Strabismus requires Flamingo and Prickle function to regulate tissue polarity in the Drosophila eye

The organization of tissues and organs based upon a precise body plan is a universal theme in metazoan development. Polarity within a cell is essential for cells to perform specialized functions. Similarly, the polarized orientation of cells within an epithelium, a phenomenon known as tissue,epithelial, or planar cell polarity, shapes the tissue into a functional organ. Until recently, genes regulating these processes have been studied almost exclusively in Drosophila. The proteins encoded by these tissue polarity genes impose a high degree of order in a variety of epithelial structures, including the uniform orientation of bristles and hairs throughout the adult body, the polarized organization of tarsi in the legs, and the precise patterning of unit eyes, or ommatidia, in the compound eye. A growing number of tissue polarity genes are being identified, yet the mechanisms by which the proteins encoded by these genes function and the intricate network of interactions that connects these genes are poorly understood.

We have identified interactions between a subset of tissue polarity genes in the developing Drosophila eye. The compound eye of the fly is a polarized epithelium composed of approximately 800 unit eyes, or ommatidia. Each ommatidium contains 20 cells, including eight photoreceptors (R1-R8). The photosensitive organelles of the photoreceptors, the rhabdomeres, are arranged in a characteristic trapezoid in which photoreceptor R3 defines the `point' of the trapezoid. There are two chiral forms of the trapezoid and they fall on opposite sides of a dorsal-ventral midline known as the equator. In the dorsal hemisphere of the eye, the points of the trapezoids face the dorsal margin while those in the ventral half face the ventral margin of the eye(Fig. 1A).

This highly ordered adult pattern is established in the eye imaginal disc of the third larval instar. Initially, nascent photoreceptor clusters are uniformly oriented within this epithelium. An essential step in polarizing the retinal epithelium is a break in symmetry between the future photoreceptors R3 and R4 as they adopt distinct cell fates. The photoreceptor clusters subsequently rotate 90° within the epithelium — they rotate counterclockwise in the dorsal half and clockwise in the ventral half of the eye (reviewed by Wolff and Ready,1993).

A number of tissue polarity genes have been identified, among themfrizzled (fz), prickle (pk),dishevelled (dsh), diego (dgo),strabismus (stbm; also known as Van Gogh) andflamingo (fmi; also known as starry night)(Zheng et al., 1995;Gubb et al., 1999;Klingensmith et al., 1994;Feiguin et al., 2001;Wolff and Rubin, 1998;Taylor et al., 1998;Usui et al., 1999;Chae et al., 1999). We have focused our efforts on identifying the position of stbm in the tissue polarity pathway as a means of more precisely defining its role in setting up polarity in the eye. Flies null for stbm lack an equator because of a variety of defects in ommatidial orientation and fate misspecification(Wolff and Rubin, 1998).stbm acts cell-autonomously to define R4(Wolff and Rubin, 1998), and it co-localizes with other tissue polarity proteins at the contact between photoreceptors R3 and R4 (this report)(Strutt et al., 2002).

To identify genes that interact with stbm, we carried out a genetic modifier screen. We identified two tissue polarity genes, fmiand pk, that dominantly modify the stbm mutant phenotype. The role of fmi in tissue polarity was identified from its requirement for polarization of wing hairs(Usui et al., 1999).fmi also plays an essential role in the first asymmetric cell division of the SOP cell lineage in the PNS(Lu et al., 1999).fmi encodes a protein with a seven-pass transmembrane domain, a unique cytoplasmic tail and nine extracellular cadherin domains(Usui et al., 1999;Chae et al., 1999). The extracellular cadherin domains are capable of mediating cell-cell adhesion while the unique intracellular domain is potentially involved in signal transduction (Usui et al.,1999).

In this paper, we define a role for fmi in directing tissue polarity in the Drosophila retina. We show that loss-of-functionfmi interacts genetically with both misexpression and loss-of-function stbm. We have generated an antibody against Stbm and show that Stbm is apically localized in all cells anterior to, in, and several rows posterior to, the morphogenetic furrow. Stbm subsequently fades in R8, R2 and R5 and becomes pronounced at the contact between R3 and R4 [also reported using Sbm-YFP by Strutt et al. (Strutt et al., 2002)]. We also show that Fmi and Stbm co-localize in early,but not later, stages of ommatidial development. In addition, we show thatfmi is cell- and ommatidium-autonomously required for ommatidial polarity [as also reported by others (Yang et al., 2002; Das et al.,2002; Strutt et al.,2002)].

Like fmi, pk also acts globally to influence polarity throughout the fly (reviewed by Mlodzik,2000). Pk has several protein interaction domains and binds Dsh(Tree et al., 2002). Tree et al. have suggested this interaction is an essential component of a feedback loop that asymmetrically localizes Fz and Dsh in wing cells, ultimately leading to the polarized arrangement of hairs and bristles. Similarly, in the eye, pk is essential in establishing Fz asymmetry in R3 and R4(Strutt et al., 2002). Taylor et al. have demonstrated a genetic interaction between pk andstbm in the wing (Taylor et al.,1998). Here we show that pk enhances both misexpression and loss-of-function stbm phenotypes in the eye, and that localization of Pk is disrupted in stbm mutant ommatidia.

Fly strains used: sev-stbm^14-1, stbm¹⁵³,stbm^6cn, FRT42, fmi¹⁹²/CyO (a genetic null allele identified in a screen for novel tissue polarity mutants, T. Wolff,unpublished), fmi^frz3 (a gift from D. Gubb),fmi^E59 (a gift from T. Uemura), w,hsFLP122;FRT42,eyGAL4, UASFLP;FRT42 GMRhid, shi^2ts,hk¹¹ (a gift from H. Kramer), fz^KD4A,N^ts1, pk^pk1, pk^eq (a gift from T. Xu). Unless otherwise noted, phenotypic analyses were conducted using FRT42fmi¹⁹². fmi¹⁹² mosaic clones were generated using heat shock-induced, FLP/FRT-mediated recombination(Xu and Rubin, 1993). Entirely mutant fmi eyes were generated using the EGUF(eyeless-Gal4 UAS-FLP)system of recombination (Stowers and Schwarz, 1999), a modification of the FLP-FRT system in which recombination is driven selectively in the eye by the eyelesspromoter and wild-type cells are killed by the expression of GMR-hidand uncharacterized cell lethal mutations.

sev-stbm homozygous flies were crossed to the 250 second and third chromosome deletion lines that constitute the Bloomington Deficiency Kit. The degree of eye roughness in F1 transheterozygotes was analyzed under the dissecting microscope and compared to that of sev-stbm heterozygotes. A secondary screen, in which adult eyes were fixed and sectioned (as described by Wolff, 2000) and the phenotype quantified, was conducted on 27 candidate interactors. Six enhancers of sev-stbm were confirmed.

Adult eyes were fixed, embedded and sectioned according to standard protocol (Wolff, 2000). The number of ommatidia and eyes scored is as follows:Df(2R)E3363/sev-stbm^14-1: 383 ommatidia from 5 eyes;Df(2R)Jp4/sev-stbm^14-1: 480 ommatidia from 3 eyes; Roote 276/sev-stbm^14-1: 533 ommatidia from 3 eyes; Roote 2-42/sev-stbm^14-1: 557 ommatidia from 3 eyes;Df(2L)s1402/sev-stbm^14-1: 608 ommatidia from 3 eyes;Df(2R)pk78k/sev-stbm^14-1: 497 ommatidia from 3 eyes;fmi^frz3/sev-stbm^14-1: 1135 ommatidia from 14 eyes; fmi¹⁹²/sev-stbm^14-1: 877 ommatidia from 15 eyes; fmi^E59/sev-stbm^14-1: 872 ommatidia from 12 eyes; fmi^frz3/fmi^frz3: 1788 ommatidia from 17 eyes;stbm¹⁵³,fmi^frz3/stbm¹⁵³,fmi^frz3:1024 ommatidia from 15 eyes;stbm¹⁵³,+/stbm¹⁵³,fmi^frz3:1123 ommatidia from 16 eyes;+,fmi^frz3/stbm¹⁵³,fmi^frz3: 976 ommatidia from 10 eyes; EGUF-fmi¹⁹²: 1612 ommatidia from 18 eyes; fmi¹⁹²/fmi¹⁹² mosaic clones: 792 ommatidia from 23 clones in 23 eyes;pk^pk1/sev-stbm^14-1: 1030 ommatidia from 11 eyes; and pk^pk1, stbm¹⁵³/pk^pk1,stbm¹⁵³: 729 ommatidia from 7 eyes.

The anti-Stbm polyclonal antibody was raised against the N-terminal 143 amino acids of the protein. A 429 base pair PCR product was generated and subcloned in frame into the EcoRI/XhoI site of pGEX-4T-1(Pharmacia). The fusion protein was purified on glutathioneagarose beads and used to immunize rabbits. Immunization and subsequent production were carried out by Pocono Rabbit Farm and Laboratory, Inc.

Third instar larval eye discs were dissected and processed as described previously (Wolff, 2000). Tissue was incubated in primary antibody overnight at 4°C at concentrations of 1:10 for anti-Fmi [mouse monoclonal; generous gift from T. Uemura (Usui et al., 1999)],1:600 for anti-Pk [rabbit polyclonal; generous gift from D. Tree and J. Axelrod (Tree et al., 2002)],1:500 for anti-Stbm (rabbit polyclonal, see above) and 1:10 for anti-Armadillo(Arm) antibody (mouse monclonal, Developmental Studies Hybridoma Bank). Secondary antibodies conjugated to Alexafluor fluorescent dyes were used(Molecular Probes).

Fmi localization was studied in fz^KD4A null larval escapers. For immunostaining of Fmi in shi^2ts animals,third instar larvae were heat shocked at 31°C for 1 hour, eye discs removed and fixed immediately and immunostained as described above. Recovery experiments were conducted by allowing the larvae to recover at room temperature for 1, 2 or 6 hours following heat shock.

pk^eq has not been characterized at the molecular level,however it fails to complement known pk^pk-sple alleles. This allele was chosen over pk^pk1 to characterize Stbm localization because it produces an obvious eye phenotype, unlike thepk^pk1 allele.

Fluorescent images were collected using a Leica TCS SP2 confocal microscope.

The Bloomington deficiency kit was screened to identify genes that interact with the tissue polarity gene, stbm. The screen was carried out in a misexpression stbm background in which the sevenless(sev) promoter was used to drive expression of stbm(sev-stbm) in photoreceptors R3, R4 and R7 and the four cone cells. The phenotype of eyes of flies carrying the sev-stbm transgene mimics the loss-of-function stbm phenotype, although it is milder (Wolff,unpublished): in transgenic flies carrying a single copy of sev-stbm,10% of ommatidia exhibit disruptions in polarity(Fig. 1B). Externally, these eyes appear mildly rough, so both enhancement and suppression of this phenotype can be readily detected on the surface of the eye.

In an F₁ genetic modifier screen, 250 deletion lines from the Bloomington deficiency kit, which uncover approximately 70-75% of the second and third chromosomes (Berkeley Drosophila Genome Project), were crossed to sev-stbm and the progeny scored for dominant modification of the sev-stbm phenotype. Six deficiency lines were identified as dominant enhancers of the sev-stbm phenotype(Table 1); no suppressors were identified. We have identified the interacting gene in two of these six deletions (Bloomington deficiencies Df(2R)E3363 and Roote 276).

Df(2R)E3363, which uncovers chromosomal region 47A-47F, enhances thesev-stbm phenotype three-fold, increasing the percentage of ommatidia with polarity defects from 10% to 31%(Table 1). The most likely candidate for interaction within this region was a second tissue polarity gene, fmi, which maps to 47B4. To determine if fmi was the gene responsible for enhancement of the sev-stbm phenotype, three alleles, fmi^frz3 (a hypomorph), fmi¹⁹²and fmi^E59 (genetic nulls) were crossed tosev-stbm. These three alleles dominantly enhance thesev-stbm phenotype to the same extent as the original deficiency(Fig. 1C-E;Table 2), suggesting thatfmi is the interacting locus.

Genetic interactions in a misexpression background can be unreliable, as the observed effect may be the consequence of, for example, non-specific effects on the promoter. To definitively demonstrate that fmiinteracts genetically with stbm, the interaction was confirmed in astbm loss-of-function background in which a recombinant line carrying the hypomorphic alleles stbm¹⁵³ andfmi^frz3 was analyzed. The fmi^frz3phenotype is dominantly enhanced approximately threefold instbm¹⁵³, fmi^frz3/+,fmi^frz3 flies, confirming the misexpression result(Fig. 2A,B;Table 3). In contrast, the presence of a single copy of fmi^frz3 in astbm¹⁵³ homozygous background has no effect on thestbm¹⁵³ phenotype (data not shown). Phenotypes that are never seen in fmi^frz3 or stbm¹⁵³homozygotes appear in the double homozygotes. In these flies, viability drops from 100% to about 5%, there is a significant increase in the number of symmetric ommatidia (both R3/R3 and R4/R4), and lastly, some ommatidia are missing photoreceptors (Table 3).

Consistent with its previously characterized roles in establishing polarity in cuticular structures, we report that fmi is also required to establish ommatidial polarity. The fmi mutant eye phenotype resembles that of previously described tissue polarity mutants, such as fz, pk, dsh,dgo and stbm (Zheng et al.,1995; Gubb et al.,1999; Klingensmith et al.,1994; Feiguin et al.,2001; Wolff and Rubin,1998). Since null fmi alleles are lethal, we examined the mutant phenotype in FLP/FRT-generated clones and found that 37% of genetically mutant or mosaic ommatidia exhibit disrupted polarity(Fig. 3A,B;Table 4). The majority of mutant ommatidia display inversions on the anterior-posterior (AP) axis and the dorsal-ventral (DV) axis, although ommatidia that are inverted on both their AP and DV axes, as well as partially rotated ommatidia, symmetrical ommatidia (both R3/R3 and R4/R4) and unrotated ommatidia, are also present. It is interesting to note that the phenotype is stronger in clones that lie on or very close to the equator (Fig. 3B). Recently, the fmi eye polarity phenotype was also reported by others (Yang et al.,2002; Das et al.,2002; Strutt et al.,2002). The data reported here are qualitatively consistent with data presented in these reports.

The phenotype described above is similar to that seen in eyes entirely mutant for fmi (Fig. 3C, Table 4). Furthermore, the phenotype of the viable hypomorphic allelefmi^frz3 is similar to, but weaker than, that described for null alleles (Fig. 2A;Table 3).

Das et al. (Das et al.,2002) reported that fmi^E59 mutant clones have a cell death phenotype in which 20% of fmi ommatidia lack photoreceptors. This is in contrast to observations that neitherfmi^E45/fmi^E59 transheterozygotes(Strutt et al., 2002) nor ommatidia in fmi¹⁹² clones (reported here) exhibit a photoreceptor death phenotype. While the basis of this difference is not known, it is interesting that stbm, fmi double homozygotes are missing photoreceptors (Table 3), perhaps suggesting a previously unrecognized role for these loci in photoreceptor specification or survival. [A small percentage of ommatidia are missing photoreceptors in EGUF-fmi eyes, however this is an artifact of the EGUF system(Rawls et al., 2002).]

Gene products that act at a distance generally exert their influence in a non-autonomous fashion, while gene products that exert their effects intracellularly act autonomously. fmi has been shown to act autonomously in the wing (Chae et al.,1999). Since tissue-specific differences have been observed in the autonomy of some gene products (for example, fz), we analyzed the polarity of ommatidia in and near fmi mutant clones to confirm thatfmi also acts autonomously in the eye. This analysis demonstrated that the presence or absence of fmi does not affect neighboring ommatidia, suggesting that fmi acts autonomously within ommatidia. In other words, genetically mutant and mosaic ommatidia have no effect on wild-type ommatidia outside the clone, nor does wild-type tissue rescue ommatidia that are mutant or mosaic for fmi(Fig. 3A,B). Similar findings reported by Das et al. support the autonomous requirement for Fmi in developing ommatidia (Das et al.,2002).

To determine if fmi acts in one specific photoreceptor to establish the orientation of an ommatidium, we carried out an analysis of mosaic ommatidia, ommatidia that contain a mixture of wild-type and mutant photoreceptors. We scored these ommatidia to determine the requirement for Fmi in each photoreceptor to direct polarity and found that Fmi is not required in any specific photoreceptor(s) for normal rotation; rather, if at least one photoreceptor expresses fmi, the ommatidium can, but not necessarily will, rotate correctly (data not shown). As with stbm, proper rotation is only guaranteed when all photoreceptors within an ommatidium are wild type for fmi. Das et al. (Das et al., 2002) propose that Fmi function in R3 and R4 is necessary and sufficient for polarity. While this may be true, we cannot rule out the possibility that Fmi may also play a role in the remaining six photoreceptors since we have seen a small fraction of mosaic ommatidia in which both R3 and R4 are wild type for fmi, yet these ommatidia still rotate incorrectly (data not shown).

To determine if Fmi is required in any specific photoreceptor for cell fate specification, we examined developmental pairs of photoreceptors (R1/R6, R2/R5 and R3/R4) that were mosaic within the pair. If Fmi is required to specify the R3 photoreceptor, for example, then there would be a trend such that in the majority of mosaic R3/R4 pairs, the photoreceptor that expresses fmiwould become the R3. We do not see any such trends for any of the pairs,consistent with the findings of Das et al.(Das et al., 2002), suggesting that Fmi is not required for binary photoreceptor fate decisions in developing ommatidia.

Stbm is expressed in a dynamic pattern in the third larval instar. Anterior to, in, and immediately posterior to the morphogenetic furrow, Stbm is uniformly expressed on the apical membranes of all cells(Fig. 4A-C). [In the discussion that follows, row numbers are as defined by Wolff and Ready(Wolff and Ready, 1993); each row is equivalent to 1.5-2 hours of development.] Four to five rows posterior to the furrow, at about the time ommatidial rotation first becomes apparent,Stbm begins to undergo an intriguing change in its pattern of localization. First, it becomes prominent at the membranes of photoreceptors R3 and R4,except where they contact R2 and R5, respectively, while simultaneously dropping to undetectable levels in photoreceptors R8, R2 and R5 [this stage is also described by Strutt et al., using Stbm-YFP(Strutt et al., 2002)]. Second, no protein is detectable at the interfaces between photoreceptors R3/R2 or R5/R4. A restricted region of Stbm staining is evident at the posterior tip of R8 where it contacts R1, R7 and R6, and likely reflects the presence of Stbm in R1, R7 and R6, but not in R8(Fig. 4D-F). While the level of resolution of these images does not reveal if Stbm is present in only R3 or R4, or in both cells, studies of Stbm-YFP mosaic clones have demonstrated that Stbm is present only in R4 at the R3/R4 boundary(Strutt et al., 2002).

Given that photoreceptors R3 and R4 seem to act as the predominant compass in determining ommatidial polarity in the eye, it is intriguing that Stbm remains abundant at the interface between these two cells. It is equally interesting that Stbm is removed from photoreceptors R8, R2 and R5. At this point it is not known if removal of Stbm from these photoreceptors is essential to achieve normal ommatidial polarity.

Later in development, Stbm is localized in the cone cells. Initially, it is most prominent at the points of contact between the cone cells. It continues to be expressed at high levels in the cone cells once they meet centrally(Fig. 4G-I). The functional relevance of Stbm at these sites is not obvious since cone cell assembly is virtually normal in stbm mutant eyes (a small percentage of ommatidia have only three cone cells, but this phenotype may be a secondary effect of improper recruitment by the underlying photoreceptors). No anti-Stbm staining is evident in eye discs that are null for stbm(stbm^6cn) (data not shown).

Fmi co-localizes with Stbm anterior to, within and for several rows posterior to the furrow (Fig. 5A, parts a,b; 5B, parts a,b). [The pattern of Fmi localization was also recently reported by others (Yang et al.,2002; Das et al.,2002; Strutt et al.,2002); here, we extend these observations by providing a developmental time-course for the dynamic distribution of Fmi.] The co-localization persists until approximately seven rows posterior to the furrow, at which point the patterns diverge in two intriguing ways. First,approximately 7-8 rows behind the furrow, Fmi becomes diminished in photoreceptor R3 and simultaneously becomes enhanced in photoreceptor R4(Fig. 5Ac,Bc). As an intermediate step in this change in protein distribution, Fmi becomes weaker in the polar region of R3, resulting in a transient asymmetry in which Fmi is more prominent in the equatorial region of R3. It is not clear why Fmi undergoes a shift from expression in both R3 and R4 to expression in R4 alone. Perhaps there are Fmi-dependent qualities to being an R4 that cannot be detected in a standard mosaic analysis, such as the placement or morphology of the cell's rhabdomere.

The second way in which Fmi localization differs from that of Stbm is that two to four rows after Fmi becomes conspicuous in R4, Fmi protein becomes internalized into large, punctate structures, whereas Stbm does not(Fig. 5Ad,Bd). These vesicular structures are found in the photoreceptor cell bodies, slightly above the level of the R3 and R4 nuclei. Based on size (approximately 700-800 nm), they resemble multivesicular bodies (MVBs). The majority of these vesicles accumulate centrally within the ommatidium at the junction where photoreceptors R8, R2, R5, R3 and R4 meet, although a small number are evident at the anterior end of the ommatidium in R3 and/or R4. The internalization of Fmi is not essential for rotation, since this event takes place once rotation is well underway.

The predominance of Fmi-containing vesicles at the junction of photoreceptors R8, R2, R5, R3 and R4 suggested that this internalization may be the means by which Fmi is removed from these cells, or at least from a subset of these cells. In an effort to identify the process underlying this internalization, we tested the efficiency of this process in two mutants that interfere with the endocytic pathway: shibire (shi) andhook (hk). We found that the vesicularization of Fmi is altered in these mutants, indicating that Fmi is internalized via endocytosis.

shi, which encodes the Drosophila dynamin, is required early in the endocytic pathway for the budding of clathrin-coated pits from the membrane upstream of the fusion of these structures with endosomes(Chen et al., 1992) (reviewed by Narayanan and Ramaswami,2001). Temperature-sensitive shi^2ts larvae were heat shocked for 1 hour at the restrictive temperature, sacrificed immediately and immunostained with an antibody against Fmi. Inshi^2ts larvae, the MVB-like, Fmi-containing vesicles normally found in wild type (Fig. 6A) are abolished; Fmi is instead found in small puncta on cell membranes (Fig. 6B). The large Fmi-containing vesicles reappear in larvae allowed to recover for 1-6 hours at room temperature (data not shown).

hk, which encodes a novel component of the endocytic pathway, acts downstream of shi in this pathway and is required for the formation and/or maintenance of MVBs (Kramer and Phistry, 1996; Kramer and Phistry, 1999). In hk¹¹ mutants, the large,MVB-like vesicles are absent; instead, Fmi is localized to smaller cellular puncta at the junction where the large vesicles normally localize(R8/R2/R5/R3/R4) (Fig. 6C). The failure of Fmi to accumulate in vesicles in shi and hkmutants suggests that the Fmi-containing vesicles in wild-type eyes result from the endocytosis of Fmi. While the internalization of Fmi into vesicles is dependent on endocytosis, earlier changes in distribution of the protein (for example, its removal from R8, R2, R5 and subsequent accumulation in R4) are not.

The functional significance of Fmi endocytosis in the eye is not known. Clearly, this internalization is taking place too late to initiate or mediate rotation. Perhaps it is necessary for rotation to stop. It could also be important for other aspects of development given that the endocytosis of membrane-associated receptors is required for signaling in key developmental pathways [for example, Notch, Dpp, and Wg(Parks et al., 2000) (reviewed by Narayanan and Ramaswami,2001)].

The regulation of Fmi localization in the larval eye disc shows a dependency on fz (Fig. 5C,D) (Strutt et al.,2002; Das et al.,2002) and Notch (N) (data not shown)(Das et al., 2002), genes implicated in R3 and R4 cell fate determination, respectively. The dependency of Fmi localization on fz has also been described for Fmi localization in the wing (Usui et al.,1999; Chae et al.,1999).

The early pattern of Fmi localization is unaffected in the absence of Fz— it is still localized to all cell membranes anterior to the furrow(data not shown) and in nascent photoreceptor clusters(Fig. 5Ca,Da). Furthermore,slightly later in development, Fmi is still abundant in photoreceptors R3 and R4. However, whereas Fmi would ordinarily be removed from photoreceptors R8,R2 and R5 at this stage in wild type, it is only partially removed from these cells in the fz^KD4A mutant(Fig. 5Cb,Db). The most notable change in Fmi localization is that it no longer accumulates asymmetrically in R4 (Fig. 5Cc,Dc). The size,number and location of Fmi-containing vesicles are also disrupted infz^KD4A larvae: there are more vesicles, they are smaller and they accumulate approximately four rows earlier in development(Fig. 5Cd,Dd). We observe similar defects in Fmi localization in N^ts1 larvae heat-shocked for 6 hours (data not shown). Additionally, Das et al.(Das et al., 2002) show that Fmi localization is also perturbed when N-mediated signaling is knocked down via overexpression of the sev-Su(H)-EnR transgene. While these data do suggest a role for N in the asymmetric localization of Fmi, one cannot yet be assigned, given the abundance of roles for N throughout development.

The observations that fmi and stbm have similar phenotypes, that they interact genetically and that their products colocalize,suggested that they may act in the same pathway to specify tissue polarity. To explore the possibility that Stbm and Fmi define a complex, we investigated both the localization of Fmi in a null stbm background and the localization of Stbm in EGUF-fmi eyes. In neither case was the localization affected (data not shown), demonstrating that Stbm is not required for Fmi localization, nor is Fmi required for Stbm localization. Furthermore, we have been unable to demonstrate a physical interaction between Fmi and Stbm using co-immunoprecipitation assays (data not shown).

In the deficiency screen described earlier, a second tissue polarity gene,pk, was identified as a dominant genetic modifier of stbm. The original deficiency, Roote 276, which uncovers 42E4-43E7, enhances thesev-stbm phenotype from one in which 10% of ommatidia have defects in polarity to one in which 27% have defects(Table 1). The best candidate interactor was pk, a tissue polarity gene that maps to 42F2-43A1. We demonstrated that pk was the gene responsible for the dominant enhancement of the sev-stbm phenotype: haploinsufficiency ofpk^pk1, a pk allele with no eye phenotype,enhances the sev-stbm phenotype to the same degree as the original deficiency (Fig. 1F;Table 2). We confirmed this genetic interaction in a loss-of-function stbm background: the percentage of symmetrical defects in stbm¹⁵³,pk^pk1 double homozygotes is significantly enhanced relative tostbm¹⁵³ homozygous flies(Fig. 2C,D;Table 3).

The genetic interaction between stbm and pk may have its basis in a physical interaction that enhances or stabilizes these proteins at the R3/R4 boundary. To explore this possibility, we examined Stbm localization in a pk mutant background, and Pk localization in a stbmmutant background. Stbm localization does not appear to be affected in apk^eq background (a genetic null that fails to complementpk^pk-sple alleles, data not shown). However, Pk localization is disrupted in a stbm^6cn null background. We have characterized the distribution of Pk in wild-type eye imaginal discs(Fig. 7B-D) and find that it is indistinguishable from that of Stbm (Fig. 4). Pk is significantly reduced overall in thestbm^6cn background. While some protein does accumulate at the boundary between R3 and R4, Pk is not detectable at the R8/R1/R7/R6 boundary (Fig. 7E-G). Physical interactions have not been demonstrated between either of these proteins, nor have genetic interactions between fmi and pk been shown. These data are consistent with the possibility that Stbm, Fmi and Pk may all function together in a complex.

The proteins encoded by the tissue polarity genes, fz, pk, dsh, dgo,stbm and fmi are thought to make up a dynamically regulated membrane-associated complex. The asymmetric localization of this complex is thought to be required for the regulation of Fz and N activity, which ultimately results in the appropriate specification of photoreceptors R3 and R4 (Strutt et al., 2002;Das et al., 2002). This pathway also regulates the polarization of wing hairs, although there appear to be tissue-specific differences (Tree et al.,2002).

In an attempt to define more precisely the role of Stbm in the tissue polarity pathway, we have identified genetic interactions betweenstbm and two other tissue polarity genes, fmi andpk. Characterization of the fmi-stbm interaction revealed a requirement for Fmi in ommatidial polarity and a dynamic pattern of Fmi localization that depends on Fz and N. We have also raised an antibody against Stbm, characterized its subcellular localization, and shown that the localization of Fmi and Stbm differs in two ways: first, Fmi is enriched in R4, whereas Stbm is not, and second, Fmi, but not Stbm, is endocytosed. Characterization of the pk-stbm interaction showed that pkenhances the stbm phenotype and that Pk localization requires Stbm.

Three alternatively spliced transcripts are encoded by the pklocus: pk^pk, pk^M andpk^pk-sple. Although these three isoforms differ in the 5′ region, they all contain the single PET and three LIM domains characteristic of the Pk protein (Gubb et al., 1999). PET and LIM domains are thought to mediate protein-protein interactions (Dawid et al.,1998). Isoform-specific mutations in the 5′ region of the transcript result in the pk^pk phenotype, affecting only the wing and notum, whereas mutations in the LIM- or PET-encoding domains result in pk^pk-sple alleles, null alleles that affect the eye, legs and abdomen in addition to the wing and notum(Gubb et al., 1999).

Our observation that Pk distribution is altered in a null stbmbackground suggests that its localization is, at least in part, dependent on Stbm. The possibility that Pk localization is mediated directly by Stbm has not yet been explored, but the PET and LIM domains are candidates for domain-specific interactions with Stbm. Disruption of these domains would result in genetic null alleles, consistent with thepk^pk-sple phenotype described above.

Although ommatidial polarity is not affected in individuals carrying thepk^pk1 allele, this allele enhances the stbm eye phentoype. Functional redundancy could account for the ability of pkto enhance the stbm phenotype such that there is no phenotype whenpk is knocked out but a reduction in pk gene dose can be detected by Stbm. Furthermore, Gubb et al.(Gubb et al., 1999) have indicated that the balance of Pk isoforms contributes to the establishment of tissue polarity. Perhaps this balance is also required for Stbm function.

Cadherins, or Ca²⁺-dependent cell adhesion molecules, have traditionally been recognized for their role in adhesion and the resulting tumorous phenotype. Fmi, Fat (Ft) and Dachsous (Ds), members of a class of cadherins that contain a large number of extracellular cadherin domains(atypical cadherins), have recently been shown to contribute to the polarization of ommatidia (Fig. 3) (Das et al.,2002; Strutt et al.,2002; Rawls et al.,2002; Yang et al.,2002). While the ability of cells to adhere to one another is clearly essential for the establishment of polarity within epithelia, recent work suggests the role of cadherins extends beyond adhesion.

Several lines of evidence suggest atypical cadherins may be involved in signaling. For example, Ft is required in the haltere to inhibit DV signaling and ft mutants display haltere to wing transformations(Shashidhara et al., 1999). In the fly eye, Ft and Ds have been proposed to be required for the transduction of a dorsal-ventral positional signal via cell-cell relay(Rawls et al., 2002). In addition, Yang et al. (Yang et al.,2002) have shown that gradients of Ds and Four-jointed (Fj)activity may regulate Ft to establish this dorsal-ventral cue. It has been suggested that the combined activities of Ds, Fj and Ft, which appear to be functionally conserved in the wing, leg and abdomen(Ziedler et al., 2000),constitute the `elusive' factor `X' in the morphogen model for tissue polarity(Casal et al., 2002).

The data described here are consistent with the notion that Fmi also plays a role in the intracellular signaling required for the establishment of tissue polarity. Given that Fmi is capable of mediating homophilic association between S2 cells (Usui et al.,1999), its role in signal transduction may be indirect and a consequence of a primary role in cell adhesion. However, fmi clones in the eye do not give rise to tumors, nor is the tissue grossly disrupted as has been noted in clones of genes that maintain the integrity of tissue [for example, epithelial phenotypes described for shg mutant embryos(Tepass et al., 1996;Uemura et al., 1996)]. Therefore, it is possible that the primary role of fmi is not to maintain the integrity of tissue via cell adhesion, but rather to maintain sufficient contact between cells to mediate signaling, or even to signal directly.

Ommatidial polarization is thought to rely heavily upon the proper specification of two photoreceptors: R3 and R4. Although these two photoreceptors are recruited into the growing ommatidium as a pair and they morphologically resemble one another in early stages of development, they have long been known to be distinct from one another based on their adoption of distinct sets of contacts early in development(Tomlinson, 1985). Recent work on a number of tissue polarity genes provides genetic and molecular evidence that the complexes of tissue polarity proteins are not identical in photoreceptors R3 and R4. The asymmetric regulation of N by these complexes may ultimately lead to low levels of N activity in R3 and high levels in R4,the combination of which is thought to be essential for the specification of the R3 and R4 cell fates.

Fmi has been shown to interact homophilically, and while current data do not establish that Fmi is present in both R3 and R4 at the junction between R3 and R4, in the model that follows, we assume homophilic interactions between the extracellular cadherin domains of Fmi help to anchor Fmi in R3 and R4 on both sides of the R3/R4 interface (Fig. 8). Furthermore, we suggest the intracellular tail of Fmi is involved in signaling, and that it signals through a complex that is made up of at least three proteins: Fmi, Diego (Diego localization depends on Fmi) and Dsh (Dsh co-localizes with Fmi) (Das et al., 2002). Dsh has also been shown to interact physically with two proteins required for R4 specification, N and Stbm(Axelrod et al., 1996;Strutt et al., 2002; Park and Moon, 2001) and with Pk (Tree et al.,2002). Finally, our stbm-pk genetic and protein localization data suggest Pk and Stbm could physically interact within a complex.

As mentioned above, in order to differentially affect signal transduction through the N pathway, the assembly and/or activity of proteins that set up polarity must be different in R3 and R4. The model presented below requires that Stbm and Pk be restricted to the R4 cell to properly modulate N signaling. [Strutt et al. (Strutt et al.,2002) have shown that Stbm is restricted to R4 at the R3/R4 boundary; the subcellular location of Pk in the eye has not yet been determined.]

We propose that the direct interaction between N and Dsh blocks N signaling, and that the different subset of proteins bound to Dsh is the basis of the asymmetry of the complex. In the future photoreceptor R3, N binds Dsh(which is part of the Fmi/Diego/Dsh scaffold) thereby inhibiting N activity in R3 (Fig. 8). In the future R4 cell, where Stbm and perhaps Pk are localized, Fmi, Diego and Dsh also form a complex. However, in this case, the re-organization of the Fmi/Diego/Dsh complex to include Stbm and Pk bound to Dsh may prevent N from binding to Dsh,leading to high levels of N-mediated signaling in R4(Fig. 8). Ultimately, these differences in gene activity in the R3 and R4 precursors direct the fate specification of these cells.

We express our gratitude to D. Gubb, T. Uemura, H. Kramer, T. Xu and the Bloomington Stock Center for fly stocks; T. Uemura, D. Tree and J. Axelrod for generously providing antibodies; H. Chang and S. Spencer for critical comments on the manuscript; H. Collis for assistance in preparation of the anti-Stbm antibody, and members of the Wolff lab for discussion. T.W. would like to express her gratitude to G. M. Rubin, in whose lab some of this work was initiated. This work was supported by a Lucille P. Markey Pathway fellowship to A.S.R. and NIH grant R01 EY13136 to T.W.