Mosaic Eyes is a novel component of the Crumbs complex and negatively regulates photoreceptor apical size

Drosophila Crumbs (Crb) and vertebrate Crumbs orthologs are important regulators of epithelial morphology and apical polarity(Wodarz et al., 1995; Tepass et al., 1990; Roh et al., 2003) and are important for normal photoreceptor morphogenesis and zonula adherens junction formation and/or maintenance in Drosophila(Pellikka et al., 2002; Izaddoost et al., 2002; Mehalow et al., 2003; van de Pavert et al., 2004). Mutations in human Crumbs Homolog 1 (CRB1) are associated with the inherited retinal diseases retinitis pigmentosa 12 and Leber's congenital amaurosis(LCA) and retinal lamination defects have been observed in LCA(den Hollander et al., 1999; den Hollander et al., 2001; Lotery et al., 2001; Gerber et al., 2002; Jacobson et al., 2003). Deficiencies in mouse Crb1 are also associated with retinal abnormalities(Mehalow et al., 2003; van de Pavert et al., 2004),but these are less severe than those observed in humans, and photoreceptor development appears largely normal under normal light conditions. Loss-of-crb function in Drosophila photoreceptors causes defects in zonula adherens formation, stalk length and rhabdomere morphology, but polarity appears normal (Izaddoost et al.,2002; Pellikka et al.,2002).

The small intracellular domains of Crumbs and vertebrate Crumbs orthologs are highly conserved and encode two important protein interaction domains, a predicted FERM-(Band 4.1, Ezrin, Radixin, Moesin) binding domain and a PDZ-binding domain (Wodarz et al.,1995; den Hollander et al.,2000; Izaddoost et al.,2002; Roh et al.,2003). The orthologous Maguk proteins, Drosophila Stardust and mammalian Pals1, have been shown to bind to the PDZ-binding domain in Drosophila Crb and mammalian CRB1 and CRB3(Bachmann et al., 2001; Hong et al., 2001; Kantardzhieva et al., 2005; Roh et al., 2002). The FERM protein, DMoesin, has been shown to be in a complex with Crb and shows some subcellular overlap with Crb in embryonic epithelia(Medina et al., 2002), but a direct interaction has not been demonstrated.

We previously reported that zebrafish mosaic eyes (moe),which encodes a FERM protein, is required for normal retinal lamination and suggested that it might interact with the predicted FERM-binding domain in Crumbs proteins (Jensen et al.,2001; Jensen and Westerfield,2004). The phenotype of moe mutants is similar to the phenotype of mutants in nagie oko (nok) (mpp5 -Zebrafish Information Network), which encodes Pals1(Wei and Malicki, 2002), and mutants deficient in both loci are indistinguishable from the single mutants,suggesting a genetic interaction between the two genes(Jensen and Westerfield,2004). In this study we show that Moe interacts directly with Crb proteins and Nok (Pals1), and also forms a complex with Has (aPKC λ;Prkci - Zebrafish Information Network). Morpholino knockdown of one of the zebrafish Crumbs orthologs, crb2a, phenocopies the moemutations. Finally, we show that moe is required for normal photoreceptor morphology; the apical domain is expanded in rod photoreceptors that lack moe function, suggesting that Moe may negatively regulate Crumbs protein function.

AB wild-type strain, nok^m520, has^m567moe^b476, moe^b781 and moe^b882alleles were maintained and staged as previously described(Jensen et al., 2001). The moe^b476 and moe^b781 alleles were crossed into the Rhodopsin-GFP (Xop-GFP) transgenic line(Fadool, 2003).

UV-opsin GFP transgenic fish were generated by cloning a 7 kb SacI fragment from a UV-opsin⁺ PAC clone, including ∼870 bases of the coding region plus 6 kb upstream, into the pG1/pESG GFP vector to generate a fusion protein that includes most of the UV opsin protein with GFP at the C-terminus. Linearized plasmid (60 ngμl) was injected into single-cell embryos; germline founders were identified by PCR.

Whole-mount and section in situ hybridizations were performed as previously described (Jensen and Westerfield,2004). cDNAs were linearized and transcribed: crb2b XhoI,SP6; crb2a NotI, SP6; mouse ymo1 BglII, T3; mouse crb1 (Accession number BM941539) Not1, T3; mouse crb2 (Accession number BI738283) EcoRI, T7.

crb2a splice-blocking morpholinos were targeted to the intron(∼810 bases) between the second and third coding exons. Donor morpholino sequence was TTGCACTTCAATTACCTGTATATCC and acceptor was ACAGTTTACACCTACAGAGATCACA. Injection of donor morpholino (200 μmol/l)produced no discernable abnormality, while injection of acceptor morpholino(200 μmol/l) produced a weak moe-like phenotype in about 50% of injected embryos. Co-injection of both morpholinos (each 200 μmol/l)produced a moe-like phenotype in all injected wild-type embryos. Morpholino activity was tested by RT-PCR on RNA isolated from 60 30-hpf injected embryos. Primers used for RT-PCR were: RT, GCGGTCGTGGCAAAGTC); PCR,GGCGAGACCTGTGAAGAAGACC and CCGTTTTGACAGGGATTTGACTC. Co-injection of nok splice-blocking morpholinos (each 200 μmol/l; donor,GTTTATGACACCCACCTAGTAAAGC; acceptor, CTCCAGCTCTGAAAGTACAAACACA) produced a phenotype indistinguishable from nok mutants (not shown).

GST-tagged Crb^intra proteins: sequences encoding the intracellular domains of Crb proteins plus variable amounts of transmembrane domain were cloned into pGEX-4T-1 (Amersham); Crb1, Crb2a, Crb2b, Crb3a and Crb3b are 40, 42, 42, 61 and 65 amino acids, respectively. A Moe fragment (EST accession number CD750925) encoding amino acids 1-434 was cloned into pRSET-A to make His-Moe_FERM. For MBP Moe_FERM and MBP-Moe_C-terminus sequence encoding residues 59-383 and sequence encoding residues 383-772, respectively,was cloned into pMal C2X (NEB). To make His-Nok proteins, full-length Nok and a fragment encoding the first 411 amino acids (Nok-N, including the PDZ domain) were cloned into pRSET-A or -B (Invitrogen). To make GST-Nok-Int, a 468 bp internal StuI-PstI fragment of Nok encoding the predicted Band 4.1 interacting region was cloned into pGEX4T-1.

Polyclonal antibodies to Moe were generated by immunizing rabbits (UMASS antibody facility) and guinea pigs (Rockland Immunochemicals) with purified His-Moe_C term fusion protein (amino acids 383-772, Accession number CD750925).

About 190 adult zebrafish eyes were homogenized with 5 ml cold IP lysis buffer, incubated for 1 hour at 4°C, centrifuged, and supernatant collected. For each reaction, 500 μl lysate was pre-cleared with 20 μl normal rabbit serum and 50 μl protein A/G Plus-Agarose (Santa Cruz Biotech). Pre-cleared lysate was collected by centrifugation and incubated with 20 μl normal rabbit serum or 10 μl anti-CRB3 antibodies at 4°C overnight. Twenty-five microliters of protein A/G Plus-Agarose and 300 μg BSA were added subsequently to capture the immunocomplex and incubated for 2 hours at 4°C. Resin was washed six times with lysis buffer. Coimmunoprecipitated proteins were eluted with reducing sample buffer and analyzed by western blotting.

Western blot analysis was performed on whole embryo or larval lysates,affinity-purified proteins or purified fusion proteins. Proteins were resolved by SDS-PAGE and transferred to nitrocellulose membrane, blocked in blotto,incubated overnight in primary antibody at 4°C; anti-CRB3 (1:1500),anti-GST-HRP (1:9000-1:20000; Amersham), rabbit anti-Moe (1:1500), guinea pig anti-Moe (1:750), anti-Nok (1:2500), anti-PKCζ (1:1000; Santa Cruz Biotech, C-20), mouse anti-α-tubulin (1:2000; 12G10 DSHB), washed,incubated in anti-rabbit-HRP (1:70000-100000; Pierce), anti-guinea pig-HRP(1:15000, Jackson), or anti-mouse-HRP (1:20000; Pierce), washed, detected with SuperSignal West Dura (Pierce) or West Pico (Pierce) substrate and exposed to X-ray film.

One microgram of bait fusion protein was resolved on SDS-PAGE, transferred to nitrocellulose, blocked with blotto and incubated with interacting fusion protein (2.5 μg/ml) in 3% milk/PBSTw at 4°C overnight. Bound fusion protein was detected by using anti-GST or anti-MBP (1:10000; NEB) or anti-Omni(1:1000; Santa Cruz Biotech). Immunoreactivity was revealed as above.

About 1000 3d zebrafish were homogenized in 1 ml cold lysis buffer plus protease inhibitors, centrifuged, supernatant isolated, and TritonX-100 added to 1%. One milligram of purified His-Moe_FERM or His-lacZ was immobilized on Ni-NTA resin (Qiagen), washed, incubated with 1 ml lysate for 1 hour at 4°C, resin was washed with 10 ml HEPES column buffer (with and without 1% TritonX-100) and proteins eluted with reducing sample buffer, 20-30μl eluant (total eluant volume, 700 μl), resolved by SDS-PAGE, 0.5-1.5%of the lysate was included on the gel and analyzed by western blot. In parallel, 500 3d Nok morpholino-injected zebrafish were homogenized and 200μg His-Moe_FERM was used for pulldown as described above. Five hundred microliters of adult eye lysate was incubated with either 200 μg immobilized His-Moe_FERM or His-lacZ overnight at 4°C. Resin was washed and proteins eluted with 90 μl sample buffer.

For Moe-Crb protein interactions, 10 μg His-Moe_FERM was bound to NHS resin (Amersham). Residual reactive sites were blocked with 100 mmol/l ethanolamine. The resin was incubated with 10 μg GST-Crb1, GST-Crb2a or Crb2b for 1 hour at 4°C. Bound proteins were eluted with sample buffer. For Moe-Nok interactions, 10 μg GST or GST Nok-fusion proteins were bound to GST microspin columns (Amersham), washed and incubated with 20 μg BSA in PBS for 2 hours at 4°C, incubated overnight at 4°C with the GST-tagged protein in PBS (plus 20 μg BSA). Columns were washed, protein glutathione-eluted and analyzed by western blot with tag antibodies.

Zebrafish were fixed in 4% PFA, and sections (18 or 30 μm) permeabilized with 0.1% SDS for 15 minutes, washed in PBSTw, and incubated in blocking solution then with primary antibody overnight at 4°C: rabbit anti-Moe 1:1000; guinea pig anti-Moe 1:500; anti-CRB3 1:400; anti-Nok 1:400;anti-aPKCζ 1:1000; mouse anti-ZO-1 1:10; rabbit anti-GFP 1:300 (Molecular Probes); mouse anti-Rhodopsin 1:100; mouse Zn5 1:5 (ZIRC); anti-phospho-H3 1:1000 (Upstate Biotech). Sections were washed, incubated with secondary antibody [-FITC/-TRITC (Molecular Probes) 1:200, -CY5 1:100 (Jackson)] and analyzed with a Zeiss LSM 510 Meta Confocal System. Confocal images are a single scan (four to eight averaged), optical thickness ∼1 μm. Images in Fig. 2 were taken with the same settings.

Donor embryos were produced by incrossing Rhodopsin-GFP;moe^b781/+ fish. At the high stage to dome stage, ∼20-40 cells were transplanted from labeled-donor embryos into wild-type host embryos (Jensen et al.,2001). Larvae were placed in the dark for 2 hours and were fixed and sectioned (30 μm) as described above. Confocal z-stack images (∼1μm optical thickness) were acquired every 0.38 μm. Cell volume was approximated by accumulating the area of the cell on each z-section: cell area was outlined and measured in each z-section and for each cell the area from the z-sections was summed to represent volume. Only cells completely captured in the z-stack were included in the calculations. Measurements were repeated and averaged to reduce sampling error. Donor cells were not lineage traced with rhodamine dextran for the experiments localizing panCrb and ZO-1 in GFP⁺ rods so that rhodamine could be used for antibody localization.

There are three vertebrate Crumbs orthologs, Crb1, Crb2 and Crb3(den Hollander et al., 1999; van den Hurk et al., 2005; Makarova et al., 2003). In order to test our hypothesis that Moe forms a complex with Crumbs proteins, we had to clone zebrafish crumbs genes. We used sequences from the zebrafish genome project to clone crb1, crb2a, crb2b and crb3b, and found an EST for crb3a. These are the same crumbs genes recently described by Omori and Malicki(Omori and Malicki, 2006). To determine whether any of these crumbs orthologs are coexpressed in the same cells and tissues as moe, we performed in situ hybridization in embryos and eyes. At 30 hours post-fertilization (hpf), moe and crb2a were expressed in the skin, nose and epiphysis, and ubiquitously in the brain and eye; crb2b was very weak in the brain but highly expressed in the epiphysis and nose(Fig. 1A). At 7 days post-fertilization (dpf), moe and crb2a were expressed in photoreceptors, cells in the inner nuclear layer, proliferating cells in the peripheral margin, and cells in the anterior chamber(Fig. 1B); crb2b was detected in a small number of newly differentiating photoreceptors in the periphery and cells in the anterior chamber(Fig. 1B). At 3 dpf, we observed high expression of crb2b in newly differentiating photoreceptors and no expression of crb1, crb3a or crb3b in the retina (data not shown) like that observed by Omori and Malicki(Omori and Malicki, 2006). In adults, moe and crb2a were expressed in photoreceptors and cells in the inner nuclear layer (Fig. 1C).

We also examined the expression of the mouse ortholog of moe, ymo1(Drosophila yurt and zebrafish moe like 1; annotated Erythrocyte Protein Band 4.1 Like 5 in the human genome), along with crb1 and crb2 in the adult mouse retina. We found that ymo1 was expressed in the same population of cells as crb1and crb2 and was expressed in photoreceptors and a subset of cells in the inner nuclear layer (Fig. 1D), similar to that expression previously shown by den Hollander and colleagues (den Hollander et al.,2002) and van den Hurk and colleagues(van den Hurk et al., 2005),although we did not detect crb2 expression in the retinal ganglion layer or in bipolar cells like that reported by van den Hurk and colleagues(van den Hurk et al., 2005). Coexpression in the same populations of cells of moe and crb2a/b genes in zebrafish and the orthologous genes in mouse is consistent with the idea that the genes/proteins interact.

To test whether loss-of-function of zebrafish crumbs orthologs would phenocopy the moe mutations, we targeted knockdown of crb2a gene function, because its expression most closely resembles moe expression. Injection of crb2a splice-blocking morpholinos into one to two-cell wild-type embryos (hereafter called crb2a morphants) inhibited splicing as tested by RT-PCR (see Fig. S1A in the supplementary material) and produced a phenotype that is indistinguishable from moe mutants, including reduced brain ventricles, edema of the heart cavity and patchy retinal pigmented epithelium(see Fig. S1C,D in the supplementary material). The similarity in phenotype between crb2a morphants and moe mutants extended to loss of retinal lamination and the ectopic localization of retinal ganglion cells and rod photoreceptors (see Fig. S1E-P in the supplementary material). crb2a morphants died around 5-6 dpf, similar to moe mutants. The observation that moe and crb2a loss of function cause similar phenotypes further supports the idea that moe and crumbs genes/proteins interact. Omori and Malicki(Omori and Malicki, 2006)reported recently that the ome locus is crb2a and our ome morphant phenotype was like their ome morphants.

If Moe forms a complex in vivo with Crb proteins, the proteins should show some colocalization. To examine the localization of Moe protein, we generated Moe polyclonal antibodies and showed by western blot that a protein of∼110 kDa was recognized by the antibody in wild type and was absent in all moe mutant alleles (Fig. 2A). The molecular weight of Moe protein is larger than predicted(∼89 kDa), suggesting that it may be modified post-translationally. To examine the localization of Crumbs proteins, we determined by western blot that the antibody raised against the intracellular domain of human CRB3(Makarova et al., 2003)recognized the recombinant intracellular domains of all the zebrafish Crumbs proteins we identified (Fig. 2B), and thus anti-CRB3 should be considered a pan-Crb antibody in zebrafish, and hereafter is referred to as anti-panCrb.

We examined the localization of Moe and panCrb in wild-type embryos and whether their localization depends on the function of moe, nok(pals1), crb2a/ome and has (aPKCλ). We included has mutants(Horne-Badovinac et al., 2001)because its phenotype is similar to moe, ome and nok mutants and crb2a morphants. In addition, the Par complex, which includes aPKCλ, interacts with Crumbs complexes(Wodarz et al., 2000; Hurd et al., 2003; Lemmers et al., 2004; Nam and Choi, 2003), and in Drosophila, Crumbs is a target of DaPKC(Sotillos et al., 2004). In 30 hpf wild-type embryos, both Moe and panCrb were concentrated at the apical surface of the brain and retinal neuroepithelium; Moe also localized cortically in all neuroepithelial cells(Fig. 2C,D). In moemutants, Moe was lost, and apical panCrb in the brain and retinal neuroepithelium was lost (Fig. 2E,F). In crb2a morphants, ome and nokmutants, apical Moe and panCrb were lost, although in nok mutants disorganized plaques of Moe labeling were observed near the brain midline(Fig. 2G-J,M,N). In has mutants, apical Moe and panCrb in the brain and retinal neuroepithelium were reduced and patchy compared with wild-type embryos(Fig. 2K,L). Double labeling and colocalization of Moe and panCrb is provided in Fig. S2 in the supplementary material. Both Moe and panCrb localize to the apical neuroepithelium and their localization is co-dependent, further supporting the idea that the proteins might interact.

Loss of panCrb labeling at the apical surface in moe, nok and ome mutants could be due to loss of Crb proteins; or, alternatively,Crb proteins may be present but no longer localized apically and instead are diffuse. To distinguish between these two possibilities, we performed western blot analysis of 30 hpf wild-type, moe, nok and ome mutants. In wild-type embryos, a protein of about 150 kDa was recognized by anti-panCrb antibody. The predicted molecular weight of Crb2a is about 156 kDa, and it is by far the most abundantly expressed crumbs gene at 30 hpf; thus this protein is probably Crb2a. Levels of this protein were unaffected in moe^b781 mutants (and moe^b476deficiency, not shown), barely detectable in crb2a morphants and nok mutants, was absent in ome mutants(Fig. 2O) and moderately reduced in has mutants (data not shown). Expression of crb2aand crb2b mRNA by in situ hybridization was unaffected in crb2a morphants, moe, nok and ome mutants (data not shown). These results suggest that Moe is required either for apical localization (or retention) or for trafficking of Crumbs proteins.

Because Crumbs proteins are important for retinal development, we examined whether Moe colocalizes with panCrb in the retina as well as with Nok (Pals1)and Has (aPKCλ). At 7 dpf, Moe localized to the photoreceptor layer,retinal progenitors in the periphery and cells in the anterior chamber(Fig. 3A-F), while panCrb localized to the photoreceptor layer apical to the outer limiting membrane(OLM) and the apical surface of progenitor cells in the periphery(Fig. 3A). Merging the Moe and panCrb images shows that anti-Moe and anti-panCrb strongly colocalized in newly differentiating photoreceptors (Fig. 3A, Moe+anti-panCrb, bracket) at the periphery of the photoreceptor layer. Higher magnification of Moe and panCrb colocalization in the photoreceptor layer was shown at 4 dpf, a time when most photoreceptors are forming outer segments (Fig. 3B). Moe also colocalized with Nok and aPKCλ in the photoreceptor layer and with aPKCλ in the outer plexiform layer(Fig. 3C-F). Colocalization of Moe with panCrb, Nok and aPKCλ places these proteins in a position to potentially interact.

We also examined the localization of Moe and panCrb relative to markers for photoreceptors and Müller glia, which send processes into the photoreceptor layer. Moe and Crb proteins appeared to be in all photoreceptor types examined, and Moe was in Müller processes that project into the photoreceptor layer (Fig. 4A-G). We also examined Moe and panCrb localization in the photoreceptor region relative to the OLM, a specialized adherens junction between photoreceptor cells and Müller glia and between individual Müller cells and individual photoreceptors(Williams et al., 1990). In mouse retina, Crb1 localizes just apical to OLM and deficiencies in Crb1 result in OLM defects (Mehalow et al.,2003; van de Pavert et al.,2004). The highest level of Moe labeling was basal to the OLM, as marked by anti-ZO-1, but anti-Moe labeling was also observed apical to the OLM, where panCrb labeling localizes (Fig. 4H,I).

Because Crumbs proteins contain a predicted FERM-binding domain and Moe is a FERM protein, we sought to determine whether Moe and Crumbs proteins physically interact. We found that anti-panCrb (anti-CRB3) antibodies co-immunoprecipitated proteins recognized by anti-Moe (∼110 kD),anti-panCrb (∼150 kDa) and anti-Nok (∼80 kDa) from adult eyes(Fig. 5A). We also found that a fusion protein of Moe that includes the FERM domain (Moe_FERM) pulled down a∼150 kDa protein from larval lysates that was recognized by anti-panCrb antibody (Fig. 5B), further supporting the idea that Moe and Crumbs proteins form a complex. We also showed that Moe_FERM can pull down a ∼80 kDa protein recognized by anti-Nok from 3 dpf wild-type larvae and adult eyes but not from nokmorphants (Fig. 5C,E). Further,we showed by western blot that this ∼80 kDa protein was absent in nok^m520 (Fig. 5D), indicating that this protein is encoded by the noklocus.

Given that Moe and aPKCλ colocalized in tissue, we also tested whether aPKCλ (Has) forms a complex with Moe. Moe_FERM pulls down a protein of about 72 kDa that was recognized by antibody to the highly related protein aPKCλ (Fig. 5F). The ∼72 kDa protein recognized by anti-aPKCλ was absent in has^m576 (Fig. 5G), suggesting that this protein is encoded by the haslocus. A protein of about 52 kDa was also pulled down, but it is unclear whether this represents an isoform of aPKCλ, an aPKCλdegradation product (Coghlan et al.,2000), or another protein that crossreacts with the aPKCλantibody that was pulled down by Moe_FERM.

To test whether Moe directly binds to Crumbs proteins, we performed in vitro GST pull-down and far western experiments using purified recombinant proteins. Immobilized His-Moe_FERM interacted with GST-Crb2a^intra,GST-Crb2b^intra and, to a lesser extent, GST-Crb1^intra(Fig. 6A). Furthermore, we showed by far western analysis that GST-Crb1^intra,GST-Crb2a^intra and GST-Crb2b^intra bind to His-Moe_FERM immobilized on nitrocellulose (Fig. 6B). Taken together, our biochemical data suggest that Moe interacts directly with Crumbs proteins.

Both Nok and Pals1 have predicted Band 4.1-binding domains consisting of several lysine residues (Kamberov et al.,2000; Wei and Malicki,2002) and Stardust has a similar stretch of residues following the SH3 domain. We sought to determine whether Moe, a Band 4.1 protein, interacts directly with Nok (Pals1). We showed that immobilized GST-Nok-Int, containing the predicted Band 4.1-binding motif, pulled down MBP-Moe_FERM but not MBP-Moe C-terminus (Fig. 6C). We also showed in far western experiments that MBP-Moe_FERM bound to nitrocellulose-bound full-length Nok (His-Nok-FL), but not His-Nok-N, which does not contain the predicted Band 4.1-binding domain(Fig. 6D). We also showed that His-Nok-N (including the PDZ domain) directly interacted with both GST-Crb2a^intra, GST-Crb2b^intra proteins by far western(see Fig. S3 in the supplementary material). Our biochemical analyses show that Moe interacts directly with Crumbs proteins and Nok (Pals1).

Given the important role of Crumbs proteins in photoreceptors in Drosophila, mice and humans, we sought to determine whether Moe,probably through its interaction with Crumbs proteins, plays a role in vertebrate photoreceptor morphogenesis. Because the position of photoreceptors is so abnormal in moe mutants (see Fig. S1O in the supplementary material) (Jensen et al.,2001), examination of their morphology is uninformative. To overcome this limitation, we used genetic mosaic analysis and a transgenic line that expresses GFP in rods (Fadool,2003). When moe mutant cells are transplanted into wild-type hosts at the blastula stage, moe mutant cells are almost invariably found in their normal laminar position in the retina(Fig. 7) (Jensen et al., 2004),and so we used this strategy to examine the morphology of GFP⁺ rods that lack moe function. Wild-type rods have a stereotypical morphology: a basal synaptic terminal, a round cell body, a thin apical inner segment (IS) and a thick apical outer segment (OS) filled with rhodopsin(Fig. 7A,B,G,H).

At 6 dpf, the morphology of moe^- rods seemed largely normal, but the cells were almost 50% larger than wild-type rods(Fig. 7C-F,O). We measured the accumulated area of the OS versus the IS and cell body and found that the size increase in moe^- rods was due largely to an increase in the size of the OS (Fig. 7O). By 10 dpf the morphology of moe^- rods was markedly abnormal and the cells were about 50% larger than wild-type rods(Fig. 7I-N,P). Most often moe^- rods displayed a coiled apical structure that seemed to encompass both the IS and OS and seemed larger than the combined area of the IS and OS area of wild-type rods (Fig. 7I,K,L-N). We could not accurately measure the OS versus the IS and cell body at 10 dpf because the morphology was so distorted. We grouped transplanted moe^- rods into two groups by examining the genotype of neighboring cells; one group included moe^-rods that had few moe^- neighbors(Fig. 7I-K) and the other group moe^- rods had large numbers of moe^-neighbors (Figure L-N). Generally, moe^- rods with large numbers of moe^- neighbors were more abnormal(Fig. 7L-N) than those with mostly wild-type neighbors (Fig. 7I-K). Rhodopsin remained localized to the most distal portion of the cell (Fig. 7I-N),suggesting that apical-basal polarity is preserved. Three-dimensional movies of moe^- rods are provided (see movies 1-5 in the supplementary material).

We examined the localization of Crumbs proteins in wild type and moe^- rods in our transplant experiments. At 6 and 10 dpf,anti-panCrb labeling in the inner segment in moe^- rods did not seem different from wild-type rods(Fig. 7Q-X). Localization of ZO-1 appeared normal around moe^- rods in genetic mosaics(see Fig. S4 in the supplementary material). The observations that the apical region was expanded in moe^- rods suggest that Moe may normally act to inhibit apical size in photoreceptors. Interestingly, proper localization of Crumbs proteins to the photoreceptor IS does not appear to require Moe function, contrary to that observed in embryos (see Fig. 2).

The present study shows that Moe, a FERM protein, colocalizes with vertebrate Crumbs orthologs, co-immunoprecipitates with Crb2a, can directly interact with three zebrafish Crumbs proteins (Crb1, Crb2a and Crb2b) and is required for embryonic localization of Crumbs proteins. The similarity between the embryonic phenotypes of moe mutants and crb2a morphants suggests that Moe is a crucial regulator of crb2a function. We further show that Moe negatively regulates the size of the apical domain in rod photoreceptors. We propose that Moe, through its interactions with Crumbs proteins, has three independent roles in central nervous system development:brain ventricle formation, retinal lamination and photoreceptor morphogenesis. The role of Moe in brain ventricle formation may be to localize proteins that promote inflation of the ventricles by regulating ion transport and fluid dynamics. Consistent with this potential role is our observation that the peptide in the FERM domain of Band 4.1 that binds an anion exchanger(Jons and Drenckhahn, 1992) is highly conserved in Moe. It seems that ion exchangers play a crucial role in brain ventricle formation, as mutations in a Na+K+ATPase cause failure of brain ventricle formation that is very similar to the moeand nok mutations (Yuan and Joseph, 2004; Lowery and Sive,2005). Localization of such proteins in the retinal epithelium would be disruptive, because formation of a lumen would separate the retinal epithelium from the retinal pigmented epithelium. We still do not fully understand why moe loss of function has such a dramatic effect on retinal epithelial integrity while the integrity of the brain epithelium seems relatively normal even though moe and crb2a are expressed in both tissues. Protein function redundancy seems unlikely to account for the differences in retina brain defects, as we detected no expression of the paralog of moe, ehm2, in the brain (A.M.J. and Y.-C.H., unpublished),and panCrb labeling, which we showed recognizes all zebrafish Crumbs proteins,is completely lost in the brain of ome (crb2a) mutants.

The most remarkable observation made is that the apical membrane in rods is expanded by moe loss of function, and moe^- rods are almost 50% larger than wild-type rods. The morphology of moe^- photoreceptors was largely normal at 6 dpf, which is 3 days after the onset of rod IS and OS formation(Schmitt and Dowling, 1999),but by 10 dpf, most rods exhibited an abnormal morphology that included a distinctive coiled shape. By morphology we were unable to always distinguish the IS from the OS at 10 dpf, but Rhodopsin remained localized distally,suggesting that apical/basal polarity is preserved. The conspicuous coiled morphology of the rods could be intrinsic, caused by adhesion defects or due to space constraints imposed by neighboring cells.

The vertebrate photoreceptor OS is a highly modified cilium(Röhlich, 1975). Vertebrate Crumbs proteins have been shown to be important for ciliogenesis;siRNA knockdown of crb3 leads to a dramatic reduction in the number of ciliated MDCK cells (Fan et al.,2004), and in zebrafish inhibition of crb3a shortens auditory kinocilia, and morpholino knockdown of crb2b shortens nephric cilia and also shortens photoreceptor ISs, which lie below the OS(Omori and Malicki, 2006). Mice with a Crb1 mutation have shortened ISs and OSs(Mehalow et al., 2003). The Drosophila photoreceptor stalk region, which may be a homologous structure to the IS in vertebrate photoreceptors, is shortened by crbloss of function and is expanded by overexpression of full-length crb(Izaddoost et al., 2002; Pellikka et al., 2002). Overexpression of crumbs also expands the apical domain of ectodermal epithelia in the Drosophila embryo(Wodarz et al., 1995). Our observations that OSs are expanded in moe^- rods, taken with those above, suggest that Moe may be a negative regulator of Crumbs protein function in photoreceptors. We did not observe a significant increase in IS size at 6 dpf, and at 10 dpf we were unable to confidently identify the different compartments (cell body, IS and OS) in moe^-rods, so it remains to be determined whether the IS is affected by loss of moe function. Drosophila Yurt (Moe ortholog) also appears to act as negative regulator of apical membrane size and is shown to interact directly with Crumbs (Laprise et al.,2006). Collectively, our observations and those of others lead us to propose that Crb proteins are good candidates to be part of the molecular mechanism that regulates daily apical renewal in photoreceptors and that Moe may be an important negative regulator of this mechanism.

Photoreceptors, both in invertebrates and vertebrates, periodically shed distal apical membrane and then renew to replace the shed material. It has been estimated that as much as 10% of the OS is shed and renewed daily in mammals (Young, 1967), and both processes are regulated by cyclic light(Besharse et al., 1977; Hollyfield and Rayborn, 1979; Stowe, 1980); experiments in the locust suggest these processes are controlled locally(Williams, 1982). To our knowledge, no molecules or molecular mechanisms have been identified that regulate the process of apical membrane renewal in photoreceptors, although many genes, including that encoding Rhodopsin, are required for the formation of the OS (Lem et al., 1999). A prediction of this idea is that Crb1/Crb2 complex should be regulated by light. This prediction is supported already by the observation that exposure to bright light accelerates photoreceptor death in the mouse crb1knockout (van de Pavert et al.,2004) and Drosophila crb mutant retina(Johnson et al., 2002). The potential role of Crb1/Crb2 and its regulation by Moe has important implications in the etiology of photoreceptor degeneration diseases, which often are marked first by a shortening of the IS and OS.

Levels of Crumbs proteins may be especially critical for photoreceptors. Mouse and zebrafish photoreceptors express two crumbs genes(Fig. 1)(den Hollander et al., 2002; van den Hurk et al., 2005),and in Drosophila the stalk is slightly shorter in crb^-/+ photoreceptors(Pellikka et al., 2002). Perhaps the differences between Crb1 loss of function in humans and mice reflect differences in the compensatory function of Crb2. Although crb1 is not expressed in the zebrafish retina (data not shown) (Omori and Malicki,2006), photoreceptors still express two crumbs genes, crb2a and crb2b; perhaps one of these crb2 genes may have adopted the function of crb1 in mammals.

We showed that Moe and panCrb localization at the brain and retina ventricular surface depend on reciprocal Moe/Crb protein function and Nok function. The intracellular punctate panCrb labeling in the cell bodies of the wild-type brain and retinal neuroepithelium is reduced or absent in moe mutants (Fig. 2D,F), but overall protein levels are unaffected, suggesting that Moe may be required for the intracellular trafficking of Crb protein through organelles. Interestingly, disruption of Crb trafficking through the endosomal pathway leads to an upregulation of cell-surface Crb protein(Lu and Bilder, 2005), and recently two other FERM proteins, Merlin and Expanded, have been implicated in regulating cell-surface receptor localization, abundance and turnover(Maitra et al., 2006). The loss of apical panCrb labeling in moe^- embryos contrasts with the normal Crumbs protein localization observed in moe^- rods, suggesting that additional proteins or cellular or molecular mechanisms operate to localize Crumbs proteins in photoreceptors. Crumbs-expressing wild-type Müller glia, which send processes into the IS region, may help to localize Crumbs protein in moe^-rods.

We also show that Moe forms a complex with Nok (Pals1) and Has(aPKCλ) and that Moe can interact directly with Nok. The interaction between Moe and Nok may serve to regulate the interaction between Nok and Crumbs proteins or to bring Nok into the Crumbs complex. The former hypothesis is supported by studies of the Glycophorin C (GPC) ternary complex, which includes the Maguk protein, p55, and the FERM protein, Band 4.1, showing that the inclusion of Band 4.1 in the complex increases the affinity of p55 for GPC by an order of magnitude (Nunomura et al.,2000). The interaction of Moe with aPKCλ may be mediated by the Par3/6 complex (Wodarz et al.,2000; Hurd et al.,2003; Lemmers et al.,2004; Nam and Choi,2003); perhaps aPKCλ regulates the interaction between Moe and Crumbs proteins, as DaPKC phosphorylates Drosophila Crumbs in the FERM-binding domain and phosphorylation of Crb is required for apical localization of Crb in Drosophila embryos(Sotillos et al., 2004). In addition, Moe itself may be a target for aPKCλ regulation, as there are several potential serine and threonine phosphorylation sites (A.M.J.,unpublished).

We thank the following: Ben Margolis (anti-CRB3); Xiangyun Wei (anti-Nok);Paul Linser (anti-Carbonic Anhydrase II); Mikio Furuse (anti-ZO-1); Paul Hargrave (anti-Rhodopsin, B6-30); Jim Fadool (EGFP transgenic fish); Jeff Lee and Kathryn Anderson (mouse ymo1 (lulu) cDNA); Arianna Bruno for genotyping help; Judy Bennett for fish care; Ulli Tepass, Rolf Karlstrom and Barbara Osborne for comments. Supported by the NIH, EY015420.