Genetic control of cell morphogenesis during Drosophila melanogaster cardiac tube formation

Biological tubes are fundamental structural and functional units of tissue architecture. The cardiovascular system and most internal organs, such as kidney, liver, heart, and lungs, are composed of simple tubes or of a network of tubes that transport fluids or gases. During tubulogenesis, cell polarity, shape, and size must be precisely controlled and cellular adherens junctions need to be continuously remodeled. Unraveling the mechanisms underlying such membrane dynamics is crucial to understand various pathologies including metastasis and tumor progression.

Our knowledge of tubulogenesis has increased considerably during the last decade. Several studies proposed that the main steps may be shared by diverse pathways of tubulogenesis (Lubarsky and Krasnow, 2003; Kerman et al., 2006).

However, it is not clear whether these general features are relevant to the formation of all tubes. In particular, morphogenesis of the dorsal aorta, the posterior cardinal vein, and the primitive vertebrate heart tube appear to involve different mechanisms. In fish, major axial vessels are formed by the migration of angioblasts originating from the lateral plate mesoderm, which coalesce in the midline (Weinstein, 1999; Jin et al., 2005). Recently, Jin et al. (2005) reported a cellular and molecular analysis of vascular tube and lumen formation in zebrafish, showing the coalescence of angioblasts at the midline to form aggregates or solid cords. Within these aggregates, endothelial cell–cell contacts are established, and subsequently a tube with a lumen becomes apparent. The membrane walls of the lumen display some characteristics of basal membranes, as they express, for example, integrins and extracellular matrix components (Davis and Senger, 2005). However, the mechanisms of cell migration, polarity, and shape remodeling underlying lumen formation remain largely unknown.

Drosophila melanogaster cardiac tube morphogenesis shares remarkable similarities with the formation of primary axial vessels in vertebrates. Indeed, it has been recently proposed (Hartenstein and Mandal, 2006) that the D. melanogaster cardiovascular system is phylogenetically related to the vertebrate vascular system. The cardiovascular system in flies is formed by a simple linear tube, which constitutes the unique vessel of an open circulatory system (Rizki, 1978; Rugendorff et al., 1994). The D. melanogaster cardiac tube is made of two rows of 52 myoendothelial cells (cardioblasts [CBs]) enclosing a lumen. The cardiac myoendothelium originates from migrating mesodermal cells, which undergo a mesenchymal–epithelial transition to form two bilateral rows of cells attached to each other by adherens junctions (Rugendorff et al., 1994; Tepass and Hartenstein, 1994; Fremion et al., 1999). During dorsal closure, the two rows of CBs, together with adjacent pericardial cells, migrate as a sheet of cells in association and coordination with the overlying ectoderm (Chartier et al., 2002). They eventually meet each other at the dorsal midline, make new adherens junctions, and start forming a lumen that enlarges during the late stages of embryogenesis (Rugendorff et al., 1994; Haag et al., 1999).

The genetic control of the D. melanogaster cardiac tube morphogenesis has been extensively studied (Zaffran and Frasch, 2002; Monier et al., 2007; Tao and Schulz, 2007). These studies have provided a better understanding on how affecting gene function can perturb general organ morphogenesis, cell number, and cell identity. However, only few studies have characterized, at a cellular level, the consequences of gene inactivation on cardiac cell morphogenesis.

In this study, formation of the D. melanogaster cardiac tube lumen was revisited by providing a detailed analysis of cardiac cell morphogenesis. Using in vivo 3D and time-lapse imaging and analyzing the distribution of various molecular markers led to the definition of distinct membrane domains to which specific functions in lumen formation can be attributed. To evaluate the functional importance of cell shape changes, membrane specification, and remodeling, we searched for mutations affecting these aspects of cardiac cell morphogenesis. We have identified slit, roundabout (robo), dystroglycan (dg), and held out wings (how) as key components of a genetic pathway that controls cardiac cell morphogenesis and is required for correct lumen formation. Moreover, our data provide evidence for a mechanism of tube formation substantially distinct from the so-far described mechanisms of epithelial tubulogenesis.

We have restricted our analysis to the CBs, as the cardiac tube lumen is exclusively formed by the membrane walls of these cells. To investigate CB morphogenesis during formation of the cardiac tube, the expression of Dmoesin-GFP (Fig. 1), which binds to cortical actin (Polesello et al., 2002), or directly to actin-GFP (see Fig. 2, A–L), was targeted to CBs using the 24B-Gal4 driver and the binary Gal4–upstream activating sequence (UAS) system. Time-lapse confocal imaging was used to record GFP and to follow actin dynamics during the coalescence of the two bilateral rows of CBs (Videos 1 and 2). To focus this analysis on the membrane domains that are directly implicated in lumen formation, transverse Z sections were generated from the reconstructed 3D views at different developmental time points. This allowed us to efficiently follow CB shape changes at different steps of cardiac tube formation (Fig. 1).

At the onset of dorsal closure, CBs adopt a pearlike shape consecutive to constriction of their domain facing the dorsal midline (Fig. 1, F and F′, compared with Fig. 1, E and E′). Actin-rich cytoplasmic extensions grow from this membrane domain, which constitutes the leading edge of the dorsally migrating CBs (Fig. 1 B, arrowheads). CBs from each of the two rows come progressively into contact at their leading edges and join at the dorsal midline (Fig. 1, B, G, and G′). Subsequently, CBs adopt a crescentlike shape (Fig. 1, H and H′), which allows their bases to join ventrally and thus to close the tube (Fig. 1, I and I′), creating an internal lumen. During this step, which takes ∼60 min (see Fig. 2 and Videos 1 and 2), the dorsal part of CBs detaches from the dorsal ectoderm. Throughout this process, CBs increase their size and keep growing after tube closure; as a result, the lumen enlarges progressively (Fig. 1, D and I). Thus, our in vivo analysis provides for the first time a step by step description of the dynamics of cardiac tube formation. It confirms previous observations (Rugendorff et al., 1994; Haag et al., 1999) made on fixed preparations and provides, in addition, a highly tractable new method for analyzing in vivo CB cell behavior in wild-type and mutant backgrounds.

To investigate the genetic control of cell behavior during cardiac tubulogenesis, we tested several candidate genes known to be expressed in migrating CBs and in the cardiac tube. We first analyzed the function of slit, encoding an extracellular protein that binds to the membrane receptors Robo and acts either as attractant or repellent (Dickson and Gilestro, 2006). Slit is expressed in CBs (Rothberg et al., 1988) and, recently, three studies have reported that Slit–Robo pathway function is required for normal assembly of the cardiac tube, CB migration, and lumen formation (Qian et al., 2005; MacMullin and Jacobs, 2006; Santiago-Martinez et al., 2006). Loss of slit function leads to composite phenotypes, such as misaligned or twisted tube portions, gaps in one row of CBs, and, at some places, when the two opposite CB rows have coalesced, lack of lumen or formation of an abnormal lumen. Thus, Slit–Robo signaling plays a key role during the late steps of cardiac development, but none of the previous studies have addressed precisely how cell morphogenesis is affected when the Slit–Robo pathway is inactivated.

To better understand this issue, we used CB-targeted live imaging of actin-GFP. Changes in the CB cell shapes were recorded in wild type, slit², or robo/robo2 double mutant embryos (Fig. 2 and Videos 2–4). We found, in contrast to the wild type, that the mutant CBs (in both slit and robo/robo2 mutant backgrounds) do not modulate their shape during migration (Fig. 2, A′ and A″, compare with Fig. 2 A). Mutant CBs keep their initial round shape (Fig. 2, A′–C′ and A″–C″), do not constrict to form a dorsal leading edge with filopodia, and do not detach from the overlying dorsal ectoderm. As a consequence, CBs come into contact with the opposite row by their entire nonconstricted dorsal membrane surfaces, forming a large dorsal cell–cell contact extending ventrally (Fig. 2, F′–L′ and F″–L″ [bracket], compared with Fig. 2, F–L). Thus, when the two rows of mutant CBs succeed in joining at the dorsal midline, the presumptive lumen domain is either absent or displaced ventrally, preventing its normal juxtaposition with the contralateral luminal surface and blocking lumen formation (Fig. 2, L′ and L″, compared with Fig. 2 L).

The in vivo observations have been confirmed by an electron microscopy analysis of wild-type and slit² mutant cardiac tubes showing the lack of cell shape changes in slit² mutant CBs, which, in contrast to wild-type CBs, display a round shape (Fig. 3, compare B with A). The progressive shrinking of the CB cytoplasm at the site of initial cell–cell contact, which contributes to the lumen formation in wild type, is strongly affected (Fig. 3 A and Fig. S1, yellow arrows, compared with Fig. 3 B). Cell–cell contacts between the two opposite CBs spread over a much larger area than in the wild-type situation. This extended area of cell–cell contact in slit² mutant embryos is most probably caused by the lack of cell shape remodeling in the absence of slit function. Electron-dense dots corresponding to adherens junctions between two CBs are detected in slit² mutants (Fig. 3, D′ and E′) in the same position as in the wild type (Fig. S1 B), indicating that slit mutants are able to differentiate cell junctions. This general phenotype is recovered along the entire anterior–posterior axis, at least in the region where the two rows of cells have coalesced (Fig. 3, C–F). However, slight variation in this general phenotype is observed consisting essentially of interruptions in the firm cell–cell contacts between CBs that sometimes form vacuolelike structures inside the CB cytoplasm (Fig. 3, B–F, blue arrows). Similar structures are also observed at early stages in wild-type embryos (Fig. S1 A, blue arrows).

To better understand the cellular mechanisms regulating cardiac lumen formation, we analyzed the distribution of cell polarity markers in developing wild-type CBs. CB precursors originate from nonpolarized mesenchymal cells of the dorsal mesoderm. After germ band retraction, they form two bilateral rows of polarized cells sharing some polarity features with epithelial cells (Fremion et al., 1999). CBs possess a basal domain adjacent to the overlying ectoderm, expressing classical markers of basement membranes; extracellular matrix proteins, including laminin A (Yarnitzky and Volk, 1995), perlecan (Terribly reduced optic lobes [Trol] in D. melanogaster; Fig. 4 B; Voigt et al., 2002), pericardin (a type IV collagenlike protein; Chartier et al., 2002), Slit (Fig. 4 C; Rothberg et al., 1988), and their receptors, Dg (Fig. 4 A); integrins; and Robo (Stark et al., 1997; Qian et al., 2005). CBs also contain basal-lateral domains expressing Discs large (Dlg; Fig. S2, A and C), α-spectrin, and adherens junction markers such as β-catenin (Armadillo [Arm] in D. melanogaster), and DE-Cadherin (DE-Cad; or Shotgun [Shg] in D. melanogaster; Fremion et al., 1999; Haag et al., 1999), from which adherens junctions among cells of the same CB rows are formed (Fig. 4 D′, arrowheads).

In the following, we refer to luminal domains (L domains) and adherent domains (J domains) to designate the membrane domains involved in the formation of the lumen walls and adherens junctions responsible for the dorsal and ventral sealing of the tube, respectively.

CBs were assumed to display “apical-basal” polarity, based on the absence of α-spectrin expression in the domain facing the dorsal midline (Fremion et al., 1999). However, this membrane domain never shows any expression of the known classical apical markers of epithelial cells, including Crumbs (Tepass et al., 1990; Qian et al., 2005), β-heavy-spectrin, Bazooka, or atypical PKC (unpublished data), indicating that CBs cannot be considered as classical apical-basal polarized epithelial cells. The apical-like spectrin-free domain of CBs corresponds to the F-actin–rich leading edge or J domain that extends filopodia and contacts dorsally with the CBs of the opposite row.

In contrast, the future lumen membrane domain expresses classical basal components such as Slit (Fig. 4, C and C′), Dg (Fig. 4, A and A′; Qian et al., 2005), integrins (not depicted), and Trol (Fig. 4, B and B′). This membrane domain, the L domain, constitutes the wall of the lumen and actively contributes to the deposition of an extracellular matrix within the growing lumen (Fig. S1, asterisks; Rugendorff et al., 1994; Chartier et al., 2002).

We complemented this analysis of CB polarity by examining the localization of Arm as a marker for potential cell–cell junctions. Arm starts to localize at the future J domains (dorsally, at the leading edge, and ventrally; Fig. 4, E–J and E′–J′, arrowheads) significantly before the joining of the two CB rows at the dorsal midline. In the course of CB row migration, Arm becomes excluded from the future L domains, where Dg (Fig. 4, E″–J″), Trol, and Slit are localized (not depicted). This specific localization of Arm is maintained during CB migration and after fusion of the bilateral cardiac primordia (Fig. 4, F–J and F′–J′) and marks the sites of the adherens junctions, seen by electronic microscopy (Fig. S1 B″), which are responsible for tube sealing. In a similar manner, another adherens junction marker, DE-Cad, is also observed at the J domains of CBs (not depicted), as well as classical basolateral markers, Dlg (Fig. S2, B and D) and Lethal giant larvae (Lgl; not depicted).

Altogether these observations define distinct membrane domains involved in lumen formation: the L domain expressing Dg, Slit, and Trol in which cell–cell adhesion is prevented, surrounded by two J domains where Arm is localized and responsible for sealing the tube dorsally and ventrally (Fig. 4 K). D. melanogaster CBs appear to show specific cell polarity with prepatterned L domains with basal membrane properties. In contrast to so-far described models of tubulogenesis, the cardiac lumen is formed from a basallike and not from an apical membrane domain.

As CB cell shape changes do not occur in slit² mutant embryos (Figs. 2 and 3), we decided to investigate the involvement of the Slit–Robo pathway in specifying CB membrane domains by analyzing Arm and Dg expressions in slit² mutant embryos (Fig. 5). As in wild-type embryos, Arm is observed in the J domains (Fig. 5, A′–F′), as well as Dlg (Fig. S2, B and D) and Lgl (not depicted). However, from the onset of CB migration onward, the Arm-positive domain is clearly expanded and maintained, or even increased when CBs come into contact (Fig. 5, E′ and F′). Correlatively, Arm is first found to partially colocalize with Dg (Fig. 5, A–C). During CB migration, Dg is progressively excluded from the Arm-positive domain to become completely absent from this domain (Fig. 5, D–F) in the cases when the two contralateral CBs make contact (mean of 60% of the CBs constituting the mutant cardiac tube can reach this stage), Dg is displaced toward the ventral boundary of the Arm J domain, underlining, in some cases, an ectopic, small lumenlike structure (Fig. 5 F). Interestingly, Dg is also sometimes recovered at the periphery of these lumenlike structures located inside the cytoplasm of the CB cells (Fig. S3 A, arrows). Together, these observations suggest that Slit–Robo function is required for setting the nonadherent L domain, which is critical for the formation of a correct lumen (Fig. 5 G).

The aforementioned results indicate that the establishment of specialized CB membrane domains is a crucial step for cardiac lumen formation. To gain further insights into this process, we searched for other genes whose activity could be required for setting CB membrane domains. Previous studies demonstrated that dg and how are involved in late aspects of cardiac development or function (Zaffran et al., 1997; Qian et al., 2005; unpublished data), and thus we decided to test their role in CB morphogenesis.

How encodes an RNA-binding protein of the Star family implicated in mRNA translation control and gene splicing (Nabel-Rosen et al., 2002; Volohonsky et al., 2007). The formation of the cardiac lumen has been investigated in the hypomorphic how18 mutants (Zaffran et al., 1997), in which no How expression in the cardiac tube can be detected (Fig. S4). Electron microscopy analysis in how18 mutant CBs (Fig. 6) shows the lack of cell shape changes and the extension of cell–cell contacts (Fig. 6, B and C, yellow arrows). These mutant CBs form nonswollen lumenlike gaps filled with extracellular matrix material (Fig. 6 C, asterisk) that intrude into the CB cytoplasm and are highly convoluted (Fig. 6 compared with Fig. S1).

The misspecification of the lumen membrane domain in how mutants is supported by the reduced and displaced ventral staining of the L domain marker Dg (Fig. 7 F, compare with Fig. 4 J). As a consequence, the Arm-positive domain is now extended along almost the entire region of cell–cell contacts between CBs of the opposite rows (Fig. 7, A–F). A partial colocalization of Arm and Dg is observed, principally during the migration step (Fig. 7, A′ and B′, compared with Fig. 7, A″ and B″). Importantly, as in slit² mutants, the CB shape and membrane domains remodeling observed in the wild type does not occur (Fig. 7, E and F). CBs do not constrict their leading edge at the onset of migration and remain rounded, with large Arm-positive J domains facing the dorsal midline (Fig. 7, C and D).

In a similar way we tested lumen formation in dg mutants. Dg is an extracellular matrix receptor and is a part of the widely expressed and evolutionary conserved Dystrophin complex (Winder, 2001). In D. melanogaster, Dg has been shown to be required for epithelial organization, polarity, and muscle viability (Deng et al., 2003; Schneider et al., 2006; Shcherbata et al., 2007).

As in dg deficiency Df(2R)JP6Dg (Qian et al., 2005), earlier cardiac defects make the analysis of lumen formation difficult, so we opted to use partial loss-of-function dg alleles, dg62, dg323, and dg248 (Deng et al., 2003; Schneider et al., 2006; Shcherbata et al., 2007). They give rise to essentially the same type of phenotypes, so we present only data for the dg62 allele. We observed that most of the homozygous dg mutants (80%, first group) form a tube with a general wild-type appearance; however, the L domains remain stacked, with no or a very small lumen formed (Fig. 7, G–L). A small proportion (5%, second group) of dg mutant embryos shows a strong phenotype in cardiac tube morphogenesis: migration defects, gaps in the cardiac rows, and twisted tubes similar to phenotypes recovered in slit² mutant and in dg Df(2R)JP6Dg deficiency (Fig. S2, E and F;Qian et al., 2005). The last 15% of dg mutants display a wild-type phenotype. Precise localization of cell polarity markers has therefore been investigated only in the first group of mutant embryos. All along the anterior–posterior axis, the L domain, revealed by Trol, is considerably reduced compared with wild type, at the expense of the J domain, probed by Arm expression (Fig. 7, G–L). Moreover, as in how18 and slit² mutants, the dynamics of cell shape remodeling observed in wild-type embryos does not occur (Fig. 7, G–L); the CBs remain rounded with no constriction of the dorsal J domain. A partial colocalization of Arm and Trol is also observed during the CBs migration step (Fig. 7, G′–L′ and G″–L″).

Because slit, dg, and how mutants share common phenotypes, we analyzed Slit localization in how18 and dg62 mutant embryos (Fig. 8). At the end of CB migration (Fig. 8, I, J, and N, compared with Fig. 8 D) and when the tube is closed (Fig. 8, K–M and O–R, compared with Fig. 8, E–H), Slit expression is no longer recovered at the L domain or into the cytoplasm just underneath the L domain, but is found to predominantly localize randomly in the cytoplasm in both mutants. Occasionally, Slit expression is still present within the small and generally ectopic lumen (Fig. S3 B). Slit is also found in association with Dg or Trol in cytoplasmic vacuolelike structures, also observed in slit² mutants (Fig. S3 A).

To investigate genetic epistasis between slit, how, and dg, we first analyzed the effects of Slit overexpression in the cardiac tubes of how18 and dg62 homozygous mutant embryos. Remarkably, both Slit localization at the L domain and lumen formation (in 60% of how18 mutants) are rescued by Slit overexpression in the CBs of how18 homozygous mutants (Fig. 9). In contrast, Slit overexpression is not able to rescue lumen formation in dg62 cardiac tubes (not depicted). These results suggest that how acts upstream of slit and that dg functions parallel to slit to control lumen formation.

In addition, we observed that slit genetically interacts with how and dg for the formation of the cardiac lumen. A mutant phenotype in lumen formation is observed in 80% of slit²/how18 transheterozygotes (Fig. 5 A) and in 73% of slit²/dg62 transheterozygotes (Fig. 5 B), which is never observed in single heterozygotes. Finally, overexpression of Dg in CBs leads to a similar phenotype as dg loss of function, showing a strong ectopic Slit localization in the cytoplasm and a very narrow lumen (Fig. S5, C and D).

Altogether, these observations support that slit, how, and dg participate in the same pathway to control cardiac lumen formation.

The analysis provided here, using both in vivo cell imaging and subcellular localization of molecular markers, establishes the cellular basis of lumen formation of the D. melanogaster cardiac tube. As previously reported, the lumen of the tube is formed from the migration of two bilateral rows of polarized CBs, which join at the dorsal midline (Rugendorff et al., 1994; Fremion et al., 1999; Haag et al., 1999). One main result of our study is the characterization of two types of cell membrane domains directly involved in lumen formation, the J domains and the L domain. Adherens junctions that are responsible for sealing the tube originate from the J domain, whereas the membrane walls of the lumen originate from the L domain.

Remarkably, the L domain displays characteristics of basal membranes, revealed by expression of molecular markers normally associated with a basal membrane. Furthermore, specification of the L and J domains takes place very early in the tubulogenesis process, significantly before coalescence of the bilateral rows of CBs at the dorsal midline. Finally, during CB migration, membrane domains undergo remodeling, concomitant with profound cell shape changes. These two cellular processes appear to be closely connected and are probably regulated by the cellular environment of the CBs composed by the overlying dorsal ectoderm and the amnioserosa cells. These interactions will be investigated in a future work.

The mechanism of D. melanogaster cardiac lumen formation reported here (Fig. 10 A) is thus notably different from the previously described mechanisms of epithelial tubulogenesis (Myat, 2005). In epithelial tubulogenesis, after receiving a polarization signal that sets apicobasal polarity, the cells or group of cells establish a basal surface and generate vesicles carrying apical membrane proteins. The vesicles are targeted to the prospective apical region, where they fuse with the existing membrane or with each other to form a lumen (Kamei et al., 2006; Kerman et al., 2006). Finally, continued vesicle fusion and apical secretion expand the lumen.

In contrast, constriction of the leading edge domain during CB migration, precise control of cell shape changes, and delimitation of specific membrane domains appear to be the driving forces of D. melanogaster cardiac lumen formation. Cells forming the dorsal vessel have the features of migrating cells. In contrast to epithelial tubulogenesis, which involves apical membrane domains, the apex of polarized CBs constricts, forms adherens junctions, and consequently does not constitute the L domain. Instead, the luminal membrane domain possesses basal membrane characteristics, as is also the case in endothelial cells (Davis and Senger, 2005). Moreover, the size of the cardiac lumen is determined by the isotropic growth of CBs, and not, as in other models, by anisotropic extension of the L domain involving apical membrane vesicles. Finally, the genetic control of the process involves gene products of slit, robo, how, and dg, which are not known regulators of lumen formation in epithelial tubes.

Our study leads to the identification of a genetic pathway, including slit, robo, how, and dg, controlling membrane domain specification and dynamics during cardiac lumen formation. Within this pathway, Slit appears to play a central role and a previously unrecognized function in cell morphogenesis.

Several studies have shown that Slit–Robo function is essential for cardiac tube formation by controlling the proper migration, cohesion, and alignment of the two rows of CBs (Qian et al., 2005; MacMullin and Jacobs, 2006; Santiago-Martinez et al., 2006). The results reported here show that Slit is also involved in the correct specification of the L domain and its distinct features with respect to the adjacent J domains. Activation of Slit–Robo signaling determines the respective size of these two types of domains.

Our data suggest that activation of this pathway inhibits the formation of adherens junctions. This possibility is supported by recent findings in chick retina cells, where activation of the Slit–Robo pathway leads to the inactivation of β-catenin (Arm in D. melanogaster), resulting in the dissociation of N-cadherin from the junctional complex and preventing the formation of adherens junctions (Rhee et al., 2002, 2007). Consistent with these observations, DE-Cad (Shg) is expressed in the J domains of CBs (unpublished data) and is required for cardiac tube morphogenesis (Haag et al., 1999). Moreover, slit and shg show genetic interaction in cardiac tube morphogenesis (Qian et al., 2005). In the absence of slit function, the size of the L domain is strongly reduced (Fig. 5, A–C), suggesting that Slit–Robo signaling prevents the formation of Arm/DE-Cad–mediated adherens junctions in the L domain.

How encodes an RNA-binding protein involved in mRNA metabolism, and given its exclusive nuclear localization at this stage of development (Fig. S4), How may regulate slit splicing. In the absence of the How protein, the gene splicing could be affected, producing a Slit protein unable to correctly localize at the L domain. This hypothesis is consistent with the fact that expression of wild-type Slit in CBs can suppress the effect of how18 mutation on Slit localization and lumen formation. How has also recently been shown to regulate the splicing of neuronal membrane proteins such as neurexin (Edenfeld et al., 2006). Moreover, How is expressed in the midline glia with Slit and Dg (Fig. S4 C), suggesting that interaction among these three genes is part of a general mechanism by which junctions and lumen formation are controlled.

A model for the genetic control of lumen formation in the cardiac tube, based on our observations, is proposed in Fig. 10 B. According to this model, How could directly regulate Slit by controlling its splicing and targeting the luminal compartment. Consequently, Slit binds to Robo activating the signaling pathway, which in turn inhibits Arm/DE-Cad–mediated adherens junction formation in the luminal compartment, leading then to the specification of distinct J and L domains. Parallel to this, activation of Slit–Robo signaling modulates the actin cytoskeleton and triggers CB cell shape remodeling required for lumen formation and growth. As How is able to act on many targets, it could also directly control the actin cytoskeleton by targeting an actin-binding molecule. Concerning Dg, we have shown that dg and slit genetically interact (Fig. S5 B); however, overexpression of Slit does not rescue the lumen phenotype observed in dg mutants, contrasting with how mutations. Thus, we propose that Dg could regulate Slit localization at the L domain by its function in the specification and differentiation of the L domain, and therefore acts parallel to slit for lumen formation, behaving, for example, as a coreceptor of Robo. In addition, Dg could control actin cytoskeleton dynamics via its interaction with Dystrophin.

Our data clearly show that cardiac tube formation in D. melanogaster differs substantially from all other described mechanisms of tubulogenesis. Is this mechanism of tubulogenesis unique or is it shared with other organs and/or other organisms? Primary vasculogenesis in vertebrates leads to the formation of large median vessels, the dorsal aorta and the cardinal vein (Pardanaud et al., 1987; Torres-Vazquez et al., 2003; Jin et al., 2005). These vessels arise from migrating mesenchymal cells of the lateral mesoderm, termed angioblasts, that are organized in bilateral groups of cells. Angioblasts migrate toward the midline as a cohort of cells, coalesce, and form a lumen. At this stage, as in flies, cells around the lumen show a crescentlike shape (Risau, 1995; Jin et al., 2005) and an extracellular matrix is deposited at the internal face of luminal membranes (Davis and Senger, 2005). Similar cellular events are also observed during the formation of the primitive cardiac tube in vertebrates (Harvey, 2002), suggesting that a common mechanism of tubulogenesis might exist for all tubes that arise from the coalescence of migrating bilateral mesenchymal cells.

As proposed by Hartenstein and Mandal (2006), the D. melanogaster cardiac tube, or dorsal vessel, shares many similarities with the cardiovascular system of vertebrates. A significant fraction of genes expressed in the D. melanogaster cardiac tube are also annotated to be expressed in vertebrate blood vessels, suggesting that vasculogenesis and dorsal vessel morphogenesis might share common genetic regulators.

Finally, components of the genetic pathway controlling cardiac lumen formation that we describe here have potentially similar functions in vertebrates. It has been shown previously that numerous proteins involved in axon guidance are expressed in vertebrate blood vessels (for review see Weinstein, 2005). In particular, the Slit–Robo signaling pathway has been involved in promoting tumor vascularization, hSlit2 being expressed in tumor cells and hRobo1 in endothelial cells (Wang et al., 2003). Moreover, mSlit3 has been implicated in mammalian cardiogenesis (Liu et al., 2003), and Quaking, the mouse homologue of How, is required for vasculogenesis and expressed in the developing heart (Noveroske et al., 2002).

In conclusion, our analysis of CB morphogenesis during development of the D. melanogaster cardiovascular system provides evidence for a new model of biological tube formation. We propose that this mechanism might also be used for the formation of the large median vessels and primitive heart tube in vertebrates.

For live experiments, we used a UAS Dmoesin-GFP; 24B-Gal4 line and UAS actin-GFP; 24B-Gal4 line to follow cardiac cell behaviors. The UAS Dmoesin-GFP was obtained from F. Payre (Centre de Biologie du Développement, Toulouse, France; Polesello et al., 2002) and the UAS-actin GFP was obtained from the Kyoto Stock Center. The flies slit²; UAS-actin GFP and robo^GA285, robo2⁴; UAS actin-GFP were crossed with the slit² or robo^GA285, robo2⁴ mutant flies carrying 24B-GAL4 driver. Wild-type flies are Oregon lines. Trol-GFP line has been obtained from the L. Cooley laboratory (L. Cooley, Yale University Medical School, New Haven, CT; Kelso et al., 2004). As mutant flies, we used slit²/CyOWg-LacZ, (Bloomington Stock Center), double mutant robo^GA285, robo2⁴/CyOWg-LacZ and robo^GA285, robo2⁸/CyOWg-LacZ (provided by B. Dickson, Institute für Mechanik, Vienna, Austria; Rajagopalan et al., 2000), how18/TM3actGFP (Zaffran et al., 1997), dg62/CyOWg-lac-Z, dg248/CyOWg-lac-Z, and dg323/CyOGFP (provided by S. Baumgartner, Lund University, Lund, Sweden; Deng et al., 2003). Mutants were selected by the absence of βGal/GFP staining. To overexpress Dg we used 24B-Gal4 and UAS-Dg full length obtained from M. Schneider (Lund University; Schneider et al., 2006). To overexpress Slit, we used 24B-Gal4 and Hand-Gal4 (DHandGal4 4.2) from A. Paululat (Osnabrück University, Osnabrück, Germany) as drivers and UAS Slit from B. Dickson.

For the primary antibodies, we used rabbit anti-Dg, 1:300 (provided by W.M. Deng, Florida State University, Tallahassee, Florida; Deng et al., 2003); rabbit anti-Trol, 1:1,000 (Schneider et al., 2006); rabbit anti-β3-tubulin, 1:1,000 (Developmental Studies Hybridoma Bank [DSHB]); mouse anti-Arm, 1:100 (DSHB); mouse anti-Dlg, 1:100 (DSHB); mouse anti-Slit, 1:50 (DSHB; amplification with Renaissance TSA Biotin system [PerkinElmer]); rabbit anti-Lgl, 1:500 (provided by J. Knoblich, Institute für Mechanik; Betschinger et al., 2003); rat anti-How, 1:100 (provided by T. Volk, Weissman Institute, Rehovot, Israel; Nabel-Rosen et al., 1999); mouse anti-Prc, 1/3 (Chartier et al., 2002); mouse anti-βGal, 1:500 (Promega); and mouse and rabbit anti-GFP, 1/500 (Invitrogen).

For secondary antibodies, we used Biotin SP conjugated anti–mouse (Jackson ImmunoResearch Laboratories), 1:500, and Cy5 anti–rat and anti–rabbit, 1:100 (Jackson ImmunoResearch Laboratories); Alexa 546 anti–mouse and anti–rabbit, 1:500; and Alexa 488 anti–mouse and anti–rabbit, 1:500 (Invitrogen).

Embryos were collected from 12 to 16 h, dechorionated in bleach for 5 min, fixed in PBS/heptane (1:1), 4% formaldehyde, for 20 min, and devitellinized in heptane/methanol (1:1), washed in methanol and ethanol, and rehydrated progressively in PBS. Embryos were blocked in PBS, 0.1% Tween, and 10% BSA and incubated overnight at 4°C with the primary antibody. Embryos were washed with PBS, 0.1% Tween, and 0.1% BSA and incubated for 1–2 h with the secondary antibody, washed again, and mounted in Fluoromount medium (SouthernBiotech).

After dechorionation, embryos were fixed by heat fixation method: they were plunged for 5–7 s in a boiling solution with 68 mM NaCl and 0.04% Triton X-100. They were put for a few minutes on ice and prepared for devitellinization as described in the previous section.

Embryos expressing UAS actin-GFP (or UAS Dmoesin-GFP) under the control of 24B-Gal4 driver in wild-type, slit², and robo^GA285, robo2⁴ mutant backgrounds were dechorionated in bleach (2 min), rinsed in water, placed on a coverslip in the appropriate orientation, and then mounted in 3 M of oil (Voltalef) for live imaging using a confocal microscope. Images compiled from 0.5-μm optical sections were collected every 6 min. All the preparations were visualized on confocal microscopes (LSM 510 or 510 Meta; Carl Zeiss, Inc.) at room temperature using a plan Apochromat 20× 0.8 NA or a C Apochromat (water) 40× 1.2 NA lens (Carl Zeiss, Inc.). 4D reconstructions and image analysis were performed using LSM browser, Zen 2007 light edition (Carl Zeiss, Inc.), Volocity 4.0 (Improvision), Imaris 4.0 (Bitplane), and Photoshop CS2 (Adobe) softwares.

Embryos were dechorionated in bleach for 3 min. They were transferred into a microfuge tube on ice and fixed for ∼20 h with 2% glutaraldehyde and 2% paraformaldehyde in 0.1 M cacodylate buffer, pH 7.4. They were washed several times in cacodylate buffer and postfixed for 2 h on ice in 1% osmium tetroxide and 2% glutaraldehyde in 50 mM cacodylate buffer. After dehydratation the specimens were embedded in epon 812. Ultra-thin sections (80 nm) were done on a UCT ultramicrotome (Leica). They were analyzed with an electron microscope (912; Carl Zeiss, Inc.) and the images were taken with a camera (Bioscan 792; GATAN) using digital micrograph software.

The model of cardiac tube morphogenesis (Fig. 10 A) has been built using Blender 4.22.

Video 1 depicts the CB morphogenesis in the course of cardiac lumen formation and is the basis for building Fig. 1. Videos 2–4 allow the building of Fig. 2. Video 2 shows actin dynamics during cardiac lumen formation, and Videos 3 and 4 reveal the requirement of slit and robo/robo2 functions, respectively, in cardiac lumen formation.

Fig. S1 presents the ultrastructure of wild-type CBs at different stages of cardiac tube development. Dlg localization in wild-type and slit², dg62, and how18 mutants is described in Fig. S2. Additional examples of cardiac tube phenotypes in slit², dg62, how18, and dg323 mutant embryos are observed in Fig. S3. Fig. S4 describes How expression in the cardiac tube and in the midline glia. Finally, transheterozygous mutant embryos, slit²/how18 or slit²/dg62, have been analyzed for their cardiac lumen defects in Fig. S5.

We thank S. Baumgartner, W.M. Deng, J. Knoblich, and M. Schneider for antibodies and all the helpful material and advice they provided. We are grateful to T. Volk and to the Developmental Studies Hybridoma Bank for antibodies; B. Dickson, A. Paululat, the Bloomington Stock Center, and the Kyoto Stock Center for fly stocks; and to J. Gros for his help in building the model of cardiac tube formation using Blender. We also thank R. Kelly, T. Lecuit, and S. Zaffran for advice and helpful discussions; M. Gettings for critical reading of the manuscript; J.P. Chauvin, A. Aouane, and J.P. Arsanto for electron microscopy studies; P. Weber for imaging; and S. Long and F. Graziani for fly assistance.

This work was supported by the Centre National de la Recherche Scientifique, the Institut Nationale de la Santé et de la Recherche Médicale, the Association de la Recherche contre le Cancer, and the Association Française contre les Myopathies. C. Medioni was supported by a postdoctoral fellowship from the Association de la Recherche contre le Cancer and M. Zmojdzian by a postdoctoral Institut Nationale de la Santé et de la Recherche Médicale fellowship.