Dynamin- and Rab5-dependent endocytosis is required to prevent Drosophila photoreceptor degeneration

Late differentiation of the rhabdomeric Drosophila photoreceptor poses significant challenges. These highly polarized cells have to elongate along the proximo-distal axis (i.e. from the lens to optic lobe) by ~tenfold; they also have to grow a rhabdomere at their apical tip (Ready, 2002). The rhabdomere is the light-gathering organelle of the cell and comprises a stack of ~60,000 tightly packed microvilli. Each microvillus contains two microfilaments of actin, which project their fast-growing pointed end towards the cytoplasm and away from the plasma membrane (Arikawa et al., 1990). The visual pigment Rhodopsin 1 (Rh1), encoded by the neither inactivation nor afterpotential E (ninaE) locus, is expressed in the terminally differentiating photoreceptors when pupal development (pd) reaches 70% and is required for proper rhabdomere morphogenesis and maintenance (Cowman et al., 1986; Kumar et al., 1997; Kumar and Ready, 1995; O'Tousa et al., 1985; O'Tousa et al., 1989; Zuker et al., 1985). Rhodopsin is the principal membrane protein found in the rhabdomere (Paulsen and Schwemer, 1979; Schwemer and Henning, 1984), although rhabdomere morphogenesis is initiated well before Rh1 is expressed (Kumar and Ready, 1995). At the onset of ninaE expression, the apical stack of microvilli is already well defined and is separated from the subapical membrane, which is called stalk membrane (Ready, 2002). Rhabdomere morphogenesis relies heavily on apically targeted membrane delivery. During this process, Rh1-Rab11-Rip11-positive vesicles, called rhodopsin-bearing transport carriers (RTCs), traffic from the trans Golgi network towards the base of the rhabdomeres in a Myosin V (MyoV)-dependent manner (Li et al., 2007; Satoh et al., 2005). In this context, Rh1 is required to generate a rhabdomere terminal web (RTW) that consists of a meshwork of F-actin cables, which occupies the sub-rhabdomeric space and merges with the base of the rhabdomere (Chang and Ready, 2000). Importantly, the RTW has been proposed to serve as a conduit for the Rab11 –Rip11-MyoV trafficking route to properly target rhabdomeric membrane to the apical pole of the cell (Li et al., 2007). In addition, the RTW might serve as a mechanical fence, supporting the enormously amplified rhabdomeric membrane (Chang and Ready, 2000; Kumar and Ready, 1995; Ready, 2002). Accordingly, a complete loss of ninaE expression leads to the collapse of rhabdomeric microvilli inside the photoreceptor cytoplasm, a phenotype that is observed at ~90% pd and exacerbates within the first hours of the life of the adult fly (Kumar et al., 1997; Kumar and Ready, 1995; O'Tousa et al., 1989).

Rh1 function in promoting rhabdomere and RTW morphogenesis is independent to its sensory role, in that it does not rely on Phospholipase C (PLC), Transient receptor potential (TRP) channels or neither inactivation nor afterpotential C (ninaC), which are major components of the phototransduction cascade (for a review, see Wang and Montell, 2007). However, it is at present not clear whether the heterotrimeric protein G_qα_eβ_eγ_e is involved in this function. Furthermore, Rh1 function in RTW morphogenesis is linked to that of the small Rho GTPases Rac1 and Cdc42, as activated forms of Rac1 or Cdc42 are able to rescue the RTW and hence the rhabdomere phenotype of a ninaE-null mutant (Chang and Ready, 2000). In addition, a dominant-negative version of Rac1 leads to the same phenotype as in ninaE-null flies (Chang and Ready, 2000). However, the phenotype of the rac1 loss of function has not been analysed during rhabdomere morphogenesis.

To investigate the nature of the Rh1-Rac1-Cdc42 pathway (Chang and Ready, 2000) during the late phase of photoreceptor rhabdomere development, we used the null mutant ninaE^l17 (O'Tousa et al., 1985). Although rhabdomere development occurs well before the onset of ninaE expression (Ready, 2002) (Fig. 1A–C), the first rhabdomere development defects are observed at ~85% pd in the ninaE^l17 mutants (Kumar and Ready, 1995) (Fig. 1D,E). At this stage, short intrusions of closely apposed membranes emanating from the base of the rhabdomere microvilli are readily detected in the cytoplasm of the cell, disrupting the normally well-defined boundary (compare Fig. 1B with 1E) between the base of the rhabdomere and the cytosol. As described previously, these short intrusions of closely apposed membrane further develop with time to eventually represent a significant proportion of the rhabdomeric membrane collapsing inside the cell (Fig. 1F,G).

The ninaE^l17 phenotype can be rescued by a pulse of activated forms of the small GTPase Rac1 (V12Rac1) or Cdc42 (V12Cdc42) (Chang and Ready, 2000). Conversely, overexpression of a dominant-negative version of Rac1 (N17Drac1) (Luo et al., 1994) leads to a rhabdomeric phenotype that resembles that of ninaE^l17 (Chang and Ready, 2000). However, rac1 might function redundantly with cdc42 downstream of ninaE. To address this issue, we generated eyes deficient for rac1, rac2 and Mig-2 like (mtl), the three loci encoding Rac proteins (Hakeda-Suzuki et al., 2002; Ng et al., 2002). Although the previously described axon guidance phenotype was present in the corresponding optic lobes (Hakeda-Suzuki et al., 2002; Ng et al., 2002) (Fig. 1J,K), no rhabdomeric defect was observed at the base of the rhabdomere at eclosion (Fig. 1H,I). These data demonstrate that rac function is dispensable for rhabdomere and RTW morphogenesis. This suggests that, in the absence of rac function, cdc42 is able to function downstream of ninaE. However, as cdc42 function is crucial for early phases of photoreceptor development (Walther and Pichaud, 2010), this prevented us from further testing for this gene function during the later rhabdomere and RTW morphogenesis.

Rh1 function in rhabdomere and RTW morphogenesis is not mediated by the phototransduction cascade (Bloomquist et al., 1988). However, the detailed nature of the coupling between Rh1 and rhabdomere morphogenesis remains elusive. In particular, a role for the heterotrimeric G-proteins and in particular Gqαe (officially known as Gα49B) remains to be tested. The phenotype of a loss-of-function allele (the strong hypomorphic mutation Gqαe¹) does not resemble the ninaE^l17 phenotype (Han et al., 2007) (data not shown). However, this might be because Gqαe¹ is not a null allele (Lee et al., 1990). This is particularly important, as it has been estimated that 1% Rh1 expression is sufficient to achieve proper rhabdomere morphogenesis (Leonard et al., 1992). To address this issue, we made use of an embryonic lethal null allele for Gqαe, dgq^221c (Banerjee et al., 2006). Generating whole mutant eyes for this null allele did not reveal any detectable phenotype in 1-day-old flies (Fig. 2A), demonstrating that Gqαe is not required for Rh1 function during rhabdomere and RTW morphogenesis. Furthermore, we could not rescue the ninaE^l17 phenotype by expressing a constitutively active form of Gqαe (Gqαe^Q203L) (Fig. 2B,C), which localizes to the rhabdomere (supplementary material Fig. S1).

Rh1 is part of a large family of G-protein-coupled receptors (GPCRs) that includes invertebrate and vertebrate opsins. To test whether the ability of Rh1 to promote morphogenesis is conserved throughout evolution, we first sought to assay the ability of invertebrate opsins to rescue the rhabdomeric phenotype observed in ninaE^l17 flies. We used the blue- and UV-sensitive opsins from honeybee (Apis mellifera) (Townson et al., 1998), which has evolved separately from Drosophila for ~250 million years (Riek, 1970). We also used the lateral eye opsin from horseshoe crab (Limulus polyphemus) (Knox et al., 2003; Smith et al., 1993), which diverged from Drosophila over 520 million years ago. Expressing this set of opsins individually in otherwise ninaE^l17 mutant flies led to a rescue of the corresponding rhabdomere morphology, including the characteristic well-defined base of the rhabdomere (Fig. 3A,B). These data demonstrate that the ability of Rh1 to promote RTW morphogenesis is conserved through the arthropod phyla. We next extended our assay to human melanopsin, which belongs to the family of rhabdomeric opsins and is involved in circadian rhythm entrainment in retinal ganglion cells (RGCs) (Melyan et al., 2005; Panda et al., 2005). Expressing human melanopsin in ninaE^l17 eyes led to a substantial rescue of the architecture at the base of the rhabdomere (Fig. 3C); however, this base was distorted and small, indicating that human melanopsin does not fully support rhabdomere development. Finally, we expressed the bovine ciliary opsin (rhodopsin) in ninaE^l17 eyes (Ahmad et al., 2006). It is known that bovine rhodopsin can also rescue the early developmental failure in rhabdomere morphogenesis (Ahmad et al., 2007), and the corresponding photoreceptors indeed appeared identical to those in wild-type flies (Fig. 3D). Altogether, these experiments demonstrate that the Rh1 function in promoting rhabdomere and RTW morphogenesis is highly conserved. Because bovine rhodopsin cannot activate the heterotrimeric G proteins (Gqαe) involved in invertebrate phototransduction (Ahmad et al., 2006), this also confirms that the Rh1 function in rhabdomere and RTW morphogenesis is independent of heterotrimeric G-protein function.

Rh1 function during rhabdomere and RTW morphogenesis implies an interface between Rh1 and cytoplasmic effectors. In this context, the NPxxY motif in the seventh transmembrane domain of Rh1 represents an ideal candidate. Indeed, this domain has previously been shown to bind small GTPases of the Rho family (Balasubramanian and Slepak, 2003; Elsaesser et al., 2010; Mitchell et al., 1998) and is conserved in the invertebrate and vertebrate opsins that rescue the ninaE^l17 mutant. We, therefore, tested whether this domain is necessary for Rh1 to rescue the ninaE^l17 phenotype. We replaced Asn325, Pro326 and Tyr329 in the NPxxY motif with alanine residues. The corresponding recombinant protein did not traffic normally to the rhabdomere and seemed to reside predominantly in the cell cytosol (Fig. 4A–D). It nevertheless significantly rescued the collapsing of the closely apposed membrane inside the cells that is seen in ninaE^l17 mutant photoreceptors (Fig. 4E–G). However, there were numerous vesicles occupying the sub-rhabdomeric space. These were not characterized by closely apposed membranes and did not appear to be coated. However, they were often associated with the base of the rhabdomere. This suggests that the NPxxY motif is not necessary for the interaction between Rh1 and the downstream effectors involved in preventing the collapse of rhabdomeric membranes into the cytosol. However, this motif might be required for other aspects of rhabdomeric membrane apposition or sub-rhabdomeric organization.

The RTW has been proposed to act as a conduit for directing membrane delivery from the trans Golgi network to the base of rhabdomere. In photoreceptors, Rab11–Rip11-MyoV vesicles, which contain Rh1 and represent the main delivery mode for the visual pigment, are likely to be moving onto the F-actin cables that are part of the RTW (Li et al., 2007; Satoh et al., 2005). Because localized F-actin morphogenesis is a key parameter for regulating membrane endocytosis (Engqvist-Goldstein and Drubin, 2003; Smythe and Ayscough, 2006), we wondered whether the RTW could also represent a conduit for rhabdomeric membrane endocytosis and thus overall membrane turnover. We first made use of a conditional dominant-negative mutation in Dynamin (Shibire thermo-sensitive, Shi^ts) (Poodry and Edgar, 1979). Dynamin is a main effector of early endocytic vesicle scission from the plasma membrane (Sever, 2002). Blocking this process by shifting the shi^ts fly strain to restrictive temperature for 4 hours led to significant and specific disruption of the base of the rhabdomere. In particular, we readily detected the characteristic apposed rhabdomeric membrane collapsing inside the photoreceptor cytoplasm, similar to the phenotype of ninaE^l17 mutants (Fig. 5A,B). This demonstrates that Dynamin activity is required to generate the proper architecture of the rhabdomere base. To complement this experiment, we used a dominant-negative form of Rab5 (Rab5^DN), a small GTPase required for early endocytic steps including early vesicle formation and the transition towards more mature endocytic compartments (Markgraf et al., 2007; Rink et al., 2005). When expressed in the developing photoreceptor, using the GMR-Gal4 driver, Rab5^DN caused a phenotype identical to that of ninaE^l17 (Fig. 5C,D). Sheets of apposed plasma membrane, emanating from the rhabdomere, were readily detected. Importantly, this phenotype was light-independent, as it still arose in dark-raised flies. In addition, in our experimental conditions, we did not detect apposed membrane in the cell cytosol before the onset of Rh1 expression (supplementary material Fig. S1). There are potential caveats of using a dominant negative approach; however, taken together with our data demonstrating that Dynamin function is also required in this pathway, these data indicate that endocytosis is crucial for the proper development of the interface between the rhabdomere and the cytosol.

Coupling of GPCRs to the cell cytoskeleton has been documented in several instances of directed cell migration (Cotton and Claing, 2009). In addition, sensory GPCRs, such as mouse olfactory receptors, have been implicated in axon guidance, a function that is likely to be independent from their odorant-dependent sensory function (Feinstein and Mombaerts, 2004; Imai et al., 2006). Similarly, ninaE is required upstream of the small Rho GTPases rac1 and possibly cdc42, to promote morphogenesis of the RTW (Chang and Ready, 2000), which serves as a conduit for Rab11-dependent delivery of Rh1-rich vesicles from the trans Golgi network to the rhabdomere (Li et al., 2007; Satoh et al., 2005). Such remarkable function for a sensory receptor is independent of the phototransduction cascade and is not well understood. Here, we have focused our study on the early onset of the rhabdomeric defect observed in the total absence of ninaE function.

There are several examples of GPCRs being involved in promoting directed cell migration though the associated Gα or Gβγ G-proteins (Cotton and Claing, 2009). In Dictyostelium, the cAMP receptor utilizes Gα–GTP to promote pseudopod formation in response to cAMP stimulation (Parent and Devreotes, 1999). This depends upon F-actin remodelling at the leading edge of the cell, a process typically relying on localized Rho GTPase activity. Similarly, in human leukocytes, chemokine GPCRs play a role during directed cell migration and hence polarized F-actin remodelling (Viola and Luster, 2008). This process is classically associated, at least in part, with Gα subunit activation, although other routes including ARF6-dependent regulation of Rac1 and RhoA have been reported (Cotton et al., 2007). Our work clearly demonstrates that Rh1 function during rhabdomere and RTW morphogenesis is independent from G_qα_e. Rh1 might be directly coupled to Drosophila Rac1 and Cdc42. Interestingly, both ARF6 and RhoA co-immunoprecipitate with rhodopsin-family receptors in N1321N1 cells, an association that depends upon the conserved NPxxY motif located immediately distal to the seventh transmembrane helix (Mitchell et al., 1998). In addition, both vertebrate (Balasubramanian and Slepak, 2003) and invertebrate opsins (Elsaesser et al., 2010) have been shown to co-immunoprecipitate with Rac1 and Rac2, respectively. In vertebrate photoreceptors, the association between the opsin and Rac1 also depends upon the presence of the NPxxY motif (Balasubramanian and Slepak, 2003). Finally, although in vertebrates Rac1 activation is light-dependent, work in Drosophila indicates that the association of Rh1 with Drosophila Rac2 is light-independent (Elsaesser et al., 2010). Altogether, this suggests that the NPxxY family of opsin receptors is able to associate with Rac proteins. However, in the case of rhabdomere morphogenesis, our data indicate that the NPxxY domain is largely dispensable. Nevertheless, rescue experiments performed with the corresponding Rh1^AAXXA transgene suggest a function for this motif during membrane apposition at the base of the rhabdomere. In addition, an excess of vesicles was present in the photoreceptors and at the base of their rhabdomeres. These could represent either early endosomes failing to fuse and form multivesicular bodies (MVBs), or alternatively an excess of exocytic vesicles. Furthermore, we failed to demonstrate any involvement of rac1, rac2 and mtl during rhabdomere morphogenesis. This is puzzling given that an activated form of Drosophila Rac1 is able to rescue the early onset of rhabdomere defects observed in ninaE^l17 mutants (Chang and Ready, 2000). Interestingly, an activated form of Cdc42 (V12cdc42) can also rescue the early onset of rhabdomere defects. However, although cdc42 can act upstream of the rac genes (Hall, 1998), our data indicate that, in the case of Rh1 function, during rhabdomere morphogenesis, these two GTPases might instead function redundantly. Interestingly, systematic screening of the fly genome for all guanine exchange factors (GEFs) and GTPase-activating proteins (GAPs) using a collection of RNA interference (RNAi) lines (NIG-Fly and VDRC) and electron microscopy did not lead to any ninaE^l17-like phenotype (data not shown). This suggests that more than one GEF is involved in this pathway, a hypothesis that is compatible with the idea that rac1 function might be redundant with that of cdc42.

The ninaE^l17 allele is a null allele for the ninaE locus and it shows rhabdomeric phenotypes, including apposed membrane intrusions in the cell cytosol. Because these appear before the animal has eclosed, it is compatible with an early developmental role for this opsin during rhabdomere morphogenesis (Chang and Ready, 2000). However, examining other ninaE alleles has demonstrated differences in the onset of rhabdomere degeneration (Leonard et al., 1992). In addition, rescue experiments using bovine rhodopsin expression in ninaE^l17 photoreceptors, reveals that even though newly eclosed flies have wild-type rhabdomeres (Ahmad et al., 2006) (Fig. 3D), these start degenerating 7–10 days after eclosion. Overall, this suggests that there are different phases during which the opsin is required. In this model, the early onset of the ninaE^l17 phenotype is attributed to a failure in building the RTW of F-actin (Chang and Ready, 2000; Colley, 2000). The late onset of degeneration might in turn, be due to a suboptimal coupling of the opsin to structural components of the cells, such that, with time, a more progressive degeneration is observed.

Here, we find that interfering with the function of Dynamin or Rab5 in the developing photoreceptor, leads to a phenotype that is indistinguishable from ninaE^l17. Interestingly, such a phenotype is also observed in adult wild-type flies and is associated with the light-dependent reversible translocation of Gqαe from the rhabdomere to the cytosol (Kosloff et al., 2003). In all cases, this is accompanied by a profound alteration of the base of the rhabdomere (Kosloff et al., 2003; Kumar et al., 1997). However, the molecular basis and, in particular, a link between such phenomena and the endocytic pathway is not clear. On the one hand, it has been proposed that rhabdomeric membrane turnover relies on the formation of coated pits and coated vesicles originating from the base of the rhabdomere and merging into MVBs (Sapp et al., 1991; Stark and Sapp, 1987). Indeed, Rh1-containing MVBs have been reported during early phases of rhabdomere development, suggesting that the endocytic pathway could eventually exceed the rate of rhabdomeric membrane delivery during rhabdomere development (Satoh et al., 2005). On the other hand, the ninaE^l17 mutant photoreceptors do not present MVBs (Stark and Sapp, 1987), a situation that is also observed upon interfering with Rab11 and Rab5 function (Satoh et al., 2005), and this is consistent with the idea that these MVBs are derived from rhabdomeric membrane. Altogether, this raises the possibility that by blocking Dynamin and/or Rab5 function in the photoreceptor, we are preventing MVB morphogenesis, a situation resulting in the accumulation of rhabdomeric membrane in the cell. This hypothesis implies that, in wild-type photoreceptors, an important part of Rh1 function during rhabdomere morphogenesis relies on the endocytic pathway. In particular, we favour a model in which Rh1 function in promoting RTW morphogenesis is required for proper rhabdomeric membrane endocytosis, including formation of MVBs. Indeed, F-actin morphogenesis has previously been shown, in yeast, flies and vertebrates (Merrifield et al., 2002; Merrifield et al., 2005), to be instrumental in modulating endocytosis. In this context, it has been shown in yeast that F-actin turnover (i.e. regulated rounds of polymerization and depolymerization) is an important parameter during endocytosis (Ayscough, 2000; Kaksonen et al., 2005). A role for F-actin during endocytic vesicle scission is also compatible with the observations of long tubular structures in yeast (Jonsdottir and Li, 2004; Wendland et al., 1996) and, more recently, in Drosophila epithelial cells, upon interfering with component of the F-actin cytoskeleton (Georgiou et al., 2008; Leibfried et al., 2008). Our data demonstrate that the strong requirement for the opsin in organizing the base of the rhabdomere and the RTW is paralleled by a strong requirement for the endocytic pathway. This finding demonstrates that Dynamin- and Rab5-dependent endocytosis is required for setting up the interface between the rhabdomere and the sub-rhabdomeric space, a feature that is crucial for preventing early onsets of photoreceptor apical organelle degeneration. In this model, Rh1 could serve as a passive contributor to endocytosed membrane or, alternatively, could play a more active role in promoting rhabdomere membrane endocytosis. Further work will be required to address this issue.

The following genotypes were used: ninaE^l17, sr, e (Chang and Ready, 2000) yw; UAS-YFPRab5^S43N (Zhang et al., 2007) pRh1-Rho, ninaE^l17, sr, e (Ahmad et al., 2006) yw, pRh1-Limulus [y+];;ninaE^l17, sr (Knox et al., 2003) yw, p Rh1-BeeBlue [y+];;ninaE^l17, sr and yw, p pRh1-BeeUV [y+];;ninaE^l17, sr (Townson et al., 1998) GMRGal4; ninaE^l17, sr, e (Freeman, 1996) shi^ts1 (van der Bliek and Meyerowitz, 1991). The dgq^221c mutation (Banerjee et al., 2006) was recombined onto an FTR42D chromosome and crossed with w;FRT42D, GMRhid,cl/Cyo; EGUF (Stowers and Schwarz, 1999). The mtl^Δ mutation (yw;;FRT82B, mtl^Δ/TM3, Sb) was recombined onto the right arm of the GMRhid, cl, FRT2A (eyflp; GMRhid, cl, FRT2A/TM6B, Tb) chromosome. Recombinants were tested using PCR on the mtl locus. Two independent recombinants were used to generate whole mutant eyes for the three rac loci crossed with yw;;rac1^J11, rac2^Δ, FRT2A, mtl^Δ/TM6B, Tb (Hakeda-Suzuki et al., 2002; Ng et al., 2002). The overexpression of YFPRab5^S43N was performed by crossing the UAS-YFPRab5^S43N transgene with the GMRGal4 driver. The corresponding flies were raised at 18°C until 65% pd and then shifted to 25°C and kept in the dark until dissection at eclosion. The effect of this transgene on rhabdomere morphogenesis, before the temperature shift, was performed in the same conditions as the assays, with retina being dissected at approximately 65% pd.

The cDNA encoding human melanopsin (tagged with His₆) (Melyan et al., 2005) was amplified using PCR, subcloned in the PCR R 2.1-TOPOR vector (Invitrogen) and then cloned in the pUAST vector. The pUAS-Gqαe^Q203L and the pninaE-Rh1^AAXXA transgenes were generated by two rounds of PCR, introducing the Q203L mutation and AAxxA mutations in the PCR primers. The following primers were used: Gqαe^Q203L transgene, ATGforward, 5′-ATGGAGTGCTGTTTATCGGAGG-3′ and MUTreverse, 5′-CGGATCGCAGACCACCGACG-3′, and TGAreverse, 5′-TGCTCTAGAGCACCAGTTTAGACCAAATTATATTCCTTAAGG-3′ and MUTforward 5′-CGTCGGTGGTCTGCGATCCG-3′; Rh1^AAXXA mutant transgene, ATGforward, 5′-ATGGAGAGCTTTGCAGTAGCAGCCGC-3′ and MUTreverse 5′-GCCGGCTACAATTGCAGCGTAGCAGGCG-3′, TGAreverse, and 5′-TTATGCCTTGGACTCGGCCTCGCTGG-3′ and MUTforward, 5′-CGCCTGCTAC-GCTGCAATTGTAGCCGGC-3′. The PCR products were subcloned in the PCR R 2.1-TOPOR vector (Invitrogen) and then cloned in the pUAST vector (Gqαe^Q203L) and in the pCHAB containing the minimal ninaE promoter (Rh1^AAXXA).

Immunochemistry on adult retinas was performed as described previously (Walther and Pichaud, 2006). The following antibodies were used: mouse anti-Rh1 antibody (at 1:50) and the 24B10 (at 1:50) antibody (Developmental Studies Hybridoma Bank) with the corresponding Alexa-Fluor-488-conjugated secondary antibodies (1:100) (Molecular Probes). Phalloidin–TRITC (Sigma-Aldrich) was used to stain F-actin. Imaging was performed with a Leica SP5 confocal microscope, and the images were edited with ImageJ and Adobe Photoshop.

Electron microscopy was performed according as described previously (Tomlinson et al., 1987) on 1-day-old flies. Images were acquired with a Tecnai G2 Spirit transmission electron microscope (FEI) equipped with a Morada charge-coupled device camera (Olympus Soft Imaging Systems).

We thank S. Britt, J. O'Tousa, G. Hasan, R. Lucas, the Bloomington Drosophila Stock Center and the Developmental Studies Hybridoma Bank for providing reagents, and the members of our laboratory for their comments on the manuscript. We also thank J. Burden and I. White for their help and support with electron microscopy. N.P. is a recipient of an MRC Career and development fellowship. Work in the F.P. laboratory is funded by Medical Research Council, UK. Deposited in PMC for release after 6 months.