Fat2 acts through the WAVE regulatory complex to drive collective cell migration during tissue rotation

Collective and directed cell movements are essential for diverse developmental processes in animals. Recent advances in live imaging and ex vivo culture conditions of Drosophila melanogaster ovaries have led to the discovery of a new type of morphogenetic movement that drives egg elongation during fly oogenesis (Haigo and Bilder, 2011; Bilder and Haigo, 2012). Elongating follicles, so-called egg chambers, rotate around their anterior–posterior (A-P) axis during the initial stages of oogenesis. As follicle cells migrate around their circumferential axis, they build up a polarized ECM (Haigo and Bilder, 2011). The planar-polarized ECM has been proposed to function as a molecular corset to restrict radial expansion and to force egg chamber elongation (Haigo and Bilder, 2011; Bilder and Haigo, 2012). Like the ECM, the basal actin cytoskeleton forms a corset composed of parallel actin bundles aligned perpendicular to the A-P axis. Thus, the timing and orientation of follicle rotation coincide with the formation of a planar polarized ECM and basal actin cytoskeleton.

Which molecules drive the global egg chamber rotation? Two recent studies unraveled an important role of the Arp2/3 complex–activating WASP family verprolin homologous (WAVE) and its interaction partner Ena/Vasodilator-stimulated phosphoprotein (Ena/VASP) in egg chamber elongation, two known key actin regulators promoting single-cell migration in different systems (Cetera et al., 2014; Chen et al., 2014b). WAVE forms a heteropentameric complex, the WAVE regulatory complex (WRC), comprising the Abelson interacting protein Abi, Kette/Nap1, Specifically Rac1 associated protein 1 (Sra-1), and the Hematopoietic Stem Progenitor Cell 300 (HSPC300; Eden et al., 2002; Chen et al., 2010). The loss of either the WRC or its conserved interaction to Ena in Drosophila results in abnormally shaped, round eggs (Cetera et al., 2014; Chen et al., 2014b). In vivo membrane labeling using a Neuroglian-GFP trap revealed that migrating follicle cells form membrane protrusions extending in the direction of rotational movement (Cetera et al., 2014). WAVE and Ena are enriched at the tips of these membrane protrusions, and a model has been proposed in which WAVE defines the protrusive edge of each follicle cell and WAVE-dependent lamellipodia drive collective follicle cell migration (Cetera et al., 2014). Consistently, RNAi-mediated depletion of Abi in all follicle cells completely blocks egg chamber rotation (Cetera et al., 2014).

By contrast, less clear is how global tissue rotation is coordinated and how the directionality of migrating follicle cells as an epithelial sheet is controlled. Elongating egg chambers always rotate perpendicular to the A-P axis, either clockwise or counterclockwise (Haigo and Bilder, 2011). Genetic studies in Drosophila have already revealed a close functional relationship between the ECM, cell adhesion receptors and actin dynamics during egg chamber elongation. Mutations in genes encoding components of basement membrane such as laminin, collagen IV, their receptors such as βPS integrin, and the receptor phosphatase Dlar and the loss of the atypical cadherin Fat2 result in a common round-egg phenotype (Gates, 2012). These mutations lead to a misorientation of the basal actin bundles (Viktorinová et al., 2009; Gates, 2012). However, not all of the round-egg mutants show a fully penetrant phenotype varying in the frequency of round eggs produced. For example, loss of dlar function results in a moderate phenotype with ∼14% round eggs (Bateman et al., 2001). Mosaic analysis further revealed that Dlar acts nonautonomously to control the basal actin cytoskeleton polarization (Bateman et al., 2001). A similar nonautonomous requirement was found for several round egg genes such as fat2/kugelei (Viktorinová et al., 2009; Gates, 2012). Remarkably, only fat2 mosaic egg chambers containing more than 60% mutant follicle cells show a disruption in the planar actin alignment in relation to the A-P axis (Viktorinová et al., 2009). Thus, the mutant cells seem to be dragged along by remaining wild-type cells. This also suggests that cell–cell interactions are required for tissue-level alignment of the basal actin filaments. Fat2 is indeed a transmembrane cell adhesion receptor that might function in contact-based cell–cell signaling as known for the related Fat protein, an important mediator of planar cell polarity (PCP) signaling in Drosophila wing and eye development. Like all members of the Fat cadherin subfamily Drosophila Fat2 contains a large extracellular region containing 34 cadherin repeats, one laminin A subunit G domain and five EGF motifs in the membrane proximal region that might mediate either heterophilic or homophilic interactions as recently demonstrated for rat Fat2 (Nakayama et al., 2002).

Given the disrupted planar-polarized basal actin cytoskeleton in fat2/kugelei mutants, it has been proposed that Fat2 might control an alternative PCP pathway in follicle cells (Bilder and Haigo, 2012). Localization studies with a GFP-tagged transgene further support this idea (Viktorinová et al., 2009; Gates, 2012). A Fat2-GFP transgene has been reported to localize at tricellular junctions in developing egg chambers until stage 5, and concomitant with the proposed onset of egg chamber rotation, Fat2 relocalizes to the lagging edge of cells (Viktorinová et al., 2009; Gates, 2012). However, recent live-imaging analyses demonstrate that egg chamber rotation begins much earlier, before an asymmetric localization of Fat2-GFP can be observed (Cetera et al., 2014). Thus, the relocalization of Fat2 seems not to be the instructive signal for directional follicle cell migration.

In mammalian cells, the most similar mammalian Fat2 orthologue (Fat1) localizes at the tips of membrane protrusions and cell–cell contacts, where it regulates junctional actin dynamics and cell polarity through Ena/VASP proteins (Moeller et al., 2004; Tanoue and Takeichi, 2004). The cytoplasmic tail of the Drosophila Fat2 protein, however, lacks all three EVH1 binding sites, which mediate Ena/VASP binding of mammalian Fat1, suggesting that Drosophila Fat2 might be linked to the actin cytoskeleton by an alternative pathway. We recently described a novel mechanism directly linking the WRC to diverse cell adhesion molecules, ion channels, and G-protein–coupled receptors (Chen et al., 2014a). All these transmembrane receptors contain a conserved peptide motif, the WRC interacting receptor sequence (WIRS) that directly binds a conserved interaction surface on the WRC formed by the Sra-1 and Abi subunit (Chen et al., 2014a). In flies, disruption of the WIRS binding surface in Abi results in impaired photoreceptor axon targeting, defective egg morphology, and partial female sterility (Chen et al., 2014a). Here, we identify the atypical cadherin Fat2 as a novel WIRS ligand and explore a molecular pathway in which Fat2 and Dlar act through the WRC to control collective cell migration during Drosophila oogenesis.

Recent data indicate that egg chamber rotation requires WAVE-dependent protrusions at each follicle cell's leading edge (Cetera et al., 2014). However, imaging of the actin cytoskeleton of rotating egg chambers has been largely limited to fixed samples or protrusion dynamics were visualized in vivo by membrane markers (Haigo and Bilder, 2011; Cetera et al., 2014). To better understand WRC function in actin-driven cell protrusions during follicle cell migration, we used a LifeAct-EGFP transgene that enables the visualization of actin-rich protrusions of migrating follicle cells at high resolution. As previously reported, a polarized rotation of wild-type egg chambers can clearly be observed from stage 4 until the end of stage 8 when border cell migration starts (Video 1; He et al., 2011; Bilder and Haigo, 2012; Cetera et al., 2014). LifeAct-EGFP not only marks cortical F-actin but also highly accumulates at actin-rich protrusions (Fig. 1 A, arrowheads; and Video 2 A). Follicle cells migrate as an epithelial sheet forming thin filopodial protrusions at the cell front rather than sheet-like lamellipodial protrusions found in single migrating cells (Fig. 1 A, open arrowheads; and Video 2 A). Strikingly, the largest cell protrusions emerge predominantly at the tricellular junctions at the cell’s leading edge (Fig. 1 A, closed arrowheads; and Video 2 A). Unlike the small filopodial protrusions that uniformly extend from the cell membrane, junctional protrusions form highly dynamic tails that propel against the direction of movement (Fig. 1 A, closed arrowheads; and Video 2, A and B). The length of these whip-like protrusions remarkably increases during egg chamber maturation, and they assume a flagella-like shape and dynamics at the end of stage 8, when rotations slow down (Fig. 1 B, closed arrowheads; and Video 2 C). Whip-like protrusions at junctional contacts also can be observed with a fluorescent F-actin probe based on the calponin homology domain of utrophin (Utr-CH; Burkel et al., 2007) expressed under the endogenous spaghetti-squash (sqh; Rauzi et al., 2010) promoter, suggesting that these structures are not an artifact of LifeAct-EGFP overexpression (Video 2 D). All protrusive actin structures of migrating follicle cells are highly labile and thus are poorly preserved by standard fixation methods (Fig. 1, C and C′). However, we found optimal preservation of acute wild-type egg chamber preparations after fixation without any subsequent permeabilization and washing steps (Fig. 1, D and D′). High resolution structured illumination microscopy (SIM) of fixed wild-type egg chambers stained with fluorescently labeled phalloidin reveals numerous long filopodial and whip-like protrusions as similarly observed in vivo (Fig. 1, D and D′).

To better visualize migration and protrusion dynamics of single follicle cells within the epithelium we made use of the da-Gal4 driver, which drives LifeAct-EGFP in a mosaic pattern (Fig. 1 E and Video 3, A and B). 4D live imaging analysis of thicker z-stacks through the mosaic follicle epithelium reveals that actin-rich filopodial protrusions are also visible along the baso-lateral cell border and at the apical side (Fig. 1 E and Video 3 C). In contrast to the basal side, filopodial protrusions at the apical side are not polarized along the migration direction but are rather radially symmetrical (Fig. 1, E1–E4; and Video 3 C). Whip-like protrusions are formed most basally or just beneath the basal surface at tricellular junctions (Fig. 1, E1–E4; and Video 3 C; compare scheme in Fig. 1 F). Both types of actin-rich protrusion, the filopodial and the junctional whip-like protrusions, depend on WRC function, because they are completely missing in sra-1–deficient follicle cells (Fig. 2 A and Video 4). Suppression of sra-1 function also completely abolishes egg chamber rotation (see Fig. 6 G), confirming that WAVE-induced protrusions provide the driving force for follicle cell migration and egg chamber rotation (Cetera et al., 2014). Supporting this idea, wave and abi mutant cell clones completely lack actin-rich protrusions, large junctional and additionally cortical F-actin accumulations (Figs. 2 B and S1). Consistent with our previous studies, all WRC components are required for the integrity of the complex and the Abi protein level is strongly reduced in wave mutant cell clones (Stephan et al., 2011; Fig. 2 B). Mutant clones lacking the WRC show no obvious defects in the formation or alignment of basal actin bundles as previously reported (Fig. 2 B, F-actin staining, asterisk; and Fig. S1 A; Viktorinová et al., 2009; Cetera et al., 2014). Thus, the remaining migrating wild-type cells might provide sufficient forces for cell migration driving the tissue-level based alignment of the actin cytoskeleton (Viktorinová et al., 2009; Horne-Badovinac, 2014).

We next analyzed the localization of the central WRC component Abi in developing egg chambers. Consistent with the observation that the WRC is required for actin-rich protrusions of migrating follicle cells, the WRC subunit Abi indeed localizes at the right time and space in follicle cells. Endogenous Abi as well as an EGFP-tagged Abi transgene expressed in abi mutant follicle cells are highly enriched at dynamic protrusions emerging from tricellular junctions and along the membrane (Fig. 2, B and C; and Video 5 A). At later stages (9 and10), follicle cells initiate oscillating acto-myosin contractions driving further egg chamber elongation (He et al., 2011). At this advanced stage of egg chamber maturation, when follicle cell migration ceases, the polarized localization of Abi changes into a more even distribution at basal cell–cell junctions (Fig. 2 D). Live-imaging analysis of stage 10 egg chambers expressing EGFP-tagged Abi documents these rhythmic cell contractions but also uncovers dynamic finger-like protrusions between follicle cells marked by Abi-EGFP (Fig. 2 D, arrowheads; and Video 5 B). Genetic mosaics of stage 10 egg chambers lacking a functional WRC reveal an additional late function of the WRC in the organization or the maintenance of basal actin stress fibers mediating oscillating contractions (Fig. S1 B). abi mutant cell clones show a strong disorganization of basal actin bundles (Fig. S1 B), as previously described in clones deficient for the receptor tyrosine phosphatase Dlar (Bateman et al., 2001).

The prominent localization of the WRC at tricellular junctions resembles the distribution of Fat2 and Dlar reported previously (Bateman et al., 2001; Viktorinová et al., 2009). Because no specific anti-Fat2 antibody is available, we used a functional fosmid-based Fat2-GFP transgene (Viktorinová et al., 2009) to follow the localization of Fat2 with respect to the WRC in follicle cells. Structured illumination microscopy reveals an increasing overlap of Abi and Fat2-GFP at tricellular contacts in egg chamber development (Fig. 3, A–C; and Fig. S2). A reliable expression of Fat2-GFP is first seen in stage 3 follicle cells (Fig. 3 A). However, most prominent overlap of Abi and Fat2-GFP at tricellular contacts can be observed between stages 5 and 7 (Fig. 3, B and C; and quantification in Fig. 3, D and E). Live imaging of this weakly expressed Fat2-GFP confirms its accumulation at tricellular contacts of migrating stage 4/5 follicle cells (Video 6 A). Fat2-GFP remains enriched at tricellular contacts in vivo, although the initially weak expression signal increases over time, whereas Fat2-GFP becomes visible along the plasma membrane in stage 6–8 egg chambers (Fig. 3, F and G; and Video 6, A–C). Quantification of the relative GFP fluorescence intensity in living follicle cells confirms a significant accumulation of Fat2 at tricellular junctions between stages 5 and 9; the strongest localization at tricellular junctions is found at stage 6/7 (Fig. 3, F and G). In later stages, clones of follicle cells from stage 7/8 egg chambers expressing Fat2-GFP imply an enrichment of Fat2 at the lagging edge as recently reported (Fig. S3; Viktorinová and Dahmann, 2013).

Co-labeling with LifeAct-RFP further shows a striking overlap of Fat2-GFP with junctional actin-rich protrusions in migrating follicle cells (Fig. 4 A and Video 7, A and B). Fat2-GFP localizes at the tips of whip-like protrusions (Fig. 4 B and Video 7, A and B).

The overlap of Fat2 with the WRC at tricellular junctions suggests that Fat2 might be involved in directing WRC-dependent follicle cell protrusions. Supporting this notion, the C-terminal intracellular tail of Fat2 (dFat2-CT) contains three highly conserved WIRS motifs that could mediate a direct interaction with the WRC (Fig. 4 C). A purified GST-fusion of dFat2-CT binds to the wild-type WRC, but not to a mutant whose WIRS binding surface is disrupted by mutations (Fig. 4 D; 2MBP-WRC AW). Binding assays with distinct double mutant C-terminal Fat2 proteins (WIRS Δ1, 2; Δ1, 3; Δ2, 3; Δ1, 2, 3; see loading control in Fig. 4 E, right) further revealed that the second and, to a lesser extent, the third WIRS motif are crucial for WRC binding (Fig. 4 E). Thus, Fat2 is a novel WIRS ligand.

Loss of WRC components affects egg shape (Cetera et al., 2014; Chen et al., 2014b). To test the functional relevance of the Fat2–WRC interaction in egg elongation, we first analyzed flies lacking the conserved WIRS binding surface in Abi (AbiΔWIRS; Chen et al., 2014a). In these rescue experiments, we reexpressed either a wild-type Abi (Abi-WT) or a mutant Abi transgene bearing the two point mutations in the WIRS binding surface (AbiΔWIRS, Abi-AW). In contrast to Abi-WT rescue, female flies deficient for WIRS binding (AbiΔWIRS) show significant defects in egg shape (Fig. 5, A and B; Chen et al., 2014a). Quantifications by measuring the length-to-width ratios revealed a bimodal phenotype. The majority shows a normal shape (Fig. 5 C), but ∼20% of all AbiΔWIRS egg chambers are spherical as shown in Fig. 5 D. These 20% spherical egg chambers resemble the phenotype observed in kugelei mutants deficient for fat2 (egg index <2; Fig. 5, A and D). In contrast, loss of fat2 function or suppression of WRC subunits results in a fully penetrant round-egg phenotype (Fig. 5 A; Viktorinová et al., 2009; Cetera et al., 2014). The efficacy of RNAi-mediated knockdown depends not only on the transgenic line but also on the Gal4 driver used. As recently described, the GR1-Gal4 is expressed efficiently in follicle cells at all stages (Berns et al., 2014). Follicle cell–specific suppression of the WRC subunit Sra-1 using the GR1-Gal4 driver results in the strongest round-egg phenotype, whereas RNAi-mediated downregulation of either abi or wave function results in a milder round-egg phenotype (Fig. 5 A; Cetera et al., 2014; Chen et al., 2014b). The level of gene activity suppression by transgenic RNAi also depends on the Gal4 driver. Supporting this notion, GR1-driven suppression of the WRC interaction partner Ena significantly affects egg elongation, whereas expression of the same ena RNAi transgene by the weaker TJ-Gal4 driver has no significant effect on egg shape as recently described (Fig. 5 A; Cetera et al., 2014).

Like fat2 mutants, round AbiΔWIRS stage 7–9 egg chambers show strong defects in the planar polarized basal follicle cell actin cytoskeleton (Fig. 5 E). Unlike Abi-WT rescued follicle cells, actin filaments are no longer aligned perpendicular to the A-P axis in AbiΔWIRS egg chambers (Fig. 5 E). Analysis of actin bundle orientation documents strong orientation defects of actin bundles in AbiΔWIRS and fat2 mutant egg chambers (Fig. 5 F). The majority of actin bundles in AbiΔWIRS and fat2 mutant is not preferentially orientated perpendicular to the A-P axis (+90/−90°) as observed for actin filament bundles in Abi-WT rescued follicle cells (Fig. 5, E and F). This suggests that egg chamber elongation requires not only the WRC and Fat2 but also functional WIRS interactions.

Our previous data indicate that WIRS ligands can directly recruit the WRC to membranes (Chen et al., 2014a). To test whether the localization of the WRC depends on Fat2, we stained fat2 mutant mosaic egg chambers for endogenous Abi. In contrast to wild-type cells (positively marked by GFP expression), in fat2 mutant cells, Abi localization is strongly reduced at the basal follicle side (Fig. 6 A). Loss of fat2 also substantially reduces actin-rich protrusions at the basal side (Fig. 6 A). Remaining protrusions show some residual Abi localization at tricellular contacts (Fig. 6 A, arrowheads). No changes in F-actin and Abi staining can be observed at baso-lateral cell membranes in the fat2 mutant cells (Fig. 6 A, bottom). Because neither loss of the WRC nor disruption of its interaction to Fat2 in AbiΔWIRS egg chambers affects the localization of Fat2-GFP (Fig. S4), these data indicate that Fat2 acts upstream of the WRC to regulate the localization of the WRC, which might in turn control the formation of polarized cell protrusions.

We further analyzed possible changes in the protrusive behavior of follicle cells lacking either fat2 function or the interaction between the WRC and Fat2 in vivo. The comparison of migrating wild-type (Fig. 6 B and Video 8 A) and nonmigrating fat2 depleted follicle cells (Fig. 6 C and Video 8 B) reveals striking differences. As shown, migrating wild-type follicle cells form preferentially cellular protrusions at the cell front and at tricellular junctions, but not at the cell back (Figs. 1 E and 6 B and Video 8 A). In contrast, fat2-depleted cells show a loss of tricellular protrusions and remaining filopodial protrusions are distributed among the cell membranes pointing in opposite directions (Fig. 6 C, arrowheads; and Video 8 B). Disruption of the WRC–Fat2 interaction phenocopies fat2 (Fig. 6 D and Video 8 C). Those 20% round AbiΔWIRS mutant egg chambers (already visible at stage 6/7) show a loss of whip-like protrusions and multiple nonpolarized filopodial protrusions. For a quantitative measure of this finding, we analyzed the numbers of filopodia emerging either perpendicular or along the A-P axis over 15 min as illustrated in Fig. 6 E. We found that in Abi-WT rescued follicle cells, filopodia are preferentially formed at the cell front (Fig. 6 F). In contrast, AbiΔWIRS mutant and fat2-depleted cells show a strongly reduced bias for protrusion orientation (Fig. 6 F). Concomitantly, the AbiΔWIRS and fat2 mutant cells completely failed to migrate, whereas Abi-WT rescued follicle cells migrate at ∼0.5 µm/min, similar to wild-type cells (0.6 µm/min; Fig. 6 G). The loss of polarized cellular protrusions in fat2 und AbiΔWIRS mutant follicle cells might explain their loss of rotational movement (Video 8 D). Supporting this notion, disruption of polarized protrusions in wild-type egg chambers adhered on concavalin A–coated surfaces immediately blocks cell migration and egg chamber rotation (Video 9 A).

The finding that only 20% of the AbiΔWIRS egg chambers exhibit an egg elongation defect implies additional regulatory mechanisms that enable Fat2 to act through the WRC on actin dynamics. The ECM binding receptor phosphatase Dlar might be part of this regulatory network, because loss of dlar function also results in a moderate round egg phenotype (Bateman et al., 2001). Supporting this notion, we found a similar overlap in the localization of endogenous Dlar and Abi throughout egg chamber development (Fig. 7 A). Interestingly, we found that Abi localization also depends on Dlar (Fig. 7). In dlar mutants, the localization of Abi is not as severely affected as in fat2 mutants, but significantly reduced at the basal side (Fig. 7, compare A and B; C and D; and E and F). Compared with fat2 mutants, higher levels of endogenous Abi can be still detected at tricellular contacts (compare arrowheads in Fig. 7, B, D, and F, Abi staining). F-actin still accumulates at tricellular junctions, but actin-rich protrusions along the membrane are clearly reduced (Fig. 7, B, D, and F, F-actin, arrowheads). These effects on Abi localization and F-actin are only seen in egg chambers that are completely mutant for dlar or in large dlar mutant clones as inferred from our mosaic analysis (Fig. 8, A and B). The global alignment of contractile actin bundles in most dlar egg chamber mutants is not affected. Thus, the remaining protrusions in dlar mutants might be sufficient to drive global alignment in most cases. Consistently, loss of Dlar function results in a moderate round-egg phenotype (Fig. 8 C; Bateman et al., 2001). Remarkably, removal of one copy abi in dlar mutant background strongly enhances the round egg phenotype, indicating that Dlar and Abi functionally interact in egg elongation (Fig. 8 C). Dlar might act through Ena on Abi. Dlar directly binds and dephosphorylates Ena (Wills et al., 1999), which in turn binds the WRC to promote actin polymerization and single-cell migration (Chen et al., 2014b). Supporting this notion, a GST fusion of the Ena-binding domain of Dlar (D2 domain; Wills et al., 1999) can pull down Abi from a cell extract of Drosophila S2 cells (Fig. 8 D). These data suggest that Ena might provide an important link between the ECM, tricellular junctions, and the actin cytoskeleton in egg elongation (Fig. 8 E). Live-imaging analysis of mutant egg chambers lacking the conserved interaction between WRC and Ena indeed revealed a complete loss of egg chamber rotation and the formation of multiple nonpolarized protrusions reminiscent of fat2 or AbiΔWIRS filopodia (Video 9 B).

In summary, we identified the atypical cadherin Fat2 as a novel WRC interacting receptor and define a molecular pathway (Fig. 8 E) in which Fat2 acts through the WRC to induce the formation of polarized cellular protrusions that drive collective cell migration during Drosophila oogenesis.

What drives symmetry-breaking in collective follicle cell migration? By analogy with the conventional PCP signaling in the wing, previous studies have led to a model in which the asymmetric localization of Fat2 might be required for the polarization of the cytoskeleton (Viktorinová and Dahmann, 2013). However, recent data already revealed that egg chamber rotation begins much earlier, suggesting that the relocalization of Fat2 to the lagging edge in stage 6–8 egg chambers cannot be the instructive signal for directional follicle cell migration (Cetera et al., 2014). Our present data further support this assumption and indicate an important role of Fat2 at tricellular contacts that might serve as a signaling platform initiating the polarization of the actin cytoskeleton (Fig. 8 F). Our live-imaging analyses revealed that large actin-rich protrusions are formed at tricellular junctions where Fat2 and the WRC already accumulate at early stages of developing egg chambers. They strongly resemble the whip-like microtubule-dependent appendages called flagella found in algae or cilia on the surface of many cells. A recent study revealed that whip-like appendages are able to synchronize cell movements in the green alga Volvox solely through interactions with the fluid that surrounds them (Brumley et al., 2014). In follicle cells, whip-like protrusions might interact similarly with the ECM to synchronize directed cell migration and to drive this novel type of morphogenetic movement.

A similar rotational cell movement has been recently identified in cultured epithelial cysts (Tanner et al., 2012; Wang et al., 2013). Nonmalignant mammary gland cell lines such as MCF10A cells form in 3D matrix spherical cysts (acini), which undergo multiple rotations (Tanner et al., 2012). Remarkably, their rotational movements also correlate with the formation of a planar polarized actin cytoskeleton and the assembly of an endogenous laminin matrix (Tanner et al., 2012). Thus, rotational motions seem to be a general feature of epithelial cells during tissue morphogenesis. To analyze the protrusion dynamics that drive rotational motions during mammary acinar morphogenesis, we performed 4D live imaging of cultured MCF10A cells infected with a LifeAct-EGFP lentivirus (Fig. 9 and Video 10). As recently reported, MCF10A cells form spherical cysts and undergo multiple rotations when cultured under spatially confined conditions (Tanner et al., 2012; Fig. 9 A and Video 10 A) Interestingly, we found that F-actin strongly accumulates at junctional interfaces at the basal side of rotating acini (Fig. 9, A and B; and Video 10, B and C). Similar to migrating follicle cells, rotating MCF10A acini form long actin-rich protrusions at basal intercellular junctions (Fig. 9, A and B; and Video 10, B and C). The similar morphology of the protrusions observed in rotating MCF10A acini suggests that the molecular mechanisms driving the tissue rotation are evolutionarily conserved. It is tempting to speculate that molecules such as human Fat2 homologue (Fat1) and Ena/VASP that are known to localize to filopodial protrusions and mature intercellular epithelial junctions might be part of the conserved molecular machinery driving rotational motions in epithelia (Moeller et al., 2004).

Different from all other round-egg genes identified so far, fat2 encodes a cell-adhesion molecule of the cadherin subfamily. Fat2 might interact homophilically or might form heterodimers as previously demonstrated for rat Fat2 or mouse Fat1 and Fat4 (Nakayama et al., 2002; Badouel et al., 2015). These homophilic interactions might modulate the intercellular shear stress between follicle cells and therefore might contribute to the orientation of cell migration as recently demonstrated by stress microscopy of epithelial monolayers (Tambe et al., 2011). The large extracellular domain of Fat2 may also mediate heterophilic interactions. One possible binding partner or coreceptor of Fat2 could be the membrane receptor phosphatase Dlar that directly binds ECM components such as laminin and nidogen (O’Grady et al., 1998). Previous studies already demonstrated a reciprocal influence of Dlar and Fat2 on their localization at tricellular junctions (Viktorinová et al., 2009). Our data suggest that both pathways, Fat2-WRC and Dlar-Ena, are tightly linked and converge to the WRC. Consistently, loss of the Fat2–WRC interaction might be at least partially compensated by Dlar–Ena interactions, a pathway in which Dlar acts through Ena to recruit the WRC.

Further understanding of how Dlar and Ena functions are integrated within the Fat2-WRC regulatory pathway could provide insights into the common mechanism by which these conserved molecular machineries coordinate collective cell migration that might also drive elongation of other developing organs, including those in vertebrates.

The following strains were used: w¹¹¹⁸ (Bl 6326), act5C-Gal4, da-Gal4, GR1-Gal4 (Bl 36287), mata4-Gal4 (Bl 7063), TJ-Gal4 (provided by V. Riechmann, Heidelberg University, Mannheim, Germany), tub-Gal4, hs-Flp (Bl 42725), FRT40A scarΔ37 (Bl 8754), FRT40A tubP-Gal80 (Bl 42725), FRT80B ubiGFP (Bl 5630), FRT82B tubP-Gal80 (Bl 5135), Fat2 RNAi (Bl 40888), Dlar RNAi (Bl 43979), Ena RNAi (Bl 39034), Abi RNAi (Bl 36707), Sra-1 RNAi (Bl 38294), Wave RNAi (Bl 36121), UAS-mCD8-GFP (Bl 5136), UAS-LifeAct-EGFP (Zobel and Bogdan, 2013), fat2^103C, FRT80B fat2^58D and Fat2-GFP (provided by C. Dahmann; Viktorinová and Dahmann, 2013; Bl indicates Bloomington Stock Center). abiΔ20, FRT82B abiΔ20 were generated by imprecise excision of the EY20423 transposon located in the abi gene (Stephan et al., 2011). Full-length Abi-WT and AbiΔWIRS fragments were amplified by PCR and cloned into Gateway Entry Vectors (pENTR D-TOPO; Invitrogen). The inserts were sequenced and subcloned into pUASt-attB-rfA, pUASp-attB-rfA, pUASt-attB-rfA-EGFP (Drosophila Genomics Resource Center) by LR in vitro recombination (Invitrogen) as previously described (Chen et al., 2014a). The UAS-Abi-EGFP, UAS-Abi-WT, and UAS-Abi-ΔWIRS transgenes were generated by the germline-specific ΦC31 integration (Bischof et al., 2007) into the M{3xP3-RFP.attP’}ZH-68E landing site. pUAST-based expression vectors were injected into an early embryo that incorporates the DNA into the germline. To identify positive integration events, the injected DNA carries besides its coding region additionally a marker gene. All Drosophila strains and crosses were performed at 25°C on standard medium. RNAi crosses were performed at 29°C.

Marked clones of cells were generated by using the FLP-FRT system or the MARCM system (Lee and Luo, 2001). Cell clones for fat2 were induced by crossing a null allele FRT80B fat2^58D/TM6b to hs-Flp;;FRT80B ubi-GFP/TM6b. Marked clones lacking abi were induced by crossing a null allele hs-Flp;;FRT82B abiΔ20/TM6b (Stephan et al., 2011) to act5C-Gal4,UAS-mCD8-GFP/CyO;FRT82B tubP-Gal80/TM6b (positively marked mutant clones). Marked clones lacking wave were induced by crossing a null allele FRT40A scarΔ37/CyO^{twist-Gal4>UAS-GFP};TM2/TM6bubi^GFP to hs-Flp;FRT40A-tubP-Gal80/CyO;tub-Gal4-UAS-mCD8-GFP/TM6b. Cell clones within the germline were induced by subjecting 1-d-old adult females to a 37°C heat shock (water bath) for 2 h in three successive days. After heat shock, flies had been placed in a fresh vial with yeast. Ovaries were dissected 1–2 d after the last heat shock.

Antibody stainings were performed on ovaries dissected in Drosophila Schneider medium (Greiner), fixated for 15 min in 4% paraformaldehyde (Sigma-Aldrich) in PBS, permeabilized for 1 h in 0.1% Triton X-100 (Invitrogen) in PBS, blocked for 30 min in 3% BSA and stained with first antibody overnight at 4°C and for 2 h at RT with the second antibody (Zobel and Bogdan, 2013). The following primary antibodies were used rabbit α-AbiA8 (1:1,000; this work), guinea pig α-WAVE (1:1,000; Bogdan et al., 2005), α-Dlar (1:500; DSHB), mouse α-GFP (1:1,000; Invitrogen). Secondary antibodies were Alexa Fluor 488– or 647–conjugated anti–rabbit, anti–guinea pig, and anti–mouse (Molecular Probes), all used at 1:1,000. F-actin was stained by either Alexa Fluor phalloidin 568 or Alexa Fluor phalloidin 488 (1:100; Invitrogen). Samples were mounted in Fluoromount (SouthernBiotech).

Light microscopic images of eggs were taken at an Axioplan 2 (Carl Zeiss) using a Plan-Neofluar 10×/0.30 objective. Fluorescently labeled specimens were analyzed on an AxioExaminer.Z1 microscope equipped with an LSM 710 confocal unit (Carl Zeiss) using a Plan-Apochromat 63×/1.4 oil-immersion objective. Further image processing was performed using the Zen 2009 software (Carl Zeiss). SIM images were taken with a CellObserver SD microscope equipped with ELYRA S.1 system using a 63×/1.4 oil-immersion objective (Carl Zeiss); five grid rotations were used with a mean of two as previously described. 4D datasets were analyzed using Imaris software (Bitplane).

Live imaging of egg chambers was performed in Schneiders medium (Sigma-Aldrich) containing 0.2 mg/ml insulin (Invitrogen) and 15% FCS (Thermo Fisher Scientific) at room temperature as previously described (Prasad et al., 2007), using an inverted Observer.Z1 microscope equipped with a Yokogawa CSU-X1 spinning disk scanning unit (YOKOGAWA), an Axiocam MRm CCD camera (6.45 µm × 6.45 µm; Carl Zeiss AG) and a 488-nm laser line (LASOS). Egg chambers were imaged through a Plan-Apochromat 25×/0.8 (UV) Vis-IR oil-immersion objective, Plan-Apochromat 63×/1.4 oil-immersion objective or Plan-Neofluar 100×/1.30 oil-immersion objective (Carl Zeiss). ZEN software 2012 (Carl Zeiss) and Fiji (ImageJ) software was used for acquisition and processing of the images. Volume-rendered 4D reconstructions (gamma 1.0) were performed on the z-series (Fig. 1, E1–E4, and corresponding Video 7, A and B, were reconstructed based on 125 single sections over 15 min) was performed using Imaris software (Bitplane).

MCF-10A cells (provided by K. Matter, Institute of Ophthalmology, University College London, London, England, UK) were grown in DMEM/F12 (1:1) supplemented with 5% horse serum, 20 ng/ml EGF, 0.5 μg/ml¹ hydrocortisone, 100 ng/ml cholera toxin, 10 μg/ml insulin, 100 U/ml penicillin, and 100 U/ml streptomycin, as recently described (Debnath et al., 2003). MCF-10A cells expressing LifeAct-EGFP were generated using a lentiviral expression construct (LifeAct-EGFP in pFUGW; provided by H.-J. Schnittler, University of Münster, Münster, Germany). The LifeAct-EGFP plasmid was cotransfected with psPAX2 (virus-packaging plasmid) and pMD₂G (envelope plasmid; 4:3:1 ratio) into Hek293T cells using X-Fect Transfection reagent (Clontech), as described previously (Tuncay et al., 2015). Cellular supernatants containing virus particles were harvested and used for MCF-10A infection. After overnight incubation, the virus containing culture medium was removed and the cells were cultivated in standard growth medium.

Morphogenesis assays with MCF-10A cells were performed essentially as described elsewhere (Debnath et al., 2003). Briefly, MCF-10A cells were resuspended in assay medium (DMEM/F12 supplemented with 2% horse serum, 5 ng/ml EGF, 0.5 µg/ml hydrocortisone, 100 ng/ml cholera toxin, 10 μg/ml insulin, 100 U/ml penicillin, and 100 U/ml streptomycin). Eight-well chamber slides (Ibidi) were coated with 40 µl growth factor reduced Geltrex (Thermo Fisher Scientific) per well. After solidification of the matrix, 5 × 10³ cells in assay medium containing 2% Geltrex were seeded into each well. Acini were analyzed after 2–3 d using a confocal microscope (LSM 780; Carl Zeiss) equipped with an objective lens Plan-Apochromat 63×/1.4 oil differential interference contrast (Carl Zeiss). 4D datasets were analyzed using Imaris software (Bitplane).

MBP pull-down was performed as previously described (Chen et al., 2014a). In brief, 1 ml reactions containing 40 pmol 2MBP-tagged WRC, 400 pmol GST-dFAT2 CT, and 15 µl amylose beads in binding buffer (20 mM Hepes, 100–120 mM NaCl, 5% [wt/vol] glycerol, 1 mM EDTA, and 5 mM β-mercaptoethanol, pH 7) were mixed at 4°C for 30 min. The beads were centrifuged and washed three times with the binding buffer. Bound proteins were eluted with 0.5% (wt/vol) maltose added in the binding buffer and examined by SDS-PAGE.

The Dlar-D2 domain (amino acids 1,191–1,480) was amplified by PCR, sequenced, and cloned into the pGEX-6P1 expression vector. GST-Dlar-D2 was expressed using Escherichia coli Arctic Express cells. Lysis was performed by sonification in (1× PBS, 2 mM MgCl₂, 2 mM DTT, 10% glycerin, and Protease Inhibitor Cocktail; Roche). After centrifugation GST-Dlar-D2 proteins were immobilized on glutathione resin according to the manufacturer’s instructions (GE Healthcare). S2 cells transfected with pUAST-Abi and Act5c-Gal4 were harvested and lysed in 1 ml ice-cold lysis buffer (50 mM Tris, pH 7,4, 150 mM NaCl, 1 mM DTT, 1.5 mM MgCl₂, 4 mM EDTA, 1% Triton X-100, 10% glycerin, and Protease Inhibitor Cocktail; Roche). After centrifugation, 500 µl cytoplasmic supernatant was added to 50 µl (GST-Dlar-D2 or GST-control) glutathione resin. After incubation for 1 h at RT, unbound proteins were washed away using four times 1 ml 1× PBS. 20% of the preparation was used per lane on standard SDS-PAGE and analyzed by Western blot analyses.

To quantify the egg index, the length and width of stage 14 eggs were measured manually. The quotient of length divided by width represents the egg index. For each genotype, 50–100 eggs were quantified. Each sample was tested for normal distribution using the Pearson d’Agostino Omnibus test. For statistical analysis of the egg index, an analysis of variance (ANOVA; P < 0.001) was performed, when all samples were distributed normally. If one sample was not distributed normally, all samples were compared with the Kruskal-Wallis test (P < 0.001) using PRISM software (GraphPad). The amount of round eggs produced by a certain genotype was counted manually (between 250 and 500 eggs depending on the genotype). Upon dissection, stage 14 eggs were sorted into two groups, round and elongated eggs. The number of each fraction was counted. The amount of each fraction in respect to the total number of eggs was given in percentage terms. The samples were compared statistically using the χ² test (χ² = 0.001), which compares the frequency distribution of two samples.

Migratory speed of moving follicle cells was measured manually using Fiji software. One point of interest was set at a tricellular junction, which was followed over time. The distance between the resulting two points of interest was measured and the velocity was calculated according to distance per time (µm/min).

The number of filopodia was counted manually over a time course of 15 min. Fluorescence intensities were measured using the Fiji software. The Pearson coefficient for tricellular junctions was measured using the coloc2 plug-in for Fiji software.

Fig. S1 shows confocal images from the basal surface of abi mosaic egg chambers. Fig. S2 shows SIM images from the basal surface of progressively developing egg chambers expressing a Fat2-GFP transgene stained for endogenous Abi. Fig. S3 shows single frames of spinning disc microscopy time-lapse movies of ex vivo cultured stage 7/8 egg chambers expressing Fat2-GFP in a mosaic pattern. Fig. S4 shows SIM images from the basal surface of developing egg chambers expressing a Fat2-GFP transgene in sra-1 RNAi and AbiΔWIRS background stained for endogenous Abi. Video 1 shows ex vivo–cultured wild-type ovarioles with rotating egg chambers (stage 7, 6, and stage 4/5) expressing LifeAct-EGFP in the follicle epithelium. Video 2 shows whip-like protrusions of migrating wild-type follicle cells expressing LifeAct-EGFP (GR1-Gal4) or Utrophin-EGFP in developing egg chambers. Video 3 shows ex vivo–cultured egg chambers expressing LifeAct-EGFP in a mosaic pattern (da-Gal4) and a 4D movie of migrating wild-type follicle cells. Video 4 shows nonmigrating sra-1–depleted follicle cells (stage 7) expressing LifeAct-EGFP (GR1-Gal4). Video 5 shows two abi mutant egg chambers at different stages reexpressing an Abi-EGFP transgene (da-Gal4). Video 6 shows migrating follicle cells from egg chambers at different developmental stages expressing a Fat2-GFP transgene under the endogenous promoter. Video 7 shows migrating follicle cells expressing a Fat2-GFP transgene and LifeAct-RFP in a mosaic pattern (da-Gal4 driver). Video 8 shows migrating wild-type follicle cells (stage 6/7) expressing LifeAct-EGFP in a mosaic pattern (da-Gal4) and movies of nonmigrating follicle cells (stage 6/7) either expressing a Fat2-RNAi transgene (da-Gal4) or lacking the WIRS interaction. Video 9 shows a wild-type egg chamber adhered to a concavalin A–coated surface and a movie of nonmigrating abi mutant follicle cells lacking the Ena interaction domain (stage 6/7). Video 10 shows rotational movement and 3D presentation of LifeAct-EGFP–infected MCF10A acini.

We thank Ivana Viktorinova for helpful discussions and sharing unpublished data and Christian Klämbt and Meike Bechtold for helpful discussions and critical reading of the manuscript. We thank Hans-Joachim Schnittler for providing the lentiviral LifeAct-EGFP expression construct and Christian Dahmann for providing fat2 mutant and Fat2-GFP stocks. We thank the Bloomington Stock Center, Vienna Drosophila RNAi Center, National Institutes of Genetics–Fly for fly stocks.

This work was supported by grants from the Deutsche Forschungsgemeinschaft to S. Bogdan (Heisenberg Program) and K. Ebnet (EB 160/4-2) and the Cells-in-Motion Cluster of Excellence (EXC 1003 – CiM), Westfälische Wilhelms-Universität Münster (to S. Bogdan).

The authors declare no competing financial interests.