Dynamic analysis identifies novel roles for DLG-1 subdomains in AJM-1 recruitment and LET-413-dependent apical focusing.

Cell-cell junctions are composed of a diverse array of specialized proteins that are necessary for the movement and integrity of epithelia. Scaffolding molecules, such as membrane-associated guanylate kinases (MAGUKs) contain multiple protein-protein interaction domains that integrate these proteins into macromolecular complexes at junctions. We have used structure-function experiments to dissect the role of domains of the Caenorhabditis elegans MAGUK DLG-1, a homolog of Drosophila Discs large and vertebrate SAP97. DLG-1 deletion constructs were analyzed in directed yeast two-hybrid tests as well as in vivo in a dlg-1 null mutant background. Our studies identify novel roles for several key domains. First, the L27 domain of DLG-1 mediates the physical interaction of DLG-1 with its binding partner, AJM-1, as well as DLG-1 multimerization. Second, the PDZ domains of DLG-1 mediate its association with the junction. Third, using dynamic in vivo imaging, we demonstrate that the SH3 domain is required for rapid lateral distribution of DLG-1 via a LET-413/Scribble-dependent pathway. Finally, we found that inclusion of the SH3 domain can ameliorate dlg-1 mutant phenotypes, but full rescue of lethality required the complete C terminus, which includes the GUK and Hook domains, thereby demonstrating the importance of the C-terminus for DLG-1 function. Our results represent the first in vivo analysis of requirements for the L27 domain of a Discs-large/SAP97 protein, identify a crucial LET-413/Scribble regulatory motif and provide insight into how MAGUK subdomains function to maintain epithelial integrity during development.


Introduction
During development, epithelial tissues undergo finely orchestrated morphogenetic movements to create organized epithelial sheets that protect the organism from its external and internal environments. Apical cell-cell junctions are crucial for these epithelial functions in several ways. First, they mediate strong adhesive linkages between adjacent epithelial cells, and they form connections to the cytoskeleton, which are essential for the transmission of forces during development (Goodwin and Yap, 2004). Second, junctions provide a paracellular permeability barrier that regulates diffusion of solutes between epithelial cells (Anderson et al., 2004). Third, junctions are essential in establishing and maintaining cell polarity, which allows epithelial cells to specialize their apical and basolateral surfaces (Shin et al., 2006). These three functions are distributed between specialized junctional complexes. The bulk of cell-cell adhesion is performed by the adherens junction (Goodwin and Yap, 2004;Kobielak and Fuchs, 2004), whereas tight junctions in vertebrates (Aijaz et al., 2006;Shin et al., 2006) and septate junctions in invertebrates (Tepass et al., 2001) maintain cell polarity and regulate intercellular diffusion (Anderson et al., 2004).
Scaffolding molecules are required to integrate a wide variety of proteins into junctions (Funke et al., 2005). The presence of multiple protein-protein interaction domains enables scaffolding molecules to bind several proteins simultaneously and link them together into functional complexes (Caruana, 2002). Thus, scaffolding molecules provide the 'molecular glue' that binds junctional proteins together into a functional network.
One group of scaffolding molecules central to the organization of epithelial junctions is the MAGUK (membrane-associated guanylate kinase) family. MAGUKs contain several conserved protein-protein interaction domains, including a non-enzymatic GUK (guanylate kinase) domain, a SH3 (Src homology 3) domain and multiple PDZ [PSD95 (DLG4), Discs large (DLG), ZO-1] motifs (Funke et al., 2005). Some MAGUKs, including DLG/SAP97 and CASK, also contain one or more L27 (LIN-2 AND LIN-7) domains, which have been shown to mediate heterodimerization between proteins containing the two L27-domain subtypes (Harris et al., 2002;Feng et al., 2004;Petrosky et al., 2005). MAGUKs have been demonstrated to play a number of essential roles in junction formation and function. Mutants of the Drosophila MAGUKs Stardust (SDT) or DLG exhibit disrupted adherens or septate junctions, respectively, leading to an overall loss of cell polarity in both cases (Perrimon, 1988;Tepass et al., 2001). In vertebrate systems, the MAGUK ZO-1 plays important roles at the tight junction via regulation of junction assembly (Umeda et al., 2006) and by providing connections to the actin cytoskeleton (Fanning et al., 2002). Understanding how MAGUKs organize protein complexes is therefore essential for providing insight into how epithelial junctions function.
The DLG subgroup of MAGUK proteins have been shown to perform essential roles during development in multiple organisms (Perrimon, 1988;Bossinger et al., 2001;Caruana and Bernstein, 2001). Structure-function studies of DLG homologs in different contexts have revealed surprisingly disparate requirements for the conserved MAGUK domains for their localization and function. For example, in Drosophila epithelia, the DLG Hook (Hk) region -located between the SH3 and GUK domains -as well as the PDZ2 domain are required for septate junction localization (Hough et al., 1997), and the SH3, PDZ2 and PDZ3 domains are required to rescue Dlg epithelial-polarity and cell-proliferation defects, respectively (Hough et al., 1997). By contrast, DLG requires the GUK domain, PDZ1 and PDZ2 for localization to synapses (Thomas et al., 2000). The N-terminal region of DLG, containing the L27 domain, has only recently been described in some isoforms of Drosophila DLG that are expressed in the CNS and at neuromuscular junctions (Mendoza et al., 2003). Little is known about the requirement of the L27 domain in those tissues.
In contrast to Drosophila, the PDZ, SH3 and GUK domains are all dispensable for SAP97 localization to the lateral membrane in vertebrates. Instead, the conserved N-terminus of SAP97 is sufficient for localization, and the PDZ1 and PDZ2 domains provide increased localization efficiency (Wu et al., 1998). This N-terminal targeting probably occurs via L27-domain hetero-dimerization between SAP97 and mouse CASK (Karnak et al., 2002;Lee et al., 2002;Sanford et al., 2004). Determining the function of DLG homologs in vertebrates is somewhat difficult, however, because of redundancy between the four DLG homologs [PSD95, PSD93 (DLG2), SAP102 and SAP97] (Funke et al., 2005), which is demonstrated by the relatively weak phenotypes displayed by knockout mice for DLG homologs (McGee et al., 2001;Migaud et al., 1998;Caruana and Bernstein, 2001).
The Caenorhabditis elegans homolog of DLG, DLG-1, provides an excellent model for the function of this MAGUK subclass. First, DLG-1 is the sole homolog of DLG, eliminating the possibility of closely related redundant molecules (Bossinger et al., 2001). Second, DLG-1 possesses one primary isoform that contains all of the conserved domains, thus simplifying structure-function studies . Finally, GFP-tagged dlg-1 transgenes can be imaged in vivo to analyze the dynamics of DLG-1 function (Köppen et al., 2001). In C. elegans, DLG-1 is essential for organizing the AJM-1-DLG-1 complex, which is localized adjacent and basal to the cadherin-catenin complex (Köppen et al., 2001;McMahon et al., 2001). Loss of either DLG-1 or AJM-1 results in developmental arrest during embryonic elongation, a process whereby the embryo is squeezed into a worm-like shape (Ding et al., 2004).
Initial structure-function experiments using C-terminally truncated DLG-1 constructs suggested that the N-terminus of DLG-1 is sufficient for junctional localization ). However, a major limitation of these experiments is that they were performed in the presence of endogenous DLG-1, making it impossible to assess the functional importance of DLG-1 subdomains during development. Moreover, the presence of endogenous DLG-1 could potentially alter localization results. To analyze the functional DLG domains dynamically in a living organism, we have undertaken structure-function experiments in a dlg-1 null mutant. Our results provide the first in vivo evidence for the functional importance of the L27 domain of a DLG protein in epithelia. We also clarify roles for the N-terminus and PDZ domains of DLG-1 during junctional localization and AJM-1 recruitment, and show that the GUK domain is required for viability. Finally, we identify a role for the SH3 domain, acting via LET-413/Scribble, in mediating rapid apical 'focusing' of DLG-1.

Results
The L27 domain is necessary for the AJM-1-DLG-1 interaction and for DLG-1 multimerization Previous work from Köppen et al. (Köppen et al., 2001) showed a direct physical interaction between the N-terminal half of DLG-1 (amino acids 1-483) and a portion of the coiled-coil region of AJM-1 (amino acids 180-809). We set out to further map the domains involved in this interaction via directed two-hybrid tests in yeast. We constructed a series of dlg-1 deletion constructs, removing portions of the N-terminus and PDZ regions of dlg-1 fused to the LexA DNA-binding domain (Fig. 1A). These constructs were cotransformed into yeast with constructs containing the ajm-1 coiledcoil domain (encoding amino acids 180-809) fused to the Gal4 activation domain and then assayed for physical interaction of the proteins. Removal of the first PDZ domain (amino acids 200-290), the flanking N-terminal sequence (amino acids  or the Cterminal (amino acids 298-354) sequence did not affect the AJM-1-DLG-1 interaction (Fig. 1A,B). However, constructs deleting Journal of Cell Science 121 (9) Fig. 1. The L27 domain of DLG-1 is necessary for the AJM-1-DLG-1 physical interaction and for DLG-1 multimerization. (A) Schematic showing the structure of the DLG-1 deletion proteins fused to the LexA DNA-binding domain. Constructs were co-transformed with a Gal4-ajm-1 (encoding amino acids 180-809) fusion. Deletions of the L27 domain abrogated the AJM-1-DLG-1 interaction. DLG-1 multimerization was also assayed by transformation of yeast with LexA-dlg-1 deletion constructs and a Gal4-dlg-1 fusion. Physical interaction was observed in constructs containing the Nterminus and the first two PDZ domains of DLG-1; this interaction was disrupted by removal of the L27 domain. (B) lacZ activity of yeast cotransformed with LexA-dlg-1 deletion constructs and Gal4-ajm-1. Numbers on the streaks correspond to the bait constructs from A. lacZ activity was detected in all streaks except 2 and 3. (C) lacZ activity of DLG-1 multimerization. Yeast were co-transformed with Gal4-dlg-1 and LexA-dlg-1 deletion constructs. lacZ activity was observed in streaks 1, 4, 5 and 6. Both streaks 2 and 3 lacked detectable activity. portions of the L27 domain (amino acids 17-45 and amino acids 53-116) both failed to interact with AJM-1 (Fig. 1A,B). These results suggest that the L27 domain is necessary for DLG-1 to bind AJM-1, whereas the remaining N-terminus and first PDZ domain are dispensable for this interaction.
Previous experiments have demonstrated the capability of human DLG to form homotypic multimers in vitro (Marfatia et al., 2000). Oligomerization was found with both full-length human DLG and with the first 229 N-terminal amino acids, suggesting that the Nterminus is sufficient for multimerization. To determine whether C. elegans DLG-1 is capable of forming multimers, we assayed for homotypic interactions between fragments of DLG-1 via the yeast two-hybrid system. We co-transformed yeast with the N-terminus of dlg-1 fused to the Gal4 activation domain and various dlg-1 fragments fused to the LexA binding domain (Fig. 1A). We found that DLG-1 was indeed capable of self-interaction, and, surprisingly, deletion constructs showed a pattern of interaction similar to that required for AJM-1-DLG-1 binding (Fig. 1A,C). Specifically, constructs lacking the PDZ1 domain and surrounding regions were able to interact with the N-terminal half of DLG-1 (amino acids 1-483), whereas deletions that disrupted the L27 domain failed to interact. These data suggest that DLG-1 is capable of forming multimers via an L27-domain-mediated interaction.
In order to assay the function of DLG-1 deletion constructs in a strong dlg-1 mutant background, we characterized the dlg-1(ok318) allele. Sequencing of dlg-1(ok318) revealed a pair of deletions. The first is a 109-basepair (bp) deletion spanning the third exon-intron boundary; the second, larger, deletion of 1368 bp has endpoints at the beginning of exon 8 and near the end of exon 10 ( Fig. 2A). The first deletion is predicted to cause a frameshift and produce a product containing only the first 153 amino acids of DLG-1. dlg-1(ok318) is 100% zygotic lethal (Table 1) and is phenotypically similar to dlg-1(RNAi), with embryos terminating at the 1.5-to 2-fold stage of elongation and forming fluid-filled vacuoles (Fig. 2E). Immunofluorescence using anti-AJM-1 antibodies revealed disruption of the AJM-1-DLG-1 domain in mutant embryos (Fig.  2G). Immunostaining for DLG-1 protein using an antibody raised against the PDZ domains of Drosophila DLG failed to detect any protein in homozygous mutants (Fig. 2F), whereas staining in heterozygotes was unaffected. Thus, dlg-1(ok318) acts as a strong loss-of-function allele that is useful for assessing the in vivo function of DLG-1 deletion constructs.
The L27 domain and PDZ domains are sufficient for DLG-1 junctional association in the absence of wild-type DLG-1 Firestein and Rongo ) created a series of GFP-tagged C-terminal DLG-1 deletion constructs and demonstrated that the conserved N-terminus of DLG-1 was able to localize in a wild-type background. However, given that N-terminal fragments of DLG-1 can interact with both full-length DLG-1 and AJM-1, the presence of wild-type DLG-1 in these experiments is a significant confounding factor. We therefore set out to further elucidate the function of conserved DLG-1 domains by analyzing the constructs in a dlg-1(ok318) mutant background. Transgenic constructs were maintained as extra-chromosomal arrays by injecting GFP-tagged dlg-1 deletion constructs ( Fig. 3A) into dlg-1(ok318)/dpy-8(q287) heterozygotes. We then assayed whether the resulting DLG-1 fragments were able to localize in dlg-1(ok318) homozygotes. Significantly, unlike in a wildtype background (Fig. 3B, supplementary material Reslicing through the z-axis identified the defect as a failure in lateral distribution of DLG-1, because the GFP was focused to the proper apical-basal region (data not shown). Further imaging of the remaining constructs in arrested embryos demonstrated that the N-terminus and the first PDZ domain are sufficient for DLG-1 association with the junction in the absence of wild-type DLG-1. With the exception of DLG-1(1-186)::GFP, Journal of Cell Science 121 (9)  all proteins showed strong junctional localization in dlg-1(ok318) homozygotes (Fig. 3G,K, supplementary material Movies 2A,B). Thus, we found that the N-terminus of DLG-1 is insufficient for localization in the absence of wild-type DLG-1. Instead, our results suggest that, in the absence of N-terminal binding by either wildtype DLG-1 or AJM-1, the PDZ domains of DLG-1 are required for its association with the membrane. To determine whether the L27 domain is also required for DLG-1 localization, we analyzed the localization of a DLG-1 fragment that lacked the L27 domain but retained the remaining DLG-1 domains [DLG-1(Δ17-116)::GFP]. This fragment localized to the apical junction of epithelia ( Fig. 4D and Fig. 5J, and supplementary material Movie 6), suggesting that the PDZ motifs are sufficient for DLG-1 junctional association.
The N-terminus of DLG-1 is incapable of mediating lateral distribution, as demonstrated by the punctate DLG-1(1-186)::GFP localization in dlg-1(ok318) homozygotes (Fig. 4A). Conversely, the SH3 domain was capable of restoring DLG-1 accumulation rates to wild-type levels. DLG-1(1-710)::GFP was localized into a continuous junction before the beginning of elongation in dlg-1 mutants (Fig. 4C, supplementary material Movie 4B) and wild-type embryos (Fig. 3J, supplementary material Movie 4A), and its localization was indistinguishable from that of full-length DLG-1::GFP (data not shown). Thus, we infer, from dynamic localization of DLG-1 deletion constructs, that the PDZ domains of DLG-1 are capable of mediating association with the junction but cannot confer wild-type rates of lateral distribution along the junction. This latter function requires the SH3 domain. Our results also demonstrate the dispensability of the GUK and Hk domains for DLG-1 localization, because localization of constructs that terminate after the SH3 domain was indistinguishable from that of full-length DLG-1::GFP. Based on these results, we performed a final set of experiments to test whether the N-terminus of DLG-1, which contains the L27 domain, is dispensable for localization if the rest of the protein is intact. As shown in Fig. 4D (also see supplementary material Movie 6), DLG-1(Δ17-116), which lacks the L27 domain, is capable of localizing in a dlg-1(ok318) mutant background. Taken together, these results show that the L27 domain is not absolutely required for DLG-1 localization.
The delayed rate of localization observed in constructs truncated C-terminally to the PDZ domains was similar to the phenotype observed in let-413 mutants (Köppen et al., 2001). LET-413 is the C. elegans homolog of Drosophila Scribble and is expressed along the basolateral membranes of epithelial cells (Legouis et al., 2000). Loss of let-413 function causes disruption of DLG-1 and AJM-1 localization (McMahon et al., 2001). Previous dynamic analyses of let-413 mutants from our laboratory revealed a failure of apicalbasal 'focusing' of junctional proteins during early stages of junction formation (Köppen et al., 2001). To determine whether the delayed rate of localization observed for DLG-1 constructs lacking the SH3 domain was due to a LET-413-dependent pathway, we assayed the effect of let-413 loss-of-function on the localization of DLG-1::GFP deletion constructs. Significantly, we found that the localization of DLG-1(1-710)::GFP, which contains the SH3 Journal of Cell Science 121 (9) domain, in let-413(RNAi) embryos (Fig. 4E, supplementary material Movie 5A) resembled the mislocalization of DLG-1(1-468)::GFP in dlg-1(ok318) mutants (Fig. 4B). Moreover, the rate of localization of DLG-1(1-468)::GFP in dlg-1(ok318);let-413(RNAi) embryos (Fig. 4F, supplementary material Movie 5B) was similar to that observed in dlg-1(ok318) embryos (Fig. 4B). These results strongly suggest that the delayed localization rate observed for DLG-1 deletion constructs lacking the SH3 domain is the result of a failure to interact with a LET-413-dependent localization pathway.

AJM-1 localization is rescued by the L27 domain and the first two PDZ domains
Because the conserved N-terminus and PDZ domains were sufficient for junctional DLG-1 localization and interaction with AJM-1 in a yeast two-hybrid assay, we assessed AJM-1 localization in deletionconstruct backgrounds to determine what domains of DLG-1 were required to restore AJM-1 localization in dlg-1(ok318) mutants. Embryos expressing the deletion-construct arrays were fixed and stained with anti-AJM-1 antibodies to visualize AJM-1. Consistent with its own pattern of mislocalization in dlg-1 mutants, DLG-1(1-186)::GFP was unable to rescue AJM-1 localization. Instead, both the GFP (Fig. 5A) and AJM-1 (Fig. 5B) colocalized (Fig. 5C) into puncta along the junctional belt in dlg-1(ok318) embryos.
Fragments that truncated DLG-1 after the PDZ domains were junctionally localized in arrested mutants (Fig. 5D,G) and showed normal AJM-1 distribution (Fig. 5E,H). By contrast, although the deletion fragment removing only the L27 domain but possessing C-terminal regions was localized along the junction, it failed to rescue AJM-1 localization (Fig. 5J-L). These results demonstrate the requirements of the PDZ domains and the N-terminal L27 domain for junctional integration of DLG-1 and localization of AJM-1, respectively.
Significantly, we were able to rescue dlg-1(ok318) to viability via expression of full-length (FL) DLG-1::GFP (Table 1). dlg-1(ok318) mutants rescued by the full-length array appeared completely wild-type and were able to be maintained as homozygotes. Thus, the GUK domain and/or the entire Hk region are required for complete rescue of DLG-1 function during embryonic development.

Discussion
We report here the first in vivo analysis of the N-terminus of a DLG family member in a null background. In contrast to previous studies of DLG-1 and vertebrate SAP97, we found that the N-terminus is insufficient for DLG-1 localization in mutant backgrounds and instead functions to bind a downstream complex component, AJM-1. Dynamic imaging of DLG-1::GFP fragments showed that the PDZ and SH3 domains are required for normal DLG-1 localization. Specifically, the PDZ domains mediate DLG-1 membrane association, whereas the SH3 domain is required for rapid distribution of DLG-1 at the apical junction. Loss-of-function experiments suggested that this SH3-mediated pathway acts via LET-413, suggesting a novel function for the SH3 domain. Although fragments that restore localization partially rescued dlg-1 mutant phenotypes, viability was only restored with full-length DLG-1 constructs, suggesting that the C-terminus plays essential roles in DLG-1 function as well. A schematic based on our results is shown in Fig. 7; a related model diagram summarizing our results is provided in supplementary material Fig. S2.

The L27 domain binds AJM-1 and mediates DLG-1 multimerization
Previous studies of DLG family proteins in epithelia have not assessed the requirements for the L27 domain in vivo. We have shown for the first time that the L27 domain plays roles in mediating the AJM-1-DLG-1 physical interaction and probably in DLG-1 multimerization. Yeast two-hybrid assays using DLG-1 deletion constructs identified the necessity of the L27 domain for DLG-1 to interact with AJM-1, as well as for DLG-1 selfassociation. These interactions were further supported by our structure-function analysis in vivo. A DLG-1 deletion construct containing residues N-terminal to the first PDZ domain, DLG-1(1-186)::GFP, was mislocalized in dlg-1(ok318) and dlg-1(ok318);ajm-1(RNAi) backgrounds. Specifically, the GFP colocalized with AJM-1 puncta in dlg-1(ok318) homozygous mutants, whereas, in a double loss-of-function background, DLG-1(1-186)::GFP mislocalized entirely away from the junction into the cytoplasm. Localization of this fragment was unaffected in ajm-1(RNAi) embryos. This pattern of localization is consistent with the L27 domain of DLG-1 functioning to mediate both the AJM-1-DLG-1 physical interaction as well as DLG-1 multimerization (Fig. 7A). We hypothesize that, in the absence of wild-type DLG-1, the L27 domain of the DLG-1(1-186)::GFP construct binds to and colocalizes with AJM-1 protein, which is itself mislocalized into punctate aggregates (Fig.  7B). In embryos lacking only AJM-1, DLG-1(1-186)::GFP is properly localized, because it can bind wild-type DLG-1 at the junction (Fig. 7C). However, in dlg-1(ok318);ajm-1(RNAi) embryos, the DLG-1 construct lacks a binding partner and thus mislocalizes to the cytoplasm (Fig. 7D). Further support for a role of the L27 domain in AJM-1 localization comes from a deletion fragment lacking the L27 domain but retaining C-terminal regions. This fragment was sufficient for localization of DLG-1, but was unable to rescue AJM-1 localization in a dlg-1 mutant (Fig. 5J-L).
To date, L27 domains have been implicated in the formation of heterodimer pairs consisting of two different L27 sub-types, A and B, which are classified by the presence of key electrostatic residues (Feng et al., 2004;Petrosky et al., 2005). Our data suggest that a novel interaction occurs between the DLG-1 L27 domain and a coiled-coil protein, AJM-1. Interestingly, both coiled-coil and L27 domains adopt their helical structures only upon forming dimers (Harris et al., 2002), which suggests a possible mechanism for the AJM-1-DLG-1 interaction. The coiled-coil region of AJM-1 might act as a pseudo type-B L27 domain to complement the type-A L27 domain of DLG-1. We also found unexpected evidence for a homotypic interaction between the L27 domains of DLG-1 molecules. Because the L27 domain is thought to dimerize in a 1:1 stoichiometry (Harris et al., 2002), it is likely that the AJM-1-DLG-1 interaction and DLG-1 multimerization are mutually exclusive. This raises interesting questions: do the DLG-1 oligomers form only in the absence of AJM-1? Do both DLG-1-and AJM-1-bound pools of DLG-1 coexist in the cell? Additional experiments to determine the functional relevance of the DLG-1-DLG-1 interaction will help elucidate the answers to these questions.

The PDZ domains are necessary for junctional association
We have shown that constructs truncated after any of the PDZ domains are capable of localizing to the junction in dlg-1(ok318) mutants, but their rate of lateral distribution is retarded when compared with wild type. Although immunofluorescence experiments with deletion constructs suggest that constructs truncated after the three PDZ domains were well-distributed along the junction in terminal mutants, dynamic analysis reveals subtle defects. Using time-lapse confocal microscopy, we were able to demonstrate a significant difference in the rate at which the DLG-1 deletions formed a continuous junctional belt in dlg-1(ok318) mutants compared with wild-type embryos (Fig. 7E,F). A similar phenotype was seen in dlg-1(ok318);ajm-1(RNAi) double loss-offunction embryos, suggesting that lateral distribution of DLG-1 occurs independently of AJM-1. Because the membrane association of deletions truncated after the PDZ domains is independent of both wild-type DLG-1 and AJM-1, the first PDZ domain must be acting via a third protein that is capable of functioning as an adapter protein, associating either directly or indirectly with the cell membrane. In addition, we observed a decrease in the amount of cytoplasmic localization for constructs with multiple PDZ domains compared with the construct containing only one. These data suggest that the PDZ domains of DLG-1 function to increase the efficiency of junctional localization, similar to the function of PDZ domains in vertebrate SAP97.
Although DLG-1 fragments containing the PDZ domains mediated normal junctional localization of themselves and AJM-1 by the twofold stage of elongation, they failed to rescue the retarded elongation rate and lethality associated with dlg-1(ok318) embryos. Thus, proper lateral distribution of DLG-1 during the early stages of elongation is essential for viability and it is likely that the loss of DLG-1 protein at this crucial stage is responsible for the elongation defects observed in dlg-1(ok318) mutants.

The SH3 domain facilitates lateral distribution via a LET-413dependent pathway
For full rescue of dlg-1(ok318) lethality, we found that only the full-length DLG-1::GFP construct was sufficient. A DLG-1 deletion construct that lacked the GUK domain and part of the Hk region [DLG-1(1-710)::GFP] was unable to rescue dlg-1(ok318) mutants to viability, but was able to restore wild-type rates of junctional DLG-1 distribution (Fig. 7G) and partially ameliorated the severity of the dlg-1(ok318) mutant arrest phenotype. Thus, the C-terminus of DLG-1 is essential for its function, but is dispensable for localization.
The similarity between the delayed rate of full-length DLG::GFP localization in let-413 loss-of-function embryos and in the presence of deletions lacking the SH3 domain in dlg-1(ok318) mutants, as well as the lack of significant enhancement in dlg-1(ok318);let-413(RNAi) double mutants, suggests that the SH3 domain functions via a LET-413-dependent pathway (Fig. 7H-K). How LET-413 acts on the SH3 domain to modulate DLG-1 localization is unclear. In C. elegans, LET-413 and DLG-1 are thought to occupy different regions of the epithelial membrane, with DLG-1 localizing to the apical junction and LET-413 localizing along the basolateral membrane (Köppen et al., 2001;McMahon et al., 2001). Thus, LET-413 might act to regulate DLG-1 localization from an adjacent position at the membrane, perhaps by shuttling DLG-1 apically to its proper junctional position or by forming a complex at the interface between the AJM-1-DLG-1 and basolateral compartments. Another possibility is that the basolateral domain might extend into the AJM-1-DLG-1 region, thus causing LET-413 expression to overlap with DLG-1 at the apical junction. This overlap would necessarily be quite small, because no obvious colocalization is observed via light microscopy (Köppen et al., 2001;McMahon et al., 2001;our unpublished work). If DLG-1 and LET-413 do partially overlap, then LET-413 might positively affect DLG-1 localization via the formation of a macromolecular complex containing both proteins. This hypothesis is supported by genetic studies in Drosophila, in which Scribble and Discs large are mutually required to localize (Humbert et al., 2003;Tepass et al., 2001). Further experiments to finely resolve LET-413 and DLG-1 localization, and studies to determine whether they coexist in an in vivo molecular complex, will be instrumental in determining the mechanism of the LET-413/SH3-domain-mediated lateral distribution.

Morphogenetic roles for DLG-1
In addition to rescuing the rate of GFP localization, DLG-1(1-710)::GFP partially rescues the dlg-1(ok318) embryonic-arrest phenotype, and restores a wild-type rate of elongation. Because the DLG-1(1-710)::GFP fragment contains the domains necessary for localizing DLG-1 and AJM-1, it might rescue the AJM-1-specific functions of DLG-1. Both dlg-1 and ajm-1 loss-of-function results in similar phenotypes, with embryos arresting at the 1.5-to 2-fold stage of elongation. Therefore, AJM-1 and DLG-1 might be involved in the early stages of elongation, before muscle activity and attachment become important (Hresko et al., 1999). Because we were able to fully rescue the defects of dlg-1(ok318) with a fulllength DLG-1::GFP construct, the remaining lethality in partially rescued mutants probably arises from defects in an AJM-1independent pathway that acts via the DLG-1 GUK domain or Hk region and is necessary for later stages of elongation. Identification of additional DLG-1 binding partners that interact with specific DLG-1 subdomains will be essential for determining how DLG-1 and other MAGUKs carry out their functions during embryonic development.

Strains
Strain Bristol N2 was used as wild type and handled as described previously (Brenner, 1974). dpy-8(q287) was a gift from Judith Kimble (University of Wisconsin-Madison, WI).

Yeast two-hybrid
Directed yeast two-hybrid assays were performed by transforming dlg-1 deletion constructs fused to the LexA binding domain along with DNA encoding the coiledcoil region of AJM-1 (encoding amino acids 180-809) fused to the Gal4 activation domain. The vectors pBTMKnDB and pACT2 were used for the LexA and Gal4 domains, respectively. Transformed L40 yeast were selected on SD Trp -Leumedia, and physical interaction was assayed by growth on SD Trp -Leu -Hisplates and by β-galactosidase activity.
Immunostaining was visualized on a spinning disc confocal microscope consisting of a Yokogawa CSU10 spinning disc confocal scan head attached to a Nikon Eclipse E600 microscope. Data were collected using Perkin Elmer Ultraview software and a Hamamatsu ORCA-ER CCD camera.
RNAi dsRNA-mediated interference was performed by injection, as previously described (Walston et al., 2004). Single-stranded RNA was transcribed and purified using the Ambion MegaScript T3, T7 and SP6 kits, then dsRNA was obtained by incubating single-stranded mixtures at 70°C for 10 minutes and at 37°C for 30 minutes. RNA was injected at 2 μg/μl for single gene knockdown and 1 μg/μl per gene for double knockdown (2 μg/μl total). Partial cDNA clones were used as templates for the RNA transcription reaction as follows: yk25e5 (dlg-1), yk285a2 (ajm-1), and yk524b9 (let-413). All clones were a kind gift from Yuji Kohara (National Institute of Genetics, Japan).

Light microscopy
4D Nomarski microscopy was performed as follows. Embryos were collected from gravid adults and mounted on a 5% agar pad in M9 solution (Raich et al., 1999). Mounts were covered with a glass coverslip and sealed with a paraffin-vaseline-lanolin mixture. Embryos were filmed on a Nikon Eclipse E600 or Optishot-2 microscopes. Nomarski data were collected at 3-minute time points. Custom ImageJ plugins were then used to compress the 3D data into 4D movies (see supplementary material Movies 1-6). The plugins are available on request from J.H.

Live imaging of GFP transgenes
Embryos were collected as outlined for light microscopy, and then 4D live imaging of GFP transgenes was performed by spinning disc confocal microscopy. Data were collected using the Perkin Elmer Ultraview software and a Hamamatsu ORCA-ER CCD camera. Time points were collected at 3-or 5-minute intervals and then projected and saved as 3D movies of projected stacks.

Lethality assays
Transgene rescue of mutant lethality was assayed for all constructs, except the dlg-1 (Δ17-116), as follows. Embryos from dlg-1(ok318)/dpy8(q287) heterozygotes were isolated and mounted on 5% agar pads as described for light microscopy. Embryos were allowed to develop for ~8 hours and then assayed for the presence of the transgene via GFP fluorescence. Lethality was then assayed approximately 16-18 hours later and correlated with GFP expression. For dlg-1(Δ17-116), lethality was scored by determining the percentage of dead progeny from dlg-1(ok318)/dpy8(q287) heterozygotes, or from wild-type animals carrying the transgene.