fated to die, support a new model of how ICD contributes to limb morphogenesis(Fig. 10B).

The requirement of Bmp members in ICD has been determined by three strategies: (1) overexpression of dominant-negative forms of Bmpr(Pizette and Niswander, 1999);(2) adding Bmp antagonists such as noggin and gremlin(Guha et al., 2002; Merino et al., 1999; Wang et al., 2004); and (3)mutations in Bmp or Bmpr genes(Bandyopadhyay et al., 2006; Pajni-Underwood et al., 2007). Nonetheless, in most of these studies, the spatial and/or temporal requirement of Bmp activity for ICD was not determined. The most common view is that Bmp members act directly on mesenchymal cells of interdigits; however, recent studies in which Bmpr1a or Bmp2/Bmp4 were specifically inactivated in the AER suggest that AER Bmp activity can indirectly regulate ICD (Bandyopadhyay et al., 2006; Pajni-Underwood et al., 2007). Our data support this latter view in which Bmp activity causes the Fgf8 downregulation observed in the ectoderm overlying the interdigital tissue. However, it is important to consider that Fgf8 could also have, in addition to a role in survival, a mitogenic activity on mesenchymal cells; thus, the persistence of Fgf8 might have a double effect that promotes the interdigital tissue growth.

In the mouse limb, we did not find evidence supporting a direct role of Bmp in ICD. Application of noggin in the interdigital mesenchyme did not prevent ICD, even after 24 hours of culture (data not shown). Furthermore, blocking Smad phosphorylation with dorsomorphin in S9 limbs produced the same result. Bmp also does not mediate ectopic cell death induced by RA, as the expression pattern of Bmp7, the main candidate to play a role in ICD in the mouse, did not correlate with the observed pattern of cell death (see Fig. S4A in the supplementary material), and noggin did not affect the increase in cell death. Furthermore, expression of Msx2, which has been considered the end effector for Bmp-mediated apoptosis(Marazzi et al., 1997; Rodriguez-Leon et al., 1999),is downregulated at the beginning of ICD and is strongly induced by ectopic Fgf8 in the interdigits of both mouse and chick limbs in association with the survival activity (see Fig. S5 in the supplementary material). In the chick,our data support the Bmp role in ICD suggested in previous reports but also suggest the possibility that Bmp plays different roles in distal and proximal ICD. Bmp might induce distal ICD by downregulating Fgf8 expression in the distal ectoderm, whereas induction of proximal ICD by Bmp does not appear to affect Fgf8 expression. In the mouse, ICD activation appears to be similar to the distal ICD proposed for the chick. Indeed, it has been observed that ICD in the chick begins from distal and proximal regions, whereas in the mouse, proximal ICD is scarcely detected(Fernandez-Teran et al.,2006). It is possible that the large interdigital region in the chick requires different mechanisms to activate ICD in order to completely separate the digits.

In the limb, the distal region is the most sensitive to the RA death effects (Alles and Sulik,1989). Our data show that, at around the time interdigital regression occurs, all distal mesenchyme is competent to die by RA. This competence is associated with the expression of several genes encoding RA receptors, such as Rarg, Rarg and Rxra(Dollé et al., 1989). It is interesting to note that the domain of Rarb expression in the limb that was the first indication of a possible role of RA in ICD is not within the competence region observed at early stages(Dollé et al., 1989; Salas-Vidal et al., 2001). However, the Rarb; Rarg double mutant has the most penetrant syndactyly phenotype (Dupé et al.,1999). Accordingly, the antagonistic inhibition of Rarα or Rarβ does not affect ICD in the time window studied here (R.H.-M. and L.C., unpublished), suggesting that Rarγ might be the essential active receptor promoting cell death. At this time, with the information available,an indirect interaction between Rarb and Rarg cannot be discounted.

In agreement with previous reports(Dupé et al., 1999; Rodriguez-Leon et al., 1999),our data indicate that RA is essential for ICD. However, these reports propose that RA activates cell death indirectly by upregulating Bmp7 (mouse and chick) and Bmp4 (chick) expression. Our data do not support this possibility in the mouse, as RA activated cell death in regions in which Bmp7 expression was not induced and noggin did not reduce RA-induced cell death. We propose that RA activates cell death directly without the intervention of Bmp members. In the medial epithelial seam of the developing palate, which also expresses Bmp4 and Bmp7, the cell-death-inducing activity of RA is also Bmp independent(Cuervo et al., 2002). In agreement with this proposal, we show that RA positively regulates the expression of the pro-apototic gene Bax. Distal mesenchyme Bax (and likely also Bak) expression is at least in part controlled by RA, supporting the idea that a survival factor is required to counteract the persistent RA-induced death activity. Our data suggest that Fgf8 is this survival factor, the effects of which are also regulated by RA.

Although previous reports indicate that Fgf members are survival factors for mesenchymal cells of the developing limb, they do not indicate how this survival activity contributes to the initiation of ICD. We confirmed that Fgf8 has survival activity on mouse interdigital mesenchyme cells, but more importantly, we showed that Fgf8 expression is downregulated at the time ICD begins. In agreement with a role of endogenous Fgf8 in interdigital cell survival, several conditions that reduce ICD and cause syndactyly are associated with preserving Fgf8 expression in the ectoderm overlying the interdigital areas. For instance, reduction in Bmp activity by both noggin protein (Guha et al., 2002; Wang et al., 2004) and Bmpr1a- or Bmp2/Bmp4-specific knockout in ectoderm(Pajni-Underwood et al., 2007)prevents the normal downregulation of Fgf8 expression in the interdigital distal ectoderm. Similarly, the Msx1^-/-; Msx2^-/- double mutant retains Fgf8 expression in interdigital distal ectoderm (Lallemand et al., 2005). Supporting the idea that Fgf8 expression is the direct cause of the lack of ICD and syndactyly in previous reports,constitutive Fgf4 expression in the distal ectoderm was found to cause syndactyly (Lu and Martin,2006). Our data, and a study in the chick(Ganan et al., 1998) showing that Bmp7-induced cell death is associated with downregulation of Fgf8 expression, are also in concordance with the crucial role of Fgf8 for survival of distal interdigital mesenchyme.

Fgf8 survival activity appears to be mediated by two mechanisms(Fig. 10A). Our data suggest that Fgf8 indirectly prevents cell death by decreasing RA synthesis and increasing its inactivation, which results in a global reduction in RA levels. During limb development, RA diffuses from proximal to distal regions. At around the time ICD begins, RA is synthesized in the proximal region of the interdigits [defined by Raldh2 expression(Niederreither et al., 1999)],and diffuses to the distal area where Fgf8 presumably reduces RA levels by increasing the amount of Cyp26b1. In vivo, induction of Cyp26b1expression in the presence of Fgf8 is apparent when ICD begins and Cyp26b1 expression restricts to the digits. Antagonistic regulation between Fgf8 and RA appears to be a common mechanism in the control of developmental processes (Abu-Abed et al.,2001; Diez del Corral et al.,2003; Mercader et al.,2000).

Our data suggest that Fgf8 also promotes its survival activity by a direct mechanism through the Mapk pathway. This conclusion comes from the evident requirement of Mek activity for the Fgf survival activity and the direct association of this activity with Erk1/2 phosphorylation and Mkp3expression. Given that RA activates Bax expression, the end target of this signaling pathway might be anti-apoptotic Bcl2 members. Bcl2 and BclxL do not appear to be good candidates, but more members and post-translational modifications need to be studied in order to define whether this is a plausible antagonistic mechanism controlling ICD. On the other hand, RA has no perceptible effect on Fgf8 expression but does promote a reduction in Fgfr1 expression. In addition, RA negatively regulates Erk1/2 phosphorylation by a mechanism apparently independent of upregulation of Mkp3 expression. Therefore, RA might also be able to antagonize Fgf8 survival activity and, in this way, promote cell death.

The areas with phosphorylated Erk and Mkp3 expression define the domain of action of Fgf8 survival activity. This domain does not cover the whole interdigital area, suggesting that proximal ICD is triggered by a different mechanism. Our model proposes that most interdigital cells derive from the distal mesenchyme, the survival of which depends on Fgf8. Thus, all cells that die in the interdigital region must be at some point within the domain of Fgf8 action (estimated as no more than 100 μm of the underlying mesenchyme based on the zone of proliferating cells). Interestingly, our data with mouse limbs show that the interdigital tissue, defined as the region with abundant cell death before digit individualization, is generated within a 12 hour time window after the initial detection of dying cells. Proximal dying cells in the interdigital mesenchyme included those cells that were underlying the distal ectoderm at the time ICD began. As is the case with digits,interdigits also grow but distal projection is restricted by cell death. According to this view, cell death is initiated in the distal region, and, as the limb grows, dying cells acquire a proximal position(Fig. 10B). Our model proposes that ICD in the mouse, rather than being a process of massive cell death, is mostly a progressive mechanism. In the chick, distal ICD might use the same mechanism described above for the mouse limb, but the most proximal ICD appears to follow a different mechanism that is dependent on interdigital Bmp activity.