growth arrest specific gene 1 acts as a region-specific mediator of the Fgf10/Fgf8 regulatory loop in the limb

In vertebrates, limb buds are formed from the lateral plate mesoderm by an early induction from the adjacent somites and intermediate mesoderm(Harrison, 1921) (reviewed byStocum and Fallon, 1982). The limb mesoderm then induces the overlying ectoderm to thicken and form the apical ectodermal ridge (AER) (Zwilling,1961). The AER in turn provides signals that drive limb outgrowth(Saunders, 1948) along the proximodistal (PD) axis. The signals responsible for these activities have been assigned molecularly (reviewed byMartin, 1998). Fgf8expressed in the somitic and intermediate mesoderm is presumed to be the limb bud initiation signal (Crossley et al.,1996; Vogel et al.,1996); it consolidates the expression of Fgf10 in the lateral plate mesoderm to the prospective limb domains(Ohuchi et al., 1997).Fgf10 in the limb mesenchyme then induces the AER-specificFgf8 expression (Ohuchi et al.,1997; Min et al.,1998; Sekine et al.,1999). Fgf8 in the AER in turn maintains Fgf10expression in the mesoderm (Mahmood et al., 1995; Vogel et al.,1996; Ohuchi et al.,1997; Moon and Capecchi,2000). Fgf4, Fgf9 and Fgf17 are expressed in the posterior AER and likely join forces with Fgf8 to act upon the mesenchyme (Lewandoski et al.,2000; Sun et al.,2002). The Fgf8/Fgf10 regulatory loop thus fulfills the documented mutual interactions between the mesenchyme and the AER to stimulate limb outgrowth and generate the appropriate amount of cellular mass for the formation of all the limb skeletal elements.

There are many modulators of AER function. Shh, expressed in the zone of polarizing activity (ZPA), regulates Fgf4 in the posterior AER by acting through the intermediate component Gremlin, a BMP inhibitor(Zuniga et al., 1999). Inhibition of BMP signaling also can increase the thickness of the AER(Pizette and Niswander, 1999). These studies indicate that BMP negatively modulates AER function and that SHH counteracts this by modulating the BMP signal. However, at an earlier stage,BMP signaling is necessary for AER formation(Ahn et al., 2001;Pizette et al., 2001).Fgf10 appears to act through the Wnt/β-catenin pathway in the ectoderm to activate Fgf8 in the AER(Kawakami et al., 2001). Consistently, the null mutant mouse embryos of Tcf1 andLef1, the Wnt downstream mediators, do not expressFgf8 in the AER (Galceran et al.,1999). FGFs from the AER serve to maintain Fgf10expression in the mesenchyme, but whether they are the sole and direct input or require additional modulators and intermediate components is not yet known.

Sufficient mesenchymal mass is required for the formation of the skeletal elements of the appropriate number and size. The aforementioned growth regulators are important in the generation of a defined amount of mesenchymal mass. From this will be produced the mesenchymal condensations representing the limb skeletal elements, including the proximal segment (stylopod;femur/humerus), the medial segment (zeugopod) with two elements(tibia-fibula/radius-ulna), and the distal segment (autopod) with, in the mouse, the five elements (digits)(Hinchliffe and Griffiths,1983; Shubin and Alberch,1986). These skeletal elements arise by endochondral cartilage formation, starting with a group of mesenchymal cells that condense and differentiate into chondrocytes(Erlebacher et al., 1995). The growth and size of each element are then coordinately regulated by successive transitions in differentiation that are locally controlled, for example,through the activities of BMP and of IHH, which activates a negative feedback relay system by regulating PTHrP (a negative growth regulator)(Ganan et al., 1996;Vortkamp et al., 1996;Zou et al., 1997;St-Jacques et al., 1999).

We have previously proposed that GAS1, a GPI-anchored membrane glycoprotein(Stebel et al., 2000), acts as an inhibitor of SHH via direct physical interaction(Lee et al., 2001a). However,Gas1 mutant mice do not display phenotypes related to those of SHH overexpression (Lee et al.,2001b; Liu et al.,2001). Instead, we report here that Gas1 mutant limbs have defects caused by reduced proliferation in the AER and mesenchyme, and develop with small autopods, missing phalanges and anterior digit syndactyly. We provide several lines of experimental evidence supporting the model thatGas1 is a necessary mesenchymal factor that positively regulatesFgf10 in a regional- and temporal-specific manner to maintain theFgf10/Fgf8 regulatory loop.

Gas1 mutant mice were described previously(Lee et al., 2001b). After backcrossing to CD1 and to 129sv females for five successive generations, the mutant limb defects remained in both backcrosses. For this study, animals used were from the CD1 backcrosses because the mutants were viable in this background. PCR was used to determine the genotype(Lee et al., 2001b).

Skin and internal viscera were removed and the bodies fixed overnight by St. Marie's fluid, followed by standard Alcian Blue and Alizarin Red staining procedures (Bancroft and Cook,1994). Whole-mount fetal Alcian Blue staining was performed according to the same protocol omitting the Alizarin Red.

ISH on paraffin sections (8 μm) with ³⁵S-UTP-labeled probes was performed as described previously (Fan and Tesseir-Lavigne, 1994). Photographs of the detected transcripts were taken as dark-field images with a red filter. Phase images were taken with a blue filter and overlayed with the dark-field images. Whole-mount ISH using DIG-labeled probes was performed following standard procedures (Wilkinson,1992). Probes used were: Gas1(Lee and Fan, 2001),Shh (gift from Dr McMahon), Fgf8 (gift from Dr Martin),Fgf4, Gli3 (gift from Dr Hui), Fgf9, Fgf17 (gift from Dr Ornitz), Bmp2, Bmp4, Bmp7 (gift from Dr Hogan), Gremlin(gift from Dr Zeller), Hoxb8, Hoxd13 (gift from Dr Duboule) andAlx4 (gift from Dr Wisdom).

Mice were injected with 10 mg/ml BrdU (Sigma) at 0.01 ml/g body weight, 1 hour before sacrifice. BrdU-positive cells were detected by using a BrdU staining kit (Zymed). Cell death assays were performed using the In Situ Cell Death Detection Kit, Fluorescein (Boehringer Mannheim) according to the manufacture's protocol.

Embryonic limbs were cultured as described previously(Zuniga et al., 1999). E9.75-E11.5 embryos from heterozygous mating were dissected in L-15 medium. The heads and tails were removed for genotyping and the trunks including forelimbs were used for injection. Recombinant FGF10 and FGF8 proteins(Research Diagnostic Inc.) resuspended in PBS were delivered by glass needles. 1-25 μg/ml of FGF10 and 1-100 μg/ml of FGF8 were used for injection and a pulse of 9.2 nl was injected into the right forelimb mesenchyme underneath the AER. The left forelimb was not injected and served as an internal control. Mock injection was performed using PBS. The injected limb buds were cultured in BGJb medium with 0.2 mg/ml ascorbic acid (Gibco/BRL) at 37°C/5%CO₂. After overnight culturing, the trunk fragments were fixed and subjected to whole-mount ISH.

The mouse Gas1 transcript is detected in the early limb bud(Lee and Fan, 2001;Lee et al., 2001c). Here we extend the documentation by whole-mount and section in situ hybridization(ISH). At E9.25 and E9.5, Gas1 is expressed in the entire lateral plate mesoderm and in the forelimb bud(Fig. 1A,B). It is preferentially expressed in the anterior part of the limb mesenchyme. This asymmetric pattern continues to E10.5 (Fig. 1C) and is also observed in the hindlimb (not shown). Sections through various planes of the limb bud indicate that Gas1 is expressed in approximately the anterior two-thirds of the mesenchyme. Adjacent sections of an E10.5 forelimb hybridized to Gas1 and Fgf8,an AER marker, confirm that Gas1 is expressed up to the distal tip of the mesenchyme abutting the anterior-to-central AER but not in the AER(Fig. 1H). At E11.5,Gas1 expression retains its anterior preference and apparently corresponds to the mesenchymal condensations of the radius and ulna(Fig. 1D). At E12.5,Gas1 is expressed in the condensing mesenchyme and the interdigital mesenchyme (Fig. 1E,I)(Lee and Fan, 2001;Lee et al., 2001c). One day later, it is expressed at the outer edges of the condensed digital rays, the prospective joints, and weakly in the interdigits(Fig. 1F,J)(Lee and Fan, 2001;Lee et al., 2001c). From E14.5 to E15.5, its transcripts are restricted to the perichondrium, the articular surface and the joints (Fig. 1G,K). Its mesenchymal expression pattern suggests its role in limb growth, patterning and/or skeletogenesis.

To define the role of Gas1 in limb development, we analyzed the limb phenotype of Gas1 null mutant (Gas1^-/-) mice(Lee et al., 2001b). NewbornGas1^-/- mice have smaller fore- and hind-limb paws(Fig. 2A-B'). Skeletal preparation using the dyes Alcian Blue (for cartilage) and Alizarin Red (for calcified bone) revealed that this is due to a size reduction of all phalanges, metacarpals and metatarsals(Fig. 2C-D'). The calcified regions were of normal size but the chondrogenic regions were reduced in proportion. Digits I-III were disproportionately reduced in size. In addition,the second phalange of digit II was greatly reduced(Fig. 2C') or absent(Fig. 2D'), and there was a high rate of soft tissue fusion between digits II and III(Fig. 2A',B'). Syndactyly between digits II and III was also observed by histology(Fig. 2D') and by X-ray imaging(not shown). The frequency of these defects is summarized inFig. 2E. The front and hind paws of the adult Gas1^-/- mice were almost of normal size(not shown), indicating a compensatory growth postnatally. Carpals, tarsals and long bones (radius, ulna, tibia, fibula) were either slightly shorter or not affected. Macroscopic and histological analyses revealed no apparent defects in the patterns of muscles and tendons (not shown). Thus,Gas1 contributes to proper formation of the autopodial skeletal elements, in particular the phalangeal elements and the anterior digits.

We next determined the ontogeny of the mutant phenotype by histology(Fig. 3). As early as E11.5,the mutant limb bud width across the AP axis is noticeably reduced(Fig. 4B'). The prechondrogenic condensations of digits II-IV in the forelimb are normally visible by E12.5(Fig. 3A). In the mutant,condensations of digits III and IV are smaller and digit II is less evident(Fig. 3A'); the autopod is also narrower along the AP axis and shorter along the PD axis. At E13.5, the metacarpals and first phalange are individualized by the onset of joint formation (Fig. 3B). In the mutant, separation of the phalanges is ill-defined and the phalanges of digit II and metacarpal I are not apparent (Fig. 3B'). From E13.75 to E14.5, the third and second phalanges of all digits become clearly defined (Fig. 3C,D); whereas in the mutant, the second phalanges of digits II and III are not individualized (Fig. 3C',D').

At E16.5, ossification of metacarpals III and IV is delayed in the mutant(Fig. 3E,E'). At E17.5,ossification of the first phalanges of digits III and IV(Fig. 3F,F') is clearly delayed. While ossification of digit III second phalange is delayed by about 24 hours, digit II second phalange does not appear to ever ossify(Fig. 3F' and not shown). Intriguingly, as soon as the ossification assumes, the bone segments appear normal in size at the expense of the chondrogenic domains(Fig. 2C',Fig. 2D',Fig. 3F'). The mutant hindlimb phenotype is similar (Fig. 2E). These manifestations suggest that Gas1^-/- limbs have reduced mesenchymal mass as early as E11.5, which at least in part leads to the small and delayed condensations. Delayed or absent joint formation may affect the size of the chondrogenic region, but Gas1 is not essential for chondrogenic differentiation.

To detect alterations in cell proliferation, we assayed bromodeoxyuridine(BrdU) incorporation in vivo. At E10.5, the rate of BrdU incorporation in the limb mesenchyme adjacent to the AER (cells within 150 μm of the AER were counted and compared) was not altered in the mutant(Fig. 4A,A'). Note that the mutant AER is present but thinner than normal. AER morphology was confirmed by scanning electron microscopy (not shown). Importantly, the mutant AER has fewer BrdU⁺ cells (Fig. 4A',D), indicating a non cell-autonomous effect of Gas1on the AER. The most marked difference in the mesenchymal proliferation rate is observed at E11.5 (Fig. 4B,B'D). The reduction is observed preferentially in the distal mesenchyme at the anterior-to-central portion of the mutant limb: mild anteriorly and severe in the central region(Fig. 4D). This cellular mass reduction prefigures the delay in digit formation. When the digit ray is visible at E12.5, we found no marked difference in the rate of proliferation between digit III/IV and digit IV/V of mutant and control limbs. However,between digit II/III, there was a significant reduction (∼10%) in proliferation rate in the mutant (Fig. 4C,C',D). This regional-specific defect is surprising given the more general Gas1 expression in all interdigits at this stage. These findings suggest that the main proliferation defect occurs around E11.5 and the perduring smallness of the embryonic limb is due to this early deficiency of precursor population.

PCD is found in the following areas of the developing chick limb: anterior and posterior necrotic zones, the opaque patch, the interdigital mesenchymes and the joints (Hinchliffe,1982; Hurle et al.,1996). Gas1 expression overlaps with the opaque patch at E11.5, the interdigits at E12.5-13.5 (albeit weakly), and the joints(E13.5-E15.5) (see also Lee and Fan,2001; Lee et al.,2001c). Furthermore, overexpression of Gas1 can trigger PCD in cultured limb mesenchymal cells(Lee et al., 2001c). To assess whether Gas1 normally plays a role in PCD, we performed TUNEL-fluorescence labeling. At E11.5, fewer apoptotic cells were detected in the central mesenchyme area (opaque patch) in the mutant compared to control forelimb (Fig. 5A,A'), even though the limb is already smaller. In the control E13.5 forelimbs, cell death was observed in the interdigital zones and the prospective joints of the digits (Fig. 5B). Interdigital cell death in Gas1^-/- forelimbs was relatively normal posterior to digit III but greatly reduced anterior to digit III(Fig. 5B'), correlating with the anterior soft tissue syndactyly. Joint PCD was delayed in digits II and III in the mutant, consistent with the delay in phalange separation seen histologically. Since the mutant limbs appear delayed in development, we also examined PCD at E14.0. There was still little PCD between digits II and III,but PCD appeared relatively normal between other digits (compareFig. 5B,C,C'). By contrast, PCD in the mutant joints at this time appeared at a higher rate than those of the wild type at E13.5 and E14.0 (of digits II-IV inFig. 5C').

Reduced PCD between digits II and III suggests that Gas1 normally facilitates PCD and supports the claim by Lee et al.(Lee et al., 2001c), but in other interdigits, PCD appeared to be relatively normal. However, more apoptotic cells were detected in the mutant joints, suggesting thatGas1 is anti-apoptotic. In other affected regions of theGas1 mutant, such as the eyes and the cerebellum(Lee et al., 2001b;Liu et al., 2001), the PCD rate is not altered. Either Gas1 regulates PCD in a cell-context-dependent manner or these region-specific alterations of PCD are a secondary consequence of deregulated growth and heterochrony of the mutant limb.

Owing to the fact that GAS1 can physically interact with SHH and IHH(Lee et al., 2001a), we examined whether there are patterning abnormalities related to deregulated SHH signaling in the Gas1 mutant, using a battery of functional marker genes in the SHH pathway. However, we did not observe expression pattern changes. Shh expression in the ZPA(Riddle et al., 1993;Echelard et al., 1993) was activated and maintained correctly at E10 and E10.5(Fig. 6A' and not shown).gremlin, a downstream target of Shh(Zuniga et al., 1999), was expressed in the normal posterior domain albeit at apparently reduced levels(Fig. 6B'). The expression ofBmp2 and Bmp4 in the AER and the mesenchyme (reviewed byHogan, 1996) was also apparently normal in positions and levels(Fig. 6C',D'). These results suggest that the thinner AER is not due to misregulation of Bmps orgremlin. Expression of Ptc1 (Ptch)(Fig. 6E')(Marigo et al., 1996) andGli1 (not shown) (Hui et al.,1994) also appeared normal in the posterior domain. Alx4(Qu et al., 1997) andGli3 (Hui et al.,1994) expression was confined to the anterior domain in the mutant as in the control (not shown). These analyses were extended to E11.5 and E12.5 and no obvious alterations in these expression patterns were found. Lastly,Hoxd13 (Dolle et al.,1993) (Fig. 6F')and Hoxb8 (Charite et al.,1994) (not shown) expression was activated at a normal distal position in the Gas1^-/- limbs at E10.5 (not shown), E11.5 and E12.5 (not shown). The smaller domains of expression appear to be proportional to the smaller size of the limb bud.

Since in the mutant autopod chondrogenesis is delayed, we examined the expression patterns of Ihh, Bmp2, Bmp4, Bmp7, BmpRIA, BmpRIB andPTHrP (reviewed by Hogan,1996) in the developing cartilage. Their expression was delayed corresponding to the delayed progression of the limb (by ∼12 hours). Once initiated, these genes were expressed in normal patterns with proportionally smaller domains (data not shown). As Gas1^-/- displays neither expanded nor reduced expression domains of Shh andIhh downstream reporters and the Gas1^-/- limb phenotype is unrelated to those of Shh^-/-(Chiang et al., 2001) andIhh^-/- (St-Jacques et al., 1999), Gas1 does not appear to modulate the activity of the hedgehog pathway in the limb.

Because the Gas1 mutant AER is thinner(Fig. 4A') and compromised AER function is a potential cause of reduced mesenchymal mass (reviewed byMartin, 1998), we reasoned that expression of the AER-specific Fgfs may be affected in theGas1 mutant. Fgf4, Fgf9 and Fgf17 expression in the AER at E10.5 (Fig. 7A-C) and E11.5 (not shown) was normal in the mutants when compared to the controls (not shown). In addition, Fgf8 expression at E9.5 was also normally initiated (Fig. 7D,D'). However, at E10.0 and E10.5, AER-specific Fgf8 expression was lost in the mutant (Fig. 7E',F'). At E11.5, variable small patches of Fgf8 expression were regained in the mutant AER (Fig. 7G'). ThisFgf8 reappearance was restricted: when observed, it was most frequent in the posterior region, rarely in the anterior region, and never in the central AER. Loss of the FGF8 input from the AER may be the main cause of theGas1 mutant limb defects.

Fgf10 is necessary and sufficient to initiate Fgf8expression in the AER (Ohuchi et al.,1997; Min et al.,1998; Sekine et al.,1999). Whether it is continuously required to maintainFgf8 in the AER has not been established. It is possible thatGas1 maintains Fgf10, which in turn maintains Fgf8in the AER. We therefore examined whether Fgf10 expression is altered in the mutant. A small region of cells at the most distal tip mesenchyme immediately underneath the AER showed a clear absence of Fgf10expression at E9.5 both in the control and mutant forelimb(Fig. 8A,A'). At E10.0 and E10.5 (Fig. 8B,B',D,D'),Fgf10 expression extended to the extreme tip of the control limb mesenchyme, whereas mutant cells located at the distal tip mesenchyme lackedFgf10 expression. This reduction of Fgf10 expression could also be observed by whole-mount ISH (Fig. 8C',E'), but only in very few mutant limbs (∼8% of the mutant limbs) — in these cases, the Fgf10 loss appeared to be more extensive than that seen in sections. We reasoned that most mutants had a small affected domain (consistently detected by section ISH) which was not easily discerned by whole-mount ISH. Note that the anterior Fgf10expression domain was also slightly down regulated in the mutant. At E11.5,there was a moderate recovery of Fgf10 expression in the mutant distal mesenchyme (Fig. 8F'),temporally corresponding to the reappearance of small patches of Fgf8expression in the AER. These results support a model in which Gas1 is required to activate Fgf10 expression in the distal tip mesenchyme and Fgf10 in this distal tip mesenchyme is crucial for maintainingFgf8 expression in the AER.

In the above model, GAS1 deficiency should be overcome by supplementing FGF10 at the tip region. To test this, we applied FGF10 protein to the distal tip region of the Gas1 mutant limb and examined restoration ofFgf8 expression in the AER. The trunk segments containing the forelimbs of embryos between E9.75 and E11.5 were cultured using an in vitro system (Zuniga et al., 1999). At E9.75-E10.0 (Fgf8 is already lost in the mutant AER), FGF10 protein ranging from 9.2 to 230 pg was delivered into the anterior-central mesenchymal tip region (where Gas1 is normally expressed) underneath the AER, by microinjection. Only the right limb was injected, hence the left limb served as an internal control. The injected embryo trunks were cultured for 16 hours before harvesting for assessment of Fgf8 expression(diagram in Fig. 9A). Injection of PBS into control (Fig. 9B)and mutant (Fig. 9E) right limbs did not alter their Fgf8 expression when compared to the uninjected left side. Injection of 9.2 pg FGF10 protein (but not lower amounts) into the mutant right forelimb rescued Fgf8 expression in the AER (Fig. 9F), while the uninjected side showed no Fgf8 expression. At this amount of FGF10,only a weak and small domain of Fgf8 expression was observed in the central AER of the mutant limbs. At 230 pg, FGF10 caused the entire length of the mutant AER to express high levels of Fgf8 similar to the control limb injected with the same amount of FGF10(Fig. 9D,G). There was also a FGF10 dosage-dependent increase in AER height in the injected mutant limbs. Notably, high levels of FGF10 injected into control limbs caused an increase in Fgf8 expression as well as AER height compared to the uninjected side (Fig. 9D). At E11.5, after the Fgf8 expression was lost for more than a day in the mutant, FGF10 injection still rescued its expression(Fig. 9I), indicating thatFgf8 expression in the AER requires continuous input of FGF10. Injection of FGF10 to the proximal region (∼200 μm from the AER) did not rescue Fgf8 expression (not shown). Finally, injection of FGF8 protein (up to 1.8 ng) into the E9.75-E10.5 mutant limb did not rescue its own expression in the AER (Fig. 9Jand not shown), suggesting that FGF8 cannot restore Fgf10 expression at the distal tip. Thus, supplementation with FGF10, but not FGF8, at the distal tip mesenchyme can overcome the requirement of Gas1.

We report here the limb defects of the Gas1 null mutant mouse. Deregulation of Fgf10 in the mesenchyme and loss of Fgf8 in the AER correlate with the proliferation defect and developmental delay of theGas1^-/- limb. Our data support a model in whichGas1 acts as a novel regulator to maintain optimal levels ofFgf10 in a previously unappreciated and specialized mesenchymal region to maintain Fgf8 in the AER. These results provide several novel insights into limb development.

Gas1^-/- limbs display reduced proliferation preferentially in the anterior-to-central limb bud around E11.5. Gas1is normally expressed in the anterior-to-central region between E9.5-E11.5. Superficially, Gas1 appears to be a cell autonomous positive regulator of proliferation. Paradoxically, Gas1 overexpression is known to inhibit the cell cycle in cultured fibroblast(Del Sal et al., 1992). However, reduced proliferation in the Gas1^-/- limb does not completely correlate with the time and pattern of Gas1expression. Reduced proliferation only became measurable after theFgf8 expression in the AER was lost, indicating that the proliferation defect is more likely the consequence of compromised AER function (reviewed by Johnson and Tabin,1997). The preferential reduction in the anterior-to-central domain may reflect the fact that Fgf4, 9 and 17 are all expressed in their normal posterior AER domain, whereas Fgf8 is lost throughout the AER and thus the anterior-to-central domain does not continue to receive the FGF signal. However, we cannot exclude the possibility thatGas1 also directly contributes to anterior limb mesenchyme proliferation.

The mutant AER also has a decreased rate of proliferation. Our data suggest that Gas1 acts indirectly to promote AER proliferation by establishing high levels of FGF10 at the distal tip mesenchyme. Consistently,injection of FGF10 at a high dosage can rescue the mutant AER such that it corresponds to wild-type AER in height and levels of Fgf8 expression(Fig. 9G). This increase of AER height leads us to propose that FGF10 regulates not only the level ofFgf8 expression in the AER but also the proliferation of the AER.

Both gain-of-function (Ohuchi et al.,1997) and gene inactivation studies(Min et al., 1998;Sekine et al., 1999) have provided evidence that Fgf8 activation requires Fgf10. However, it was not clear whether Fgf10 continues to be required to maintain Fgf8 expression in the AER after initiation. The loss ofFgf10/Fgf8 in the Gas1^-/- limb and our FGF10 injection data in the Gas1^-/- background strongly indicate that Fgf10 at the distal region is continuously required forFgf8 expression and that the AER retains the potential to respond to FGF10 long after the loss of Fgf8 expression.

Gas1 is required for Fgf10 transcription at the distal tip mesenchyme of the limb between E10 and E11.5. Normally, from E10,Fgf10 is expressed in a broad contiguous domain directly underneath and extending 150-200 μm away from the AER. In Gas1^-/-limbs, Fgf10 expression is lost in the distal-most 3-5 cell layers(or more cell layers in rare cases) next to the AER(Fig. 8B′-E′). In the chick, FGF8 or AER signals including a combination of FGFs induceFgf10 expression over a broad domain(Ohuchi et al., 1997). InGas1^-/- limbs, the proximal Fgf10 domain is relatively normal (though weaker anteriorly), suggesting that the remaining AER FGFs can still act over a long distance. However, it also indicates that these remaining FGFs are not sufficient to maintain Fgf10 at the distal tip, even though FGFRI expression is normal (including the tip region,not shown) in the mutant limb. Reciprocally, our data indicate that the distal tip Fgf10 expression is necessary and sufficient, as shown by FGF10 rescue injection, to maintain Fgf8. Thus Gas1 is necessary to maintain the distal Fgf10 domain and this is required to maintainFgf8 expression in the AER.

This finding provides several novel insights(Fig. 10). First, there are two distinct regulatory mechanisms for Fgf10 expression in the limb mesenchyme: the distal tip domain, which requires Gas1 function, and the proximal larger domain, which does not. Second, only this tip region ofFgf10 expression is responsible for maintenance of the Fgf8expression in the entire AER. Although the proximal Fgf10 expression domain in the mutant extends to the anterior and posterior borders next to the ectoderm, it is not sufficient to maintain Fgf8 there. Third,Fgf4, 9 and 17 expression is present in the mutant,suggesting that either the FGF10 in the proximal region is sufficient to maintain their expression or their expression does not depend on FGF10. One possible mechanism whereby the three Fgfs are expressed in the absence of the distal FGF10 is that they are regulated by SHH/ZPA.

Gas1 maintains Fgf10 expression at the tip mesenchyme,either directly or indirectly. Gas1 is normally expressed in the anterior two-thirds of the limb (Fig. 1H). It is possible that Gas1 directly controls distalFgf10 expression in this region(Fig. 10, model a). In this model, two signals are required to maintain Fgf10 at the tip,Gas1 in the mesenchyme and the Fgf8 in the AER. In a model of indirect control, Gas1 may help to mediate Fgf8's feedback regulatory loop, which maintains the tip Fgf10 expression(Fig. 10, model b). In this model, GAS1 is an obligatory component of FGF8 signaling as injection of high doses of FGF8 fails to overcome the Gas1^-/- phenotype. However, it should be noted that except for their similar limb defects,Gas1 mutants and Fgf mutants do not share any common defects in other tissues outside of the limb (Sun et al., 1999), suggesting a specialized function of Gas1in the limb in relation to Fgf8. In either model, it is intriguing that the Fgf10 expression is more affected in the central tip than the anterior region in the Gas1^-/- limb and that this causes the entire domain of Fgf8-AER expression to be lost. Nonetheless, the discovery that Gas1 in the mesenchyme is an additional component in the regulatory loop between Fgf10/Fgf8 adds a new dimension to this molecular network.

The Gas1 mutant has phenotypes less severe than the two types ofFgf8/AER-KO mutants reported. For simplicity, only the forelimb phenotype is discussed. When Fgf8 is inactivated prior to its expression in the AER (Moon and Capecchi,2000), there is a severe growth defect and a loss ofFgf10 expression in the anterior limb. When Fgf8 is inactivated shortly after its initiation(Lewandoski et al., 2000), the limb defect is milder and the Fgf10 expression is normal. In both cases, the forelimbs are observably smaller at E10.5 and develop with shortened or missing proximal bones in addition to the autopod defects. In contrast, the Gas1 mutant's limb size reduction is not measurable prior to E10.5 and the phenotype is restricted to the autopod. One possible explanation is that the Gas1 mutant loses Fgf8 expression later and has higher levels of residual FGF8 than both Fgf8/AER-KO mutants. The three mutants may thus represent Fgf8 deficiency at different stages and/or of different levels. Together, these data suggest a progressively diminishing requirement of Fgf8 activity for the proximal elements during the PD growth and patterning of the limb.

One difference between the Gas1 and the Fgf8/AER-KO mutants is puzzling: high levels of Fgf4 are activated in the entire AER in both types of Fgf8/AER-KO mutants(Moon and Capecchi, 2000;Lewandoski et al., 2000). This does not occur in the Gas1 mutant. While the precise reason for this difference is unknown, we suggest that activation of compensatoryFgf4 expression along the entire length of the AER still requiresFgf10 at the specialized distal tip mesenchyme, which is missing only in the Gas1 mutant.

Missing phalanges and syndactyly, the phenotype ofGas1^-/-, can also be induced by chemical inhibitors of proliferation, presumably because the autopod elements are most vulnerable as they are formed late (Shubin and Alberch,1986). The anterior digit condensations appear even later than the posterior ones (Burke and Alberch,1985), making them more sensitive to growth disruption. The small phalanges and metacarpals in the forelimb of Hoxd13^-/-mice (Dolle et al., 1993) were attributed to decreased proliferation(Duboule, 1995). We propose that the Gas1^-/- autopod defects are also a consequence of insufficient precursor mesenchyme generated earlier, thereby causing delayed chondrogenesis and small condensation sizes. The anterior digit defects may thus be due to depletion of a smaller pool of mesenchymal cells by early condensing posterior elements. Since the alterations of the proliferation and PCD patterns in the mutant do not strictly correlate with Gas1expression, we suggest that they reflect a secondary consequence of the heterochrony of the mutant limb caused by the deregulation of Fgf8. The loss of Fgf8 expression may also account for reduced PCD at E11.5(Montero et al., 2001). Finally, the relatively normal patterns of the condensations and expression ofBmp2, Bmp4, Bmp7, BmpRIB, Ihh, and PTHrP in theGas1^-/- limb further argue for a defect in early mesenchymal mass rather than a disruption in patterning or bone growth per se. As the chondrogenic growth regulatory network appears relatively normal, this may explain the post-natal compensatory growth of the mutant autopod elements.

We are in debt to Dr Lee Niswander for her encouragement and insights to help us to complete this project, and her tireless effort to correct the manuscript. We also thank Drs A. Fire, D. Koshland and a member of the Tabin lab for critical reading of the manuscript. This work is supported by the Beckman Foundation, the Dammon-Runyon Cancer Research Fund and an NIH grant(R01-HD 35596).