FGF-regulated BMP signaling is required for eyelid closure and to specify conjunctival epithelial cell fate

Eyelid formation from the surface ectoderm and the underlying periocular mesenchyme involves four processes: eyelid specification, growth, closure and re-opening. In mice, eyelid specification begins by embryonic day (E) 9, when the expression of the transcription factor Foxl2 defines the future location of the eyelids dorsal and ventral to the globe(Swindell et al., 2008). At E11.5, invagination of the dorsal and ventral periocular ectoderm signals the beginning of the period of eyelid growth. The resulting eyelid folds grow towards each other across the surface of the eye between E11.5 and E15.5. At E15.5, a projection of the outer, peridermal layer of the ectoderm extends from the eyelid margins across the cornea until the periderm extensions meet and fuse. The two lids separate at `eye opening' on about postnatal day 10(P10) (Findlater et al.,1993).

The closing eyelids are constituted by a loosely organized mesenchyme and the overlying epithelium. The eyelid epithelium differentiates into the palpebral epidermis (outer surface of the eyelid) and the palpebral conjunctiva (inner surface of the eyelid). The palpebral conjunctiva is continuous with the bulbar conjunctiva (the epithelium covering the anterior periphery of the globe), which is continuous with the corneal epithelium on the most anterior surface of the globe(Fig. 1).

After eyelid closure, the palpebral epidermis differentiates as part of the skin. Stratification and keratinization begin, and the regularly spaced hair follicles of the eyelashes form at the margins of the lids. However, the conjunctival epithelium does not stratify until after the eyelids re-open and it remains non-keratinized throughout life. The mature conjunctival epithelium contains abundant goblet cells, which produce mucus that is important for the properties of the tear film. Soon after birth, the Meibomian glands, which produce a lipid component of the tears, form by in growth of the conjunctival epithelial cells near the inner surface of the lid margin(Findlater et al., 1993).

Defects in eyelid growth or fusion may cause the eyelids to be open at birth (EOB). A surprising number of genes and signaling pathways are required for eyelid closure. An EOB phenotype is seen in mice with germline deletion of activin β-B (Inhbb), MEK kinase1 (Map3k1), c-Jun N-terminal kinase (Mapk8), c-Jun (Jun), the epidermal growth factor (EGF) family members HB-EGF (Hbegf) and transforming growth factor α (Tgfa), and their receptor (Egfr), fibroblast growth factor 10 (Fgf10), its receptor (Fgfr2), the forkhead transcription factors Foxc1 and Foxc2, and the Wnt antagonist Dkk2 (Gage et al.,2008; Kidson et al.,1999; Kume et al.,1998; Li et al.,2001; Li et al.,2003; Luetteke et al.,1994; Luetteke et al.,1993; Miettinen et al.,1995; Mine et al.,2005; Smith et al.,2000; Takatori et al.,2008; Tao et al.,2005; Vassalli et al.,1994; Weston et al.,2004; Zenz et al.,2003; Zhang et al.,2003). Previous studies suggested that activin β-B promotes eyelid closure by activating a Smad-independent cascade involving MEK kinase1,Jun N-terminal kinase (JNK) and c-Jun(Takatori et al., 2008; Weston et al., 2004; Zhang et al., 2003). EGF family members contribute to periderm migration by activating the ERK signaling pathway (Mine et al.,2005). Upstream of the EGF cascade, c-Jun increases EGF receptor expression (Li et al., 2003; Zenz et al., 2003). FGF10 controls eyelid epithelial proliferation and periderm migration by stimulating the expression of activin β-B and TGFα, and by modulating the expression of sonic hedgehog (Shh)(Tao et al., 2005). The administration of a short-acting Shh antagonist at E9 results in EOB(Lipinski et al., 2008). Recently, mice lacking the Wnt antagonist Dkk2, showed EOB,indicating that Wnt activity must be properly tuned during eyelid development.

It has not been clear whether bone morphogenetic protein (BMP) signaling plays a role in eyelid closure. An EOB phenotype was detected in one mouse strain in which Bmpr1a was conditionally deleted in the ectoderm by using a keratin 14-driven Cre-recombinase transgene(Andl et al., 2004). However,mice overexpressing the BMP antagonist noggin under the control of the human K14 or K5 promoters had eyelid defects, but no EOB phenotype(Plikus et al., 2004; Sharov et al., 2003). Mice overexpressing the inhibitory Smad Smad7, driven by the bovine K5 promoter,did have an EOB phenotype (He et al.,2002). However, whether this phenotype was attributable to blocking TGFβ, activin, or BMP signaling has not been clarified. In addition, overexpression of BMP signaling antagonists or deficiencies in the BMP signaling pathway cause other epithelial defects that may indirectly result in EOB. For example, in some of these cases, the epidermal,conjunctival and corneal epithelia were hyperplastic, and sweat glands transdifferentiated into hair follicles(He et al., 2002; Plikus et al., 2004).

To clarify the possible function of BMP signaling in eyelid development, we conditionally deleted two type I BMP receptors, two of the BMP-activated R-Smads or the co-Smad Smad4 in the prospective eyelid epithelium beginning on E9. In each case, the mice showed normal eyelid formation and adequate growth,but the eyelid epithelia did not fuse, resulting in an EOB phenotype. Deletion of the sole type II TGFβ receptor or the two activin- and TGFβ-activated R-Smads did not interfere with eyelid closure. Further analysis suggested that Fgf10 from the mesenchyme activates Fgfr2 in the lid ectoderm. Fgfr2 signaling modulates Shh levels, resulting in Bmp4expression in the mesenchyme. FGF signaling also inhibits Wnt signaling in the eyelid ectoderm, independently of its effects on BMP expression. BMPs are required for the expression of the transcription factors Foxc1 and Foxc2 in the ectoderm, the nuclear translocation of activated c-Jun in periderm cells,the proper timing of conjunctival epithelial differentiation and the establishment of conjunctival epithelial cell fate. In the absence of BMP signaling, ectopic hair follicles formed on the inner edges of the eyelid at the expense of the Meibomian glands, a feature of human lymphedema-distichiasis syndrome.

The following genetically modified mice were used in this study: Le-Cre(Ashery-Padan et al., 2000), Acvr1 flox (Dudas et al.,2004), Bmpr1a flox(Gaussin et al., 2002), Smad4 flox (Yang et al.,2002), Smad1 flox(Huang et al., 2002), Smad5 flox (Umans et al.,2003), Smad2 flox(Piek et al., 2001), Smad3 germline knockout (Roberts et al., 2006), Fgfr2 floxed(Yu et al., 2003), Tgfbr2 floxed (Chytil et al.,2002), presenilin 1 floxed (Yu et al., 2001), presenilin 2 germline knockout(Steiner et al., 1999) and TOPGAL, a Wnt reporter strain (DasGupta and Fuchs, 1999). Matings between mice that were homozygous for the floxed allele, only one of which was Cre-positive, resulted in litters in which about half of the offspring were Cre-positive (conditional knockout,CKO). The others were Cre-negative (wild type). Noon of the day when the vaginal plug was detected was considered as embryonic day (E) 0.5 of development. Embryos were collected at the desired stages (n=3 to 5 for each genotype and stage).

Embryo heads were fixed in 4% paraformaldehyde/PBS overnight at 4°C,dehydrated through a series of ethanol concentrations, embedded in paraffin and sectioned at a thickness of 4 μm. For morphological studies, sections were stained with Hematoxylin and Eosin (Surgipath, Richmond, IL, USA).

Frozen sections were fixed in 4% paraformaldehyde/PBS, treated with proteinase K (10 ug/ml), post-fixed in 4% paraformaldehyde/PBS and acetylated in triethanolamine-acetic anhydride solution. Samples were pre-hybridized in 50% formamide, 5×SSC, 5 mM EDTA, 1×Denhardt's, 100 ug/ml heparin,0.3 mg/ml yeast tRNA and 0.1% Tween-20, incubated in the same solution with riboprobes overnight, washed with 0.2×SSC, blocked in 10% lamb serum and incubated with anti-digoxigenin antibody overnight. The color reaction was developed using NBT and BCIP in the dark. After the reaction was completed,the slides were washed in PBS, fixed in 4% paraformaldehyde/PBS and mounted in 100% glycerol.

Digoxigenin-labeled riboprobes were synthesized from cDNA generated from RNA isolated from wild-type E15.5 eyelids using the following PCR primer pairs:

• Foxc1, 5′-CCAGAAAGTGTTCCAAAAGC-3′and 5′-GAAACCACCCCAGACTAATG-3′;

• Foxc2, 5′-GCCACCTCCTGGTATCTGAAC-3′ and 5′-CTGGGCAAGAACAAAATAGCC-3′;

• BMP4, 5′-TGGTAACCGAATGCTGATGG-3′ and 5′-GGCGACGGCAGTTCTTATTC-3′;

• Sfrp1, 5′-ATCCCCCTCTTTCTGCCTTAG-3′ and 5′-GAAATACCTCTGGGCACTTGG-3′;

• Dkk2, 5′-TTTACAAAGTGGGTTCCCTTG-3′ and 5′-CTCCATTTTCACATCACAAAGC-3′.

Probe for patched 1 was a kind gift from Dr David Ornitz (Washington University, St Louis, MO, USA). Gene expression patterns were compared between CKO and wild-type littermates and each in situ hybridization was performed at least twice.

Frozen sections were warmed to room temperature and then fixed in 4%paraformaldehyde/PBS. After three washes in PBS, the samples were treated with 3% H₂O₂ in methanol to quench endogenous peroxidase activity, blocked in 5% goat serum/0.1% Triton X-100, incubated in primary antibody overnight, washed and processed with tyramide amplification. The antibodies for pSmad1/5/8 and p-c-Jun were from Cell Signaling Technology(Danvers, MA, USA). The keratin 14 and keratin 10 antibody was from Covance Research Products (Denver, PA, USA). The keratin 4 antibody was from Sigma-Aldrich (St Louis, MO, USA). The Dkk2 antibody was from Santa Cruz Biotechnology (Santa Cruz, CA, USA).

Staged embryos expressing a lacZ reporter gene were fixed 4% in paraformaldehyde/PBS at 4°C for 30 minutes, washed twice in PBS with 2 mM MgCl₂, 0.02% NP-40/0.01% deoxycholate (DOC), and stained with X-gal solution [5 mM K3Fe(CN)₆, 5 mM K4Fe(CN)₆, 1 M MgCl₂, 0.02% NP-40/0.01% DOC NP-40, 1 mg/ml X-gal in PBS] for 5 hours at 37°C, post-fixed with 4% paraformaldehyde for 1 hour,cryoprotected and, when required, 10-μm sections were prepared.

Pregnant female mice were injected with 50 mg/kg of a mixture of 10 mM BrdU(Roche, Indianapolis, IN, USA) and 1 mM 5-fluoro-5′-deoxyuridine (Sigma,St Louis, MO, USA) and sacrificed after 1 hour. A monoclonal anti-BrdU antibody (diluted 1:250; Dako, Carpinteria, CA, USA) was used with a Vectastain Elite Mouse IgG ABC kit. Sections were counterstained with Hematoxylin. Terminal deoxynucleotidyl transferase (TdT)-mediated deoxyuridine triphosphate nick end-labeling (TUNEL) was done with an Apoptag kit (Chemicon,Temecula, CA, USA). The deparaffinized slides were treated with 3%H₂O₂ in methanol for 30 minutes, followed by proteinase K treatment (20 μg/ml) for 15 minutes. Slides were incubated with TdT enzyme in equilibration buffer for 1 hour at 37°C. The reaction was terminated with wash buffer provided by the manufacturer for 10 minutes at room temperature. Anti-digoxigenin-peroxidase conjugate was added for 30 minutes at room temperature, followed by DAB and H₂O₂treatment. Slides were counterstained with Hematoxylin.

BrdU and TUNEL-positive cells were counted in the ocular surface epithelia at E14.5 in 4-5 sections from each embryo (wild type, n=6; Smad4^CKO, n=8). The means and standard error (s.e.m.) were calculated from the pooled data. Differences were considered significant if P<0.05, as determined by Student's t-test.

To study the functions of BMPs in early eye development, we used Cre recombinase to delete floxed alleles of key components of the BMP signaling pathway. Transgenic Cre expression was driven by a promoter that is first expressed at E9 in the lens placode and in the ectoderm that later differentiates into the ocular surface epithelia (Le-Cre)(Ashery-Padan et al., 2000). The ocular surface epithelia targeted by the transgene include the palpebral epidermis, palpebral conjunctiva, bulbar conjunctiva and corneal epithelium(Ashery-Padan et al., 2000)(Fig. 1). Le-Cre mice were mated to mice with floxed alleles of two of the three type I BMP receptors(Acvr1 and Bmpr1a), two of the BMP-activated R-Smads(Smad1 and Smad5), the two activin/TGFβ-activated R-Smads (Smad2 and Smad3), the sole type II TGFβreceptor (Tgfbr2) or the co-Smad Smad4. Matings were between Cre-positive homozygous flox and Cre-negative homozygous flox animals,assuring that about half of the offspring were Cre-positive and no animals received two copies of the transgene.

In each knockout targeting the BMP pathway (Acvr1/Bmpr1a, Smad1/5and Smad4), Cre-positive animals had an eyelid-open-at-birth (EOB)phenotype (Fig. 2B-D). Offspring with conditional deletion of one allele of the BMP pathway genes(Acvr1/Bmpr1a, Smad1/5 and Smad4; not shown), both Tgfbr2 alleles (Fig. 2G), or both Smad2 and Smad3 alleles (data not shown) had normal-appearing, closed eyelids at birth and normal-appearing conjunctival epithelium between E15.5 and birth. By examining embryos between E16.5 and birth, we found that eyelids from the Cre-positive embryos with EOB never closed, indicating that the phenotype resulted from the failure of eyelid closure, not from premature eyelid opening (not shown). Because only knockouts in the canonical BMP-Smad pathway had an EOB phenotype, in the remainder of the studies described we show only the phenotype of Smad4^CKO mice to represent the function of BMP signaling in eyelid closure. The `control' eyelids shown are all from homozygous flox,Cre-negative littermates.

Besides the EOB phenotype, eyes with BMP signaling deficiencies showed hyperplasia and what appeared to be keratinization of the conjunctival epithelium. As opposed to the two-layered epithelium seen in wild-type embryos, the conjunctival epithelium comprised several cell layers, including a stratum granulosum with dark, Hematoxylin-stained keratohyalin granules, and an eosinophilic stratum corneum, which suggested keratinization(Fig. 2E,F).

We investigated whether loss of BMP signaling contributed to the failure of eyelid closure by affecting cell proliferation or cell death. At E12.5, BrdU and TUNEL labeling were similar in wild-type and Smad4^CKOeyelids (not shown). However, a significant increase in BrdU labeling occurred in the conjunctival epithelia of Smad4^CKO eyes at E14.5(Fig. 3A-E). This observation was consistent with the conjunctival hyperplasia seen at later stages. No increase in BrdU labeling was detected in the palpebral epidermis(Fig. 3A-E) and no change was seen in programmed cell death in any of the ocular surface epithelia (not shown).

Expression of the forkhead transcription factors Foxc1 and Foxc2 is required for eyelid closure (Kidson et al., 1999; Kume et al.,1998; Smith et al.,2000). Foxc1 mRNA was present in wild-type upper and lower eyelid epithelia (Fig. 4A), but undetectable in Smad4^CKO eyelids(Fig. 4C). Similarly, in wild-type embryos, Foxc2 mRNA was expressed in palpebral conjunctival epithelial cells (Fig. 4D), but could not be detected in Smad4^CKO palpebral conjunctiva(Fig. 4F).

c-Jun and the signaling cascade that leads to its phosphorylation are required for eyelid closure (Li et al.,2003; Zenz et al.,2003). In wild-type mice, a shelf of periderm cells begins to extend from the margin of the eyelids at E15.0(Fig. 5A), covering much of the cornea by E15.5 (Fig. 5B). The nuclei of these periderm cells were strongly stained by an antibody to phosphorylated c-Jun (Fig. 5A,B, insets). In the Smad4^CKO eyelid epithelium, the appearance of the lid margin was comparable to that of wild type at E15.0 (Fig. 5C),although, by E15.5, fewer migrating periderm cells were present than in control eyes (Fig. 5D). In the Smad4^CKO embryos, the levels of phosphorylated c-Jun appeared to be lower than in wild-type periderm cells and p-c-Jun staining was present in the perinuclear cytoplasm, but not in the nuclei(Fig. 5C,D, insets). Thus, BMP signaling is required for the full activation of c-Jun and for its translocation into the nucleus to exert its function as a transcription factor.

Fgf10 signaling via Fgfr2 is essential for eyelid growth and closure(Li et al., 2001; Tao et al., 2005). To determine whether there is a relationship between FGF and BMP signaling, we deleted Fgfr2 in the prospective eyelid epithelium using Le-Cre. Mice deficient in Fgfr2 in the ectoderm showed an EOB phenotype, as described previously (Garcia et al.,2005), and deficiencies in BMP expression and function. In wild-type lower eyelids at E15.5, Bmp4 mRNA was expressed in a cluster of mesenchymal cells underlying the palpebral conjunctival epithelium(Fig. 6A, arrows), but Bmp4 transcripts were undetectable in the lower eyelids of Fgfr2^CKO mice (Fig. 6C). Bmp4 mRNA was expressed in two groups of mesenchymal cells in the wild-type upper eyelid: in a cluster corresponding to those found in the lower eyelid (Fig. 6A,arrows) and in a cluster underlying the palpebral epidermis(Fig. 6A, arrowheads). In the Fgfr2^CKO upper eyelid, Bmp4 mRNA accumulation in the mesenchyme underlying the palpebral conjunctiva was not affected, but Bmp4 transcripts were not detectable in the mesenchyme underlying the palpebral epidermis (Fig. 6B).

A previous study found that Fgf10 maintains Shh expression in the eyelid mesenchyme (Tao et al.,2005). We found that hedgehog function in the mesenchyme, as measured by the expression of the hedgehog receptor patched 1(Ptch1), a direct target of Shh signaling, was remarkably similar to the distribution of Bmp4 transcripts(Fig. 6D). Moreover, the pattern of residual Ptch1 expression in the Fgfr2^CKO eyelid was similar to the pattern of residual Bmp4 expression. Although Ptch1 expression was preserved in the upper lid ectoderm and mesenchyme, it diminished greatly in the lower lid ectoderm and mesenchyme of Fgfr2^CKO mice(Fig. 6E). Ptch1expression was not affected in Smad4^CKO mice(Fig. 6F).

In agreement with the dependence of Bmp4 expression on Fgfr2,nuclear staining for phosphorylated Smad1/5/8, the receptor-activated Smads that transduce BMP signals, was strong in wild-type epithelial cells(Fig. 6G,H), but greatly diminished in the upper and lower eyelids of Fgfr2^CKO mice(Fig. 6I,J). As in the Smad4^CKO eyelids, Foxc1 and Foxc2 mRNA was not detectable in Fgfr2^CKO conjunctival epithelia(Fig. 4B,E). Thus, FGF signaling controls eyelid closure, at least in part, through the activation of BMP signaling.

Loss of Dkk2, a Wnt signaling antagonist, causes EOB, revealing that excessive Wnt signaling can prevent eyelid closure(Gage et al., 2008). To further assess the regulation of Wnt pathway signaling in the ocular surface epithelia, we produced Fgfr2 and Smad4 conditional knockouts in the ectoderm in a TOPGAL background, in which canonical Wnt signaling activates a β-galactosidase reporter transgene(DasGupta and Fuchs, 1999). In wild-type (Cre-negative) eyelids at E15.5, β-galactosidase staining was abundant in hair follicles of the epidermis and in the conjunctival epithelium in a band near the edge of the upper eyelid. Weaker staining was present in a band along the edge of the lower eyelid(Fig. 7A,D). In Fgfr2^CKO eyes, TOPGAL reporter activity increased in intensity and spread over the conjunctival epithelium of the upper and lower eyelids (Fig. 7B,E). In Smad4^CKO eyelids, β-galactosidase expression was not increased, but had a different distribution from wild type. Instead of localizing in a continuous band in the peripheral conjunctival epithelium,staining was present in an extra row of ectopic hair follicles in the upper and lower eyelids (Fig. 7C,F). A double row of eyelashes is called distichiasis. Because distichiasis occurs in humans and mice haploinsufficient for FOXC2(Fang et al., 2000; Kriederman et al., 2003), this finding is consistent with our observation that BMP signaling was required for Foxc2 expression.

Although loss of Dkk2 expression causes EOB(Gage et al., 2008), Dkk2 mRNA or protein expression was not affected in Fgfr2^CKO or in Smad4^CKO eyelid epithelial cells (Fig. 8A-F). FGF signaling modulates Shh expression in the eyelid and, in other tissues, hedgehog signaling induces the expression of the Wnt antagonist,secreted frizzle-related protein 1 (Sfrp1)(He et al., 2006; Katoh and Katoh, 2006). We examined the levels of Sfrp1 mRNA in wild-type, Fgfr2^CKO and Smad4^CKO eyelids. In wild-type eyelids, Sfrp1 mRNA was expressed in the entire eyelid epithelium, with strongest expression in the conjunctival epithelia(Fig. 8G). Consistent with the effects of FGF and BMP signaling on TOPGAL activity, Sfrp1transcripts were undetectable in Fgfr2^CKO eyelids(Fig. 8H), but were present at normal levels in Smad4^CKO eyelids(Fig. 8I). Thus, FGF signaling suppresses Wnt signaling, at least in part, through the activation of Sfrp1. However, the control of Sfrp1 expression by Fgfr2 is independent of BMP signaling.

Although the BMP-Smad pathway is not involved in suppressing Wnt signaling,it appears to be required to suppress ectopic hair follicle formation from the conjunctival epithelial cells at the inner margin of the eyelid(Fig. 7C). In its absence, the uniform band of Wnt signaling near the eyelid margin was replaced by a row of eyelash follicles (Fig. 7C,F). These observations suggest that BMP signaling maintains the pattern of Wnt pathway activity required for the differentiation of the Meibomian gland progenitor cells at the inner margins of the eyelids.

The conjunctival epithelium in mice deficient in BMP signaling developed features that were reminiscent of epidermis, including keratinization and ectopic hair follicles. We, therefore, examined the levels of the epithelial cell differentiation marker keratin 14 (K14) at E15.0, before eyelid closure,and at E15.5, during closure. We also stained for a specific epidermal differentiation marker, keratin 10 (K10), and a conjunctival epithelial differentiation marker, keratin 4 (K4), at E17.5, after eyelid closure. In wild-type mice, the conjunctival epithelium first expressed K14 at about the time of eyelid closure (Fig. 9A). However, in conjunctival epithelial cells deficient in BMP signaling, expression of K14 was precocious(Fig. 9C), with the mutant conjunctiva differentiating at the same time as the epidermis. In wild-type E17.5 eyes, K10 is expressed by epidermis, and K4 is expressed by conjunctiva(Fig. 9D,G). However, in Smad4^CKO conjunctival epithelium, we found ectopic K10 expression, with K4 staining detected only in a few residual cells(Fig. 9F,I), suggesting that most mutant conjunctival cells transdifferentiated into epidermal cells. Although conjunctival differentiation in Fgfr2^CKO eyes was premature, as determined by K14 expression(Fig. 9B), the transdifferentiation of conjunctiva to epidermis was not evident, because Fgfr2^CKO conjunctiva did not express K10(Fig. 9E,H). It seems possible that residual BMP signaling in the Fgfr2^CKO conjunctiva maintained conjunctival cell fate, but was unable to suppress the premature differentiation of this tissue.

Conditional deletion of the genes encoding two of the three type I BMP receptors, two of the three BMP-associated R-Smads or the co-Smad Smad4 in the eyelid epithelia resulted in an EOB phenotype, precocious conjunctival epithelial differentiation and the transdifferentiation of conjunctiva to epidermis, including ectopic eyelash follicle formation and epidermis-specific keratin expression. Although activin signaling is important for eyelid closure, a previous investigation indicated that activin β-B signals are transduced to c-Jun by MEKK1 and JNK, not through the canonical Smad pathway(Zhang et al., 2003). Our results are consistent with this interpretation, as deletion of the activin/TGFβ-stimulated R-Smads Smad2 and Smad3 did not cause an EOB phenotype. Deletion of the TGFβ type II receptor also did not show an eyelid phenotype, indicating that the Smad4 phenotype was due to defects in signaling through the canonical BMP pathway, and not in another of the TGFβ superfamily pathways that share Smad4 function.

The EOB phenotype in eyelids lacking epithelial BMP receptors and BMP-activated R-Smads is consistent with the results of Bmpr1adeletion using a K14-Cre transgene (Andl et al., 2004) and the overexpression of the inhibitory Smad Smad7 using the bovine K5 promoter (He et al.,2002). The inability of K14-driven noggin overexpression to cause EOB may be due to the late expression of K14 in the conjunctiva, just as eyelid closure is occurring, giving insufficient time for noggin to prevent the activation of BMP receptors. EOB in mice deficient in BMP signaling was not due to decreased cell proliferation or increased cell death in the palpebral epithelia. On the contrary, increased cell proliferation was found in the conjunctival epithelia, consistent with the hyperplasia observed in the BMP receptor knockouts or when Smad7 is driven by the K14 promoter(Andl et al., 2004; He et al., 2002). Thus, BMPs normally inhibit the proliferation and differentiation of the conjunctival epithelium, and specify conjunctival epithelial cell fate.

The phosphorylation and function of c-Jun in the eyelid epithelium depends,at least in part, on Smad-independent signaling by activin β-B(Zhang et al., 2003). However,in Acvr1;Bmpr1a^DCKO, Smad1/5^DCKO and Smad4^CKO eyelids, c-Jun appeared to be more weakly phosphorylated than in wild type and failed to translocate into the nuclei of periderm cells. Previous studies showed that Smads can bind to c-Jun when it is not associated with DNA, that c-Jun is phosphorylated after treatment of cells with TGFβ and that Smad-c-Jun complexes promote AP-1-dependent transcription (Liberati et al.,1999; Qing et al.,2000; Verrecchia et al.,2001; Zhang et al.,1998). Our data suggest that BMPs, acting through the canonical R-Smad/Smad4 pathway, cooperate with activin β-B to promote maximal phosphorylation of c-Jun, and that association with the R-Smad/Smad4 complex mediates the translocation of phosphorylated c-Jun to the nucleus. To our knowledge, this is the first report suggesting that activated Smads are required for the nuclear translocation of c-Jun. Further studies are required to determine whether this function of BMP signaling is important in other examples of epithelial fusion, such as closure of the neural tube and ventral closure of the optic cup, for which proper function of the JNK pathway is essential (Xia and Karin,2004).

BMP signaling was required for the accumulation of Foxc1 and Foxc2,transcription factors shown in previous studies to be required for eyelid closure (Kidson et al., 1999; Kume et al., 1998; Smith et al., 2000). Foxc1 and Foxc2 are also expressed in periocular mesenchyme(Gage et al., 2005) and limb bud mesenchyme (Nifuji et al.,2001), and treatment with Bmp4 and Bmp7 increases the transcription of Foxc2 in limb bud mesenchyme in organ culture(Nifuji et al., 2001). The function of Foxc1 and Foxc2 in eyelid closure is not clear. However, it has been shown that Foxc1 and Foxc2 control somitogenesis by regulating Notch signaling (Kume et al., 2001). Foxc transcription factors also directly stimulate the production of the Notch ligand Dll4, to activate Hey2 accumulation in vascular endothelial cells(Hayashi and Kume, 2008). In experiments undertaken for another study, we found that deletion in the surface ectoderm of the presenilins Psen1 and Psen2 resulted in an EOB phenotype (see Fig. S1 in the supplementary material). Because, in most tissues, presenilin activity is required for Notch activation, Foxc1 and Foxc2 might promote Notch signaling during eyelid closure. However, the presenilins are also required for the activity of other transmembrane proteins. For this reason, the function of Notch signaling in eyelid closure requires further exploration.

Although the functions of BMPs that were identified in this study were dependent on Fgfr2, BMPs do not mediate all effects of FGF signaling in eyelid development. Fgfr2 is required in the palpebral ectoderm to suppress Wnt signaling, at least in part by inducing the Wnt antagonist Sfrp1. Loss of BMP signaling does not affect the expression of Sfrp1. This arm of the FGF signaling pathway might involve Shh, as expression of Shh in the eyelid is modulated by Fgf10 (Tao et al., 2005), and the expression of Ptch1, an Shh receptor that is a direct target of Shh signaling, was greatly decreased in the lid ectoderm and the underlying mesenchyme in eyelids lacking Fgfr2 in the ectoderm.

Wnt signaling must be suppressed in the palpebral epithelia to effect eyelid closure (Gage et al.,2008) (this study). However, a band of Wnt signaling activity is normally present in the palpebral conjunctiva near the margins of the upper and lower eyelids. These cells appear to include the precursors of the Meibomian glands. BMP signaling is required to maintain this domain of Wnt signaling and Meibomian gland cell fate(Plikus et al., 2004) (this study). Thus, FGF signaling is required to suppress Wnt signaling and promote BMP signaling, yet BMPs maintain local Wnt signaling in the Meibomian gland precursor cells. The factors that specify the location and extent of BMP and Wnt signaling in Meibomian gland formation remain to be studied.

In mice deficient in BMP signaling, conjunctival epithelial cells in both eyelids formed an extra row of eyelashes, a characteristic called distichiasis. This phenotype is similar to that of mice that overexpress noggin in the ectoderm, in which ectopic eyelashes are formed at the expense of the Meibomian glands (Plikus et al.,2004). Human distichiasis syndrome is characterized by the presence of an aberrant second row of eyelashes in place of the Meibomian glands (Fox, 1962). As a consequence, patients have Meibomian gland dysfunction, corneal irritation,conjunctivitis and photophobia. Most families presenting with distichiasis have lymphedema in common, or lymphedema-distichiasis (LD) syndrome (OMIM 153400). LD syndrome is an autosomal dominant disease caused by mutations in FOXC2. Foxc2 heterozygous mice mimic LD syndrome, demonstrating distichiasis and hyperplasia of lymphatic vessels and lymph nodes(Kriederman et al., 2003). The distichiasis seen in our studies is consistent with a dependence of Foxc2 expression on BMP signaling.

Dkk2-null mice also develop distichiasis, with decreased Foxc2 expression in the eyelids(Gage et al., 2008). Dkk2 expression is promoted by the transcription factor Pitx2, which functions in the neural crest-derived eyelid mesenchyme. Surprisingly, we detected Dkk2 mRNA and protein in the eyelid epithelia and in the mesenchyme. This observation suggests that another pathway regulates Dkk2 expression in the ectoderm. Excessive Wnt signaling, whether resulting from defects in the Pitx2→Dkk2 pathway(Gage et al., 2008), or the Fgf10→Fgfr2→Sfrp1 pathway, appears to be sufficient to suppress Foxc2 expression, prevent eyelid closure and cause distichiasis. Because loss of BMP signaling also leads to distichiasis, it seems possible that excessive Wnt pathway activity inhibits Bmp4 expression in the eyelid mesenchyme or the function of the BMP pathway in the conjunctival epithelium. These possibilities remain to be tested.

Different temporal and spatial expression of keratin intermediate filaments is an important aspect of the differentiation and function of many epithelia(Kurpakus et al., 1994). In wild-type eyelids, the palpebral epidermis expresses K14 before eyelid closure, whereas the conjunctival epithelia begin expressing K14 as the eyelids close. However, in mice deficient in BMP signaling, the conjunctival epithelium expressed K14 at the same time as the epidermis. In addition,conjunctival cells expressed K10, which is normally restricted to the epidermis, and K4 expression was reduced. Together with the transdifferentiation of the Meibomian gland precursor cells to hair follicles,these observations suggest that BMP signaling normally prevents conjunctival cells from adopting the epidermal cell fate(Fig. 10A).

Previous studies and the results described here reveal complex epithelial-mesenchymal interactions in eyelid development. Activation of Fgfr2 in the surface ectoderm by Fgf10 from the underlying mesenchyme(Tao et al., 2005) is required for the localized expression of Bmp4 in the palpebral mesenchyme. The signal from the epithelium that promotes Bmp4 expression in the mesenchyme is likely to be Shh, as Shh is expressed in the eyelid margin from E13.5 and active Shh signaling (indicated by Ptch1expression) has an expression pattern similar to that of Bmp4 in the wild-type eyelid. Moreover, the pattern of residual Shh signaling in the Fgfr2^CKO eyelid is similar to the pattern of residual Bmp4 expression. These observations, together with the fact that preventing BMP signaling did not alter Shh function, suggested that Shh mediates the Fgfr2-dependent cross-talk between epithelium and mesenchyme.

Bmp4 signals to the overlying ectoderm, as demonstrated by the phosphorylation of the BMP-dependent R-Smads in the eyelid epithelia. Phosphorylated R-Smads are also seen in the mesenchyme, suggesting that BMPs may affect additional targets there. In an additional level of complexity, the Bmp4 and Ptch1 expression patterns and Wnt activity were of different magnitude and location in the upper and lower eyelids, and Bmp4 and Ptch1 expression were differentially affected in the upper and lower eyelids by deletion of Fgfr2.

In addition to the functions of the FGF, BMP, Shh and Wnt pathways examined in this study, germline deletion showed that activin β-B, TGFα,HB-EGF and the EGF receptor are each required for eyelid closure(Luetteke et al., 1994; Luetteke et al., 1993; Miettinen et al., 1995; Vassalli et al., 1994). Notch signaling might also be involved (see Fig. S1 in the supplementary material). The pathways activated by these morphogens interact in a remarkably complex web to assure the proper migration and fusion of a small population of periderm cells (Fig. 10B). Further studies are needed to fully define the functions and interactions of these pathways. Such studies will provide a more complete understanding of eyelid fusion and, perhaps, other morphogenetic events that depend on epithelial fusion, such as closure of the neural tube, the optic fissure, the lens vesicle and the palatal shelves.