Asymmetric localisation of cytokine mRNA is essential for JAK/STAT activation during cell invasiveness

In many cell types, the switch from stationary to migratory behaviour is associated with changes in gene expression (Thiery et al., 2009; Yang and Weinberg, 2008). During this transition, migrating cells have to detach from the epithelium, develop novel interactions with the extracellular environment, become competent to extend protrusions and decipher new guidance cues (Yilmaz and Christofori, 2009). Studying migration under physiological conditions is crucial to understand the molecular mechanisms that control how subsets of cells become motile during development or tumour metastasis.

Border cell (BC) migration in Drosophila ovaries provides a powerful model with which to study regulated invasive cell migration in vivo. The egg chamber is the functional unit of the fly ovary and is composed of an oocyte, 15 nurse cells and surrounding follicle cells assembled into an epithelial monolayer. BCs are organised into a cluster of about eight epithelial cells [two polar cells (PCs) and six outer border cells (oBCs)], which delaminate and migrate in between the nurse cells to reach the oocyte and contribute to the formation of the micropyle, a structure required for fertilisation (Fig. 1A). Delamination and migration are initiated at stage 9 and occur over a period of 6 hours. This process has been well characterised both in fixed and living egg chambers and has proven to be robust and stereotyped (Bianco et al., 2007; Jang et al., 2007; Prasad et al., 2007; Prasad and Montell, 2007; Rorth, 2007).

The well-conserved JAK/STAT (Janus kinase/Signal transduction and activator of transcription) signalling pathway is known to promote cell proliferation, survival and cell fate differentiation in different organisms. In Drosophila, JAK/STAT signalling is essential for embryonic segmentation, eye and wing development, immune response, stem cell development and BC specification and migration (Hombria and Brown, 2002; Hou et al., 2002). oBCs are recruited by a pair of specialised cells, the PCs, which secrete the cytokine Unpaired (Upd; Outstretched – FlyBase), the main ligand of the pathway. Upd binding to its receptor Domeless (Dome) (Chen et al., 2002; Harrison et al., 1998; Ghiglione et al., 2002) in neighbouring cells activates the JAK/STAT cascade. Activated Hopscotch (Hop), the JAK orthologue, phosphorylates the transcription factor Stat92E, which then enters the nucleus and induces the expression of target genes including slow border cells (slbo), which encodes a transcription factor related to mammalian C/EBPs, the expression of which is essential to promote BC invasiveness (Montell et al., 1992). Indirect evidence suggests that Upd activates adjacent cells in a graded manner. Indeed, receptor endocytosis shows a gradient of internalisation, with its maximal level close to the PCs (Devergne et al., 2007; Ghiglione et al., 2002; Xi et al., 2003).

It is essential that the correct number of oBCs be specified because an excess or a deficit of recruited cells prevents migration and reduces fertility (Silver et al., 2005). Hence, loss of function of any component of the JAK/STAT pathway impairs oBC recruitment, whereas overactivation of the pathway is sufficient to induce ectopic follicle cells to become invasive (Silver and Montell, 2001). These findings suggest that JAK/STAT activity needs to be tightly controlled and that a specific level of signalling must be reached close to the ligand source, allowing the selection of an appropriate number of oBCs. One control mechanism involves the DNA-binding protein Apontic (Apt), which spatially regulates JAK/STAT signalling in the egg chamber follicular epithelium and prevents the recruitment of cells distant from the ligand source (Starz-Gaiano et al., 2008). In addition, it was shown that JAK/STAT activity has to be maintained during the entire process of migration to prevent splitting of the cluster (Silver et al., 2005). However, no mechanism involved in the positive regulation of JAK/STAT activation has been demonstrated.

Many signalling pathways, including JAK/STAT, have been shown to be partially polarised (Hombria and Sotillos, 2008), and polarisation or subcellular localisation of pathway components emerge as key mechanisms to control signalling specificity and/or efficiency. In the ectoderm, JAK/STAT signalling is modulated by the cell polarity protein Par3 (Bazooka – FlyBase) (Sotillos et al., 2008). Furthermore, in egg chambers, both Dome and Dome-containing endocytic vesicles localise apicolaterally, suggesting that the ligand has to be presented to the apical membrane for correct pathway activation (Devergne et al., 2007; Ghiglione et al., 2002). Finally, upd mRNA is apically enriched in the PC, suggesting a polarised synthesis and release of the protein (Ghiglione et al., 2002). mRNA localisation is a well-characterised and conserved mechanism that plays an important role in targeting proteins to their site of function in different organisms (Kardon and Vale, 2009; Martin and Ephrussi, 2009; Tekotte and Davis, 2002; Wade, 2009). Recent studies revealed that the substantial majority of Drosophila mRNAs have specific subcellular localisations (Lecuyer et al., 2007), indicating that mRNA targeting is more widely used than first anticipated. Nonetheless, it is not yet clear whether mRNA localisation can be involved in signalling regulation.

In this study, we identify a novel positive-regulatory mechanism that controls JAK/STAT signalling through intracellular trafficking and apical localisation of the upd cytokine mRNA. We show that PCs possess a dense and polarised microtubule network, the integrity of which is required for BC specification and migration. upd mRNA appears to be apically concentrated in the PC in a Dynein- and Bicaudal D (BicD)/Egalitarian (Egl)-dependent manner. We show that in the absence of upd transcript localisation, secretion of the Upd protein is affected and JAK/STAT signalling is dramatically reduced, leading to recruitment and migration defects. We identified the upd localisation element in the open reading frame (ORF) region, suggesting an important selection constraint on this motif. We propose that upd transcript localisation is essential to optimise Upd apical secretion as well as JAK/STAT activation in epithelial cells that require the highest levels of ligand. This work reveals a novel post-transcriptional regulatory mechanism controlling cell invasiveness.

D. melanogaster stocks were raised on standard corn meal agar medium at 25°C. w¹¹¹⁸ or nlsGFP strains were used to inject fluorescently labelled mRNA. The following strains were used: UAS-SpastinGFP (Jankovics and Brunner, 2006), Slbo-gal4, Upd-gal4 (gift from D. Montell), 10×STAT92E-GFP (Bach et al., 2007), Nod-lacZ (Clark et al., 1997), UAS-Upd (M. Zeidler and N. Perrimon, Sheffield, UK and Boston, MA) and Sas-4 null mutants (gift from J. Raff, Oxford, UK). UAS-BicD RNAi, UAS-Egl RNAi, UAS-Glued RNAi and UAS-Lis-1 RNAi strains are from the Vienna Drosophila RNAi Center. Dynein mutant clones were generated by crossing the Dhc4-19, FRT80/TM6b females with e22cgal4, UAS-FLP/CyO; FRT80, UbiGFP males. e22cgal4, UAS-FLP/+; Dhc4-19, FRT80/UbiGFP, FRT80 females were kept for 48 hours at 30°C before dissection (Li et al., 2008).

Ovaries were dissected in 1×PBS and fixed in 3.7% formaldehyde. The samples were blocked in 1×PBS containing 0.3% Triton X-100 and 2% BSA prior to incubation with antibodies and then mounted in Vectashield (Vector Laboratories). For extracellular staining, ovaries were dissected on ice and incubated with the primary antibody in Schneider medium containing 10% foetal calf serum (FCS) for 1 hour 30 minutes at 4°C. Ovaries were washed three times in 1×PBS at 4°C and fixed for 20 minutes in 3.7% formaldehyde. The following primary antibodies were used: mouse anti-α-tubulin-FITC (1/250; Sigma); mouse anti-α-tubulin clone DM1A (1/500; Sigma); rabbit anti-Slbo (1/2000); mouse anti-Fas3 7G10 [1/100; Developmental Studies Hybridoma Bank (DSHB)]; rabbit anti-Upd [1/1000 for conventional staining, 1/300 for extracellular staining; gift from D. Harrison (Ghiglione et al., 2008)]; sheep anti-digoxigenin-POD (1/1000; Roche); and rabbit anti-PLP (1/500; gift from J. Raff). Cy3-, Alexa Fluor 546-, Alexa Fluor 488- and Cy5-conjugated secondary antibodies (Molecular Probes, Jackson ImmunoResearch) were used. Hoechst was used to visualise DNA. In situ hybridisation signal was revealed using the Fluorescent Tyramide Detection Kit (TSA, NEN LifeSciences). Digoxigenin-labelled antisense probes were prepared using full-length upd cDNA (gift from D. Harrison, Lexington, KY).

Microtubule alteration was induced by Colchicine treatment (65 μg/ml; Sigma). Colchicine was mixed with dry yeast and added to the corn meal agar medium at room temperature. Flies were fed for 48 hours on this medium before dissection.

RNA was transcribed in vitro using T3, T7 or SP6 RNA polymerase and UTP-Alexa 546 (Molecular Probes) (Wilkie and Davis, 2001). Plasmids containing gfp cDNA were used as a negative control. Plasmids containing full-length or truncated fragments of upd cDNA were used. Constructs were generated by PCR amplification and cloned into the pGEM-T Easy vector (Promega). The plasmids were sequenced to confirm the final products. Embryo injections were performed as described (Wilkie and Davis, 2001) at 250-500 ng/μl. P1H4 antibody was pre-injected 10 minutes before RNA injection.

Imaging was performed on a Zeiss LSM 510 confocal microscope or on a wide-field DeltaVision microscope (Applied Precision, DV 3.0 system, IX70 Olympus microscope and CH350 camera) and deconvolved using SoftWorx software. Particle tracking was performed manually using ImageJ and SoftWorx Explorer. The speed of particles was calculated in micrometers per second according to the distance travelled in each time interval. ImageJ was used to quantify the GFP level relative to Hoechst staining. Experiments were performed at least twice in parallel, with the control processed under the same conditions and on the same day. Image acquisitions were performed using the same parameters. Quantifications were made with ImageJ. Statistical analysis of the number of oBCs was performed in Excel using a paired t-test to compare wild-type and mutant conditions. Statistical significance of migration defects was evaluated using a χ² test of the mutant conditions versus control (5% critical value, one degree of freedom).

In order to characterise microtubule (MT) organisation in BC clusters, we examined their overall distribution and density as well as their polarity using anti-α-tubulin antibodies and a Nod-lacZ construct, respectively. The Nod-lacZ flies carry a MT motor fusion leading to β-gal accumulation at the minus end of MTs (Clark et al., 1997). At stage 8 (i.e. before BC cluster delamination), the α-tubulin staining was apically enriched and revealed a dense network of MTs in the PCs and oBCs, confirming previous observations (Dahlgaard et al., 2007; Doerflinger et al., 2003) (Fig. 1B). Double β-gal and α-tubulin stainings revealed a colocalisation between apical α-tubulin and the minus ends of MTs that was particularly visible in the PCs (Fig. 1B). At stage 9, when migration starts, PCs rotate 90° from their initial anteroposterior orientation and are found roughly orthogonal to the direction of migration (Pinheiro and Montell, 2004). Interestingly, their MT network maintained the same strongly polarised architecture (Fig. 1C). By contrast, the oBC MT cytoskeleton showed no clear orientation as revealed by the dispersed β-gal staining (Fig. 1C). When the cluster reached the oocyte at stage 10, both the PCs and oBCs showed an accumulation of MT minus ends toward the posterior (Fig. 1D). Hence, in comparison to the oBCs, which seem to have a very dynamic cytoskeleton, the PCs exhibit a very dense and stable MT network during the entire process of migration. In order to analyse centrosome positioning, we stained egg chambers using anti-PLP (Pericentrin-like protein; Cp309 – FlyBase) antibody (Martinez-Campos et al., 2004). Pericentrin was found at the apical side, confirming that MTs emerge from the apical surface of follicle cells (Fig. 1E). However, it appears that centrosomes are not essential for BC migration because acentriolar Sas-4 mutant females (Stevens et al., 2007) showed normal MT organisation and BC migration (data not shown).

To test whether MTs are required for oBC specification and/or migration, we used Colchicine to prevent MT polymerisation. Wild-type females were fed for 2 days on standard medium supplemented with Colchicine and their ovaries analysed. We observed that MTs were strongly affected and that BC clusters arrested before they reached the oocyte in 70% of the chambers (compare Fig. 1F,G with 1H). At early stage 9, the number of oBCs expressing Slbo, a marker of oBCs in delaminating clusters, was identical to that of the control (6), but dropped to 3.3 at stage 10 (Fig. 1K). Thus, after Colchicine treatment, the Slbo marker is not efficiently maintained (see below). As Colchicine feeding alters MTs in the whole egg chamber, we expressed Spastin, a MT-severing protein, to disrupt MTs specifically in the BCs. Spastin expression has been shown to cause very efficient MT disassembly in larvae and embryos (Jankovics and Brunner, 2006; Sherwood et al., 2004). Expression of Spastin in the oBCs (using a Slbo-gal4 driver) did not cause any phenotype (Fig. 1K,L). By contrast, MT disruption in the PCs (using a Upd-gal4 driver) specifically induced a decrease in oBC recruitment (4.5) as well as migration defects, with ~30% of the clusters not reaching their final destination at stage 10 as compared with 5% in the control (Fig. 1J-L). These results demonstrate that a polarised MT network is required in the PCs for normal BC specification and migration.

The dynein motor complex is responsible for cargo transport towards the minus ends of MTs in all eukaryotic cells. To investigate whether intracellular transport along MTs is involved in BC specification and migration, we analysed the effects of cytoplasmic Dynein motor loss-of-function in egg chambers. Dynein heavy chain 64C (Dhc64C) is a component of the dynein complex, the disruption of which prevents all forms of cargo transport and is embryonic lethal (Vale, 2003). To bypass lethality, we performed clonal analysis using two Dhc64C lethal alleles, Dhc^4-19 and Dhc⁹⁰² (Li et al., 2008), which showed similar phenotypes. Interestingly, clusters of Dhc64C mutant BCs did not migrate and were found at the anterior tip of the egg chambers (Fig. 2A,B). The number of cells per cluster was determined by observing the Slbo and Fas3 markers, which are expressed in the oBCs and the PCs, respectively. Dhc64C mutant clusters comprised only 3.5 oBCs on average as compared with six in the wild type (Fig. 2E), indicating that Dynein is involved in BC specification and migration. To determine whether Dynein function is required in the PCs and/or the oBCs, we analysed mosaic BC clusters consisting of wild-type and mutant cells. Interestingly, clusters in which only oBCs were mutant for Dhc64C showed no migration defects (Fig. 2C,D,G) and had the normal number of oBCs. Furthermore, Dhc64C mutant oBCs were randomly distributed at the front or at the rear, indicating that oBC migration and shuffling behaviour does not require Dynein (Fig. 2F) (Bianco et al., 2007; Prasad et al., 2007). We conclude that Dynein function is required in the PCs, but not in the oBCs, for normal specification and migration of the cluster.

Dynein associates with adaptor proteins, such as the large multi-protein dynactin complex, which facilitates Dynein processivity and interaction with its cargoes, and Lissencephaly-1 (Lis-1), which modulates Dynein association with the MTs (Kardon and Vale, 2009). We tested the role of the dynactin component Glued (also known as p150) and Lis-1 using double-stranded RNA-mediated knockdown specifically in the PCs or oBCs. Interestingly, when Glued or Lis-1 was depleted in the PCs (Upd-gal4 driver), but not when depleted in the oBCs (Slbo-gal4 driver), the number of oBCs was reduced by half and migration defects were observed as for the Dhc64C mutant clones or after Upd RNAi expression (Fig. 2E,G). BicD and Egl proteins are required for the Dynein-dependent localisation of many mRNAs in Drosophila. Egl is an mRNA-binding protein that associates with the BicD Dynein co-factor, thus bridging mRNAs to motors (Claussen and Suter, 2005; Dienstbier et al., 2009; Nashchekin and St Johnston, 2009). To test their role in BC migration, BicD and Egl were specifically depleted in the PCs or oBCs by RNAi. Absence of BicD or Egl in the PCs induced a reduction in oBC recruitment as well as migration defects (Fig. 2E,G). By contrast, depletion of BicD or Egl in the oBCs alone showed no defect as compared with the wild type (Fig. 2E,G). Altogether, these data indicate that the dynein complex as well as the adaptor proteins BicD and Egl are specifically required in the PCs for cluster differentiation and migration.

Slbo is a known target of the JAK/STAT pathway. Therefore, reduction in the number of Slbo-expressing cells in egg chambers through alteration of the MT network or Dynein function (Fig. 1G,J,K) suggests that JAK/STAT signalling might be affected. To further investigate the role of MT or Dynein in oBC specification through JAK/STAT signalling we used the 10×STAT92EGFP reporter line, which accurately reflects JAK/STAT activity (Bach et al., 2007; Starz-Gaiano et al., 2008). At stage 8, GFP accumulates in the nuclei of follicular cells surrounding the PCs in a graded manner. We quantified the level of nuclear GFP relative to Hoechst staining in anterior follicular cells (fc) as a function of their position, fc1 being the closest to the PC. In the absence of Upd ligand there was a strong reduction in GFP intensity from fc1 to fc6 compared with the control, which correlated with the reduction in Slbo expression in this context (Fig. 2E, Fig. 3A,B,F). Interestingly, when MT integrity or Dynein motor function was altered in the PCs using SpastinGFP or by BicD, Egl or Lis-1 RNAi, we observed a substantial decrease in GFP intensity from fc1 to fc3 (Fig. 3A,C-F). These data show that, in the absence of Dynein transport along MTs, JAK/STAT signalling is dramatically decreased leading to a reduction in oBC recruitment. To confirm the genetic interaction between the JAK/STAT pathway and the transport machinery, we tested whether the Egl and Glued RNAi phenotypes were enhanced in a upd heterozygous mutant condition. Using three different upd alleles, we found that the number of oBCs per cluster was significantly lower in all three conditions than in the control (see Fig. S1 in the supplementary material).

We next looked directly at the distribution of the Upd protein by immunostaining and found that in the wild type, Upd is present in the PCs as well as in two to three adjacent cells in a punctate pattern, most likely representing exocytic and endocytic Upd vesicles, respectively (Fig. 4A-C,I). Strikingly, the staining revealed an apical enrichment of Upd protein in PCs throughout the entire process of migration, suggesting that the protein is preferentially secreted from their apical region (Fig. 4I,J).

This result is consistent with previous indirect evidence suggesting that Upd is secreted from the apical surface of the PCs to act on adjacent cells in a graded manner (Devergne et al., 2007). In order to confirm this hypothesis, a specific immunostaining protocol (Strigini and Cohen, 2000) was performed that allows the exclusive visualisation of secreted Upd protein in the extracellular space. An accumulation of extracellular Upd was detected in the apical region of the PCs (Fig. 4D,I,J; data not shown) from delamination to the end of migration, confirming that Upd is constantly apically secreted to activate the JAK/STAT pathway (Silver et al., 2005). Antibody specificity was validated using Upd RNAi knockdown, a condition in which no antibody signal was detected (Fig. 4H-J). After either Colchicine or Spastin treatment, or in Glued, BicD and Egl loss-of-function conditions, apical secretion of the Upd protein was completely lost and the punctate signal in adjacent cells reduced, suggesting that less Upd was available in the extracellular space for signalling and for endocytosis by receiving cells (Fig. 4E-J; data not shown). Altogether, these results show that in the absence of Dynein transport Upd apical secretion from PCs is much less efficient, leading to a reduction in JAK/STAT signalling and oBC specification.

Interestingly, overexpression of Spastin protein using Upd-gal4 led to an outstretched wing phenotype similar to the upd mutant phenotype (see Fig. S1B,C in the supplementary material). The outstretched phenotype was rescued by co-expression of upd (see Fig. S1E in the supplementary material), whereas misexpression of upd alone had no effect on wing posture (see Fig. S1A in the supplementary material). Similarly, expression of Glued RNAi (Upd>Glued RNAi) produced outstretched wings, which were rescued by co-expressing upd (see Fig. S1D,F in the supplementary material). Hence, an excess of ligand is able to compensate for the lack of transport resulting from a loss of Glued function or from Spastin treatment. These data indicate that intracellular trafficking involving Dynein transport positively regulates JAK/STAT signalling and ligand delivery efficiency in different developmental contexts.

We next investigated the basis for abnormal Upd secretion in the absence of Dynein cargo transport. BicD and Egl are required for Dynein-dependent localisation of many mRNAs in Drosophila (Dienstbier et al., 2009; Nashchekin and St Johnston, 2009), facilitating protein targeting to their site of function (Lecuyer et al., 2007; Martin and Ephrussi, 2009). This prompted us to investigate upd mRNA subcellular distribution by performing fluorescent in situ hybridisation coupled with MT staining. We found that upd mRNA is apically enriched and colocalises with the MTs at all stages of BC migration (Fig. 5A-C). Hence, apical localisation of upd transcript provides a means to concentrate the Upd protein towards the apical side of the PCs, thus promoting efficient activation of the JAK/STAT pathway in adjacent cells. To determine whether MTs and the Dynein motor are required for upd mRNA localisation, their functions were altered through pharmacological or genetic means. After Colchicine treatment, upd mRNA was no longer apically enriched (compare Fig. 6A with Fig. 5B). Moreover, although Spastin expression in the oBC had no effects on upd mRNA localisation (Fig. 6B), its expression in the PCs led to a homogenous distribution of the mRNA in these cells (Fig. 6C). Finally, in loss-of-function conditions for Lis-1, Glued or the BicD-Egl complex, upd mRNA was distributed evenly throughout the cell (Fig. 6D-F), consistent with the loss of apical Upd protein secretion (Fig. 4). These results indicate that the apical enrichment of upd mRNA is MT and Dynein dependent.

To test whether upd mRNA localisation is a unique property of the PCs, we ectopically expressed upd in follicular cells. Interestingly, upd transcripts localised apically in follicle cells, which also have a polarised MT network and a functional dynein complex (Horne-Badovinac and Bilder, 2008; Li et al., 2008) (Fig. 5D). Furthermore, early during embryonic development upd mRNA is expressed broadly throughout the trunk of the embryo and in a dorsal crescent in the head (Harrison et al., 1998) to activate the JAK/STAT pathway and regulate the expression of pair-rule genes. At the same stage, Dynein is known to transport wingless (wg) and pair-rule mRNAs to the apical cytoplasm of blastoderm embryos (Bullock and Ish-Horowicz, 2001; Simmonds et al., 2001; Wilkie and Davis, 2001). We found that upd mRNA is enriched apically above the nuclei of blastoderm embryos, where MT minus ends are concentrated (Fig. 5E-H), indicating that upd mRNA apical localisation is conserved in different tissues.

To confirm that Dynein is required for upd mRNA localisation, fluorescently labelled RNAs were injected into syncytial embryos and assayed for their localisation potential. At this stage, the nuclei form a monolayer beneath the apical cortex from which MTs grow and extend into the basal cytoplasm. It has been reported that any transcript carrying a localisation element injected into syncytial embryos is recognised by the transport machinery and moved apically (Bullock and Ish-Horowicz, 2001; Wilkie and Davis, 2001). Injected RNAs assemble into particles, the movement of which can be followed in living embryos. After injection, the full-length upd RNA rapidly localised at the apical side of the nucleus, confirming that upd RNA is recruited by the localisation machinery and transported along MTs (Fig. 7A,C). gfp RNA did not show any apical localisation 20 minutes after injection (Fig. 7E). The movement of upd-labelled RNA particles was followed in time-lapse experiments, tracking their speed and direction (Fig. 7D,D′). Interestingly, most particles moved towards the apical side with an average speed of 0.5 μm/second, consistent with Dynein active transport (MacDougall et al., 2003). To further test whether this transport was Dynein dependent, embryos were pre-injected with anti-Dhc64C antibody (P1H4), which completely blocked upd RNA apical transport (Fig. 7F). These results indicate that Dynein recruits and transports upd mRNA cargoes to the minus ends of MTs.

mRNAs are thought to be sorted in the cytoplasm according to signals present in their sequence (cis-acting elements or ‘Zip’ code) that are recognised by trans-acting factors associated with molecular motors (Martin and Ephrussi, 2009; Van de Bor and Davis, 2004). These signals do not share any obvious consensus sequence or common features, making them difficult to identify. To locate the upd mRNA localisation element, different fragments of the upd transcript were injected into embryos. No apical localisation of the 3′UTR or 5′UTR fragments was observed (Fig. 7G,H). However, fragments containing the 5′UTR-ORF or the ORF alone did localise apically (Fig. 7I,J), indicating that the upd localisation element is located within the ORF region. We further dissected the ORF into five fragments (ORF1 to ORF5) and found that only the 450 bp ORF4 fragment localised apically, indicating that it carries a upd localisation signal (Fig. 7K,L). Because ORF4 does not localise as efficiently as the complete ORF (50% of injected embryos showed localisation as compared with 88% for the full-length ORF), we conclude that other element(s) of the mRNA might act in cis with the signal in ORF4 for full recruitment and/or transport by the machinery.

The Drosophila JAK/STAT pathway is activated by the cytokine Upd, which is a secreted glycosylated protein that is thought to act as a diffusible ligand in several tissues (Arbouzova and Zeidler, 2006; Harrison et al., 1998; Hombria and Sotillos, 2008). In the Drosophila adult female gonad, activation of JAK/STAT signalling is essential for BC specification and migration (Beccari et al., 2002; Ghiglione et al., 2002; Silver and Montell, 2001). In this process, JAK/STAT is activated by a high concentration of Upd and is maintained active throughout migration (Silver et al., 2005). However, the mechanisms that ensure proper signalling strength, which is essential for BC determination, remain unclear. Here, we report a new mechanism of JAK/STAT positive regulation that relies on polarised accumulation of the ligand mRNA, which in turn allows apical Upd protein secretion from PCs and the strengthening of JAK/STAT activation in oBCs (Fig. 8). The asymmetric accumulation of upd mRNAs is driven by an active, Dynein-dependent transport along polarised MTs. We show that this process is essential for BC specification, as conditions that affect the MT network or associated motor complex mimic a loss of JAK/STAT function. Hence, the polarised secretion of Upd protein from PCs provides a peak of ligand that induces efficient activation of the JAK/STAT pathway in neighbouring cells, which is important for oBC determination and migration. These findings reveal a novel post-transcriptional regulatory mechanism controlling JAK/STAT signalling during a process of epithelial cell invasiveness.

mRNA subcellular localisation prior to translation is a conserved mechanism that restricts protein function both spatially and temporally (Martin and Ephrussi, 2009). It plays an important role in the establishment of cell polarity and in the specification of embryonic axes. Here, we show that mRNA localisation also plays a role in the fate determination of migratory cells within an epithelium. Transcripts that are known to be transported encode a wide variety of proteins, including transcription factors, components of the cytoskeleton and signalling molecules. Some well-characterised examples include Veg1 and An1 mRNAs in Xenopus, the localisation of which determines the vegetal and animal poles of the embryo (Mowry and Melton, 1992), the Ash1 mRNA which localises to the bud site in yeast (Long et al., 1997), and Actin mRNA localisation at the leading edge of migrating fibroblasts and growth cones (Condeelis and Singer, 2005). Transcript localisation can also act as a mechanism to regulate signalling pathways. One example is the Wnt-like protein encoded by the Drosophila wg gene, which plays an important role in segment patterning; apical localisation of wg transcripts in syncytial embryos restricts the diffusion of the Wg protein and therefore contributes to the definition of sharp boundaries (Simmonds et al., 2001). Our data show that, within the follicular epithelium, mRNA transport and localisation regulate the JAK/STAT signalling cascade to provide a peak of ligand activity that is important for follicle cell patterning and oBC specification. Furthermore, our results show that upd mRNA localisation is essential for accumulation of the Upd protein at the apical side of epithelial cells. Hence, as for Wg, Upd is a secreted protein that needs to be restricted to a narrow region, a process that depends on BicD and Egl function (Figs 3 and 6). BicD and Egl form a complex that binds the Dynein motor as well as various minus end-directed mRNA signals (Dienstbier et al., 2009). Interestingly, unlike BicD, Egl is specifically required for mRNA transport and for no other cargoes in Drosophila (Dienstbier et al., 2009), confirming that upd mRNA localisation is a key factor in the control of JAK/STAT pathway activity. It was shown previously that the Drosophila JAK/STAT pathway receptor Dome is apically enriched in the follicular epithelium (Ghiglione et al., 2002) and that Dome undergoes ligand-dependent endocytosis apically (Devergne et al., 2007). In addition, a pool of STAT (Stat92E – FlyBase) protein is apically enriched in a Par3-dependent manner in the Drosophila embryonic ectoderm (Sotillos et al., 2008). These observations suggest that polarisation and/or subcellular localisation represent a key aspect of JAK/STAT signalling efficiency that serves to avoid misrouting and dilution of the ligand. This is likely to represent a general mechanism utilised in other major signalling pathways, as it has been observed that the Notch and EGF receptors, as well as Patched and Frizzled, also localise apically and/or basolaterally (Assa-Kunik et al., 2007; Levitan and Greenwald, 1998; Strutt, 2003; Whitfield et al., 1999).

Proper patterning of the anterior follicular epithelium by the JAK/STAT pathway requires the establishment of a robust gradient, with an appropriate ligand concentration for follicle cells to become properly determined. The transport of upd mRNA and the subsequent accumulation of the Upd protein apically contribute to setting the high concentration of ligand required for oBC determination. Recently, a new negative-feedback mechanism has been identified in oBCs that involves a regulatory circuit between Slbo, STAT and Apt (Starz-Gaiano et al., 2008; Starz-Gaiano et al., 2009). This feedback mechanism inhibits JAK/STAT activity in cells that receive low levels of Upd ligand, i.e. those distant from the source. Therefore, JAK/STAT signalling requires several regulatory mechanisms to select the correct number of migratory BCs. First, in cells sending the signal (PCs), post-transcriptional control of upd mRNA and asymmetric protein accumulation are important for setting the correct ligand levels. It is possible that subcellular localisation of the transcript could also control the translation of the protein spatially, which in turn could affect translation efficiency, post-translational modifications such as glycosylation, co-factor association and/or degradation. Second, in receiving cells mechanisms are required to read the Upd gradient and limit the number of migratory cells, which involves spatial regulation of JAK/STAT signalling in anterior follicle cells receiving the Upd ligand through Apt. Additional mechanisms that involve sequestration of the ligand after its secretion might also exist. Indeed, in tissue culture, Upd is mainly found associated with the extracellular matrix (Harrison et al., 1998), which might help to limit Upd diffusion in vivo. Interestingly, it has been reported that a transient cap of extracellular matrix forms at the apical side of PCs when BCs are being recruited (Medioni and Noselli, 2005), raising the possibility of an association of Upd with the extracellular matrix cap for the purposes of building up the gradient.

Future work will need to integrate all these various aspects of signalling regulation in order to fully understand how invasive cells are determined during development and how they can escape control mechanisms in the context of disease and cancer.

We thank all members of the laboratory for fruitful discussions; J. B. Coutelis, R. Delanoue, N. Founounou, C. Géminard, N. Parassol and R. Rousset for their comments on the manuscript; the PRISM platform for providing state of the art imaging resources; D. Harrison and T. Hays for reagents; the DSHB for providing antibodies; and the Vienna Drosophila RNAi Center for providing Drosophila lines. V.V.D.B. and S.N. were supported by the Centre National de la Recherche Scientifique, the Agence Nationale pour la Recherche and the Association pour la Recherche contre le Cancer (ARC).