During Drosophila leg development, the distal-most compartment (pretarsus) and its immediate neighbour (tarsal segment 5) are specified by a pretarsus-specific homeobox gene, aristaless, and tarsal-segment-specific Bar homeobox genes, respectively; the pretarsus/tarsalsegment boundary is formed by antagonistic interactions between Bar and pretarsus-specific genes that include aristaless (Kojima, T., Sato, M. and Saigo, K. (2000) Development 127, 769-778). Here, we show that Drosophila Lim1, a homologue of vertebrate Lim1 encoding a LIM-homeodomain protein, is involved in pretarsus specification and boundary formation through its activation of aristaless. Ectopic expression of Lim1 caused aristaless misexpression, while aristaless expression was significantly reduced in Lim1-null mutant clones. Pretarsus Lim1 expression was negatively regulated by Bar and abolished in leg discs lacking aristaless activity, which was associated with strong Bar misexpression in the presumptive pretarsus. No Lim1 misexpression occurred upon aristaless misexpression. The concerted function of Lim1 and aristaless was required to maintain Fasciclin 2 expression in border cells and form a smooth pretarsus/tarsal-segment boundary. Lim1 was also required for femur, coxa and antennal development.
Vertebrate limbs and invertebrate appendages are formed through subdivision of the corresponding developing field and each subdomain, possibly with its own particular properties such as specificity in local cell adhesivity, may be specified by a combinatorial region-specific expression of transcription factors. Thus, it is important to clarify the manner in which transcription factor expression domains are generated and the mechanisms by which they determine local fates of developing limbs or appendages.
Drosophila adult leg consists of several segmental units, which, in proximal-distal direction, are the coxa, trochanter, femur, tibia, tarsal segments 1-5 and pretarsus – the latter bearing claws, pulvilli and an empodium. These segments develop through concentric subdivision of the leg disc epithelium, a mono-layered cell sheet that invaginates from the epidermis during embryogenesis. Distal segments are derivatives of the central region of the leg disc, while proximal segments, derivatives of the peripheral region. The concentric subdivision of the disc epithelium occurs in multiple phases. At the earliest stages of leg disc development, disc epithelium is divided into a distal region expressing Distal-less (Dll) and a proximal region expressing homothorax (hth), escargot (esg) and teashirt (tsh) (AbuShaar and Mann, 1998; Erkner et al., 1999; Goto and Hayashi, 1999; Wu and Cohen, 1999). Dll and hth are homeobox genes (Cohen et al., 1989; Rieckhof et al., 1997), while esg and tsh encode zinc-finger proteins (Fasano et al., 1991; Whiteley et al., 1992; Fuse et al., 1994). As development proceeds, these domains undergo further subdivision into smaller domains through the action of dachshund (dac), encoding a novel nuclear factor (Mardon et al., 1994; Abu-Shaar and Mann, 1998; Wu and Cohen, 1999), and finally, into the regions corresponding to adult leg segments or components (Fig. 1A). Regulatory interactions between transcription factor genes expressed in neighbouring domains have been implicated to be essential for precise subdomain determination (Abu-Shaar and Mann, 1998; Erkner et al., 1999; Wu and Cohen, 1999; Kojima et al., 2000).
BarH1 and BarH2, a pair of homeobox genes at the Bar locus (Kojima et al., 1991; Higashijima et al., 1992a), are essential for distal leg segmentation and specification of tarsal segments 3-5 in a functionally redundant manner (Kojima et al., 2000); they are hereafter collectively referred to as Bar. aristaless (al) is a homeobox gene expressed in the distalmost segment, pretarsus, and required for normal pretarsus development (Campbell et al., 1993; Schneitz et al., 1993; Campbell and Tomlinson, 1998). al and Bar expression domains are initially determined as broad domains that partially overlap each other but at slightly later stages, a line of demarcation between al and Bar expression domains becomes evident through auto-regulation of Bar and mutually exclusive interactions between Bar and pretarsus factors (Kojima et al., 2000). Although al is included in pretarsus factors, other pretarsus genes may also be necessary for effective repression of Bar. Indeed, little or no reduction in Bar expression could be detected in future tarsal segments 4 and 5 following al misexpression, while Bar misexpression brought about considerable reduction in al expression in the prospective pretarsus region (Kojima et al., 2000).
Here, we show that Lim1, which encodes a protein similar in sequence to a vertebrate LIM-homeodomain protein, LIM1 (Taira et al., 1992; Fujii et al., 1994; Barnes et al., 1994), serves as a pretarsus element. Our results also indicated that, in the future pretarsus, al expression is positively regulated by LIM1, while Lim1 expression is under the negative control of Bar. In addition, Lim1 was found to be required not only for pretarsus development but the formation of femur, coxa and antennal structures as well.
MATERIALS AND METHODS
Flies used in this study were raised on standard medium at 25°C. Fly strains used are Canton-S (wild-type), ptc-GAL4 (559.1; Hinz et al., 1994), blk-GAL4 (40C.6; Morimura, et al., 1996), UAS-BarH1M6 (Sato et al., 1999b), and UAS-al6 (Kojima et al., 2000), al1, alice and alex (Campbell and Tomlinson, 1998). alex and alice are a null and a strong hypomorphic mutant, respectively. alice/alex flies exhibit the same leg and antennal phenotypes as alex mosaic clones (Campbell and Tomlinson, 1998) and no detectable level of AL was observed by immunostaining in alice/alex leg and antennal discs (data not shown). Thus, the genotypes of alice/alex are referred to as al− in this paper. P0092 (Lim1-lacZ) were obtained from FlyView (http://pbio07.uni-muenster.de/). UAS-Lim1 was generated by inserting an EcoRI-XhoI fragment of GH04929 (Berkeley Drosophila Genome project (BDGP; http://fruitfly.berkeley.edu/)) into pUAST (Brand and Perrimon, 1993). For FRT/FLP mosaic analyses, FRT19A (Xu and Rubin, 1993) and eyFLP5 (Newsome et al., 2000) were used.
FRT/FLP mosaic analysis
Lim1– clones were generated in larvae whose genotype are Lim17B2FRT19A/ y w arm-lacZ FRT19A; eyFLP5/+. eyFLP5 seems to express FLPase also in the leg and antennal discs from early stages of development and can induce mosaic clones before the onset of third instar.
Ectopic expression of Bar, al and Lim1
Both ptc-GAL4 and blk-GAL4 can drive gene expression along the anterior-posterior (A/P) compartment boundary (Hinz et al., 1994; Morimura, et al., 1996). In most experiments, UAS-BarH1M6 and UAS-al6 were driven by ptc-GAL4. Any appreciable mutant phenotype was given by UAS-al driven by neither ptc-GAL4 nor blk-GAL4, while blk-GAL4-driven UAS-Bar gave phenotypes somewhat less severe than those given by ptc-GAL4-driven UAS-Bar, suggesting that ptc-GAL4 is a slightly stronger driver than blk-GAL4. Among 13 independent UAS-Lim1 lines so far generated, nine lines, showing essentially the same phenotype when driven by blk-GAL4, were chosen and used for further experiments. UAS-Lim1 was driven only by blk-GAL4, since flies with ptc-GAL4 and UAS-Lim1 were mainly larval lethal under our experimental conditions.
Immunohistochemistry and in situ hybridisation
X-Gal staining and antibody staining were carried out according to Sato et al. (1999b). Primary antibodies used were rat anti-AL (Campbell et al., 1993), mouse anti-DLL (Diaz-Benjumea et al., 1994), rabbit anti-BarH1 (Higashijima et al., 1992b), mouse anti-FAS2 (Lin et al., 1994), rabbit anti-lacZ (anti-β;-galactosidase; Cappell), and mouse anti-lacZ (Promega). As secondary antibodies, Cy3, Cy5 (Amersham Pharmacia Biotech), or biotin (vector) conjugated antibodies followed by avidin-FITC (Promega) were used. Images were obtained using MRC-1000 confocal microscopy (Bio Rad) and processed using Photoshop 5.0 (Adobe). In situ hybridisation was carried out as described previously (Sato et al., 1999a). RNA probe was prepared using the GH04929 insert as a template.
Xenopus embryo injection assay
Plasmids for mRNA injections were constructed by inserting PCR amplified 1.5 kb Lim1-coding sequences between BamHI and XbaI sites of pCS2+ (Turner and Weintraub, 1994). Xlim1 constructs, mRNA synthesis, Xenopus embryo injections and antibody staining for 12/101 antibody (Kintner and Brockes, 1984) have been described previously (Taira et al., 1994).
Identification of Lim1 as a gene expressed in the distal tips of developing legs and antenna
To isolate genes that possibly act with al in pretarsus specification, a search was made for genes expressed in the pretarsus but not the segment immediately adjacent to it at late third instar stages. P0092 is an enhancer trap line, in which lacZ expression in leg and antennal discs was found to be similar to al expression in these tissues. As shown in Fig. 1, lacZ was coexpressed in virtually all AL-positive cells in the pretarsus, tibia, femur and possibly coxa in leg discs (Fig. 1B-D”), and the arista and first antennal segment in antennal discs (Fig. 1E-E”). Although AL expression was restricted to ventral cells in the tibia and dorsal cells in the femur, coxa and first antennal segment, lacZ expression was noted in both ventral and dorsal cells uniformly, which gave rise to complete circular expression. In wing and haltere discs, in which al is also expressed (Campbell et al., 1993), no appreciable expression of P0092-lacZ was observed (data not shown).
In P0092, P-element insertion occurs at 8B on the X chromosome (FlyView). Genomic DNA clones spanning about 70kb region surrounding the P insertion site were isolated using a plasmid rescue fragment as an initial probe (Fig. 2A). Two relevant EST (Expressed Sequence Tag) clones (GH04929 and LD27231) were identified in the Berkeley Drosophila Genome project (BDGP) database using genomic DNA sequence information. As shown in Fig. 2B,D, GH04929 gave in situ hybridisation patterns almost identical to P0092-lacZ expression. Nucleotide sequence analysis indicated that the putative P0092 gene encodes a LIM-homeodomain protein identical in amino acid sequence to LIM1, a Drosophila homologue of vertebrate LIM1 (Accession number, AB034690; Lilly et al., 1999).
Xlim1 is known to initiate the formation of a secondary axis when its mRNA is coinjected with XLdb1 mRNA, which encodes a LIM domain-binding protein homologous in amino acid sequence to Drosophila Chip (Agulnick et al., 1996; Morcillo et al., 1997). Thus, a study was undertaken to determine whether Lim1 possesses activity similar to Xlim1 by injecting Lim1 mRNA into fertilised Xenopus eggs with XLdb1 mRNA. Figure 2F-I show that, as with Xlim1, Lim1 is capable of effectively inducing a secondary axis in Xenopus in a XLdb1-dependent manner. It may thus follow that the P0092 gene product or Drosophila LIM1 is similar not only in amino acid sequence but also in association with LIM domain-binding protein and target sequence recognition to vertebrate LIM1.
Lim1 is essential for leg and antennal distal structure formation
Flies neither homozygous nor hemizygous for the P0092 P insertion showed any obvious morphological defects. Thus, Lim1 loss-of-function mutants were generated by imprecise P-element excision and six independent larval or pupal lethal mutant lines were obtained. These frequently produced pharate adults with apparent defects in mouth parts, leg and antennal morphology (for detailed mutant phenotypes, see below), making it possible to examine the roles of Lim1 in leg and antennal development.
Lim17B2 was the severest in our Lim1 mutants. In this mutant, the predicted RNA start site, the first exon and a portion of the first intron were found to be lost (Fig. 2A). No appreciable Lim1 RNA signals could be detected in Lim17B2 leg and antennal discs (Fig. 2C,E) and embryos (data not shown), indicating that it is a transcriptional null mutant allele. In the following, Lim17B2 is referred to as Lim1− and used as the Lim1 mutant.
In legs and antenna completely lacking al activity, all pretarsus structures and arista are lost, respectively (Fig. 3B; Campbell and Tomlinson, 1998). In moderate hypomorphic al mutants such as al130, al1/Df(2L)al and al2/Df(2L)al flies, claws are frequently lost without loss of other pretarsus structures such as pulvilli and empodia (Fig. 3C; Schneitz et al., 1993; Campbell and Tomlinson, 1998; Kojima et al., 2000), while in weak hypomorphic mutants (e.g. homozygotes for al1), claws and aristae were not lost but only reduced in size (Fig. 3D). Figure 3E shows that, in Lim1− legs, pulvilli and empodia were normally present but claws are frequently lost. It may thus follow that Lim1− mutants are very similar in leg phenotype to moderate al hypomorphic mutants. In about half of all cases (n=24), the antenna was absent from the Lim1− half head (Fig. 3H). When antennae was present, arista was deformed and reduced in size (Fig. 3I). That is, dim1− arista are morphologically similar to those of weak hypomorphic al mutants. These findings indicate that Lim1 is essential for proper development of pretarsus and arista as well as al, although Lim1− mutant phenotypes are much less severe than a1− mutant phenotypes. In Lim1− legs that were simultaneously homozygous for al1, not only claws but also empodia and pulvilli were frequently lost (Fig. 3F). The concerted function of Lim1 and al would thus appear to be required for normal pretarsus/aristal development.
Absence of Lim1 expression in early pretarsus and arista precursor cells expressing al
Lim1 may affect the morphogenesis of distal parts of the leg and antenna by modulating al and/or Bar expression or their mutually antagonistic interactions (Kojima et al., 2000). Lim1 expression may be affected by al and/or Bar activity. As a first step to clarify these points, examination was made of Lim1-lacZ, al and Bar expression in the centre of developing wild-type leg and antennal discs. Gene expression was examined using the corresponding antibodies for the gene products.
As indicated previously, AL and BAR expression begins in the central region of early-third instar leg discs in a partially overlapping manner (Fig. 4A; Kojima et al., 2000). As shown in Fig. 4A”, no Lim1-lacZ expression was noted to occur at the earliest stages of AL and BAR expression in early third instar. Just prior to initiation of central fold formation along the outer circumference of the BAR ring (Kojima et al., 2000), Lim1-lacZ expression first became detectable in the centre of the AL domain (Fig. 4B-B”). But, unlike the AL domain, the Lim1-lacZ expression domain did not overlap the surrounding BAR expression domain (Fig. 4B′). By mid third instar, the central region of the leg disc has divided almost completely into two non-overlapping regions; the central domain expressing AL and Lim1-lacZ but not BAR, and the surrounding domain expressing only BAR (Fig. 4C-C′′′). A similar relationship between AL, BAR and Lim1-lacZ expression was observed in antennal discs (Fig. 4D-F′).
Requirements of al for Lim1 expression and Bar repression in the pretarsus
That Lim1-lacZ expression is initiated in the AL domain would suggest that al is required for Lim1 expression. Thus, we examined whether Lim1-lacZ expression would be affected by misexpressing al along the A/P border using ptc-GAL4 and UAS-al or by abolishing al activity from the presumptive pretarsus. Although no Lim1-lacZ misexpression was induced by ectopic al expression along the A/P border (Fig. 5A-B), pretarsus Lim1-lacZ expression was virtually completely eliminated in al− leg discs (Fig. 5C′), indicating that al is directly or indirectly required for pretarsus Lim1 expression.
Since partial Bar misexpression was previously observed in al hypomorphic mutants (Kojima et al., 2000), Bar misexpression might be induced throughout the presumptive pretarsus in al− leg discs. We tested this hypothesis and found that this is the case. As shown in Fig. 5C, Bar was misexpressed strongly over the entire presumptive pretarsus region of al− leg discs. Loss of Lim1 expression in al− leg discs may thus arise from a secondary effect of Bar misexpression in the absence of al activity.
Repression of Lim1 expression by Bar misexpression
To clarify whether Bar is capable of repressing the pretarsus Lim1 expression, UAS-BarH1 was driven by ptc-GAL4 to determine the effects of Bar misexpression on the pretarsus Lim1-lacZ expression. As shown in Fig. 5D,E, endogenous Lim1-lacZ expression was almost completely abolished along the A/P border where Bar was misexpressed, thus confirming that Bar is capable of repressing Lim1 expression in the presumptive pretarsus. Hence, the idea that loss of Lim1 in al− leg discs is caused by Bar misexpression, which is induced in the absence of al activity, was again supported, although the possibility that al activates Lim1 expression independently of Bar cannot be formally excluded.
Requirement of Lim1 for al expression
Lim1 expression requires al activity but this does not necessarily rule out the possibility that al expression is governed by Lim1. To confirm this, Lim1 was misexpressed along the A/P border using blk-GAL4 and UAS-Lim1 (see Materials and Methods) or mosaic clones mutant for Lim1 were made. Fig. 5F-G” demonstrate that AL misexpression results from Lim1 misexpression not only in the BAR domain but more proximal regions as well. In contrast, although AL expression in the pretarsus was not completely eliminated in Lim1− leg discs (Fig. 5H), it was evident that AL expression was significantly reduced in a cell-autonomous fashion in Lim1− clones generated in the pretarsus (Fig. 5I,I′). Moreover, AL expression in the region other than the pretarsus was also substantially reduced or completely eliminated, as described below (see Fig. 6). Therefore, Lim1 probably activates al expression in all al-expressing leg and antennal cells, including those in the pretarsus.
Involvement of Lim1 and al in normal smooth border formation between the pretarsus and tarsal segment 5
Formation of the tarsal segment 5/pretarsus boundary requires antagonistic interactions between Bar and al (Kojima et al., 2000). To determine whether Lim1 is involved in this process, the effects of the absence of Lim1 activity on Bar expression were examined. As with al hypomorphic mutants, BAR expression appeared virtually normal in nearly all cases (Fig. 5J,K, Table 1). However, about 80% of leg discs showed BAR misexpression in the pretarsus in double mutants of Lim1− and al1 (Fig. 5L, Table 1), indicating the involvement of Lim1 in the repression of Bar expression.
Fasciclin 2 (FAS2), a putative protein involved in cell-cell connection (Grenningloh et al., 1991), is strongly expressed in border cells separating the pretarsus and tarsal segment 5 cells (Kojima et al., 2000). Although FAS2 expression was almost normal in al1 discs (Fig. 5J) and only slightly reduced in Lim1− discs (Fig. 5K), most FAS2 expression was eliminated in double mutants (Fig. 5L), indicating that both al and Lim1 are involved in the regulation of FAS2 expression in border cells. Interestingly, the normal smooth boundary between the pretarsus and tarsal segment 5 was replaced by irregularly zigzagged one in the double mutant discs (Fig. 5L). However, it should be noted that any appreciable change in morphology of the pretarsus/tarsus boundary cannot be brought about solely by eliminating FAS2 activity (data not shown), suggesting the involvement of unknown factors functionally redundant to FAS2 in normal pretarsus/tarsus boundary formation.
Requirement of Lim1 for normal development of the femur and coxa
Apart from the future pretarsus, Lim1 was expressed circularly in proximal segments such as the coxa, femur and tibia (see Fig. 1). In Lim1− flies, the femur was extensively reduced in size (Fig. 6A,B) and the coxa was missing for the most part or present only as a small bulb-like structure (Fig. 6B, left inset), suggesting the requirement of Lim1 for proper development of the femur and coxa. Although the tibia was bent and fused with the femur, morphological analysis indicated the presence of essentially normal characteristic structures of the tibia, such as transverse rows of bristles, preapical bristles, tibial sense organs and tibial sensilla trichodea (Bryant, 1978; data not shown); tibial sense organs and tibial sensilla trichodea are structures situated near the proximal tibial end (Fig. 6A,B, right inset). The tibial phenotype may thus possibly derive from secondary effects of the femoral deformation. In late third instar, DLL expression is evident in the central region spanning from the most distal tip to distal half of the tibia along with in the future trochanter (Fig. 6C, see also Fig. 1A; Diaz-Benjumea et al., 1994). Consistent with shortening of the femur, appreciable reduction in mass has already taken place in the region flanked by the central DLL domain and the proximal DLL ring at late third instar (Fig. 6C-F).
In Lim1− leg and antennal discs, AL expression in the proximal region, such as in the femur, coxa and first antennal segment, was virtually absent (Fig. 6G-I). In Lim1 mosaic clones in the femur or coxa, AL expression was abolished cell autonomously (Fig. 6J,J′). Tibial AL expression remained in Lim1 discs (Fig. 6G,H) but mosaic analysis clearly indicated substantial reduction in AL expression in Lim1 clones (Fig. 6K,K′). But loss of AL expression would not completely explain the femoral and coxal defects, since al is dispensable for normal development of the femur and coxa (Campbell and Tomlinson, 1998).
Possible roles of Lim1 in pretarsus development
We showed here that Lim1 is coexpressed with al in the future pretarsus (Figs 1 and 4) and required for proper pretarsus development (Fig. 3E). Since the pretarsus phenotype of Lim1− legs was similar to that of moderate al hypomorphic mutant legs (see Fig. 3C,E), the requirement of Lim1 for pretarsus formation may be less than that of al. The pretarsus phenotype of Lim1− legs was enhanced in double mutants of al1 (a very weak hypomorphic al allele) and Lim1− (Fig. 3F), indicating that Lim1 and al are cooperatively involved in pretarsus development.
According to this, and the fact that Lim1 expression in the future pretarsus is completely eliminated in al− leg discs (Fig. 5C′), Lim1 might be considered to lie downstream of al and be involved in only some al functions. However, normal levels of pretarsus AL expression required Lim1 activity (Fig. 5I,I′) and Lim1 misexpression induced AL misexpression (Fig. 5F-G”), indicating that LIM1 rather serves as an activator of al expression. Furthermore, Bar misexpression in the pretarsus caused repression of Lim1-lacZ expression (Fig. 5D) while al misexpression failed to induce ectopic Lim1-lacZ expression (Fig. 5A,B), implying that the elimination of pretarsus Lim1-lacZ expression in al− leg discs is an indirect consequence of the absence of al activity through strong Bar misexpression (Fig. 5C). All these findings and considerations are consistent with the idea that Lim1 lies upstream of al and at least some Lim1 functions in the pretarsus are mediated by activation of al expression (see a solid arrow in Fig. 7), although the possibility that the pretarsus Lim1 expression is partly under the direct positive control of al cannot be formally excluded (see a broken arrow in Fig. 7). al is expressed considerably prior to that of Lim1 (Fig. 4A-A”) and Lim1 may thus be involved in maintenance of pretarsus al expression. The incomplete elimination of pretarsus al expression in Lim1− discs (see Figs 5H-I′ and 6G,H,K,K′) indicates the involvement of one or more positive factors (designated as X in Fig. 7) other than Lim1 in pretarsus al expression.
Mutually antagonistic interactions between al and Bar were previously shown to be essential for the strict separation of AL and BAR domains, leading to localized Fas2 induction by Bar in border cells (Kojima et al., 2000). Although the absence of Lim1 shows little BAR misexpression in the pretarsus (Fig. 5K, Table 1), increased BAR misexpression in Lim1−; al1 leg discs (Fig. 5L, Table 1) could indicate the involvement of Lim1 in the repression of Bar expression in the pretarsus (Fig. 7). Remarkable decrease in FAS2 expression in putative Lim1−; al1 mutant border cells (Fig. 5L) indicates that Fas2 expression requires al and Lim1 functions, in addition to cell non-autonomous functions of Bar (Kojima et al., 2000; Fig. 7). Lim1 may be involved in pretarsus specification and boundary formation only through its activation of al, as shown by the unbroken lines in Fig. 7. Low al expression in Lim1 single mutants may still be sufficient for maintaining the normal expression of Bar and Fas2, but with further reduction in al expression in Lim1−; al1 double mutants, Bar misexpression and loss of Fas2 expression may result. Alternatively, as shown by broken lines in Fig. 7, Lim1 may act independently of al, and simultaneous reduction in al and Lim1 expression may cause Bar misexpression and reduction of Fas2 expression in the double mutants. These considerations are not mutually exclusive.
Previous experiments have shown that pretarsus al expression is partially repressed by misexpressed Bar (Kojima et al., 2000). UAS-Bar driven by ptc-GAL4 repressed pretarsus Lim1 expression along the A/P border almost completely (see Fig. 5D), while al expression was abolished only partially in Lim1 mutants. Thus, Bar is likely to repress al expression indirectly through the repression of Lim1 expression as depicted by unbroken lines in Fig. 7, although direct repression of al expression by Bar cannot be formally excluded. Taken together, our results indicate that the interactions between al, Lim1 and Bar are important for precisely defining the pretarsus region.
Lim1 misexpression indicated no appreciable reduction in Bar expression, even though AL misexpression was induced in the BAR domain (Fig. 5G-G”). al and Lim1 may thus not be sufficient for repressing Bar expression in the future pretarsus. Interestingly, an additional locus (‘clawless’ locus) showing pretarsus defects similar to al mutants when mutated was recently found (T. K. and K. S., unpublished). In clawless mutant leg discs, Bar is misexpressed in the presumptive pretarsus region as in the case of al mutants.
We thank B. J. Dickson, C. S. Goodman, G. Campbell and S. M. Cohen for antibodies and/or fly stocks. We also thank FlyView and Bloomington Drosophila stock centre for fly strains. A. S. is supported by the Research Fellowships of the Japan Society for the Promotion of Science for Young Scientists. This work was supported in part by grants from the Ministry of Education, Science and Culture of Japan to K. S. and T. K.