Repulsion by Slit and Roundabout prevents Shotgun/E-cadherin–mediated cell adhesion during Drosophila heart tube lumen formation

Lumen morphogenesis is an essential process during the formation of vascular tissue. Blood vessels are tubes formed by a single layer of polarized endothelial cells with junctional contacts enclosing a central lumen (Davis et al., 2000). Although several models have been proposed for lumen formation (Lubarsky and Krasnow, 2003), the cellular and molecular mechanisms that drive this process are poorly understood. The Drosophila melanogaster embryonic heart, a simple linear tube resembling a vertebrate capillary, provides a straightforward genetic model for vessel morphogenesis and lumen formation.

Several studies, both in vitro and in vivo, show that the vascular lumen is established via the formation of intracellular vacuoles, which fuse together with the plasma membrane to produce a lumen between cells (Vega-Salas et al., 1987; Davis and Camarillo, 1996; Beitel and Krasnow, 2000; Kamei et al., 2006). Presumably, dynamic changes in cell adhesion and cell–cell contacts must also be regulated to account for morphogenetic changes accompanying this process. The cadherin family of molecules (Carthew, 2005) plays an important role in vessel formation via both their adhesive and signaling functions (Dejana et al., 1999). In vertebrate endothelial cells, VE-cadherin is localized specifically to sites of cell contact known as adherens junctions (Bach et al., 1998; Vincent et al., 2004). Recent evidence suggests the adhesive function of cells is dynamically regulated through E-cadherin (E-Cad) endocytosis (Gumbiner, 2000). However, what remains unclear is how intracellular cell adhesion mechanisms are regulated by extracellular signals to control complex cell shape changes.

Slit is an extracellular matrix protein that is the ligand for the Roundabout (Robo) family of transmembrane receptors (Rothberg et al., 1990; Brose et al., 1999; Kidd et al., 1999). Genetic data from D. melanogaster and Caenorhabditis elegans together with in vivo studies in mice indicate a conserved role for Slit and Robo proteins in repulsive axon guidance (for review see Dickson and Gilestro, 2006). In the D. melanogaster central nervous system, expression of slit by the midline glial cells creates a repulsive cue for robo-expressing central nervous system axons (Kidd et al., 1999). Here, we present genetic evidence for a repulsive activity for Slit in the heart. Specifically, we show that Slit/Robo signaling is required at the apical plasma membrane of heart cells for inhibiting E-Cad–mediated cell adhesion, thus permitting cell shape changes required for lumen formation.

The alignment of cardioblasts (CBs) into rows on either side of the midline is the first of several steps during heart tube assembly. After alignment, the two rows of CBs, which are each flanked by a row of pericardial cells (PCs), migrate dorsally. Prior to merging at the midline, CBs undergo a mesenchyme-to-epithelium transition and acquire apical-basal polarity (Fremion et al., 1999). At the dorsal midline, contralateral pairs of CBs make specific dorsal and ventral contacts between their opposing apical membranes to form the lumen (Fig. 1, D and H). To better understand the steps leading to lumen formation, we performed EM on cross sections (XSs) of wild-type embryos at three steps during the late stages of heart development. At early embryonic stage 16, each CB initiates contact with its contralateral counterpart at the dorsalmost leading edge of the apical membrane (Fig. 1 A). After dorsal contact, CBs undergo a shape change and make contact at their ventral apical surfaces (Fig. 1 B). In this way, a lumen is formed between two opposing CBs (Fig. 1 C). The fact that CBs specifically make contact at dorsal and ventral attachment points (Fig. 1 D) suggests that an inhibitory mechanism may prevent the centralmost apical surfaces between these points from coming into contact.

During the early stages of heart morphogenesis (embryonic stage 14), Slit is uniformly distributed on CB surfaces (Qian et al., 2005; Santiago-Martínez et al., 2006). We reported that Slit and its receptors Robo and Robo2 play an important role in the dorsal migration of CBs and PCs (Santiago-Martínez et al., 2006). Prior to merging at the dorsal midline, Slit becomes restricted to CB apical surfaces (Santiago-Martínez et al., 2006). This dynamic change in Slit localization suggests that during this late phase of heart morphogenesis, Slit has a second function in lumen formation.

To explore the function of Slit and Robos in lumen formation, we performed a detailed analysis of Slit, Robo, and Robo2 localization in the heart, focusing on stage 17 after CB migration has occurred. Whole mount antibody staining revealed that Slit's localization is polarized at the apical surface of CBs (Fig. 1 E). Robo is also polarized at CB apical surfaces (Fig. 1 F). Robo2 is absent from the CBs and is restricted to the two rows of PCs, which flank the CBs (Fig. 1 G). To precisely localize these proteins, we performed immunohistochemistry on embryos that we examined in XS. Both Slit and Robo are localized to the CB apical surfaces surrounding the central lumen of the heart (Fig. 1, I and J). XSs also confirm that Robo2 is localized to PCs and excluded from the lumen (Fig. 1 K). The concurrent expression of Slit and Robo at CB apical surfaces suggests that these proteins may play an important role in lumen formation.

A notable aspect of heart tube formation is that although the dorsal and ventral leading edges of CB apical surfaces make connections with their contralateral partners, the innermost surfaces fail to come in contact (Fig. 1 D). Based upon Slit's localization at the CB apical domain and Slit's well-characterized role in repulsive guidance (Brose et al., 1999; Kidd et al., 1999; Kramer et al., 2001), we hypothesized that Slit may be responsible for preventing CB membranes from coming into contact, permitting the formation of a lumen. To test this idea, we used the Gal4-UAS system (Brand and Perrimon, 1993) to overexpress slit in the heart. Using the Mef-Gal4 driver to drive UAS-slit in a slit mutant background (slit gain of function [GOF]), we achieved rescue of the heart cell positioning phenotype in which the CBs fail to migrate to the dorsal midline of the embryo (Santiago-Martínez et al., 2006). However, at stage 16, Slit is no longer restricted to the apical surface of CBs and is now also localized basally and on the PCs as well (Fig. 2 B). In slit GOF embryos, Robo is now also mislocalized to CB basal surfaces (Fig. 2 D). Examination of the heart in slit GOF embryo XSs by EM revealed two lumens (Fig. 2 F) rather than the single lumen observed in the wild type (Fig. 2 E). We observed this phenotype in all of the embryos we sectioned (n = 9). The ectopic lumens were not continuous along the entire length of the heart tube but rather were confined to the pair of CBs in the section. By examining serial sections, we observed a series of disconnected lumens. Both lumens are lined with extracellular matrix, a recognizable feature in EM and shown to line the D. melanogaster heart (Fig. 2, G and H; Haag et al., 1999). To identify between which cell types ectopic lumens were forming, we performed Mef2 staining on slit GOF embryos that we examined in XS. Mef2 labels CB nuclei, and in wild-type embryos, the lumen is visible between these cells (Fig. 2 I). In slit GOF embryos, ectopic lumens are formed between CBs and neighboring non-Mef2–positive PC or amnioserosal cells (Fig. 2 J). These results demonstrate that when Slit is no longer polarized, CBs are impaired in their ability to confine formation of the lumen to their apical domains.

We hypothesize that Slit localization correlates with lumen formation because of the selective repulsion of CB membranes. However, an alternative explanation may be that lumen defects in slit GOF embryos are secondary to general defects in CB polarity. In slit embryos, the rows of CBs do not assemble properly and exhibit abnormal localization of several cell polarity markers (Qian et al., 2005; Santiago-Martínez et al., 2006). To confirm that the slit GOF phenotype was not secondary to defects in cell polarity, we examined cell polarity markers in these embryos. Discs-large (Dlg), a membrane-associated guanylate kinase protein (Woods et al., 1996), normally localizes to the apical-lateral surface of CBs in stage 16 embryos (Fig. S1 A, available). In slit GOF embryos, this pattern is not significantly altered (Fig. S1 B), which indicates that the CBs are correctly polarized. In addition, the localization of α-spectrin, a membrane-bound cytoskeletal protein (Lee et al., 1993) that preferentially localizes to the basal-lateral surface of CBs (Santiago-Martínez et al., 2006), was normal in slit GOF embryos (Fig. S1 G).

In embryos mutant for slit or both robo and robo2, the rows of CBs do not properly align, resulting in gaps at the dorsal midline (Santiago-Martínez et al., 2006). It has been found that slit mutant embryos have lumen defects (MacMullin and Jacobs, 2006). However, because of the severe CB alignment phenotype in slit or robo,robo2 mutants, we found it difficult to separate slit's earlier role in ipsilateral CB cell alignment with what we believe to be a later role lumen formation between contralateral CBs. Embryos mutant for robo alone have very mild migration defects, and the majority of CBs are able to align at the dorsal midline (Santiago-Martínez et al., 2006). EM of robo mutants (n = 11) reveals that contralateral CBs are inappropriately adhered, and the lumen fails to form between these cells (Fig. 3 D). In addition, Slit is no longer concentrated at CB apical domains (Fig. 3 C), which suggests that the Robo localization is important for the polarized accumulation of Slit. Moreover, the mislocalization of Slit in a robo mutant is not sufficient to induce ectopic lumen formation as in slit GOF embryos (Fig. 2 F), which suggests that Robo is required for this process. To further explore the functional connection between Slit and Robo, we show that removal of robo in slit GOF embryos completely suppresses the slit GOF phenotype and blocks lumen formation (Fig. 3, E and E′). Finally, we found that Dlg and α-spectrin are properly localized in robo mutants (Fig. S1, E and J), providing additional evidence that during lumen formation, Slit and Robo are not required for CB cell polarity but instead function at a later step once the initial polarity of the cell has already been established.

One of the only genes known to affect lumen formation in the D. melanogaster heart is shg, which encodes E-Cad. (Tepass et al., 1996; Uemura et al., 1996). It was previously shown that E-Cad is expressed by the CBs, and that in shg mutants, the rows of CBs align but fail to attach to each other across the midline (Haag et al., 1999). We further investigated these findings by first examining the localization of E-Cad in cross-sectioned embryos. In stage 17 embryos, E-Cad is concentrated at the dorsal and ventral sites of contact between opposing pairs of CBs (Fig. 3 F). Next, we examined the lumen in shg/E-Cad mutants by EM (n = 2). In shg/E-Cad mutants, CBs fail to attach and form a lumen (Fig. 3 G), and we observed the presence of extracellular space between contralateral CBs (Fig. 3 H).

The loss of CB adhesion in shg/E-Cad mutants (Fig. 3 H) is interesting in comparison to robo mutants, in which CBs are inappropriately adhered along the entire apical face (Fig. 3 H′). These results suggest that robo and shg/E-Cad play opposing roles in lumen formation. Robo functions to prevent the CBs from sticking together at the central apical surfaces of the CBs, whereas E-Cad is required to maintain cell adhesion dorsally and ventrally. To further investigate their combined functions, we tested for genetic interactions by examining the lumen in embryos transheterozygous for robo and shg/E-Cad. Although genetic interactions between slit and shg/E-Cad during the earlier step of CB alignment have been explored (Qian et al., 2005), it remains to be shown whether shg/E-Cad interacts specifically with robo during lumen formation. Embryos that were missing a single copy of either robo or shg/E-Cad did not have obvious lumen defects (unpublished data). However, in 89% of robo/shg transheterozygotes (n = 9), the lumen was misshapen (Fig. 3 I), which suggests that the functions of robo and shg/E-Cad are linked.

Next, we took a GOF approach to investigate the respective roles of robo and shg/E-Cad during lumen formation. When we overexpressed shg/E-Cad in the CBs using Mef-Gal4, we observed significant heart defects. However, unlike what we observed in shg/E-Cad mutants where the CBs failed to attach (Fig. 3 G), the failure in lumen formation in shg/E-Cad GOF embryos is caused by the inappropriate adhesion of contralateral CBs. By EM, we observe CBs that are inappropriately attached with no extracellular space in between (n = 9; Fig. 3 J). We also found that Dlg and α-spectrin were properly localized (Fig. S1, D and I), which indicates that the defects were not secondary to loss of CB cell polarity.

robo loss of function (LOF) embryos have a phenotype similar to shg/E-Cad GOF embryos, in which CBs are inappropriately attached (Fig. 3, D and H′). We hypothesize that Robo's function at this stage is to repel contralateral CBs at the central apical domain, enabling the formation of a lumen in this region. Consistent with this idea, we found that overexpression of robo in CBs using Mef-Gal4 results in severe lumen defects. Although CBs were properly aligned (Fig. 3 K) and polarized (Fig. S1, C and H), with EM (n = 5), we observed a loss of cell contact between the dorsal and/or ventral CB apical surfaces (Fig. 3 L). These results indicate that high levels of Robo prevent CBs from initiating or maintaining adhesion at the specific sites of contact and suggest that robo overexpression can overcome E-Cad–mediated cell adhesion at these sites. Interestingly, in robo GOF embryos we found that although Slit remains polarized in CBs, we see Slit staining in intense patches (Fig. 3 K) as compared with more uniform localization in the wild type (Fig. 3 A).

Based upon our findings, we hypothesize that Slit and Robo function to negatively regulate E-Cad–mediated cell adhesion between opposing CBs. Next, we examined the localization of E-Cad in our mutant backgrounds. In the wild type, E-Cad localizes to the specific sites of cell–cell contact between contralateral CBs (Fig. 3 F). The concentration of E-Cad at CB apical membranes can also be seen in dorsal view (Fig. 4 A). In slit GOF embryos where we observe ectopic lumen formation, we noticed changes in E-Cad expression. E-Cad fails to accumulate at high levels at the apical domain in CBs (Fig. 4 B). In XS, ectopic E-Cad accumulation is seen at sites of cell–cell contact surrounding the ectopic lumens (Fig. 4 E). However, in robo mutants, we observed higher–than–wild type levels of E-Cad at the apical membrane as seen in dorsal view and in XS (Fig. 4, C and F). These results are consistent with our EM studies revealing that in robo mutants, the CBs are tightly adhered (Fig. 3 H′). We obtained similar results when we looked at Baz localization (Fig. S1, K–O). Baz is a PDZ domain–containing protein that may provide a landmark for adherens junction assembly by recruiting E-Cad to sites of cell–cell contact (Harris and Peifer, 2005).

The Ena/VASP family of proteins regulate actin dynamics during cell motility (Krause et al., 2003) and, more recently, has been implicated in cell adhesion (Vasioukhin et al., 2000; Grevengoed et al., 2001; Scott et al., 2006). Here, we find that Ena, similar to E-Cad and Baz, is localized at the apical domain of CBs in regularly spaced puncta (Fig. 4, G and J). In robo mutants, Ena is no longer concentrated at CB apical domains (Fig. 4, I and L). Embryos mutant for ena have lumen defects as observed by EM (n = 4) that are consistent with a loss of CB cell contact (Fig. 4 D). What is the function of Ena during lumen formation? One possibility is that, similar to its role during D. melanogaster dorsal closure (Gates et al., 2007), Ena may be required for protrusive behavior of the CBs to permit cell–cell contact. Alternatively, Ena may be required for regulating E-Cad–based adhesion. This hypothesis is supported by our findings that Ena localizes to ectopic sites along CB basal surfaces in slit GOF embryos (Fig. 4, H and K), where we observe ectopic sites of E-Cad localization (Fig. 4, B and E). Interestingly, Ena has been shown to directly interact with Robo to mediate repulsive signaling during axon guidance (Bashaw et al., 2000). Our data suggest that Ena may provide a critical link between Robo and E-Cad during lumen formation, and provide an avenue for further study.

For D. melanogaster heart lumen formation, CBs must be attached at specific sites but also have apical surfaces that are not adhered (Fig. 5). Our studies show that Slit/Robo signaling is required for lumen formation between CBs by inhibiting E-Cad–mediated cell adhesion. Slit and Robo localize to the apical surface of the CBs facing the lumen. In robo LOF or shg/E-Cad GOF embryos, contralateral CBs are tightly bound, resulting in a failure of lumen formation (Fig. 5). However, when Slit is ectopically expressed on basal and lateral surfaces, we observe an inappropriate loss of cell adhesion in these regions, resulting in the formation of ectopic lumens (Fig. 2, F and H). These phenotypes are accompanied by mislocalization of the cell junction markers E-Cad, Ena (Fig. 4), and Baz (Fig. S1) to these sites. How might Slit and Robo regulate cell adhesion and lumen formation? Much of our understanding of Slit and Robo signaling has emerged from studies of how Robo's activation by Slit regulates downstream cytoskeletal changes during axon guidance (Dickson and Gilestro, 2006). Two recent in vitro studies support our findings that, in addition to their known role in cell migration, Robo proteins also function to either positively or negatively regulate cell adhesion (Rhee et al., 2002; Kaur et al., 2006). Here, we provide in vivo genetic evidence that Slit and Robo negatively regulate cell adhesion during lumen formation. The next step will be to identify the downstream mechanisms by which Slit and Robo signaling negatively regulate E-Cad–mediated cell adhesion.

Fly crosses and experiments were performed at 25°C. The w¹¹¹⁸ strain was used as the wild type. The following mutations have been described previously: null alleles for slit (sli¹,sli²) and robo (robo^Z570, robo⁵; Kidd et al., 1999). UAS-slit and UAS-robo were also described previously (Kramer et al., 2001). The UAS-GAL4 system was used to drive expression of the slit transgene with Mef-GAL4. slit²/Cyo,pActGFP;UAS-slit was crossed with sli²/Cyo,pActGFP; Mef-GAL4 to get slit²/slit²; Mef-Gal4/UAS-slit embryos. sli¹,robo⁵/Cyo,WgBgal; Mef-Gal4 was crossed with sli¹,robo⁵/Cyo,WgBgal; UAS-slit to get sli¹,robo⁵/ sli¹,robo⁵; Mef-Gal4/UAS-slit embryos. The ena allele used was ena²³ (BL stock No. 8571) and the shg/E-Cad allele used in this study was shg^K03401 (BL stock No. 10377); they were obtained from the Bloomington Stock Center. The wild-type UAS-shg transgene (Oda and Tsukita, 1999) was obtained from the Drosophila Genetic Resource Center (Kyoto Institute of Technology).

Embryos were fixed and stained as described previously (Santiago-Martínez et al., 2006). The following antibodies were used: mouse anti-Slit (1:10), mouse anti-Robo (1:10) and rabbit anti-Robo2 (1:1,000; 16), rabbit anti-Mef2 (1:2,000; a gift from B. Paterson, National Cancer Institute, National Institutes of Health, Bethesda, MD), and rabbit anti-Baz (1:1,000; gift from A. Wodarz; University of Göttingen, Göttingen, Germany). Mouse anti-Dlg (1:10), mouse anti-Enabled (1:10), mouse anti–α-spectrin (1:10), rat anti–DE-cadherin (1:10; obtained from the Developmental Studies Hybridoma Bank, developed under the auspices of the National Institute of Child Health and Human Development, and maintained by the University of Iowa), FITC anti–mouse (1:500), Alexa 488 anti–rabbit (1:500), Cy3 anti–mouse (1:500) and –rabbit (1:500), Biotin anti–rat (1:500) and Streptavidin 488 (Invitrogen), and anti–β-Gal (1:10,000; MP Biomedicals). Embryos were mounted in Vectashield mounting media for fluorescence (Vector Laboratories), and confocal z sections were collected at ambient temperature on a microscope (IX81; Olympus) with a CARV Nipkow disc confocal unit (BD Biosciences) equipped with a VApo/340 40× 1.15 NA water immersion objective (Olympus) and a SensiCam QE camera (the Cooke Corporation). Image processing and analysis was done with IPLab image analysis software (BD Biosciences) and Photoshop CS2 (Adobe).

Mutant embryos were selected based on the absence of GFP-marked balancer chromosome. Embryos were fixed and embedded as described previously (Santiago-Martínez et al., 2006). 90-nm sections were cut using a microtome (Ultracut E; Leica) from the posterior to the anterior end of the embryo and picked up on a copper or a carbon support film on specimen grids. Sections were stained with uranyl acetate and lead citrate. XSs were examined and using an electron microscope (CM-12; Philips) operating at 80 kV. Digital images were collected using a digital camera (XR611/BZ; Advanced Microscopy Techniques) and Image Capture Engine V600 software (Advanced Microscope Techniques) and processed using Photoshop CS2 (Adobe).

Embryos were fixed and stained at ambient temperature using anti-Slit, anti-Robo, anti-Robo2, anti-Mef2, and anti–E-Cad as described previously (Santiago-Martínez et al., 2006). Biotin and conjugated streptavidin-HRP (Invitrogen) were used as secondary antibodies, and the signal was enhanced using a Metal Enhanced DAB Substrate kit (Pierce). Embryos were staged and embedded in epon-Spurr resin. 4-μm sections were cut using the Ultracut E microtome from the posterior to the anterior end of the embryo and subsequently imaged on a Axio Imager.A1 equipped with a Plan-APO 40× 0.95 NA objective and an AxioCam MRc5 digital camera (all from Carl Zeiss, Inc.). Image processing and analysis was done with AxioVision 4.6 (Carl Zeiss, Inc.) and Photoshop CS2 (Adobe) software.

Fig. S1 shows the expression of the cell polarity markers Dlg and α-spectrin and the cell junctional marker Baz in wild-type, slit GOF, robo GOF, shg GOF, and robo LOF backgrounds.

We thank Kevin Ang for assistance with embryo stainings, Jeff Roberto for assistance with line art, and Colleen Guerin, Raymond Habas, Joe Kramer, Haichang Li, Ramani Ranchandran, and William Wadsworth for helpful discussions and critical reading of the manuscript.

This study was supported by a Scientist Development Grant from the American Heart Association (0435395T) and a National Science Foundation grant (award ID 0744165) to S.G. Kramer, and a National Institutes of Health Predoctoral Fellowship (F31/GM75393) to E. Santiago-Martínez.