Expression of Signaling Components in Embryonic Eyelid Epithelium

Closure of an epithelium opening is a critical morphogenetic event for development. An excellent example for this process is the transient closure of embryonic eyelid. Eyelid closure requires shape change and migration of epithelial cells at the tip of the developing eyelids, and is dictated by numerous signaling pathways. Here we evaluated gene expression in epithelial cells isolated from the tip (leading edge, LE) and inner surface epithelium (IE) of the eyelid from E15.5 mouse fetuses by laser capture microdissection (LCM). We showed that the LE and IE cells are different at E15.5, such that IE had higher expression of muscle specific genes, while LE acquired epithelium identities. Despite their distinct destinies, these cells were overall similar in expression of signaling components for the “eyelid closure pathways”. However, while the LE cells had more abundant expression of Fgfr2, Erbb2, Shh, Ptch1 and 2, Smo and Gli2, and Jag1 and Notch1, the IE cells had more abundant expression of Bmp5 and Bmpr1a. In addition, the LE cells had more abundant expression of adenomatosis polyposis coli down-regulated 1 (Apcdd1), but the IE cells had high expression of Dkk2. Our results suggest that the functionally distinct LE and IE cells have also differential expression of signaling molecules that may contribute to the cell-specific responses to morphogenetic signals. The expression pattern suggests that the EGF, Shh and NOTCH pathways are preferentially active in LE cells, the BMP pathways are effective in IE cells, and the Wnt pathway may be repressed in LE and IE cells via different mechanisms.


Introduction
Formation of the eyelid is one of the last major morphogenetic events in mammalian prenatal development. Though for the most part data are scarce in humans, histological analyses of available embryos/fetuses have shown that eyelid development proceeds through four distinct phases, namely, lid formation, growth, fusion and re-opening [1,2]. In mice, eyelid development follows similar steps but has been characterized in greater detail. Mouse eyelid formation begins at around embryonic day 11.5 (E11.5). At this time, the surface ectoderm adjacent to the developing cornea folds to form the lid buds, which are a simple structure consisting of loose periocular mesenchyme (POM) covered by undifferentiated ectoderm [3][4][5][6]. The eyelid buds grow from E12 onward, and they extend across the ocular surface, undergoing proliferation and differentiation. The eyelid at this stage is covered by epidermis, overlaid by periderm at the anterior surface and conjunctiva at the posterior surface. The epithelial margins of the superior and inferior lid fuse between E15 -E16. Lid fusion begins when the periderm cells become rounded and piled up at the leading edges of the eyelids, and then stream out across the corneal surface. The eyelids meet at the inner and outer canthi and temporarily fuse across the cornea [3,4]. Once contact is established between the apposed eyelids, the cells at the fusion junction flatten and form a strip along the fusion line, and they slough off with the rest of the periderm [4,7,8]. Mouse eyelid remains closed between E16.5 and postnatal day [12][13][14]. Cells at the eyelid fusion junction undergo desquamation and/or apoptosis, resulting in separation of the upper and lower eyelids at around postnatal day 14 [4,9].
Much is known about the molecular factors involved in eyelid formation and fusion. This is because, although mice are normally born with a closed eyelid, a large number of genetic mutant strains display a distinct ''eye open at birth'' (EOB) phenotype. The Mouse Genome Informatics (MGI) (http://www.informatics.jax. org/) has a collection of .138 genotypes associated with the phenotype; the number is likely to increase with complete or partial knockout of new genes.
The majority of the EOB phenotype is caused by failure of eyelid fusion at E15-E16. One of the most significant findings made by the analysis of EOB mice is that multiple signaling pathways are involved in the regulation of eyelid closure. Some pathways, such as RA-RXR/RAR and PITX2-DKK2, and the FOXL and OAR2 transcription factors, seem to operate in the periocular mesenchyme [10][11][12]; others, such as the FGF10-FGFR and BMP-BMPR pathways, act through crosstalk between mesenchyme and epithelium [6,13]. Furthermore, a number of pathways, including MAP3K1-JNK, EGFR, ROCK and PCP, are specifically effective in the eyelid epithelial cells . There is also evidence for signal compartmentalization and spatial segregation, so that the signaling pathways are activated in distinct cell population in the developing eyelids [21,36].
Though the outline of the pathways is more or less drawn, the role that the actual players involved in signal transduction has not  [13]. The most remarkable feature of lid closure is the shape change and migration underwent by the epithelial cells at the ''tip'' of the eyelid. This is accompanied by activation of specific morphogenetic pathways. It is possible that the tip cells have unique surrounding tissues, i.e., microenvironments, which produce morphogens for specific activation of signaling pathways. Alternatively, the tip cells may have unique gene expression thereby acquiring new signaling and morphogenetic properties. Gene expression is a crucial facet of its function, and many genes essential for eyelid closure, such as Tgfa, Hb-egf, Activinbb and Map3k1, are indeed up-regulated in the developing eyelid epithelium [6,20,38,39].
In the present work, we applied a global approach to compare gene expression profiles in epithelial cells isolated from the tip (leading edge, LE) and the inner surface (inner epithelium, IE) of the embryonic eyelid. We evaluated the relative abundance in expression of genes whose products might constitute the major ''eyelid closure pathways''. Results may help to understand how signals are distinctly regulated in the LE cells and provide guidance for selecting ''genes of interest'' for expression and knockout studies.

Materials and Methods
Experimental animals C57BL/6 fetuses were collected at E15.5. Euthanasia of the E15.5 fetuses was done by decapitation with surgical scissors, and genotypes were determined by PCR. Experiments conducted with these animals were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of Cincinnati (Protocol no. 06-04-19-01).

Tissue and cell preparation, RNA and cDNA generation and microarray
This process was done as previously described [40]. Briefly, the heads of E15.5 fetuses were embedded in Tissue-Tek OCT medium (Sakura Finetek USA) and stored in 280uC. Eight mm coronal sections were mounted on plain uncoated glass slides, dehydrated and stained with HistoGene LCM frozen section staining kit, and were used for LCM following the manufacturer's protocol (Molecular Devices). Cells from 4 sections were collected  on one LCM cap and lysed for RNA harvesting. The lysates from each fetus were pooled and processed as one biological sample. It was estimated that 10 ng and 15 ng total RNA were obtained from LE and IE eyelid epithelium, respectively, per fetus. RNA was analyzed by Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA) and samples with RNA Integrity Number (RIN) .5.5 were processed for cDNA amplification. cDNA amplification and biotinylation was done using Ovation Pico WTA System (NuGEN, San Carlos, CA) following the manufacturer's instructions. Specifically, RNA (10 ng) was processed into first strand cDNA, a DNA/RNA heteroduplex, and thereafter a linear isothermal amplified cDNA. The amplified cDNA was purified with a PCR Purification Kit (QIAGEN, Valencia, CA).
The cDNAs from each fetus were considered one biological sample and 3 samples were used for triplicate hybridization on the Affymetrix GeneChip Mouse Gene 1.0 ST array (P/N 901168, Affymetrix, Santa Clara, CA). The arrays were hybridized with 15 mg of fragmented aRNA. The hybridization, staining, and washing are carried out using the Affymetrix GeneChip Hybridization Wash and Stain Kit (P/N 900720) following the manufacturer's protocols. The arrays were hybridized for 16 hr at 45uC using Affymetrix Hybridization Oven 640 (P/N 800139). FS450-0001 protocol was used for staining and washing the GeneChips using the Affymetrix Fluidics Station 450 (P/N 00-0079). The GeneChips were scanned with Affymetrix GeneChip Scanner 3000 7G Plus using Affymetrix GeneChip Command Console 3.2.3.1515 software and Affymetrix preset settings.

Quantitative RT-PCR
Quantitative PCR was performed using an MX3000p thermal cycler system and SYBR Green QPCR Master Mix (Stratagene), using conditions optimized for each target gene primers with efficiency greater than 85%, cycles less than 29 and sample locations on the plates been randomized. The PCR products were subjected to melting curve analysis and the relative cycle differences in qRT-PCR were determined using DCt, as described

Statistical and bioinformatics analyses
Array data (GEO repository, accession no. GSE39240) were analyzed at gene level using statistical software R and the limma package of Bioconductor [42] with custom CDF downloaded from BrainArray [43]. Data pre-processing, including background correction and normalization, was performed using RMA. Array quality was assessed using the Array Quality Metrics package of Bioconductor [44]. Statistical significance of differential gene expression between LE and IE samples were established based on empirical Bayes linear model as implemented in limma package [42].
Functional enrichment analysis of differentially expressed genes was performed using the logistic regression based LRpath methodology [45] as implemented in the R package CLEAN [46]. The gene list used in the functional enrichment analysis came from genes associated with Gene Ontology terms [47]. The statistical significance of gene list enrichment was determined based on the False Discovery Rate (fdr) cut-off of 0.1. The statistical significance of deviations of average gene expression levels for genes within the same group were established by calculating gene specific z-statistics and comparing it to the standard Normal distribution. The z-statistic was calculated by subtracting the average of expression levels of all genes in the group from the expression level of the gene and dividing the difference by the standard deviation of the expression levels within the group.

Results and Discussion
Gene expression profiles in the developing eyelid epithelium To identify the molecular signatures of eyelid closure, we collected mouse fetuses at E15.5, a developmental stage immediately before the eyelid beginning to close. We used laser capture microdissection (LCM) to isolate epithelial cells from the leading edge (LE) and inner surface epithelium (IE). The samples were used for expression array and gene expression signatures were analyzed as described [40].
To determine whether the LE and IE cells were different at E15.5, we analyzed the expression data by Gene Ontology (GO). The LE cells were enriched for genes involved in epidermis development, transcription factor activity, pattern specification and odontogenesis. By contrast, the IE cells were enriched for genes for muscle development, RNA splicing, microtubule organization and centrosomes ( Table 1). The GO signatures suggest that the E15.5 LE and IE cells have already departed to distinct paths from their common origin -the ocular surface ectoderm.

Expression of signaling molecules in the FGF and EGF pathways
To evaluate whether the LE and IE cells had differential expression of signaling molecules, we examined genes involved in the FGF and EGF pathways, known to be involved in eyelid closure. The fibroblast growth factor (FGF) family has 22 ligands and four membrane-bound receptors, FGFR 1-4, with different ligand binding affinities [48,49]. In LE and IE cells, the Fgfr2 was the most abundantly expressed receptor gene, and Fgf9 was the highly expressed ligand gene ( Table 2). Between LE and IE, there was no major difference in the expression of genes belonging to the families of FGF ligands and receptors, except for Fgfr2 ( Table 2). The level of Fgfr2 was 1.8-fold higher in LE cells, suggesting that the LE cells might be more responsive to FGF signals than the IE cells.
Previously, we have shown that FGF9 expression was decreased in LE cells of Map3k1 knockout fetuses corresponding to failure of eyelid closure [40]. FGF9 could act in an autocrine fashion to induce epithelial branching, or it could send signals to the mesenchyme to induce PITX2 and FGF10. FGF10 in turn could trans-activate FGFR in epithelial cells and stimulate epithelial budding [50,51]. Genetic studies show that FGF10 is crucial for eyelid closure, but FGF9, though required for sex determination and reproductive system development, lung embryogenesis, and inner ear morphogenesis, is dispensable for eyelid development   [52][53][54]. Since FGF10 was almost undetectable in LE and IE cells, it is possible that this ligand is produced by the underlying mesenchymal cells, responsible for activation of FGFR2 in the eyelid epithelium [5,6]. The epidermal growth factor (EGF) pathway operates in an autocrine fashion, such that ligands produced by the epithelial cells can activate receptors on the same or nearby cells [6,38,55,56]. The mammalian system has nine ligands, which are first expressed as transmembrane proteins comprising a signal sequence, a transmembrane domain and the EGF domain(s). The ligands are then activated by ectodomain shedding that releases the EGF domain from the membrane-bound precursors. This is carried out by members of disintegrin and metalloproteases (ADAMS) family of type I transmembrane Zn-dependent proteases. There are four EGF receptor tyrosine kinases, including EGFR/ERBB1, ERBB2, ERBB3 and ERBB4 [57]. Activation of the receptors is also facilitated by members of the leucine-rich repeat containing Gprotein coupled receptor (LGR) and G protein-coupled receptor (GPCR) families.
In LE and IE cells, the Egfr and Erbb2, and several genes in the GPCR families, such as Lgr4, Gpr125, Gpr20, Gpr180, Gpr89 and Gpr3, were abundantly expressed (Tables 3 and 4). Expression of Adams10 was also abundant ( Table 5). Expression of Gpr56 was relatively abundant in LE cells, whereas expression of Adam 17, Lgr4, Gpr107 and Gpr137b-ps was more abundant in IE cells. Compared to the IE cells, the LE cells had significantly higher expression of Erbb2 (1.8-fold) and Gpr56 (1.3-fold), but less expression of Adamts1 (-2.6-fold).
The ligands specific for ERBB2 are unknown, but ERBB2 can dimerize with EGFR. The heterodimers, similar to the EGFR homodimers, can be activated by amphiregulin (AREG), heparinbinding EGF-like growth factor (HB-EGF) and transforming growth factor a (TGFa) [58]. Activation of the EGFR signaling is essential for embryonic eyelid closure [59]. Based on the relative abundance of receptor gene expression, the EGFR/EGFR and EGFR/ERBB2 dimers are likely to form in the developing eyelid epithelium. Specifically, the EGFR/ERBB2 may be the dominant form in LE, whereas the EGFR/EGFR is likely to be the predominant form in IE cells.
ADAMS10 is important for the development of blood vessels and central nervous system, as well as in pathological conditions such as inflammation and cancer [60]. Recently, it was shown that ADAMS10 may be the sheddase of notch receptors, involved in the release of the extracellular domain and mediating skin development; however, its role in eyelid development has not been established. On the other hand, the Adams17 knockout mice exhibit the open eye phenotype [61]. ADAMS17 is the major sheddase of TGFa, amphiregulin, HB-EGF and epiregulin, and is essential for activation of EGFR during development [62,63]. Of the Lgr/Gpcr families, only the Lgr4 (2/2) mice have defective keratinocyte motility and produce the EOB phenotype. The Lgr4, also known as Gpr48, was known to play a role in HB-EGFinduced EGFR activation [64,65]. The expression of Adams17 and Lgr4 was both relatively abundant in the IE cells (Tables 4 and 5).
The most surprising observation made by the RNA array was that expression of EGFR ligands was scarce in the LE and IE cells (Table 3). This was in clear contrast to previous findings made by in situ hybridization and immunohistochemistry, which showed that expression of TGFa and HB-EGF was up-regulated in a group cells located at the tip of the developing eyelid [6,38,66]. The discrepancy could be explained if induction of ligand is a temporal-spatial event, taking place in a small number of cells and in a narrow window during embryogenesis. Hence, either ligand up-regulation was insignificant at E15.5, or the expression signals were masked or under-represented in the collectives of the LCM captured cells, exemplifying the limitations of this approach.
Taken together, the gene expression data confirm that many genetically identified ''eyelid closure'' factors, such as FGFR, EGFR, ADAMS17 and LGR4, are also relatively abundant in the LE and/or IE cells, but some highly expressed genes, including Fgf9 and Adam10, are not known to be involved in eyelid closure. In comparison to the IE cells, the LE cells have higher expression of Fgfr2 and Erbb2, which may contribute to differential signaling responses of these cells.

Expression of genes involved in the TGFb signaling
The TGFb superfamily consists of more than 30 structurally related ligands. They belong to the Bone Morphogenetic Proteins (BMPs), TGFbs and Activin/Inhibin subfamilies [67]. These ligands act selectively on seven type I and five type II receptors, resulting in receptor dimerization and activation. The receptors in turn activate two sets of so called R-SMAD. SMAD 1, 5, and 8 are  substrates of Type I receptors for BMPs, whereas SMAD2 and 3 are substrates for Type I receptors for TGFbs and Activins. Once activated, R-SMADs assemble with SMAD4, also known as co-SMAD, and the heterodimer translocates into the nucleus to regulate responsive gene expression. In LE and IE cells, the Acvr2a was the significantly expressed receptor gene, while Smad2 was the abundantly expressed gene for intracellular transmitter. In addition, expression of Bmp7 was relatively abundant in LE, and Growth differentiation factor 10 (Gdf10) was abundant in IE cells (Table 6). Furthermore, the IE cells had a slightly higher expression of inhibin beta-B, but much higher Bmp5, Bmpr1a and Acvr1.
Previous genetic studies in mice have implicated TGF b signaling in eyelid closure. Huang et. al. carried out a methodical gene knockout study, in which each TGFb cascade was specifically inactivated in ocular surface epithelium [13]. The results showed that BMP, but not TGFb or activin, signaling was required for eyelid closure. The EOB phenotype was observed in mice lacking the type I BMP receptor genes, Acvr1 and Bmpr1a, the R-Smad genes, Smad 1 and Smad5, and the Co-Smad gene, Smad 4, but not in mice lacking the type II TGFb receptor gene Tgfbr2 and the activin/TGFb-activated R-Smad genes, Smad2 and Smad3. Conditional deletion of Bmpr1a in the ectoderm and overexpression of the inhibitory SMAD7 in keratinocytes also led to an EOB phenotype [68,69]. Our data showed that although the LE and IE cells had type II BMP receptor expression, only the IE cells expressed abundantly the type I receptor BMPR1A. Hence, activation of the BMP pathway can be carried out mainly in the IE cells.
Of the ligands highly expressed in IE cells, BMP5 is required for chondrocytic activity during endochondral ossification, and its deficiency leads to a number of skeletal defects [70]. GDF10 is expressed in skeletal muscles but is dispensable for fetal development [71]. Recently, it was shown that GDF10, similar to TGFb, can activate Smad2/3 and counteract the BMP signals [72]. Of the ligands highly expressed in LE cells, BMP7 is required for eye development, but is dispensable for eyelid closure [73]. The inhibin bB is required for embryonic eyelid closure; however, it may do so through a mechanism independent of SMAD [13,20,39]. These observations seem to support the idea that activation of the BMP pathways for eyelid closure is initiated by BMP4 produced by the the mesenchymal cells, but not ligands produced in the epithelial cells [13]. Collectively, the gene expression pattern has identified differential expression of Bmpr1a, Inhbb and Bmp5 in the LE and IE cells, and suggests that the BMP pathways may be preferentially activated in the IE cells.

Expression of genes involved in the canonical Wnt pathways
The canonical Wnt pathway is activated by binding of ligands to the Frizzled (FZD) receptors, seven-transmembrane proteins with 10 family members (FZD 1-10), and co-receptors, such as the lowdensity lipoprotein-related receptor protein-5 or -6 (LRP5/6) [74,75]. The receptor signal is transduced by the Dishevelled (DVL), which are scaffold proteins that interact with diverse  proteins, including kinases, phosphatases and adaptor proteins. Intracellular transduction of the Wnt signal is carried out by stabilization and cytosolic accumulation of the critical mediator, bcatenin. The b-catenin then translocates to the nucleus, binds with members of the T-cell factor (TCF)/lymphocyte enhancer factor (Lef) family of transcription factors to regulate target gene expression [76]. Wnt ligands are a family of secreted signaling proteins, consisting of 19 members in mammals [77]. Their activities are antagonized by the Secreted frizzled-related proteins (SFRPs) and the dickkopf homologs (DKKs). The SFRP is a family of secreted glycoproteins that may antagonize Wnt-mediated signaling by direct competitive interaction with Wnt ligands or by formation of non-signaling complexes with Frizzled proteins [78,79]. The DKKs, also secreted cysteine-rich proteins, interact with and inhibit the Wnt co-receptor Lrp5/6 [80].
The array data showed that the Fzd3 was the most abundant receptor and Ctnnb1 and Tcf4 were abundant intracellular transducers expressed in LE and IE cells. While Sfrp2 was highly expressed in LE and IE cells, Dkk2 and Sfrp1 were abundantly expressed in the IE cells, and Apcdd1 was abundant in the LE cells (Table 7). In addition, Dkk2 was 4-fold more abundant in the IE cells, conversely, Apcdd1 was 1.7-fold more abundant in the LE cells.
Among the receptors highly expressed, FZD9 is required for bone morphogenesis and is a receptor for non-canonical Wnt that activates JNK, while DVL3 is required for cardiac outflow tract development [81][82][83][84]. Neither, however, is known to be involved  [87][88][89]. Although the nuclear factor TCF4 has not been implicated in eyelid closure, TCF3, through interactions with b-catenin, is shown to be crucial for eyelid closure [36,90]. Using the Wnt reporter mice, it was shown that Wnt activity is repressed overall in eyelid epithelium [36]. The repression is likely to be mediated by the expression of Wnt antagonists. On the one hand, the retinoic acid (RA)-Pitx2 pathway can induce the expression of Wnt antagonists in the periocular mesenchyme; while on the other hand, the BMP and FGFR2 pathways can activate the expression of Wnt antagonists in ocular surface epithelium [13,91]. Our results showed that antagonists could indeed be produced in the LE and IE cells. Of the antagonists, SFRP4 is dispensable for fetal development; SFRP1 and SFRP2 have redundant functions in regulating embryonic patterning, and DKK2 is required for epithelial differentiation and eyelid closure [12,[92][93][94]. In addition, APCDD1 is a membrane-bound glycoprotein that can interact with WNT3A and LRP5 and inhibit Wnt signaling in a cell-autonomous manner [95]. Our data also suggested that the LE and IE cells might use distinct antagonists for Wnt inhibition.
In the Wnt reporter mice, it is also shown that the canonical Wnt pathway is activated in restricted areas of the developing eyelids [36]. Specifically, Wnt activity is induced in a small group of epithelial cells positioned at the transition zone between the palpebral conjunctiva and eyelid tip epidermis, so called mucocutaneous junction (MCJ) [96,97]. Repression of Wnt in the MCJ cells results in failure of eyelid closure [36]. Hence, Wnt may establish distinct morphogenetic fields within the developing eyelids, so that activation takes place in MCJ, but repression occurs elsewhere. Isolation of the MCJ cells and characterizing their molecular signatures may help to understand the developmental roles of the temporal-spatial Wnt activity.

Genes in the SHH, NOTCH and the PCP pathways
The Sonic Hedgehog ligands bind to the transmembrane receptor Patched (Ptch) to initiate pathway signaling [98]. In its inactive state, PTCH exerts an inhibitory effect on the signal transducer Smoothened (SMO), but upon ligand binding, the inhibition on SMO is released and downstream signaling occurs. This leads to the activation of the Gli transcription factors. We found that expression of Ptch1, Smo and Gli2, but not the ligand genes, was relatively abundant in IE and LE cells (Table 8). This is in agreement with the idea that activation of Shh pathway is dependent on Ptch1 expression induced by the FGFR signaling in the eyelid epithelial cells, and the SHH expression induced by FGF10 in the periocular mesenchyme [6,13]. Furthermore, many of the genes were expressed slightly but significantly higher in LE than in IE cells, suggesting that this pathway may be differentially activated in these cells.
The NOTCH cascade consists of NOTCH, its ligands, and intracellular signal transmitters. Mammals possess four different notch receptors, including NOTCH 1-4, which are membranetethered transcription factors. They are activated by the ligands of the Delta, Serrate, Lag-2 families. In LE and IE cells, expression of NOTCH ligands and receptors was overall low, but Jag1 was 1.5fold and Notch 1 was 1.5-fold more abundant in the LE than in the IE cells (Table 9). The role of NOTCH in eyelid development however has been inconclusive. On the one hand, constitutive activation of NOTCH in periocular mesenchyme leads to abnormalities in cranial facial development and incomplete eyelid closure; on the other hand, genetic ablation of NOTCH signaling in ocular surface epithelium does not cause an EOB phenotype [12,13,[99][100][101].
The non-canonical Wnt/planar cell polarity (PCP) pathway regulates cell orientation within the plane of a cell sheet and is involved in convergent extension during development [28,102]. WNT5A, WNT5B, and WNT11 are the non-canonical WNT ligands, and FZD 3/6 and DVL are the receptors, which transmit signals through the core PCP proteins. The core is composed of cytoplasmic Prickled (PK), the transmembrane protein Van Gogh, the cadherin Starry/Flamingo (STAN/FMI), and the Ankyrin repeat protein Diego (DGO) [103,104]. In addition, SEC24B is a cargo-binding component of the COPII vesicle coat [105]. The COPII vesicles are the primary pathway for active transport of secretary proteins from the ER to the Golgi. Though SEC24B is not a PCP core component, it selectively sorts VANGL2 into COPII vesicles thereby controlling PCP assembly and activity.
Expression of non-canonical Wnt ligands and core receptors was overall low in LE and IE cells with a few exceptions (Table 10). While expression of Fzd3 and Dvl3 was relatively abundant in LE and IE cells, expression of naked cuticle 1 homolog (Nkd1) was higher in LE, and expression of Sec24b was higher in IE cells. Genetic inactivation of many PCP genes, including Fzd3/6, Dvl2, Vangl2, Scrb1, Ptk7 and Celsr1, as well as Sec24b, causes craniofacial developmental abnormalities, including open eyelids [27][28][29][30][31][32][33]35,106]. It is yet to be determined whether the eyelid defect is secondary to craniofacial abnormalities resulting from inactivation of the PCP pathways.

Validation of differential gene expression by qRT-PCR
Collectively, the microarray studies identified 20 genes of the morphogenetic signaling pathways were differentially expressed in the LE and IE cells (Fig. 1). To validate the results, we used qRT-PCR to examine 7 relatively abundant genes ( Fig. 2A and 2B). Consistent with the array data, qRT-PCR showed that the LE cells had significantly more expression of Erbb2, Gli2 and Notch1, but significantly less expression of Adamts1, Bmpr1a and Dkk2 than the IE cells. Also consistent with the array data, qRT-PCR showed that the LE cells had a slight but insignificant decrease in expression of Tcf4 and Adam17 than the IE cells (Fig. 2C, Tables 4  and 7). Different from the array data, however, qRT-PCR detected no difference of Fgfr2 expression in LE and IE cells ( Fig. 2A). Hence, most gene expression pattern observed by cDNA array can be validated by qRT-PCR.

Conclusions
The LE and IE cells have the same ontogenic origin, but different developmental fate. The fate divergence can be detected at E15.5, as the IE cells develop gene expression signatures towards the muscle lineage, while the LE cells express epidermal markers. The LE cells also undergo morphological changes and migrate at E15.5 to eventually form the closed eyelid. This morphogenetic event is thought to be dictated by specific activation of signaling pathways. Our results show that the LE and IE cells are overall quite similar in the compositions for the major ''eyelid closure pathways'', but there are a few differences (Fig. 2D). The LE cells have a slight but significant increased expression of Erbb2 of the EGF pathway, Pach1and 2 and Gli2 of the Shh pathway, Jag1 and Notch 1 of the Notch pathway, and Nkd1 of the PCP pathway, but the IE cells have higher expression of Bmpr1a, Acvr1 and Bmp5 of the BMP pathway. In addition, we find higher expression of Apcdd1 in the LE cells, but higher expression of Dkk2 in the IE cells of the Wnt pathway. Differential expression of signaling molecules in the eyelid epithelium may be one of the mechanisms for ectopic activation of morphogenetic pathways. The contributions of the eyelid mesenchyme should also be crucial and can be evaluated using the similar approach. Combination of LCM, cDNA array and pathway analyses can serve as a preliminary screening tool for identifying critical developmental genes for further expression and knockout studie.