The apical ectodermal ridge is a timer for generating distal limb progenitors

The apical ectodermal ridge (AER) is an essential signaling center governing vertebrate limb development(Capdevila and Izpisua Belmonte,2001; Martin,1998; Niswander,2003). The importance of the AER was demonstrated by classic experiments in chicken embryos showing that AER removal at progressively later stages of limb development causes a progressive loss of distal elements of the limb (Saunders, 1948; Summerbell, 1974). Although different models have been proposed to explain limb skeletal patterning along the proximal-distal (PD) axis, AER function in this process has remained largely unclear. For example, the Progress Zone (PZ) model postulates that the AER provides permissive signals to keep PZ cells labile until they exit the zone, at which time these cells are autonomously specified by ceasing to acquire `positional information'(Summerbell et al., 1973). By contrast, the Early Specification (ES) model proposes that PD elements are specified, rather than progressive, at the earliest stages of limb bud development and that the AER regulates subsequent expansion of progenitor pools by promoting cell proliferation and survival(Dudley et al., 2002). Furthermore, recent models suggest that PD elements are specified via dynamic interactions between the flank of the lateral plate mesoderm and the AER(Mercader et al., 2000; Tabin and Wolpert, 2007).

A transient structure, the AER undergoes a series of morphogenetic changes consisting of four stages - initiation, maturation, maintenance and regression- each characterized by distinctive changes of cell shape and gene expression(Altabef et al., 1997; Kimmel et al., 2000). Briefly,AER initiation in the mouse forelimb starts at approximately embryonic day (E)9, when the limb bud is first discernable and the gene expression in the ventral ectoderm that marks the future AER becomes apparent. During AER maturation, pre-AER cells in the ventral ectoderm migrate towards the distal tip and undergo a compaction process, whereby a distinctive narrow band of stratified columnar epithelium forms at ∼E10(Loomis et al., 1998). The mature AER is then maintained for an additional 2-3 days, while mesenchymal skeletal progenitors continue to proliferate and differentiate until a fully patterned limb emerges. The AER then regresses via programmed cell death and eventually flattens to a simple cuboidal epithelium(Guo et al., 2003).

At the molecular level, AER initiation involves interactions of several major signaling pathways. The prevailing model holds that Wnt2b-Wnt8c/β-catenin signaling in the lateral plate mesoderm is required for Fgf10 expression in the presumptive limb bud mesenchyme (LBM)(Kawakami et al., 2001), which in turn regulates Wnt3/β-catenin signaling in the overlying ectoderm to induce AER formation (Barrow et al.,2003; Kengaku et al.,1998). BMP signaling also plays a role in this process, as mice lacking Bmpr1a in the ectoderm fail to form the AER(Ahn et al., 2001). The molecular basis of the remaining stages of AER morphogenesis is less clear. Studies have shown, however, that engrailed 1 (En1) plays a role during migration and compaction of AER progenitor cells(Loomis et al., 1998), whereas Wnt/β-catenin signaling is required to maintain the AER after AER initiation and maturation (Barrow et al.,2003). By contrast, BMP signaling promotes destruction of the AER during the regression process (Pizette and Niswander, 1999), despite its early role in initiating the AER.

FGFR2 functions in both limb ectoderm and mesenchyme during limb development (Itoh and Ornitz,2004). Although mouse embryos completely lacking Fgfr2function fail to develop beyond implantation stages(Arman et al., 1998), those with partial loss of Fgfr2 function, including ones specifically lacking the 2b isoform (De Moerlooze et al., 2000; Revest et al.,2001), survive to later embryonic stages, but fail to develop limbs (Arman et al., 1999; Gorivodsky and Lonai, 2003; Xu et al., 1998). These results suggest that AER initiation requires ectodermal FGFR2 function to respond to mesenchymal FGF10 signaling, as mice lacking mesenchymal Fgf10 are also limbless (Min et al., 1998; Sekine et al.,1999). However, mesenchymal expression of FGFR2(Coumoul et al., 2005), as well as of FGFR1 (Li et al., 2005; Verheyden et al., 2005), is essential for skeletal progenitor cells to respond to AER-FGFs to ensure normal skeletal formation and patterning. In this study, we used a conditional approach based on the Cre/lox system to modify the AER. We found that AER maintenance requires FGFR2 function and is essential for distal limb development.

Mice carrying a conditional Fgfr2 null allele, Fgfr2^fl (Yu et al.,2003), the Msx2-cre transgene(Lewandoski et al., 2000) or a conditional gain-of-function (GOF) allele of β-catenin,Catnb^lox(ex3) (Harada et al., 1999) were kindly provided by Drs David Ornitz, Gail Martin and Makoto Mark Taketo, respectively. Mice carrying the BAT-Gal transgene, a reporter of Wnt/β-catenin signaling activities(Maretto et al., 2003), were purchased from the Jackson Laboratory (Stock #005317). All mice were maintained on a mixed genetic background and genotyped based on previously published reports.

Noon of the day when a vaginal plug was detected was considered ∼E0.5. Embryos were collected in cold PBS, fixed in 4% paraformaldehyde (PFA), and stored in 100% methanol at -20°C. 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. Each somite stage is ∼2 hours. E10.5 is equivalent to ∼35-sominte stage and E11.5 is equivalent to ∼45-somite stage. Standard protocols were used for RNA in situ hybridization and skeletal preparations as previously described(Lu et al., 2006). The area of Hoxa13 expression was measured using ImageJ. β-galactosidase(β-Gal) activity was conducted using standard protocol. Bright-field images were captured with a Leica DMR HC microscope using a 40×/0.75 Plan Apochromat air objective.

Embryos were embedded in agarose, and 40 μm sections were cut using a Leica vibratome. Samples were blocked in PBS containing 5% BSA and 0.5% Tween 20 for 1 hour before they were subjected to incubation with primary antibodies, anti-β-catenin mAb (BD-Biosciences), anti-CD44(BD-Biosciences) and anti-phospho-Histone H3 (pH3, Upstate Biotechnology). Goat-anti-rabbit secondary antibodies (AlexaFluor488, Invitrogen) were used to detect the primary antibodies.

Cell proliferation analyses were based on either detection of 5-bromodeoxyuridine (BrdU) incorporation or pH3 immunofluorescence, which marks nuclei in cells undergoing mitosis. For BrdU analysis, mice were injected with BrdU (5 mg/100 g body-weight) 1 hour prior to euthanasia. Embryos were harvested and fixed in 4% PFA. Transverse paraffin sections (5μm) were cut using microtome and five sections, 50 μm apart from the middle region of each limb bud were used for BrdU detection. Antigen was retrieved in 10 mM sodium citrate (pH 6.0) in a microwave oven and BrdU was detected using a kit from Roche (Cat#1296736). Alternatively, 40 μm vibratome sections were used for pH3 immunofluorescence. All samples were counterstained with To-Pro-3 (Invitrogen). The total number of nuclei and BrdU-positive nuclei or pH3-positive nuclei in a 3.1×10⁴μm² area (φ=200 μm) of the sub-AER limb mesenchyme were counted in each section. BrdU incorporation and pH3 immunofluorescence were quantified using ImageJ. Confocal microscopy was performed on a ZeissLSM510 confocal.

Assays for cell death via TdT-mediated dUTP nick-end-labeling (TUNEL)analysis on vibratome sections were performed according to manufacturer's protocol (Roche Cat#1684817), while those via LysoTracker(MolecularProbesL-7528) staining were conducted as previously reported(Lu et al., 2006). A total of over 180 embryos were examined between the 28- and 45-somite stages by TUNEL or Lysotracker staining (at least five mutant embryos were examined for each stage).

The Msx2-cre transgene expresses cre in the ventral ectoderm and the AER of early limb buds(Lu et al., 2006; Sun et al., 2000). Importantly, in the hindlimb Msx2-cre acts before limb bud induction,while in the forelimb cre is not expressed until ∼20 hours after AER initiation (Sun et al.,2002). We crossed male mice homozygous for the Msx2-cretransgene and heterozygous for the Fgfr2-null allele, Fgfr2^Δ (Yu et al., 2003), with female mice homozygous for the Fgfr2conditional allele, Fgfr2^fl(Yu et al., 2003). All Msx2-cre;Fgfr2^fl/Δ progeny(Fgfr2^AER-KO) were viable and displayed anomalies,including hair overgrowth (not shown), when compared with their morphologically normal `control' littermates(Msx2-cre;Fgfr2^fl/+).

We found that Fgfr2 removal via Msx2-cre before AER initiation eliminated the hindlimb of Fgfr2^AER-KO embryos at E18.5 (Fig. 1A,B; n=18). By contrast, Fgfr2 removal at the 26- to 28-somite(s) stage after AER initiation in the forelimb eliminated only distal bones(Fig. 1C-F). In these mutants,although the humerus, radius and ulna were present and of normal sizes, the autopod (hand-plate) was absent except for a few carpal bones(Fig. 1C-F, n=18). Thus, FGFR2 is required for later stages of mouse forelimb development.

The loss of distal elements in Fgfr2^AER-KO forelimbs resembles the limb truncations of chicken embryos in AER removal studies(Saunders, 1948; Summerbell, 1974). To determine whether the AER is maintained in mutant forelimb buds, we examined the histological and molecular consequences of FGFR2 removal during AER morphogenesis. AER histology was assessed by the immunofluorescence of CD44,an AER marker (Sherman et al.,1998), on vibratome sections of early limb buds at stages after cre activation. In control embryos, we noted a distinctive AER composed of stratified epithelium at the 29 s stage, which became progressively more compact along the dorsal-ventral (DV) axis at later stages(33-45s) (Fig. 1G,I,K,M). In mutant limb buds, the AER was also distinctive at the 29 s stage, although it was slightly thinner than normal (Fig. 1H). During later stages, the mutant AER became progressively thinner and did not compact along the DV axis(Fig. 1J,L). By the 45 s stage,the AER in mutant limb buds was lost, as shown by the lack of stratified epithelium in limb ectoderm and the absence of CD44 expression(Fig. 1N). The loss of the AER was also shown by the lack of expression of additional AER markers, including Bmp4, Dlx2, Sp8 and En1, in mutant limb buds at E11.5 (not shown). Thus, FGFR2 is required for AER maintenance during mouse limb development.

We then analyzed how expression of Fgf8, the primary AER-Fgf mediating AER function(Lewandoski et al., 2000; Moon and Capecchi, 2000), is affected by premature AER loss. Using RNA in situ hybridization on whole-mount embryos, we found that Fgf8 was expressed at a relatively normal level in mutant limb buds at the 28-somite stage(Fig. 1O-P′). However, by the 33-somite stage, Fgf8 expression was noticeably less in mutant limb buds than in control (Fig. 1Q-R′). Moreover, Fgf8 expression became discontinuous, and gaps lacking Fgf8 expression were often observed in mutant limb buds at later stages, including the 35 s (not shown) and 39 s stages (Fig. 1S-T′). By the 45 s stage, Fgf8 expression was essentially absent, except occasionally in a tiny patch of the residual AER(Fig. 1V). No Fgf8expression was observed afterwards (not shown). Thus, FGFR2 removal causes premature AER regression and progressive AER-FGF reduction.

To determine how the mesenchymal gene expression is affected, we examined Mkp3 expression, a downstream target of FGF signaling(Kawakami et al., 2003). We found that Mkp3 expression was not obviously different in Fgfr2^AER-KO and control forelimb buds at the 27 s (not shown) or 30 s stage (Fig. 2A,B). However, the Mkp3 expression domain was reduced by the 32 s stage in mutant limb buds (Fig. 2C,D) and became progressively smaller at later stages(Fig. 2E-H). By the 46 s stage,while the proximal domain of Mkp3 expression was not affected, the Mkp3-positive domain in the sub-AER mesenchyme was absent(Fig. 2G,H) because the distal limb bud failed to form in mutant limbs (see below). The Mkp3expression domain was also lacking in the distal mesenchyme of mutant limbs at E12.5, when extensive Mkp3 expression was evident in the autopod of control limb buds (not shown). Thus, FGF signaling activities continuously decrease in mutant limb buds.

AER-FGFs influence limb mesenchyme in part via a positive-feedback loop involving Shh and its downstream target Gremlin(Gre). We found that at the 35 s stage Shh expression in the mutant was not significantly different from that in control limb buds(Fig. 2I,J), but that at E11.5 its expression domain was smaller than control in mutant limb buds(Fig. 2K,L). Likewise, Gre expression was not greatly changed at E10.5 (not shown) but its expression domain was significantly reduced by E11.5(Fig. 2M-N′). Moreover,the thin strip of Gre-negative cells in the sub-AER mesenchyme or the mid-section along the DV axis was absent in mutant limb buds by E11.5(Fig. 2M-N′). Thus, Shh and Gre expression domains are reduced as a result of decreasing AER-FGF signaling activities.

In addition, we found that expression of Alx4, Meis2 and Hoxa11, markers for the anterior mesenchyme(Fig. 2O), the proximal mesenchyme (Fig. 2Q), and the slightly more distal mesenchyme (Fig. 2S), respectively, was not changed in mutant limb buds at E11.5(Fig. 2P,R,T). Likewise, the proximal domain of Hoxd11 expression, which marks the proximal mesenchyme, was not affected in mutant limb buds(Fig. 2U,V). By contrast, the distal domain of Hoxd11 expression, which marks the future autopod(Tarchini and Duboule, 2006)(see below), was completely absent in mutant limb buds at E11.5(Fig. 2U,V). As Hoxd11expression in these two domains results from two separate transcriptional events, first in the proximal then in the distal mesenchyme(Tarchini and Duboule, 2006),the lack of Hoxd11 expression in the distal domain could be due to either delayed transcription in the distal mesenchyme or an absence of the future autopod. No Hoxd11 expression in the distal domain was observed at E12.5 (not shown), thus arguing against the possibility of a transcriptional delay of Hoxd11. These data suggest that the molecular events causing loss of the distal autopod in mutant embryos have occurred by E11.5.

What events might cause loss of the distal autopod in Fgfr2^AER-KO embryos? There are two obvious possibilities,e.g. increased cell death and decreased cell proliferation in the distal mesenchyme, either or both could reduce the pool of autopod progenitors in mutant limb buds. First, we examined cell death in mutant limb buds at every somite stage between the 28 s and 45 s (E11.5) stages (at least five mutant embryos were examined for each stage) by TUNEL or Lysotracker staining. However, we did not observe abnormal cell death in the distal mesenchyme at these stages (Fig. 3A-D, not shown). Therefore, loss of the distal autopod in mutant forelimbs is unlikely to be due to death of autopod progenitors in the distal mesenchyme. Interestingly, we found that the number of dying cells in the AER and ventral ectoderm, where both Msx2-cre (Lu et al., 2006) and Fgfr2(Arman et al., 1999; Min et al., 1998; Revest et al., 2001) are expressed, was greater in mutant forelimb buds than in control limb buds at the 30 s stage (Fig. 3A,B; n=8). A similar pattern of cell death was observed in mutant limb buds up to the 36 s stage (Fig. 3C,D; n=6), when AER morphology was no longer distinct. Thus, FGFR2 acts as a survival factor to maintain the AER and, in its absence,the AER regresses prematurely owing to increased cell death.

Next, we examined cell proliferation in mutant limb buds by assaying the percentage of cells in an area [3.1×10⁴ μm²,approximately the size of the hypothetical PZ(Dudley et al., 2002; Summerbell et al., 1973)] of the sub-AER mesenchyme that incorporated BrdU after a 1-hour pulse. We found no significant differences between the percentages of BrdU-positive cells in the distal mesenchyme of mutant and control limb buds either at the 34 s or 44 s stage (Fig. 3E-H′,M). We also examined the mitotic index as indicated by pH3 immunofluorescence in the sub-AER mesenchyme but again found no significant differences between mutant and control limb buds at the 36 s, 39 s or E11.5 stages(Fig. 3I-L,N). Thus, based on two independent methods, our results indicate that global reduced cell proliferation is unlikely to account for loss of the distal autopod in mutant limbs.

A third possibility to explain loss of the distal autopod in Fgfr2^AER-KO embryos is that the early events of autopod progenitor formation may be compromised when the AER prematurely regresses in mutant limb buds. Consequently, a reduced number of autopod progenitors may be generated, which may be able to proliferate and expand similarly as those in control limb buds but fail to produce a sufficient number of skeletal progenitors required to form normal autopod elements. To test this possibility, we first reanalyzed several Hox genes, including Hoxd11, Hoxd13 and Hoxa13, that are known to be expressed in the autopod at late stages as potentially useful markers of early autopod progenitors. As shown previously, Hoxd11 was expressed in the distal autopod at E12.5 (Fig. 4C);however, this distal domain of Hoxd11 expression arose between E10.5(Fig. 4A) and the 40 s stage(E11.0; Fig. 4B). Thus, Hoxd11 expression at E10.5 probably marks progenitors of both the proximal limb and the autopod. Unlike Hoxd11, Hoxd13 expression was confined to the distal autopod at E12.5(Fig. 5O). However, at the 34-35 s stage, Hoxd13 was also expressed in the posterior-proximal limb bud (Fig. 4D). This proximal domain of Hoxd13 expression was gradually reduced as the limb bud grew out (Fig. 4E, Fig. 5M, not shown) and was lost after the ∼49 s stage (Fig. 4F, Fig. 5O). Thus, Hoxd13 expression at E10.5 is also likely to mark progenitors of both the proximal limb and the autopod.

Hoxa13 was expressed only in the distal autopod as Hoxd13at E12.5 (Fig. 5K). Unlike Hoxd13, however, Hoxa13 was not expressed in the proximal limb bud at the 35 s stage (Fig. 4G). Rather, at this stage, it was expressed in a thin strip of the posterior-distal mesenchyme, correlating with the area that is going to give rise to the future autopod based on fate-mapping studies of the chicken limb buds at the equivalent stages (Sato et al., 2007; Vargesson et al., 1997). Looking more closely over time, we found that Hoxa13 was first expressed in the posterior-distal mesenchyme in forelimb buds at the 31-32 s stages (Fig. 5A). As the limb bud grew out at later stages, the Hoxa13-positive domain expanded concomitantly and extended anteriorly until a well-patterned limb emerged at E12.5(Fig. 4G-I; Fig. 5C,E,G,I,K,U,U′). Importantly, we have never observed a loss of expression domain for Hoxa13, as we did for Hoxd13, during the course of early limb bud development, suggesting that Hoxa13 expression is unlikely to mark progenitors of the proximal mesenchyme. Thus, consistent with studies showing that Hoxa13 is initially expressed in a subpopulation of autopod progenitors (Nelson et al.,1996; Sato et al.,2007), our data suggest that Hoxa13 expression in the early limb bud marks autopod progenitors of the distal limb.

Next, we examined development of autopod progenitors by analyzing Hoxa13 expression in mutant limb buds. We found that the onset of Hoxa13 expression was delayed by two somite stages beginning at the 33-34 s stages instead of at the 31-32 s stages in Fgfr2^AER-KO limb buds(Fig. 5C,D; Table 1). At later stages, the Hoxa13 expression domain expanded at a relatively stable pace, as in the control. Thus, at the stages examined after its initiation, the area of the Hoxa13-expression domain in the mutant was ∼45% of that in the control until about the 45 s stage, when it reached a plateau and barely expanded afterwards (Fig. 5F,H,J,L,U,U′). Hoxd13 expression, marking slightly more distal autopod, confirmed that autopod expansion was indeed stalled by the 47 s and E12.5 stages (Fig. 5M-P).

Interestingly, the outline of the mutant limb bud, which was smooth and regular at the 45 s stage (Fig. 5H), became rugged and irregular by the 47 s stage(Fig. 5J). These changes of limb bud shape may reflect ongoing cellular events, e.g. chondrocyte condensation of the autopod in the underlying mesenchyme. To test this possibility, we examined Sox9 expression to directly visualize skeletal progenitors in early limb buds. We found that at the 40 s stage, when skeletal condensation had yet to start, the total number of skeletal progenitors was not grossly different between mutant and control limb buds(Fig. 5Q,R). By the 46 s stage,skeletal condensation had occurred, and although stylopod and zeugopod rudiments in mutant forelimb buds appeared normal(Fig. 5S,T), autopod rudiments were absent. Thus, consistent with previous studies suggesting that skeletal condensation occurs on a fixed schedule in a proximal to distal wave, our data indicate that the autopod progenitor pool in mutant limb buds is insufficient,owing to delayed generation and expansion, to form a normal autopod at the onset of autopod condensation.

To examine whether Wnt/β-catenin signaling is affected by Fgfr2 removal, we used the reporter allele of the BAT-Gal transgene to directly examine Wnt/β-catenin signaling(Maretto et al., 2003). We found that Wnt/β-catenin signaling was greatly reduced in the AER of Fgfr2^AER-KO forelimb buds at the 33 s stage(Fig. 6A, n=6; Fig. 6B, n=4),although ectopic activities was also observed in the distal-dorsal mesenchyme(Fig. 6B, arrow). In addition,expression of Lef1, which plays a role in limb development(Galceran et al., 1999), was reduced in mutant limb buds than in control at E10.5 (∼35 s stage)(Fig. 6C,D). Tcf1 and Wnt3 expression were not significantly changed (not shown). Thus,FGFR2 is required for normal Wnt/β-catenin signaling during AER maintenance.

Next, we asked whether gain of Wnt/β-catenin signaling in the AER may prevent premature AER loss and restore autopod bones in the mutant limb. We crossed male mice homozygous for the Msx2-cre transgene and heterozygous for the Fgfr2^Δ allele with female mice homozygous for the Fgfr2^fl allele and heterozygous for the Catnb^lox(ex3) allele, a conditional GOF allele ofβ -catenin (Harada et al.,1999), to generate Msx2-cre;Fgfr2^fl/Δ;Catnb^lox(ex3)embryos (referred to as Fgfr2^AER-KO;β-cat^GOFhereafter). In these embryos, Msx2-cre-mediated recombination inactivates Fgfr2 by converting Fgfr2^fl to a null allele and concomitantly activates β-cat^GOFexpression from the Catnb^lox(ex3) conditional allele. We found that β-catenin overexpression(Fig. 6E-H) caused ectopic Fgf8 expression in the ventral ectoderm at E11.75(Fig. 6I,K), when it was already absent in the mutant limb bud (Fig. 1N,V; Fig. 6J). In Fgfr2^AER-KO;β-cat^GOFlimb buds, Fgf8 expression was retained in the mutant AER and was ectopically expressed in the ventral ectoderm(Fig. 6L). Moreover, we found that β-catenin GOF prevented ectopic cell death in the ventral ectoderm(Fig. 6N, n=3) and restored normal expression of Hoxa13 in mutant limb buds(Fig. 6O,P; Table 1). Furthermore, we found that the distal limb elements, based on Sox9 expression marking skeletal progenitors (Fig. 6Q,R) and skeletal staining(Fig. 6T, n=5), were present in Fgfr2^AER-KO;β-cat^GOFembryos. Together, these data showed that β-catenin GOF prevents AER loss and restores the autopod in Fgfr2^AER-KO limb buds.

In this study, we ablated the AER by genetically removing FGFR2 function and demonstrated that AER maintenance is essential for distal limb development. We showed that FGFR2 promotes survival of AER cells and that it interacts with Wnt/β-catenin signaling during AER maintenance. Interestingly, we found that neither cell survival nor cell proliferation was affected in the distal mesenchyme of Fgfr2^AER-KO forelimb buds. This is in contrast to the current model suggesting that the AER regulates limb skeletal progenitors by promoting cell survival and proliferation in the mesenchyme. To uncover the role of the AER in autopod development, we validated Hoxa13 expression as a marker of autopod progenitors and described a dynamic morphogenetic process of autopod development. In mutant limb buds, generation of autopod progenitors was delayed, which in turn failed to reach a critical mass required to form a normal autopod. Thus, we have serendipitously discovered a novel mechanism whereby the AER regulates the number of autopod progenitors by determining the onset of their generation.

There are several lines of evidence suggesting that Hoxa13expression is an early marker of autopod progenitors. First, the initial expression domain of Hoxa13 resides in a thin strip of the posterior-distal mesenchyme at the 31-32 s stage, which is the area of the early limb bud shown by fate-mapping studies in the chick at comparable stages to give rise to the autopod (Sato et al.,2007; Vargesson et al.,1997). In addition, despite close examination, we never observed a loss of Hoxa13 expression domain during early limb bud development. These results suggest that, in contrast to Hoxd13, the initial Hoxa13 expression domain is unlikely to mark progenitors of the proximal limb. Indeed, studies have shown that this initial Hoxa13expression domain marks only a subpopulation of autopod progenitors. For example, using DiI labeling in chick embryos, Nelson and colleagues demonstrated that Hoxa13-positive cells in early limb buds ultimately give rise to only the posterior two-thirds of the autopod(Nelson et al., 1996). Likewise, recent studies suggested that this initial Hoxa13expression domain contributes to the distal ∼two-thirds of the future autopod (Sato et al., 2007). The most anterior and proximal autopod progenitors, which express Hoxa13 at E12.5, must acquire Hoxa13 expression de novo,thus highlighting a heterogeneous origin of autopod progenitors.

By contrast, Hoxd13, which is ultimately expressed in the emerging autopod similar to Hoxa13, is initially expressed in a subpopulation of zeugopod progenitors. The domain of Hoxd13 expression in the autopod, like that of Hoxd11, is the result of a second-wave transcriptional event controlled by an enhancer element shared by the HoxD family of genes (Tarchini and Duboule, 2006). Thus, the final pattern of Hoxd13expression in the autopod is a combined result of downregulation of its expression in the proximal domain and upregulation of its expression in the distal mesenchyme (Vargesson et al.,1997).

Based on data showing that AER removal in chicken limb buds at the stage 20/21 causes loss of Hoxa13 expression, it was suggested that the AER directly regulates Hoxa13 expression(Hashimoto et al., 1999; Vargesson et al., 2001). In light of recent studies (Dudley et al.,2002), an alternative explanation is that loss of Hoxa13expression may be caused by death of autopod progenitors because of AER removal. Indeed, our data show that Hoxa13 expression is initiated,albeit with a delay, and its domain is able to expand in the presence of dwindling AER-FGF signaling activities. In addition, Hoxa13expression is maintained for at least 24 hours after the AER is lost and sub-AER mesenchymal FGF signaling is absent by the 45 s stage. Together, these results suggest that the AER indirectly regulates Hoxa13 expression by directly controlling development of autopod progenitors (see below).

Our data support a model in which the AER influences autopod development by regulating `generation' (see below) of autopod progenitors(Fig. 7). Previous studies suggest that skeletal condensation occurs on a fixed schedule in a proximal to distal wave (Dudley et al.,2002; Lewandoski et al.,2000; Sun et al.,2002; Wolpert et al.,1979) (reviewed by Mariani and Martin, 2003). Thus, whether a normal autopod forms depends on the number of progenitors at the time of autopod condensation. In Fgfr2^AER-KO limbs, the onset of generation of autopod progenitor is delayed by two somite stages because of premature AER regression and reduction of AER-FGF signaling activities in the LBM(Fig. 7B). This delay of progenitor generation causes an insufficient number of autopod progenitors to be available at autopod condensation to form normal autopod bones. Remarkably,autopod progenitors in mutant limbs are initially able to expand at a rate comparable with that in the control when AER-FGF signaling activities progressively diminish in the mesenchyme. These data thus suggest that the primary function of the AER is to promote generation of autopod progenitors during vertebrate distal limb development.

There are several major distinctions between our model and the existing models, in particular the PZ model(Summerbell et al., 1973), the ES model (Dudley et al., 2002)and other recent models (Mercader et al.,2000; Tabin and Wolpert,2007), regarding limb skeletal patterning along the PD axis. Unlike these models, which are concerned with and differ at when limb elements are specified, our model deals with events after the distal skeletal elements are specified. Indeed, Hoxa13 is unlikely to specify the autopod directly, even though it is required for autopod development(Fromental-Ramain et al.,1996). At present, specification markers of the PD elements are still elusive. It is thus unclear precisely when and where the autopod is specified, how many progenitor cells are present in the initial pool, to what extent cell proliferation is involved and whether these earliest autopod progenitors share similar proliferative properties to cells in other parts of the mesenchyme. For the sake of simplicity and clarity in discussing our data,we have used the generic term `generation' to distinguish these early events,which include autopod specification, of autopod progenitor development proceeding Hoxa13 expression with the later `expansion' event, which our data suggest is initially less sensitive to reduced FGF signaling activities.

To explain loss of distal elements in AER removal experiments, the PZ model proposes that, in the absence of the AER, PZ cells loss their lability, stop acquiring distal `positional information' and fail to be specified as distal elements (Summerbell et al.,1973). By contrast, our data show that autopod progenitors in Fgfr2^AER-KO forelimb buds express markers, including Hoxd13 and Hoxa13, of events that occur after autopod specification. Our model suggests that the loss of the distal limb is due to an insufficiency of skeletal progenitors at the onset of autopod condensation,rather than to failure of specification according to the PZ model. Moreover,by regulating generation of AER progenitors, the AER may play more than a permissive role as the PZ model suggests. Finally, whereas the ES model proposes that the AER regulates skeletal progenitor pools by promoting cell proliferation and survival (Dudley et al.,2002), our model shows that, at least for the autopod, this is accomplished by controlling the number of progenitors that are initially generated.

It is noteworthy that our model provides new mechanistic insights on several puzzling skeletal phenotypes reported previously. For example,although mice lacking AER-Fgf4/8 fail to form the hindlimb autopod,neither cell proliferation nor cell death is affected in the distal mesenchyme(Sun et al., 2002). In light of our results, it is possible that generation of autopod progenitors may be delayed, or prevented in these mutants. Furthermore, it remains possible that the AER, in addition to promoting survival of stylopod and zeugopod progenitors (Barrow et al.,2003; Dudley et al.,2002; Sun et al.,2002), may also regulate their generation. Consistent with this notion, mouse limb buds that lack Fgf4/8, which fail to form normal stylopod and zeugopod, are smaller than normal at the earliest stages of limb development without changes of cell death or cell proliferation in the mesenchyme (Sun et al.,2002).

The lack of cell death in the distal mesenchyme of Fgfr2^AER-KO forelimbs is consistent with results from previous studies. As mentioned, cell death is increased only in the proximal but not in the distal mesenchyme, where autopod progenitors reside, of embryos lacking AER-Fgf4/8 (Sun et al.,2002). In addition, whereas AER removal before stage 24 in the chick [which eliminates proximal limb elements up to the wrist(Niswander et al., 1993)]causes cell death in the distal mesenchyme, AER removal after stage 25 (which results in loss of the distal autopod) does not(Dudley et al., 2002). It is interesting to note that, although abnormal cell death was observed in the limb bud mesenchyme of certain mutant mice(Revest et al., 2001; Sun et al., 2002), it was not observed in the proximal mesenchyme of Fgfr2^AER-KOforelimbs, as reported here and by a recent study(Yu and Ornitz, 2008). These differences in the cell death patterns, in addition to the final skeletal phenotypes, are due to the fact that AER-FGF signaling is reduced earlier and/or more greatly in limb buds where AER fails to initiate(Revest et al., 2001) or in limb buds where AER-Fgf4/8 are directly ablated(Sun et al., 2002) than in Fgfr2^AER-KO forelimb buds, where AER-Fgf is,although progressively less, continuously expressed until the 45 s stage. Together, these data suggest that AER function is required for the survival of stylopod and zeugopod progenitors, but not of autopod progenitors.

Interestingly, our analyses detected no significant differences of cell proliferation in the distal mesenchyme between mutant and control limb buds. These data thus join several previous studies, which also show that cell proliferation is not affected in mutant limb buds that fail to form skeletal elements, including the autopod (Barrow et al., 2003; Revest et al.,2001; Sun et al.,2002; Verheyden et al.,2005). Furthermore, it was observed over three decades ago(Janners and Searls, 1971)that cell proliferation is unchanged after AER removal in chicken limb buds. Taken together, these results support our conclusion that reduced cell proliferation in the distal mesenchyme is unlikely to account for loss of the distal autopod. Finally, although the lack of abnormal cell death and cell proliferation in the distal mesenchyme led us to discovering AER function in generation of distal limb progenitors, we do not exclude the possibility that,earlier, the AER may also regulate cell proliferation of the distal progenitors.

Previous studies have shown that the FGF10-FGF8 loop, presumably via ectodermal FGFR2, operates to initiate the AER during vertebrate limb development (Crossley et al.,1996; Ohuchi et al.,1997). It was further suggested that FGF10 may be the `AER maintenance factor' in the posterior limb mesenchyme to maintain AER morphology (Ohuchi et al.,1997). In this study, by conditionally removing FGFR2 function, we demonstrate that loss of Fgfr2 function causes failure of AER maintenance and, as a consequence, loss of the forelimb autopod. Our results therefore support the model that FGFR2 signaling maintains the AER after AER initiation.

Msx2-cre-mediated loss of Fgfr2 function in the AER and ventral ectoderm increased cell death in these tissues, where both Fgfr2 and the cre transgene are expressed. These results suggest that FGFR2 acts as a survival factor to maintain the AER. Interestingly, cell death is also increased in limb ectoderm of mice lacking Fgfr2b function, in which the AER fails to initiate(Revest et al., 2001). Thus,these results suggest that FGFR2 promotes cell survival in both AER initiation and maintenance during limb development.

Several lines of evidence indicate that FGFR2 interacts with Wnt/β-catenin signaling during AER maintenance. Wnt/β-catenin signaling activities, as measured by β-Gal staining using the BAT-Gal transgene, are greatly reduced in the AER of Fgfr2^AER-KOlimb buds. Expression of Lef1, which together with Tcf1 is required for AER initiation, is also downregulated in mutant limb buds. Furthermore, loss of β-catenin (Barrow et al., 2003) and FGFR2 function via Msx2-cre cause similar patterns of cell death increase in the forelimb ectoderm, suggesting that they interact to promote cell survival during AER maintenance. Finally, a gain of Wnt/β-catenin signaling rescues AER loss and this in turn restores the autopod bones missing in Fgfr2^AER-KO limbs. FGF signaling interacts with Wnt/β-catenin signaling during early limb bud development (Barrow et al.,2003; Galceran et al.,1999). Together with our data, these studies suggest that interactions of FGF and Wnt/β-catenin signaling pathways are required at multiple stages of vertebrate limb development.

We thank J. C. Belmonte, D. Duboule, R. Harland, J. Hébert, J. Helms, J. Innis, N. Itoh, A. Joyner, V. Lefebvre, A. McMahon, D. Ornitz, J. Rubenstein and J. Wozney for providing plasmids from which the probes for in situ hybridization were made. We thank Helen Capili for excellent technical assistance, and Gary Freeman, Sylvain Provot and Mimi Zeiger for critical reading of the manuscript. Supported by an NIH NRSA HL07731 (P.L.) and NIH grants AR046238 and CA057621 (Z.W.).