ELMO and Sponge specify subapical restriction of Canoe and formation of the subapical domain in early Drosophila embryos

Cortical domains (apical, subapical, lateral, basal) are a characteristic feature of epithelial cells and are crucial for their diverse functions (Hermiston and Gordon, 1995; Martín-Belmonte et al., 2008; Stephenson et al., 2010). Cortical domains are defined by integral membrane proteins, such as Crumbs and E-Cadherin, as well as membrane-associated proteins such as the Par proteins (Chalmers et al., 2005; Cohen et al., 2004; Izumi et al., 1998), in addition to lipid composition (reviewed by Gassama-Diagne and Payrastre, 2009). In many cases, the separation of the domains is maintained by mutual exclusion of determinants, such as the lateral exclusion of Bazooka (Baz; also known as Par-3) by Par-1 and Scribbled (Scrib or Scrb) (Benton and Johnston, 2003; Bilder et al., 2003). By contrast, the mechanisms for initial establishment and arrangement of cortical regions are less clear.

During embryonic development of Drosophila, epithelial cells emerge at the cellular blastoderm stage in a process called cellularization. Following a stage of syncytial development that includes 13 nuclear cycles, ∼6000 cortical nuclei are synchronously enclosed into individual cells in interphase 14 as the plasma membrane invaginates between adjacent nuclei (reviewed by Foe et al., 1993). Cellularization leads to a monolayered columnar epithelium with four distinct cortical regions and adherens junctions positioned at the subapical region. Prior to cellularization, only two cortical regions can be differentiated, namely the cap and intercap regions in syncytial embryos (Warn et al., 1980, 1984). During mitosis of the nuclear cycles, the spindles are separated by a metaphase furrow up to 10 µm deep (Sherlekar and Rikhy, 2017). Three cortical regions are found within the metaphase furrow: the apical, lateral and basal regions (Mavrakis et al., 2009). The change of cellular organization and switch from syncytial to cellular development is a central feature of the mid-blastula transition (MBT), in addition to cell cycle remodeling, degradation of maternal RNA and activation of zygotic transcription (reviewed by Farrell and O'Farrell, 2014; Liu and Großhans, 2017; Yuan et al., 2016).

Here, we focus on the subapical region, which, in contrast to the other cortical regions, emerges during cellularization. It has been suggested that the subapical region matures gradually, since Baz, the typical subapical determinant, and the E-Cadherin (Shotgun) complex gradually accumulate during the course of cellularization and are properly positioned only by the end of cellularization (Harris and Peifer, 2004; McGill et al., 2009). Subapical restriction of Baz and E-Cadherin depends on the actin-binding protein Canoe (Cno)/Afadin and its regulator, the Rap1 GTPase (Choi et al., 2013; Sawyer et al., 2009). Histological analysis of fixed embryos suggested that Canoe marks the subapical domain prior to Baz (Choi et al., 2013). The timing of initial subapical accumulation of Canoe has not been retraced. As a GTPase, Rap1 is potentially regulated by an upstream guanyl nucleotide exchange factor (GEF) or a GTPase activating protein (GAP). The unconventional GEF complex ELMO (Ced-12)-Sponge (Spg)/DOCK represents a candidate (Geisbrecht et al., 2008; Postner et al., 1992; Winkler et al., 2015). Biochemical analysis indicates that ELMO-DOCK complexes are GEF-specific for Rac and Rap1 (Biersmith et al., 2011, 2015; deBakker et al., 2004; Yajnik et al., 2003; Wu and Horvitz, 1998). Although a role of sponge in the organization of actin caps in syncytial embryos was reported some time ago (Postner et al., 1992), a function in the context of the subapical domain or of Canoe has not been investigated.

In this study, we show by live imaging that the cortical markers Canoe, Scribbled and Slam segregate during the first minutes of cellularization. Analyzing the mutant phenotypes of ELMO (Ced-12) and sponge, we identify them as an upstream factor in Canoe localization. We report the dynamics of Sponge and ELMO protein in fixed and live embryos. Sponge and ELMO mark the apical regions, which are characterized by strong F-actin staining (ʻactin caps') of the plasma membrane during the syncytial blastoderm stage prior to cellularization. Strikingly, this localization pattern changes at the onset of cellularization. ELMO and Sponge redistribute to the rims of the actin caps and surround the apical regions and thus generate the information for a new region. Essentially, the distribution changes from a disc-like to a ring-like pattern. Based on this dynamic, we propose a model for pattern formation in which the subapical domain is positioned adjacent to the apical domain.

Stereotypic and stage-specific changes in cortical organization are linked to early embryonic development (Nance, 2014). In Drosophila, a uniformly structured cortex is characteristic for the preblastoderm stage (Karr and Alberts, 1986). When the first nuclei reach the cortex in nuclear cycle 9, their associated centrosomes induce cortical differentiation and segregation of cortical markers into disc-like caps or cytoplasmic buds, rich in F-actin, and the region between the caps (intercap), which is marked by Slam and Toll (Fig. 1A) (Mavrakis et al., 2009; Raff and Glover, 1989; Warn et al., 1984). The cortex is further differentiated during syncytial mitoses. By immunostaining, we detected three cortical domains: Slam at the furrow tip, Discs large (Dlg, or Dlg1) at the furrow, and Canoe apically and at the furrow (Fig. 1B).

An important feature of MBT and the transition from syncytial to cellular development is the emergence of a subapical domain and of a typical epithelial cortex with four domains. The subapical domain is inserted between the apical and lateral domains. Canoe is the most upstream acting subapical marker (Choi et al., 2013). During syncytial mitoses. We detected Canoe at the apical and lateral regions. Staining was excluded from the basal domain (Fig. 1C). The cortical distribution profoundly changes in interphase 14, when Canoe became restricted to the subapical domain and segregated from the lateral marker Dlg (Fig. 1D,E). In contrast to Canoe, other subapical markers such as Baz, Par-6, aPKC and E-Cadherin accumulate gradually at the subapical region during the course of cellularization (Harris and Peifer, 2004; McGill et al., 2009; Wang et al., 2004).

To achieve a precise timing and reveal the dynamics of subapical accumulation of Canoe, we conducted time-lapse imaging of embryos expressing CanoeYFP from its endogenous locus in comparison with the lateral marker ScribbledGFP or the basal marker CherrySlam (Fig. 2, Figs S1 and S2). As we reported previously (Acharya et al., 2014), Slam continuously marks the transition from the retracting metaphase furrows to cellularization furrows (ʻold' furrow). Within a few minutes, Slam accumulates at the ʻnew' furrow. For comparative timing, we refer to the formation of the furrow between corresponding daughter nuclei and to the retraction of the metaphase furrow. During telophase, the retracting metaphase furrow (old furrow) encloses corresponding daughter nuclei, while a shallow groove (new furrow) emerges between them (Fig. 2A).

We found that CanoeYFP (Fig. 2A,B, t=3 min) and ScribbledGFP (Fig. 2A,C, t=1 min) were uniformly dispersed following retraction of the metaphase furrow. Within a few minutes, however, a restriction of both markers became apparent at the prospective furrow (Fig. 2B,C, t=7 min). CherrySlam accumulated slightly later than CanoeYFP at the new furrow. A clear signal was observed starting at t=8-12 min (Fig. 2B,D,E). This difference between CanoeYFP and CherrySlam was also observed along the apical-basal axis. CanoeYFP reached high levels at a position between 2 and 3 µm (Fig. 2D,E). CherrySlam showed the strongest signal a few minutes later at a position of ∼4 µm (Fig. 2D,E). The appearance of CherrySlam in the basal layer is probably linked to the invagination of the new furrow. These data describe a clear segregation of the subapical and basal markers already at t=8 min. This represents the earliest time that the two markers are clearly separated. We also observed a difference in the accumulation of the lateral marker ScribbledGFP. We detected a strong signal at position 3-5 µm (Fig. 2F), which was more basal than the peak of CanoeYFP. Domain restriction and furrow formation also became obvious when considering the width of the new furrow (Fig. 2G, Fig. S3). Within 10 min, the width of the CanoeYFP region was gradually reduced from 2.5 µm to 0.5 µm. In summary, our live imaging analysis shows that the cortical domains form within ∼10 min after exit from mitosis 13. Analyses of both fixed and live embryos show that Canoe marks the subapical domain throughout cellularization, suggesting that the subapical domain forms together with the basal and lateral domains and does not gradually arise during cellularization.

Staining and live analysis revealed a mutually exclusive distribution of Canoe, Scribbled and Slam. We examined whether this pattern depended on the function of these proteins. We first confirmed previous reports that membrane association of Canoe depends on Rap1 (Choi et al., 2013) (Fig. 3A,B). By contrast, subapical Canoe restriction depended neither on slam (Fig. 3E) nor scribbled (Fig. 3D). We stained embryos from scribbled or slam germline clones (hereafter referred to as scribbled and slam mutants) for Canoe and the lateral marker Dlg. In scribbled mutants, we detected subapically restricted Canoe (Fig. 3D). Consistent with previous reports on the mutual dependency of Scribbled and Dlg (Bilder et al., 2003), Dlg was spread over and loosely associated with the membrane in scribbled mutants. Similarly, Canoe was subapically restricted in slam mutants (Fig. 3E). Although furrow invagination is impaired, furrows are specified in slam mutants (Acharya et al., 2014). These data show that the initial accumulation of the subapical marker Canoe does not depend on scribbled and slam, and that independent pathways at the level of canoe, scribbled and slam might define the respective cortical domains.

The GTPase Rap1 is presumably controlled by GEFs or GAPs, such as the GEF protein Dizzy (Dzy; also known as PDZ-GEF) (Huelsmann et al., 2006; Spahn et al., 2012) or the heteromeric GEF complex ELMO-Sponge (Fig. 4E) (Biersmith et al., 2011; Yajnik et al., 2003). To test this hypothesis, we analyzed dizzy mutant embryos for subapical restriction of Canoe. In embryos from dizzy germline clones, we did not detect a deviation of Canoe staining as compared with wild-type embryos (Fig. 3C). By contrast, Canoe was spread along the cellularization furrow in embryos from ELMO germline clones (Fig. 5B,D) and embryos from sponge mutant females (Fig. 5C); these embryos are hereafter designated as ELMO and sponge embryos. ELMO embryos passed through a normal number of nuclear divisions, but failed to form actin caps and metaphase furrows during syncytial cycles (Fig. 4B) (Winkler et al., 2015). Associated with this morphological phenotype were chromosomal segregation defects and subsequent fallout of nuclei. In interphase 14, they cellularized at the anterior and posterior termini but not in the medial region (Fig. 4C). We observed a similar phenotype in sponge embryos, which is consistent with a previous report (Postner et al., 1992).

Canoe spread along the cellularization furrows in ELMO and sponge embryos, whereas the lateral marker Dlg and the basal marker Slam localized normally (Fig. 5B,C). We quantified the distribution of Canoe along the cellularization furrow by plotting the relative fluorescence along the apical-basal axis. We did not compare absolute protein levels. Whereas restriction to the subapical domain was observed in wild-type embryos (Fig. 5A,D), Canoe was spread all along the furrow in ELMO embryos (Fig. 5B,D). These data show that ELMO and sponge are required for subapical restriction of Canoe.

Functioning upstream of Canoe, the ELMO-Sponge complex might confer positional information for the subapical domain, in that ELMO or Sponge would localize to the prospective subapical region no later than when subapical restriction of Canoe is observed. To test this hypothesis, we analyzed the dynamics and localization pattern of ELMO and Sponge. Since a suitable antibody was available (Biersmith et al., 2011), we fixed and stained wild-type embryos for Sponge (Fig. 6). We detected Sponge at the caps in syncytial cycles. During syncytial mitoses, uniform Sponge staining was detected at the apical and lateral membranes. A strikingly different pattern was observed in embryos during cellularization. We detected a staining pattern clearly restricted to the subapical domain and largely separated from the lateral marker Dlg throughout cellularization (interphase 14). The subapical restriction of Sponge was not as clear as that observed for Canoe, however. Importantly, Sponge staining was depleted in the apical region, leading to a grid-like pattern in the surface view, even when the apicalmost layers were included in the projections (Fig. 6).

The subapical staining pattern of Sponge did not depend on scribbled, canoe or Rap1. We detected subapical restriction of Sponge in scribbled, Rap1 and canoe germline clones (hereafter scribbled, Rap1 and canoe mutants) (Fig. 7A-D). These data indicate that Sponge functions independently and possibly upstream of the Rap1-Canoe pathway. In contrast to Rap1, canoe and scribbled, the subapical localization of Sponge depends on ELMO. When we stained ELMO mutants for Sponge, the Sponge signal was uniformly distributed and no membrane association was detected (Fig. 7E). The loss of membrane association and restricted localization are consistent with the structure of the ELMO-Sponge complex, in which ELMO provides membrane binding and Sponge the GEF activity (Fig. 4E) (Komander et al., 2008).

To record the dynamics of the ELMO-Sponge complex at high temporal resolution, we generated an ELMO-GFP fusion protein expressed from a genomic transgene (Fig. 4A). The ELMO-GFP fusion protein was expressed at comparable levels as from the wild-type allele (Fig. 4D) and was functional, as it rescued the germline clone phenotype of ELMO. During syncytial cycles, we detected ELMO-GFP at the caps (Fig. 8A). With the onset of cellularization, the localization pattern of ELMO-GFP changed profoundly from the disc-like pattern in syncytial interphases to a ring-like pattern (Fig. 8B,C, arrowheads; Fig. S4). ELMO-GFP signal at new furrows was first detected at about the time when CanoeYFP accumulates at the apical region (Fig. 2). We quantified the relative fluorescence intensity along the apical-basal axis of new furrows and compared the profile to CherrySlam fluorescence (Fig. 8D). We observed an early subapical accumulation and subsequent restriction of ELMO-GFP, whereas CherrySlam accumulated a few minutes later at a basal position.

In this study, we have focused on analysis of the functional relationships of cortical domains and their associated proteins in terms of the insertion of a new domain at a specific time and place. Based on in vivo and in vitro studies of ELMO-Sponge/DOCK, Rap1 and Canoe/Afadin in other experimental systems, we have some understanding of their biochemical activities (Biersmith et al., 2011; Boettner et al., 2003; Sawyer et al., 2009). In biochemical assays, the ELMO-Sponge complex specifically activates Rac and Rap1 among the small GTPases (Yajnik et al., 2003). ELMO-Sponge might activate Rap1 in the subapical region during the onset of cellularization (Fig. 9). As a consequence, activated Rap1-GTP may restrict Canoe to the subapical domain. We analyzed the distribution of Rap1 employing Rap1-GFP as a proxy. As we detected a wide and uniform distribution of Rap1-GFP over the plasma membrane (Fig. S5), Rap1 distribution cannot be used as a test for whether ELMO-Sponge would activate Rap1. Alternatively, ELMO-Sponge might act on Canoe independently of Rap1 (Fig. 9), as we have not demonstrated that Rap1 links ELMO-Sponge and Canoe or that Rap1 is activated by ELMO-Sponge.

During the course of cellularization, Canoe will gradually recruit the polarity protein Baz (Par-3) and subsequently the E-Cadherin complex to this domain in order to make it into a generic subapical domain. When comparing their distributions, Canoe is more clearly localized than ELMO-Sponge. This difference might be due to the signaling process. In the simple model, the pathway initiated by ELMO-Sponge via Rap1 and Canoe is linear. We do not rule out reinforcing feed-back interactions within the pathway that lead to enhanced signals. Such feedback interactions are likely to be important for maintenance of the subapical domain. For example, Baz influences Canoe localization later in cellularization (Choi et al., 2013). In the accompanying paper, Bonello et al. (2018) describe a function of dizzy in the apical restriction and especially localization of Canoe to tricellular junctions via activation of Rap1 during late cellularization.

When studying the initial formation of the subapical domain, we have not observed an influence of the lateral determinant Scribbled or the basal protein Slam. Such interactions are likely to be important later in cellularization and development for maintenance of the cortical domains or sharpening of the boundaries, as it is well established that lateral and subapical components interact by mutual exclusion (Bilder et al., 2003; Tanentzapf and Tepass, 2003).

The change in ELMO-Sponge distribution from a disc-like pattern during the nuclear cycles to a ring-like pattern during initial cellularization suggests a model for the origin of positional information for the emerging subapical domain (Fig. 9A). The organization in cap and intercap regions already contains the information for a third domain, namely the interface between the two regions (Fig. 9B). The dynamic localization pattern of ELMO-Sponge makes use of this information in initial cellularization. Whereas uniformly distributed within the caps during syncytial cycles, ELMO-Sponge accumulates at the rims of the apical region during the onset of cellularization. Subsequently, when the furrows invaginate, adjacent rings around the apical regions meet to form a grid-like pattern.

The determinants for specific membrane localization of ELMO-Sponge are not clear. Within the ELMO-Sponge complex (Fig. 4E), membrane association is assumed to be conferred by ELMO. The role of the conserved ELMO domain is unclear (Komander et al., 2008). The PH domain might mediate membrane association by binding phospholipids. However, no corresponding subapical localization pattern of phospholipids has been reported. Sensor proteins for PI(4,5)P₂ and PI(3,4,5)P₃ were reported to be widely distributed along the invaginating furrow (Reversi et al., 2014). As an alternative to phospholipids, a membrane protein might serve as an anchor for ELMO.

The change of ELMO-Sponge distribution from a disc-like to a ring-like pattern coincides with the MBT and thus might ultimately depend on one or more zygotic genes, on cell cycle regulators such as checkpoint kinases, which change their activity during MBT, or other MBT-associated processes (reviewed by Liu and Großhans, 2017). Among the early zygotic genes, the best candidate for contributing to ELMO-Sponge redistribution might be the zygotic gene dunk, which controls apical myosin contractility and flow at the onset of cellularization. However, a potential role of dunk or other zygotic genes in cortical domain formation and segregation has not been analyzed (He et al., 2016). Future experiments analyzing the detailed dynamics and factors controlling the localization of ELMO-Sponge are likely to provide insight into the underlying molecular and biophysical mechanism.

Fly stocks were obtained from the Bloomington Drosophila Stock Center, unless otherwise noted. The following fly strains and mutations were used: UASp-CherrySlam driven by maternal Gal4 (Acharya et al., 2014), Df(2L)slam (Acharya et al., 2014), Ced-12/ELMO³⁶⁷ (Winkler et al., 2015), cno[R2] (Choi et al., 2013), CanoeYFP (PBac{602.P.SVS-1}cno^CPTI000590, Drosophila Genomics and Genetic Resources, Kyoto), scrb-GFP^CA07683 (Buszczak et al., 2007), scrb[1] (Bilder and Perrimon, 2000), rap1[P5709] (R. Reuter, University of Tübingen, Germany; Knox and Brown, 2002), sponge[242] (Postner et al., 1992), Df(3R)3450 (deficiency uncovering sponge), dizzy(Δ8) (R. Reuter; Huelsmann et al., 2006) and GFP-Rap1 (E. Knust, Max Planck Institute, Dresden, Germany; Knox and Brown, 2002). ELMO-GFP transgenes were generated according to standard protocols by φC31 integrase-mediated site-specific insertions in the landing site ZH-86Fb (Bischof et al., 2007). All fly cages and crosses were maintained by standard methods at 25°C unless otherwise specified. Germline clones were induced by heat shock (each 1 h, at 24-48 h and 48-72 h of development) of first and second instar larvae and selected by ovoD transgenes on corresponding Frt chromosomes.

The genomic ELMO-GFP construct was generated by ligation of multiple fragments, which were amplified by PCR. Fragment 1-GFP (optimized for Drosophila codon usage), using upstream primer (5′-3′) ZL95 (tggcggccgcACTGGAAGTTCTGTTCCAGGGGCCCGGCTCCGCCGGCTCC), introducing a NotI site and a PreScission cleavage site; and downstream primer ZL75 (TAAAGCTTGTACAGCTCGTCCATGCC), inserting a HindIII site. Fragment 2-ELMO3′ with a leading stop codon (chromosome 2L, 12,100543-12,101,303), using upstream primer ZL76 (ACAAGCTTTAAGCATAACGAGCACAATTAC), adding a stop codon and HindIII site; and downstream primer ZL73 (atctcgaGTCTGCCTGCCGGACCGG), adding an XhoI site to the 3′ end. Both fragments were cloned (NotI/XhoI) into pBSK (Stratagene) to create BSK-GFP-ELMO3′ and transferred to the transformation vector pAttB (Bischof et al., 2007) to create pAttB-GFP-ELMO3′. Fragment 3, a 3633 bp genomic fragment encompassing the 5′ region of the ELMO locus to the 3′ end of ELMO exon 4 (chromosome 2L, 12,096,904-12,100,537). NotI sites were introduced on both ends by upstream primer ZL99 (AACAGATCTGCGGCCGGAAGACAAGCGATCGGATGC) and downstream primer ZL104 (ACTTCCAGTgcggccgcgCTCTCAAAGCAAAAATCATAG). Fragment 3 was cloned into the NotI site of pAttB-GFP-ELMO3′ leading to the final transformation plasmid pAttB-ELMO-GFP-ELMO3′, which comprises DNA from the ELMO locus region (breakpoints are 12,096,904-12,101,303) with a PreScission cleavage site and GFP in front of the stop codon.

Embryos were fixed with 4% formaldehyde or heat fixed according to standard procedures as described previously (Großhans et al., 2005) and stored in methanol. Fixed embryos were rinsed three times in PBS with 0.1% Tween 20 (PBT), and blocked in 5% bovine serum albumin (BSA) in PBT for 1 h at room temperature. Blocked embryos were incubated with primary antibodies in 0.1% BSA in PBT overnight at 4°C or at room temperature for 2 h with constant rotation. After rinsing three times and washing four times for 15 min each with PBT, the embryos were incubated with fluorescently labeled secondary antibodies in PBT for 2 h at room temperature. Following another round of rinsing and washing, embryos were stained with DAPI (0.2 µg/ml) for 10 min, rinsed three times in PBT, washed in PBT for 10 min and mounted in Aqua-Poly/Mount (Polysciences). Embryos for phalloidin staining were fixed in 8% formaldehyde and the vitelline membrane was manually removed. Primary antibodies were rabbit anti-Canoe (1:1000; Choi et al., 2013), mouse anti-Dlg (1:100, 4F3, Developmental Studies Hybridoma Bank), rabbit/guinea pig anti-Slam (1:5000; Brandt et al., 2006), guinea pig anti-Sponge (1:1000; Biersmith et al., 2011). Phalloidin coupled to Alexa Fluor 488 was obtained from Thermo Fisher. Secondary antibodies were coupled to Alexa Fluor 488, 568 and 647 (1:500, Thermo Fisher).

Western blotting was conducted as previously described (Wenzl et al., 2010). Briefly, lysis samples corresponding to ten staged embryos (0-2 h) were separated by SDS-PAGE and transferred by wet transfer to a nitrocellulose membrane. Blots were imaged with an Odyssey CLx infrared imaging system (LI-COR Biosciences) with 16-bit depth. Primary antibodies were goat anti-ELMO (1:3000; Biersmith et al., 2011), mouse anti-α-Tubulin (1:50,000, B512, Sigma). Secondary antibodies were 800CW- or 680CW-coupled donkey anti-guinea pig/mouse/rabbit IgG (Invitrogen). Images were processed in Adobe Photoshop and Fiji/ImageJ (Schindelin et al., 2015).

Live embryos were handled as previously described (Kanesaki et al., 2011). Fluorescent time-lapse movies and images of fixed embryos were recorded with a Zeiss LSM780 confocal microscope equipped with Airyscan detection, and objectives LCI Plan Neofluar 63×/water NA 1.3 for fixed samples, Plan Neofluar 63×/oil NA 1.4 for live imaging. Movies of embryos expressing CanoeYFP and CherrySlam were obtained with a frame size of 256×256 pixels (28.7×28.7 µm) and a lateral pixel size of 110 nm at an interval of 60 s. Channels were changed after recording of every z-stack; each z-stack had 19 slices with a step size of 0.5 µm. Embryos expressing ScribbledGFP were imaged with Airyscan detection with a frame size of 488×488 pixels (36×36 µm) and a lateral pixel size of 73 nm. Each z-stack included 11 slices with a step size of 1 µm, imaged at an interval of 60 s. Embryos expressing ELMO-GFP and CherrySlam were imaged with Airyscan detection with a frame size of 488×488 pixels (32×32 µm) with a lateral pixel size of 66 nm. Each z-stack contained 17 slices with a step size of 0.5 µm, obtained at an interval of 60 s. Images of embryos expressing GFP-Rap1 were obtained with Airyscan detection with a frame size of 476×476 pixels (32.1×32.1 µm; 67.5 nm lateral pixel size). z-stacks were conducted with a step size of 0.2 µm and orthogonal views were conducted with Fiji/ImageJ as well as measurement of furrow length. Fixed embryos were imaged with a frame size of 512×512 pixels (67.5×67.5 µm; 130 nm lateral pixel size) for top views and 512×200 pixels (96.4×29.4 µm; 190 nm lateral pixel size) for sagittal views. Images were processed with Fiji/ImageJ and Adobe Photoshop and Adobe Illustrator.

Measurements were conducted with Fiji/ImageJ, with calculation in Microsoft Excel. For the quantifications in Fig. 2E,F and Fig. 8D, fluorescence intensities were measured along three furrows in each z position along the apical-basal axis. The maximal intensities for each protein were normalized to 1 and plotted as a graph with apical-basal position on the y-axis and normalized fluorescence intensities on the x-axis. The heatmap in Fig. 5D was prepared by measuring the fluorescence intensity distribution along the furrows in side views in a total of nine furrows in three embryos. Intensities were normalized to 1 for every furrow and displayed as heatmaps using the conditional formatting function in Excel. Averages of normalized intensities were plotted with apical-basal position on the y-axis and normalized fluorescence intensities on the x-axis. For quantifications in Fig. 2G, the distribution of CanoeYFP fluorescence intensity at one furrow was measured in top view of live images at different time points using the line plot function of Fiji/ImageJ. Position zero on the x-axis was defined by the peak of the curve at the latest time point.

We are grateful to E. Geisbrecht, D. Bilder, M. Peifer, R. Reuter and E. Knust for materials or discussions. We acknowledge service support from the Developmental Studies Hybridoma Bank created by the NICHD of the NIH, USA, and maintained by the University of Iowa; the Bloomington Drosophila Stock Center (supported by NIH P40OD018537); the Drosophila Genomics and Genetic Resources, Kyoto; the BACPAC Resources Center at Children's Hospital Oakland; and the Drosophila Genomics Resource Center at Indiana University (supported by NIH 2P40OD010949-10A1).