Follistatin modulates a BMP autoregulatory loop to control the size and patterning of sensory domains in the developing tongue

Many vertebrate epithelia display regularly spaced specializations, such as hair follicles, feather buds, mammary glands, tooth buds and taste papillae(Chuong et al., 2000), which develop from discrete epithelial swellings known as placodes(Baker and Bronner-Fraser,2001; Mikkola and Millar,2006; Schlosser,2006; Streit,2007). The patterning and differentiation of placodes appears to be the result of complex epithelial-mesenchymal interactions, often involving secreted factors such as Wnts, bone morphogenetic proteins (BMPs) and sonic hedgehog (SHH) (Jiang et al.,1999; Jung et al.,1998; Mikkola and Millar,2006; Plikus et al.,2008). Many placode-derived tissues also regenerate, making them ideal for investigating epithelial stem cells and their niches(Blanpain et al., 2007; Kawauchi et al., 2004; Murray and Calof, 1999; Plikus et al., 2008; Watt et al., 2006). In mammals, placode-derived sensory systems provide some of the few examples of robust neuronal regeneration following injury(Murray and Calof, 1999); the tongue, for example, readily regenerates its sensory (gustatory) cells following damage or denervation (Beidler and Smallman, 1965; Cheal and Oakley, 1977).

The tongue contains two classes of epithelial appendages, called papillae,which develop from an initially uniform field of lingual epithelial cells: (1)taste papillae (circumvallate, fungiform, foliate), which contain taste buds,the gustatory sensory cells of which signal to the nerves of the geniculate ganglion (Barlow, 1999; Barlow, 2003; Krimm, 2007; Mistretta and Liu, 2006); and(2) filiform papillae, keratinized structures that are innervated by mechanosensory neurons of the trigeminal root ganglion(Oakley and Witt, 2004). Taste papillae are distributed on the dorsal surface of the tongue in a regularly spaced pattern, a characteristic shared with other types of epithelial appendages, such as hair follicles.

In recent years, much has been learned about the signals passed between epithelium and mesenchyme in hair follicle formation and regeneration(Alonso and Fuchs, 2006; Bitgood and McMahon, 1995; Chiang et al., 1999; Lyons et al., 1989; Plikus et al., 2008; Reddy et al., 2001; Suzuki et al., 2009; Wall et al., 1993), but in sensory systems such as the tongue, our understanding is much less complete. Within the lingual epithelium, Wnts, BMPs and SHH are all expressed and participate in taste papilla/bud development(Barlow, 1999; Barlow, 2003; Hall et al., 2003; Hall et al., 1999; Iwatsuki et al., 2007; Jung et al., 1999; Liu et al., 2007; Liu et al., 2004; Okubo et al., 2006; Zhou et al., 2006). However,within the lingual mesenchyme, few signals of potential importance have been identified (e.g. Alappat et al.,2005), despite evidence for their existence(Barlow, 2003; Mistretta and Liu, 2006).

Here we identify follistatin (FST), an antagonist of a subset of TGFβsuperfamily molecules, as a mesenchyme-derived secreted polypeptide that is crucial for regulating the size, patterning and innervation of taste papillae and the gustatory cells within them. We show that loss of FST leads to changes in the regions where taste papillae form and affects the spatial patterning of taste papillae, the levels of Wnt activity and Shh expression, and the production and differentiation of sensory cells. We present evidence that these alterations are the consequence of elevated BMP7 signaling, caused (1)directly, by loss of BMP7 antagonism (consequent to loss of FST) and (2)indirectly, via positive autoregulation of Bmp7 expression. Finally,results of computational modeling experiments lead us to propose that one role of an FST-BMP7 regulatory circuit is to suppress spatial noise in the patterning of taste papillae.

Fst^+/- mice (Matzuk et al., 1995) and Bmp7^lacZ/+ mice(Godin et al., 1998) were maintained on a C57BL/6J background. BAT-Gal mice(Maretto et al., 2003) were maintained on a Swiss Webster background (Harlan). Sox2β-geo mice(Avilion et al., 2003), used to report Sox2 expression, were maintained on a CD1 background (Charles River); the Sox2β-geo background does not affect the Fst^-/- phenotype in the tongue (see Fig. S1 in the supplementary material). The day of vaginal plug appearance was designated embryonic day 0.5 (e0.5). All protocols for animal use were approved by the Institutional Animal Care and Use Committee of the University of California,Irvine.

Whole-mount and section ISH were performed according to published protocols(Kawauchi et al., 1999; Murray et al., 2003). Probes were generated from: 318 bp mouse Fst(Wu et al., 2003); 642 bp mouse Shh (Echelard et al.,1993); 748 bp mouse Sox2 (1281-2029 bp of GenBank#X94127); 566 bp mouse Foxa2(Puelles et al., 2003); and 860 bp mouse Bmp7 (Furuta et al.,1997). For β-galactosidase (β-gal) histochemistry (X-Gal staining), tissue was processed as described(Murray et al., 2003). After staining, tongues were dehydrated, cleared in 100% methyl salicylate for 30 minutes and photographed immediately. For detection of cells in S phase,timed-pregnant (e17.5) dams were given one intraperitoneal injection of 5-bromo-2′-deoxyuridine (BrdU; 50 μg/gm) 1 hour before sacrifice, and immunostaining for BrdU was performed as described(Murray et al., 2003), except that rat anti-BrdU was detected using a biotin-conjugated rabbit anti-rat secondary antibody and the Vectastain ABC Kit (Vector Laboratories), with HRP-avidin visualized using DAB histochemistry (Leinco, K107).

For immunostaining, sections were incubated overnight at 4°C with:mouse anti-β-gal (Developmental Studies Hybridoma Bank; 1:2); goat anti-SHH (R&D, AF464; 15 μg/ml); rabbit anti-keratins (Dako, Z0622;1:400); rabbit anti-β-gal (5′-3′, 7-063110; 1:1000); rabbit anti-gustducin (Santa Cruz, sc-395; 1:100); rabbit anti-PGP9.5 (Abcam, 10404;1:1000); or rabbit anti-phospho-Smad1,5,8 (Cell Signaling, 9511; 1:50). For immunofluorescence, sections were counterstained for 10 minutes in bisbenzimide (Hoechst 33258; 20 μg/ml) and visualized using appropriate secondary antibodies [Cy2-conjugated goat anti-rabbit (Jackson); TRITC goat anti-mouse (Zymed); Cy3-conjugated donkey anti-rabbit (Jackson);Cy2-conjugated donkey anti-goat (Jackson)]. For anti-PGP9.5 immunohistochemistry, biotinylated goat anti-rabbit antibody and the Vectastain ABC Kit were employed as described above.

Dissected tongues were stained with Bromophenol Blue according to published protocols (Jonker et al.,2004). For histological analysis, tongues were cryosectioned at 12μm and stained with Hematoxylin and Eosin(Kawauchi et al., 2009). For SEM, tongues were processed as described(Zhou et al., 2006).

Tongues from e12.0 embryos were dissected free of surrounding mandible in cold Earle's Balanced Salt Solution (Invitrogen) and cultured in Millicell-HA chambers (0.45 μm, Millipore) in an airtight chamber with 5%CO₂, 3% O₂, 92% air for 2 days. The medium (DMEM/F12)contained 0.5% heat-inactivated fetal bovine serum, 2% B27 additive and Pen/Strep (Invitrogen). Where indicated, recombinant human BMP7 (PeproTech;1.5 μg/ml) was added.

A rectangular domain 0<y<1 and 0<x<1.5 was divided into a 50×75 grid, and the value at each grid point assigned an initial value equal to the sum of the function 0.8[sin(15x)sin(15y)]⁶, which generates an evenly spaced array of 35 peaks, maximum value=0.8, and a noise value equal to 0.24 times a random number chosen from a log-normal distribution with a mean of 0.3 and a variance of 0.7. The grid was used as the initial conditions for solving the differential equation:

where u represents the value at each grid point (a concentration of BMP7), d is a diffusion coefficient, v is a (uniform)concentration of FST, and k is a decay constant. This equation describes a system in which BMP7 exerts saturating auto-activation with half-maximal activity at a concentration of √2 and a Hill coefficient of 2, and FST exerts linear inhibition. Parameter values were d=0.0005, k=3 and v=0.0875 in the wild-type case, and v=0.07 in the Fst^-/- mutant (this corresponds to a situation in which loss of FST results in a 25% increase in BMP7 specific activity). Boundary conditions were Neumann-type at all boundaries. The equation was solved using the Reaction Diffusion Lab Package (version 2.0) written by Selwyn Hollis in Mathematica (Wolfram Research), which is available at www.appliedsymbols.com.

Fst mRNA is first observed in the tongue anlage at day 11.5 of gestation (e11.5) (Fig. 1A). By e12.5, Fst is expressed throughout the tongue mesenchyme subjacent to lingual epithelium, with patches of expression also evident in deeper developing muscle. At e13.5, a region of especially high Fstexpression is evident in posterior tongue mesenchyme underlying the developing intermolar eminence (IE) (Fig. 1A, e13.5 inset) [the IE, which is located between the anterior fungiform papillae and the posterior circumvallate papilla (CV), contains only non-gustatory filiform papillae (Mistretta and Liu, 2006; Zhou et al.,2006)]. After e14.5, Fst expression declines, and is barely detectable at e16.5 except in deep lingual muscle. The timing of Fst expression in lingual mesenchyme corresponds to the period during which taste placodes form and the size and patterning of taste papillae are determined (Mistretta and Liu,2006).

When tongues of mice null for Fst were examined, we noted no major difference in their overall size as compared with those of wild-type littermates (Fig. 1B,C; data not shown). However, a test to assess formation of the keratinized permeability barrier on the lingual surface (Bromophenol Blue dye exclusion;see Materials and methods) showed that the tongues of P0 Fst ^-/- pups had a dramatic increase in dye permeability(Fig. 1B). Notably, a region of the posterior tongue, anterior to the CV and normally completely impermeable to dye, was deeply stained by Bromophenol Blue in Fst ^-/-mice (Fig. 1B), indicating that the keratinized barrier of the epithelium is compromised. This finding prompted us to examine the epithelium at higher resolution.

Scanning electron micrographs (SEMs) performed at e17.5 revealed several interesting changes in Fst^-/- tongue. First, whereas the posterior surface of the tongue normally exhibits a uniform cobblestone appearance owing to the presence of numerous filiform papillae, in Fst^-/- tongue the posterior surface contained a prominent,smooth, raised region; this region was similar in shape to that which stained aberrantly with Bromophenol Blue (Fig. 1C, compare with 1B). This suggested a lack of filiform papillae in this region, an idea confirmed in histological sections(Fig. 1D, IE). Thus, a distinctive phenotype of the Fst^-/- tongue is a lack of filiform papillae on the epithelial surface of the posterior tongue. Also noted were changes in the size and spacing of fungiform papillae in the anterior region of the Fst^-/- tongue. Although fungiform papillae were present in the mutant, they were irregular in size and often present in tight clusters, rather than showing their normal regular spacing(Fig. 1C, asterisks).

Histological staining revealed that the epithelium covering both the anterior and posterior tongue in Fst^-/- mice was uneven in thickness, with multiple cell layers invaginating into underlying mesenchyme,as well as thinned areas of epithelium overlying domes of mesenchyme(Fig. 1D). These domes were reminiscent of fungiform papillae in shape, but tended not to protrude above the tongue surface (Fig. 1D,arrowhead). BrdU incorporation (to label cells in S phase) demonstrated that the epithelium of both wild-type and Fst^-/- tongues contained proliferating cells, with no noticeable increase in the density of these cells, even in regions where the epithelium was thick and invaginated in the mutant (Fig. 1D).

Despite the morphological changes in filiform and fungiform papillae, we observed no major alterations in the morphology of either the CV or foliate papillae, both of which are present and contain identifiable taste buds in Fst^-/- tongue (see Fig. S2A,B in the supplementary material). Lingual serous glands are also present, and of normal size and number, in Fst^-/- tongue (see Fig. S2C in the supplementary material).

To determine the cellular composition of the dysplastic epithelium in Fst^-/- tongue, we used Sox2 as a marker for taste bud stem/progenitor cells (Okubo et al.,2006). For these experiments, we crossed a Sox2 reporter allele [Sox2β-geo(Avilion et al., 2003)] onto the Fst^-/- background to track Sox2 expression using X-Gal histochemistry and anti-β-galactosidase immunostaining (see Materials and methods). In control tongue at e12.5, a low level of X-Gal staining (indicative of Sox2 expression) was present in the surface epithelium, with higher levels apparent in developing taste placodes(Fig. 2A). As development proceeded, X-Gal staining became confined to the taste buds proper(Fig. 2A, e17.5, inset), and was still evident in the taste buds of adult Sox2β-geomice (data not shown).

At e14.5, the tongues of Fst^-/-;Sox2β-geo mice displayed noticeable disruptions in the pattern of X-Gal staining (Fig. 2E-G′). The X-Gal-stained domains in the apices of the developing fungiform papillae were visibly smaller in the mutant tongue than in the control (Fig. 2D′,G′, arrowheads). The epithelium in the anterior two-thirds of the mutant tongue was also thickened, and was lightly stained throughout (Fig. 2G′),foreshadowing the overt dysplasia and expanded staining evident a few days later. Interestingly, whereas the region containing the IE was free of Sox2-expressing taste papillae and devoid of X-Gal staining in control tongue at e14.5, in Fst nulls the same region displayed ectopic X-Gal staining: Sox2β-geo expression was evident in large clusters of epithelial cells(Fig. 2F) that invaginated into the underlying mesenchyme (Fig. 2F′, arrow). It is notable that the cells within these clusters expressed keratins, a known characteristic of lingual epithelium and taste buds (Luo et al., 2008; Mbiene and Roberts, 2003; Wong et al., 1994), despite their ectopic location, aberrant number and morphology (see Fig. S3 in the supplementary material).

At e17.5, expanded X-Gal staining and Sox2-expressing, invaginated epithelial cell clusters were prominent in the mutant tongue(Fig. 2K-M′). Disruption of the spacing of the fungiform papillae in the anterior region of the Fst^-/- tongue was also more evident at e17.5 (compare Fig. 2J with 2M). In addition, large clusters of invaginating, X-Gal-stained cells were apparent in both the anterior and posterior regions of the mutant tongue at this stage(Fig. 2L′,M′,arrows); these were never observed in controls. Remarkably, the pattern of expanded X-Gal-stained cells in the posterior (IE) region of the mutant tongue corresponded closely to the raised, smoothened region of lingual epithelium seen by SEM and Bromophenol Blue staining (compare Fig. 2K with Fig. 1B,C).

Previous studies have identified an early requirement for SHH and Wnts in initiating and directing morphogenesis of fungiform papillae(Hall et al., 2003; Iwatsuki et al., 2007; Liu et al., 2004). Fig. 3 shows the effect of FST absence on these two signaling systems. In wild-type mice, Shhexpression is detectable throughout the lingual epithelium at the time that taste placodes are initially forming (e12.5), in a diffuse pattern with elevated expression in placodal domains (similar to Sox2)(Fig. 2A; Fig. 3A). By e14.5, Shh expression becomes restricted to regions where fungiform taste papillae will develop in the anterior tongue, and by e17.5 only low-level expression in the taste buds, located in the apices of fungiform papillae,remains (Fig. 3A). In Fst mutants, by contrast, Shh expression in the anterior tongue was lower than normal at e12.5, markedly reduced at e14.5, and below the limits of detection by e17.5 (Fig. 3A). Interestingly, expression of Shh by the CV did not appear to be affected in Fst^-/- tongue at any age examined, suggesting that development of this papilla is not controlled by FST.

To evaluate Wnt-β-catenin activity in Fst^-/-tongue, the BAT-Gal reporter allele was bred onto the Fst^-/- background. In BAT-Gal mice, lacZ is expressed under the control of β-catenin/TCF-response elements, so X-Gal staining reveals endogenous Wnt-β-catenin activity(Maretto et al., 2003). As shown in Fig. 3B,Wnt-β-catenin activity was decreased, compared with controls, in taste placodes in the anterior tongue of Fst^-/- embryos at e12.5. Later in development, at e14.5 and e17.5, X-Gal-stained cell clusters were present in the anterior tongue of Fst^-/-; BAT-Galmice, as well as in BAT-Gal^+/- controls; however, the stained clusters in Fst nulls were noticeably reduced in size and overall X-Gal staining was reduced. X-Gal staining also revealed that the minimum distance between fungiform papillae was altered (reduced) in Fst^-/- tongue after e14.5(Fig. 3B, arrowheads). This was quantified by counting the number of X-Gal-stained papillae per unit area in the anterior one-third of tongues of control (BAT-Gal^+/-)and Fst^-/-; BAT-Gal mice. The results (see Fig. S4 in the supplementary material) show a significant increase in the density of fungiform papillae in the tongue of Fst nulls as compared with control littermates [Fst^-/-; BAT-Gal, 47.9 (±3.8 s.e.m.) papillae per mm²; BAT-Gal^+/-, 28.4(±1.8 s.e.m.)].

Altogether, these findings are consistent with the observed reduction in the expression of Sox2 and Shh in the developing fungiform papillae in the anterior region of Fst^-/- tongue(Fig. 2M′; Fig. 3A). Moreover, they suggest that the defective formation and spacing of anterior taste papillae might be due to misregulation of Wnt and/or SHH signals in Fstnulls.

The IE of the tongue is normally devoid of any taste papillae/buds. The observations that (1) the IE of Fst^-/- tongue contains structures that morphologically resemble fungiform papillae(Fig. 1D), and (2) these ectopic `papillae' express Sox2(Fig. 2), led us to speculate that the epithelium of the IE might take on some characteristics of true taste bud sensory cells (i.e. acquire gustatory competence) in the absence of FST. To test this, we examined the IE region of Fst^-/- tongue for expression of taste bud/progenitor cell markers, for known components of the cytoplasmic taste-transduction pathway, and for innervation by cranial nerves. The results are shown in Fig. 4.

As demonstrated above (Figs 1 and 2), the IE region of Fst^-/- tongue loses filiform papillae and takes on expression of Sox2, a marker for stem/progenitor cells of the fungiform papillae (Okubo et al.,2006). In sections through the IE of Fst^-/-tongue processed for ISH, Sox2 expression was present in clusters of invaginating epithelial cells that resembled ectopic, dysplastic taste papillae (Fig. 4A). The epithelium of the IE region in Fst^-/- tongue also displayed discrete patches of both Shh expression and Wnt-β-catenin activity: these were never observed in the IE of wild-type controls (Fig. 4A).

Functional taste papillae are innervated by lingual nerves(Krimm, 2007; Ringstedt et al., 1999). To address whether the ectopic papillae in the IE of Fst^-/-tongue acquire this characteristic, we used an antibody to the pan-neuronal marker PGP9.5 (UCHL1 - Mouse Genome Informatics)(Liu et al., 2007). As shown in Fig. 4B, mutant IE contained nerve fibers that extended into the cores of ectopic, SOX2-expressing papillae. We detected ingrowing PGP9.5-positive nerve fibers in 44% (173/394)of IE fields in Fst nulls, whereas innervation was never detected in the IE of wild-type tongue (0/143 fields). In addition to providing evidence for the acquisition of another functional characteristic by the ectopic taste papillae of Fst^-/- IE, these findings provide evidence that it is the presence of taste papillae within the tongue that controls the ingrowth of nerves into the presumptive sensory field.

Transduction of chemical signals by the taste buds in fungiform papillae requires not only innervation, but also maturation of taste receptor cells within the taste buds. This process includes expression of gustducin (a component of the T1R family of G-protein-coupled taste receptors; also known as GNAT3). We found that gustducin is expressed in the ectopic taste papillae that develop in the IE of Fst^-/- tongue, whereas gustducin is never expressed in the IE of wild-type tongue(Fig. 4C). It is notable that gustducin was expressed only in small subsets of cells in the apical region of the much larger, SOX2-expressing domains of the ectopic taste papillae(Fig. 4C, arrowheads). Similarly, the ectopic expression of Shh and Wnt-β-catenin activity observed in the IE of Fst^-/- tongue was also seen in the apical region of the epithelium(Fig. 4A). These observations raise the possibility that although SOX2 expression marks cells as competent to develop into functional taste bud cells(Okubo et al., 2006), and FST is crucial for normal expression of SOX2(Fig. 2, above), other factors- such as Wnts and SHH - are required for full differentiation of SOX2-expressing progenitors into functional taste receptor cells(Iwatsuki et al., 2007; Liu et al., 2007; Liu et al., 2004; Okubo et al., 2008; Okubo et al., 2006).

We also noted that gustducin appears to be expressed prematurely in SOX2-expressing papillae in the anterior tongue of Fst^-/-mice, being detectable as early as e17.5(Fig. 4C, Ant; data not shown). Published reports indicate that gustducin expression normally does not occur until after birth in rodents (Wong et al.,1999; Zhang et al.,2007). This suggests that taste receptor cells in anterior fungiform papillae undergo accelerated differentiation in the absence of FST. Additionally, the observation that SOX2 and gustducin are coexpressed in individual cells of ectopic taste papillae(Fig. 4C, arrowheads) suggests a lineage relationship between SOX2-positive taste bud stem/progenitor cells and mature taste receptor cells that express gustducin, a finding in agreement with recent fate-mapping studies (Okubo et al., 2008).

Foxa2 encodes a transcription factor that is expressed both in developing lingual epithelium and mature taste buds(Besnard et al., 2004; Luo et al., 2008). We used ISH to determine whether the ectopic taste papillae that form in the IE and anterior regions of the tongue of Fst^-/- mice acquire this additional marker of sensory (taste) epithelium. As shown in Fig. 4D, Foxa2 is not expressed in the posterior/IE region of the tongue in wild-type mice at e18.5. However, Foxa2 was expressed in the invaginated clusters of epithelial cells present in the IE of Fst^-/- mice at this age, and its pattern of expression mimicked that of Sox2(Fig. 4A). In agreement with others (Luo et al., 2008), we found that Foxa2 is expressed within definitive taste buds in the anterior tongue at e18.5 (Fig. 4D). Foxa2 was also expressed within the ectopic epithelial cell clusters that invaginate into the underlying mesenchyme in Fst nulls, a further indication that the cells in these clusters possess characteristics of definitive taste bud cells.

In summary, the data indicate that in the absence of FST, epithelial cells of the IE region of the posterior tongue acquire many characteristics of taste papillae, including innervation by lingual nerves and expression of markers characteristic of taste bud progenitor cells and gustatory sensory cells. Since (1) the CV remains unchanged in Fst^-/- tongue (Figs 1, 2, 3), and (2) the ectopic papillae that form in the posterior region of Fst^-/-tongue resemble, both histologically and by expression of a number of molecular markers, those that form in the anterior tongue of Fstnulls (Fig. 4), these observations suggest that an important function of FST is to suppress the formation of ectopic fungiform taste papillae in the IE region.

FST binds to and inhibits several ligands of the activin subfamily, members of which signal via phosphorylation of cytoplasmic SMAD2 and SMAD3; FST also functions as an inhibitor of BMP7, which signals via phosphorylation of SMADs 1, 5 and 8 (Amthor et al.,2002; Khokha et al.,2005; Schneyer et al.,2003; Schneyer et al.,1994; Sidis et al.,2006). In agreement with others(Nie, 2005), we found that Bmp7 expression is first detectable in lingual epithelium at e11.5;later, expression extends throughout the anterior epithelium and then becomes progressively restricted, in anterior tongue, to taste buds of the fungiform papillae, with some patches of low-level expression in mesenchyme and muscle(see Fig. S5 in the supplementary material). In posterior tongue and IE, Bmp7 expression is normally detectable only in the lingual glands by the time of birth (see Fig. S2 in the supplementary material; Fig. 5B).

Since BMPs have been shown to inhibit taste papilla development in culture(Zhou et al., 2006), whereas loss of FST causes ectopic taste papilla formation, we postulated that changes in BMP signaling might underlie the alterations in taste papilla development observed in Fst^-/- tongue. To test this, we used anti-phospho-SMAD1,5,8 immunostaining. As shown in Fig. 5, elevated and ectopic phospho-SMAD1,5,8 immunostaining was evident in localized patches of invaginating epithelial cells (ectopic taste papillae) in both the anterior(Fig. 5A) and posterior(Fig. 5B, IE) regions of Fst mutant tongue. We also observed ectopic expression of Bmp7 in these same invaginating epithelial cell clusters(Fig. 5). Together, these data indicate that FST suppresses both Bmp7 expression and BMP7 signaling during normal tongue development, and suggest that this serves two functions:(1) to control the size and patterning of developing taste papillae in the anterior tongue; and (2) to suppress the formation of ectopic taste papillae in the posterior (IE) tongue. A summary of the developmental phenotypes in the tongue that result from the absence of Fst expression is given in Table S1 in the supplementary material.

To directly test the effects of elevated BMP7 signaling on taste papilla development, we cultured tongues in recombinant BMP7 and looked for changes in Wnt-β-catenin and SHH activity (both are suppressed in the anterior tongue of Fst^-/- mice)(Fig. 3). The data (see Fig. S6 in the supplementary material) demonstrate that increasing BMP7 levels causes a significant reduction both in Wnt-β-catenin activity and in the number and size of SHH-positive taste buds on the dorsum of the anterior tongue. These effects strongly resemble the phenotypes observed in vivo in Fst^-/- tongue (Fig. 3), supporting the idea that a primary role of Fst in taste bud development is the suppression of BMP7 activity.

If this idea is correct, then loss of Bmp7 should compensate for loss of Fst. As shown in Fig. 6, this is the case. In Fst^-/-;Bmp7^-/- mice, two of the most severe phenotypes observed in Fst^-/- tongue -loss of the surface permeability barrier of the IE resulting from an absence of filiform papillae (Fig. 6A),and loss of Shh expression in developing taste buds(Fig. 6B) - were rescued. This result provides strong evidence that the actions of Fst in the development of taste papillae/buds are mediated through antagonism of BMP7.

The observation that Bmp7 expression is markedly elevated and appears at ectopic locations in Fst^-/- tongue(Fig. 5) suggests that BMP7 might positively regulate its own expression. This idea is further supported by the observation that substantial rescue of Fst^-/-phenotypes can be achieved by loss of a single allele of Bmp7(Fig. 6A,B), a result which suggests that high BMP7 signaling might be required to maintain Bmp7expression. To directly test for Bmp7 autoregulation, we cultured e12.0 tongues dissected from Bmp7^lacZ/+ mice, which contain a lacZ reporter gene inserted into the first coding exon of Bmp7, in the presence or absence of BMP7. X-Gal staining was then used as an assay for Bmp7 promoter activity. As shown in Fig. 6C, tongues cultured in the presence of BMP7 showed upregulation of Bmp7 promoter activity in both anterior and posterior lingual epithelium. These results indicate that Bmp7 stimulates it own expression, most likely at the transcriptional level.

Interestingly, despite the ability of Bmp7 mutations to rescue the tongue phenotypes of Fst mutants, we have so far observed no lingual abnormalities in mice mutant for Bmp7 alone. Such mice have normal expression of taste bud markers in both anterior and posterior tongue (see Fig. S7 in the supplementary material), and SEM analysis demonstrates that both fungiform taste papillae and IE are of normal morphology in the mutant tongues (Fig. 6D).

The regular spacing of taste papillae on the tongue is reminiscent of the regular spacing of other epithelial appendages, such as hair and feathers. In recent years, evidence has emerged that such patterning is controlled by the mechanism of diffusion-driven instability, in which a diffusible molecule activates both its own expression and that of a diffusible inhibitor that acts over a greater distance (Jiang et al.,1999; Jung et al.,1998; Maini et al.,2006; Sick et al.,2006). The ectopic formation and aberrant spacing of papillae in the Fst^-/- tongue suggests that FST, through its role in suppressing BMP7 action, plays an important role in such a patterning process. However, as Bmp7^-/- mice have well-patterned tongues(Fig. 6C; see Fig. S7 in the supplementary material), it seems unlikely that BMP7 and FST themselves form the activator/inhibitor pair that drives papilla patterning. In addition,although BMP7 does activate its own expression in the tongue(Fig. 6D), we find no evidence that it regulates expression of Fst (data not shown).

A more likely role for FST and BMP7 is in modulating a patterning system that is dependent on other factors, such as SHH and Wnts, both of which, along with SOX2, are crucial for taste papilla/bud development in the anterior tongue (Iwatsuki et al., 2007; Liu et al., 2007; Liu et al., 2004; Okubo et al., 2006). Our concept of how FST and BMP7 actions might regulate these processes to control development of fungiform taste papillae in the anterior tongue is illustrated in Fig. 7A. We conceive of an analogous process occurring in the IE, with the important difference that in the developing IE Bmp7 expression is normally low or nil(Fig. 5B), whereas Fstexpression is very high (Fig. 1A). This results in a lack of Sox2 expression, so that the region normally produces only filiform (non-gustatory) papillae, rather than fungiform (gustatory) papillae (Fig. 1) (Okubo et al.,2006).

Using computational simulations, it is straightforward to show that any system with an auto-activating, diffusible signaling molecule (e.g. BMP7) and a ubiquitously expressed inhibitor (e.g. FST) will tend to act as a spatial noise filter. The findings from such a simulation for fungiform papilla development in anterior tongue are presented in Fig. 7B. They demonstrate how a weak and noisy pre-pattern of initial Bmp7 expression converts itself, over time, into one that is strong and relatively noise-free(`wildtype'). Loss of Fst activity (`follistatin^-/-')under the same simulation conditions leads to domains of Bmp7expression that are abnormally large and less regularly spaced, much like those observed in the Fst^-/- tongue(Fig. 5). Thus, a circuit comprising epithelial BMP7 and mesenchymal FST could play an important role in regulating the fidelity of patterning.

A role for epithelial-mesenchymal interactions in taste papilla morphogenesis has been proposed (e.g. Barlow, 2003; Mistretta and Liu, 2006), but relevant mesenchymal factors have not been identified previously. In this study, we show that FST, a secreted protein expressed in tongue mesenchyme,controls cell fate in the lingual epithelium and influences formation of the taste papillae that provide a niche for presumptive taste bud stem cells. FST regulates numerous biological processes through its ability to inhibit the activities of certain members of the TGFβ superfamily of secreted proteins (Balemans and Van Hul,2002). In the tongue, we show that FST inhibits BMP7 activity,which in turn modulates signaling through the Wnt pathway, controls expression of Shh, Sox2 and Bmp7 itself, and influences spatial patterning. Support for the conclusion that FST acts primarily by regulating the function of BMP7 comes from the fact that loss of Bmp7 can rescue the Fst^-/- phenotype. However, we cannot exclude the possibility that FST also acts as an antagonist of other TGFβ family members to which it binds and that are present in the developing tongue(Nie, 2005; Sidis et al., 2006; Zhou et al., 2006).

Based on the available data, we propose that BMP7 plays at least two distinct, and partly antagonistic, roles in fungiform papilla development(Fig. 7A). The first role is in promoting the specification or stabilization of Sox2-positive progenitors that give rise to taste buds. Sox2, which is necessary and sufficient to drive lingual epithelial cells to differentiate into neural taste bud progenitors rather than keratinocytes [filiform papillae, in the case of the tongue (Okubo et al.,2006)], is initially expressed diffusely in lingual epithelium,but becomes progressively restricted to placodes of the developing taste papillae in the anterior tongue, where it is maintained at high levels in taste buds and taste papillae postnatally. Neighboring cells that differentiate into the keratinocytes of filiform papillae normally downregulate Sox2 expression. Here we show that, in the absence of FST, cells that normally differentiate into keratinocytes maintain elevated levels of Sox2 and, in at least some cases, appear to develop into taste buds expressing markers of taste transduction. Large numbers of ectopic Sox2-expressing cells also develop in the IE of Fst mutant tongue, and many of these cells are incorporated into fungiform papilla-like structures. The IE of Fst^-/- tongue also loses its keratinized filiform papillae, suggesting that conversion of filiform papillae to gustatory, fungiform papillae is favored in the absence of FST. Furthermore, we found that ectopic taste papillae of both anterior and posterior Fst^-/- tongue are innervated, suggesting that when the fate of the epithelium is altered, projecting nerve fibers can be redirected to new targets. These data advocate that neural competence is not an intrinsically determined aspect of tongue development, but rather one that is directed by interactions between the tongue epithelium and its underlying mesenchyme.

The second proposed role for BMP7 is in inhibiting a Wnt-SHH signaling circuit that maintains Sox2-positive cells within developing taste papillae that have already been specified. Within the fungiform papillae of Fst mutants, we observe coordinate downregulation of canonical Wnt signaling, Shh expression and Sox2 expression(Fig. 2M′; Fig. 3A,B). Recent studies have indicated that canonical Wnt signaling promotes Sox2 expression,which is required for taste papillae and taste bud development(Okubo et al., 2006) and induces expression of Shh (Liu et al., 2007). SHH in turn appears to act in a negative-feedback loop to antagonize Wnt activity (Iwatsuki et al., 2007). Therefore, the simplest explanation for the results we observe in the anterior tongue of Fst^-/- mice is that reduction or loss of Wnt signaling, resulting from increased BMP7 activity, is the primary defect, with changes in Shh and Sox2 expression being downstream consequences.

Because the effects of FST loss are rescued by a reduction in Bmp7dosage (Fig. 6A,B), and can be mimicked in vitro by the addition of BMP7 to wild-type tongues (see Fig. S6 in the supplementary material), we infer that BMP7 is a negative regulator of Wnt signaling in developing taste papillae. The fact that developing taste buds in the Fst^-/- tongue are not only reduced in size, but also show the premature appearance of gustducin-positive cells(Fig. 4C), suggests that one mechanism of BMP7 action might be to promote stem/progenitor cell differentiation at the expense of self-replication. Inhibition of the Wnt pathway by BMP signaling has been reported in various tissues: for example, in hair follicles, BMPs suppress LEF1 activation via E-cadherin induction(Jamora et al., 2003); and in muscle, BMPs inhibit TCF expression(Bonafede et al., 2006).

The dual nature of the effect of BMP7 - promoting the allocation of progenitor cells to the taste bud fate while at the same time limiting the expansion of those cells - helps to explain why the Fst^-/-phenotype is characterized both by too much Sox2 expression in places where it should not be, and not enough Sox2 expression in places where it should be. It also provides potential explanations for the regional differences in tongue patterning in wild-type and mutant animals. For example,in the IE, in which Fst expression levels are normally very high(Fig. 1A), inhibition of the role of BMP7 in specifying Sox2-expressing progenitors could explain why taste papillae are normally absent, yet appear in large numbers in the Fst mutant. In the anterior tongue, where Fst expression is normally not as high as in the IE (Fig. 1A), there is normally sufficient BMP7 signaling to allow taste papilla formation. Consequently, when BMP7 signaling is increased in the Fst mutant, a major phenotype is the reduced expression of taste bud markers, such as Shh, within those papillae(Fig. 3).

From the data presented here we may also infer that although FST/BMP7 might be necessary to induce the taste bud stem/progenitor cells of taste papillae,this is not sufficient to induce the morphological specializations of the papillae themselves. In Fst^-/- tongue, many of the supernumerary Sox2-positive cells remain in clusters near the basal surface of the epithelium and do not become associated with morphological papillae. This suggests that taste papilla formation requires additional signals. Interestingly, our data indicate that such signals would need to be present in the IE, even though taste papillae normally never form there.

The notion that signals other than BMP7 dictate the locations of taste papillae also agrees with our suggestion that BMP7 does not primarily drive pattern formation, but rather improves its accuracy by suppressing noise(Fig. 7B; see Fig. S8 in the supplementary material). Patterns formed by diffusion-driven instability are intrinsically sensitive to spatial noise; indeed, the amplification of small,spatial non-uniformities is what can drive such systems to develop patterns spontaneously. In the tongue, the gradual conversion of uniform, low levels of Wnt signaling and Shh expression into regularly spaced puncta with high levels of Wnt activity and Shh expression, also suggests that pattern formation works through the amplification of subtle initial asymmetries. That the pattern becomes partially, but not completely,disorganized in the Fst mutant supports the view that BMP7 improves the ability of the patterning mechanism to distinguish such asymmetries from random variations in gene expression. Whether that patterning mechanism consists solely of the Wnt/SHH activator/inhibitor pair, or involves other molecules (such as DKK) (c.f. Sick et al.,2006), remains to be determined. Regardless, the results of this study clearly show that FST plays a major role in the epithelial-mesenchymal interactions that govern the morphogenesis of tongue epithelium.