IMP regulates Kuzbanian to control the timing of Notch signalling in Drosophila follicle cells

RNA-binding proteins (RBPs) play diverse roles in the post-transcriptional regulation of gene expression by controlling the splicing, stability, translation or subcellular localisation of specific mRNAs. One of the best studied classes of RBPs is the conserved family of IGF2 mRNA-binding proteins (IMPs, also known as the VICKZ family), which are characterised by four conserved KH domains, with KH3 and KH4 being most important for RNA binding, and two N-terminal RRM domains (Degrauwe et al., 2016). Initial studies on IMPs pointed to an important role in mRNA localisation. The Xenopus IMP3 orthologue, Vg1RBP/Vera (Igf2bp3), binds to the localisation signal in vg1 (gdf1) mRNA and colocalises with it to the vegetal cortex of the Xenopus oocyte (Deshler et al., 1997; Havin et al., 1998). Similarly, the chicken IMP1, ZBP1 (IGF2BP1), binds to the 54-nucleotide localisation signal in β-actin mRNA to mediate its localisation to the periphery of fibroblasts and the dendrites of neurons (Farina et al., 2003; Tiruchinapalli et al., 2003). However, IMPs also regulate mRNA translation and mRNA stability. Mammalian IMP1-3 were initially identified as translational regulators of insulin-like growth factor II (Igf2) mRNA and ZBP1 represses the translation of β-actin mRNA until it reaches its destination (Hüttelmaier et al., 2005; Nielsen et al., 1999; Yao et al., 2006). One mechanism by which IMPs regulate mRNA translation and stability is by preventing the binding of siRNAs and miRNAs to their targets, either by masking the binding sites or by sequestering the mRNA away from the Argonaute/RISC complex (Degrauwe et al., 2016). In many cases, IMPs have been found to play an important regulatory role, although the relevant RNA targets have not been identified. For example, IMP1 and 3 are upregulated in a number of tumours, with their expression levels correlating with increased metastasis and poor prognosis (Degrauwe et al., 2016; Ioannidis et al., 2001; Nielsen et al., 2000).

Vertebrates contain three closely related IMP paralogues, which has hampered functional analysis, whereas Drosophila contains a single IMP orthologue with four well-conserved KH domains, allowing the genetic analysis of IMP function (Nielsen et al., 2000). IMP was found to bind directly to oskar and gurken mRNAs and localise with them to the posterior and dorsal sides of the oocyte, respectively (Geng and Macdonald, 2006; Munro et al., 2006). Although the IMP-binding sites are required for oskar mRNA translation and anchoring, loss of IMP has no obvious phenotype, suggesting that it functions redundantly with other proteins in the germ line. IMP is strongly expressed in the developing nervous system and RNAi knockdown causes neuronal loss and axon-pathfinding defects and a reduced number of boutons at the neuromuscular junctions (Boylan et al., 2008; Koizumi et al., 2007). imp mutant clones in the developing adult brain cause similar defects in axon elongation in mushroom body neurons, at least in part through IMP's role in regulating the localisation of chic mRNA (Medioni et al., 2014). These neural phenotypes may be related to IMP's function as temporal identify factor that acts in opposition to Syncrip to specify early-born neuronal fates and to promote neuroblast proliferative capacity (Liu et al., 2015; Narbonne-Reveau et al., 2016). IMP also acts as part of a temporal programme that controls the aging of the testis hub cells. IMP protects unpaired mRNA from repression by miRNAs in these cells and, as IMP levels fall with age, Unpaired signalling, which maintains the male germline stem cells, declines, leading to stem cell loss (Toledano et al., 2012).

Here, we analyse the function of IMP during the development of the somatic follicle cells of the Drosophila ovary and show that it also controls the temporal programme of development in this tissue. Unlike other well-characterised roles of IMP, we find that IMP functions independently of the microRNA pathway to regulate the timing of Delta/Notch signalling.

To investigate the role of IMP in the follicle cell layer, we generated clones that were homozygous for the null allele imp⁷, marked by the loss of RFP (Munro et al., 2006). The mutant cells showed no phenotypes during early oogenesis up until the end of stage 6. However, phalloidin staining of actin revealed that mutant cells at later stages were smaller in size and were more densely packed than the surrounding wild-type cells (Fig. 1A,A′). We observed the same phenotype with a second null allele, imp⁸ (Munro et al., 2006). imp mutant cells also have smaller nuclei (Fig. 1A,C,C′). The size and number of follicle cells is determined by the timing of the mitotic-to-endocycle transition, which takes place at stage 6, when the germ cells in the egg chamber produce the DSL ligand Delta to activate the Notch pathway in the follicle cells (Deng et al., 2001; Lopez-Schier and St Johnston, 2001). Analysis of 56 imp mutant clones revealed that there are twice as many mutant cells in each clone than there are wild-type cells in the twin spot clone induced at the same time (Fig. 1B). Thus, imp mutant cells go through one extra round of mitosis, suggesting that Delta/Notch signalling is delayed.

Notch activation controls both the mitosis-to-endocycle switch and follicle cell differentiation. The late differentiation in the absence of IMP leads to delays in several other aspects of follicle cell development. For example, the lateral follicle cells move posteriorly to form a columnar epithelium around the oocyte during stages 8-9, but imp mutant cells do this more slowly and later than normal (Fig. 1C,C′). During stage 9 of oogenesis, the anterior-most follicle cells adopt the border cell fate, delaminate from the epithelium and migrate between the nurse cells to the anterior of the oocyte (Fig. 1D,E). When the entire border cell cluster is mutant for imp, the cells frequently fail to delaminate and remain at the anterior of the egg chamber, whereas those that do delaminate often only migrate part of the way to the oocyte (Fig. 1F,H). When the cluster contains both mutant and wild-type cells, the wild-type cells lead the migration with the mutant cells trailing behind (Fig. 1G). Thus, loss of IMP affects the timing of all aspects of follicle cell behaviour, suggesting that it plays a general role in this process. The localisation of IMP does not give any clues as it its function, however, as IMP protein is uniformly distributed throughout the cytoplasm of the follicle cells (Fig. S1A-A″).

To test whether IMP is required for the proper timing of Notch pathway activation in the follicle cells, we stained imp mutant clones for Cut and Hindsight (Hnt; also known as Pebbled). Cut is expressed from stages 1 to 6 of oogenesis and is downregulated at stage 7 in response to Notch activation (Sun and Deng, 2005). By contrast, Hnt is expressed only in post-mitotic cells that already received the Delta signal from the germ line (Sun and Deng, 2007). imp clones continue to express Cut at stage 7 in contrast to wild-type cells in the same egg chamber (Fig. 2A,A′). However, Cut expression is lost in the majority of mutant cells at stage 8 (Fig. 2B,B′). On the other hand, Hnt is not expressed in imp mutant cells at stage 7 as in wild type, but appears one stage later (Fig. 2C-D′). Both the continued expression of Cut and the late activation of Hnt in imp⁷ MARCM (mosaic analysis with a repressible cell marker) clones are rescued by expressing UAS-IMP, indicating that these phenotypes are caused the loss of IMP (Fig. S2A-D′). Cut and Hnt both give a binary readout of whether Notch signalling has reached a threshold and we therefore examined the level of Notch signalling more directly by monitoring the expression of the E(spl)m7lacZ reporter (Assa-Kunik et al., 2007). Staining for β-galactosidase revealed that imp mutant follicle cells show much weaker expression of the reporter than wild-type cells at stage 6, but express nearly normal levels by stage 7 (Fig. 2E-F′). As β-galactosidase is very stable, this suggests that Notch signalling is impaired but not abolished by the loss of IMP, so that it takes longer to reach the threshold for the follicle cell response to Delta signalling from the germline, resulting in the delay in follicle cell differentiation.

To test directly whether impaired Notch signalling is responsible for the imp phenotype, we used the MARCM system to express a constitutively active form of Notch, the Notch intracellular domain (NICD), in imp mutant cells (Go et al., 1998; Lee and Luo, 2001). Control imp mutant MARCM clones do not turn on Hnt at stage 7, but expressing NICD in the mutant cells restores timely Hnt expression (Fig. 2G-H′). This indicates that IMP controls Notch activity upstream of NICD production.

Upon binding to its ligand, Delta, the extracellular domain of Notch is cleaved at the S2 site by the ADAM10 protease, Kuzbanian, to produce a transient form of the receptor, NEXT, which contains the transmembrane and intracellular domains of Notch (Lieber et al., 2002; Pan and Rubin, 1997). NEXT then undergoes a second cleavage at the S3 site mediated by the Presenilin/γ secretase complex (De Strooper et al., 1999; Struhl and Greenwald, 1999; Vaccari et al., 2008; Ye et al., 1999). This releases the NICD, which translocates to the nucleus to regulate transcription in association with Suppressor of Hairless protein [Su(H)] (Bray, 2016).

Staining with antibodies that detect the Notch intracellular and extracellular domains reveal that full-length Notch accumulates on the apical side of the follicle cells during early oogenesis and is then cleared from the membrane at stage 6, when signalling occurs (Fig. 2I,K). However, both antibodies detect high levels of Notch at stage 7-8 in imp mutant cells (Fig. 2J-J″,L-L″). The Notch extracellular domain is removed by the first cleavage and then disappears, perhaps because it is endocytosed with Delta into the signalling cell, as in other tissues (Langridge and Struhl, 2017; Nichols et al., 2007; Parks et al., 2000). The persistence of the extracellular domain signal in the mutant cells therefore indicates that loss of IMP inhibits Notch signalling before or at the first Kuzbanian-dependent cleavage of the Notch extracellular domain (NECD).

To test whether IMP is more generally involved in Notch signalling, we generated mutant clones in the wing imaginal disc. Notch activity is required for Cut expression in two rows of cells along the dorsal-ventral midline of the wing disc (Fig. 3A) (Micchelli et al., 1997). Large imp mutant clones that include the dorsal-ventral boundary have no effect on Cut expression (Fig. 3B,B′). Furthermore, the Notch signalling reporter, NRE-GFP (Housden et al., 2012) (Fig. 3C), is expressed at the same level in imp mutant and wild-type cells, indicating that the loss of IMP does not impair or delay Notch signalling in this tissue (Fig. 3D,D′). Consistent with this, adult wings containing imp clones have a normal bristle pattern and never show any of the wing-notching characteristic of Notch mutants, although they have some wing venation defects. imp null mutant clones in the eye imaginal disc also showed no phenotype, with the cone cells being specified normally as shown by Cut expression (Tomlinson et al., 2011) (Fig. 3E-E″). Thus, IMP seems to be specifically required for Notch activation in the ovary and is not a general component of the Notch signalling pathway.

In addition to binding to Notch in trans to activate its cleavage and signalling, Delta expressed in the same cell can bind to Notch in cis to inhibit signalling (Cordle et al., 2008; de Celis and Bray, 1997; Micchelli et al., 1997; Miller et al., 2009; Sakamoto et al., 2002). Indeed cis-inhibition by Delta expressed in the follicle cells controls their competence to respond to Delta from the germ line, as Delta mutant follicle cell clones switch from mitosis to the endocycle too early and undergo precocious differentiation (Poulton et al., 2011). This inhibition is regulated by the microRNA pathway, which represses Delta expression in the follicle cells to relieve cis-inhibition at stage 6. Mutants in the conserved components of the microRNA pathway, belle and Dicer-1, therefore cause a delay in follicle cell development similar to that observed in imp mutants (Poulton et al., 2011). As IMP is an RNA-binding protein that modulates the miRNA pathway in other contexts (Toledano et al., 2012), this raises the possibility that it is required for the miRNA-dependent repression of Delta in the follicle cells.

We compared the phenotypes of belle and Dicer-1 mutants with that of imp to confirm that they cause a similar inhibition of Notch activation. Like imp mutants, belle null mutant follicle cells go through one extra division and border cell migration is disrupted (Fig. 4A-C″). Furthermore, belle and Dicer-1 mutant cells accumulate uncleaved Notch at their apical surfaces, as shown by the persistence of staining with antibodies against the NECD and NICD (Fig. 4D-E″; Fig. S3A-B″). If IMP functions in the same microRNA pathway as Belle and Dicer-1, cells carrying null mutations in two of these genes should show a delay in follicle cell differentiation identical to that observed in imp, belle and Dicer-1 single-null mutants, as the removal of a second essential component of the pathway should have no further effect. On the other hand, one would expect an additive effect if IMP functions in a parallel pathway. We therefore generated imp null mutant clones marked by the loss of RFP and belle or Dicer-1 null mutant clones marked by the loss of GFP in the same egg chambers. This allowed us to compare directly the phenotypes of cells mutant for either gene with the double mutant cells that express neither GFP nor RFP. Although the single-mutant clones produced the expected phenotypes, the double-mutant cells showed a more severe delay in Notch activation, as illustrated by their continued expression of Cut after the single mutant cells had lost expression (Fig. 4F-F‴; Fig. S3C-C‴). This suggests that IMP functions in a different pathway from Belle and Dicer-1 to control the timing of Notch activation in the follicle cells.

Although IMP does not appear to function in the microRNA pathway, it could still control Notch activation by repressing Delta expression in the follicle cells to relieve cis-inhibition. If this is the case, Delta should be epistatic to imp, with double-mutant clones showing the Delta phenotype of early differentiation and exit from the cell cycle. We first confirmed that Delta clones cause premature differentiation of the follicle cells and observed that mutant cells have larger nuclei, indicating premature entry into endocycle (Fig. 5A,A′). Furthermore, Delta mutant cells turn off Cut expression before stage 6, unlike wild-type cells, indicating that the cells lacking Delta differentiate precociously (Fig. 5B,B′).

To examine the epistatic relationship between Delta and imp, we generated clones of null mutants in each gene in the same egg chambers, marked by the loss of GFP and RFP, respectively. This revealed that Delta imp double-mutant cells show the same delay in Notch signalling as imp single-mutant cells (Fig. 5C-E‴). Double-mutant clones express Cut protein at stage 6/7, as do neighbouring wild-type cells, whereas Delta single-mutant clones have already turned Cut off at this stage (Fig. 5D-D‴). Moreover, Hnt expression is not switched on prematurely in Delta imp clones, as it is in Delta clones (Fig. 5C-C‴). More importantly, like single imp mutant clones, the Delta imp double-mutant clones express Cut longer than wild-type cells (Fig. 5E-E‴). To confirm this result, we also examined whether IMP regulates Delta mRNA stability or translation, using a sensor line that contains the 3′UTR of Delta downstream of the GFP coding sequence (Poulton et al., 2011). However, imp clones expressed the same level of GFP from the Delta sensor as wild-type cells (Fig. S4). The observation that the imp mutants are epistatic to Delta mutants in the follicle cells indicates that IMP regulates the timing of Notch activation independently of and in parallel with Delta cis-inhibition.

Because IMP is only required in the follicle cells, it cannot affect the expression of Delta in the germ line and must therefore affect either the ability of Notch to bind to Delta in trans or the first Delta-dependent cleavage of Notch at the S2 site, which is mediated by the ADAM family metalloprotease Kuzbanian (Lieber et al., 2002; Pan and Rubin, 1997). To confirm that Kuzbanian is required for Notch signalling in the follicle cells, we generated clones of an amorphic allele, kuz^e29-4 (Rooke et al., 1996). kuz mutant follicle cells show an identical phenotype to null mutations in Notch: mutant cells continue dividing after stage 6, leading to an increase in cell number and a decrease in cell size, and fail to differentiate, as they continue to express Cut and never turn on Hnt (Fig. 6A-B′). Staining for the extracellular and intracellular domains of Notch showed that kuz mutant cells maintain high levels of NICD and NECD at the follicle cell apical membrane, consistent with its role in the first cleavage of Notch, a phenotype that resembles that of imp (Fig. 6C-D′).

As kuz and imp mutants impair the same step in Notch activation, we investigated whether the imp phenotype results from a deficit in Kuzbanian activity by overexpressing Kuzbanian in imp mutant cells using the MARCM system (Lee and Luo, 2001). Control imp MARCM clones continue to express Cut and not Hnt at stage 7, as expected (Fig. 6E,E′,G,G′). By contrast, expression of Kuz eliminates the delay in Cut repression and restores timely Hnt expression (Fig. 6F,F′,H,H′). Furthermore, although Notch remains at high levels in the apical plasma membrane of imp mutant cells at stage 7, it disappears on schedule in mutant cells overexpressing Kuzbanian, as it does in wild type (Fig. 6I-J′). Thus, increasing the levels of Kuzbanian rescues the Notch signalling defect in imp mutant follicle cells, suggesting that IMP is required for normal Kuzbanian activity. As a control, we also overexpressed Fringe, which modifies Notch to make it more responsive to Delta (Brückner et al., 2000; Goto et al., 2001; Moloney et al., 2000). However, Fringe overexpression has no effect on the imp mutant phenotype (Fig. S6). The observation that the delay in Notch signalling in imp mutants is not rescued by the overexpression of another component that acts upstream of Notch cleavage reveals that rescue is specific to Kuzbanian overexpression.

Because no anti-Kuz antibodies are available, we took advantage of a GFP-tagged kuzbanian BAC transgene that rescues the lethality of kuz mutants (Dornier et al., 2012). Kuz-GFP is expressed at very low levels in the follicle cells, with the highest expression at stages 5-6. Live imaging of egg chambers with two copies of this transgene revealed that Kuz-GFP is enriched at the apical side of the follicle cells and in intracellular speckles (Fig. 7A-A″). This weak apical enrichment is reduced or lost in the majority of imp mutant cells (63% of 38 clones; Fig. 7B,B′). Instead, most mutant cells show a marked increase in the size and brightness of the speckles, with 56% of clones containing large intracellular foci (Fig. 7C-D′). Unlike the signal at the apical membrane, the intracellular foci of Kuzbanian remain intact after fixation, which allowed us to stain for markers for different vesicular compartments. The Kuzbanian-positive punctae showed the strongest colocalisation with Rab7 (Fig. 7E-F″) (34/35 punctae), a marker for late endosomes, whereas they showed minimal colocalisation with the Golgi marker GM130 (2/20) (Fig. 7G-G″). The formation of these endosomal foci of Kuz-GFP is not a consequence of impaired Notch signalling, because they did not form in bel mutant clones (Fig. S5A-A″), which show a very similar phenotype to imp mutants, or in follicle cells adjacent to Delta^revF10 germline clones, in which Notch signalling does not occur (Fig. S5B-C′). These results suggest that loss of IMP disrupts the intracellular trafficking of Kuzabanian, leading to a reduction in its levels at the apical plasma membrane, where Notch cleavage occurs.

Although Delta/Notch signalling uses a very simple signal transduction pathway, its activity can be regulated at several levels to control the timing and direction of signalling in a context-dependent manner. Factors controlling the endocytosis of Notch, such as Numb, can determine the direction of signalling, whereas the ubiquitin ligases that induce Delta endocytosis, Neuralized and Mind bomb, determine where and when signalling occurs (Deblandre et al., 2001; Itoh et al., 2003; Lai et al., 2001, 2005; Le Borgne et al., 2005; Pavlopoulos et al., 2001). Whether cells respond to a particular DSL ligand can also be regulated by the glycosylation of the NECD by the Fringe family of proteins, making Drosophila Notch more responsive to Delta and less responsive to Serrate (Brückner et al., 2000; Goto et al., 2001; Moloney et al., 2000). Indeed, the modification of Notch by Fringe in the polar/stalk follicle cell precursors renders these cells more responsive to Delta than the other follicle cells, thereby restricting polar cell fate to the ends of the early egg chamber (Grammont and Irvine, 2001). Finally, signalling can be modified by cis-inhibition by DSL ligands, which impede Notch interactions with activating ligands in trans. This mechanism plays a role in controlling the timing of follicle cell differentiation until Delta is downregulated in the follicle cells by the microRNA pathway (Poulton et al., 2011). Here, we present evidence for a new mechanism that regulates the activity of the Notch pathway through the localisation of the ADAM metalloprotease Kuzbanian.

Our results show that mutants in the RNA-binding protein IMP have impaired Delta/Notch signalling in mid-oogenesis, which results in one extra round of follicle cell division. This leads to a delay in follicle cell differentiation that is likely to be the cause of some of the later imp mutant phenotypes, which include impaired border cell delamination, a delay in the formation of the columnar epithelium around the oocyte and delayed migration of the centripetal follicle cells. Continuing Delta-Notch signalling has also been found to be required for border cell migration after the cells have been specified, and this correlates with higher levels of Kuzbanian expression (Wang et al., 2007). Thus IMP may also play a role in the later Notch signalling that is needed for normal border cell migration.

Mutants in the microRNA pathway give a very similar phenotype to imp because of a failure to repress Delta translation, but our results show that IMP functions in parallel to this pathway. Firstly, double mutants between imp and belle or Dicer-1 show an additive delay in follicle cell differentiation. Secondly, imp mutants are epistatic to Delta mutants, indicating that IMP does not function by relieving cis-inhibition. Instead, we find that the delay in Notch signalling can be efficiently rescued by overexpression of Kuzbanian. Furthermore, loss of IMP disrupts the enrichment of Kuzbanian at the apical plasma membrane, which is where Notch must be cleaved for signalling to occur. Thus, IMP is required in some way for the localisation of Kuzbanian to the site where germline Delta binds to Notch in trans to trigger the first cleavage, identifying a new regulatory step in the Notch pathway.

The loss of IMP does not disrupt other Delta/Notch signalling processes, such as the formation of the dorsal ventral boundary in the wing, indicating that it is not a general component of the pathway. This raises the question of why IMP is specifically required in the follicle cells. One possibility is that this relates to the different geometry of the Delta/Notch interaction in the follicle cells compared with other signalling events. Most examples of Delta/Notch signalling occur between adjacent cells in epithelial tissues, where Delta and Notch can only interact at the lateral membrane, usually at the level of the adherens junctions (Andersson et al., 2011; Bray, 2016; de Celis et al., 1996; Kidd et al., 1989; Micchelli and Blair, 1999; Woods and Bryant, 1993). By contrast, the germ cells of the egg chamber, which produce the activating Delta signal, contact the apical side of the follicle cells, and both Notch and Kuzbanian therefore need to be localised apically for signalling to occur. Notch localisation is not affected in imp mutants, however, as it accumulates at high levels at the apical side of mutant cells. Thus, IMP appears to disrupt Kuzbanian localisation specifically rather affecting apical trafficking more generally.

As IMP is a cytoplasmic RNA-binding protein, it presumably acts by regulating the stability, translation or localisation of specific mRNAs. One possibility is that IMP acts on kuz mRNA directly. Indeed, we have observed by qPCR that kuz RNA is enriched in immunoprecipitations of IMP (data not shown), although we cannot detect the mRNA by fluorescent in situ hybridisation, presumably because it is present at very low levels. It seems unlikely that IMP regulates the translation of kuz mRNA, as Kuzbanian protein levels do not appear to change in imp mutants. There is also no evidence for a role of IMP in the localisation of kuz mRNA, as IMP itself is not localised and Kuzbanian is a secreted transmembrane protein that must be translated at the endoplasmic reticulum and trafficked to the cell surface through the Golgi complex. One possibility is that IMP controls Kuzbanian localisation through 3′UTR-dependent protein localisation in a similar way to that in which human HuR (ELAVL1) binds to the long 3′UTR of CD47 mRNA to recruit SET protein, which then facilitates CD47 protein trafficking to the cell surface (Berkovits and Mayr, 2015). In this scenario, IMP binding to the 3′UTR of kuz mRNA would recruit a co-factor, which then associates with the cytoplasmic domain of Kuzbanian protein to direct its trafficking to the apical plasma membrane.

It seems more likely that IMP enhances the translation or stability of another mRNA that encodes a factor that either directs the apical trafficking of Kuzbanian or anchors it in the apical plasma membrane. Possible candidates include the TspC8 Tetraspanins, Tsp86D and Tsp3A, which have been shown to enhance the trafficking of Kuzbanian to the cell surface in S2 cells and in the migrating border cells (Dornier et al., 2012). However, the accumulation of Kuzbanian protein in Rab7-positive late endosomes in imp mutant cells suggests that Kuzbanian is being endocytosed from the apical membrane in the absence of IMP, arguing that the phenotype results from the loss of a factor that stabilises Kuzbanian at the membrane and prevents its endocytosis, rather than a factor that facilitates its delivery there. Various cross-linking approaches, such as RIPseq, i-CLIP, PAR-iCLIP and eCLIP have shown that IMPs bind to hundreds to thousands of mRNAs, predominantly in their 3′UTRs (Conway et al., 2016; Hansen et al., 2015). IMP-binding sites are particularly enriched in mRNAs encoding cytoskeletal components and trafficking factors, many of which are plausible candidates for the relevant target of IMP in regulating Kuzbanian localisation and Notch signalling in the follicle cells.

We used the following mutant alleles and transgenic constructs: imp⁷ and imp⁸ (Munro et al., 2006), GFP-IMP trap line 126.1 (a gift from Alain Debec, University Paris Diderot, Paris, France; Morin et al., 2001), UAS-NICD (Go et al., 1998; a gift from S. Bray, University of Cambridge, Cambridge, UK), bel⁴⁷¹¹⁰ and Dl-3′UTR sensor (Poulton et al., 2011), Delta^revF10 and Delta^M1 (Sun and Deng, 2005; a gift from S. Bray), Dicer-1^Q1147X (Lee et al., 2004; a gift from A. Brand, The Gurdon Institute, Cambridge, UK), Kuz-GFP and kuz^e29-4 (Dornier et al., 2012; Rooke et al., 1996), UAS-kuz (Sotillos et al., 1997; Kyoto Stock Center 108440), E(spl)M7-lac (a gift from Sarah Bray; Assa-Kunik et al., 2007) and NRE-GFP (a gift from Sarah Bray; Housden et al., 2012). The following stocks were used to generate mitotic clones: ubiRFP-nls, hsflp, FRT19A (BDSC 31418), FRT40A ubiRFP-nls (BDSC 34500), FRT82B, ubiRFP-nls (BDSC 30555) and FRT82B ubiGFP (BDSC 5188). MARCM, following the method of Lee and Luo (2001), was carried out using UAS-GFP-mCD8 (Lee and Luo, 1999) as the marker.

The following antibodies were used: mouse anti-Cut [Blochlinger et al., 1990; Developmental Studies Hybridoma Bank (DSHB), 2B10], mouse anti-Hnt (Yip et al., 1997; DSHB, 1G9), mouse anti-NICD (Fehon et al., 1990; DSHB, C17.9C6), mouse anti-NECD (Diederich et al., 1994; DSHB, C458.2H). All primary antibodies from DSHB were used at a dilution of 1:100. Anti-GM130 was purchased from Abcam (Sinka et al., 2008; ab30637) and used at 1:500. Anti-Rab7 (rabbit) was kindly provided by the Nakamura lab and used at 1:1000 (Tanaka and Nakamura, 2008). AlexaFluor 488- and AlexaFluor 647-conjugated secondary antibodies were purchased from Jackson ImmunoResearch and used at a dilution of 1:1000 (115-545-003, 711-545-152, 115-607-020 and 111-605-003). F-actin was stained with Alexa Fluor 568 or Alexa Fluor 647 phalloidin (Invitrogen) at 1:1000. Ovaries were mounted in Vectashield with DAPI (Vector Laboratories). The cell membranes were labelled with CellMask Orange Plasma Membrane Stain or CellMask Deep Red Plasma Membrane Stain (Thermo Fisher Scientific).

Ovaries from adult flies or imaginal wing discs from third instar larvae were dissected in PBS and fixed for 20 min in 4% paraformaldehyde and 0.2% Tween 20 in PBS. The tissues were then incubated in 10% bovine serum albumin (BSA) in PBS to block for 1 h at room temperature. The incubation with primary antibody was performed at 4°C overnight in PBS, 0.2% Tween 20 and 1% BSA.

Immunostaining on pupal eye discs was performed as described by Richard et al. (2006).

Fixed preparations were imaged using an Olympus IX81 (40×/1.3UPlan FLN Oil or 60×/1.35 UPlanSApo Oil). Live imaging was performed using a Leica SP8 (63×/1.4 HCX PL Apo CS Oil) or Olympus IX81 (40×/1.3 UPlan FLNOil or 60×/1.35 UPlanSApo Oil) inverted confocal microscope. For live observations, ovaries were dissected and imaged in 10S Voltalef oil (VWR Chemicals).

Follicle cell clones of imp, Dl, bel, Dcr-1 and kuz were induced by incubating larvae or pupae at 37°C for 2 h every 12 h over a period of at least 3 days. Adult females were dissected at least 2 days after the last heat shock. Wing imaginal disc clones and eye imaginal disc clones of imp were induced by heat shocking first and second instar larvae for 30 min per day over a period of 2 days. Larvae were dissected at least 1 day after the last heat shock.

We thank Sarah Bray, Francois Schweisguth, Wu-Min Deng, Alain Debec and Andrea Brand and their labs for fly stocks and Akira Nakamura for antibodies; the Developmental Studies Hybridoma Bank, Kyoto Stock Center (DGRC) and Bloomington Drosophila Stock Center (BDSC) for antibodies and fly stocks; and members of the D.St J. lab for technical assistance, helpful comments and criticism.