Cooperative activities of Drosophila DE-Cadherin and DN-Cadherin regulate the cell motility process of ommatidial rotation

Cell polarization is essential for cellular function, patterning and cytoskeletal organization. Epithelial cells are polarized along the apicobasal axis. In addition, many tissues are polarized within the plane of the epithelium, generally called planar cell polarity (PCP). The conserved signaling events that regulate PCP are mediated by the Frizzled (Fz) pathway(Adler, 2002; Klein and Mlodzik, 2005; Mlodzik, 2002; Shulman et al., 1998; Veeman et al., 2003). Although genetic screens have identified many signaling molecules that participate in PCP signaling, the relationship between signaling events and changes in cytoskeletal organization and/or cellular adhesion remains unclear.

The read-out of PCP signaling is often the organization of cytoskeletal elements or orientation of the mitotic spindle(Adler, 2002; Bellaiche et al., 2001; Klein and Mlodzik, 2005; Shulman et al., 1998). PCP is also evident in neural sensory epithelia, most prominently in the Drosophila eye (Mlodzik,1999; Strutt and Strutt,1999) and mammalian inner ear(Dabdoub et al., 2003; Montcouquiol et al., 2003). In the Drosophila eye, PCP establishment includes transcriptional events linked to photoreceptor cell fate specification and complex cellular movements that cause ommatidial precursors to undergo a 90° rotation towards the midline of the developing eye field(Mlodzik, 1999; Strutt and Strutt, 1999).

Drosophila eyes are composed of ∼800 ommatidia, each consisting of eight photoreceptor cells, and 12 accessory cells. Ommatidial assembly begins posterior to an organizing center, the morphogenetic furrow(MF), an indentation that forms at the posterior of the eye disc and moves across the developing disc. Photoreceptor precursors are assembled and specified behind the MF in a stereotyped manner, with R8 being the first,followed by pairwise recruitment of R2/R5, R3/R4 and R1/R6, and finally R7(Wolff and Ready, 1993).

Correct R3/R4 specification is crucial for PCP establishment. Initially,this photoreceptor pair is symmetrical. Then, the cell located closer to the dorsoventral (D/V) midline, the equator, receives a higher level of Fz signal,which specifies it as R3, and causes a transcriptional upregulation of Delta in the same cell. Delta then activates Notch signaling in the neighboring cell of the pair, specifying it as R4(Cooper and Bray, 1999; Fanto and Mlodzik, 1999; Tomlinson and Struhl, 1999). As a result of distinct R3/R4 specification, ommatidial preclusters lose their symmetry and adopt two opposing mirror-image chiral forms in the ventral and dorsal eye hemispheres (Mlodzik,1999; Strutt and Strutt,1999).

Following R3/R4 specification, ommatidial preclusters rotate towards the equator (Mlodzik, 1999; Strutt and Strutt, 1999; Wolff and Ready, 1993). Each cluster rotates as a unit; this process is highly coordinated as clusters of the same maturity rotate simultaneously. At its end, ommatidia have rotated 90° away from their original position. R3/R4 specification is a prerequisite for the correct direction of rotation, resulting in dorsal and ventral ommatidia rotating in opposite directions(Mlodzik, 1999; Strutt and Strutt, 1999). Both the establishment of asymmetry within nascent ommatidial clusters and the subsequent rotation event lead to the final ommatidial arrangement, reflected in the mirror-image symmetry across the equator(Fig. 1).

How ommatidial rotation initiates is unknown. The signal to stop rotation at 90° is also unknown, but likely involves EGF-receptor (Egfr) and Notch(N) signaling (Brown and Freeman,2003; Chou and Chien,2002; Gaengel and Mlodzik,2003; Strutt and Strutt,2003). Fz/PCP signaling may regulate rotation through effects on cytoskeletal elements and cell-adhesion molecules. This is supported by the observation that rok (Drosophila Rho-associated kinase), an effector of the RhoA GTPase that has been placed genetically downstream of Fz has a rotation-specific eye phenotype(Winter et al., 2001). Finally, based on genetic evidence, initiation of the second rotation step(45° to 90°) is regulated by nemo, a MAP kinase(Choi and Benzer, 1994). A link of nemo, Egfr or N-activity to Fz signaling or rok-mediated events has not been established. Rotation still takes place in Fz/PCP mutants,albeit at random and mutations in rotation-specific genes do not affect the R3/R4 fate decision and ommatidial chirality. It thus appears likely that distinct signaling inputs are required downstream of Fz/PCP to regulate the mechanistic aspects of ommatidial rotation.

The role of cell-adhesion in ommatidial rotation has not yet been addressed. Cadherins are a conserved family of transmembrane proteins that mediate cell-cell adhesion. They maintain tissue integrity and are required for dynamic processes including cell polarization, cell sorting, tissue segregation, axonal patterning, synaptogenesis and cell migration(Oda et al., 1998; Tepass et al., 2000; Yeaman et al., 1999). Classical cadherins are single pass transmembrane glycoproteins that mediate cell-cell adhesion through their Ca-dependent extracellular domains, which are composed of `cadherin repeats'. Their cytoplasmic tails interact with the cytoskeleton by binding to catenins, which then recruit other cytoskeletal or signaling proteins (Perez-Moreno et al.,2003). The cadherin-catenin complex, which forms adherens junctions in vertebrates and invertebrates, is dynamically regulated to participate in cell adhesion, sorting and signaling.

Drosophila has three cadherins that are functionally related to vertebrate classical cadherins: DE-cadherin (DE-cad/shotgun),expressed in almost all epithelia; and two DN-cadherins (DN-cad1/cadNand DN-cad2/cadN2), encoded by adjacent genes and predominantly expressed in mesodermal and neural tissues(Iwai et al., 1997; Tepass et al., 1996; Uemura et al., 1996; Prakash et al., 2005). Vertebrate E- and N-cadherins often show complementary expression patterns. Similarly, Drosophila DE-cad is early expressed ubiquitously and later is downregulated in newly forming mesoderm and neural tissues, and replaced with DN-cad. This `cadherin switch' from an epithelial cadherin,maintaining cells in an epithelial layer, to a N-cad that promotes cell migration is a required step during development(Hazan et al., 2004). As cadherins are required to maintain tissue integrity, it is difficult to demonstrate an instructive role in tissue remodeling. However, there is evidence that, in addition to N-cad, E-cad also promotes cell movement(Montero et al., 2005; Niewiadomska et al.,1999).

Studies with Egfr signaling in ommatidial rotation suggest that DE-cad is involved in this process. In particular, decreasing the dose of endogenous DE-cad enhances the rotation specific phenotype of Egfr-signaling components(Brown and Freeman, 2003; Gaengel and Mlodzik, 2003),suggesting that DE-cad is required during rotation. Here, we demonstrate that shotgun (shg/DE-cad) promotes ommatidial rotation. By contrast, the DN-cadherins appear to restrict rotation, most probably by preventing ommatidia from rotating too fast. Our data indicate that the balance between DE-cad and DN-cad is crucial for the precision required to achieve correct ommatidial rotation. Consistent with this conclusion is the largely complementary subcellular accumulation of DE-cad and DN-cad1 in developing ommatidial preclusters. The dynamic cell-cell adhesion mediated by the two cadherins regulates the extent of motility in several developmental contexts and disease states.

UAS-DE-cad transgenes were as described(Lee et al., 2003). For UAS-DN-cad1, two independent lines were tested with medium to high expression levels (T. Uemura, personal communication). All were driven with the UAS/Gal-4 system (Brand and Perrimon,1993). sevGal4 drives expression in R3/R4, R1/R6 and R7 precursors (gift from K. Basler). The flip-out gain-of-function clones of UAS-DN-cad and UAS-DE-cadDN were generated with hs-FLP,actin>CD2>Gal4 (Pignoni and Zipursky, 1997) and marked with UAS-GFP. Interaction crosses were grown at 25°C and eyes from female flies examined. The transgenes that modified sev>DE-cadDN (UAS-Ecad^WT,UAS-EcadEx, UAS-EcadΔB, UAS-Arm^S2) showed no phenotype with sevGal4 by themselves.

For loss-of-function clonal analyses, shg^P34-1 (gift from U. Tepass) was recombined onto the w^-; FRT42chromosome and clones were induced with eyFLP (marked with ubiGFP) (Newsome et al.,2000). Minute (M) clones [giving mutant tissue a growth advantage;Garcia-Bellido et al. (Garcia-Bellido et al., 1976)] were induced with eyFLP; FRT42, ubiGFP, M. For rescue experiments, eyFLP; shg^P34-1, FRT42, ubiGFP;sev>DEcad^WT genotype was analyzed.

Mutant alleles and UAS lines: shg^P34-1, shg^R6,shg^Ig29 (from U. Tepass), DE-cad^WT (UAS-DECH and UAS-DEFL), DE-cadDN (dCR3h 7-1), DE-cad-dEx (UAS-dEx), DE-cadDN (UAS-dCR4h,UAS-dCR4, UAS-dCPc3, UAS-mNcGSP) (see Oda and Tsukita, 1999); UAS-DE-cadΔβ(Pacquelet et al., 2003),ubi-DE-cad-GFP (from F. Schoeck), UAS-DN-cad, Ncad^M19(Iwai et al., 1997),UAS-RhoA^IR (Billuart et al.,2001), Ncad^Δ14(Prakash et al., 2005),UAS-DE-cad^wt (Lee et al.,2003), stbm⁶, and dgo, stbm double mutant(Das et al., 2004). shg², mys¹, UAS-Arm^S2, arm^YD35,arm^XM19, wg^CX4, spi²⁵⁹, Egfr^ElpB1,pnt^D88, fz^R52, dsh¹, fmi^E59,dgo³⁸⁰, nmo^P, dRok², sca^BP1,rhoA^72R and UAS-GFP are as described in FlyBase(http://flybase.bio.Indiana.edu).

Primary antibodies used were: rat anti-DE-cad (DCAD2; gift from H. Oda);rabbit anti-Bar (gift from K. Saigo); rat anti-Spalt (from B. Mollereau);rabbit anti-Dlg (gift from Z. H. Chen); mouse anti-γ-tubulin (Sigma),rhodamine-phalloidin (Molecular Probes); anti-β-Gal (Cappel, Promega),rat anti-Elav, mouse anti-Armadillo (N27A1), rat anti-DN-cad (Dn-Ex#8) and mouse anti-Flamingo from Hybridoma Bank. Secondary antibodies coupled to fluorochromes were from Jackson Laboratories. Imaginal discs were stained as described (Fanto et al.,2000), mounted in 80% glycerol/2% N-propyl-gallate and viewed with a Zeiss 510 confocal laser-scanning microscope. Eye sections were prepared as described (Tomlinson et al.,1987). For genetic interactions, eyes were sectioned near equatorial region and only ommatidia with correct photoreceptor number were scored; a rotation defect was defined as being either less than 80° or more than 110° (compared to 90° in wild type).

To address the role of DE-cad in ommatidial rotation, we first analyzed the eye phenotypes of mutant alleles of shotgun (shg, the Drosophila E-cad gene). A complete loss of DE-cad causes severe cellular defects because of the loss of tissue integrity and, thus, precludes the analysis of ommatidial rotation. However, several hypomorphic shgalleles exist that retain enough activity to allow for normal early epithelial development and, thus, the analysis of PCP associated defects. Eye sections from a viable weak hypomorphic allele, shg^R6, show areas where many ommatidia have rotated less than 90°(Fig. 2A; stronger alleles do not allow for an analysis in adult eyes, as many photoreceptors are malformed or missing, Fig. 2B). Next, we analyzed somatic eye imaginal disc clones of an intermediate shgallele (shg^P34-1) using the FRT/FLP method. This allowed us to test whether the effect on rotation is a consequence of DE-cad/shg function at the time when ommatidial rotation is taking place, and to analyze the phenotypes of stronger shg alleles. Using several markers, including anti-Bar staining, we were able to highlight the rotation angle of the R1/R6 cell pair and, thus, the whole ommatidial cluster(Fig. 1C,D). In late 3rd instar larval discs, when ommatidial rotation occurs, clones of shg^P34-1 show robust rotation defects(Fig. 2C-C″). Within mutant tissue, about 50% of ommatidia do not initiate rotation and several others rotate less in comparison with wild-type clusters of the same stage(Fig. 2C-C″,E-E″and Table 1). Rotation defects are apparent both in small and large mutant clones, where most of the eye is homozygous for shg^P34-1(Fig. 2C-C″ and not shown). Rotation defects are apparent from its initiation as detected with the early R3/R4 marker Spalt (Fig. 2D,D′). We also tested for apicobasal polarity defects with specific markers (see Materials and methods), and found no evidence for significant apicobasal polarity defects(Fig. 2F,F′ and not shown). Specification of R3/R4 cell fates was also largely normal(Fig. 2D).

These data indicate that the rotation phenotype is a primary defect and not due to a general disorganization of the epithelium. Taken together, this indicates that shg/DE-cad is required to promote ommatidial rotation.

To further examine the role of DE-cad in rotation, we used the UAS/Gal4 system (Brand and Perrimon,1993) to analyze the consequences of eye specific expression of different forms of the DE-cad protein (Oda and Tsukita, 1999) (see Materials and methods). In addition to full-length protein, transgenic lines with deletions of either the entire extracellular, intracellular or deletions in cadherin (CAD) repeats were tested (see Materials and methods). Expression of full-length DE-cad under the control of sevGal4 (expressed in the R3/R4 pair at the time of rotation initiation; see Materials and methods) had no effect at moderate levels, and caused epithelial defects at high levels (not shown). By contrast,a sevGal4 driven dominant-negative construct [DE-cadDN, with an internal deletion of two CAD repeats;(Oda and Tsukita, 1999)]resulted in specific rotation defects (sev>DE-cadDN; Fig. 3A). Importantly, as photoreceptor loss and patterning defects were minimal and PCP associated chirality defects were not observed, this suggested that DE-cad specifically affects ommatidial rotation. This phenotype is comparable with rok or nemo (Choi and Benzer,1994; Winter et al.,2001), which also specifically affect rotation but not the R3/R4 fate decision. We confirmed that DE-cadDN indeed acts as a dominant negative, as removing one copy of endogenous DE-cad/shgenhanced the sev>DE-cadDN phenotype(Fig. 3B). Moreover,coexpression of DE-cad^WT (moderate levels) with DE-cadDN (under sevGal4 control) rescued the rotation defects (Fig. 3C).

A striking feature of all rotation defects associated with the DE-cad loss-of-function phenotypes (either in hypomorphic allele, shg^R6, shg^P34-1, or DE-cadDNcombinations) is that the majority of ommatidia rotate less than 90°,displaying variable levels of under-rotation or lacking rotation completely(Fig. 2A-E; Fig. 3A,B; Table 1). The severity of the under-rotation correlates with allelic strength (see Fig. S1 in the supplementary material). The above data indicate that DE-cad/shg acts as a positive regulator of ommatidial rotation.

We next analyzed the expression pattern of DE-cad during ommatidial rotation. E-cad is expressed uniformly in epithelia close to the apical cell surface, establishing adherens-junctions through its homophilic cell-adhesion behavior and its intracellular interactions with the cytosolic catenins(Perez-Moreno et al., 2003). Similarly, in the eye disc epithelium, prior to photoreceptor recruitment and adjacent to the furrow (MF), DE-cad is equally distributed as an apicolateral membrane `ring' in all cells (Fig. 4A-A′, left of MF). Posterior to MF, DE-cad expression is patterned (Fig. 4A,A′,right of MF). As photoreceptor precursors are recruited into preclusters,DE-cad becomes enriched in membranes bordering R8, and the R2/R5 pair(Fig. 4A,A′, see Fig. 4D,D″ for detail),and interestingly it is detected at much lower levels at the membranes between the R3/R4 pair, as well as membranes between precluster cells and surrounding undifferentiated cells (Fig. 4A,A′,4D,D″). DE-cad remained upregulated at all membranes between precluster cells, except the R3/R4 border, throughout the process of rotation and precluster assembly(Fig. 4A).

The Drosophila β-catenin protein Armadillo (Arm) colocalizes with DE-cad through its interaction with cytoplasmic DE-cad sequences(Tepass et al., 2000). Interestingly, Arm is still enriched at the R3/R4 border where DE-cad expression is low (Fig. 4A,arrows; Fig. 4D,D′),suggesting that Arm interacts with another cadherin. We, thus, analyzed DN-cad1 expression and observed that its distribution within the precluster is largely complementary to DE-cad; DN-cad1 is enriched at R3/R4 cell border membranes (Fig. 4B,C,4E-E″). DN-cad1 is not detected ahead of and close to the MF(Fig. 4C). Later, DN-cad1 is also detected in axonal projections of all photoreceptors as these extend out of the retinal layer (Lee et al.,2001; Prakash et al.,2005). At the apical R3/R4 border, DN-cad1 colocalizes with the atypical cadherin Flamingo [Fmi; a core PCP-signaling and R3/R4 specification factor (Das et al., 2002)](Fig. 4E).

To gain insight as to how the DE-cad and DN-cad1 complementary accumulations are generated within the precluster, we analyzed their localization in loss-of-function and overexpression clones of each other (see Materials and methods). First, we determined the contribution of DE-cad and DN-cad1 to apicolateral Arm localization. DE-cad levels strongly affect Arm levels (Fig. 4G), suggesting that the bulk of endogenous apicolaterally localized Arm is anchored through DE-cad, the exception being the R3/R4 cell border membranes. By contrast,decrease in DN-cad1 and 2 levels (Fig. 4F) has no effect on DE-cad (not shown) or Arm, except at the R3/R4 border, where Arm levels are reduced(Fig. 4F,F′). The DN-cad1 expression pattern is unchanged in DE-cad/shg(shg^P34-1) loss-of-function clones(Fig. 4G,H). Arm is significantly decreased in such clones, but still accumulates apicolaterally between the precluster cells, where DN-cad is present(Fig. 4G). Clones overexpressing DE-cadDN do not affect endogenous DN-cad1 or Fmi (not shown). However, an increase in DN-cad1 expression lowers DE-cad levels(Fig. 4I,I′), without affecting Fmi localization (Fig. 4I,I″; the same is observed anterior to MF, not shown). Both DN-cad1 and DE-cadDN overexpression causes an increase in membrane associated Arm (Fig. 5E; not shown). These data indicate that there is largely mutually exclusive localization of DE-cad and DN-cad1 in wild type. The regulation of this pattern might be in part mediated through a repressive effect of DN-cad on DE-cad, as observed in the gain-of-function DN-cad clones (Fig. 4I). However, it is more complex, as in DN-cad mutant clones DE-cad is not upregulated.

The complementary accumulation of DE- and DN-cad1 within the precluster and the concentration of DN-cad1 at the R3/R4 border prompted us to examine a potential role for DN-cad in rotation. As noted previously, Drosophila has two DN-cadherin genes(Prakash et al., 2005). Although single loss-of-function analysis did not show a retinal phenotype,mosaic eyes of mutants lacking both genes[DN-cad^Δ14 deletes DN-cad1 and DN-cad2(Prakash et al., 2005)]revealed retinal phenotypes, including misrotated ommatidia, symmetrical clusters and loss of photoreceptor cells(Fig. 5A, and not shown).

To gain better insight into the role of DN-cad in rotation, we analyzed N-cad^Δ14 clones in 3rd instar eye discs, during ommatidal rotation and prior to terminal photoreceptor differentiation [also affected in the N-cad^Δ14 mutant(Prakash et al., 2005)]. Interestingly, the disc phenotype of N-cad^Δ14 mutant clusters is the opposite of DE-cad/shg mutant clones, with the majority of mutant ommatidial clusters reaching a 90° angle prematurely as compared with wild-type neighbors (Fig. 5B; Table 1). Therefore, DN-cad function appears to provide the correct rate of movement.

We next tested the effects of DN-cad1 overexpression, driven with sevGal4 (sev>DN-cad). The sev>DN-cad genotype showed rotation defects reflected in ommatidia having rotated to a random degree (Fig. 5F). Random rotation angles are also observed in 3rd instar sev>DN-cad eye discs (Fig. 5C), consistent with the loss-of-function analysis and the suggestion that the DN-cadherins function in controlling the rate of rotation.

To gain further insight into cadherin requirements in ommatidial rotation,we used the dose-sensitive sev>DE-cadDN and sev>DN-cadgenotypes for dominant genetic interaction studies. This approach is often used in Drosophila to identify members of a signaling pathway or interacting factors involved in the regulation of a particular developmental or morphological process (e.g. Boutros et al., 1998; St Johnston,2002). Our initial analysis of sev>DE-cadDN indicated that the phenotype is dose sensitive and modified by decreasing endogenous DE-cad levels (Fig. 3B).

We first tested mechanistic aspects of DE-cad-regulated rotation. The sev>DE-cadDN phenotype is enhanced by dose reduction of an arm null allele (arm^YD35; Fig. 3D), but not an arm allele that maintains its cell architecture function(arm^XM19, only defective for Wg-signaling; Fig. 3E). This suggests that the cytoskeletal anchoring function of DE-cad, via Arm/β-catenin, is crucial for ommatidial rotation, and is supported by several observations. First, the enhancement is, in addition to rotation, in defects associated with ommatidial architecture and adhesion (Fig. 3D; e.g. causing gaps between ommatidia; also Fig. 5G). Second, unlike full-length DE-cad that rescues the rotation defects of sev>DE-cadDN (Fig. 3C), co-expression of sev>DE-cadDN with DE-cadΔβ, which lacks Arm interaction sequences, does not rescue the defects (Table 2). Third, DE-cad lacking the extracellular Cad-repeats (DE-caddEx) does not rescue sev>DE-cadDN when co-expressed(Table 2). Taken together,these data suggest that the cell adhesion and cytoskeletal anchoring aspects of DE-cad are crucial in ommatidial rotation.

To further define the relationship of DE-cad and DN-cad in rotation, we tested their genetic interactions. Whereas sev>DE-cadDN associated under-rotation is enhanced by co-expression of DN-cad, it is not affected by dose reduction of DN-cad loss-of-function alleles(N-cad^Δ14; Table 2). The sev>DN-cad phenotype is enhanced both by DE-cad/shg dose reduction and by co-expression of DE-cad^WT (not shown). The enhancement of sev>DE-cadDN under-rotation by co-expression of DN-cad may be due to competition for Arm (as suggested from DN-cad gain-of-function clones; Fig. 5E,E′). However, the sev>DN-cad rotation phenotype is not enhanced by arm reduction(Fig. 5G) and co-expression of UAS-Arm^S2 (full-length) enhances both the rotation defects and defects in epithelial integrity of sev>DN-cad(Fig. 5H). Taken together,these observations suggest that a proper balance between DE-cad, DN-cads, and Arm levels is crucial for the precise 90° ommatidial rotation.

The direction of rotation is regulated by Fz/PCP signaling(Adler, 2002; Klein and Mlodzik, 2005; Mlodzik, 1999; Strutt and Strutt, 1999). Egfr-signaling is also required for the process, although its specific regulatory input remains unclear (Brown and Freeman, 2003; Gaengel and Mlodzik, 2003; Strutt and Strutt, 2003). We thus tested for genetic interactions of components of these pathways with sev>DE-cadDN.

Previous studies identified shg/DE-cad alleles as enhancers of rotation defects caused by reduction in Egfr signaling(Brown and Freeman, 2003; Gaengel and Mlodzik, 2003) (a decrease in DE-cad enhances the randomness of rotation in Egfr-signaling mutants). Our data show that reduced Egfr-signaling levels (e.g. reduced gene dose of the Egfr-ligand Spitz) specifically enhanced the under-rotation of sev>DE-cadDN (Table 2), suggesting that Egfr-signaling promotes rotation by positively regulating DE-cad function. This is supported by the observation that the gain-of-function Egfr^Elp allele appears to suppress sev>DE-cadDN under-rotation (not shown; owing to the dominant effects of Egfr^Elp on photoreceptor numbers and survival this effect cannot be directly compared with the other interactions). The enhancement of sev>DE-cadDN by pointed(Table 2; pnt, a common nuclear effector of RTK/MAPK pathways) suggests that the entire RTK/MAPK cascade is functioning in this context (see Discussion).

The complementary DE- and DN-cadherin accumulation within the R3/R4 pair,which is crucial for PCP establishment (see Introduction), suggested a possible link between PCP components and cadherin function. Although reduction in Fz and Dsh, the upstream components of Fz/PCP-signaling did not modify sev>DE-cadDN, mutations in the PCP genes stbm and dgo enhanced sev>DE-cadDN, as did reduction of RhoA, an effector of Fz/PCP-signaling downstream of Dsh(Table 2; Fig. 6A,B).

We next wished to determine whether DE-cad/shg during rotation is required in a specific cell or cells within the precluster, and in particular tested for R3/R4 requirement. Mosaic analysis of shg^P34-1clones (in discs) suggested that mosaic ommatidia with both R3 and R4 wild type do not have a higher probability of rotating normally. Largely, the overall number of wild type R-cells correlates with correct rotation: the more wild-type cells are present per cluster, the higher is the probability of normal rotation (see Tables and Fig. S1 in the supplementary material). Furthermore, we have also observed fully wild-type preclusters directly adjacent to mutant interommatidial cells that failed to rotate properly(Fig. 2C-D″; orange arrowheads), suggesting that DE-cad/shg function is required not only in R-cells but also in adjacent interommatidial cells. Nevertheless,expression of sev>DEcad^WT (expressed in R3/R4 at the onset of rotation) increased the number of clusters that initiated rotation in shg^P34-1 clones (Fig. 2E,E′; Table 1), suggesting that sufficient levels of DE-cad in R3/R4 are crucial. Taken together, these data suggest that in rotation DE-cad/shg is required in more cells than the R3/R4 pair, including cells that are adjacent but outside the precluster. Moreover, the genetic interactions suggest that parallel input from PCP and Egfr-signaling regulates DE-cad activity during rotation.

Classical cadherins have multiple links to the cytoskeleton, primarily through α- and β-catenin(Perez-Moreno et al., 2003). Reduction of the Fz/PCP effector RhoA enhanced the sev>DEcadDNphenotype. Mutants in rok, a downstream effector of RhoA, have a rotation-specific loss-of-function eye phenotype, and RhoA has been linked to the regulation of DE-cadherin apical localization and clustering at the membrane (Bloor and Kiehart,2002; Magie et al.,2002; Yap and Kovacs,2003).

Our genetic interactions have revealed that decreasing endogenous RhoA levels enhanced the rotation and adhesion defects of sev>DEcadDN(Table 2; and not shown). To minimize the RhoA-associated defects that are not linked to rotation [e.g. general cellular architecture as RhoA null clones are cell lethal(Strutt et al., 1997)], we expressed UAS-RhoA^IR [lowering endogenous RhoAlevels through RNA interference (Billuart et al., 2001)] in the R3/R4 precursors. Co-expression of sev>DEcadDN with UAS-RhoA^IR causes a specific enhancement of the sev>DE-cadDN under-rotation phenotype(Fig. 6A,B; Table 2). Whereas a general decrease in endogenous RhoA affects cell adhesion (in addition to rotation),the specific RhoA RNAi knock-down in a subset of R-cell precursors at the onset of rotation only enhances ommatidial under-rotation(Table 2; and not shown). Clones of hypomorphic RhoA alleles display an irregular DE-cad localization in eye discs, with malformed preclusters and a general reduction of DE-cad levels outside the precluster(Fig. 6C,C′) (see also Bloor and Kiehart, 2002; Magie et al., 2002; Yap and Kovacs, 2003). Similar results were obtained with dRok² clones (not shown). These data show that in addition to the proposed role for RhoA in DE-cad localization and maintenance of tissue integrity, the dynamic rearrangements of DE-cad and DN-cad during rotation require RhoA function. Fz/PCP-signaling thus probably regulates DE-cad function and/or localization through RhoA-Rok. The fact that we do not observe gross mislocalization of DE-cad in mutant clones of primary PCP genes (e.g. fmi or stbm; not shown)suggests that effects of RhoA on DE-cad localization/activity in rotation are subtle, influenced by several inputs and not affecting tissue integrity.

Here, we have demonstrated that a balance of the classical cadherins DE-cad and DN-cad regulate ommatidial rotation during Drosophila eye development. DE-cad promotes the movement of photoreceptor preclusters relative to the surrounding epithelial cells and DN-cadherins act to restrict the rate of movement. Furthermore, the data indicate that DE-cad is affected by PCP and Egfr signaling, thus integrating input from several pathways. A similar requirement for DE-cad in cell movement is also observed in border cell migration during Drosophila oogenesis(Niewiadomska et al., 1999). We propose that such conserved molecular mechanisms drive cell movement processes in general during development and disease (see below).

Although the role for DE-cad in tissues undergoing rearrangements during development is established (Tepass et al.,2000; Perez-Moreno et al.,2003), a direct role for DE-cad in cell and tissue movement has been more difficult to study in vivo owing to its essential role in maintenance of epithelial integrity. Analysis of adult eye phenotypes of a homozygous viable shg/DE-cad allele and a dominant-negative DE-cad construct (DE-cadDN), expressed in the R3/R4 and later R1/R6, R7 precursors, indicate that DE-cad is required throughout the rotation process. The ability of ommatidia to complete the precise 90° rotation directly depends on DE-cad activity. Both the extracellular domain, responsible for cell-cell adhesion, and the intracellular domain, linking DE-cad to the actin cytoskeleton, are required for rotation. DE-cad associates with the actin cytoskeleton primarily through interactions with Arm/β-catenin. Althoughβ-catenin has a dual role in cell adhesion and Wg signaling [which can be separated (Orsulic and Peifer,1996; Sanson et al.,1996)], our data indicate that during ommatidial rotationβ-catenin acts through its role in cell adhesion.

Ommatidial rotation represents the final step in establishing PCP during eye development. The direction of rotation depends on proper R3/R4 cell fate specification, which is determined by PCP signaling. The Egfr pathway and input by rotation-specific genes, e.g. nemo, are thought to function in parallel to Fz-PCP signaling. We observed an enhancement of the sev>DE-cadDN rotation defects by dose reduction in core regulatory PCP genes dgo and stbm(Table 2; see Fig. S1 in the supplementary material; ommatidial under-rotation and the number of ommatidia that did not initiate rotation in sev>DE-cadDN/dgo^-/+, stbm^-/+ was comparable with the enhancement of sev>DE-cadDN by heterozygosity of a shg null allele). The localization of PCP protein complexes at the level of adherens junctions (Djiane et al., 2005; Wu et al.,2004) is consistent with the idea that PCP factors can influence DE-cad function. The mechanism of this regulation remains unclear. The RhoA-RNAi transgene, which was expressed only in R3/R4 precursors during the initiation of ommatidial rotation, enhanced sev>DE-cadDN associated under-rotation defects. Although a RhoA requirement in multiple cellular processes makes it difficult to dissect its specific role in rotation, the specificity of the phenotype (enhanced under-rotation in sev>DEcad/RhoA^IR) suggests a role for RhoA in the regulation of cadherin-mediated cell movement.

Although Egfr signaling appears to be required for the precise 90°rotation (Brown and Freeman,2003; Gaengel and Mlodzik,2003; Strutt and Strutt,2003), its role in the process -promoting motility or antagonizing it -has remained unclear. Our genetic data suggest that Egfr signaling acts positively to promote rotation, as a reduction in Egfr signaling enhances the sev>DE-cadDN under-rotation phenotype(Table 2, Fig. 7). This may reflect a positive role for Egfr signaling in the regulation of DE-cad activity or turnover at the membrane, as suggested from human tumor cell lines(Lu et al., 2003). Affecting the function of endocytic pathway components can also have an effect on ommatidial rotation (Weber et al.,2003). This might be mediated by Egfr signaling, as is thought to be the case in human cancer cells, leading to recycling and redistribution of E-cad at the plasma membrane (Naora and Montell, 2005).

It is likely that, besides PCP and Egfr signaling, additional pathways act in parallel and/or redundantly to regulate this cell motility aspect. A number of genes essential for rotation, including nemo and scabrous, have not yet been placed within a particular signaling pathway, suggesting that rotation results from spatiotemporal integration of multiple signaling events, emanating either from within the precluster or from the dynamic cell context within the moving furrow.

Drosophila DN-cadherins, which are encoded by the adjacent cadN and cadN2 genes, are the main cadherins expressed in the nervous system. In developing photoreceptors they participate in axon guidance, and in pupal eye discs they mediate terminal patterning of the retina [through specific expression in cone cells(Hayashi and Carthew, 2004; Iwai et al., 1997; Lee et al., 2001; Prakash et al., 2005)]. During PCP establishment, DN-cad1 is concentrated at the border between R3/R4 precursors, in a pattern largely complementary to DE-cad(Fig. 4B,C,E). This suggested a possible combinatorial role for DE-cad and DN-cad in rotation, with DN-cad either providing a structural role in rotating clusters, or participating in signaling cascades that regulate cell movement. Analysis of DN-cadmutant clones in discs during rotation demonstrated a specific function, as many mutant clusters have completed rotation well before wild-type clusters of the same stage (Fig. 5B-B″). These data indicate that DN-cadherins function to slow down rotation, serving an opposing function to DE-cad.

The balance and complementary distribution of DE-cad and DN-cad appear crucial for correct rotation to occur. Mild overexpression of DN-cad1 in R3/R4(sev>DN-cad) is sufficient to interfere with the process,possibly by affecting DE-cad levels. Consistently, DN-cad1 overexpression enhances sev>DE-cadDN induced under-rotation and overexpression clones of DN-cad1 cause a decrease in endogenous DE-cad levels(Fig. 4I). Alternatively, the negative effect of DN-cad on DE-cad might be through competition forβ-catenin, as sev>DE-cadDN is partially rescued by UAS-Arm^S2 (full length; not shown), although as sev>DN-cad is not enhanced by arm dose reduction this appears less likely. Interestingly, sev>DN-cad is enhanced by co-expression of full-length DE-cad and full-length Arm. These phenotypes resemble those of a strong sev>DN-cad line (not shown), suggesting that DN-cad is stabilized by increased levels of available Arm, and also that co-overexpression of two cadherins may interfere with optimal turnover rate at the membrane.

In Drosophila `clone 8' (derived from wing discs) cell culture experiments, increased levels of DE-cad cause an increase in Arm levels without changing its mRNA levels (Yanagawa et al., 1997). This is consistent with our data, as overexpression of DN-cad induces apparent accumulation of endogenous Arm in the cell, largely at the cell membrane.

Previous work has identified some signaling input into ommatidial rotation,e.g. PCP and Egfr signaling. Here, we begin to address mechanistic aspects of rotation, defining the involvement of cell adhesion molecules that are linked to cytoskeletal elements.

Ommatidial rotation shares similarities with border cell migration (BCM)during Drosophila oogenesis(Niewiadomska et al., 1999). In both cases, DE-cad is upregulated in the cell cluster as the respective cells are being recruited from surrounding epithelial cells, and as migration is initiated, DE-cad is downregulated at the membrane between the migrating cells and surrounding `substrate' cells. Egfr signaling is required positively for both processes (Duchek and Rorth,2001) (this work). Thus, although the regulatory role for Egfr in rotation is complex, our data suggest that both Egfr and PCP components promote DE-cadherin mediated rotation. Other RTKs may act redundantly in promoting rotation (I.M. and M.M., unpublished), as they do in BCM(Duchek et al., 2001). Although the role of DN-cad in BCM remains unclear, existing data suggest that it may serve an inhibitory function there. DN-cad1 is detected in all epithelial follicular cells from which border cells are recruited but is lost after mid-oogenesis when BCM initiates(Tanentzapf et al., 2000). Our data suggest that DN-cad1 is initially present, albeit reduced, in border cells, but becomes undetectable during BCM (I.M. and M.M., unpublished).

A positive role for DE-cad in migration is not limited to Drosophila. A recent study addressing zebrafish gastrulation finds a similar role for E-cad during cell migration(Montero et al., 2005). Furthermore, BCM (Naora and Montell,2005) and ommatidial rotation (this work) share similarities with the progression of ovarian carcinoma(Naora and Montell, 2005). Ovarian cancer cells tend to cluster and metastasize as cell aggregates that show upregulation in E-cad expression. The proposed role for RTK signaling in the three processes (BCM, ommatidial rotation and ovarian carcinoma metastasis) suggests an evolutionarily conserved mechanism that can switch the function of E- and N-cad between promoting and inhibiting cell movement.

We are grateful to H. Oda, U. Tepass, T. Uemura, P. Rorth, T. Clandinin, Z. Chen, K. Saigo, K. Basler, L. Luo, F. Schoeck and the Bloomington stock center for flies and reagents; and to K. Gaengel for help with Fig. 1. We thank G. Das, D. Del Alamo, J. Delaney, T. Klein, J. Reynolds-Kenneally and U. Weber for comments on the manuscript; members of the Mlodzik laboratory for useful discussions;and C. Tan and M. Burnett for technical help. This work was supported by NIH/NEI grant RO1 EY14597 to M.M.