Rap1 acts via multiple mechanisms to position Canoe and adherens junctions and mediate apical-basal polarity establishment

Cell polarity – the ability to differentially target proteins to distinct plasma membrane domains – is a fundamental cellular property, underlying processes from bacterial motility to neuronal transmission. We explore polarity in epithelia, the most common tissue architecture in animals. Epithelia form diverse organs, from skin to gut to blood vessels. In epithelia, apical-basal polarity distinguishes the apical and basal surfaces of each epithelial sheet, allowing them to serve as barriers between body compartments and to selectively transport molecules across this barrier (Campanale et al., 2017).

Cadherin-based cell-cell adherens junctions (AJs) reside at the junction between the apical and basolateral domains, separating them. Cadherins mediate cell-cell adhesion, while proteins bound to their cytoplasmic tails, broadly known as catenins, interact with the actomyosin cytoskeleton (Meng and Takeichi, 2009). Blocking cadherin-catenin function disrupts cell adhesion and epithelial polarization in cultured cells and developing embryos (Cox et al., 1996; Gumbiner et al., 1988; Johnson et al., 1986). Defining roles for AJs in cell polarity raised a new question: how are cadherin-catenin complexes positioned at the apical-basolateral boundary during polarity establishment?

Early Drosophila development provides an outstanding model of polarity establishment (Harris, 2012). Flies begin development as a syncytium, in which nuclear division occurs without cytokinesis. Nuclei move to the egg cortex and undergo several more rounds of synchronous division. They then exit the cell cycle and undergo cellularization, during which the actomyosin cytoskeleton pulls in membrane around each nucleus, creating 6000 polarized cells. The original egg cortex becomes the apical membrane, and AJs are positioned in a polarized manner at the apicolateral interface.

In the absence of AJ proteins, embryos cellularize but cells then lose adhesion for one another and concurrently lose apical-basal polarity (Cox et al., 1996; Harris and Peifer, 2004). While AJs are key for polarity initiation, they themselves must be positioned apically as membranes invaginate. The polarity protein Bazooka (Baz; fly Par3) plays a key role. It colocalizes with cadherin-catenin complexes as polarity is established, in large multiprotein complexes called spot AJs (SAJs) (Harris and Peifer, 2004; McGill et al., 2009; Tepass and Hartenstein, 1994). Initial small cadherin-catenin protein clusters are present early in cellularization. Baz clusters accumulating at the apicolateral interface engage these precursory cadherin-catenin complexes as membranes invaginate (McGill et al., 2009; Harris and Peifer, 2004), leading to the robust assembly of nascent SAJs. Smaller cadherin-catenin clusters are present all along the lateral membrane. While Baz localizes correctly in the absence of cadherins, cadherin-catenin complexes require Baz to be apically restricted. In the absence of Baz, small cadherin-catenin complexes localize all along the basolateral axis and fail to assemble into larger SAJs (McGill et al., 2009). This placed Baz at the top of the polarity hierarchy, raising the question of how Baz is positioned. One clue came from the fact that syncytial nuclear divisions involve polarized actin and microtubules. Strikingly, apical Baz positioning requires both dynein-based microtubule transport toward the apical surface and an intact actin cytoskeleton, which may anchor nascent SAJs (Harris and Peifer, 2005). However, the protein(s) linking nascent SAJs to the cytoskeleton remained unclear.

In the conventional model, cadherins link to actin via α- and β-catenin [Armadillo (Arm) in Drosophila]. Recent work revealed that this linkage is mediated by a far more sophisticated set of molecules, facilitating tension-sensing feedback mechanisms (Lecuit and Yap, 2015). The actin-binding protein Canoe (Cno; ortholog of mammalian Afadin) is part of the molecular machinery linking AJs to the cytoskeleton (Mandai et al., 2013). Cno's multidomain structure allows it to directly interact with the cytoskeleton via its F-actin-binding domain and to bind AJ proteins, including nectin, E-cadherin and α-catenin, via its PDZ and proline-rich domains.

Unlike cadherin/catenins, Drosophila cno is not essential for cell-cell adhesion (Sawyer et al., 2009), but it is required for many processes driven by AJ/cytoskeletal linkage. Mesoderm apical constriction during gastrulation offers a good example. In the absence of Cno, AJs lose connection to the contractile apical actomyosin cytoskeleton, hampering cell shape change and mesoderm internalization (Sawyer et al., 2009). Cno plays similar roles in other actomyosin-driven processes, including germband convergent elongation and dorsal closure, helping link force-generating myosin cables to AJs (Boettner et al., 2003; Boettner and Van Aelst, 2007; Choi et al., 2011; Sawyer et al., 2011).

Afadin plays similar roles to Cno. Afadin null mice are embryonic lethal, with highly disorganized ectoderm and impaired mesoderm migration (Ikeda et al., 1999; Zhadanov et al., 1999). Thus, Afadin also is not essential for cell-cell adhesion but regulates morphogenesis. Tissue-specific knockouts implicated Afadin in morphogenic events depending on cadherin function, including synaptogenesis (Beaudoin et al., 2012), lymphangiogenesis (Majima et al., 2013) and nephron lumen formation (Yang et al., 2013). In the intestine, Afadin is required for epithelial barrier function (Tanaka-Okamoto et al., 2011) and for maintaining adhesion between Paneth and intestinal crypt cells (Tanaka-Okamoto et al., 2014).

The diverse events in which Cno/Afadin are involved and their roles in linking AJs to actomyosin suggested that Cno might mediate AJ/cytoskeletal interactions during polarity establishment in cellularization. Consistent with this, Cno localizes to nascent SAJs as they form, with special enrichment in supramolecular structures at tricellular junctions (TCJs), where three forming cells meet (Choi et al., 2013). Strikingly, in cno maternal/zygotic mutants, apical enrichment of both Baz and cadherin-catenin complexes is lost – they localize to the cortex but are not apically enriched. This placed Cno at the top of the polarity establishment hierarchy, and thus raised questions about the cues localizing Cno to nascent SAJs as polarity is established.

An intact actin cytoskeleton is crucial for Cno cortical localization (Sawyer et al., 2009). Cno localization to nascent SAJs is also dependent on the small GTPase Rap1 (Choi et al., 2013; Sawyer et al., 2009). Cno binds active Rap1 via two N-terminal Ras-associated (RA) domains (Boettner et al., 2003). Rap1 maternal/zygotic mutants also fail to apically enrich Baz and cadherin-catenin (Choi et al., 2013). Thus, Rap1 is an essential cue in establishing polarity, acting upstream of Cno. Consistent with a role for Rap1 in mediating AJ organization, in wing disc epithelia Rap1 helps maintain uniform AJ distribution around the cell periphery (Knox and Brown, 2002). Rap1 also helps regulate vertebrate cadherin-mediated cell adhesion (Asuri et al., 2008; Price et al., 2004; Sato et al., 2006).

Like other small GTPases, Rap1 is pleiotropic. Although Rap1's effectors vary between organisms and tissues, many act through the actin cytoskeleton (reviewed by Frische and Zwartkruis, 2010). Drosophila Rap1 is required for many events that depend on cell shape change and dynamic junctional remodeling (Asha et al., 1999). Rap1 loss-of-function mutants, like cno mutants, exhibit defects in mesoderm invagination (Sawyer et al., 2009; Spahn et al., 2012) and dorsal closure (Boettner et al., 2003; Boettner and Van Aelst, 2007). In both events, Dizzy (Dzy, or PDZ-GEF; mammalian ortholog RapGEF2) is the GEF regulating Rap1 (Boettner and Van Aelst, 2007; Spahn et al., 2012).

These observations place Cno and its regulator Rap1 at the top of the hierarchy of proteins that regulate apical-basal polarity establishment during Drosophila embryogenesis, one of our premier models of this process. They frame the questions that we set out to answer here. (1) How is Cno positioned at nascent AJs as cells form, allowing it in turn to position Baz and thus AJ proteins? Which Cno domains are important for apical positioning and polarity establishment and how do they influence Cno function at cellularization and during later development? (2) By what mechanisms does the small GTPase Rap1 act in this process, what role does GTPase activity play, and what molecules act as Rap1 regulators?

Our goal was to define mechanisms mediating apical-basal polarity establishment, with a focus on how Rap1 regulates Cno, the most upstream known player in the process. We first extended our earlier description of Cno localization (Sawyer et al., 2009; Choi et al., 2013). At the onset of cellularization, small endogenous Cno puncta are positioned apically (Fig. 1A,F). As the invaginating furrows approach the bottom of the nuclei, Cno puncta are retained in the more apical part of the membrane and begin to assemble into nascent SAJs (Fig. 1B′, arrow), which are ∼2-3 µm below the apical surface. At this stage, Cno localizes equally to bicellular junctions and TCJs (Fig. 1G, arrows), colocalizing with DE-cadherin (Shotgun) and Baz (Sawyer et al., 2009). However, if one images 1.5 µm more basally, one begins to see Cno enrichment at TCJs (Fig. 1G′, arrows). By mid-cellularization, Cno puncta are spatially excluded from the apicalmost domain (Fig. 1C), with some TCJ enrichment (Fig. 1H,H′). As cellularization is completed, Cno remains enriched at nascent SAJs (Fig. 1D,E) and TCJ enrichment intensifies, both at the SAJ level (Fig. 1I,J, red arrows; quantified in Fig. 1L) and 2.1 µm basally to them (Fig. 1I′,J′, arrows). Quantification revealed that while Cno is present at roughly equal levels at bicellular junctions and TCJs at mid-cellularization, it becomes ∼5-fold enriched at TCJs by the end of cellularization (Fig. 1L). Three-dimensional reconstruction suggested that, as cellularization is completed, Cno enrichment extends several microns basally along TCJs (Fig. 1K, arrow; apical-basal extension at TCJs at mid- and late cellularization is quantified in Fig. 1M and N). Given the limitations of imaging with light microscopy, especially in the z-axis, we cannot determine whether these are actually continuous structures or closely aligned puncta; for simplicity, we refer to them as cables below.

Cno recruitment/maintenance at nascent SAJs is Rap1 dependent. In Rap1 maternal/zygotic mutants, Cno membrane accumulation is lost, disrupting apical-basal polarity and leading to defects in epithelial integrity (Choi et al., 2013). We utilized Valium20 shRNAs targeting Rap1 expressed under control of the GAL4-UAS system. Expression in the female germline knocks down maternal mRNA, and maternally expressed GAL4 persists zygotically, driving zygotic shRNA expression, often leading to knockdown mimicking maternal/zygotic mutants (Staller et al., 2013). RNAi reduced Rap1 below levels detectable by immunoblotting (Fig. S1G,H), effectively reproducing the Rap1 null phenotype (Fig. S2A,B). In late cellularizing Rap1 RNAi embryos, Cno is virtually absent from the membrane in apical-basal cross-sections (Fig. 2A,B), maximum intensity projections (MIPs) of cross-sections (Fig. 2C,D) or apical en face sections (Fig. 2I,J; all images were intensity matched with wild type; when single images are presented they are chosen to be representative, with the number of embryos visualized for key conclusions presented in Table 1). Reduced Cno at the membrane is a consequence of altered localization and not a change in absolute Cno levels (Fig. S1G).

The polarity regime established during cellularization is elaborated on during gastrulation. At this point, new polarity cues come into place, and, for example, apical Baz and AJ localization is partially restored in cno mutants (Choi et al., 2013). We examined whether Cno localization was restored in Rap1 RNAi embryos when gastrulation began. Low-intensity Cno puncta reappeared by late stage 6, which were apically enriched (Fig. 2E-H, arrows) and associated with TCJs (Fig. 2K,L, arrows), but overall Cno intensity at the membrane remained very low. Thus, Rap1 still plays a role during early polarity maintenance. However, as germband extension neared completion, Cno localization to AJs resumed (Fig. 2M,N), although at reduced levels. Thus, Cno localization is strongly Rap1 dependent during polarity establishment but becomes less so in the polarity maintenance phase. This Rap1-independent Cno relocalization is not sufficient to maintain ectodermal integrity however, as Rap1 RNAi embryos exhibit severe epithelial disruption (Fig. S2B).

Rap1 localization during cellularization contrasts with the polarized localization of Cno at the apicolateral boundary. Endogenous Rap1 is not polarized (Choi et al., 2013), instead localizing along the entire length of the cellularizing membrane, extending down to the furrow front where actin and myosin are enriched. Rap1 physical interaction with Cno/Afadin is dependent on Rap1's activity state (Boettner et al., 2000, 2003). Since Rap1 itself is not polarized, we hypothesized that Rap1 activation is spatially restricted to the apicolateral boundary and consequently defines the site for initial Cno localization and SAJ formation.

To manipulate Rap1 activity we created three GAL4-UAS-driven GFP-tagged Rap1 transgenes (Ellis et al., 2013): full-length wild-type Rap1 (referred to as Rap1WT); Rap1S17A, which is locked in the GDP-bound off state (referred to as Rap1GDP); and the constitutively GTP-bound and thus active Rap1Q63E (referred to as Rap1CA). Small GTPase mutants like Rap1S17A trap GEFs in high-affinity complexes, effectively sequestering them from acting on downstream GTPases (Dupuy et al., 2005). Conversely, constitutively active mutants are locked in the GTP-bound state and rendered insensitive to GAPs, resulting in continuous signaling to downstream effectors (Kitayama et al., 1990; Scrima et al., 2008). When expressed using maternal GAL4, Rap1WT, Rap1CA and Rap1GDP were 11-, 15- and 3.5-fold above endogenous Rap1 levels, respectively (1-4 h, Fig. S3A,B). All three localized along the entire lateral membrane during cellularization (Fig. 3A-C), like endogenous Rap1 (Choi et al., 2013). Expressing Rap1WT had no effect on embryo viability (Fig. S2E). By contrast, expressing either GDP-locked or GTP-locked Rap1 mutants was embryonic lethal (Fig. S2E). Rap1GDP strongly disrupted epithelial integrity, with the ventral epidermis fragmented (Fig. S2C,F), similar to, although slightly more severe than, Rap1 maternal/zygotic loss (Choi et al., 2013) or RNAi (Fig. S2B). By contrast, Rap1CA embryos had a largely intact epidermis, with defects in head involution alone or in conjunction with epidermal holes (Fig. S2D,F).

We next examined how altering Rap1 activity affects Cno localization. In embryos overexpressing Rap1WT, Cno localizes normally to nascent SAJs during cellularization (Fig. 3D,G MIP). At gastrulation onset, Cno clustered at the extreme apical end of the cell, as in wild type (Fig. 3N). By contrast, eliminating Rap1 activation by expressing Rap1GDP resulted in almost complete loss of Cno from the membrane (Fig. 3E,H), thus mimicking Rap1 null or RNAi embryos (Fig. 2) (Choi et al., 2013). Small Cno puncta were observed along the length of the invaginating membrane, but their low-level intensity suggests they do not account for all Cno displaced from nascent SAJs (Fig. 3E,H). We quantified this using ImageJ Plot Profile to measure pixel intensity in projected cross-sections compared with intensity-matched controls (Fig. 3L; as in Choi et al., 2013). These very low levels of Cno at the membrane in Rap1GDP embryos meant it was impossible to assess relative enrichment in TCJs. When polarity maintenance initiated, Cno levels at the membrane were not restored in Rap1GDP embryos (Fig. 3O,Q). However, as in Rap1 RNAi embryos, low-intensity Cno puncta appeared on the membrane by late stage 6 (Fig. 3O′).

Our hypothesis that a spatially confined pool of active Rap1 is an essential cue for Cno polarization predicted that expressing Rap1CA along the length of the cellularizing membrane would broaden Cno membrane recruitment basally. Consistent with this, Rap1CA induced several changes in endogenous Cno localization. In Rap1CA embryos, a subset of membrane-associated Cno did not exhibit its characteristic apically restricted, punctate appearance, but rather localized uniformly all along the basolateral membrane (Fig. 3F, arrows). This Cno localization is reminiscent of that of Rap1CA (Fig. 3C). A second Cno pool remained apically enriched (Fig. 3F, bracket), but relative to wild type there were two differences. First, the Cno puncta in Rap1CA embryos extended more basally, involving, in part, basal extension of TCJ Cno cables (Fig. 3G versus I, quantified in Fig. 3M). Second, Cno puncta were found more apically in Rap1CA embryos (Fig. 3F′, yellow arrows; J versus K, arrowheads; en face views at SAJs versus more apical). However, Rap1CA expression did not eliminate enrichment at TCJs (Fig. 3K′). As Rap1CA embryos gastrulate, mislocalized Cno puncta became excluded from the apical domain and more tightly focused at the AJs (Fig. 3P), although full apical translocation was delayed. Together, these data are consistent with the idea that an early apical asymmetric distribution of active Rap1 along the cellularizing membrane regulates Cno recruitment to nascent SAJs.

Cno and Rap1 regulate polarity establishment by regulating apical Baz positioning. In the absence of either Cno or Rap1, Baz puncta lose apical restriction and redistribute all along the basolateral membrane (Choi et al., 2013). Whereas expressing Rap1WT did not affect Baz polarization (Fig. 4B,F versus A,E), Rap1GDP expression depolarized Baz. Membrane-associated Baz puncta extended all along the basolateral membrane and up into the apical domain above the nascent SAJs (Fig. 4C,G versus A,E, quantified in Fig. 4I). Baz mislocalization was partially rescued as embryos entered gastrulation; Baz was gradually excluded from the lateral domain, accompanied by enrichment at the apicolateral interface (Fig. 4L versus K). Expressing Rap1GDP also disrupted cadherin-catenin complex assembly into SAJs (Fig. 5A versus B, quantified in Fig. 5D), thus resembling Rap1 or cno null embryos (Choi et al., 2013). The pool of smaller Arm puncta along the basolateral membrane and enrichment at basal junctions (Fig. 5A,B, arrows) appeared unaffected. AJ protein polarization was not substantially restored at gastrulation onset (Fig. 5F,I versus E,H). Thus, Rap1 activity is required to initiate Baz and AJ polarization and plays a role in AJ polarity maintenance.

We next examined how unrestricted Rap1 activity affects Baz polarization. In Rap1CA embryos, Baz puncta became enriched in the apical half of the forming cell but did not concentrate as tightly at nascent SAJs (Fig. 4D,H versus A,E). Pixel intensity plots confirmed this flattening of the Baz peak relative to wild type (Fig. 4J). Apical Baz clustering was largely restored at gastrulation onset (Fig. 4M versus K), although focusing of Baz into AJs was less complete. In cellularizing Rap1CA embryos, AJ proteins remained somewhat enriched apically but were not as tightly apically restricted, localizing along the upper half of the lateral membrane in a manner similar to Baz (Fig. 5A versus C′, quantified in Fig. 5D). Mislocalized Arm puncta largely colocalized with the extended Cno cables (Fig. 5C, arrows), suggesting that although Rap1CA disrupts the precise spatial organization of AJs, molecular interactions between AJ proteins remain at least partially intact. Arm colocalization with Cno did not extend to the smoothly localized pool of Cno along the basolateral domain; instead, Arm localized to small clusters as in wild type. AJ assembly began to be restored at gastrulation, although apical translocation was delayed (AJs 2.6±1.0 µm from the apical surface in wild type versus 4.8±1.1 µm in Rap1CA; Fig. 5G,J; Fig. S3F versus G). This occurs in addition to the normal difference in timing of wild-type AJ translocation between dorsal and ventral ectoderm (Weng and Wieschaus, 2016). These data suggest that apical Cno clustering and subsequent Baz and Arm recruitment to nascent AJs are regulated by localized Rap1 activity. Thus, localized Rap1 activity is necessary for establishing polarity, but is less essential for maintaining the polarized state.

One mechanism to spatially compartmentalize Rap1 activity is via specific GEFs. The Rap1GEF Dzy (PDZ-GEF) is a plausible candidate, given its role in mesoderm invagination, which immediately follows cellularization and requires Rap1 and Cno (Spahn et al., 2012; Sawyer et al., 2009). We generated embryos lacking maternal/zygotic Dzy, crossing females with germlines homozygous for the null allele dzy^Δ1 (Huelsmann et al., 2006) to dzy^Δ1 heterozygous fathers. 64% of progeny died as embryos, suggesting partial zygotic rescue of the 50% of embryos receiving paternal wild-type dzy. Dead embryos had defects in head involution, dorsal closure and epidermal integrity (Fig. S2G,H). We then examined effects on Cno localization. In contrast to Rap1 mutants (Choi et al., 2013), Cno was not lost from the plasma membrane. Cno was apically restricted to nascent SAJs [Fig. 6B,D versus A,C; localized to the apical 50±3% of the cell in dzy (n=7) versus to the apical 45±6% of the cell in wild type (n=3)]. However, there was one striking difference from wild type. At late cellularization, wild-type embryos have strong Cno enrichment at TCJs, both at the level of SAJs (Fig. 6C) and 2 µm basally (Fig. 6C′). By contrast, in dzy embryos Cno localized to bicellular contacts but was not enriched at TCJs (7/7 late stage 5 embryos; Fig. 6D versus C). This defective enrichment of Cno at TCJs was even more obvious in 3D reconstructions (Fig. 6E,F). The TCJ defect persisted into gastrulation (3/5 stage 6 embryos; Fig. 6H,K versus G,J), but zygotic rescue was observed in ∼50% of embryos (2/5 embryos; Fig. 6I,L). Thus, Dzy is not essential for initial Cno membrane recruitment or apical-basal positioning, but is required for effective TCJ clustering. These data suggest that, during cellularization, Rap1 might be regulated by more than one GEF, with different regulatory inputs coordinating Rap1's role in Cno polarization versus junctional maturation.

In our earlier analysis, we observed that Rap1 mutants had an additional phenotype during cellularization that was not shared by cno mutants. In wild type, apical cell shapes during cellularization are trapezoidal, with straight bicellular borders and relatively uniform cell apical areas (Fig. 6M). In Rap1 mutants, cell borders are less straight and apical areas become more variable (Choi et al., 2013). dzy mutants also exhibited more irregular apical cell shapes during cellularization (Fig. 6M versus N), consistent with Dzy also regulating this Rap1 activity. dzy mutants also mimicked effects of cno mutation (Choi et al., 2011; Sawyer et al., 2009, 2011) in later events, disrupting mesoderm invagination (Spahn et al., 2012), exaggerating Baz planar polarization during germband extension and leading to deepened segmental grooves during dorsal closure (Fig. S2I-P). However, neither dzy nor cno maternal/zygotic mutants share the severe disruption of epidermal integrity seen in Rap1 maternal/zygotic mutants (Fig. S2A,C,G) (Sawyer et al., 2009). These data suggest that Dzy is an important regulator of Rap1 and Cno during many morphogenetic events, but is not the only regulator.

Our data are consistent with the hypothesis that Rap1 is activated at SAJs by Dzy. This could occur by spatially restricting Dzy to this region, or Dzy could be uniform but activated apically. To investigate this, we imaged GFP-tagged Dzy driven by its endogenous promotor (Boettner and Van Aelst, 2007; Spahn et al., 2012). We found that by the end of cellularization, Dzy is enriched at the cell cortex and restricted to the apical half of the membrane (Fig. 6P), with low-level enrichment of Dzy observed at TCJs in some, but not all, fields of view (Fig. 6O). At gastrulation onset, Dzy cortical localization and TCJ enrichment continued (Fig. 6Q) and in cross-section enrichment at apical AJs was apparent (Fig. 6R). By the beginning of germband extension, Dzy incorporated into a uniform AJ belt, as junctions moved apically (Fig. 6S,T), analogous to the localization of Cno and other AJ proteins. These data are consistent with a localized pool of Dzy regulating the activity of a junctional pool of Rap1.

The RA domains of Cno bind GTP-bound Rap1 (Boettner et al., 2003). To define RA domain roles in Cno regulation and function, we created three C-terminally GFP-tagged Cno constructs driven by the GAL4-UAS system: full-length wild-type Cno (CnoWT); Cno with the N-terminal RA domains deleted (CnoΔRA); or Cno with the F-actin-binding domain and linker region removed in addition to the RA domains (CnoFHA-PDZ). When driven by maternal GAL4, all accumulated at similar levels (Fig. S1A,B,D). Overexpressing CnoWT or CnoΔRA in a wild-type background did not appreciably affect embryo lethality (data not shown).

One simple mechanism by which Rap1 could recruit Cno to the plasma membrane and nascent SAJs is via direct interaction with the RA domains. In the simplest version of this hypothesis, deleting the RA domains would have the same effect on Cno as removing Rap1, abolishing membrane recruitment. To test this hypothesis, we wanted to express CnoΔRA in a background in which endogenous Cno was reduced/eliminated. We were fortunate that an existing shRNA targeted the 5′ region of cno mRNA deleted in CnoΔRA, allowing us to knockdown endogenous Cno without affecting CnoΔRA. Endogenous Cno was effectively knocked down during cellularization/gastrulation (Fig. S1E,F; to <5%). Cno continued to be substantially depleted during mid- to late embryogenesis, although there was some rebound in expression.

We expressed either CnoWT-GFP or CnoΔRA-GFP in embryos with endogenous Cno knocked down by RNAi. The shRNA also targeted CnoWT-GFP. However, fortuitously, the high-level expression of CnoWT-GFP partially balanced this, reducing its levels to near those of endogenous Cno in controls (Fig. S1E,F). The GFP signal was sufficiently strong to confirm that CnoWT-GFP retained apical enrichment, localizing like endogenous Cno in wild type (Fig. 7C,F versus A,D).

This knockdown-rescue approach allowed us to test whether CnoΔRA was deficient in either membrane recruitment or apical enrichment. Strikingly, CnoΔRA expressed in a cno RNAi background remained membrane associated (Fig. 7B,H versus A,G), in contrast to Cno in Rap1 mutants. However, in contrast to CnoWT expressed in a cno RNAi background (Fig. 7F), CnoΔRA apical enrichment was severely reduced (Fig. 7B,E, quantified in Fig. 7J). Furthermore, CnoΔRA was not particularly enriched at TCJs versus neighboring bicellular junctions (Fig. 7H versus G, arrows), unlike CnoWT (Fig. 7I). At gastrulation onset, CnoΔRA began to be recruited into apical AJs (Fig. 7L), but apical clustering and TCJ enrichment remained impaired (mid-stage 6; Fig. 7L,N versus K,M). By late stage 6, CnoΔRA relocalized to AJs but not TCJs (Fig. 7O-R). These data suggest CnoΔRA can localize to the membrane independently of endogenous Cno, but that it is defective in apical retention at SAJs and clustering at TCJs. This is consistent with the idea that different inputs regulate different aspects of Cno localization.

In parallel, we considered an alternative mechanism by which Rap1 could regulate Cno localization. Actin-regulatory formins provide an interesting model. Their N-terminal RA domain folds back onto the rest of the protein, locking it into a closed, inactive conformation. This is relieved by Rho binding (Rose et al., 2005). We tested the hypothesis that the RA domains of Cno are similarly autoinhibitory, with inhibition relieved by Rap1 binding. If so, deleting the RA domains might allow Rap1-independent Cno SAJ recruitment. We expressed either CnoWT-GFP or CnoΔRA-GFP in Rap1 RNAi embryos. We confirmed knockdown by immunoblotting; during cellularization, Rap1 was knocked down to levels undetectable by immunoblotting (Fig. S1G,H). As expected, Rap1 RNAi eliminated apical enrichment of CnoWT-GFP in nascent SAJs (Fig. 8A versus B, arrowheads), although weak localization of small puncta all along the lateral membrane remained (Fig. 8F, arrows). CnoWT-GFP accumulated ectopically at basal junctions (Fig. 8B,D, arrows), something we also saw weakly when expressing CnoWT in a wild-type background. Strikingly, Rap1 RNAi had similar effects on CnoΔRA-GFP localization, eliminating apical enrichment at SAJs (Fig. 8C, arrowhead) and leading to ectopic basal junction accumulation (Fig. 8C,E, arrows). There was little rescue of either CnoWT or CnoΔRA localization to AJs at gastrulation onset (stage 6; Fig. 8J,K). Thus, deleting the RA domains does not simply open up an inactive conformation of Cno to render it Rap1 independent for localization.

After gastrulation begins, initial defects in AJ protein localization in some mutants begin to be rescued, restoring normal or near-normal junctional localization, and Par1 activation can contribute to this (McKinley and Harris, 2012). After Rap1 knockdown, endogenous Cno relocalizes to AJs as gastrulation proceeds (Fig. 2; Rap1 knockdown remained effective even at 12-15 h; Fig. S1G,H). Similarly, when germband extension initiated (stage 7) both CnoWT and CnoΔRA relocalized to AJs in Rap1 RNAi embryos (Fig. 8L-O). Thus, Cno is less reliant on Rap1 for post-gastrulation localization.

A striking aspect of Cno localization is enrichment in macromolecular ʻcables' at TCJs. This suggests it might have the ability to self-interact, either directly or indirectly. We tested whether this putative interaction would allow wild-type Cno to recruit CnoΔRA to nascent SAJs, expressing our GFP-tagged Cno constructs in a wild-type background, using maternal GAL4. CnoWT-GFP largely replicated endogenous Cno localization. It was enriched at nascent SAJs as they formed (Fig. 9A,B), with particular enrichment at TCJs (Fig. 9C,D). There was also slight enrichment at the basal junctions, just apical to the actomyosin machinery driving membrane invagination; we suspect that this localization is an overexpression artifact. We next examined CnoΔRA-GFP. In contrast to CnoΔRA localization defects after cno RNAi, in a wild-type background CnoΔRA localization also strongly resembled that of endogenous Cno, with strong recruitment to forming apical SAJs (Fig. 9E,F) and marked enrichment at TCJs, including extension into cables at those sites (Fig. 9G,H). Like CnoWT-GFP, there also was some basal junction enrichment. Both CnoWT and CnoΔRA also resembled endogenous Cno in becoming more focused in the apical-basal axis at late stage 6 (Fig. 9I,J) and moving apically with AJs at stage 7 (Fig. 9K,L). Thus, in the presence of wild-type endogenous Cno and Rap1, the RA domain is dispensable for Cno localization. By contrast, deleting the F-actin-binding domain and linker in addition to the RA domains (CnoFHA-PDZ) completely abolished membrane localization (Fig. 9M). This suggests that the F-actin-binding and/or linker regions are important for Cno membrane localization.

In our final tests, we examined CnoΔRA function in morphogenesis. We first examined its ability to rescue cno RNAi knockdown, which leads to highly penetrant embryonic lethality (Fig. 10I) and strong defects in morphogenesis (Fig. 10A-E,J). Since combining two UAS-driven transgenes can reduce expression, as a control we expressed both cno RNAi and red fluorescent protein (RFP), controlling for the possibility that Cno knockdown was reduced. We observed no significant reduction in Cno knockdown (Fig. S1E,F) or embryonic lethality (Fig. 10I), although effects on morphogenesis were slightly alleviated (Fig. 10J). Expressing CnoWT in cno RNAi embryos completely rescued both embryonic viability (Fig. 10I) and morphogenesis, as assessed by cuticle phenotype (Fig. 10A-E,J) and embryo staining (Fig. 10F,G). CnoΔRA also retained substantial function, although less than CnoWT, despite accumulating at much higher levels. Embryonic lethality was reduced to 41% (Fig. 10I), and in the dead embryos morphogenesis was essentially fully restored (Fig. 10J,H).

The Rap1 RNAi background, with its severely reduced Rap1 levels, provided a sensitized background in which to compare CnoΔRA and CnoWT function. Rap1 RNAi led to completely penetrant embryonic lethality (Fig. 10S) and to strongly disrupted head involution, dorsal closure and, in severe cases, epidermal integrity (Fig. 10K-O,T). We once again controlled for two transgenes using RFP, with little or no effect on knockdown (Fig. S1G,H) or phenotype (Fig. 10S,T).

We then examined whether CnoWT or CnoΔRA rescued Rap1 knockdown. CnoWT partially rescued embryonic viability (Fig. 10S). Morphogenesis rescue was more substantial; 85% of embryos had a nearly wild-type epidermis (Fig. 10Q,T). By contrast, expressing CnoΔRA in the Rap1 RNAi background led to no rescue of embryonic viability (Fig. 10S), and the rescue of embryonic morphogenesis was significantly less complete (Fig. 10R,T). Thus, CnoΔRA retains significant but not full function.

Apical-basal polarity establishment is a key step in animal development, driving the assembly of epithelial tissues and organs and creating the architecture that enables morphogenetic movements. Early fly embryos provide a premier model of polarity establishment. Work from many labs has defined the assembly and apical positioning of cell-cell AJs as the key initial step. Efforts then focused on defining molecular and cellular mechanisms underlying this, revealing key roles for Baz/Par3 (Harris and Peifer, 2004) and, more recently, the junctional actin crosslinker Cno and its upstream activator Rap1 (Choi et al., 2013). This refocused our attention on the next level of mechanistic analysis: how does Rap1 regulate Cno localization and do Rap1-independent cues also play a role?

Our current model of apical-basal polarity establishment during fly embryogenesis suggests the key upstream step is positioning Cno at the site where AJs will form. The small GTPase Rap1 and F-actin play roles in Cno positioning (Sawyer et al., 2009). However, the mechanism by which Rap1 acts remained unclear, as Rap1 localizes all along the invaginating membrane. Our new data support a model in which apical Rap1 activity plays a key role. The ability of GTP-locked constitutively active Rap1 to recruit Cno all along the basolateral membrane and the loss of Cno from the membrane induced by GDP-locked Rap1 are both consistent with this model. In the future, it would be valuable to design a Rap1 activity sensor to confirm this hypothesis. As we discuss below, it was intriguing that although the Cno RA domain plays a role in apical Cno recruitment/retention, Rap1 also influences localization of a cno mutant lacking the RA domain. Finally, it was of interest that both the GDP- and GTP-locked Rap1 mutants altered Baz localization, raising the possibility that Baz localization requires cycling of Rap1 between the active and inactive states.

Our next task was to define Rap1 activation mechanisms. Rap1 has many GEFs, including C3G, Epac, CalDAGGEF1 and PDZ-GEF (Gloerich and Bos, 2011). Dzy (PDZ-GEF) was the most likely candidate as it has a role 30 min later in mesoderm apical constriction, which also requires Rap1 and Cno (Spahn et al., 2012). Strikingly, although our analyses suggest Dzy regulates Cno in later morphogenetic events such as germband extension, junctional planar polarization and segmental groove retraction, maternal/zygotic Dzy loss did not fully mimic effects of Rap1 loss on polarity establishment. Dzy loss replicated only a subset of these effects, as cortical Cno recruitment and apical restriction were unaffected. Instead, Dzy loss specifically affected Cno enrichment at TCJs, and also led to defects in columnar cell shape like those caused by Rap1 loss. Based on this, we hypothesize that multiple GEFs regulate Rap1 activity, each directing specific aspects of cellularization and polarity establishment. Each might create temporally or spatially restricted pools of active Rap1, with different Rap1 pools mediating different effects on Cno localization. In the accompanying paper, Schmidt et al. (2018) describe a role for the unconventional GEF ELMO-Sponge complex in regulating initial apical positioning of Cno, consistent with this hypothesis. In the future, it will be important to investigate other Rap1GEFs, such as C3G. The cortical and apical localization of Dzy that we observed during cellularization and early gastrulation are consistent with the idea that localized Dzy provides a direct input into Rap1 activation. Our data also suggest that in later events in embryogenesis, Dzy and Cno work together in many events with Rap1, but that Rap1 is also likely to regulate events where it has other activators and effectors. Another alternative is apical restriction of Rap1 activation via basolateral Rap1GAPs, as occurs at other times (Wang et al., 2013), and this will be important to explore.

The N-terminus of Cno carries two RA domains that bind Rap1 (Boettner et al., 2003). Given the essential role of Rap1 in Cno localization and function, we suspected that the RA domains would be similarly essential. We tested two hypotheses for the mechanism by which Rap1 could act via the RA domains to regulate Cno function during polarity establishment: (1) Rap1-GTP binding to the RA domain opens a closed, autoinhibited conformation; or (2) Rap1-GTP binding to the RA domain physically recruits Cno to sites where AJs will be assembled. Our data essentially rule out the first hypothesis, as it predicted that CnoΔRA recruitment to nascent AJs would be Rap1 independent. Our data are largely consistent with the second hypothesis, although the effect of Rap1 on CnoΔRA localization revealed that the Rap1-RA domain interaction is not the only means of regulation.

Our data also suggest that after gastrulation onset Rap1-independent mechanisms of Cno localization come into play, as cortical localization of both wild-type Cno and CnoΔRA begin to be restored at that stage in Rap1 knockdown embryos. Although some residual Rap1 might remain, we think this is unlikely as we cannot detect it by immunoblotting at stages when Cno localization is restored. Cno can localize to the cortex during dorsal closure in embryos expressing GDP-locked Rap1 (Boettner et al., 2003), also supporting Rap1-independent mechanisms. One potential mechanism of post-gastrulation Cno recruitment is via its known interactions with AJ proteins. Before AJs assemble, Rap1 might be essential for cortical Cno recruitment, but once AJs reappear after gastrulation onset then interactions between Cno and α-Catenin or DE-cadherin might restore Cno recruitment to AJs. Both CnoWT and CnoΔRA also retained significant function in Rap1 knockdown embryos, suggesting that Cno has Rap1-independent activity, at least when expressed at elevated levels. CnoΔRA was significantly less effective than CnoWT, confirming and extending earlier work during dorsal closure (Boettner et al., 2003). However, CnoΔRA retained significant function in Cno knockdown embryos, suggesting that the RA domain is not absolutely essential. It might facilitate some Cno/Afadin activities; for example, the RA domain of mammalian Afadin regulates interactions with p120-catenin and modulates E-cadherin endocytosis (Hoshino et al., 2005).

Several observations support the hypothesis that Cno localization responds to multiple inputs. Rap1 activity plays a key role, as during cellularization Cno cannot localize to the membrane in the absence of Rap1 or when its ability to load GTP is compromised. It seems likely that direct interactions between Rap1 and the Cno RA domain help regulate Cno localization. However, since CnoΔRA still requires Rap1 to localize correctly, this suggests additional complexity. A second Rap1 interaction site might exist in Cno outside the RA domain. Alternately, other Rap1 effectors might regulate Cno localization by different mechanisms. During cellularization, Rap1 has Cno-independent effects on apical contractility and thus columnar cell shape (Choi et al., 2013). Perhaps this postulated effector alters the actomyosin cytoskeleton in a way that promotes Cno binding, since intact actin is required for Cno cortical localization (Sawyer et al., 2009). However, Cno does not simply colocalize with actin or myosin, as both are most enriched at the leading edge of the invaginating membrane. Perhaps there is an apical pool of actin in a particular conformation or with particular binding partners that allow Cno to ʻchoose' the correct localization. Consistent with an important role for the Cno C-terminal actin-binding domain and/or intrinsically disordered linker in localization, CnoFHA-PDZ did not localize cortically in wild-type embryos, unlike CnoΔRA. Together, these data suggest a multifactorial recruitment mechanism.

Drosophila embryos afford unmatched temporal resolution, allowing us to follow AJ morphogenesis both in the context of apical-basal polarity establishment and later in polarity maintenance. This complex process begins with the appearance of cadherin-catenin clusters, which arise independently of a Baz polarization cue (McGill et al., 2009). It continues through formation of mature SAJs, which coordinate with the contractile cytoskeleton to enable the first morphogenetic movements of gastrulation. We now view these events as involving two independent but interlocked processes: apical restriction of Cno and AJ proteins, and their assembly into higher-order multiprotein complexes. Integrating our data with previous analyses prompts the following model. Baz helps ensure that small cadherin-catenin complexes present before cellularization are recruited/retained at an apicolateral position and assembled into larger complexes containing over 1000 cadherin molecules (McGill et al., 2009). Cno plays an important role, helping retain Baz at the apicolateral site of SAJ assembly. Our new data suggest that Rap1 activity guides two important aspects of AJ morphogenesis – AJ protein recruitment/retention at the apicolateral interface and the specialized assembly of larger macromolecular complexes at TCJs. Our data further suggest that these two events require at least two spatial cues directed by active Rap1, one acting via the Cno RA domains and one independent of that. Our results with the Rap1GEF Dzy suggest these two events have different modes of regulation. Finally, our data are consistent with a Cno-driven, self-reinforcing feedback loop in which correct Cno localization can recruit more Cno to the membrane. Cno clustering is particularly prominent at TCJs. Intriguingly, recent work in mammalian cells suggests that actomyosin cables anchor end-on at TCJs and Afadin acts there to maintain trapezoidal cell shapes (Choi et al., 2016). Whether Cno oligomerization is intrinsic to Cno itself or is mediated by other partners is an important question for future work. These data also have potential implications for the roles of mammalian Afadin in epithelial polarity in the kidney and intestine, another avenue for further research.

Fly stocks are listed in Table 2. Mutations are described at FlyBase (http://flybase.org). Wild type was yellow white or Histone-GFP. All experiments were performed at 25°C. The stock to make dzy germline clones and the GFP-tagged endogenous Dzy stock were kindly provided by Rolf Reuter (Universität Tübingen, Tübingen, Germany; Huelsmann et al., 2006). dzy germline clones were made by heat shocking 48- to 72-hour-old hsFLP¹; FRT2L dzy^Δ1/FTR40Aovo^D1-18 larvae for 3 h at 37°C (Chou et al., 1993). shRNA knockdown of Rap1 and cno was carried out by crossing double-copy mat-tub-GAL4 females (Staller et al., 2013) to males carrying UAS.Rap1 shRNAi or UAS.cno shRNAi (Valium20; Ni et al., 2011) constructs. Maternal expression of GFP-tagged Cno constructs or Rap1 activity mutants was carried out by crossing males to female double-copy mat-tub-GAL4 flies also carrying the Cno or Rap1 constructs.

Rap1WT, Rap1Q63E and Rap1S17A were described previously (Ellis et al., 2013). All Rap1 activity mutants are tagged with GFP at the N-terminus. Rap1S17A is inserted on the second chromosome, Rap1WT and Rap1Q63E are inserted on the third. To generate cno transgenic lines, full-length Drosophila cno cDNA was PCR cloned into the pCR8/GW/TOPO Gateway Entry Vector (Life Technologies). Using this template, Cno domain deletions were generated using the following oligonucleotide primers (forward and reverse, 5′-3′): CnoΔRA, CGTCCGGCGGACTCACAGCCCCGGAGGAGAAAGAAAAAA and TTAGTGCACCGCGTCTATATCTCG; CnoFHA-PDZ, CGTCCGGCGGACTCACAGCCCCGGAGGAGAAAGAAAAAA and TAGATGGCTCCCTGCTTGGCCACTTCC.

Cno mutant constructs were then recombined into Drosophila UASp Gateway vectors. CnoΔRA corresponds to deletion of amino acids 1-345. CnoFHA-PDZ corresponds to amino acids 346-1100. All Cno transgenic lines are C-terminally tagged with GFP and targeted to the second chromosome.

Cuticle preparation was performed according to Wieschaus and Nüsslein-Volhard (1986).

Antibodies are listed in Table 2. Dechorionated embryos were fixed in boiling Triton salt solution (0.03% Triton X-100, 68 mM NaCl, 8 mM EGTA) for 10 s followed by fast cooling on ice and devitellinized by vigorous shaking in 1:1 heptane:methanol. Embryos were stored in 95% methanol/5% 0.5 M EGTA for at least 48 h at −20°C prior to staining. Embryos were washed three times with 0.01% Triton X-100 in PBS (0.01% PBS-T), followed by blocking in 1% normal goat serum (NGS) in 0.01% PBS-T for 1 h. Primary and secondary antibody staining were carried out at 4°C overnight with nutation. A variation on this protocol was performed for anti-GFP staining, in which embryos were blocked in 20% NGS in 0.1% Triton X-100 in PBS (0.1% PBS-T) followed by primary and secondary antibody incubations in 10% NGS in 0.1% PBS-T. Antibody dilutions are listed in Table 2.

Fixed embryos were mounted in Aqua-Poly/Mount (Polysciences) and imaged on a confocal laser-scanning microscope (LSM 710 or LSM 880, 40×/NA 1.3 Plan-Apochromat oil objective, Carl Zeiss). ZEN 2009 software (Carl Zeiss) was used to process images and render z-stacks in 3D. Photoshop CS6 (Adobe) was used to adjust input levels so that the signal spanned the entire output grayscale and to adjust brightness and contrast.

The fluorescence intensity ratio of Cno puncta at TCJs and bicellular junctions was measured from the sum of projections taken through 0.9 µm of the SAJ. For mid-cellularization this equated to puncta 2.4-3 µm below the apical surface and 3.6-4.5 µm for late cellularizing embryos. Total fluorescence intensity was determined in ImageJ (National Institutes of Health, Bethesda, MD, USA), taken from integrated density corrected for background. Measurements of puncta length along the longest principal axis were made using Imaris 8.0 software (Bitplane). MIPs were generated by acquiring z-stacks through the embryo with a 0.3 µm step size and digital zoom of 2. ZEN 2009 software was used to crop stacks to 250×250 pixels along the xy-axis and to project xyz-stacks along the y-axis as previously described (Choi et al., 2013). The apical-basal position of puncta was determined from MIPs using ImageJ. Projections were rotated 90° counterclockwise and analyzed using the Plot Profile function of ImageJ to generate values of average fluorescence intensity along the apical-basal axis. These data are also displayed as heat maps, illustrating intensity along the apical-basal axis with a color gradient. Graphs and accompanying heat maps were generated using GraphPad Prism 7.0.

Expression levels of Cno transgenic proteins and knockdown efficiency of Rap1 and Cno were determined by western blotting. Embryo lysates were prepared by grinding dechorionated embryos in ice-cold lysis buffer [1% NP-40, 0.5% Na deoxycholate, 0.1% SDS, 50 mM Tris pH 8, 300 mM NaCl, 1.0 mM DTT and Halt protease and phosphatase inhibitor cocktail (Thermo Scientific)]. Lysates were cleared at 16,000 g and protein concentration determined using the Bio-Rad Protein Assay Dye. Lysates were resolved by 8 or 12% SDS-PAGE, transferred to nitrocellulose filters and blocked for 1 h in 10% dry milk powder in PBS-Tw (0.1% Tween 20 in PBS). Membranes were incubated in primary antibody for 2 h (see Table 2 for antibody concentrations), washed four times for 5 min each in PBS-Tw, and incubated with IRDye-coupled secondary antibodies for 45 min. Signal was detected over a 4-log dynamic range with the Odyssey CLx infrared imaging system (LI-COR Biosciences). Band densitometry was performed using ImageStudio software version 4.0.21 (LI-COR).

We thank W. Choi for developing the anti-Rap1 and anti-Baz antibodies, R. Reuter for key Drosophila stocks, M. Price and P. Ariel for advice on 3D imaging, M.P. lab members including A. Spracklen for helpful advice and comments, and J. Grosshans for helpful discussions.