Egfr signaling regulates ommatidial rotation and cell motility in the Drosophila eye via MAPK/Pnt signaling and the Ras effector Canoe/AF6

Epithelial planar cell polarity (PCP) in the Drosophila eye is generated when immature ommatidial preclusters acquire opposite chirality in the dorsal and ventral halves of the eye imaginal disc and subsequently rotate 90° towards the dorsoventral midline, the equator(Adler, 2002; Mlodzik, 1999; Mlodzik, 2002; Reifegerste and Moses, 1999). The signaling pathways that regulate the establishment of ommatidial chirality are beginning to be understood and require the specification of distinct cell fates within the R3/R4 photoreceptor pair. This is generated through an interplay of Frizzled (Fz)/PCP and Notch signaling. Fz/PCP signaling induces R3 and leads to the expression of the Notch ligand Delta in R3. Delta subsequently activates Notch in the neighboring cell, specifying it as R4(Cooper and Bray, 1999; Fanto and Mlodzik, 1999; Tomlinson and Struhl, 1999). The direction of the subsequent rotation follows the cell-type specification of the R3/R4 photoreceptors (Adler,2002; Mlodzik,2002). However, in fz and other PCP mutants as well as in Notch the mechanistic aspects of rotation are not abolished(Adler, 2002; Mlodzik, 1999; Mlodzik, 2002; Reifegerste and Moses, 1999),suggesting that other signaling pathways and mechanisms regulate the cell motility aspects of rotation.

During this process each ommatidium rotates 90° towards the equator and thus in opposite directions in the dorsal and ventral halves. This morphogenetic event is remarkable, as it requires groups of cells to undergo a coordinated precise movement: cells of each ommatidial precluster stick together and move as a unit with respect to the surrounding epithelial cells(Mlodzik, 1999; Reifegerste and Moses, 1999). The rotation of ommatidial preclusters is initiated in the third instar eye imaginal disc, shortly after the Fz/PCP-Notch signaling interplay specifies cell fate, and is visibly appreciated around column 6 posterior to the morphogenetic furrow (see Fig. 1A-C). It is completed by column 15-18 approximately 20-24 hours later. During this process, the rotation angles of individual ommatidia can be visualized with markers that highlight either the R3/R4 precursor pair or the R1/R6 pair. Interestingly, ommatidial rotation appears to be a two-step process as clusters stop for 3-4 columns after undergoing the first 45°rotation and before initiating the second 45° to complete the 90°rotation. In the adult eye the rotation is represented in the precise mirror image arrangement of the two chiral ommatidial forms that face each other on opposite sites of the equator (see Fig. 1D,E).

Very little is known about the regulation and mechanistic aspects of ommatidial rotation as only a few mutants that specifically affect this process have been isolated. In such `rotation specific' mutations both the direction of rotation and associated chirality-R3/R4 specification are unaffected, whereas the degree of rotation of individual ommatidial clusters is abnormal. Mutations falling into this category include nemo(nmo) (Choi and Benzer,1994), roulette (rlt)(Choi and Benzer, 1994), rok (the Drosophila Rho-kinase)(Winter et al., 2001) and scabrous (sca) (Chou and Chien, 2002).

The requirement of rok in ommatidial rotation provides the only link to the Fz/PCP pathway as Rho-kinase acts downstream of RhoA in several cellular contexts. Particularly in the Drosophila wing, rokhas been shown to act downstream of Fz/Dishevelled and RhoA in PCP establishment, affecting wing hair number (2-3 cell) but not their orientation(Winter et al., 2001). rok has been reported to affect rotation (and photoreceptor number),but not the R3/R4 cell fate decision (which is equivalent to wing hair orientation). This might suggest that some aspects of wing hair growth and rotation are shared through similar regulatory input into cytoskeletal regulation.

The nmo gene encodes a distant member of the Map kinase family,and in nmo mutants most ommatidia arrest at a rotation angle of approximately 45°, failing to execute the second 45° turn(Choi and Benzer, 1994). In sca, mutant clusters generally rotate more than 90°(Chou and Chien, 2002). In contrast, rlt mutant ommatidia rotate both less and more than the normal 90° (approximately 50% each) (see Fig. 1F), suggesting a general regulatory role in this process. The observation that nmo is epistatic to rlt and that nmo, rlt double mutants show the nmo phenotype (with ommatidia arresting at around 45°), lead to the hypothesis that rlt is required to stop rotation(Choi and Benzer, 1994).

Here we show that rlt is a rotation-specific allele of the Egfr-inhibitory ligand Argos. Both reduction and increase in Egfr signaling lead to defects in ommatidial rotation, suggesting that Egfr signaling is generally required to regulate this process. Using a combination of mutant analysis, genetic interactions and specific Ras-effector loop mutations we have identified both the Raf/MAPK cascade and the novel Ras effector Canoe/AF6 as a mediator of Egfr/Ras signaling in this process(Kuriyama et al., 1996). Our data indicate that Canoe/AF6 provides a link from Egfr to cytoskeletal elements in this developmentally regulated cell motility process. Furthermore,we provide evidence that Egfr signaling acts on cell adhesion via effects through the cadherin Flamingo, thus providing a link between Egfr and the PCP genes.

If not otherwise mentioned, fly stocks are as described in Flybase(http://flybase.bio.indiana.edu/),provided by The Bloomington Stock Center(http://flystocks.bio.indiana.edu/). The wild-type (WT) stock used was OregonR. rlt¹ and nemo^P1 mutants were kindly provided by K.W. Choi(Choi and Benzer, 1994). The argos^W11, argos^Δ7flies (Freeman et al., 1992)were kindly provided by C. Klaembt and sevargos by M. Freeman(Freeman, 1994). Deficiencies used that failed to complement the rlt rough eye phenotype were: Df(3L)st7P, Df(3L)st-j7, Df(3L)st-f13, Df(3L)81k19, Df(3L)st7,whereas the adjacent deficiencies Df(3L)D-5rv12, Df(3L)th102, Df(3L)std11,Df(3L)w10, Df(3L)cat and deficiencies more proximal or distal to the 70C-75F interval did complement rlt.

The Egfr alleles top¹ and Egfr^EC20(Clifford and Schupbach, 1989),as well as the S^48-5 allele and UAS-λ-top(Queenan et al., 1997) were kindly provided by T. Schupbach. UAS-PI3K and Dp110A flies were from Sally Leevers (Leevers et al.,1996). The GMR-Ral^G23V flies are as published(Sawamoto et al., 1999). The cno^Mis1 allele(Miyamoto et al., 1995) was provided by D. Yamamoto. UAS-cno, cno², cno³(Matsuo et al., 1997) and the UAS-RasV12 isoforms (Karim and Rubin,1998) were kindly provided by U. Gaul. The signaling specificity of the RasV12 effector loop isoforms was confirmed in Drosophila(Karim and Rubin, 1998; Prober and Edgar, 2002).

Embedding and sectioning of eyes was performed as described (Tomlinson,1987). Crosses were performed at 25°C (except for aos^w11 at 18°C). Multiple eyes were sectioned and analyzed for each genotype.

The hypomorphic behavior of the rlt allele and the absence of any other obvious phenotype led us to the conclusion that rlt is probably a regulatory mutant of aos. As regulatory mutations often affect untranslated regions of a gene, whereas the protein is not altered, we analyzed the genomic aos region in homozygous rlt flies using overlapping PCR primers. Two of the overlapping primer sets gave PCR products that were approximately 1 kb larger than the WT control. Sequencing of these PCR products revealed a truncated P-element inserted in the 5′UTR of aos within the proposed transcriptional start site at bp 55 according to the published aos sequence M91381. Primers used to amplify the truncated P-element from genomic DNA were: 5′CACAGACACGCACATACCG 3′ and 5′ CCCTCGCTCTATCGTTGTTC 3′.

The mΔ0.5-Gal4 construct was generated by cloning a XhoI-EcoRI mΔ0.5-fragment (PCR amplified from a plasmid, kindly provided by S. Bray using primers:5′-CCGCTCGAGTGCCATCAGATGTCAGCAAATG-3′ and 5′-CGGAATTCCTTTTGGCGCACAGTCACAC-3′) upstream of an EcoRI-BamHI fragment containing the P-element minimal-promoter (PCR amplified from pCasper 4 using the primers:5′-CGGAATTCAAAGCCGAAGCTTACCGAAGT-3′ and 5′-CGGGATCCTTTTTTTTTATTCCACGTAAGG-3′).

The Gal4 gene was added as a BamHI-NotI fragment (from pGaTB). The final construct was cloned as an Asp718-NotI fragment from pBSSK^– into pCasper4. Several fly lines were established according to standard procedures. mΔ0.5-Gal4 drives expression mainly in R4 (and also weaker early in R3 and later in R7), as confirmed by ablation experiments and expression of an UAS-GFP transgene.

Imaginal discs were dissected, fixed and stained as described(Freeman et al., 1992). Antibodies used were: rat anti-Elav (1:50), mouse anti-Boss (1:1000) (from Developmental Studies Hybridoma Bank), rabbit anti-Bar (1:100) (a generous gift from K. Saigo) (Higashijima et al.,1992), mouse anti-Fmi (1:10) (gift from Tadashi Uemura), rat anti-DE-cadherin (DCAD2, 1:20) (gift from H. Oda)(Oda et al., 1994) and mouse(Promega) or rabbit anti-βGal (Molecular Probes) (1:2000). Images were acquired on a Leica TCSSP (UV) confocal microscope and processed in Adobe Photoshop.

We mapped rlt, originally described as a spontaneous mutation on chromosome III (Choi and Benzer,1994), by complementation analysis with overlapping deficiencies to the 72D10-73B1 interval on chromosome 3L (not shown). Complementation analysis with candidate genes within this region revealed that alleles of the secreted Egfr inhibitory ligand argos(Freeman et al., 1992) failed to complement rlt (Fig. 1G). Transheterozygous rlt/aos and rlt/Df(3L)st7P(aos^–) flies displayed a very similar phenotype with rotation defects resembling homozygous rlt flies(Fig. 1G,H). In addition, the extra photoreceptor cell phenotype reported in a small number of rltommatidia (Choi and Benzer,1994) is enhanced in rlt/aos and rlt/Df(3L)st7P(aos^–) combinations(Fig. 1G,H). This is intriguing, as the most prominent phenotype of aos null clones and strong hypomorphic alleles are 1-2 extra photoreceptors per ommatidium caused by a lack of Egfr inhibition (Freeman et al., 1992). Thus, in complementation analysis rlt behaves as a hypomorphic allele of aos.

To confirm these results we attempted to rescue rlt with an argos transgene. The sev-argos transgene rescues the argos loss-of-function (LOF) eye phenotype and has no dominant effect on eye development (Freeman,1994). Strikingly, sevargos completely rescued the rlt eye phenotype to WT (Fig. 1I). These data indicated that rlt is a hypomorphic,rotation-specific allele of aos. In addition, molecular analysis of aos^rlt revealed an insertion of a truncated P-element in the 5′ untranslated region within the proposed transcription start site of aos (see Materials and methods for details), indicating that rlt is most probably a regulatory mutation of aos. We will subsequently refer to the rlt allele as aos^rlt.

To refine the phenotypic features of aos^rlt we have analyzed aos^rlt eye imaginal discs using markers that highlight cell-type identity and orientation of each cluster at the time when chirality is being established and ommatidial rotation takes place(Fig. 2 and not shown). Whereas in WT discs a fairly regular pattern is observed with all markers(Fig. 2A,C and not shown; see Figure legend for markers), in aos^rlt discs many clusters reveal rotation abnormalities (Fig. 2B,D). We observe clusters that are both overrotated and underrotated with respect to their neighbors and developmental stage(Fig. 2B,D). Discs of the strong aos^w11 allele showed, in addition to the extra R-cell phenotype, even more severe rotation defects(Fig. 2E), again indicating that aos is generally required during ommatidial rotation. Based on the relatively regular appearance of the first svp-lacZ and Bar-positive rows (Fig. 2B,D),it appears that the initial 45° rotation step is less affected.

We found no evidence for an aos effect on the R3/R4 cell-fate decision and ommatidial chirality, as the expression of the mΔ0.5lacZmarker, highlighting the R4 photoreceptor, appears normal within the R3/R4 pair (not shown; see below). Thus we conclude that argos is generally required for the process of ommatidial rotation, but has no effect on chirality and associated R3/R4 cell-fate decisions.

The requirement of Argos as an inhibitory ligand of Egfr(Jin et al., 2000) and the identification of argos^rlt suggested that Egfr signaling plays a critical role in ommatidial rotation. As Egfr is very pleiotropic with multiple requirements throughout development and eye patterning in particular (Bogdan and Klambt, 2001; Casci and Freeman, 1999; Freeman,1997; Schweitzer and Shilo,1997; Van Buskirk and Schupbach, 1999), it is not possible to analyze the role of Egfr in ommatidial rotation using mitotic clones of null alleles. However, hypomorphic alleles of Egfr allow the analysis of rotation defects. Eyes of the Egfr^top1 allele(Clifford and Schupbach, 1989),for example, are mildly rough and reveal, beside the expected photoreceptor loss (Freeman, 1997), many ommatidia with rotation defects (Fig. 3A). Similarly, rotation is affected in hetero-allelic Egfr^top1/Egfr^EC20 combinations(Fig. 3B).

Further evidence for an involvement of Egfr signaling in ommatidial rotation came from the analysis of Star (S) mutants. Star is a chaperone required for processing of the secreted Egfr ligand Spitz(Lee et al., 2001), and Egfr signaling is thus reduced in S mutants. S is haplo-insufficient for eye development and loss of one gene copy leads to a rough eye phenotype because of the reduction in Egfr activation(Kolodkin et al., 1994). S^–/+ eyes show a very mild loss of photoreceptors and, in addition, rotation abnormalities(Fig. 3C). Interestingly, the S^–/+ rotation phenotype is enhanced when one copy of Egfr or Ras is removed(Fig. 3D,E, Table 1), and is suppressed by aos^rlt heterozygosity (not shown). In addition, we see a specific dominant enhancement of the S^–/+ rotation phenotype by the removal of one copy of spitz, whereas the other known Egfr ligands vein and gurken had no effect(Table 1), suggesting that Spitz is the activating Egfr ligand in this context. The recently described Egfr ligand Keren could not be tested, because no mutants are available(Reich and Shilo, 2002). Taken together, these observations indicate that Egfr/Ras signaling is directly involved in regulating ommatidial rotation, a requirement that is masked by other effects seen in null clones or severe hypomorphic mutations.

Because loss of the inhibitory ligand Aos results in gain of Egfr signaling, we asked if reduction of Egfr signaling components could suppress the rotation defects in aos^rlt. Given that the argos rough eye phenotype is dosage-sensitive to Egfr-signaling components (Sawamoto et al.,1996), we analyzed aos^rlt flies in a heterozygous background of Egfr or its downstream effectors to obtain an insight into the effector pathways employed in ommatidial rotation. We found that the rotation defects in aos^rlt eyes were suppressed in an Egfr^top1/+ or Ras^e2F/+ and pnt^Δ88/+ background(Fig. 4), arguing for a requirement of the conserved Ras/MAPK/Pnt cascade in rotation downstream of Egfr. This is further supported by the fact that aos is a direct transcriptional target of the ETS domain transcription factor Pointed(pnt), which is itself activated by the MAPK Rolled(Golembo et al., 1996). Surprisingly, removing one copy of raf and rl/Mapk did not suppress the rotation defects of aos^rlt (not shown),suggesting that these kinases are not rate limiting in this process. Nevertheless, the facts that pnt acts as the strongest suppressor of aos^rlt and that Pnt is activated by Rl/MAPK argue for a requirement of the conserved Ras/MAPK/Pnt pathway in rotation. In addition,removal of one copy of the nmo gene also suppresses the aos^rlt phenotype (Fig. 4E), suggesting a link between Egfr signaling and nmo. This is in line with previous observations proposing that nmo is epistatic to rlt: the nmo, rlt double mutant exhibits a nmo phenotype (Choi and Benzer,1994).

As ommatidial rotation is a cell motility process requiring cytoskeletal rearrangements, we wondered if other effectors of Egfr beside the Raf/MAPK cascade play a role in this process. The Ras GTPase, the main transducer of Egfr signaling, can utilize distinct effectors in different contexts. In addition to nuclear signaling, mediated by the Raf/MAPK/Pnt cascade, Ras can affect cell growth and cytoskeletal rearrangements via its effectors Rgl/Ral,Phospho-inositol-3-Kinase (PI3K) and Canoe, whose human homologue (AF6) is known as the critical partner of ALL1 in a chimeric protein associated with myeloid leukemia (Joneson et al.,1996; Karim and Rubin,1998; Kuriyama et al.,1996; Prasad et al.,1993; Prober and Edgar,2002; Rodriguez-Viciana et al., 1997).

Several point mutations have been identified within the Raseffector loop(amino acids 34-41) that abrogate the binding to and activation by Ras of specific effectors (Joneson et al.,1996; Karim and Rubin,1998; Rodriguez-Viciana et al., 1997) (see also Fig. 5). The specificity of the existing Ras-effector loop mutations has been thoroughly tested in Drosophila imaginal discs(Prober and Edgar, 2002). Prober and Edgar show that RasV12[S35], able to interact with Raf in cell culture, can activate ERK/Rolled (via Raf) and induce Ras/Raf/ERK-specific transcriptional responses in wing and eye imaginal discs (figures 3,7 in Prober and Edgar) (Prober and Edgar,2002). In contrast, RasV12[G37], unable to bind Raf in cell culture, cannot activate these responses, but is still capable of activating PI3K-specific read-outs (figures 3,7 in Prober and Edgar)(Prober and Edgar, 2002). Therefore we tested the Raseffector loop mutations in constitutively activated RasV12 for their effects on ommatidial rotation.

Expression of RasV12 in developing photoreceptor precursor cells using common eye-specific drivers causes induction of many extra photoreceptors(Fortini et al., 1992), and thus does not allow the analysis of rotation (the orientation of individual ommatidial clusters cannot be determined unambiguously in the presence of extra photoreceptor cells). To circumvent this problem RasV12 and its effector loop isoforms were expressed in a limited photoreceptor subset after these have been determined as photoreceptors using the mΔ0.5-Gal4 driver (see Materials and methods for detail). RasV12 effects on photoreceptor number and fate were thus strongly reduced. To determine the full effect of activated Ras expression under the control mΔ0.5-Gal4, we expressed constitutively active RasV12 that activates all known Ras-effectors. This gave rise to eyes with some gain and loss of photoreceptors and severe misrotations(Fig. 5A). The equivalent expression of RasV12[S35], thought to activate mainly Raf, also resulted in rotation abnormalities and occasional gain or loss of photoreceptors, again supporting a requirement of the Raf/MAPK cascade(Fig. 5B).

Strikingly, expression of RasV12[G37] and RasV12[C40] also caused misrotations, suggesting an involvement of additional Ras-effectors in this process. In particular, RasV12[G37], in which Raf activation is abolished (or at least strongly reduced), resulted in severe rotation defects(Fig. 5C,E), suggesting that PI3K, Rgl/Ral or Canoe might play a role in this cell motility process. Similarly, RasV12[C40], eliminating Raf activation, but maintaining weaker activation of other effectors, also showed rotation abnormalities, albeit weaker than RasV12[G37] (Fig. 5D). Moreover, expression of RasV12[C40] under the control of the sevenless (sev) promoter (in R3/R4, R1/R6 and R7) resulted in strong rotation defects that were comparable to those seen with RasV12[G37]under sev control (Fig. 5E,F). Taken together, these data suggested an involvement of PI3K, Rgl/Ral or Canoe in ommatidial rotation.

To confirm the RasV12[G37] effect and determine which of the three known effectors activated by RasV12[G37] is required in ommatidial rotation, we analyzed PI3K, Ral and Canoe directly. UAS-PI3K expressed under mΔ0.5-Gal4 had no effect on rotation (not shown),suggesting that PI3K is not required in this process. This is further supported by the lack of rotation defects in dPI3K mutant clones(Leevers et al., 1996). In contrast, expression of activated Ral(Sawamoto et al., 1999) as well as mΔ0.5-Gal4>UAS-Cno exhibits rotation defects (Fig. 6A and not shown).

We next tested for a direct canoe (cno) requirement in ommatidial rotation using LOF alleles. First, we asked whether cnoheterozygosity could interact with the S^48-5/+rotation phenotype. Strikingly, similar to the enhancement observed with Egfr or Ras, the cno^2/+ and cno^3/+ genotypes enhanced the S^48-5/+ rotation phenotype(Fig. 3F, Table 1 and not shown). cno is required for cone cell and photoreceptor differentiation and thus clones of null and strong alleles cause a general disorganization of the eye (Miyamoto et al., 1995)and are difficult to analyze for rotation defects. However, the hypomorphic cno^mis1 allele is subviable in trans to the strong alleles cno² and cno³ with mildly rough eyes,allowing an analysis of ommatidial rotation. Eye sections of such transheterozygous cno flies (e.g. cno^mis1/cno²) reveal severe rotation defects (Fig. 6B). To test whether such defects are already observed at the time when rotation takes place, we analyzed cno mutant third instar eye discs. Strikingly,rotation defects, comparable in strength to the stronger aos alleles,are apparent in cno eye imaginal discs(Fig. 2F). The discs were counterstained with anti-Elav (not shown) to ensure that the photoreceptor complement is normal in such cno mutant discs and the observed rotation abnormalities are primary defects, which was indeed confirmed. A similar analysis of Ral/Rgl is precluded by the lack of suitable alleles. In summary, these data indicate that cno plays a critical role in ommatidial rotation and acts as an effector of Egfr/Ras signaling in this context (see also Discussion).

As ommatidial rotation is a cell biological event, it is probable that among the main read-outs affected are cell-adhesion properties of the precluster cells and effects on cytoskeletal elements. This is further supported by our observations that (1) Raf/MAPK-independent and thus transcription-independent Egfr/Ras signaling pathways are important, and (2)that canoe is required in this context. To address this further, we performed two sets of experiments. First, we tested for genetic interactions between the dosage-sensitive Star^–/+ rotation phenotype and selected factors required in cell adhesion and cytoskeletal regulation; and second, we directly analyzed whether cell-adhesion components such as cadherins and integrins are normally localized in aos^rlt and cno^Mis1 mutant backgrounds.

To specifically test the involvement of cytoskeletal elements and adhesion as well as junctional components, we tested candidate genes for dominant interaction of the mild Star rotation phenotype(Table 1). This genetic data argue for an involvement of DE-Cadherin/shotgun, the atypical cadherin Flamingo (Fmi) (Usui et al.,1999), the adherens junction protein canoe, non-muscle myosin II (zipper), the septin peanut, and capulet,a protein with actin and adenylate cyclasebinding ability.

Next we examined the expression of Fmi and DE-cadherin in ommatidial preclusters during rotation. Strong LOF alleles of Egfr and its signaling components also affect cell proliferation, fate specification and survival, making the analysis of cell adhesion and junctional components in the context of rotation rather difficult. Thus we analyzed localization of the cadherins and Arm/β-catenin in imaginal discs of the rotation-specific aos^rlt allele.

Although the overall expression and localization of DECadherin and Arm/β-catenin are largely unaffected(Fig. 7A,A″,B,B″,C,C″,D,D″ and not shown), the localization of Fmi is changed in aos^rlt discs(Fig. 7A,A′,B,B′,G,H). In WT, Fmi is initially present apically in all cells of the morphogenetic furrow and subsequently becomes asymmetrically enriched in the R3/R4 precursor pair (column 4-5, Fig. 7A,C,G)(Das et al., 2002). In and posterior to column 6 Fmi is expressed at the membrane of R4, and largely depleted from R3 membranes that do not touch R4, forming a horseshoe-like R4-specific pattern (Fig. 7A,A′,G). In contrast, in aos^rlt discs Fmi restriction to the R4 precursor is generally delayed, and often not established even in columns 8-12, where high levels of Fmi are still seen around the apical membrane cortex of R3 and R4(Fig. 7B,H). As Fmi is thought to act as a homophyllic cell-adhesion molecule(Usui et al., 1999), its increased presence on R3 membranes should have a direct effect on Fmi localization in neighboring cells and thus possibly the adhesive properties of the precluster. It is worth noting that although Fmi is required during PCP establishment and R3/R4 cell-fate specification, the delay in Fmi restriction to R4 has no significant effect on the R3/R4 cell-fate decision. Although Fmi interacts with Fz and Notch in this context, the R4-specific mΔ-lacZ marker does not differ significantly from WT(Fig. 7E,F) and adult aos^rlt eyes also display no defects in R3/R4 specification. Thus, it appears that the delay in Fmi localization specifically affects ommatidial rotation, probably through adhesion, and possibly explains the broad range of rotation angles in aos^rlt and other Egfr pathway mutants.

Taken together with the genetic interactions, these data suggest that Egfr signaling impinges on the cell adhesive properties of ommatidial clusters during the rotation process (see below).

Our data demonstrate that Egfr signaling regulates ommatidial rotation. Loss and gain-of-function data indicate that at least two Ras-effector pathways, the Ras/Raf/MAPK cascade and Ras/Cno signaling, are required downstream of Egfr in this context of cellular motility(Fig. 8). We provide further evidence for an involvement of cadherins and non-muscle myosin II as downstream components controlling rotation. In particular, the involvement of the cadherin Fmi, a PCP gene, downstream of Egfr signaling, is of interest as it provides the first link between PCP establishment and Egfr signaling.

What is the role of Egfr signaling in the regulation of ommatidial rotation? It is of interest to note that both loss and gain-of-function phenotypes of Egfr and its signaling pathway components result in very similar phenotypes, with ommatidia being both under- and overrotated. This is in contrast to nmo and sca mutants, in which ommatidia generally underrotate (nmo) or overrotate (sca). Although rlt has been implicated to stop rotation at 90°(Choi and Benzer, 1994), our data suggest a general role for Egfr in this process affecting more than just its termination. The observed underrotation in Egfr LOF genotypes,such as Egfr^top1 alleles or Star heterozygosity,is consistent with a positive requirement of Egfr in the process. The presence of overrotated ommatidia in these genotypes questions this simple interpretation. However, the overrotations could be explained by the feedback loops that regulate the levels of Egfr signaling. An initial reduction of Egfr signaling, such as in the Egfr^top1 or Starbackgrounds, also results in a reduced induction of the negative feedback loop(mediated through aos), subsequently leading to a mild increase in Egfr signaling. Thus, as null mutations of Egfr cannot be easily analyzed in this context, it is possible that in all genotypes analyzed the absolute levels of Egfr signaling fluctuate, causing the observed phenotypic variations. In this context, it is noteworthy that the gain-of-function Egfr mutation Elp^B1 exhibits only very minor rotation abnormalities, but significant photoreceptor loss because of overproduction of Aos (Lesokhin et al.,1999).

An alternative interpretation of the observed Egfr rotation defects could be a general regulatory input from Egfr signaling into the process of rotation. Egfr/Ras signaling could serve as a `gas' or `brake' pedal throughout the process, thus regulating the strength of another signaling input. This is supported by the observations that ommatidia can be significantly overrotated before the end of the process, suggesting a continuous Egfr input.

Nevertheless, the phenotypes observed in third instar eye imaginal discs suggest that the first 45° rotation is much less affected than the second step. This indicates that Egfr signaling plays an important role in the second 45° turn, and serves a lesser role in the first rotation step (see also Brown and Freeman, 2003). In summary, we suggest that if Egfr signaling is deregulated, either through a reduction in aos or Egfr function, rotation is disturbed during the second 45° step. We propose that the restriction of Egfr signaling, required for correct rotation, is mediated by an autoregulatory feedback loop via aos. In parallel, the mechanistic cell motility aspects of rotation are probably regulated at least in part via the Ras effector Cno.

The enhancement of the rotation defects in S heterozygous eyes through several cell adhesion molecules and cytoskeletal regulators implicates Egfr signaling in controlling the mechanistic `cellmotility process'. Our results, showing that expression of the cadherin Fmi is altered in aos^rlt eye discs, suggest that Egfr signaling impinges on cell-adhesion properties of the ommatidial precluster. It is interesting to note that in WT the expression of DE-cadherin and Fmi are almost exclusive within the early precluster: membranes between the R8, R2 and R5 cells show high levels of DE-cadherin, whereas the membranes of the R3/R4 precursors show high Fmi levels (Fig. 7A,A″,B,B″,G,H). Thus the regulation of such differences among distinct cadherin proteins might be important for normal ommatidial rotation to occur. The input of Egfr signaling on cell adhesion is further supported by the genetic interactions of Star with fmi and DE-cadherin/shotgun.

These analyses of the regulation of cell adhesion molecules in the context of ommatidial rotation will help to start unravel the mechanistic aspects of rotation.

The Egfr/Ras/Cno link is intriguing for several reasons. The cnogene was originally identified as a mutation affecting the dorsal closure process during embryogenesis (Jürgens, 1984; Takahashi et al., 1998). Cno shows a genetic and molecular link to Ras: it contains two Ras-interacting domains and binds both WT Ras and activated Ras-V12(Kuriyama et al., 1996; Matsuo et al., 1997). In addition, Cno has been postulated to link cytoskeletal elements to cellular junctions via its ability to bind actin(Mandai et al., 1997), its interaction with ZO-1/Pyd and its homology with kinesin and myosin-like domains (Takahashi et al.,1998). Thus Cno could directly mediate an Egfr/Ras signal to cytoskeletal and cell architecture elements through its association with adherens junctions and its kinesin and myosin-like domains(Miyamoto et al., 1995). Interestingly, Zipper does not only show a similar interaction with Star, like Cno (Table 1), but it is also required during embryonic dorsal closure(Young et al., 1993), and thus a more general Cno-Zipper link might exist in cell motility contexts.

A second interesting feature of cno is that it has been genetically linked to sca and Notch signaling(Miyamoto et al., 1995). First, the phenotype of the sca¹ allele is strongly enhanced by cno^–/+. Second, cno alleles also display Notch-like phenotypes in the wing and a GOF Notch allele, Notch^Abruptex, is suppressed by cno (Miyamoto et al.,1995). Although the biochemical role of Sca remains obscure, it has been linked to Notch, possibly as a Notch ligand, in several contexts(Powell et al., 2001). Thus as sca has recently been implicated in ommatidial rotation(Chou and Chien, 2002), the link between Cno and Sca/Notch is intriguing. Taken together, Cno could serve as a factor integrating signaling input from different pathways, e.g. Egfr and Notch in this process, and relaying this to cytoskeletal elements. The Canoe link is also interesting from a disease point of view as its human homologue AF6 is the critical partner of ALL1 in a chimeric protein associated with myeloid leukemia (Prasad et al.,1993). Thus taken together, Cno could serve as a factor integrating signaling input from different pathways, e.g. Egfr and Notch in ommatidial rotation, and relaying this to the regulation of cell adhesion and cytoskeletal elements in the context of a developmental patterning process or disease.

We demonstrate that Egfr/Ras signaling plays a general role in the regulation of ommatidial rotation. We further identify Canoe as an effector of Ras in this context. Although much was known about how ommatidial chirality and the associated R3/R4 cell-fate decision are regulated (Fz/PCP-Notch signaling), no clear link between the mechanistic aspects of ommatidial rotation and Fz/PCP signaling previously existed. We show a first link between Egfr signaling and PCP genes, namely Fmi. A further connection between Egfr signaling and PCP establishment is provided by Zipper, which acts downstream of Fz/Dsh and Rok in wing PCP (Winter et al., 2001) and modifies the Star rotation phenotype. The identification of the Egfr pathway and its regulation of Fmi/cadherin-mediated cell adhesion will serve as an important entry point to further such studies.

We are grateful to K. Choi, M. Freeman and U. Gaul for sharing unpublished results and helpful discussion, N. Baker, K. Choi, M. Freeman, U. Gaul, Ch. Klaembt, S. Leevers, K. Moses, H. Okano, K. Saigo, T. Schuepbach, D. Yamamoto,T. Uemura, H. Oda, A. Jenny, the Bloomington Stock Center and the DSHB for flies or antibodies. We thank members of the Mlodzik lab, C. Desplan, N. Baker and L. Kockel for discussion and/or comments on the manuscript. Microscopy was performed at the MSSM-Microscopy Shared Resource Facility, supported by a NIH-NCI shared resources grant (1 R24 CA095823-01). This work was supported by NIH-NEI grant EY14597 to M.M.