The FERM protein Epb4.1l5 is required for organization of the neural plate and for the epithelial-mesenchymal transition at the primitive streak of the mouse embryo

Proteins of the FERM-domain family regulate epithelial organization by linking membrane-associated proteins to the actin cytoskeleton(Bretscher et al., 2002; Mangeat et al., 1999). The mouse genome encodes at least 50 FERM proteins, but the functions of only a few, including the actin-binding proteins ezrin (also known as Vil2 - Mouse Genome Informatics), radixin, moesin, Nf2 (also known as merlin) and the erythrocyte protein band 4.1, have been characterized(Bretscher et al., 2002; Mangeat et al., 1999). Of these proteins, only Nf2, the product of the Nf2 tumor suppressor gene, has been shown to be required for early embryonic development; embryos that lack Nf2 die shortly after the onset of gastrulation due to defects in the epithelial organization of the extra-embryonic ectoderm(McClatchey et al., 1997). Here we identify Lulu (Epb4.1l5) as another FERM protein that is essential for mouse development, and show that it has central roles in early embryonic morphogenesis.

During early postimplantation development, the single-layered columnar epithelium of the mouse epiblast is transformed over the course of 2 days into a three-layered embryo with a well-defined body plan(Tam and Gad, 2004). Beginning at embryonic day (E)6.5, cells at the primitive streak undergo an epithelial-to-mesenchymal transition (EMT) to give rise to the mesodermal and endodermal germ layers. Once generated, each germ layer undergoes a characteristic set of morphological changes: the mesenchymal cells of the mesodermal layer migrate around the embryonic circumference, endodermal cells form an epithelium that folds to generate the gut tube, and cells of the neural epithelium bend and contract their apical surfaces to generate a closed neural tube.

In an N-ethyl N-nitrosourea (ENU)-mutagenesis screen for mutations that affect the morphology of the mid-gestation mouse embryo, we isolated a mutation that we named limulus (lulu) based on the flat,plate-like morphology of the mutant embryos(García-García et al.,2005). lulu embryos have defects in the morphogenesis of all three germ layers: very little paraxial mesoderm is specified, the endoderm fails to form the gut tube, and the neural plate has an irregular shape and does not generate a neural tube. Based on positional cloning, we find that lulu is a null allele of the gene encoding the mouse FERM-domain protein erythrocyte protein band 4.1-like 5 (Epb4.1l5); the allele is designated Epb4.1l5^lulu and we refer to the allele here as lulu (Epb4.1l5 has also been called YMO1)(Laprise et al., 2006).

Homologs of Epb4.1l5 play roles in specific aspects of Drosophila and zebrafish development(Hoover and Bryant, 2002; Hsu et al., 2006; Jensen and Westerfield, 2004; Laprise et al., 2006). The Drosophila ortholog of Epb4.1l5, named yurt, is required for epithelial polarity and dorsal closure in the embryo, and for photoreceptor morphogenesis (Hoover and Bryant, 2002; Laprise et al.,2006). Yurt appears to act as a negative regulator of Crumbs, a key determinant of the apical domain of epithelial cells(Tepass et al., 1990; Wodarz et al., 1995). The zebrafish epb41l5 (also known as mosaic eyes; moe)gene is required for the layering of the retina and inflation of the brain ventricles, and, like Yurt, regulates the size of the apical membrane domain,possibly also via the negative regulation of Crumbs(Hsu et al., 2006; Jensen et al., 2001; Jensen and Westerfield,2004).

Analysis of the tissues affected in lulu mutant embryos reveals that Lulu has specific roles in the EMT at gastrulation and in the organization of the pseudo-stratified epithelium of the neural plate. In both contexts, the defects in tissue organization are associated with dramatic changes in the organization of the actin cytoskeleton. In contrast to the results in zebrafish and Drosophila, Crumbs localization appears not to be affected in lulu mutants. The results indicate that Lulu is crucial for the dynamic rearrangements of the actin cytoskeleton that occur during the morphogenesis of epithelial tissues.

lulu was generated by ENU mutagenesis of C57/BL6J mice(García-García and Anderson,2003; García-García et al.,2005; Kasarskis et al.,1998). The name limulus refers to the embryonic shape at E8.5, which resembles the horseshoe crab Limulus polyphemos. The GT1 and GT2 alleles of Epb4.1l5 correspond to BayGenomics lines AE0088 and XC282, respectively. Embryonic stem (ES) cells were injected into C57BL/6J blastocysts to produce chimeras, which were screened for transmission of the gene trap. Carriers were genotyped by PCR amplification of the neomycin resistance gene using primer sets from the Jackson Laboratories.

lulu was mapped by backcrossing to C3HeB/FeJ, using linkage to flanking simple sequence length polymorphism (SSLP) markers from the MIT database or new markers that we generated(http://mouse.ski.mskcc.org/). lulu was mapped to a 650 kb interval by analyzing approximately 1000 informative recombination opportunities. The ENSEMBL and Celera databases were used for mapping and candidate gene selection; Epb4.1l5, RalB and Ptpn4 were sequenced because of their predicted associations with the actin cytoskeleton. Complementary (c)DNAs of candidate genes were made from lulu mutant and C57BL/6J embryos using Superscript OneStep(Invitrogen). Products were subcloned into the Invitrogen pCR 2.1-TOPO vector and sequenced using T7 and SP6 primers. No sequence changes were found in the coding regions of RalB and Ptpn4; a single C to T transition was identified in the Epb4.1l5 coding sequence, in multiple lulu mutant embryos. The lulu mutation abolishes a BsaJI restriction site, which was used to confirm the mutation in carrier mice.

Whole-mount in situ hybridization, X-gal staining and immunohistochemistry were performed using standard protocols(Belo et al., 1997; Hogan et al., 1994; Yamada et al., 1993). All wild-type controls were littermates of the mutant embryos. Embryos were dissected in phosphate-buffered saline (PBS)/0.4% bovine serum albumin (BSA)and fixed in PBS/4% paraformaldehyde at 4°C overnight for in situ hybridization, or for 1 hour at room temperature for immunohistochemistry,with the exception that embryos used for N-cadherin staining were fixed for 10 minutes on ice in 100% methanol. Fixed embryos were embedded in OCT and cryosectioned at 8 μm thickness. Mutant and wild-type littermates were dissected and fixed together, and then embedded and sectioned in one block. Single slides containing wild-type and mutant sections were used for staining. Whole-mount embryos were imaged using a Zeiss Axiocam HRC digital camera on a Leica DM1RE2 inverted confocal microscope. Sections were imaged on a Leica MZFLIII microscope. In all cases, identical microscope and camera settings were used when imaging wild-type and mutant samples. Confocal datasets were analyzed using the Volocity software package (Improvision); 3D reconstructions were created using the Amira package (Mercury Computer Systems). The mitotic index was defined as the ratio of phospho-histone-H3-positive cells to total cells in anterior neural plate sections. Phospho-histone-H3-positive nuclei located 1 or more nuclear diameters away from the apical surface were scored as ectopic. Nuclear length and width were measured from 3D reconstructions.

HeLa cells were grown in Dulbecco's modified Eagle's medium (DMEM)supplemented with 10% fetal bovine serum. For transfection, cells were grown in six-well dishes and seeded onto glass coverslips at 60-70% confluency. pEGFP empty vector and lulu-pEGFP constructs were transfected into the cells using 1.5 μl of Lipofectamine 2000 and 0.5 μg of DNA per well in Opti-MeM. At 5 hours after transfection, the nucleotide-Lipofectamine mixture was removed and replaced with normal growth media. Immunocytochemistry was performed at approximately 24 hours after transfection.

The antigen used to produce the α-Lulu antibody, amino acids 669-731 of isoform B (Laprise et al.,2006), lies C-terminal to the stop codons in the lulu,Epb4.1l5^GT1 and Epb4.1l5^GT2 alleles. Theα-crumbs homolog 3 (α-Crb3) antibody was raised against human CRB3 and recognizes all zebrafish Crumbs proteins(Hsu et al., 2006; Makarova et al., 2003); it is therefore likely to recognize all mouse Crumbs proteins. Other antibodies:rabbit α-Lulu antibody, 1:300(Laprise et al., 2006); rabbit anti-Crb3, 1:250 (Makarova et al.,2003); rabbit anti-Sox2, 1:1000 (Chemicon); rabbit anti-phosphohistone H3, 1:200 (Upstate); rat anti-E-cadherin, 1:500 (Sigma);rabbit anti-Myosin IIB, 1:500 (Covance); rabbit anti-phospho-ERM, 1:50 (Cell Signaling); rabbit anti-laminin, 1:50 (Sigma); mouse anti-N-cadherin, 1:500(BD Biosciences); mouse anti-ZO-2 (anti-Tjp2), 1:250 (BD Biosciences); rabbit anti-Pals1 (anti-Mpp5), 1:100 (Upstate); FITC- and TRITC-phalloidin, 10 U/ml(Molecular Probes).

Cells were fixed in 4% paraformaldehyde, 30 mM sucrose in PBS for 30 minutes at room temperature, permeabilized with 0.1% Triton X-100 and blocked with 2% normal donkey serum (Jackson ImmunoResearch Laboratories, West Grove,PA). Cells were incubated with a rhodamine-conjugated phalloidin (Invitrogen,Carlsbad, CA) for 30 minutes at 37°C. Confocal images were acquired using an LSM510 microscope (Carl Zeiss MicroImaging, Thornwood, NY). Quantification was based on counting 93 cells for both the pEGFP control and the pEGFP-lulu transfections and scoring cells that showed increased phalloidin staining at the cell cortex.

The lulu mutation was identified because homozygous embryos showed striking morphological defects and arrested development at E8.5-9.0, prior to embryonic turning(García-García et al.,2005). lulu mutants had shortened trunks and lacked visible somites. The neural plate of E8.5 lulu embryos was raised in a series of transverse folds and failed to close to form the neural tube(Fig. 1A,B). Extra-embryonic tissues were also affected: the yolk sac appeared ruffled and the allantois did not fuse with the chorion (data not shown).

Mesoderm and definitive endoderm were specified, but abnormal, in lulu embryos. Analysis of molecular markers showed disruptions in axial, paraxial and cardiac mesoderm in lulu embryos. Brachyury(T) expression marks axial mesoderm and this gene was expressed discontinuously in the midline of E8.5 lulu mutants(Fig. 1E,F). Meox1,which marks somitic and presomitic mesoderm, was expressed in a greatly reduced, unsegmented domain in lulu mutants(Fig. 1G,H). Nkx2.5(also known as Nkx2-5), a marker for cardiac mesoderm, was expressed in the position of the lateral cardiac anlage; however, the lateral anlage failed to move to the midline to form a single heart tube (data not shown). In contrast to these tissues, the lateral plate mesoderm, assayed by Twist1 expression, appeared normal in lulu embryos (data not shown). Definitive endoderm, marked by the expression of Cerl (also known as Dand5) at E7.5 and later by the gut markers Hex1(Hhex) and Shh, was also specified (data not shown). Despite normal specification, the endoderm failed to move ventrally to form the gut tube.

The broad anterior tissue of lulu mutants was identified as neural ectoderm, based on its expression of the pan-neural marker Sox2(Fig. 1A,B). Despite its abnormal morphology, anterior-posterior patterning of the neural plate was normal. Krox20 (also known as Egr2 - Mouse Genome Informatics), which marks rhombomeres 3 and 5 of the hindbrain, was expressed in two parallel stripes in the neural plate(Fig. 1C,D). Similarly, the anterior markers Hesx1, Otx2, Wnt1 and En1 were expressed in the correct anterior-posterior order (data not shown). Thus, the phenotypes seen in lulu mutant embryos appeared to reflect defects in morphogenesis rather than in cell type specification.

The lulu mutation was mapped by meiotic recombination to a 650 kb interval of chromosome 1 that included the gene encoding the FERM-domain protein Epb4.1l5; we considered this to be a good candidate for lulubecause other FERM-domain proteins connect the actin cytoskeleton to the cell surface and could therefore influence morphogenetic movements(Bretscher et al., 2002). We identified a C to T transition mutation in lulu embryos that creates a premature stop codon near the C-terminal end of the FERM domain in Epb4.1l5 (Fig. 2A). Epb4.1l5 produces two alternatively spliced isoforms of 505 and 731 amino acids, which share a 444 amino acid N-terminal domain (including the entire FERM domain) but have divergent C-termini(Fig. 2A). The lulumutation lies within the FERM domain and thus disrupts both isoforms. Mouse lulu is orthologous to the zebrafish moe gene, which encodes two isoforms of the same structure as lulu(Jensen and Westerfield, 2004)(www.ensembl.org). Yurt is the most closely related FERM-domain protein in Drosophila;its FERM domain is 68% identical to the Lulu FERM domain. The Yurt sequence C-terminal to the FERM domain is unrelated to Lulu; however, Yurt and both isoforms of Lulu terminate with putative PSD-95/Dlg/ZO-1 (PDZ)-binding motifs.

We confirmed the identity of lulu in complementation tests with two independent alleles of Epb4.1l5 generated by BayGenomics(http://baygenomics.ucsf.edu). The GT1 allele fuses the first 149 amino acids of Epb4.1l5 withβ-galactosidase; the GT2 allele creates a fusion of the first 392 amino acids, which includes the complete FERM domain, withβ-galactosidase (Fig. 2A). lulu/Epb4.1l5^GT1 transheterozygous embryos recapitulated the lulu phenotype, including a reduced trunk that lacked somites and a misshapen open neural plate (Fig. 2B). No full-length protein was detected by immunofluorescence or western blotting in either lulu or Epb4.1l5^GT1homozygous embryos (Fig. 2F and data not shown), indicating that there was no productive splicing around the gene trap vector. Because the early truncation in the Epb4.1l5^GT1 allele produced the same phenotype as the lulu mutation, we infer that both are null alleles. The Epb4.1l5^GT2 allele was hypomorphic: lulu/Epb4.1l5^GT2 embryos arrested at E9.5, failed to complete embryonic turning, formed 10-12 pairs of abnormally shaped somites and partially closed the neural tube (Fig. 2C); Epb4.1l5^GT2 homozygotes survived to E11.5(data not shown). Because the data indicate that the lulu mutation inactivates the Epb4.1l5 gene, we will refer to the product of the gene as the Lulu protein, for ease of pronunciation.

lulu (Epb4.1l5) mRNA was ubiquitously expressed at E7.5 and E8.5 (data not shown). Lulu protein was broadly expressed but was apically enriched in epithelial tissues (Fig. 2D,E). The apical concentration of Lulu mirrored that of the Epb4.1l5^GT2-β-galactosidase fusion protein, which contained an intact FERM domain and was also localized to the apical surface of embryonic epithelia (Fig. 2G). At gastrulation, Lulu protein localization shifted from an apical localization in the epiblast to a position around the circumference of ingressing cells(Fig. 2D, inset).

One of the most striking aspects of the lulu phenotype was the deficit in paraxial mesoderm, seen in the dramatically reduced expression domains of Meox1 (Fig. 1G,H) and of the presomitic mesoderm marker Tbx6(Fig. 3A,B). Mesoderm generation depends on successful completion of the EMT at the primitive streak. Downregulation of E-cadherin (also known as Cdh1 - Mouse Genome Informatics) is required for this EMT and for mesoderm migration away from the streak region (Burdsal et al.,1993; Ciruna and Rossant,2001). Cells in the mesodermal layer of E7.5 lulu embryos correctly downregulated E-cadherin (Fig. 4A,B). However, transverse sections of the E8.5 streak showed that lulu embryos did have a defect in the EMT: although Tbx6 was expressed in the nascent mesoderm beneath the primitive streak in both wild-type and lulu embryos (Fig. 3C,D), some mutant cells that had delaminated from the epithelial layer of the streak were located in an abnormal bulge in the region above the Tbx6-expressing mesoderm (Fig. 3D, arrow). The cells in the primitive streak bulge expressed E-cadherin and Sox2, two markers of the epiblast(Burdsal et al., 1993; Damjanov et al., 1986),although they did not have an epithelial morphology(Fig. 3E,F). Thus, although some lulu epiblast cells could undergo the EMT, downregulate E-cadherin and generate mesoderm, other cells at the lulu primitive streak apparently initiated the EMT but retained some epiblast character and were trapped in the streak.

Gastrulation appeared to initiate normally in lulu embryos, but defects in the primitive streak could be detected in phalloidin-stained embryos by E7.75. By this stage, mesoderm had spread around the embryonic circumference in lulu embryos, as in wild type(Fig. 4C,D), but the anterior mesodermal wings were only a single cell thick, in contrast to three to four cells thick in wild-type littermates (Fig. 4E,F). This suggested that, by E7.75, the number of mesodermal cells was already reduced, but that mutant mesodermal cells that escaped the streak region could migrate effectively. The ability of mutant mesoderm cells to migrate was confirmed in E7.5 primitive streak explants, where lulu mutant mesoderm cells migrated efficiently away from the streak(data not shown).

The primitive streak of E7.75 lulu embryos was thicker and wider than the wild-type streak (Fig. 4C,D,G,H). Laminin marks the basement membrane that separates the epiblast and mesoderm, and there is a gap in laminin staining at the position of the wild-type primitive streak (Fig. 4I, bracket). In lulu mutants, the gap in laminin expression was wider than in the wild type(Fig. 4J, bracket). Laminin staining also highlighted the increased thickness of the E7.75 streak in lulu mutants compared with wild type(Fig. 4I,J, arrows), similar to the increased thickness that was seen at E8.5(Fig. 3D).

It has been shown that mouse Lulu can bind all three mammalian Crumbs proteins (CRB1-CRB3), and that its Drosophila homolog is required for normal apical localization of Drosophila Crumbs(Laprise et al., 2006). Therefore, it was possible that the EMT defect in lulu might represent a disruption of apical-basal polarity. We examined the expression of Crumbs proteins at the primitive streak using a pan-Crumbs antibody(Hsu et al., 2006; Makarova et al., 2003). In both wild-type and lulu embryos, Crumbs proteins were expressed apically in the epiblast at E7.5, and were absent from nascent mesoderm cells(Fig. 4K,L). Crumbs proteins were not detected in the cells that accumulated in the abnormal bulge at the lulu streak, which indicated that these cells had lost this characteristic of the epiblast. These results show that Lulu is required neither for apical localization of Crumbs in the epiblast nor for the downregulation of Crumbs in nascent mesoderm.

Phalloidin staining of E7.5 embryos revealed defects in actin organization in the mutant primitive streak. In the wild-type primitive streak, F-actin was enriched around the periphery of cells that were ingressing through the streak(Fig. 4G, arrowheads), in a pattern remarkably similar to the distribution of Lulu protein(Fig. 2D). By contrast, the circumferential F-actin was not present in cells of the lulu streak;instead, puncta of F-actin were prominent in the cells that accumulated in the lulu streak at E7.5 (Fig. 4H, arrowheads) and E8.5 (data not shown). Because some FERM-domain proteins bind actin (Bretscher et al., 2002), this abnormal distribution of F-actin could represent the primary defect that prevents the normal EMT in the luluprimitive streak.

The other tissue with particularly striking defects in luluembryos was the neural plate. The anterior neural plate of luluembryos was broad, open and folded into irregular ridges(Fig. 1B,D), and it appeared thicker than the wild-type neural plate(Fig. 5A,B). To evaluate whether the apparently large size of the neural plate could reflect excessive proliferation in that tissue, we counted the total number of cells in the anterior neural plate. Sox2 is expressed in the pseudo-stratified neural ectoderm and the columnar gut endoderm, whereas E-cadherin is expressed in the gut endoderm and the cuboidal surface ectoderm(Fig. 5A,B); therefore neural cells were defined as Sox2-positive and E-cadherin-negative (Sox2⁺,E-cadherin^-). The total number of Sox2⁺,E-cadherin^- cells in the anterior lulu neural plate was approximately the same as in wild-type littermates(Table 1). We also assayed the fraction of neural cells in mitosis, based on staining with anti-phospho-histone H3 antibodies (Fig. 5C,D and Table 1). The mitotic index in the lulu anterior neural plate was indistinguishable from that of wild type(Table 1). We conclude that the abnormal shape of the neural plate in lulu embryos is not caused by abnormal proliferation or by an increase in the number of neural cells;instead, the abnormal shape represents a defect in the organization of the neural tissue.

The nuclei of lulu neural cells appeared to be rounder than wild type (Fig. 5A,B). Nuclear shape varies with cell shape, which is a function of the cytoskeleton. Because of the dense cell packing within the pseudo-stratified neuroepithelium, we were unable to examine cell shape directly by using cell surface labeling. We therefore examined nuclear morphology as an indicator of cell shape, using DAPI staining of the nuclei and 3D reconstructions of confocal z-stacks through the neural plates of wild-type and luluembryos. Whereas wild-type neural plate nuclei were ellipsoid and elongated along the apical-basal axis, with an average length:width ratio of 2.33(Fig. 5E, Table 1), lulu neural plate nuclei were more spherical, with an average length:width ratio of 1.44,and many of the ellipsoid nuclei did not align their long axes with the apical-basal axis (Fig. 5F, Table 1). Thus, there appears to be a global distortion of nuclear shape and orientation within the lulu neural plate.

Although the mitotic index in the lulu neural plate was normal,phospho-histone H3 staining revealed another defect in the organization of the neural epithelium (Fig. 5C,D). Nuclei in columnar and pseudo-stratified epithelia, such as the neural plate,migrate along the apical-basal axis of the cell during the cell cycle, and mitosis takes place when nuclei are at the apical surface(Götz and Huttner, 2005). We observed that 20% of phospho-histone H3-positive nuclei in the lulu neuroepithelium were located away from the apical surface(Fig. 5D, arrows), whereas only 5% of phospho-histone H3-positive nuclei were located away from the apical surface in the wild-type neural plate(Table 1).

It was possible that the abnormally positioned mitoses in the luluneural plate reflected a defect in the apical-basal polarity of the epithelium, which depends on the proper organization of apical junctions(Imai et al., 2006; Koike et al., 2005; Wei and Malicki, 2002). We therefore examined the organization of tight junctions and adherens junctions in the lulu neural plate. We found that the tight junction transmembrane proteins occludin and claudin 1, as well as the tight junction-associated proteins ZO-1 (Tjp1) and ZO-2 (Tjp2), were expressed and apically localized in lulu mutant neural plates(Fig. 6A,B and data not shown). N-cadherin, the major cadherin present in the neuroepithelium(Radice et al., 1997), and its associated protein, β-catenin, were also enriched in the apical neural plate of both wild-type and lulu E8.5 embryos(Fig. 6C,D and data not shown).

Because Lulu can bind Crumbs proteins(Hsu et al., 2006; Jensen and Westerfield, 2004; Laprise et al., 2006; Omori and Malicki, 2006), we tested whether components of the Crumbs complex were localized normally in the lulu neural plate. Crumbs proteins were detected at the apical surface of the neuroepithelium in wild-type, lulu and Epb4.1l5^GT1 homozygous embryos(Fig. 6E,F and data not shown). Pals1 (also known as Mpp5 - Mouse Genome Informatics), a MAGUK protein that binds Crumbs and the zebrafish Lulu ortholog, Moe(Hsu et al., 2006), was expressed on the apical surface, as well as in part of the lateral surface between adjacent cells, in both the wild-type and the lulu neural plate (Fig. 6G,H). The Par6-Par3-aPKC (aPKC is also known as Prkci - Mouse Genome Informatics)complex interacts with the Crumbs complex(Hurd et al., 2003; Lemmers et al., 2004); we found that Par3 was present apically in both the wild-type and luluneural plate (data not shown). Thus, these markers of apical-basal polarity appeared to be localized correctly in the lulu neural plate.

Because many previously studied FERM proteins bind actin(Bretscher et al., 2000; Tsukita and Yonemura, 1999)and because we observed ectopic F-actin in the lulu primitive streak,we examined the organization of F-actin in the lulu neural epithelium. In wild-type E8.5 neural plate, phalloidin staining showed that F-actin formed a dense band near the apical surface of each cell(Fig. 7A,C). By contrast,although F-actin was enriched apically in the lulu neural plate, it was also present ectopically at more basal positions(Fig. 7B,D, arrowheads). Myosin IIB is a major non-muscle myosin and participates in apical constriction of the actin ring, an essential step during neural tube closure(Haigo et al., 2003; Hildebrand, 2005). Myosin IIB was highly enriched at the apical surface of the wild-type neural plate(Fig. 7E). By contrast, Myosin IIB was seen both apically and in ectopic basal positions in the luluneural plate (Fig. 7F,arrowheads). Spnb2 (βII-spectrin), an F-actin-binding protein(Liu et al., 1987), showed similar ectopic staining within the neuroepithelium (data not shown). Phospho-ERM (P-ERM) antibodies, which recognize the activated, F-actin-binding forms of ezrin, radixin and moesin (Gary and Bretscher, 1995; Matsui et al., 1998), strongly labeled the apical surface of the wild-type neural plate (Fig 7G),consistent with previous reports of their apical localization in other epithelia (Berryman et al.,1993; Morales et al.,2004). By contrast, P-ERM was found both apically and more basally in the neuroepithelium of lulu mutants(Fig. 7H), which again indicated the presence of ectopic F-actin at basal positions in the lulu neural epithelium.

To investigate whether Lulu might regulate the actin cytoskeleton directly,we expressed EGFP-tagged Lulu protein in HeLa cells. EGFP-Lulu was concentrated at the plasma membrane and we observed increased phalloidin staining at the cell cortex in 66% of the transfected cells, whereas only 30%of EGFP-control transfected cells had phalloidin staining concentrated at the cell periphery (Fig. 8). In addition, compared with the transfected control cells, the HeLa cells that overexpressed EGFP-Lulu appeared more spread out and showed increased numbers of small actin-rich protrusions. Thus, overexpression of Lulu was sufficient to increase cortical actin and alter the morphology of these epithelial cells.

The results presented here demonstrate that the mouse FERM-domain protein Lulu has a central role in the morphogenesis of the early mouse embryo. Embryos that lack Lulu die before mid-gestation, with a syndrome of developmental defects that affects all three germ layers. The luluphenotype is much more severe than that of its zebrafish homolog, moe: null moe mutations allow survival to the end of embryogenesis, and the only defects that have been reported in moemutants are in retinal lamination and in brain ventricle size(Jensen et al., 2001; Jensen and Westerfield, 2004). The milder moe phenotype could reflect the large contribution of maternal gene products to zebrafish development, although translation-blocking morpholinos recapitulate the moe mutations(Jensen and Westerfield,2004). Alternatively, the cellular behaviors required for gastrulation and neural tube formation are distinct in zebrafish and mouse(Adams and Kimmel, 2004; Geldmacher-Voss et al., 2003)and may depend on different molecular machines.

The absence of Lulu leads to dramatic defects in the production and morphogenesis of paraxial mesoderm, including the complete absence of somites. Some mesoderm forms in the absence of Lulu, but fewer cells are present in the mesodermal wings as soon as they surround the embryo. These defects are the consequence of the abnormal organization of the primitive streak, where cells begin to accumulate as early as E7.5.

During the EMT at the primitive streak, cells must lose epithelial cell junctions, escape the epithelial layer and acquire the properties of mesenchymal cells (Shook and Keller,2003). These morphological transitions are accompanied by changes in gene expression, as cells downregulate expression of epithelial genes, such as Sox2 and E-cadherin, and upregulate mesodermal genes, such as Tbx6. The early steps of the EMT (breakdown of the basement membrane,loss of cell-cell junctions and downregulation of Crumbs) do not require Lulu. However, in the absence of Lulu, many cells accumulate at the primitive streak. Although these abnormal streak cells have lost their epithelial organization, they continue to express the epithelial markers Sox2 and E-cadherin, and do not express the mesodermal marker Tbx6. Thus,these lulu mutant cells appear to be trapped at an intermediate state in the EMT.

Mutations in Fgfr1, Snail (also known as Snai1) and p38IP (also known as D3Ertd300e) all block the gastrulation EMT and,in each case, the defect has been attributed to the inability to downregulate expression of E-cadherin (Ciruna and Rossant, 2001; Carver et al.,2001; Zohn et al.,2006). By contrast, although E-cadherin is expressed in the cells trapped at the lulu primitive streak, E-cadherin is downregulated in the mesodermal wings in lulu mutant embryos(Fig. 4B). This suggests that the lulu phenotype defines a previously unrecognized step of the EMT that requires a Lulu-dependent reorganization of the actin cytoskeleton. The change in Lulu protein localization during the EMT suggests that it plays an active role in the EMT: Lulu is enriched at the apical surface of the epiblast and then relocalizes to the periphery of ingressing cells. This change in localization parallels the rearrangements seen in F-actin, which is also apically localized in the epiblast and surrounds the ingressing cells in the streak. In lulu mutants, this rearrangement of F-actin fails and,instead, cells in the mutant streak show ectopic foci of F-actin. We therefore propose that Lulu helps anchor F-actin to the surface of ingressing cells and that the architecture of the cells at this transition state is important for the changes in adhesion and motility that allow these cells to acquire a mesenchymal character. Some FERM domains have been shown to bind F-actin(Bretscher et al., 2002), and overexpressed Lulu is sufficient to reorganize the morphology and actin cytoskeleton in HeLa cells (Fig. 8). We therefore suggest that Lulu regulates the F-actin cytoskeleton during the EMT, either via the direct binding of actin or via an intermediary protein.

The second striking defect in lulu mutant embryos is that the anterior neural plate is broader and thicker than wild type; the irregular folding pattern in lulu mutants is strikingly different than the median and dorsolateral folds that lead to wild-type neural tube closure. These defects are not due to differences in cell proliferation or cell number,nor are they due to a loss of the apical junctions that are a hallmark of the apical domain of epithelial cells. Instead, we observe that the abnormal morphology of the neural plate is coupled to defects in organization of the apical actin network: F-actin and F-actin-binding proteins are present at basal positions in the epithelium of the anterior neural plate, in addition to their normal apical location.

Mouse Lulu, like its Drosophila and zebrafish homologs, binds Crumbs proteins, which are key determinants of the apical domain of epithelial cells (Laprise et al., 2006). Recent work in both Drosophila and zebrafish has indicated that the function of these Lulu homologs is to regulate the size of the apical domain of epithelial cells via regulation of either the localization or the activity of Crumbs (Hsu et al., 2006; Laprise et al., 2006). The effect of the loss of Drosophila Yurt varies between tissues: in the ventral ectoderm, the domain of Crumbs expression is expanded in yurtmutants, while, in yurt mutant photoreceptors, the apical domain is expanded in a Crumbs-dependent fashion without a change in the domain of Crumbs expression (Laprise et al.,2006). Loss of zebrafish moe function prevents tight junctions from forming in the retina(Jensen and Westerfield,2004), possibly due to loss of apical Crumbs proteins in the moe retina (Hsu et al.,2006). In contrast to these results, we find that loss of Lulu does not affect Crumbs localization or apical junction formation(Fig. 6), and en-face imaging did not reveal differences in the size of the apical domain of mutant cells in the neural epithelium (data not shown).

The lulu neural plate does not have detectable defects in apical junctions, but does show a clear disruption in the organization of the apical F-actin network. One possible explanation for these findings is that the loss of Lulu leads to a partial disruption in apical-basal polarity, so that some cells lose their connections to the apical surface of the pseudo-stratified neural epithelium and the F-actin network associated with these cells appears as ectopic, basal F-actin. However, the disruption of cellular organization in the neural epithelium does not appear to be limited to a sub-population of neural cells. For example, P-ERM, which marks apical F-actin, is ectopically localized throughout the mutant neural epithelium. Similarly, the nuclei throughout the mutant neural epithelium are less elongated than wild type and are not correctly oriented with respect to the apical-basal axis of epithelium(Fig. 5F, Table 1), which is likely to reflect a general change in cellular architecture. Thus, we conclude that there is a global disruption of cell organization in the lulu neural plate.

We suggest that Lulu is required to link the F-actin cytoskeleton to plasma-membrane protein complexes that are important during the dynamic cell rearrangements that take place during both the EMT at gastrulation and the folding of the neural plate. One model that reconciles the demonstrated ability of Crumbs to bind Lulu with our findings would be that mouse Crumbs acts upstream of Lulu, and Lulu helps to link Crumbs to the actin cytoskeleton. Alternatively, Lulu may act independently of Crumbs to anchor the F-actin cytoskeleton as it rearranges or contracts. In the future,characterization of the functions of the mouse Crumbs complex proteins and of genes that produce lulu-like phenotypes(García-García et al.,2005) should distinguish among these possibilities.

We thank the Sloan-Kettering transgenic facility for production of the lulu gene trap mice, and the Sloan-Kettering molecular cytology core facility for expert technical assistance during confocal imaging. We thank B. Margolis for the Crb3 antibody; and M. Baylies, J. Zallen, S. Shi and the Anderson laboratory for comments on the manuscript. Epb4.1l5^GT1 and Epb4.1l5^GT2 ES cells were obtained from BayGenomics. Genome sequence analysis used ENSEMBL and the Celera Discovery System and associated databases. J.D.L. was supported by an NRSA, and N.F.S-G by a postdoctoral fellowship from the Foundation Fighting Blindness Canada. The work was supported by the National Institutes of Health grant HD35455 to K.V.A. and a grant from the Foundation Fighting Blindness Canada to C.J.M.