boudin is required for septate junction organisation in Drosophila and codes for a diffusible protein of the Ly6 superfamily

Model organisms, such as the fruit fly, are sophisticated tools that have contributed decisively to our understanding of genetic complexity, allowing functional characterisation of new genes and novel insight into many developmental processes. We have profited from the advantages offered by Drosophila to enlarge current knowledge about a poorly characterised family of proteins present in metazoan genomes, the Ly6 superfamily. The Ly6 proteins share an extracellular motif spanning about 100 residues known as a three-finger domain, three-finger snake toxin motif or Ly6/uPAR domain. This structure, first identified in the sea-snake erabutoxin b(Low et al., 1976), features a simple inner core stabilised by disulphide bridges, which supports three protruding loops or fingers. Besides a diagnostic set of 8 or 10 cysteines found in stereotyped positions, Ly6 primary sequences are poorly conserved,but they adopt remarkably similar three-dimensional structures(Kini, 2002; Ploug and Ellis, 1994). The Ly6 module is a structural domain involved in protein-protein interactions,tolerating an unusual degree of variation and binding with high specificity to a broad spectrum of targets.

The human genome codes for 45 members of the Ly6 superfamily(Galat, 2008). These include 12 TGFβ receptors, the ectodomains of which adopt the three-finger fold,but also many glycosylphosphatidylinositol (GPI)-anchored proteins and soluble ligands. Only a few of these proteins have been studied in detail, such as the urokinase plasminogen activator receptor (uPAR; PLAUR - Human Gene Nomenclature Database), which plays important roles in cell adhesion,proliferation and migration (Blasi and Carmeliet, 2002), and CD59, an inhibitor of complement activity(Davies et al., 1989). Other members, such as Lynx1 (Miwa et al.,2006) or the soluble SLURP proteins(Grando, 2008), act as regulators of nicotinic acetylcholine receptors, and are likely to be the ancestors of the snake neurotoxins. However, although they are often used as lymphocyte and tumoural markers (Bamezai,2004), many Ly6 human and murine proteins have unknown roles.

We carried out a systematic search for members of the Ly6 superfamily in Drosophila, identifying 36 previously uncharacterised genes coding for one or more Ly6 motifs. We also explored the function of one of these proteins during Drosophila development, that encoded by the gene boudin (bou). Phenotypic analysis of bou mutants shows that this Ly6 protein participates in the formation of paracellular barriers in epithelial and neural tissues, physiological fences that regulate the passage of solutes between cells in both epithelial and glial sheaths(Banerjee and Bhat, 2007; Tepass et al., 2001). We show that bou is required for the organisation of septate junctions (SJs),invertebrate adhesion structures fulfilling an equivalent role to the vertebrate tight junctions. Differing from known SJ constituents, bourequirements are non-cell-autonomous, and, accordingly, we find that Bou can be released in extracellular particles and become incorporated into neighbouring cells. Altogether, our results indicate that Drosophilacould be an attractive system in which to study the function and general properties of Ly6 proteins in a developmental context.

We used the PSI-BLAST algorithm(Altschul et al., 1997) and the Rtv, CD59 and uPAR sequences as queries against the Drosophila RefSeq database (Pruitt et al.,2007). Newly identified Ly6 homologues were incorporated into the search matrix until no more members could be identified, typically after six to seven rounds of iterative search. Then, we used these sequences as novel queries. An identical strategy was used in the honeybee.

Full definitions of these stocks can be found in FlyBase(http://flybase.org/): bou^PG27 (bouGAL4), l(1)6Ea² (bou^let),Dp(1;Y) ct⁺y⁺, rtv¹¹, nrg¹⁴,NrgGFP, PdiGFP^74-1, UASApoLII-Myc, apGAL4, btlGAL4 UASActinGFP,nulloGAL4, enGAL4, ptcGAL4, dppGAL4, tubGAL4, hsFLP tubGAL80 FRT19A;UASmCD8GFP and GAL80^ts. The FM7c-Actin-lacZand FM7c-KrGAL4UASGFP balancers were used for genotyping. Temperature shifts at 18°C were done 24 hours before dissection in cultures containing third larval instars of the bou^PG27/NrgGFP; UASHA-Bou/+;GAL80^ts/+ genotype. Mutant clones were induced in 48-hour larvae by 1 hour heat shock at 37°C, in bou^PG27FRT19A/hsFLP tubGAL80 FRT19A; UASmCD8GFP/+; tubGAL4/+ larvae. A ywFRT19 chromosome was used as control.

Dye diffusion into trachea and chordotonal organs was analysed injecting with a micromanipulator 10 mg/ml 10 kDa rhodamine dextran (Molecular Probes)into the body cavity of stage 16 (14- to 16-hour) embryos(Lamb et al., 1998). Diffusion into the nerve cord was monitored in 22-hour embryos. Samples were visualised with a Leica SP2 confocal microscope within 20-30 minutes of injection.

Three independent PCR fragments containing the bou transcription unit were amplified from bou^let genomic DNA, cloned and sequenced. The HA-tag was introduced in frame within the Bou coding region by PCR, using specific oligonucleotides and the RE28342 cDNA (DRGC). The HA-BouΔC was generated substituting Gly128 for a stop codon. Both constructs were sequenced and subcloned into pAc5.1 (Invitrogen) for cell transfection or into pUAST (Brand and Perrimon, 1993) for transgenesis.

Cell culture, transfections and antibody staining were carried out as in Koh et al. (Koh et al., 2008). S2 cells co-transfected with pAcDMoe-GFP (kind gift from F. Payre, CBD,Toulouse, France) and pAcHA-Bou or pAcHA-BouΔC were fixed and stained in either permeabilising (PBS, 0.1% Triton-X100) or non-permeabilising (PBS)conditions. Transfected KcD26 (2×10⁶) cells were incubated at 25°C for 1 hour in PBS, with or without 1 unit of phosphatidylinositol-specific phospholipase C (PI-PLC, Sigma). Cell proteins were extracted in 1× RIPA, whereas the extracellular medium was precipitated with TCA-DOC and resuspended in 50 μl of 1× loading buffer. For each condition, 20 μg of cell extracts and 25 μl of supernatant were run in an SDS-PAGE gel and blotted with anti-HA.

Sense riboprobes were generated from clone RE28342 for in situ hybridisation (Waltzer et al.,2003). Embryos and larval tissues were fixed for 20-30 minutes in PBS 4% paraformaldehyde. Blocking, washing and overnight incubation with primary and secondary antibodies were carried out in 0.05% Triton-X100 0.1%BSA. Primary antibodies include mouse anti-β-gal (Promega), rabbit anti-β-gal (Cappel), mouse anti-HA (Covance), rabbit anti-HA (Clontech),rabbit anti-GFP (Torrey), anti-NrxIV (gift of H. Bellen, Baylor College of Medicine, Houston, TX, USA), rat anti-Crb (gift of U. Tepass, University of Toronto, Toronto, Canada), and monoclonals 9E10 anti-Myc, anti-2A12, 4F3 anti-Dlg, DCAD2 anti-DECad, BP104 anti-Nrg and 7G10 anti-FasIII, all from DSHB. Secondary FITC and TRIT conjugated antibodies and streptavidin come from Molecular Probes. We also used CBP-FITC (NEB). Samples were visualised with a Leica SP2 confocal microscope.

In general, Ly6 domains share little sequence similarity, making their identification by genomic annotation algorithms difficult. For instance, the only known Drosophila proteins containing this domain are the five TGFβ receptors (tkv, babo, sax, put and wit) and the product of the gene retroactive (rtv)(Moussian et al., 2005). Using the iterative PSI-BLAST program (Altschul et al., 1997), we carried out a systematic search for Ly6 members in the fly genome, screening for domains of about 100 amino acids containing 10 cysteines, where Cys¹ and Cys² are always separated by two residues and an Asn residue contiguous to the last cysteine (canonical 10C motif). Alignment of the Drosophila Ly6 domains revealed the presence of short intervening distances between Cys⁸ and Cys⁹ (0-3 residues) and Cys⁹ and Cys¹⁰ (4-5 residues), confirming that they belong to the Ly6 family (see Fig. S1 in the supplementary material). Thus, besides the five TGFβ receptors, we have identified in flies 72 Ly6 canonical motifs and 14 related domains encoded by 36 different genes not previously ascribed to any known family(Table 1).

Ly6 motifs are never found in combination with other extracellular domains,a principle also valid in Drosophila, where a single Ly6 domain is the only module present in 28 proteins. In the other eight cases, multiple Ly6 motifs are found, as in the human uPAR and C4.4A proteins(Galat, 2008). We also identified three different types of Ly6-related domains lacking key features of a canonical domain. The first variant found was the 8C domain (11 motifs found in two proteins), with only eight cysteines. Interestingly, 8C domains are similar to the vertebrate uPAR domain I, which lacks both Cys⁷and Cys⁸ and also the disulphide bridge formed by these residues. Nonetheless, the uPAR domain I also adopts a three-finger fold(Huai et al., 2006). Another variant is what we call `atypical 10C' domain (a10C), found only in two proteins. This motif could have arisen by replacement of Cys<sup>8</sup> by a new Cys placed two residues after the C-terminal Asn (see Fig. S1 in the supplementary material). Finally, we found a group of three contiguous genes coding for long stretches of repeated amino acids (mostly Ser, Thr and charged residues) in the region between Cys<sup>4</sup> and Cys<sup>5</sup> (see Fig. S1 in the supplementary material). We termed these long motifs `disordered 10C' (d10C), as these repeats are predicted to form flexible regions of unstable conformation, called regions of intrinsic disorder(Dyson and Wright, 2002). Hence, it is not clear whether these proteins adopt a three-finger fold.

Vertebrate members of the Ly6 family are synthesised as propeptide precursors entering the endoplasmic reticulum thanks to an N-terminal signal peptide. They have often a second C-terminal hydrophobic peptide placed after their Ly6 domain, permitting the addition of a GPI anchor to an internal sequence of the precursor. In Drosophila, all the Ly6 genes code for an N-terminal portion of 25-35 residues and a 20- to 30-residue C-terminal stretch, raising the possibility that all could incorporate a GPI anchor.

We could not establish orthology relationships between Drosophilaand vertebrate Ly6 proteins due to their low degree of sequence similarity. However, for each Drosophila melanogaster protein we identified a putative orthologue in Drosophila grimshawi, a distant drosophilid species. We found that the organisation of canonical, 8C, a10C and d10C Ly6 motifs is also conserved in this species, despite 60 mya of separate evolution(Tamura et al., 2004)(Table 1). Therefore, the whole fly Ly6 family was already present in the drosophilid ancestor. We also performed a search for Ly6 members in the honeybee genome, finding only 14 genes coding for this motif. Among these, 12 are orthologues of Drosophila genes (sequence identity above 50%). Thus, several gene duplication events followed by rapid divergence occurred in the drosophilid lineage, which nonetheless conserved most of the ancestral Ly6 members. Intriguingly, as is also the case in humans and mice(Galat, 2008), the Drosophila genes coding Ly6 proteins are often contiguous in the genome, forming six clusters that group together 24 genes(Table 1).

Existing databases of gene expression patterns allowed us to visualise during embryogenesis the transcript distribution of 21 members of the Drosophila Ly6 family (Tomancak et al., 2002). They are expressed in a dynamic and tissue-specific pattern in a wide range of contexts, from the epidermis and its derivatives to the nervous system and the gut (Table 1). Thus, Ly6 genes can potentially participate in many different developmental and physiological processes.

We analysed the function of a new member of this family, the product of the CG14430 gene, which we have called boudin (bou). The bou locus codes for a protein of 149 residues presenting all the typical features of Ly6 members. Bou is predicted to be a GPI-anchored protein by the Big-PI algorithm (Eisenhaber et al., 1998), which proposes Asn125 as the omega site of the mature protein, where the GPI moiety is attached(Fig. 1B). Unlike other members of the Drosophila Ly6 family, the Bou sequence appears conserved in other insect genomes, where we have identified clear orthologues(Fig. 1C).

The bou transcript was first detected by in situ hybridisation at the cellular blastoderm stage, first ubiquitously and then accumulating in the invaginating mesoderm (Fig. 1D-F). By stages 13 and 14, the hindgut, foregut, salivary gland and tracheal cells express high levels of bou, which is also present at lower levels in the epidermis (Fig. 1G,H). This pattern is maintained until the end of embryogenesis,although transcript levels start declining after stage 14(Fig. 1I). We did not detect bou expression in the ventral nerve cord or in mesodermal derivatives, indicating that at late stages this gene is expressed only in ectodermal tissues.

In a genetic screen we recovered a GAL4 P-element embryonic lethal insertion in the 5′ UTR of bou (bou^PG27)(Fig. 1A)(Bourbon et al., 2002). Both remobilisation of this transposon or expression of an HA-tagged Bou form(HA-Bou), using bou^PG27 itself as driver, restored the viability of bou^PG27 flies. We used this allele to carry out complementation tests with lethal mutations mapping to the same chromosomal region and identified a second bou lethal mutation, l(1)6Ea² (bou^let)(Perrimon et al., 1989). Sequencing of the bou region in the bou^letchromosome revealed a deletion of 238 nucleotides encompassing the coding region and most of the 3′ UTR (Fig. 1A). We predict bou^let to be a null allele, as this deletion truncates the Ly6 domain and eliminates the C-terminus of the Bou precursor (Fig. 1B).

As bou is expressed in the tracheal cells, we first looked for morphological defects in this tissue. Staining with the 2A12 tracheal luminal marker and labelling of tracheal cells with ActinGFP revealed that bou^PG27 and bou^let embryos display identical phenotypes, presenting tracheal tubes with abnormal shape and dimensions (Fig. 2A-H; and data not shown). At stage 16, the branch pattern of the tracheal network seemed normal, but the dorsal trunk appeared elongated and convoluted and we observed that the 2A12 luminal staining was interrupted along the dorsal branches and transverse connectives (Fig. 2B,E). These phenotypes point to tracheal lumen expansion defects(Beitel and Krasnow, 2000), and indeed, the tracheal dorsal trunk of stage 15 bou mutants did not present a uniform width (Fig. 2W), showing instead a series of bulging cysts resembling a string of sausages (hence the name `boudin', a French black sausage).

The defects observed in bou trachea were strikingly similar to those seen in mutant embryos for rtv, a gene coding for another Ly6 protein (Moussian et al.,2005). Rtv is required for the formation of an intraluminal chitin cable, which is essential for proper tube expansion of Drosophilatrachea (Devine et al., 2005; Moussian et al., 2006; Tonning et al., 2005). To determine whether bou and rtv act by similar mechanisms, we monitored chitin cable integrity in stage 16 bou mutants, using a fluorescent chitin-binding probe (CBP). Chitin forms an organised filamentous structure in the lumen of wild-type trachea, but in the rtv¹¹ null allele this structure is lost and CBP stains a diffuse luminal material (Fig. 2J-L) (Moussian et al.,2006). The chitin cable of bou^let mutants also loses its fibrous aspect, although the CBP staining is more intense than in rtv¹¹ trachea (Fig. 2K,L). Thus, chitin cable formation is affected in both bou and rtv mutants, with bou presenting a weaker phenotype.

Mutations in different Drosophila SJ components also result in embryos with abnormal trachea, presenting the same cysts observed in rtv and bou mutants(Beitel and Krasnow, 2000; Wu and Beitel, 2004). As SJs are adhesion structures required for the establishment of paracellular barriers regulating molecular diffusion through epithelia, we examined the integrity of this barrier in both rtv and bou mutants. For this, we injected 10 kDa fluorescent dextran into the body cavity of live embryos and monitored the capacity of this molecule to enter the tracheal lumen (Lamb et al., 1998). At stage 16, both wild-type and rtv¹¹ tracheal cells formed an efficient paracellular barrier, preventing dye diffusion into the lumen(Fig. 2M-R). By contrast,dextran was readily detected inside the bou^let tracheal tubes within 20 minutes of injection and we observed abnormal dye deposits trapped between contiguous cells (Fig. 2N,Q). Thus, the paracellular barrier is disrupted in boumutants, suggesting that this gene is implicated in SJ organisation. To confirm this hypothesis, we examined the subcellular localisation of an SJ component, Fasciclin3 (Fas3), in tracheal cells(Beitel and Krasnow, 2000; Wu and Beitel, 2004). In wild-type and rtv¹¹ tracheal cells, this marker accumulates in the most apical part of the lateral membrane, where SJs are present (Fig. 2S,U). By contrast, this apical accumulation was lost in bou^letembryos and Fas3 appeared uniformly distributed along the lateral membrane(Fig. 2T). Therefore, whereas bou is required for SJ maintenance, rtv seems dispensable for this process, indicating that these genes regulate tracheal morphogenesis by different mechanisms.

To further characterise the bou phenotypes, we analysed the subcellular localisation of several SJ components, including Discs large 1(Dlg1), Neurexin IV (Nrx-IV) and the protein-trap fusion NeuroglianGFP(NrgGFP) (Beitel and Krasnow,2000; Wu and Beitel,2004). Similarly to Fas3, all these markers appeared delocalised in the lateral membrane of bou^let tracheal cells(Fig. 3A,G-I,D,J-L). We also monitored the distribution of the cell polarity marker Crumbs(Tepass et al., 1990) and the apical junction component DE-Cadherin (Shotgun - FlyBase)(Oda et al., 1994). As in controls, these markers localised to the most apical part of the tracheal cells throughout development, indicating that bou specifically affects the SJ organisation rather than the general polarity of the cell(Fig. 3B,C,E,F). In addition, bou is required for the early establishment of SJ in this tissue (see Fig. S2 in the supplementary material), because a clear delocalisation of the Nrx-IV marker was already observed by stage 14, when pleated SJ begin to form(Tepass and Hartenstein,1994).

Finally, we tested if bou is required for SJ organisation in other epithelial tissues, such as the epidermis, salivary gland and embryonic hindgut. Analysis of bou^let embryos showed that NrgGFP,Dlg1 and Nrx-IV are also delocalised in these tissues(Fig. 3M-R; and data not shown). Thus, bou is generally required for SJ organisation in embryonic ectodermal derivatives.

Seeking to extend the characterisation of bou requirements to larval tissues, we analysed the contribution of this gene to the morphogenesis of imaginal discs, the epithelial precursors of the adult integument. For this, we studied mosaic individuals containing clones of homozygous bou^PG27 cells, using the MARCM technique to positively label the mutant territories (Lee et al.,2000). We found that large bou wing clones generated early in larval development did not show any obvious growth defects. Moreover,the SJ marker Fas3 protein was correctly localised in bou mutant cells (Fig. 4A-A′). Thus, bou function could be restricted to the embryonic tissues or, more intriguingly, the surrounding cells could exert a rescuing activity upon the mutant territories.

To discriminate between these possibilities, we sought to establish whether bou is required in larval tissues. To bypass embryonic lethality and recover bou^let mutant larvae, we expressed wild-type HA-Bou in bou^let embryos using nulloGAL4, a driver only active at the blastoderm stage(Coiffier et al., 2008). In this way, we obtained bou^let mutants now dying at pupariation and presenting discs with reduced size and abnormal shape. At the cellular level, we observed that Crumbs localisation was not affected in bou^let wing cells. By contrast, the SJ marker Fas3 was delocalised and distributed uniformly along the basolateral membrane(Fig. 4C-F). Thus, the bou product is also specifically required for SJ organisation in imaginal epithelia. Consistent with the idea that bou phenotypes are not cell-autonomous, we recovered morphologically normal adult bou^let mutant flies expressing HA-Bou with engrailed,patched or decapentaplegic GAL4, three drivers with clear-cut regionalised patterns. Moreover, staining for Fas3 in bou^let;enGAL4/UASHA-bou mutant discs confirmed that SJs are normal throughout the disc and not only in the engrailed domain(Fig. 4B-B′). Therefore,cells expressing the Bou protein can rescue the mutant phenotypes in surrounding territories, confirming that bou acts non-autonomously.

One possibility is that Bou itself can travel from cell to cell. Indeed,some vertebrate members of the Ly6 family have the ability to diffuse, either as soluble ligands or coupled to lipid particles via their GPI anchor(Chimienti et al., 2003; Rooney et al., 1993). To gain insight into the Bou mode of function, we generated transgenic flies expressing a C-terminal truncated form of HA-Bou (HA-BouΔC), coding for an intact Ly6 motif but missing the last 22 residues of the precursor(Fig. 1B). We predicted this molecule would behave as an active soluble form, as the C-terminal region is necessary for GPI addition in other GPI-anchored proteins. Instead, we observed that HA-BouΔC expression did not rescue the bou^PG27 lethality, indicating that the Bou C-terminus integrity is essential for its activity. Moreover, expression of HA-BouΔC in the tracheal cells driven by breathlessGAL4 could not rescue Fas3 delocalisation in bou^let mutant trachea(Fig. 4J,N,R), whereas expression of a wild-type HA-Bou form not only rescued the bou^let phenotypes in the tracheal cells but also in the salivary gland and the hindgut, tissues not expressing HA-Bou in this genetic combination (Fig. 4I,M,Q; and data not shown). This finding indicates that Bou-targeted expression can elicit non-autonomous effects in other tissues, opening up the possibility that Bou could diffuse systemically.

To characterise the Bou subcellular distribution, we first sought to confirm whether Bou is a membrane GPI-anchored protein. In DrosophilaS2 cells, HA-Bou is observed in the cell body and also the plasma membrane, as confirmed by immunostainings carried out in non-permeabilising conditions(Fig. 5A,C). By contrast, the HA-BouΔC form could only be detected in internal cell compartments after permeabilisation, showing that the Bou C-terminus is essential for cell membrane insertion (Fig. 5B,D). HA-Bou is a GPI-anchored protein, because incubation of intact cells with phosphatidylinositol phospholipase C (PI-PLC) provokes its release to the extracellular medium (Fig. 5E).

Next, we studied the HA-Bou subcellular localisation in embryonic tissues and in the wing disc, activating its expression with tissue-specific drivers. As in cultured cells, HA-Bou appeared distributed homogenously throughout the tracheal cell body and did not accumulate in any particular structure(Fig. 6A). We found that the HA-BouΔC form has a more restricted localisation, as it was excluded from contact regions between adjacent cells(Fig. 6B). Co-staining with the SJ marker NrgGFP showed that HA-BouΔC was absent from the lateral membrane, whereas the HA-Bou staining overlapped with NrgGFP in the apical part of the cells (Fig. 6A,B).

In the wing disc, HA-Bou was also present in the cell body and throughout cell contact regions (Fig. 6C,E,G,H). By contrast, the HA-BouΔC form distributed like the disulphide isomerase PdiGFP, a resident enzyme of the endoplasmic reticulum (ER) (Bobinnec et al.,2003) (Fig. 6D,F,I,J). Thus, the HA-BouΔC form could not exit the ER,whereas the full-size HA-Bou reached membrane areas from which the ER is excluded (Fig. 6C,D,G,I). Co-staining with NrgGFP revealed that HA-Bou was present at the SJ level and accumulated in an apical region, placed above the SJ, that could correspond to a secretion compartment (Fig. 6E,H, see below).

To gain insight into the dynamics of HA-Bou protein localisation, we profited from the large size of the third-larval-instar salivary gland cells. Using the bou^PG27 GAL4 driver, we drove expression of HA-Bou and HA-BouΔC in this cell type, placing a GAL80^ts thermosensitive repressor in the same genetic background (McGuire et al.,2003). At 25°C, the GAL80^ts repressor is inactive and we observed a strong accumulation of the HA-Bou forms in the salivary gland cell body (Fig. 6K,L). Then, we shifted the larvae at 18°C 24 hours before dissection, activating the GAL80^ts repressor and shutting down synthesis of HA-Bou protein. In these conditions, HA-Bou disappeared from the cell body and accumulated at high levels in the lumen of the salivary gland. In addition, we observed a weak but clear staining at the lateral membrane, coinciding with the NrgGFP SJ marker (Fig. 6M). As expected, the levels of HA-BouΔC decayed uniformly after the temperature switch (Fig. 6N). These results confirm that Bou associates with SJ membrane regions, although its localisation is not restricted to these membrane domains.

One way to explain the non-cell-autonomy of the bou phenotypes is that Bou could be secreted extracellularly. Consistently, we noticed the presence of extracellular particles containing this protein in the luminal surface of the wing disc, budding off from the apical HA-Bou-enriched domain(Fig. 6C; Fig. 7A,B). These particles were seen over apGAL4-expressing cells, but were also detected in other territories of the wing disc lumen, indicating that HA-Bou can diffuse(Fig. 7A). Interestingly,increasing the laser power of the confocal microscope, we observed a diffuse intracellular staining and the presence of dots containing HA-Bou in cells adjacent to the apGAL4 territory, showing that the secreted protein is incorporated by neighbour cells (Fig. 7D). In addition, we observed intracellular vesicles accumulating high levels of HA-Bou within the apGAL4 cells(Fig. 7C). As none of these structures was observed in HA-BouΔC-expressing discs(Fig. 7E-H), we conclude that they reflect the ongoing traffic of the HA-Bou protein in the wing epithelium. We tested if the Bou extracellular particles are lipophorin particles, as these lipid vesicles are known to contain GPI-anchored proteins(Panakova et al., 2005). However, co-expression of HA-Bou and ApoLII-Myc, the main protein component of lipophorin particles, revealed that these markers label different vesicle populations (Fig. 7I-K). Thus,HA-Bou extracellular transport is unlikely to rely on lipophorin particles.

We show that bou function is essential for SJ assembly in epithelial tissues. However, SJs also play a physiological role in the glial cells forming the blood-brain barrier and isolating the insect neural tissues(Banerjee and Bhat, 2007). This prompted us to examine if bou is involved in the maintenance of this barrier in the embryonic chordotonal organs. These sensory mechanoreceptors are made of five units, each formed by three glial cells protecting a sensory bipolar neuron. Two of these glial cells, the cap cell and the scolopal cell,form a luminal cavity encapsulating the neuron ciliar dendrite and forming SJs with each other to isolate this structure(Fig. 8A)(Carlson et al., 1997). The cell contacts between cap and scolopal cells accumulate SJ markers, such as Nrx-IV and NrgGFP (Fig. 8B-F)(Banerjee et al., 2006). We performed dextran injections in stage-16 wild-type embryos and confirmed that this dye is excluded from the chordotonal lumen(Fig. 8B-D). By contrast, the dye diffused into this structure in bou^let embryos of the same stage (Fig. 8H-J), and the SJ markers appear delocalised (Fig. 8H-L). Therefore, bou is also required for SJ organisation in the chordotonal organs.

The embryonic ventral nerve cord is also protected by a specialised layer of glial cells, the subperineural glia, which form an efficient paracellular barrier (Schwabe et al., 2005; Stork et al., 2008). Performing dye injections in 22-hour-old embryos(Fig. 8M-O), we observed that this barrier was still functional in bou^let mutants,whereas, as expected, the dye penetrated into the nerve cord of nrg¹⁴ mutants (Schwabe et al., 2005). Thus, the integrity of the central nervous system paracellular barrier does not depend on bou activity.

Our results reveal that Bou plays an essential role in the organisation of SJs and the maintenance of paracellular barriers in Drosophilaepithelia and chordotonal organs. Although some vertebrate members of the Ly6 family are known to participate in cell-adhesion processes(Bamezai, 2004), this is the first example showing that they are required for the formation of this type of cellular junction. As bou is well conserved in other insect genomes,its role in SJ organisation could have been maintained during evolution. Invertebrate SJs and vertebrate tight junctions are considered analogous structures because both participate in the establishment of paracellular barriers, although they present a different organisation. However, vertebrates have adhesion structures functionally, morphologically and molecularly similar to insect pleated SJs (Bellen et al.,1998): the so-called paranodal septate junctions, which are formed by neural axons and Schwann cells, at the level of the Ranvier's nodes(Schafer and Rasband, 2006). We show that Bou is necessary for SJ organisation in the embryonic peripheral nervous system, indicating that its activity is required in some neural tissues. Thus, our observations raise the possibility that some vertebrate Ly6 proteins could be involved in the formation of paranodal septate junctions,which are essential for axonal insulation and propagation of action potentials.

In insects, the epithelial and neural SJs share many components, so our observation that bou is not required for blood-brain barrier maintenance in the ventral nerve cord came as a surprise, revealing the existence of tissular and molecular heterogeneities in the organisation of these junctions. It will be interesting to establish whether these differences also determine different barrier selective properties. We speculate that other Ly6 proteins expressed in the nervous system could contribute to blood-brain barrier formation in the subperineural glia.

Our results show that bou inactivation specifically perturbs the organisation of SJs. As these structures are large extracellular complexes including different transmembrane and GPI-anchored proteins(Wu and Beitel, 2004), one hypothesis is that Bou could be a membrane SJ component. Consistently, HA-Bou is found at lateral contact areas in tracheal, salivary gland and wing disc epithelia, overlapping with the membrane domains that contain SJ. However,this protein does not significantly accumulate in these membrane regions and is also seen in the most apical part of the cells, opening up the possibility that it could operate in other membrane areas or act as a signalling molecule. Indeed, studies in vertebrates indicate that Ly6 proteins can assume roles in both cell signalling and cell adhesion(Bamezai, 2004). Clearly,identification of the Bou molecular partners will be a crucial step in understanding how this protein exerts its activity.

In contrast with other genes required for SJ formation(Genova and Fehon, 2003), bou functions in a non-cell-autonomous way. Accordingly, the Bou protein is found in extracellular particles and can be captured by neighbouring cells, suggesting that its diffusion is responsible for the phenotypic non-autonomy. Although it is possible that Bou could act as a secreted ligand after release of its GPI anchor, a parallelism with other members of the family suggests that the full molecule could instead become incorporated into the membrane of neighbouring cells(Neumann et al., 2007). In fact, the mammalian Ly6 member CD59, a cell-surface antigen protecting host cells from the complement attack, travels coupled to membranous vesicles called prostasomes with its intact GPI. These specialised vesicles are secreted into the seminal fluid by prostatic glands, and allow CD59 transfer to the sperm cells, which can then elude complement attack(Rooney et al., 1993). GPI-bound CD59 has also been found associated with human HDL apolipoproteins(Vakeva et al., 1994). However, we show that the Bou particles are not lipophorin vesicles, the insect equivalent to vertebrate apolipoproteins(Rodenburg and Van der Horst,2005). Therefore, the fly wing epithelium could produce a different type of vesicle, possibly similar to prostasomes, which we propose to call `boudosomes'. Unfortunately, we could not determine whether the Bou GPI anchor is required for incorporation into these particles, because the C-terminus of the protein seems essential for prior exit from the ER. Thus,future work will be needed to characterise the biochemical features of boudosomes and their function.

Little is known about how epithelial cells coordinate their activity to form efficient fences. As many SJ components are required in a cell-autonomous manner (Genova and Fehon,2003), their simultaneous expression by each individual cell seems a prerequisite for barrier assembly. A component and/or SJ regulator shared by different cells could be an element coordinating the organisation of efficient barriers in a dynamic epithelium. Alternatively, Bou extracellular traffic could be a specialised feature of this GPI-anchored protein and not have functional relevance for SJ assembly during normal development.

Besides Bou and the TGFβ receptors, the only member of the Ly6 fly family with a characterised role is the Rtv protein, which is also expressed in epidermal derivatives. We show that both bou and rtvmutants affect the organisation of the tracheal chitin luminal cable, although rtv mutants exhibit stronger phenotypes. However, SJ integrity is a prerequisite for proper assembly of the chitin cable(Swanson and Beitel, 2006),and we show that rtv is neither required for paracellular barrier integrity nor for SJ organisation. Thus, whereas our observations confirm that chitin cable deposition relies on the organisation of SJs, they demonstrate that these Ly6 proteins act in different processes.

We have carried out the first description of the Ly6 superfamily in the genome of an insect, identifying 36 new genes bearing this domain in Drosophila. The conservation of these proteins among the drosophilids indicates that the family was established before the evolutive radiation of this group. By contrast, we have identified only 14 genes coding for Ly6 domains in the honeybee genome. Most of these genes have fly orthologues, like bou and rtv, pointing out the existence in higher insects of a core of ancestral genes with potentially conserved roles. Thus, repeated events of gene duplication followed by rapid divergence of coding and regulatory sequences occurred in the drosophilid lineage. Indeed, the presence of genomic clusters grouping together different Ly6 genes is a novel evolutive acquisition, as the conserved genes tend to be in isolated positions(Table 1).

It seems that genes coding for a Ly6 motif are prone to sudden phases of extensive duplication and diversification in different phylogenetic groups. In fact, an interesting parallelism can be drawn with the evolution of three-finger elapid snake venoms. This large group of Ly6 secreted proteins operates using diverse strategies, such as forming membrane pores, targeting the activity of acetylcholine receptors, inactivating acetylcholine esterase or blocking platelet aggregation (Tsetlin,1999). Moreover, crystallographic analysis has revealed that three-finger toxins can interact with their targets via virtually any part of their solvent exposed surfaces (Kini,2002). Yet, most of them share a common ancestor(Fry et al., 2003). Given the broad diversity of expression patterns exhibited by the different Drosophila Ly6 members, it is likely that gene duplication has been followed by acquisition of new developmental and physiological functions. Analysis of this insect family from an evolutive perspective could be a way to enhance our understanding of the mechanisms underlying the generation of evolutive innovations.