The tumor-suppressor and cell adhesion molecule Fat controls planar polarity via physical interactions with Atrophin, a transcriptional co-repressor

Polarity is a fundamental characteristic of most cells and is essential for their proper function. Cellular polarity has been intensively investigated,and significant progress has been made in revealing the mechanisms that underlie the establishment of cellular polarity. Another widespread, but less understood, form of polarity is planar polarity (PP), which is the organization of cells within the plane of the epithelium. Intensive recent investigations have demonstrated that a conserved PP pathway works to coordinate the polarity of adjacent cells in such divergent systems as convergent extension movements in zebrafish and frogs, and the polarity of wing hairs in the fly (Axelrod and McNeill,2002).

Studies primarily in the fly have implicated a Frizzled (Fz) signaling pathway as being central to the control of PP (Adler and Lee, 2001). Analysis of clones of fz mutant cells reveals both cell autonomous and non-autonomous defects in PP. This has led to the suggestion that Fz, a seven-pass transmembrane receptor (Adler et al., 1990; Park et al.,1994) is involved in both the reception of a PP signal and in its transmission to adjacent cells. Recent work has demonstrated that Fz is part of an asymmetrically localized signaling complex, which also contains the signaling protein Dishevelled and the atypical cadherin, Flamingo(Axelrod, 2001;Das et al., 2002;Strutt et al., 2002;Strutt, 2001;Tree et al., 2002;Usui et al., 1999). How Fz signaling activity is controlled is not clear.

A well-studied system for understanding the control of PP is the fly eye(Wolff and Ready, 1993). The fly eye is composed of ∼800 photoreceptor clusters called ommatidia, each composed of eight photoreceptors, R1-R8, as well as a number of accessory cells (Fig. 1A). Ommatidia are organized into dorsal and ventral fields of mirror-image planar polarity,which meet at the equator (Fig. 1A,C). PP develops in the eye imaginal disc shortly after ommatidial preclusters assemble. Preclusters in the dorsal and ventral fields rotate 90° in opposite directions away from the equator, producing mirror image fields. Genetic studies have suggested that fz is necessary for interpreting and communicating a PP signal that is thought to emanate from the equator. fz has also been implicated in controlling R3 cell fate(Zheng et al., 1995). Intensive work has suggested that photoreceptors R3 and R4 are crucial in directing the rotation of preclusters. In addition it has been shown that the member of the R3/R4 pair with higher Notch activity will take on the R4 fate. This decision directs precluster rotation(Cooper and Bray, 1999;Fanto and Mlodzik, 1999;Tomlinson and Struhl,1999).

The atypical cadherins Fat (Ft) and Dachsous (Ds) act upstream offz, and may control Fz activity during the development of PP. Two separate models have been proposed to explain how Fat and Ds control planar polarity in the eye. In one model, Fat controls Fz activity, and biases the equatorially localized cell to an R3 fate(Yang et al., 2002). In another model, ft inhibits equator formation, and this inhibition is relieved, through an unknown mechanism, at the presumptive equator(Rawls et al., 2002).

We have isolated several new ft alleles in a screen to identify genes involved in cell adhesion and planar polarity in the eye(Fig. 1B). To understand howft functions in PP, we conducted a yeast two-hybrid screen to identify proteins that bind the cytoplasmic domain of Fat. One protein identified by this screen was Drosophila Atrophin (Atro; also known as Grunge), which is the sole Drosophila homolog of human Atrophins(Erkner et al., 2002;Zhang et al., 2002). Mutations in Atrophin 1 cause dentatorubal-pallidoluysian atrophy (DRPLA)(Koide et al., 1994;Nagafuchi et al., 1994), a progressive neurodegenerative disease associated with ataxia, epilepsy,myoclonus and choreoathetosis. Atro is a nuclear protein that has been recently shown to act as a transcriptional co-repressor with roles in segmentation and dorsoventral patterning. Significantly, Atro has also been shown to control PP in the eye and the wing(Zhang et al., 2002).

We show here that Fat and Atro physically interact, display strong genetic interactions, and control similar processes in development. Our analysis offt and Atro clones in the eye reveals alterations in planar polarity at the polar border of the clone as well as equatorial rescue of mutant tissue. In addition, our mosaic analysis of these clones demonstrates that both ft and Atro are needed specifically in the R3 cell. We also show that Atro (like ft) controls expression of four-jointed (fj), a gene that can control PP. These data lead us to propose that ft and Atro act at least twice in the development of PP; initially to control PP gradients and later in R3 fate decisions.

Bait 1 derives from a cDNA of the cytoplasmic region of ft(Research Genetics) cut with NcoI and BsrBI (nucleotides 14760 to 15209) and ligated in frame with the GAL4 DNA binding domain into pAS2-1 (Clontech). Bait 2 was generated from a non-overlapping region of the cytoplasmic domain of ft by PCR amplification (nucleotides 13879 to 14505). Baits were confirmed by sequencing. Bait plasmids were co-transformed into AH109 yeast with a Drosophila embryonic cDNA library in the pACT2 vector (Clontech) using the standard Lithium Acetate transformation. Interactors were selected on SD media lacking Ura, Trp, His, Leu and Ade and tested for activation of the reporter gene using X-gal filter assays(Clontech). pACT2 plasmids were extracted, inserts were PCR amplified and sequenced. Approximately 10⁵ clones were screened for each bait. Interactors were re-tested against pAS2-1 containing bait protein and against the empty pAS2-1 vector to check for autoactivation of the reporter genes.

For the screen, male w; 40FRT flies were mutagenized with ethyl methyl sulphonate using standard procedures. These flies were crossed to females hsFlp,w; 40FRT, their progeny heat-shocked to induce clones and then screened for clones that had smooth edges. Mitotic clones were generated by the Flp/FRT technique with either hsFlp or eyFlp and marked in imaginal discs by Ubi-GFP (Rawls et al.,2002; Yang et al.,2002). Mutant alleles used were ft^GRV,ft^fd, ft¹,ds^UAO71, Atro¹¹,Atro³⁵, Atro^j5A3 and mδ0.5-lacZ(described in FlyBase), UAS-Atro flies were a gift from Tian Xu.ft^alb and ft^so were obtained in our clone shape screen and shown to be ft alleles by complementation analysis. ft^chance was generated by imprecise excision of an adjacent P element (H. M., unpublished).

Third instar imaginal discs were fixed in PLP(Fanto and Mlodzik, 1999). Primary antibodies were used as follows: rabbit anti-Atro(Erkner et al., 2002) (1:100),rat anti-Elav (DHSB, 1:1000), mouse anti-β-Gal (Promega 1:1000), rabbit anti-Bar (gift from Dr Saigo, 1:1000) and mouse anti-Pros (DSHB∼ 1:10). Secondary antibodies were from Jackson Laboratories. After staining, discs were mounted in Vectastain and analyzed with a Zeiss confocal microscope. Adult eye sections were prepared as previously described(Yang et al., 1999). Wings were dehydrated in a 70% ethanol:30% glycerol solution and mounted in DPX (EMS sciences)

For the GST pull-down assay the ft fragment was amplified by PCR and cloned in-frame with the N-terminal FLAG tag provided in the vector pFTX9. FLAG-Fat was then synthesized by in vitro translation using the TNT system(Promega). GST-Atro was produced by PCR amplification of a fragment starting at tyrosine 1639 and ending at the stop codon using primers with XhoI sites. After digestion with XhoI this fragment was cloned into pGex 5x. GST and the GST-Atro fusion were produced in BL21-DE3 bacteria(Invitrogen), extracted in PBS with excess of bacterial-specific protease inhibitors cocktail (Sigma) and purified using glutathione-coupled beads(Pharmacia).

For binding assays, equal amounts of FLAG-Fat were added to pre-equilibrated beads containing 5 μg of GST fusions in HMK buffer(Zhang et al., 2002) in 400μl and were rotated at 4°C overnight. Beads were recovered, washed in HMK and analyzed by SDS-PAGE gels followed by western blotting with ECL chemoluminescence (Amersham). The following primary antibodies were used:mouse anti-FLAG (Sigma 1:2000), rabbit anti-Atro (1:500) and rabbit anti-GST(a gift from F. Miralles, 1:10000). HRP-conjugated secondary antibodies were from Amersham.

We conducted a genetic screen for genes involved in cell adhesion and planar polarity, based on the observation that alterations in cell adhesions within a clone of cells cause them to mix inefficiently with surrounding cells, resulting in a smooth-edged clone(Dahmann and Basler, 2000;Lawrence et al., 1999;Yang et al., 1999). We mutagenized flies and screened for mutations that, when homozygous, produce smooth edged clones (Fig. 1B)and, when sectioned, revealed planar polarity defects. Out of this screen we obtained two new alleles of ft, one new allele of ds, and mutations in two other genes (as yet unidentified). ft encodes a transmembrane protein that is a putative cell adhesion molecule, with multiple cadherin repeats, EGF repeats and laminin G repeats, as well as large cytoplasmic domain with no informative homologies(Fig. 2A)(Mahoney et al., 1991). Fat is an enormous molecule, with a predicted molecular weight of 560 kDa. Loss offt leads to excessive growth of imaginal tissues, and has thus been long considered a tumor suppressor gene in Drosophila(Bryant et al., 1988).ds encodes an adhesion molecule with 27 cadherin repeats and an unrelated cytoplasmic domain (Clark et al.,1995).

Fat and Ds have been shown to have roles in PP in the eye, wing and abdomen(Casal et al., 2002), and it has been proposed that Ds binding to Fat modulates Fat function. Consistent with this proposal, we see that homozygous loss of either ft ords randomizes PP (Fig. 1D,E).

A striking feature of clones of homozygous ft cells is that disruptions in PP are restricted to polar regions of the clone. In addition,entirely wild-type ommatidia on the polar side of the clone frequently show PP inversions (white arrow, Fig. 1F). Loss of ft does not disrupt polarity of wild-type tissue on the equatorial side of the clone, and in fact ft mutant tissue can be phenotypically rescued by adjacent wild-type tissue on the equatorial side of the clone (Rawls et al., 2002; Yang et al.,2002) (Fig. 1F).

Although loss of ds and ft throughout the eye produces similar PP defects, they have distinct clonal phenotypes. In dsclones, ommatida with disrupted polarity are primarily located on the equatorial region of the clone, and genotypically wild-type ommatida at the equatorial border of ds clones are often inverted in the dorsoventral axis. Remarkably, ommatidia at the polar edge of ds clones always adopt the correct polarity (Fig. 1G), suggesting that ds mutants can phenotypically rescued by wild-type tissue on the polar side of the clone.

Mosaic analysis of ft mutant clones has demonstrated a strong bias for the cell that retains ft function to become the R3 cell, whereas there is a bias for retaining ds function in the R4 cell(Rawls et al., 2002;Yang et al., 2002).

To gain insight into how Fat controls PP, we searched for proteins that bind the cytoplasmic domain of Fat. We performed a yeast two-hybrid screen,using two different fragments of the Fat intracellular domain as baits to screen a Drosophila embryonic cDNA library. The bait for the first screen was a 160 amino acid fragment, close to the C terminus, whereas a non-overlapping fragment close to the transmembrane domain (green and black arrows; Fig. 2A) was used for the second screen. Both of these fragments cover regions of high homology to a human homolog of Fat (data not shown). Drosophila Atrophin(Atro) was isolated in the first screen, and specifically interacted with the Fat bait 1 (green arrow) but not with bait 2. Atro encodes a large protein of >200 kDa, with nuclear localization signals, putative DNA-binding and chromatin remodeling domains located at its N-terminus, and a region of particularly high homology to human Atrophins at the C terminus(Fig. 2A)(Erkner et al., 2002;Zhang et al., 2002).Atro is also known as Grunge. Atrophins have been shown to act as transcriptional repressors in Drosophila and in tissue culture studies (Wood et al., 2000;Zhang et al., 2002).

As Atro has been shown to act as a transcriptional repressor in the nucleus, while Fat is located at the cell membrane, such an interaction was unexpected. We confirmed the interaction between Fat and Atro through pull-down experiments, using a GST-Atro fusion protein, and in vitro translated, Flag-tagged Fat protein. A fragment of Fat, corresponding to yeast bait 1, binds specifically to a GST-Atro fusion protein, but not to GST alone(Fig. 2B). These data confirm our yeast 2-hybrid interaction, and provides additional support for a biochemical interaction between Atro and Fat.

To determine if the biochemical interaction between Fat and Atro is important in vivo, we looked for genetic interactions between ft andAtro, and found strong dominant genetic interactions. Heteroallelic combinations of a number of ft alleles are viable. Removal of one copy of Atro in this background severely compromises viability(Table 1). By contrast, loss of one copy of Atro in a wild-type background has no obvious effect on morphology or viability. This dominant heteroallelic lethality provides strong evidence for the biological importance of the interaction between Fat and Atro. To investigate if ft and Atro also show genetic interactions in PP, we have sectioned eyes from the few escapers obtained from crosses maintained at 18°C, and see consistent but subtle PP defects inft^GRV/ft¹; Atro³⁵/+ heteroallelic combinations. The eyes of ft^GRV/ft¹ flies andAtro³⁵/+ flies raised at the same temperature are entirely wild type (data not shown).

When using a different allele of Atro, Atro¹¹, flies emerge in normal mendelian ratios for all genotypes; however, we observed a dramatic shortening of the adult flies' lifespan when one copy of theAtro¹¹ allele is present;ft^GRV/ft¹; Atro¹¹/+ flies have severely decreased viability, with most flies dying between day 4 and 8. As a result of this, a dramatic drop in viability is observed(Fig. 2C). A similarly shaped curve is obtained for ft^chance/ft¹;Atro¹¹/+ flies. Interestinglyft^GRV/ft¹; Atro¹¹/+ flies appear to have motor deficits as their general motility and climbing behaviors are severely reduced.

Atro has been shown to act as a transcriptional co-repressor in vivo inDrosophila (Zhang et al.,2002) and has been reported as a ubiquitously expressed nuclear protein (Zhang et al., 2002;Erkner et al., 2002). To gain some insight into how Atro might interact with a transmembrane protein such as Fat, we investigated its distribution in the eye imaginal disc using an affinity-purified antibody raised against the last 14 amino acids of the protein. This antibody reveals distinct nuclear staining in all photoreceptors behind the morphogenetic furrow, as well as in cells in front of the furrow(Fig. 2D). This staining is lost in Atro mutant clones (white arrow;Fig. 2E), confirming the specificity of the antibody. Reconstructed orthogonal views of confocal sections through imaginal discs that contain Atro^- clones confirms that both the nuclear and non-nuclear staining is lost inAtro clones, indicating that Atro protein is normally present both in the cytoplasm and inside the nucleus (Fig. 2F). Human Atrophins are also present in both the nucleus and in the cytoplasm (Schilling et al.,1999).

Atro has been reported to affect PP in the eye(Zhang et al., 2002), but the nature of these defects have not been investigated in depth. To understand how Fat and Atro function in PP, we analyzed the planar polarity defects inAtro mutant clones in greater detail.

In Atro^- clones, as in ft^- clones,we find non-autonomous alterations in polarity only on the polar border of the clone. Wild-type ommatidia at the polar border of dorsal Atro clones adopt a ventral appearance (Fig. 3A, white arrow), while wild-type and mosaic ommatidia at the equatorial border of the clone have normal polarity. Inside the clone there are significant defects in photoreceptor differentiation, suggesting thatAtro has other roles in the eye in addition to the control of planar polarity. Analysis of homozygous clones of weaker Atro alleles,induced by insertion of a P element, shows very similar phenotypes (not shown).

ft has been reported to be specifically required in the R3 photoreceptor (Yang et al.,2002). We assayed the requirements for Atro in different photoreceptors using clones of homozygous Atro cells generated in a heterozygous background. Because there are no strict lineage relationships among ommatidial cells, the ommatidia along the clonal border are composed of random combinations of Atro⁺ and Atro^-cells. Among these combinations were mosaic ommatidia in which only one member of the R3/R4 precursor pair possesses functional Atro. Atro⁺ ommatidia can be identified by the presence of black pigment granules in the photoreceptor rhabdomeres(Fig. 3B,C). In virtually all cases in which an ommatidium that was located on the polar border of the clone and had incorrect polarity was mosaic for Atro in the R3/R4 cell pair, the R3 photoreceptor retained Atro (black arrowsFig. 3B). This mosaic bias suggests that Atro is preferentially required in the R3 cell.

A careful quantitative analysis of these mosaic ommatidia(Table 2) shows a general bias for retaining the Atro function in all the anterior photoreceptors as previously reported for ft by Rawls et al.(Rawls et al., 2002). However,both in ommatidia that retained the correct polarity and inverted ommatidia,the bias for R3 is very strong and is significantly higher than that observed for the other anterior photoreceptors, R1 and R2.

In addition, in viable heteroallelic combinations of Atro(Fig. 3E, black arrows) we frequently see symmetric ommatidia, mostly of the R4/R4 achiral subtype. Conversely, when a wild-type form of Atrophin is transiently overexpressed in the eye imaginal disc under the control of the sevenless enhancer— alongside alteration in the degree of rotation and the polarity of several ommatidia — many elongated symmetric ommatidia of the R3/R3 achiral subtype are observed (Fig. 3D). These phenotypes have been described for mutants infz and Notch and strongly support the notion that Atro biases the R3/R4 cell fate choice towards R3.

Despite the similarity of the PP deficits, Atro clones do display some differences when compared with ft clones. As mentioned above,Atro is required for the survival or correct specification of photoreceptors. Ommatidia within the clones often appear to have lost one or more outer photoreceptors, and in many cases they have more than one inner photoreceptor with smaller rhabdomeres(Fig. 3A,E, yellow arrows).

The autonomous and non-autonomous defects in ft clones are first detectable in the eye imaginal disc, using antibodies against Bar, a marker for R1 and R6 (Higashijima et al.,1992) (Fig. 4A). Ommatidia with a dorsal orientation are indicated with red arrows; ventral orientation is indicated with blue arrows. Although PP is disrupted throughout much of the ft clones, there is often a striking rescue of ommatidial orientation on the equatorial border of the ft clone. This has led to the proposal that a ft-dependant signal is carried for one or two ommatidial rows from the Ft⁺ cells outside the clone(Rawls et al., 2002). This phenotype is reversed in ds clones, where there is a clear rescue of the mutant ommatidia at the polar border, whereas the polarity defects spread into the wild-type tissue for up to two rows at the equatorial border of the clone (Fig. 4B)

We investigated Atro function in the eye imaginal disc, using the polarity marker, E(spl)mδ0.5 (Cooper and Bray, 1999). In wild-type ommatidia E(spl)mδ0.5is expressed in the polar cell, which becomes specified as R4. Atromutant clones (marked by loss of GFP) show disruptions of the regular pattern of planar polarity, as revealed by the position of the R4 marker within an ommatidium. Within Atro^- ommatidial clusters the cell expressing the R4 marker frequently occupied the position normally taken by the equatorial cell (black arrows), indicating that these ommatidia had switched their dorsoventral PP (Fig. 4C). Strikingly, there are non-autonomous disruptions in PP in the wild-type tissue on the polar border of the clone. The alteration of ommatidial polarity in Atro^- ommatidia and in surrounding tissue is diagramed in the adjacent panel(Fig. 4C′). It is clear that Atro clones (like ft clones) show an equatorial rescue of normal PP within the clone, and non-autonomous disruptions of planar polarity at the polar border of the clone.

We also saw, at lower penetrance, alterations in the ability of Atro photoreceptors to commit to the R3 cell versus R4 cell fate in the imaginal disc. Fifteen percent of ommatidia in Atro³⁵ clones show additional photoreceptors with an R4 cell marker, suggesting that these ommatidia have adopted an R4/R4 fate. Only 2% of Atro ommatidia have lost the R4 cell marker, which may suggest they have adopted an R3/R3 fate. In flies with the weaker, viable combination of Atro alleles(Atro³⁵/Atro¹¹), the same tendency is seen, though less strongly: Out of 610 ommatidia analyzed for expression of the R4 marker, 49 were of the R4/R4 class and five were of the R3/R3 type(data not shown). This tendency of Atro ommatidia to adopt an R4/R4 configuration is seen throughout clones, without any evidence for equatorial rescue. Together, these data support a cell-autonomous function for Atro in R3 fate determination.

ft has been shown to repress expression of fj, which encodes a type II transmembrane/secreted protein, expressed in a graded fashion from the equator towards the pole of the eye imaginal disc(Fig. 4D)(Brodsky and Steller, 1996;Buckles et al., 2001;Villano and Katz, 1995;Yang et al., 2002;Zeidler et al., 1999;Zeidler et al., 2000).fj has been implicated in the regulation of ommatidial polarity,based on the observation that reversals of ommatidial polarity occur along the polar border of fj mutant clones. In addition ectopic expression of Fj in clones can also produce reversals of ommatidial polarity. Together these data suggest that ommatidia can recognize and polarize to a gradient of Fj protein.

Atro has been shown to act as a transcriptional repressor(Zhang et al., 2002). To see if the ability of ft to repress fj could be mediated byAtro, we examined Atro^- clones usingfj-lacZ to monitor changes in fj transcription. We found that loss of Atro (Fig. 4E), like loss of ft(Yang et al., 2002) results in an increase in fj transcription, demonstrating that Atro is needed for the proper regulation of the ft target fj.

Atro adult clones have clear disruptions in photoreceptor specification and/or survival (Fig. 3A). These defects include loss of outer photoreceptors and extra inner photoreceptors. To determine if Atro is functioning in cell fate decisions important for photoreceptor specification, we stainedAtro mutant clones with antibodies that recognized specific subsets of photoreceptors. Loss of outer photoreceptors is a common feature ofAtro mutant clones in the adult, therefore we examined expression of Bar, a marker for the R1 and R6 outer photoreceptors. InAtro^- clones Bar staining is variable, and at times only one cell expresses Bar (Fig. 5A, yellow arrowheads). Ommatidia in adult Atro clones often also have extra inner photoreceptors. To investigate if these extra inner photoreceptors represented extra R7-like cells, we examinedAtro clones for Prospero, which is expressed in the R7 cell and in cone cells (Kauffmann et al.,1996). In Atro clones, additional photoreceptor cells express Prospero (Fig. 5B,white arrows), compared with surrounding heterozygous tissue, suggesting that in the absence of Atro, outer photoreceptors were transformed into R7-like photoreceptors [defects that we never find in ft clones (data not shown)]. Therefore Atro is needed for fate specification for both inner and outer photoreceptor classes.

Another difference between ft and Atro function lies in the control of growth. Loss of ft leads to a loss of growth control,which is reflected in the greater size of mutant tissue (marked by the loss of GFP) compared with the twin spot, which has two copies of the GFP marker(yellow arrows; Fig. 5D) and is brighter than the surrounding heterozygous tissue. By contrast,Atro^- clones (Fig. 5C) are similar in size to their twin spot, suggesting that they lack the tumor suppressor phenotype of ft clones.

Finally, one striking feature of ft clones is their round and smooth appearance, which is visible in clones in the adult, and in the eye imaginal disc (Fig. 1B,Fig. 5D). AlthoughAtro clones do not appear smooth in the adult, Atro clones in the eye disc appear strikingly smooth when they are located anterior to the morphogenetic furrow (Fig. 5C,white arrow). Comparison with the twin spot (yellow arrow), confirms that this effect is specific to loss of Atro. This smoothness is lost after progression of the furrow, when ommatidial differentiation commences (marked by Elav staining in red), suggesting that Atro acts specifically in a pre-furrow adhesion process that is lost upon neuronal differentiation.

There are a number of similarities in mutant phenotypes that suggest thatAtro and ft function together at several stages in development. Atro is expressed ubiquitously throughout embryonic and larval development, and is also maternally contributed(Erkner et al., 2002;Zhang et al., 2002).ft is also maternally contributed, and is expressed in most cells in the embryo and the larva. Clones of ft^GRV in the wing produce blisters where the dorsal and ventral surfaces of the wing fail to fuse (Fig. 5F) (Garoia et al.,2000). Small clones of Atro have been reported to result in swirls of wing hairs and in notching when they touch the dorsoventral border(Zhang et al., 2002). We observed these defects, and in addition we noted that when we inducedAtro³⁵ clones earlier in development, these clones also formed blisters that are similar to those found in ft clones(Fig. 5G). Thoracic clones ofAtro disrupt thorax closure(Zhang et al., 2002)(Fig. 5J). Loss of ftin clones also disrupts thorax closure(Fig. 5I). Taken together,these data suggest that Fat and Atro may function together many times during development.

We describe a novel interaction between a cell adhesion molecule, Fat and a nuclear transcriptional co-repressor, Atro. We demonstrate that Fat and Atro act together in a variety of tissues, display strong genetic interactions, and are essential for cell adhesion, and cell fate choices during the development of planar polarity in the fly eye. We show that ft and Atrofunction in control of a planar polarity signal, and in the R3 fate determination. There is precedence in the eye for the reiterative role of genes acting at different stages in the development of planar polarity. Like Atro, Notch acts early at the midline to upregulate the expression offj. Later, in a separate mechanism that is coincident with ommatidial rotation, Notch defines the R4 cell fate and promotes correct rotation of the precluster.

Mosaic analysis of ft mutant clones demonstrated a strong bias for the cell that retains ft function to become the R3 cell(Rawls et al., 2002;Yang et al., 2002). This has been interpreted to indicate that Fat directly biases the cell to become an R3 cell (Yang et al., 2002). We find that Atro clones also show a bias for the R3 cell to retain Atro function, supporting the model that Atro, like Fat, works in R3 fate determination.

However, an extensive mosaic analysis found that all anterior cells (R1, R2 and R3) tend to be ft⁺, and all posterior cells (R4, R5 and R6) tend to be ft^-(Rawls et al., 2002). Rawls et al. suggested that this bias is due to spatial considerations, cells that are polar in the precluster, undergo a 90° rotation, leaving them in a posterior position in the adult. At the polar border of ft clones,ommatidia rotate in the opposite direction to wild type, therefore the bias is reversed, leading to an increase in ft^- anterior cells. Therefore, they concluded that additional data was required to show thatft function is specifically needed in R3. To determine if ftand Atro are specifically required in the R3 photoreceptor, we have undertaken mosaic analysis of Atro, ft and wild-type clones. We find that in wild-type clones, marked only by white, ommatidia at the polar border of the clone show a weak preference for posterior photoreceptors to be wild-type (60.75±4.7%; P over A). This bias is strictly due to the spatial constraints of recruitment in clones. In ft clones, we find that at the equatorial border (where polarity is unaltered) there is no discernable difference between anterior photoreceptors subclass types; 85% of all photoreceptors that retain ft are anterior class. The increase from ∼61% to 85% is probably due to the adhesive properties offt, which result in smooth edged clones.

By contrast, at the polar border, which is where planar polarity is altered, there is a marked tendency for ft function to be retained specifically in the R3 photoreceptor; 100% of R3 cells retained ft,whereas only 83% of all anterior photoreceptors retained ft. Our mosaic analysis of Atro clones showed that the bias introduced by PP alterations at the polar border of clones (similar to ft) introduces a general bias for anterior photoreceptors. However, again, the bias is stronger in the R3 cell than for other anterior photoreceptors. This increased bias in the R3 photoreceptor over the other anterior class member suggests that ft and Atro are important in R3 fate.

The conclusion that Atro function is important for the R3 cell fate is also strongly supported by our observation that loss of Atro often results in symmetric ommatidia with two R4 cells. This is reflected by the increase in R4/R4 ommatidia seen in Atro clones in the eye disc marked by expression of the R4 marker, md-lacZ. In addition the overexpression of Atro in R3 and R4 generates symmetric ommatidia with two R3 cells. Together, these data support the proposal that Atro is needed for the R3 fate.

The non-autonomous nature of the PP defects associated with ft andAtro mutant clones could have presented some problems to mosaic analysis. We might have expected that non-autonomous alterations in polarity would equally affect all photoreceptors, yet our data clearly show enhanced requirements for ft and Atro function in the R3 photoreceptor over other photoreceptors. In addition, the proposal that Atro is needed for the R3 cell fate is supported by our analysis of the R4 marker in eye discs. Interestingly, the tendency to lose the R3 cell fate inAtro clones is seen throughout the clone, and does not appear to participate in the phenomena of equatorial rescue or polar nonautonomy.

Because Fz is also needed for R3 fate decisions, it has been suggested that Fat positively affects Fz signaling (Yang et al., 2002). Our observation that Atro acts with Fat and also biases towards the R3 fate suggests that the regulation of Fz by Fat may not be direct. We propose instead that Atro is necessary for theft-dependent bias to an R3 cell fate and for the production of a diffusible PP molecule that controls Fz activity.

The proposal that Fat increases Fz activity, and thereby biases a cell towards the R3 fate (Yang et al.,2002), does not explain the non-autonomous disruptions of wild-type tissue on the polar side of ft and Atro clones, or the rescue of ft and Atro mutant tissue from wild-type tissue on the equatorial side of the clone. There are several models that could explain the non-autonomous disruptions of planar polarity. One model suggests that planar polarity is established through a `domino effect'. This model is suggested by the striking accumulation of planar polarity components,such as Fz and Dsh on the distal edge of every cell in the wing (reviewed byMcNeill, 2002). This observation, coupled with genetic data that suggests that high Fz activity on one side of the cell forces low Fz activity on the other side, leads to a model in which accumulation or loss of polarity in a cell leads to templating of that state onto the next cell, non-autonomously propogating PP defects. However in the eye, Fz and Dsh only show differential distribution on a subset of ommatidial precursor cells, and, importantly, intervening cells show no altered accumulation. These data argue against a simple templating model for PP in the eye.

An alternative model proposed by Rawls et al.(Rawls et al., 2002) suggests that the juxtapositioning of ft⁺ andft^- tissue contributes to midline determination and emphasizes the role of Fat in inhibiting DV signaling away from the equator. This inhibition would be relieved at the equator by an unidentified molecule that would inhibit Fat function. If this model was correct, a smallft clone should mimic the situation at the equator, where Fat function is predicted to be locally inhibited. We would therefore expect to find an ectopic equator in the middle of the clone. Instead, however, we observe the opposite phenotype, as the ommatidia on the two sides of the clone point towards the middle of the clone, rather than away from it.

We believe that the model that best explains both the equatorial rescue and polar nonautonomy of ft and Atro clones is that Fat and Atro together control expression of a planar polarity morphogen, here called`factor X'. We imagine that factor X is in a gradient with high levels at the equator and low levels at the poles, thus all ommatidia will appear to `point'down this gradient (see Fig. 6). If Fat and Atro are essential to the production of factor X,then the ft/Atro mutant tissue will be void of factor X, producing a sink in the gradient. The gradient will still be pointing in the same direction initially, thus the wild-type polarity of ommatidia at the equatorial side of the clone and `equatorial rescue' seen in ft andAtro clones. For ommatidia at the polar edge of the clone, the gradient will be reversed, and ommatidia will point in the opposite direction. The gradient will also be disrupted outside of the clone, leading to inversions of the polarity of wild-type tissue on the polar side of the clone and `polar nonautonomy' seen in these mutant clones. In large clones, there will be a region in the center of the clone where there is no detectable factor X, and as a result polarity will be randomized. All of these predictions are met in ft and Atro clones. Loss of Ds, which inhibits Fat function, should increase factor X. As predicted by this model,ds clones show disruptions in wild-type tissue on the equatorial side of the clone, and rescue of mutant tissue on the polar side of the clone. Without ft or ds function there would be no gradient and,consistent with this prediction, we see complete loss of planar polarity in eyes that are homozygous for strong alleles of ft or ds.

A gradient of Wg protein (which is high at the poles and low at the equator) initially establishes a gradient of Ds protein over the eye field(Yang et al., 2002). This gradient of Ds protein in turn produces a gradient of Fat activity, which, we believe, creates a gradient of Atro activity. We propose that each cell will produce factor X at a level that is proportionate to the level of Atro activity, which varies according to the position of that cell in theds and ft activity gradients. Our model assumes that Factor X is a short-range diffusible molecule, which provides polarity information to ommatidial preclusters to direct their rotation. As Fat has been shown to be upstream of Fz, we speculate that the Atro-dependant Factor X is a ligand for Fz.

Both ft and Atro also act in other, apparently unrelated,pathways. One of the prominent features of ft mutant larvae is the loss of growth control, which leads to dramatically overgrown discs(Bryant et al., 1988) and mutant clones that are markedly larger than their sister twin spots (seeFig. 5D). However,Atro^- clones do not display overgrowth in the eye,suggesting that ft restricts growth via an Atro-independent pathway. In addition, in the adult eye Atro clones (unlike ft clones)show severe defects in photoreceptor number and type, suggesting Atro has additional roles in photoreceptor specification and/or survival that are not shared by Fat. One particularly surprising result was our finding thatAtro^- clones are markedly smooth before the furrow, and that this smoothness is lost after the furrow passes. This suggests that Atro may function in a cell adhesion process that is lost upon cell differentiation.

Dentatorubral-pallidoluysian atrophy (DRPLA) is a dominantly inherited neuronal degenerative disease characterized by the variable combination of ataxia, choreoathetosis, myoclonus, epilepsy and dementia. This disease is caused by the expansion of a polyglutamine tract within the Atrophin 1 protein(Koide et al., 1994;Nagafuchi et al., 1994). Atro is the sole fly homolog of human atrophins. Atro has been shown to act as a transcriptional co-repressor in vivo in Drosophila(Zhang et al., 2002). Atro interacts genetically with even skipped, a transcriptional repressor,and is required for the in vivo repressive activity of even skipped. The transcriptional repressor activity of Atro has been localized to the highly conserved C-terminal region of Atro. This C-terminal region can bind to Evenskipped in vitro and interacts with the minimal repression domain of Evenskipped (Zhang et al.,2002).

We have shown here that the intracellular domain of Fat binds the C-terminal domain of Atro. The cytoplasmic expression of Atro and its interaction with Fat raises the possibility that instead of acting as a simple co-repressor, Atro functions in a more complex manner. Other transcriptional co-repressors are known to be converted to transcriptional activators upon cell signaling, and future work will determine if the interaction of Fat with Ds alters the transcriptional activity of Atro.

Owing to the fact that Atro binds the cytoplasmic domain of Fat, we favor a model in which Atro acts downstream of Fat, possibly relaying a Fat-dependant signal to the nucleus. However, the similarity of the ft andatro loss-of-function phenotypes makes classical epistasis experiments difficult, therefore we cannot exclude a model in whichAtro acts upstream of ft. Examination of the amount or subcellular distributions of Fat and Atro, suggest that Atro does not control Fat expression or localization, nor does ft control the levels or subcellular localization of Atro (data not shown).

In conclusion, we have demonstrated a novel interaction between a cell adhesion molecule, Fat, and a nuclear corepressor, Atro, which work together in the control of planar polarity in the fly eye. We show that Atrois essential for the regulation of a previously identified ft target,fj, and acts in R3 cell fate determination. Based on these data, we propose that Fat and Atro act together to control the production of a diffusible planar polarity signal and to bias photoreceptors towards the R3 cell fate.

We dedicate this paper to the memory of Kirsten Hardiman, who died during the course of this work. We also thank Sarah Bray, Peter Bryant, Mike Simon and Tian Xu for fly stocks and antibodies. This manuscript was improved by comments from Jeff Axelrod, Julian Lewis, David Ish Horowicz, Sally Leevers,Caroline Hill and members of the McNeill laboratory. This work was supported by Cancer Research UK (Cancer Research UK was formed by a merger between the ICRF and the CRC in February 2002). M. F. was supported by a Marie Curie EU Fellowship.