Drosophila Varicose, a member of a new subgroup of basolateral MAGUKs, is required for septate junctions and tracheal morphogenesis

The function of organs such as the lung, kidney and vascular system depends on epithelial and endothelial tubes of specific sizes. However, the cell biological and molecular processes that control tube sizes are largely unknown. The Drosophila tracheal system is a network of ramifying epithelial tubes that serves as a combined pulmonary-vascular system to directly deliver oxygen to tissues (reviewed by Uv et al., 2003). The comparative simplicity and genetic tractability of the tracheal system has made it one of the best models of tubular epithelial morphogenesis. The tracheal system develops from a series of sacs into a complex network of branches through a highly orchestrated series of cell migrations, cell shape changes and rearrangements of cell-cell junctions(Samakovlis et al., 1996). An important element of these morphogenetic events is that changes in tube size occur reproducibly during specific developmental periods(Beitel and Krasnow, 2000). Each tracheal branch has a specific size that results from the action of branch-specific signaling events that at least in some branches are known to act through transcription factors such as Spalt-Major (Spalt)(Chen et al., 1998; Chihara and Hayashi, 2000; Franch-Marro and Casanova,2002; Myat et al.,2005; Ribeiro et al.,2004). At least one additional transcription factor, Grainyhead,is required to control tube length and apical cell surface in the major tracheal branches, but the transcriptional targets that more directly mediate these functions remain to be identified(Hemphala et al., 2003). Recent work by multiple groups has produced a basic molecular framework of the mechanisms that execute the size changes of `tube expansion', a process that increases the diameter - but not the length - of the major tracheal tubes over a 2 hour period, and then gradually lengthens the tubes without changing their diameters (reviewed by Swanson and Beitel,2006). These tube size changes result from changes in cell shape and possibly cell size, but do not involve changes in cell number(Beitel and Krasnow, 2000).

The tube expansion mechanism depends upon a fibrillar, chitin-based extracellular matrix that is assembled in the tracheal lumen at the beginning of the diameter dilation (Tonning et al.,2005). As development progresses, chitin at the apical cell surface is organized into a highly patterned, multilayered cuticle. Lumenal chitin is eliminated before hatching. Defects in chitin synthesis or organization cause tracheal tube diameters to become either too large or too small, and tube lengths to become over-elongated(Araujo et al., 2005; Devine et al., 2005; Moussian et al., 2006; Tonning et al., 2005). The exact role of the chitin-based matrix in controlling tracheal cell shape is unclear. Although the lumenal matrix and cuticle may serve as structural forms or `mandrils' that mechanically shape the tracheal cells and tubes, an instructive or signaling role for the matrix is suggested by the observation that the organization of the β_H-spectrin cytoskeleton is altered in chitinsynthetase mutants(Tonning et al., 2005).

Beginning at stage 15, organization of the lumenal matrix requires the lumenal secretion of the putative chitin deacetylases, Vermiform (Verm) and Serpentine (Serp). In verm and serp mutants, the chitinbased matrix becomes disorganized and tracheal tubes become too long(Luschnig et al., 2006; Wang et al., 2006). Surprisingly, lumenal secretion of Verm requires a cell-cell junction termed the septate junction (SJ) (Wang et al.,2006). Septate junctions are complex cell adhesion junctions that have at least 15 known components (reviewed by Knust and Bossinger, 2002; Margolis and Borg, 2005; Wu and Beitel, 2004). These include transmembrane cell-adhesion proteins such as Neurexin IV (Nrx-IV;herein referred to as Nrx) and Neuroglian (Nrg), cytoplasmic proteins such as the FERM-domain protein Coracle (Cor; Cora - Flybase), the basal polarity proteins Scribbled (Scrib), Discs large (Dlg; Dlg1 - Flybase), and Lethal giant larvae (Lgl; L(2)gl - Flybase), and proteins with roles that remain to be determined, such as the Na⁺/K⁺-ATPase(Genova and Fehon, 2003; Paul et al., 2007; Paul et al., 2003). Mutations in most known SJ components cause tracheal phenotypes indistinguishable from the verm mutant phenotype, consistent with the failure of Verm to be secreted into the tracheal lumen in the SJ mutants so far examined(Wang et al., 2006). Secretion of other apical lumenal markers appears normal in SJ mutants, indicating that Verm is secreted by a specialized pathway, the mechanism of which remains to be determined.

Although the role of SJs in lumenal (apical) secretion is not understood,other SJ functions are well defined. SJs have functional and molecular similarity to vertebrate tight junctions (TJs), in that both junctions require members of the claudin protein family to create the paracellular diffusion barriers between epithelial cells that are essential to the survival of multicellular animals (Anderson et al.,2004; Behr et al.,2003; Wu and Beitel,2004; Wu et al.,2004). However, SJs are not simply the homologs of TJs, because there are significant ultrastructural, molecular and functional differences between SJ and TJs (reviewed by Wu and Beitel, 2004). For example, TJs are apical of adherens junctions(AJs) and contain conserved apical polarity complexes, while SJs are basal of AJs and contain the polarity proteins Scrib, Dlg and Lgl, which have vertebrate homologs that also localize basolaterally (reviewed by Knust and Bossinger, 2002). Thus, in some respects SJs are more related to complexes found in the basolateral regions of vertebrate epithelial cells than to TJs.

Although Scrib, Dlg and Lgl establish and currently define the similarity between SJ and vertebrate basolateral regions, it is notable that these proteins are not representative of most SJ components. DrosophilaScrib, Dlg and Lgl are maternally contributed and constitute a distinct subgroup of proteins required for initial epithelial cell polarization during embryonic stages 5-8 (Bilder and Perrimon,2000; Strand et al.,1994; Tanentzapf and Tepass,2003; Woods et al.,1996). By contrast, most SJ components are not maternally expressed, are not required for cell polarity and only function relatively late in development when SJs begin forming during stage 13 (reviewed by Bilder, 2004; Knust and Bossinger, 2002; Tepass et al., 2001). Whether the Scrib, Dlg and Lgl proteins nucleate SJ assembly, or whether the nascent SJ recruits and incorporates Scrib, Dlg and Lgl has not been determined. It also has not yet been determined how Scrib, Dlg and Lgl are localized to the basolateral membrane in either Drosophila or vertebrate epithelia. Thus the similarity between Drosophila SJ and vertebrate basolateral regions has been limited to polarity complexes, and has not extended to cell adhesion complexes.

In this report we show that vari encodes a previously uncharacterized, membrane-associated, guanylate kinase (MAGUK) scaffolding protein that is required for SJ organization and that directly binds the cell adhesion protein Neurexin IV. Importantly, Vari helps define a new subgroup of MAGUKs that includes vertebrate PALS2. Both Vari and PALS2 localize basolaterally in epithelial cells and both interact through a PDZ domain with a basolateral adhesion protein. Thus, Vari is the first late-expressed SJ component to have a vertebrate homolog, and together Vari and PALS2 extend the similarity of Drosophila and vertebrate basolateral regions from polarity complexes to adhesion complexes.

Stocks were obtained from the Bloomington Drosophila Stock Center and published sources. SJ barrier function was determined as previously described (Lamb et al., 1998; Paul et al., 2003), with the exception that a CyO dfd-YFP balancer(Le et al., 2006) was used instead of a CyO actin-GFP balancer.

The following antibodies were used: anti-tracheal lumenal 2A12 1:5 and anti-Arm N27A1 1:100 (Developmental Studies Hybridoma Bank); mouse anti-Cor C566.9c and C615.16B 1:500; guinea pig (gp) anti-Cor 1:10000(Fehon et al., 1994); rabbit(r) anti-Dlg 1:500 (Woods et al.,1997); r anti-Nrv2.1 1:500(Paul et al., 2007); r anti-Nrx 1:200 (Baumgartner et al.,1996); r anti-Veli 1:500(Bachmann et al., 2004); rat anti-DE-cadherin DECAD2 1:20 (Oda et al.,1994); r anti-Sinu 1:500 (Wu et al., 2004); r anti-Verm 1:300 and Serp 1:300(Luschnig et al., 2006); gp anti-Verm 1:1000 (Wang et al.,2006). Embryos were fixed in formaldehyde(Samakovlis et al., 1996),except for Sinu and Arm staining, which were heat fixed(Miller et al., 1989). Rat anti-Vari was used at 1:250 with formaldehyde fixation (although heat fixation can also be used) and was produced by cloning cDNA RE01836 into the pBad/His vector (Invitrogen) followed by expression in Escherichia coli,solubilization in 8 mol/l urea, purification with Ni-agarose beads, dialysis against 2 mol/l urea or PBS buffers and inoculation into rats. Guinea pig anti-Vari was produced by inoculating animals with purified 6XHis:Vari PDZ-SH3-GUK (see Protein interactions below, except pET28a, Novagen was used instead of pGEX-4T1). Secondary antibodies were used at 1:200 (Jackson ImmunoResearch and Molecular Probes). Confocal images were acquired on a Leica TCS SP2. To estimate relative levels of staining, heterozygous and homozygous embryos were imaged on the same slide in the same session and image adjustments were applied equally to matched images.

RNAi was performed as previously described(Kennerdell and Carthew, 1998; Wu et al., 2004) using the vari common ORF primers 5′-GCACCCTTTCCATTAAGAGATG,TTCAAGCCAAACATCGAACTTA, ATTGGACTCATACCATCCCAAG and ATGACAAAAGGCATCAGTTCCT,each of which was preceded by a T7 promoter sequence.

The genomic sequence of each vari allele was determined from at least 35 bp 5′ of the first common exon and through the polyadenylation site, as well as 35 bp 5′ and 3′ of each spliceform-specific exon. UAS vari short and long transgenes were constructed by insertion of cDNAs RE01836 and RE31492, respectively, into pUAST followed by germline transformation. cDNAs were obtained from the Drosophila Genome Resource Center and sequenced using an ABI dye-terminator system. GenBank accession numbers: RE35569, DQ787101; RE47555, DQ787102; RE51859, DQ787103;RE54628, DQ787104; RE58272, DQ787105; RE60702, DQ787106; RE61615, DQ787107;RH14941, DQ787108.

ClustalW and phylogenetic tree analyses were performed using the MacVector program (Accelrys) using representative full-length sequences downloaded from GenBank protein databases, expect for zebrafish ZO-1, for which a C-terminal sequence was predicted from genomic DNA sequences to produce a hypothetical protein that had more conservation with other vertebrate ZO-1 sequences. The ClustalW alignment of the sequences is presented in FASTA format (see Fig. S1 in the supplementary material). The phylogenetic tree in Fig. 2 was generated from 1000 bootstrap repetitions using the neighbor-joining method, gap site ignored,random tie breaking of branches with equal values and an uncorrected `p'. Human Carma3 was used as an outgroup to root the displayed tree.

The two-hybrid interaction screen of a 0-20 hour Drosophilalibrary was previously described in Bhat et al.(Bhat et al., 1999). Two of the 15 clones that interacted with the Nrx C-terminus encoded Vari fragments that completely contained the Vari PDZ domain and started with the 5′sequences AATCCGACCGAGCCG and GGAGGCTACCTGTTC. Both clones ended in polyA sequences.

For protein interaction experiments, vari cDNA RH61449 (GenBank Accession number AY121709) was used as a template to amplify 1323 and 251 bp fragments (nt #805-1056 and nt #805-2127) that encode the PDZ, SH3 and GUK domains or only the PDZ domain of Vari, respectively. These fragments were cloned in frame into GST expression vector (pGEX4T1, Pharmacia) to generate fusion proteins of GST-Vari or GST-Vari.PDZ, which were used in the binding experiments. A DNA fragment that encodes the Nrx C-terminal 48 amino acids was cloned in a maltose binding protein (MBP) expression vector to generate MBP-Nrx-CT fusion protein. All proteins were expressed according to the vector manufacturer's instructions and the binding assays were carried out as previously described (Bhat et al.,1999). The Nrx-CT peptide was cleaved from the MBP fusion protein with thrombin and purified over spin column and the purified peptide of approximately 7 kDa used for binding with GST, GST-Vari or GST-Vari.PDZ. The peptide was identified by anti-NRX antibody in western blotting.

The vari gene was originally defined by the vari^3953b mutation, which causes tracheal tubes to become too long and have some diameter abnormalities(Fig. 1B,F,I)(Beitel and Krasnow, 2000). To understand the molecular functions of vari, we used positional cloning to identify the vari transcription unit. Deficiency mapping narrowed the vari interval to an 80 kb region containing at least 36 genes. One of these, CG9326, was predicted to encode a MAGUK expressed at the time of SJ formation(Tomancak et al., 2002). MAGUKs are scaffolding proteins that contain SH3, HOOK, PDZ and catalytically inactive guanylate kinase (GUK) domains, and often contain other protein-protein interaction domains, such as the L27 domain (reviewed by Harris and Lim, 2001). As several known MAGUK proteins localize to, and have key roles at, cell junctions (reviewed by Harris and Lim,2001; Knust and Bossinger,2002; Margolis and Borg,2005), CG9326 was a strong candidate to be vari. Consistent with this prediction, RNAi knockdown of CG9326 caused tracheal dorsal trunk length increases resembling those caused by vari^3953b (Fig. 1D). We definitively demonstrated that CG9326 was vari by showing that the sequence of CG9326 was altered by all eight vari mutations (Fig. 1K,L, Table 1),that the CG9326 protein was reduced or eliminated in varimutants (Fig. 1M,N; Fig. 3P,S and Table 1) and that CG9326 cDNAs could rescue vari mutations(Fig. 1G,J; Fig. 4G-L).

Sequencing of 11 cDNAs revealed that the vari locus generates two major transcript forms that share seven exons encoding the core PDZ, SH3, HOOK and GUK domains, as well as a possible C-terminal PDZ-binding motif(Fig. 1K). The longer splice forms also encode an N-terminal L27 protein-protein interaction domain absent from the shorter isoforms. Together, these results show that variencodes multiple isoforms of a MAGUK required for tracheal morphogenesis.

To gain insight into possible cell biological roles of Varicose, we investigated the relationship between Vari and other MAGUKs. The MAGUK family can be divided into evolutionarily conserved subgroups, and in several cases it has been shown that subgroup members have similar functions(Fig. 2) (reviewed by Funke et al., 2005). For example, the Stardust (Sdt)/PALS1 and Dlg subgroups organize apical and basolateral cell polarity complexes in both flies and vertebrates (reviewed by Bilder, 2004; Margolis and Borg, 2005). We therefore aligned the Vari amino acid sequence with representative sequences from known MAGUK subgroups and MAGUKs that had not previously been assigned to specific subgroups. Phylogenetic trees were then generated using a `bootstrap'algorithm that more robustly indicates relationships than do `best tree'approaches (see Materials and methods). As shown in Fig. 2A, Vari does not belong to any of the previously characterized epithelial MAGUK subgroups such as the Dlg, ZO-1, Sdt/Par-3 or Lin-2/CASK subgroups, but instead belongs to a new subgroup of MAGUKs that includes PALS2(Kamberov et al., 2000), VAM-1(Tseng et al., 2001), MPP6 and MPP2. This subgroup has at least two members each in zebrafish, mice and humans, but to date the in vivo functions of the vertebrate members of this subgroup have not been determined. Importantly, although the domain structure of members of the Vari/PALS2 subgroup resembles that of P55 (MPP1) subgroup members, at the amino acid level the P55 subgroup is much more closely related to the LIN-2/CASK/Caki subgroup and is clearly distinct from the Vari/PALS2/VAM-1 subgroup (Fig. 2A; see Fig. S1 in the supplementary material). Thus, Vari is a founding member of a new subgroup of MAGUKs, the functions of which have not been previously determined.

Several additional results from the phylogenetic analysis are also notable. First, while the P55 subgroup initially appears to be vertebrate-specific because there are no corresponding genes in Drosophila or C. elegans that have the PDZ-SH3-GUK structure of the P55 subgroup members,the invertebrate equivalents of P55 may be alternative splice products of the LIN-2/Caki subgroup that lack the CAM kinase and one or both L27 domains, and thus have close amino acid sequence and domain organization similarity to the P55 subgroup (Fig. 2A). Second,it is apparent that zebrafish Humpback(Konig et al., 1999) also defines a previously unrecognized subgroup of MAGUKs that is most similar to the Sdt subgroup (Fig. 2A). Third, some MAGUK subgroups, such as the Vari/PALS2 and Lin-2/Cask, have characteristic sequences at their C-termini that could be PDZ-binding motifs,while members of other subgroups, such as the Sdt/PALS1 and Humpback families,do not have conserved C-termini and lack potential PDZ-binding motifs(Fig. 2B). Fourth, although the Dlg, Sdt, ZO-1 and Lin-2 subgroups have clear representatives in vertebrates, Drosophila and C. elegans, the Vari and Humpback subgroups do not appear to have C. elegans members. Further, C. elegans DLG lacks the conserved C-terminal amino acids present in Drosophila and vertebrate Dlgs, and the C. elegans ZO-1 C-terminus is also considerably divergent from the Drosophila and vertebrate ZO-1 C-termini (Fig. 2B). Together, these observations show that while some MAGUK subgroups have been strongly conserved, other subgroups are diverging. A practical consequence of this divergence is that Drosophila is likely to be more representative than C. elegans as a model system for investigating the roles of MAGUKs in epithelial cell junctions.

To begin investigating the functions of Vari, we determined the tissue and subcellular distribution of Vari using antibodies raised to domains common to all Vari isoforms. Vari protein was predominantly expressed in the hindgut and trachea starting at stage 14 (Fig. 4A), but was also clearly expressed in the dorsal epidermis by stage 12 (data not shown). This expression pattern corresponds to the RNA expression pattern of CG9326 determined by the Berkeley Drosophila Genome Project(Tomancak et al., 2002). By stages 15 and 16, when SJ junction assembly has been completed, Vari co-localized with the canonical SJ markers Cor and Nrx(Baumgartner et al., 1996; Fehon et al., 1994) in the trachea, hindgut, salivary gland and epidermis(Fig. 1M,N; Fig. 3A-C; Fig. 4B; and data not shown). That Vari localizes to SJs is important, because the mouse protein most closely related to Vari is PALS2, and like Vari, PALS2 localizes to the basolateral region of epithelial cells(Kamberov et al., 2000). Thus,Vari and PALS2 extend the similarity between Drosophila SJs and vertebrate basolateral regions first evidenced by the similar localizations of Scrib, Dlg and Lgl in both vertebrates and Drosophila(Knust and Bossinger,2002).

To determine if Vari organizes SJs, we tested all vari mutants for SJ barrier function and examined the subcellular localization of five SJ components in three epithelial tissues of three different varimutants: an intermediate allele vari^3953b, a strong allele vari³²⁷ and a putative null allele vari^F033 (mutants described below). The dye exclusion assay of Lamb et al. (Lamb et al.,1998) showed that all vari mutations except the semi-viable vari^38EFa2 mutation caused SJ barrier defects(Table 1). As is typical of SJ mutants, in animals homozygous for the strong or null alleles of varithe SJ components Cor, Nrx, Sinuous (Sinu) and the Na⁺/K⁺-ATPase were all mislocalized basally in the trachea, hindgut and salivary glands (Fig. 1M,N; Fig. 3F,F′,I,I′; and data not shown). However, although Dlg levels were greatly reduced in strong and null vari mutants, Dlg was nonetheless localized correctly (Fig. 3K,K′). In vari^3953b, Dlg and the Na⁺/K⁺-ATPase β-subunit Nrv2 were correctly localized, but Cor and Nrx were mislocalized basally(Fig. 3G,G′,L,L′and data not shown). By contrast to the tissue-specific effects of sinu mutations (Wu et al.,2004), but like mutations in the Na⁺/K⁺-ATPase (Paul et al., 2003), the SJs of all tissues examined were similarly affected by vari mutations.

We next investigated whether Vari was required for assembly or maintenance of septate junctions by following the subcellular localization of the canonical SJ marker Cor (Fig. 4A-F) during embryonic SJ assembly. In wild-type animals, Cor predominantly localizes to the SJ at stage 14. As SJs mature and develop barrier function during stages 15 and 16(Paul et al., 2003), the amount of Cor localized to the SJs progressively increases(Fig. 4A-C). Some increase in the amount of Cor localized to the basolateral and basal membrane surfaces is also observed. By contrast, in vari^F033 null mutants,specific localization of Cor to the SJ region was not observed at any stage. At stage 14 Cor localized fairly uniformly to the basolateral and basal surfaces and a significant amount of cytoplasmic staining was also seen(Fig. 4D). At stages 15 and 16,little cytoplasmic staining was observed, but Cor still had a basolateral distribution, and abnormal basal accumulations of Cor became apparent(Fig. 4E,F). Thus, in the absence of Vari, SJs did not begin to assemble, indicating that Vari is required for SJ formation rather than maintenance.

To investigate if Vari had functions typical of other SJ components expressed after establishment of apical-basal polarity, we asked whether Vari was required for epithelial apical-basal polarity or for the recently identified apical secretory function of SJs(Wang et al., 2006). As is the case for late SJ components such as Sinu and Lachesin (Lac), Vari is not necessary for AJ formation or for establishing apical-basal polarity, because the AJ markers Armadillo/β-catenin (Arm) and DE-Cadherin (E-cad; Shotgun- Flybase), and the apical markers DPatJ (PatJ) and Veli were localized properly in all vari mutants (Fig. 3P,R and data not shown). Similarly, and as reported for other SJ mutants (Wang et al., 2006),the lumenal levels of the matrix-associated protein Verm were reduced and punctate cytoplasmic accumulations of Verm were frequently observed in vari mutants (Fig. 5B-D). Verm cytoplasmic staining was particularly strong and penetrant in vari^3953b mutants compared with varinull and other SJ mutants, although the expressivity of the cytoplasmic accumulations varied considerably (Fig. 5C,D). Interestingly, the Verm-related protein Serp behaved differently from Verm in vari, cor or Lac mutants, because no lumenal or cytoplasmic staining of Serp was observed(Fig. 5G,H and data not shown). Together, these results show that Vari localizes to SJs and is required for their organization and function.

Because Vari is a scaffolding protein involved in SJ assembly, we determined the extent to which Vari localization depends on other SJ components. In nrx⁴⁸⁴⁶, sinu^nwu7,cor⁵ and nrv2^23b null mutants, Vari levels were greatly reduced and the remaining Vari protein was mislocalized basally(Fig. 3M,M′,N,N′and data not shown). Thus, as for all other SJ components examined to date,there is an interdependence between Vari and other SJ components for subcellular localization and SJ assembly.

If Vari and PALS2 share functional as well as sequence similarity, one would expect them to interact with similar proteins. As the PDZ domain of PALS2 binds to the C-terminal PDZ-binding motif EYFI of the basolateral cell adhesion protein Necl-2 (Shingai et al.,2003), we investigated whether Vari's PDZ domain binds the SJ component Nrx, a basolateral cell-adhesion protein with a C-terminus that ends in the related sequence EIFI (Baumgartner et al., 1996). In a yeast two-hybrid screen using the cytoplasmic 48 amino acids of Nrx as bait, we recovered 15 positive clones from 2×10⁶ colonies (Bhat et al., 1999). Two of these clones contained cDNAs encoding Vari,seven encoded the multi-PDZ-domain protein dPATJ, and the remaining six encoded proteins did not bear recognizable binding motifs (Materials and methods). The recovery of only Vari and dPATJ, but not others of the more than 125 PDZ-binding proteins in the Drosophila genome, suggests that the interaction between the Nrx C-terminus and Vari is quite specific.

To confirm the yeast two-hybrid results, we performed pull-down assays using either the short isoform of Vari or the Vari PDZ-domain fused to GST(see Materials and methods). Both Vari fusion constructs, but not GST alone,could precipitate purified Nrx C-terminus, demonstrating that the Vari PDZ-domain can directly bind Nrx (Fig. 6A). Consistent with Vari and Nrx also interacting directly in vivo, both Vari fusion constructs, but not GST or GST fused to the Vari SH3 domain, precipitated Nrx from whole embryo lysates(Fig. 6B). Significantly, the apical protein Crbs was not precipitated by Vari or the Vari-PDZ fusions,despite Crbs having the C-terminal PDZ-binding motif ERLI, which closely resembles that of Nrx EIFI. Similarly, neither Vari nor the Vari-PDZ domain precipitated the SJ component Nrg, which has isoforms that also end in potential PDZ-binding motifs (ATYV or RKGL). These results indicate that there is significant binding specificity between the Vari PDZ-domain and the Nrx C-terminus. However, we also found that a subset of SJ components, including Cor and Nrv2, co-precipitated with Vari and Nrx and remained associated even in high salt washes (Fig. 6B). While these results are consistent with previous work by Genova and Fehon(Genova and Fehon, 2003)showing that Nrx, Cor and Nrv2 co-immunoprecipitate from embryo lysates and support the hypothesis that Vari scaffolds SJ complexes, these results also leave open the possibility that the in vivo interaction between Nrx and Vari could be indirect. Attempts to use double-mutant analysis to investigate the role(s) of interactions between Vari and Nrx have been unsuccessful, because we have been unable to establish lines of the double balanced heterozygotes using a variety of balancer chromosome combinations. Despite these caveats,Vari binding to Nrx in vitro parallels PALS2 binding to Necl2 in vitro, thus extending the similarity between Vari and PALS2, and between SJs and vertebrate basolateral regions.

To investigate the functions of the different domains of Vari, we characterized the original vari^3953b allele(Beitel and Krasnow, 2000) and the vari^38EFa2 and vari^F033 alleles(Butler et al., 2001; Thibault et al., 2004), as well as five vari mutations that we generated in an ethylmethanesulfonate (EMS) non-complementation screen(Fig. 1; Table 1). vari^3953b has a 17 bp deletion in the intron after the second common exon and is an intermediate allele that is viable in certain trans-heterozygous combinations (Table 2). In vari^3953b homozygotes, Vari protein does not significantly accumulate at the SJ region and frequently has a tracheal phenotype similar to that caused by a vari null mutation(Fig. 3P,Q; Table 1). vari^F033 appears to be a null mutation that results from a transposable element insertion into the first common intronic region(Fig. 1K). vari^F033 behaves as a deficiency in genetic crosses, and in vari^F033 mutants Vari protein levels are not distinguishable from background (Fig. 1M; Fig. 3O; Table 1; Table 2; and data not shown).

Surprisingly, all four EMS mutations that cause amino acid changes affect the HOOK domain and are, to our knowledge, the first genetic mutations in a MAGUK HOOK domain (Table 1). The HOOK domain is a unique feature of MAGUK-type SH3 domains and is an unusually long loop interposed between the fourth and fifth beta strands of the standard SH3 domain (McGee et al.,2001; Tavares et al.,2001). This loop can interact with proteins such as the FERM-domain band 4.1 protein and may enable MAGUKs to homo- or hetero-multimerize via interfolded SH3 domains. Animals homozygous for the Vari HOOK domain mutations have severe tracheal defects resembling those of the vari^F033 null mutants and lack detectable accumulations of Vari at the SJs. Whether total Vari protein levels are reduced by these mutations is unclear, as immunohistochemical staining using the current anti-Vari antibodies produces a variable background. This variability makes it difficult to distinguish between the Vari protein levels appearing to be reduced because Vari is mislocalized to the cytoplasm and across the basolateral membrane, as is seen in SJ mutants such as nrv2 (Fig. 3, compare N,N′ with S), or because Vari is destabilized and degraded in the HOOK-domain mutants. Attempts to assess Vari protein levels directly using western blotting have been unsuccessful with the current antibodies. Regardless, the strong tracheal phenotypes of the HOOK-domain mutants indicate that the HOOK domain has a crucial role for Vari function.

We also investigated the role of the L27 protein-protein interaction domain in Vari function. L27 domains are distinguishing features of several MAGUK subgroups that in vitro evidence suggests are important mediators of MAGUK scaffolding functions (e.g. Kamberov et al., 2000; Tseng et al.,2001) (reviewed by Funke et al., 2005). However, in only a few (e.g. Nakagawa et al., 2004) cases has the in vivo importance of L27-mediated interactions been confirmed. For Vari, a two-hybrid screen of the Drosophila proteome detected a significant interaction between Vari and Veli(Giot et al., 2003), which paralleled the L27-mediated in vitro interactions between Pals2/Vam-1 and vertebrate Veli (Kamberov et al.,2000; Tseng et al.,2001). Despite this corroborative evidence, the functional importance of the Vari-Veli interaction in epithelial cells was suspect,because Veli localization is unaffected by vari mutations and because the subcellular distributions of Vari and Veli do not overlap(Fig. 3O). Veli localizes apically, while Vari localizes to the basolateral SJs. We confirmed that the Vari L27 domain is nonessential by showing that the short isoform of Vari,which lacks the L27 domain, can completely rescue the tracheal and lethal phenotypes of vari null mutants(Fig. 1G,J) and restore normal SJ localization of both Cor and Vari (Fig. 4G,I). Further, despite yeast two-hybrid data indicating that the Vari L27 domain interacts with Veli, expression of the long isoform of Vari rescued vari null mutants to viability, and in rescued animals the L27-containing Vari-long isoform localized to SJs rather than the apical domain where Veli is localized (Fig. 4J-L). Together, these results show that although the HOOK region of Vari has an essential role, the L27 domain is dispensable.

Vari was originally identified as a gene required for regulating the size of epithelial tubes. In vari mutants, tracheal tubes become too long without changes in tracheal cell number(Beitel and Krasnow, 2000). Here we show that Vari encodes multiple isoforms of a MAGUK that helps define a new subgroup of MAGUKs. Vari functions in the assembly of the septate junctions and is required for the apical secretion of the protein Verm, which is thought to be responsible for modifying a chitin-based lumenal matrix(Luschnig et al., 2006; Wang et al., 2006). In vari and other SJ mutants, Verm is not secreted, the lumenal matrix becomes abnormal and tracheal tubes become elongated.

The protein-protein interaction domains present in Vari suggest it acts as a scaffolding protein that helps bring together different components of the SJ complex. This hypothesis is supported by our GST-pull down assay results showing Vari's PDZ domain can directly bind the intracellular domain of Nrx, a transmembrane SJ adhesion protein. Binding of the Vari PDZ domain to Nrx would leave Vari's SH3, GUK and predicted C-terminal PDZ-binding motif available to anchor other SJ components to the membrane, or to bring together different transmembrane SJ components. One model is that Vari may help bring the Dlg-Scib complex to the membrane through interfolding of the Vari and Dlg SH3 domains, which is made possible by the unique HOOK domain insert in the MAGUK SH3 domains (McGee et al.,2001; Tavares et al.,2001). Whether or not Vari anchors the Dlg complex to the rest of the SJ, genetic evidence indicates that Vari has functions beyond simply bridging between transmembrane Nrx and intracellular SJ complexes, because vari mutations can strongly enhance the phenotypes caused by mutations in the Drosophila claudin sinuous, whereas nrx mutations do not enhance sinuous mutations(Wu et al., 2004).

By itself, the finding that Vari encodes a MAGUK was not unexpected, as many MAGUKs are associated with cell-cell junctions (reviewed by Harris and Lim, 2001). However, it is significant that Vari helps define a new subgroup of MAGUKs that includes mammalian PALS2, because Vari and PALS2 both localize basolaterally and bind the C-termini of basolateral cell adhesion proteins. Thus, Vari and PALS2 bolster the similarity between Drosophila and vertebrate epithelial basolateral regions that was first evidenced by the common basolateral localization of the Scrib, Dlg and Lgl early polarity proteins. However, by contrast to the polarity proteins, Vari is not required for cell polarity but rather is expressed late in embryonic development and is part of a cell-adhesion complex. Thus, Vari fundamentally extends the similarity of Drosophila and vertebrate basolateral regions from containing only conserved polarity complexes to containing both conserved polarity and cell-adhesion complexes.

The finding of more extensive similarity between SJ and vertebrate basolateral regions suggests that continued study of Drosophila SJs will provide insight into vertebrate epithelial basolateral regions. Further,these results support the idea that during evolution there has been conservation of different junctional functions, such as forming paracellular barriers and anchoring of polarity complexes. However, the comparison of TJs and SJs also makes it clear that there has been limited conservation of which particular functions have assorted to different junctions. An attractive explanation for these somewhat contradictory observations is that junctional functions are modular, and that the disparate junctions in different species represent alternative combinations of functional modules. For example, Drosophila SJs could be considered a combination of the claudin-based paracellular-barrier function and the basolateral polarity proteins Dlg, Scrib and Lgl. Alternatively, vertebrate TJs could be considered a combination of the claudin-based paracellular-barrier function and the apical polarity complexes of Crbs-Baz and SdtaPKC-Par-6. Thus, when comparing junctions between species, it is likely to be more useful to compare specific junctional functions, such as molecular details of polarity or barrier functions, than to attempt to directly compare junctions in their entirety.

If complex junctions such as TJs and SJs are comprised of functional modules, one would expect that these junctions should contain distinct molecular subcomplexes that mediate distinct functions. Consistent with this proposal, extensive work by many labs has shown that the polarity proteins of Crb-Sdt and Baz-cdc42-aPKC form specific complexes (reviewed by Margolis and Borg, 2005). Claudin proteins appear to be part of a `barrier complex' because claudins are required for and co-localize with the paracellular barrier in both Drosophila and vertebrates. Functional demonstration of the independence of the barrier and polarity complexes in both species is provided by the observations that cell polarity is not affected by selective disruption of the barrier complex in either mammals by knockdown of ZO-1 and ZO-2(Umeda et al., 2006), or in Drosophila by mutations in claudin genes(Behr et al., 2003; Wu et al., 2004). The Vari/PALS2 proteins could play a pivotal role in allowing cytoplasmic subcomplexes to associate different adhesion-junctional complexes, either in different cell types or during evolution, because changing which adhesion complex Vari or PALS2 associate with could be as simple as changing the four amino acid PDZ-binding motifs of one or a few transmembrane proteins. It seems likely that evolving a few unstructured amino acids would be significantly easier than evolving three-dimensional binding surfaces. Thus, Vari and its homologs could provide crucial - but malleable - links between conserved intracellular complexes and the divergent transmembrane junctional complexes found across the animal kingdom.

We are grateful to M. Ternet for initial mapping of varimutations, C. Gottardi, I. T. Helenius and members of the Beitel and Bhat labs for comments on the manuscript, to the Bloomington Stock Center, the Developmental Studies Hybridoma Bank, the Drosophila Genome Resource Center and many members of the Drosophila community for fly stocks, antibodies and cDNAs. Additional thanks to R. Engen, A. Graff, M. Kelaita, E. Klostermann, R. Lehotzky, H. Patel and M. VanGompel for isolating alleles, and W. Russin of the Northwestern Biological Imaging Facility for assistance with confocal imaging. S.M.P. was supported by NIH Lung Biology training grant 5 T32 HL076139-03. M.A.B. was supported by NIH GM63047 and NS50356 and funds from the state of North Carolina. G.J.B. was a recipient of a Burroughs Wellcome Fund Career Award in the Biomedical Sciences, NSF Career Award IBN-0133411 and NIH R01 GM069540.