Junctions and vesicular trafficking during Drosophila cellularization
Thomas Lecuit


The cellularization that converts the syncytial Drosophila embryo into thousands of distinct cells is a hybrid form of cleavage. It derives from cytokinesis and has acquired specific features required for epithelial biogenesis. Cellularization generates an epithelial layer in which adjacent cells are connected by apical adherens junctions. If this process goes awry, subsequent development is dramatically affected, in particular tissue remodelling during gastrulation. Cellularization is associated with the invagination of the plasma membrane between adjacent nuclei at the cell cortex, the formation of a basal-lateral surface and the assembly of apical adherens junctions. The regulated mobilization of intracellular pools of vesicles at defined sites of the plasma membrane underlies membrane growth and surface polarization. Genetic approaches have identified conserved core cellular pathways required for these processes, such as vesicular trafficking along the biosynthetic and endocytic routes, and vesicular insertion into the plasma membrane. The novel proteins Nullo and Slam, which are specifically induced during cellularization, represent developmental regulators of membrane growth during cellularization.


Early embryonic development in most organisms involves the formation of a sheet of cells that then undergoes dramatic rearrangements during gastrulation. The pathways that yield the first embryonic tissues can be quite different. They always comprise a phase of cell division, and the end result is often the production of an epithelial tissue that is ready to undergo extensive remodelling. In the mouse embryo, cell divisions precede compaction, the process whereby un-polarized cells together form an epithelial sphere. In amphibians such as Xenopus, mitosis produces new epithelial cells at each round of division. The polarity of a given cell is thus transmitted to its descendents. In Drosophila, and many other arthropods, the nuclei first divide in a syncytium and, later on, thousands of epithelial cells are produced simultaneously around the circumference of the embryo by invagination of the plasma membrane (Fig. 1) (Foe et al., 1993; Schejter and Wieschaus, 1993a). Cellularization is a remarkable process that seems to have appeared several times in evolution. For instance, the gametophytes in plants, such as in Arabidopsis thaliana, also form by cellularization of a large syncytium (Sorensen et al., 2002).

Fig. 1.

Cellularization links embryonic cleavage to the formation of an epithelial layer. The plasma membrane (PM, green) invaginates between the nuclei (N, black) located at the cortex of the syncytial embryo shown at the onset (A) and at the end (B) of cellularization. When cellularization is completed, 5000 columnar epithelial cells are formed. The apical adherens junctions (AAJ, red in B) have assembled and ensure the cohesion of the epithelial layer. Cellularization is thus associated with the growth and polarization of the cell surface. (C,D) Confocal sections of a cellularizing embryo during slow phase (C) and at the end of fast phase (D) corresponding to the boxed areas shown in A and B. The embryos are stained with antibodies to the integral membrane protein neurotactin (Nrt, green) that marks the entire cell surface, to the junctional protein PaTJ (turquoise) that marks the membrane front called the furrow canal (FC), to the small GTPase Rab11 (red), a marker of apical recycling endosomes (see Fig. 2), and with the DNA dye Hoechst (Hoe, dark blue) to show the nuclei. Bar, 5 μm.

Drosophila cellularization is a hybrid process. It is akin to embryonic cell division and is based on cytokinesis. The membrane invaginates between nuclei localized at the cortex of the embryo, and several genes required during standard cytokinesis are also important for cellularization. However, the main product of cellularization is an epithelium. Cellularization has thus acquired mechanisms important for the polarization of the invaginating plasma membrane and the assembly of apical adherens junctions (AAJs) between each cell in the newly formed epithelium. This hybrid nature makes cellularization a particularly interesting system for studies of cell polarization during development. Although some mechanisms may be specific to cellularization, others might share more features with other cellular processes than is superficially apparent. Recent reports show that several widely conserved proteins required for the formation of junctions and for vesicular traffic are indeed important during cellularization (Lee et al., 2003; Pelissier et al., 2003; Riggs et al., 2003). These studies confirm the idea that cellularization requires the regulated insertion of vesicles into the plasma membrane.

When the fertilized Drosophila egg is laid, it is a large (0.5 mm long), ellipsoid cell. Within about 2 hours, 13 synchronous cycles of nuclear division in a single cytoplasm (the syncytium) yield about 5000 nuclei localized at the cortex of the syncytial embryo (Fig. 1A). During mitosis, the plasma membrane invaginates slightly and partially isolates each dividing nucleus, forming a so-called metaphase or pseudo-cleavage furrow. When nuclei enter interphase 14, mitoses are blocked, the nuclei stay in interphase, and membrane invagination becomes much more significant: within about 1 hour of development from the entry into interphase 14, the invaginating membrane packages single nuclei, organelles and cytoskeletal elements into 5000 cells and forms a columnar epithelium 35 μm tall (Fig. 1B-D). Cellularization proceeds first in a slow phase (0.3 μm/minute for 40 minutes) and then in a fast phase (0.8 μm/minute for 20 minutes). At the onset of cellularization, the plasma membrane forms a dome that is rich in villous projections (VPs), the somatic bud above each nucleus (Fig. 2A,B). Adjacent somatic buds are separated by shallow invagination of the plasma membrane (Fig. 2A,B, arrowhead). These fold to form donut-shaped structures called furrow canals (FCs), which are separated from the contiguous plasma membrane by basal adherens junctions (BAJs) (Fig. 2C,C',D) (Hunter et al., 2002; Hunter and Wieschaus, 2000; Muller and Wieschaus, 1996). The FC forms the membrane front as it invaginates (Fig. 2C,C'). The FC is analogous in composition to the contractile ring of standard cytokinesis and contains F-actin (Warn and Robert-Nicoud, 1990), myosin II (Lecuit et al., 2002; Royou et al., 2003; Sisson et al., 2000; Young et al., 1991), anillin (Field and Alberts, 1995), cofilin, spectrins (Thomas and Williams, 1999), septins (Adam et al., 2000; Fares et al., 1995) and formins/diaphanous (Afshar et al., 2000). FCs form adjacent to the BAJs, which contain complexes of the cell-cell adhesion molecule E-cadherin and β-catenin (termed Armadillo in Drosophila). The FC also contains junctional proteins such as PATJ (originally called Discs lost, DLT) (Bhat et al., 1999; Pielage et al., 2003). The hybrid nature of cellularization (combining cytokinesis and epithelial polarization) is thus already reflected in the organization of the membrane front during the slow phase of cellularization.

Fig. 2.

Cell-surface remodelling and vesicular transport pathways underlying cellularization. Transmission electron micrographs (A,C,C') and schematic representations of embryos at the onset (A,B,E) and 20 minutes after the beginning (C,C',D,F) of cellularization. (A) Above each nucleus (N), the plasma membrane (PM) forms a dome called the somatic bud that is rich in villous projections (VP). Adjacent somatic buds are separated by a flat depression of the plasma membrane (arrowhead). (B) This membrane domain contains specific sets of junctional proteins such as E-cadherin/β-catenin complexes, myosin II, PaTJ and Slam, as well as DLG/SAP97 and the multi-PDZ-containing protein Scribbles (Scb). (C,C') As the plasma membrane folds back and forms the furrow canal (FC), (D) three distinct and adjacent membrane domains assemble containing MyoII/PaTJ/Slam in the FC (red), E-cadherin/β-catenin (green) at the basal adherens junction (BAJ) and DLG/Scb (blue) just apical to the BAJ. The two apposed cell surfaces (C, black arrowheads) grow progressively during the slow phase. Several vesicular organelles can be detected in the vicinity of the cell surface (C, white arrowheads). The formation of the BAJ as shown in D is dependent on Slam (Lecuit et al., 2002). Junctional stabilization requires Nullo, which localizes more broadly in the FC, at the BAJ and apical to the BAJ (Hunter and Wieschaus, 2000). Failure to form or stabilize the BAJ is always associated with an inhibition of membrane invagination in slow phase. (E,F) Trafficking through sub-apical perinuclear recycling endosomes (RE, pink) is required for membrane invagination. This process is controlled by the small GTPase Rab11 and Dynamin/Shibire. Vesicular transport from the Golgi apparatus to RE as well as from the apical plasma membrane to RE by Rab5 might also contribute to this trafficking pathway. Bar, 2 μm.

The fast phase, which produces more than half of the total lateral membrane surface area, is also associated with the formation of AAJs, which connect adjacent cells in the newly formed epithelium and are required for the integrity of the tissue. During the fast phase, E-cadherin/β-catenin complexes first appear in punctate structures in the apico-lateral region of the plasma membrane and form the spotted-adherens junctions (SAJs). At the end of cellularization, E-cadherin/β-catenin complexes form a more continuous belt that will progressively constitute the proper AAJs as it matures and stabilizes. Additional proteins important for the stabilization of the E-cadherin/β-catenin complexes are recruited at AAJs at the end of cellularization, including F-actin, atypical protein kinase C (aPKC) (Wodarz et al., 2000), and the PDZ-domain-containing proteins PAR-3 (termed Bazooka in Drosophila) (Kuchinke et al., 1998), PAR-6 (Petronczki and Knoblich, 2001) and PATJ (Bhat et al., 1999).

The prevalent model of cellularization ten years ago was largely influenced by the understanding of cytokinesis at that time, and proposed that the plasma membrane invaginates following specific rearrangements of cytoskeletal elements (Foe and Alberts, 1983; Foe et al., 1993; Schejter and Wieschaus, 1993a; Schejter and Wieschaus, 1993b). In this context, the plasma membrane was believed to follow actin and microtubule (MT) rearrangements passively. For instance, the requirement for intact actin filaments during cellularization (Foe and Alberts, 1983), and the similar biochemical compositions of the FC and of the cytokinesis front, in particular the accumulation of myosin II in this membrane domain, reinforced the idea that a contractile actin-myosin network at the membrane front might drive membrane invagination. In addition, Foe showed that MTs are essential for cellularization and that, at the onset of cellularization, astral MTs dramatically extend their + ends basally towards the dense yolk granules present about 30-40 microns inside the syncytial embryo (Foe and Alberts, 1983). Foe proposed that growth of astral MTs might push the membrane inward (Foe et al., 1993).

Since about five years ago, several studies of cellularization have progressively changed this perception in parallel with new insights into the mechanisms of standard cytokinesis. The idea that vesicular insertion in the plasma membrane is an active process that is essential during cellularization (Lecuit and Wieschaus, 2000) and more generally during cytokinesis (Drechsel et al., 1997; Jantsch-Plunger and Glotzer, 1999), which was first proposed several decades ago (Bluemink and de Laat, 1973; Fullilove and Jacobson, 1971), became more prevalent (reviewed by Glotzer, 2001; O'Halloran, 2000) and changed the model of cellularization.

The apparent increase in the plasma membrane surface area during cellularization indicates two possible scenarios: cellularization could either deploy apical membrane stored in the VPs (Fullilove and Jacobson, 1971; Turner and Mahowald, 1976), or involve the insertion of intracellular membrane vesicles into the plasma membrane (Fullilove and Jacobson, 1971; Loncar and Singer, 1995). Various studies support the notion that intracellular membrane pools are indeed used as a raw material during cellularization and that vesicular trafficking is essential for this process. First, embryos containing mutations in the SNARE protein syntaxin 1 fail to cellularize (Burgess et al., 1997). Syntaxin 1 is required for insertion of vesicles into the plasma membrane in a variety of organisms. Pulse-labelling experiments in living embryos have probed the dynamics of the plasma membrane during cellularization (Lecuit and Wieschaus, 2000). These show that, after labelling of the plasma membrane during cellularization, specific regions of the cell surface become devoid of labelling first apically then apico-laterally. These regions could either define sites of vesicle insertion or sites of vesicle internalization. Further support for the role for vesicle exocytosis comes from studies showing that injection of brefeldin A, which inhibits endoplasmic reticulum (ER)-to-Golgi transport, reduces the speed of membrane invagination (Sisson et al., 2000); injection of antibodies directed against the Golgi-associated protein Lava-lamp results in similar defects (Sisson et al., 2000). In fact, the requirement for MTs during cellularization was also shown to stem largely from their implication in secretory trafficking (Lecuit and Wieschaus, 2000). The depolymerization of MTs at the onset of cellularization blocks the post-Golgi trafficking to the plasma membrane of the transmembrane cell adhesion protein neurotactin (Nrt) (Lecuit and Wieschaus, 2000) and causes major defects in ER-to-Golgi vesicular transport (T. Lecuit, unpublished) in conjunction with the inhibition of membrane invagination.

This different interpretation of the requirement for intact MTs during cellularization is characteristic of the recent conceptual change in our understanding of cellularization. Together, these studies have substantiated the notion that cellularization relies on the mobilization of vesicles into the plasma membrane, in part from the secretory pathway. Moreover, the pulse-labelling experiments during cellularization show that membrane growth is a polarized process. The polarized remodelling of the cell surface, and in particular the polarized targeting of pools of vesicles, might be important for the growth of the plasma membrane. In addition, such a polarized targeting might underlie the polarization of the cell surface. This idea is supported by the observation that the formation of SAJs (marked by the localization of E-cadherin) coincides with the insertion of vesicles in the apicolateral region of the cell surface during the fast phase (inferred from pulse-labelling experiments) (Lecuit and Wieschaus, 2000). These observations suggest a link between membrane growth and surface polarization. However, they leave open the questions of the identity of the specific organelles through which vesicular trafficking occurs during cellularization and the precise location of where vesicular insertion finally occurs.

Trafficking through recycling endosomes is required for furrowing

The transport of vesicles along the endocytic pathway proceeds through distinct organelles. Internalized vesicles at the plasma membrane are targeted to early endosomes, also called sorting endosomes. Vesicle endocytosis requires dynamin (termed Shibire in Drosophila) (Chen et al., 1991; van der Bliek and Meyerowitz, 1991), which is a GTPase that regulates the fission of endocytic coated vesicles (McNiven, 1998), and trafficking to early endosomes is dependent on the small GTPase Rab5 (Bucci et al., 1992; Horiuchi et al., 1997). Endocytosed vesicles are then either transported back to the plasma membrane, to late endosomes and lysosomes for degradation or to the peri-centriolar recycling endosome, from which material is also recycled to the plasma membrane. In polarized epithelial cells, the apical recycling endosome is required for cell polarization and the transfer of vesicles from the apical to the basolateral surface (or vice versa) by transcytosis (Apodaca et al., 1994; Mostov et al., 2000). Trafficking through recycling endosomes is controlled in part by the small GTPase Rab11 both in mammals and in Drosophila (Dollar et al., 2002; Ullrich et al., 1996). Rab11 localizes to this organelle in vertebrates and in invertebrates (Dollar et al., 2002; Sonnichsen et al., 2000). Dynamin also controls budding of vesicles from recycling endosomes in mammalian cells (van Dam and Stoorvogel, 2002).

In rab11-hypomorphic mutant embryos, metaphase furrows form randomly around nuclei, and cellularization is not uniform, resulting in multinucleate cells. Injection of dominant-negative Rab11 (Rab11S25N) a few minutes before cellularization strongly inhibits membrane invagination during cellularization, in both the slow and fast phases. Such a defect is not seen when a wild-type form of Rab11 or a dominant-negative form of Rab1, which is implicated in ER-to-Golgi transport, is injected. Both when dominant-negative Rab11 is injected and in rab11-mutant embryos, the nuclei fail to align correctly in an apical position and `fall' from the cortex. This phenotype is reminiscent of nuclear-fallout mutant embryos. Nuclear fallout (NUF) is a member of the arfophilin family of proteins (Hickson et al., 2003), arfophilin being an ADP ribosylation factor (Arf)-binding protein that binds the small GTPases Arf5 and Rab11 in mammalian cells. NUF is required for membrane invagination during the formation of metaphase furrows as well as during cellularization (Rothwell et al., 1998; Rothwell et al., 1999). Both NUF and Rab11 localize near the centrosomes during cellularization and at earlier stages. They reside in tubulo-vesicular structures around the MT-organizing centers (MTOCs) and physically interact in Drosophila (Hickson et al., 2003; Riggs et al., 2003).

Rab11 and NUF thus probably co-operate in the exocytosis of vesicles required for the invagination of the plasma membrane during cellularization, as well as at earlier stages. Riggs et al. have emphasized the remodelling of the actinmyosin cytoskeleton at the cell surface over the need for vesicle insertion into the plasma membrane during furrow formation (Riggs et al., 2003). By contrast, Pelissier et al. favour a role for membrane insertion to explain the phenotypes. In light of the fact that the actin-myosin II cytoskeleton appears non-essential for cellularization (Royou et al., 2003), Rab11 is more likely to provide membrane material necessary for invagination. Regardless of its precise function, it clearly plays a role in trafficking through recycling endosomes during membrane invagination (Fig. 2E,F).

Dynamin/Shibire also functions during cellularization. Experiments two decades ago indicated first that shibire plays a role in the early Drosophila embryo (Swanson, 1981). It is possible to block shibire function at defined time points with a temperature-sensitive mutant (shibire-ts). Pelissier et al. shifted shibire-ts embryos to the restrictive temperature (RT) at the onset of cellularization and found that membrane invagination is blocked (Pelissier et al., 2003). However, when the shift is performed at the end of the slow phase, cellularization is only moderately affected, which shows that shibire is more critically required during slow phase. In addition to controlling endocytosis from the plasma membrane, shibire appears to control vesicular trafficking through Rab11-positive endosomes. Pelissier et al. shifted shibire-ts mutant embryos to the RT at the onset of cellularization, at a time when the cell adhesion transmembrane protein Nrt, which is synthesized de novo during slow phase (Lecuit and Wieschaus, 2000), is not synthesized yet (Pelissier et al., 2003). Surprisingly, after 20 minutes at the RT, the newly synthesized Nrt, which traffics normally through the ER and the Golgi, is not properly inserted in the plasma membrane in shibire-ts mutant embryos. Instead, Nrt accumulates in the sub-apical Rab11-positive recycling endosomes. Shibire is detected in Rab11-positive endosomes during the slow phase, which suggests that it also controls the budding of vesicles from recycling endosomes, as has been reported in some mammalian cells (van Dam and Stoorvogel, 2002). Electron microscopy has revealed a large accumulation of coated pits on sub-apical endosome-like compartments in shibire mutant embryos (Pelissier et al., 2003), further supporting this conclusion.

Rab11-containing sub-apical recycling endosomes act as trafficking intermediates during cellularization, and vesicular exocytosis from this compartment might be required for membrane invagination. Rab11 (Low et al., 2003; Skop et al., 2001) also functions in conventional cytokinesis. Two series of questions remain. Where do the vesicles upstream of Rab11-containing endosomes come from? And where are the exocytic vesicles finally inserted in the plasma membrane?

The finding that newly synthesized Nrt accumulates in Rab11-containing endosomes in shibire mutants (Pelissier et al., 2003) suggests that the secretory pathway provides vesicles upstream of Rab11-positive endosomes and that vesicular trafficking through the secretory pathway intersects to some extent with recycling compartments. This can also be inferred from studies of cell polarization in Mabin-Darby canine kidney (MDCK) cells (Ang et al., 2003; Folsch et al., 2003). The clathrin adaptor complex AP1, which is required for secretory trafficking, exists in two forms: (1) a ubiquitous form, AP1A, contains the μ1A subunit, and (2) an epithelial-cell-specific form, AP1B, contains the μ1B subunit required for basolateral vesicular targeting (Folsch et al., 1999). Interestingly, whereas AP1A is required for trans-Golgi network (TGN)-to-endosome transport and localizes in the TGN, AP1B localizes to recycling endosomes and recruits to this compartment Sec8 and Exo70 (Ang et al., 2003; Folsch et al., 2003), two components of the exocyst complex required for basolateral targeting in epithelial cells (Grindstaff et al., 1998; Yeaman et al., 2001). Since a dominant-negative mutant of Rab5, a regulator of vesicular trafficking from the plasma membrane to early endosomes, slows down membrane invagination, albeit to a lesser extent than dominant-negative Rab11, endocytosis from the plasma membrane may also be important during cellularization (Pelissier et al., 2003). The direct visualization of vesicular trafficking during cellularization will be necessary to address these questions in the future.

Junctions and membrane growth

The BAJ forms together with the FC in the first 10 minutes of cellularization (Hunter and Wieschaus, 2000; Lecuit and Wieschaus, 2000). This process involves the re-organization, within the plane of the membrane, of many junctional proteins. These are initially randomly distributed within the flat invagination of the membrane (Fig. 2B) and subsequently segregate into distinct, adjacent membrane domains when the BAJs and FCs have formed (Fig. 2D).

Two genes, nullo and slam, that are required for the invagination of the plasma membrane during cellularization are also involved in the formation or the stabilization of the BAJ. The product of the nullo gene controls the accumulation of E-cadherin/β-catenin complexes at BAJs. Nullo is a novel myristoylated protein that has a cluster of basic residues required for its localization at the plasma membrane in the BAJ and the FC (Hunter et al., 2002). Slam is a novel protein with no apparent conserved domain. Slam colocalizes with myosin II and is required for the recruitment of myosin II at the plasma membrane in the FC (Lecuit et al., 2002). Both nullo and slam are induced specifically at the onset of cellularization, the protein synthesized being of zygotic rather than maternal origin (Hunter and Wieschaus, 2000; Lecuit et al., 2002; Stein et al., 2002). In slam-mutant embryos, vesicular insertion of Nrt into the plasma membrane is compromised, which suggests that the defect in membrane invagination might be caused by a defect in vesicle insertion at junctions. This proposal remains hypothetical because vesicular insertion has never been observed directly during cellularization. However, it is supported by the fact that a conserved complex of proteins involved in vesicular insertion from budding yeast to humans, the exocyst (Lipschutz and Mostov, 2002), is recruited to AJs and is involved in the formation of the basolateral cell surface during the polarization of epithelial MDCK cells (Grindstaff et al., 1998). The Sec6 and Sec8 components of the exocyst in particular localize at mammalian AJs (Grindstaff et al., 1998; Yeaman et al., 2001). Interestingly, sec5, a gene that encodes another component of the exocyst, is required for membrane invagination during Drosophila cellularization (T. Schwarz, personal communication).

A recent report further supports the role of the BAJ in the growth of the plasma membrane during cellularization (Lee et al., 2003), implicating the junctional protein Discs large (DLG) and the transmembrane protein Strabismus (Stbm) in cellularization. Stbm is involved in planar cell polarity in epithelial cells (Fanto and McNeill, 2004). In stbm-mutant embryos, cellularization appears inhibited in a way that is reminiscent of that in nullo-mutant embryos. The membrane does not invaginate uniformly and produces multinucleate cells. In addition, the junctional proteins DLG and PaTJ are not properly localized at the BAJ, which suggests that junctions do not assemble or are not stabilized normally in the mutant, as happens in nullo mutants. Stbm and DLG physically interact (through the first two PDZ domains of DLG). However, the in vivo localization data are not that simple and do not appear to support a straightforward functional interaction. In some epithelial tissues, DLG and Stbm partially colocalize at the cell surface or on uncharacterized intracellular organelles, but in other tissues, including those that undergo cellularization, the proteins have distinct localizations. Stbm and DLG might interact in small vesicles that are difficult to observe by standard confocal microscopy, and this interaction could be linked to the formation or the stabilization of the BAJ.

Lee et al. have used mammalian cells to probe the interaction between Drosophila Stbm and the human orthologue of DLG, SAP97 (Lee et al., 2003). SAP97 and Stbm interact in vitro and display a better colocalization in intracellular organelles than do Stbm and DLG. In Drosophila imaginal disc epithelia, ectopic expression of Stbm and SAP97 causes the formation of enlarged septate junctions containing neurexin. Conversely, in dlg-mutant clones induced in imaginal discs, the epithelial junctions are disorganized. Although Lee et al. interpret these results to mean that the two proteins are essential for membrane formation, it is safer to conclude that Stbm and DLG are simply required for the formation of junctions. Given the localization of DLG just basal to the E-cadherin/β-catenin complex, DLG and Stbm might more specifically control the formation or the stabilization of the basal junctional complexes and thereby the formation of AJs.

Concluding remarks

Our picture of cellularization has considerably changed in two ways in the past five years. We have gradually moved from the idea that it is a Drosophila oddity to the realization that it is one of several pathways leading to the formation of an epithelium and that, as such, important cell biological questions about the process should be addressed (Lecuit and Wieschaus, 2002; Nelson, 2003). We have begun to identify new regulators of membrane growth and invagination during cellularization. Some of the trafficking pathways that underlie vesicular transport during cellularization are also now known. The re-organization of the cell surface that accompanies and is required for membrane growth has brought to light the importance of the BAJs during cellularization. As a result, the picture has also enlarged, leaving many important questions open. Answering them will first require better tools to follow vesicular trafficking and refine further the analysis of mutants in trafficking pathways and junction assembly. It is also important to keep in mind an essential product of cellularization that remains largely mysterious, namely the AAJs, during the fast phase. The picture will thus certainly incorporate more features and become even larger in the near future.


I thank Anne Pelissier for the staining of embryos shown in Fig. 1C,D and Jean-Paul Chauvin for his help with obtaining the electron micrographs shown in Fig. 2. Research in our laboratory his supported by an ATIP grant from the CNRS, a `subvention libre' from the Association pour la Recherche contre le Cancer, the Fondation pour la Recherche Médicale, and the EMBO Young Investigator Programme.


View Abstract