The ELMO-MBC complex and RhoGAP19D couple Rho family GTPases during mesenchymal-to-epithelial-like transitions

During mesenchymal-to-epithelial transitions (MET) and epithelial-to-mesenchymal transitions (EMT), cells change between migratory mesenchymal and cell-cell adherent epithelial states. MET and EMT processes play central roles in metazoan development and disease, and are essential for gastrulation, neural crest, heart valve formation, palatogenesis, myogenesis, tumor metastasis and wound healing (Baum et al., 2008; Le Bras et al., 2012; Nieto, 2011; Thiery et al., 2009).

In MET-related processes, opposing leading edges interact and establish a new cell-cell contact. Integrin and cadherin machineries are major regulators of MET processes and link to the actin cytoskeleton (Le Bras et al., 2012). Actin filaments, whether assembled in Arp2/3 structures or in actomyosin cables, generate cellular pushing or pulling forces (Goley and Welch, 2006; Murrell et al., 2015). Rho family GTPases regulate actin assembly pathways, which are best characterized in mammalian cells (Burridge and Wennerberg, 2004). At a leading edge, downstream of integrin signaling, Rac activation promotes branched actin network formation in the lamellipodia for membrane extension, while in the lamellum, activation of RhoA results in the formation of contractile actin bundles that buttress, anchor and flatten the cell (Burnette et al., 2014; Guilluy et al., 2011). When opposite leading edges interact, cadherins engage and actin rearranges along the nascent cell-cell contact, which coincides with a burst of Rac activity and diminished RhoA activity (Yamada and Nelson, 2007). Specific local regulation of Rho family GTPases is central to the precise control of actin-dependent processes during MET-like progression.

A large family of guanine nucleotide exchange factors (GEFs) and GTPase activating proteins (GAPs) regulate the GTP-GDP state of Rho proteins (Rossman et al., 2005; Tcherkezian and Lamarche-Vane, 2007). Ideally, unique GEFs and GAPs specify Rho protein activity within different cellular regions and at different times, yet many GEFs and GAPs have promiscuous Rho protein specificities (Goicoechea et al., 2014; McCormack et al., 2013; Tcherkezian and Lamarche-Vane, 2007). In addition, several regulators function in similar pathways that confound the notion of uniqueness, e.g. multiple Rac GEFs act at cadherin contacts (McCormack et al., 2013; Toret et al., 2014a). Last, Rac and Rho activities are often inversely correlated in cells (Burridge and Wennerberg, 2004), which complicates interpretation of loss-of-function analyses as perturbations of one pathway likely affect the other. Such challenges make it necessary to define the regulators that work together to coordinate Rac and Rho cycles during specific cellular events.

Dorsal closure during Drosophila embryogenesis bares some similarity with mammalian wound healing, but mechanistic differences between the two processes exist (Belacortu and Paricio, 2011; Harden, 2002). During dorsal closure, two opposing epithelial layers, with a leading edge enriched with filopodia and lamellipodia occur, but the underlying embryonic amnioserosa contraction and apoptosis primarily brings them together (Gorfinkiel et al., 2009; Toyama et al., 2008). A contractile actomyosin band spans the epidermal leading edge and maintains a taut cell front (Ducuing and Vincent, 2016; Pasakarnis et al., 2016; Solon et al., 2009). As the epithelial layers come together, filopodia align segments and zipper the seams (Millard and Martin, 2008). Finally, cadherin contacts sequentially form between opposite cells and seal the dorsal side (Eltsov et al., 2015). These late events are poorly understood, but the transition from a leading edge to a cadherin contact makes dorsal closure a promising model of metazoan MET-like processes.

Rho family GTPases play crucial roles during Drosophila dorsal closure (Harden et al., 1999), but the GEFs and GAPs that regulate epidermal MET-related processes are not well defined. A dorsal closure defect was originally described for myoblast city (mbc) mutants, an established Rac GEF that is in a complex with ELMO (Ced-12 – FlyBase) (Erickson et al., 1997; Nolan et al., 1998). Based on the cell migration function of the ELMO-DOCK complex, its mammalian ortholog, the ELMO-MBC complex is thought to drive dorsal epidermis migration (Harden, 2002). However, an ELMO-DOCK complex also regulates Rho GTPases transiently downstream of cadherin contact initiation in mammals (Erasmus et al., 2015; Toret et al., 2014a). With roles downstream of both integrins and cadherins, the ELMO-DOCK complex is well positioned to regulate Rho GTPases during MET-like processes. Therefore, we investigated the specific roles of the ELMO-MBC complex and a novel GAP protein in Drosophila dorsal closure.

To assess dorsal closure function of the ELMO-MBC complex, stage 14-15 embryos expressing endogenous DE-cadherin-GFP were imaged over 2 h in wild type, mbc, elmo and elmo mbc mutants. The early steps of dorsal closure were indistinguishable between wild type and mutant (Fig. 1A and Movies 1 and 2). However, at a late stage, an epidermal gap was observed in virtually all elmo and mbc mutants (Fig. 1A, arrowhead). In elmo mbc mutants, an additional enlargement of the posterior region prevented observation of a late gap (Movie 2). Upon dorsal closure completion in wild-type embryos, a DE-cadherin-labeled seam was observed along the length of the closure (Fig. 1B). In elmo and mbc mutants, a delay in sealing occurs in the final stages of dorsal closure that persisted for 15 min to over 2 h and generated the gap. This gap was surrounded by irregular-shaped epidermal cells that failed to establish DE-cadherin contacts and were separated by fragmented cellular structures that were likely amnioserosa remnants (Fig. 1B).

DE-cadherin contacts along the closure seam were imaged in more detail to analyze cadherin-specific functions. In wild type prior to cell-cell contact, DE-cadherin was strongly localized to cell contact edges near the leading edge (Fig. 1C), as previously reported (Eltsov et al., 2015). Upon cell-cell contact formation, DE-cadherin was redistributed along the new cell-cell border, and no changes in overall cell widths before and after contact formation were observed (Fig. 1C and Fig. S1). In elmo, mbc and elmo mbc mutants prior to cell-cell contact, DE-cadherin was distributed along the leading edge similar to the wild type, suggesting that the ELMO-MBC complex is not essential for this initial cadherin formation (Fig. 1C and Fig. S1). However, upon cell-cell contact formation in these mutants, new DE-cadherin contacts expanded along the new seam over a region of ∼10 cells (Fig. 1C). Thereafter, this ∼10 cell region older cell-cell contacts along the seam became irregular (Fig. 1C).

To quantify these defects, cell widths were measured for the 10 leading edge cells that preceded DE-cadherin contact formation (w1) and the 10 cells that formed new cadherin contacts (w2) (Fig. 1D). The w1 measurements were indistinguishable between all samples (Fig. 1E). Analysis of new cadherin contacts (w2) revealed an expansion in elmo, mbc and elmo mbc mutants (∼3 µm) compared with wild type (∼2 µm) (Fig. 1F). Despite this expansion, no change in the lengths (l) of the cells after cell-cell contact formation was detected, indicating that mutant cells were either larger or flatter (Fig. 1D,G). Z-stack images revealed cadherin height differences in the 10 newest contacts between wild type and mutants; wild-type cells had an average height of ∼3 µm, whereas the mutants were ∼2 µm (Fig. 1H,I). Thus, wild-type cells appeared to extend in the z-direction upon DE-cadherin contact formation, whereas mutant cells expanded along the closure axis (Fig. 1J). These data indicate that the ELMO-MBC complex functions immediately following DE-cadherin contact formation to increase the vertical height of new cell-cell contacts. Similar results with mbc, elmo and elmo mbc mutants indicate that loss-of-function of either subunit is sufficient to block activity of the complex at new DE-cadherin contacts during dorsal closure.

Actomyosin-generated tension plays a major role in regulating dorsal closure at the tissue level (Solon et al., 2009; Wells et al., 2014). At the cellular level, actomyosin tension shapes cadherin contacts (Maître and Heisenberg, 2011). To test whether differences in cellular tension between wild type and ELMO-MBC complex mutants changed cell shape, we used laser ablation of individual cell-cell contacts during dorsal closure. The initial velocity of adjacent cell border retraction is proportional to the tension of the ablated cell-cell contact, with the assumption that the viscoelastic properties around the ablation site are unchanged across samples (Hutson et al., 2003; Rauzi and Lenne, 2011). Contacts were severed either prior to contact formation (leading edge), at new cell-cell contacts (1-10 cells) or at old cell-cell contacts (>10 cells after newest cell-cell contact) (Fig. 2A). In the wild type, ablation at the leading edge of an epidermal cell resulted in the rapid retraction of DE-cadherin borders and is indicative of high tension (Fig. 2B,C and Movie 3), which in agreement with the presence of an actomyosin cable spanning the leading edge (Solon et al., 2009). Border retraction upon ablation at the leading edge of elmo and mbc mutants was indistinguishable from wild type (Fig. 2B,C and Movie 3), indicating that the ELMO-MBC complex does not regulate tension at the leading edge. Ablation of newly formed DE-cadherin contacts (1-10 cells) in wild-type, elmo or mbc mutant embryos revealed striking differences. In wild type, little or no retraction was observed (Fig. 2B,C and Movie 3), which indicates that cells were under almost no tension upon cadherin contact formation. However, in elmo or dock mutants, cell borders still retracted, albeit at a reduced rate compared with leading edge levels, indicating a persistent tension at the new cell-cell contacts (Fig. 2B,C and Movie 3). Ablation of old contacts (>10 cells) showed that retraction was restored in the wild type, though at less than leading edge levels (Fig. 2B,C and Movie 3). In contrast, in elmo or mbc mutants, there was essentially no retraction at older contacts (Fig. 2B,C and Movie 3). These data reveal that, in wild-type cells, tension at mature cell-cell contacts is restored, but this process fails in elmo mbc mutants. This may be a result of residual effects from initial cadherin contact defects or an additional ELMO-MBC complex role at mature contacts, and was not explored further. Ablation at cells directly adjacent to the newest formed cadherin contact revealed that cellular tension levels dropped rapidly upon DE-cadherin accumulation in wild-type cells (Fig. 2D). Together, these data indicate that ELMO-MBC complexes play a major role in the rapid regulation of cell-cell contact tension in response to initial cadherin engagement.

Nonmuscle myosin II and Rho1 regulate actomyosin-generated tension prior to cell-cell contact formation (Franke et al., 2005). To examine localization of myosin II at epidermal DE-cadherin contacts, mcherry-tagged myosin II regulatory light chain was imaged in conjunction with DE-cadherin-GFP in wild-type and mbc mutants. Myosin II was enriched at the leading edges in wild type, as reported previously (Franke et al., 2005). A similar localization was found at the leading edge of mbc mutants (Fig. 3A,B). Immediately upon cell-cell contact in wild-type cells, myosin II levels decreased at the newest DE-cadherin contacts, and the myosin II border progressed with new cell-cell contacts formation (Fig. 3A,B). In contrast, mbc mutants retained myosin II along new cell-cell contacts, and the myosin II signal tapered off after DE-cadherin contacts had formed (Fig. 3A,B).

Since Rho1-GTP is an activator of nonmuscle myosin II (Munjal et al., 2015), we examined Rho1 activity during dorsal closure. The anillin Rho-binding domain fused to GFP binds to Rho1-GTP and acts as a biosensor for Rho1 activity (Munjal et al., 2015). During dorsal closure, the Rho sensor formed puncta along the epidermal leading edges, which was segmentally lost upon engrailed-Gal4 expression (en2.4-GAL4) of a dominant-negative Rho1 mutant, Rho1^N19 (Fig. 3C). These results indicate that the biosensor puncta reported for active Rho1 in the epidermal leading edge.

The anillin Rho sensor was analyzed in wild type and elmo mutants to determine the levels of Rho1 activity during DE-cadherin contact formation. In both wild type and elmo mutants, the Rho sensor localized in puncta along the epidermal leading edge. However, Rho sensor accumulation decreased in wild-type cells upon DE-cadherin contact formation; in contrast, Rho sensor accumulation continued in elmo mutants after cell-cell contact formation (Fig. 3D). Quantification revealed a cumulative sharp drop in Rho sensor aggregation upon DE-cadherin contact formation in wild type (Fig. 3E), whereas the Rho sensor signal in elmo mutants, at best, tapered off gradually (Fig. 3E). These results indicate that the ELMO-MBC complex is required for the rapid inactivation of Rho1 and loss of myosin II activity at new DE-cadherin contacts, and confirm tension changes at new DE-cadherin contacts (Fig. 2).

Surprisingly, loss of ELMO-MBC complex function, a Rac GEF, resulted in aberrant Rho1 activity at new DE-cadherin contacts. Recent models suggest that GEF and GAP scaffolds are crucial to coordinate Rho family GTPase cycles (Duman et al., 2015). The original screen that identified the cadherin function of ELMO also identified the Rho GAP protein, RhoGAP19D (Toret et al., 2014b). RhoGAP19D is the Drosophila ortholog of mammalian Arhgap21/23 (Fig. S2), and Arhgap21 is demonstrated to preferentially increase the GTPase activity of RhoA and RhoC in cells (Barcellos et al., 2013; Lazarini et al., 2013). This activity makes the RhoGAP19D protein a strong candidate to coordinate Rho1 activity with the ELMO-MBC complex.

To investigate MBC and RhoGAP19D expression during dorsal closure, embryos were generated that express a UAS-LifeAct-GFP reporter with a Trojan GAL4 MiMIC insertion (Diao et al., 2015; Nagarkar-Jaiswal et al., 2015), which allowed for the expression of GAL4 under the endogenous control of MBC or RhoGAP19D. This analysis revealed a ubiquitous expression pattern for MBC and RhoGAP19D during dorsal closure, although a strong labeling of actomyosin cables and seams was present for both proteins (Fig. 4A,B).

Next, we generated a MiMIC-based GFP fusion of RhoGAP19D (Fig. 4A) that showed none of the subsequently described RhoGAP19D-related dorsal closure defects or genetic interactions with mbc mutants (data not shown). RhoGAP19D-GFP was found to have diffuse cytoplasmic puncta localization, which were weakly present at cadherin contacts (Fig. 4) and were particularly enriched at leading edges prior to the newest cadherin contact and new cadherin contacts (<10 cells after newest contact). Strikingly, in mbc mutants, a dramatic reorganization occurred, with a loss of RhoGAP19D-GFP puncta in epidermal cells and a redistribution in structures reminiscent of nuclei (Fig. 4C). In mbc mutants, a weak localization was still observed at some epidermal lateral contacts (Fig. 4C) and punctate GFP could still be detected in amnioserosa cells (Fig. 4C), suggesting that they were unaffected in mbc mutants. These results indicate that MBC function is required for proper epidermal RhoGAP19D localization and implicates the Rac GEF and Rho GAP in the regulation of MET-related Rho GTPase cycles.

A role for RhoGAP19D in dorsal closure was investigated by ectodermal expression (P{GawB}69B) of RhoGAP19D RNAi. UAS-Lifeact-RFP embryos that expressed endogenous DE-cadherin-GFP were imaged with or without RhoGAP19D depletion. Unexpectedly, RhoGAP19D-depleted embryos completed dorsal closure faster than wild-type embryos (Fig. 5A and Movie 4). This difference was apparent in the epidermis (DE-cadherin-GFP), although reduction of the amnioserosa appeared unaffected (Lifeact-RFP) (Fig. 5A,B). Analysis of the rates of epidermal leading edge closure revealed that wild-type embryos closed at a rate of ∼1 µm/min, and RhoGAP19D-depleted embryos closed at a rate of ∼1.4 µm/min (Fig. 5C). In contrast to the epidermis, the speed of amnioserosa reduction was indistinguishable between wild-type and RhoGAP19D-depleted embryos (Fig. 5D). Similar results were obtained with a truncated RhoGAP19D mutant (Fig. 4A and Fig. S3A), and a TRiP line RNAi construct (Fig. 5C,D), demonstrating that the depletion phenotype was due to a loss of RhoGAP19D function and not off-target effects. The amnioserosa cell extrusion rates over 1 h in wild-type, elmo and RhoGAP19D-depleted embryos were 11.7±1.5%, 12.3±2.5% and 10.7±3.0% (n=3 embryos each), respectively. Amnioserosa cell actomyosin pulse durations were also similar in wild-type (2.23±0.76 min), elmo mutant (2.14±0.65 min) and RhoGAP19D-depleted (1.95±0.71 min) cells (n=23 cells from three embryos each). These results are consistent with previous reports (Cormier et al., 2012; Pasakarnis et al., 2016; Solon et al., 2009; Toyama et al., 2008), and suggest that RhoGAP19D does not significantly affect aminoserosa reduction mechanisms.

We also analyzed cadherin dimensions during DE-cadherin contact formation in wild-type and RhoGAP19D-depleted cells. No significant difference was detected between wild-type and RhoGAP19D-depleted cell widths either before (Kolmogorov–Smirnov test for w1 is P=0.513) or after (Kolmogorov–Smirnov test for w2 is P=0.053) DE-cadherin contact formation (Figs 1D and 5E). However, a significant difference in new cadherin contact heights was observed (Kolmogorov–Smirnov test for h is P<0.0005). RhoGAP19D-depleted contacts showed a majority of cells that were shorter than wild type, but an additional cluster that was taller than wild type (Fig. 5E,F). These results reveal that RhoGAP19D is required for the proper heightening of DE-cadherin contacts.

Analysis of cell tensions in MET-related regions (Fig. 2A) revealed a significant difference in tension at the leading edge in wild-type and RhoGAP19D-depleted cells. In contrast to wild-type cells, RhoGAP19D-depleted cells clustered in two populations of different tension, one of which was lower than the observed wild-type tensions (Fig. 5G,H). Tension levels in wild type and mutants decreased upon new cadherin-contact formation (1-10 cell region), although not as significantly in RhoGAP19D-depleted cells (Fig. 5G,H). Last, no differences in tension were detected between wild type and RhoGAP19D RNAi at old contacts (>10 cells) (Fig. 5G,H). In summary, this analysis reveals that RhoGAP19D regulates tension at both leading edge and new cell-cell contacts.

We next tested whether there was a genetic interaction between RhoGAP19D and the ELMO-MBC complex. Embryos that ectodermally expressed RhoGAP19D RNAi in elmo mutants, and Rhogap19d mbc double mutants died prior to stage 14 (Fig. S3B and C), indicating a synthetic lethal defect occurs prior to dorsal closure. Together, these data indicate that, like the ELMO-MBC complex, RhoGAP19D functions during dorsal epidermis closure. However, despite the synthetic defect, the loss of function of the Rac GEF or Rho GAP alone yield strikingly different closure defects.

The loss-of-function phenotypes and known activities of the elmo-mbc complex and RhoGAP19D suggest a crucial role for both proteins in the regulation of the epidermal actin cytoskeleton. LifeAct-RFP expressed in the ectoderm was imaged in wild-type, elmo mutant and RhoGAP19D-depleted embryos to visualize epidermal actin cables. Cables were present as compact structures enriched in LifeAct-RFP signal at leading edges in wild-type and elmo mutant embryos (Fig. 6A). In RhoGAP19D-RNAi embryos, the LifeAct-labeled cables appeared less defined with gaps (Fig. 6A).

LifeAct-GFP expressed in engrailed stripes (en2.4-GAL4) gave epidermal-specific expression and allowed visualization of epidermal cadherin contact actin. No differences in the overall distribution of LifeAct-GFP were detected between wild-type, mbc mutant and RhoGAP19D-depleted new (1-10 cells) and old (>10 cell) contacts. For all conditions, new cell-cell contacts appeared enriched in LifeAct-GFP signal and diminished at older contacts (Fig. 6B). However, LifeAct-GFP may not effectively distinguish between Rac- and Rho-dependent actin assemblies at cadherin contacts.

Engrailed-driven UAS-LifeAct-GFP distinguished leading edge actin structures (filopodia, lamellipodia and cable) (Fig. 7A). Filopodia number quantification revealed no significant difference between wild-type, elmo or RhoGAP19D-depleted leading edges (Fig. 7B). However, minor differences occurred in filopodia dynamics (Movie 4); tracking of filopodia tips revealed that the lateral probing behavior of filopodia in mbc mutants was reduced compared with wild type, while being enhanced in RhoGAP19D-depleted cells relative to wild type (Fig. 7C). Effects on filopodia were not pursued further in this study.

Quantification of lamellipodia revealed that a few lamellipodia occurred in wild type, whereas mbc mutants rarely had any detectable lamellipodia (Fig. 7D). In contrast, RhoGAP19D-depleted cells had significantly increased numbers of lamellipodia compared with wild type (Fig. 7D). In addition, RhoGAP19D-depleted cell fronts appeared biphasic and transitioned between dominant filopodial and lamellipodial states compared with wild type and mbc mutants (Fig. 7D,E, Movie 5). In RhoGAP19D-depleted cells, when lamellipodia phases were synchronized within an engrailed positive stripe, a coordinated increase in leading edge movements was observed (Fig. 7E, Movie 5). Similar bursts in cell front progression were not detected in wild type and mbc mutants (Fig. 7E).

Rac activity is known to promote lamellipodia formation (Guilluy et al., 2011), which suggests that a transient, abnormal activation of Rac may occur in RhoGAP19D-depleted cells. We tested the effects of Rac hyperactivation on the epidermal leading edge by expressing a constitutively active Rac (UAS-Rac1^V12) in engrailed stripes. Expression of Rac1^V12 often resulted in severe disruption of epidermal leading edges (data not shown). However, engrailed-positive stripes, which maintained a well-defined leading edge, were dominated by lamellipodia and few filopodia (Fig. 7F, Movie 5). Furthermore, Rac1^V12-positive cell fronts displayed a faster progression compared with wild-type fronts, similar to lamellipodia phases of RhoGAP19D-depleted cell fronts (Fig. 7E,G). Taken together, these data imply that depletion of the putative Rho GAP results in aberrant rounds of Rac activity in leading edges (Fig. 7F).

Rac activity was directly analyzed using the Parkhurst Pak1- and Pak3-based biosensors (Verboon and Parkhurst, 2015). The Pak1 biosensor produced no detectable signal during dorsal closure (data not shown). The Pak3 biosensor resulted in faint, ubiquitous membrane localization, and 80% of Pak3 biosensor homozygous embryos (15 seams from 10 embryos) showed the formation of sparse puncta in the epidermal leading edge prior to and after cadherin contact formation (Fig. 8A,B). Pak3 biosensor puncta also appeared in the amnioserosa, with an apparent enrichment occurring during cell extrusion events (Fig. 8C), which was not pursued further. Engrailed expression of dominant-negative UAS-Rac1^N17 was consistent with a Rac dependency of the puncta in the epidermis, but the sparse nature, limited localization and 80% occurrence prevented conclusive determination. However, the ubiquitous weak Pak3 biosensor membrane localization was not affected segmentally by engrailed-driven dominant-negative Rac (data not shown). Therefore, we examined puncta formation in ELMO-MBC complex mutants. In elmo mutants, Pak3-biosensor puncta were never detected (17 seams from 20 embryos), but membrane localization persisted (Fig. 8A,B). These results suggest that the faint membrane localization was likely not specific to activated Rac, but that the Pak3 biosensor puncta accurately reported Rac activity and indicated that Rac activity in the epidermis is predominately focused before and after cadherin contact formation.

To examine Arhgap21 effects on Rac activity, the Pak3 biosensor was analyzed upon engrailed expression of RhoGAP19D-RNAi. Analysis over 20 min using the DE-cadherin signal to select epidermal-specific z-planes and exclude underlying amnioserosa biosensor signal revealed where Pak3 biosensor puncta formed along the epidermis. Control cells mainly had activity near epidermal seams (Fig. 8C). In RhoGAP19D-depleted embryos, Pak3 biosensor puncta were frequently observed at the leading edge away from the cadherin seams and these puncta occurred in segmental regions (Fig. 8C). Co-expression of mCD8.mRFP and RhoGAP19D-RNAi in engrailed stripes revealed that leading edge Rac biosensor puncta (>25 cells away from the nearest canthi) occur 90% of the time (n=52 puncta from three embryos) in engrailed-positive regions (Fig. 8D). These results indicate that Rac activity is elevated upon RhoGAP19D depletion, and consistent with the transient lamellipodia bursts of RhoGAP19D mutants (Fig. 7E).

Mammalian Arhgap21 has Rho specificity in cells and has virtually no biochemical activity on Rac (Barcellos et al., 2013; Dubois et al., 2005; Lazarini et al., 2013). Therefore, the distribution of nonmuscle myosin II and the anillin Rho sensor were analyzed to assess Rho activity in RhoGAP19D mutants. In wild-type cells, myosin II was enriched at the leading edge (Fig. 8E). In RhoGAP19D-depleted cells, myosin also localized at the leading edge, but the accumulation was less even and compact (Fig. 8E), consistent with actin cables and tensions (Figs 6A and 5H). Embryos that expressed the anillin Rho sensor ubiquitously and UAS-LifeAct-RFP in engrailed stripes were analyzed in the presence and absence of UAS-RhoGAP19D-RNAi. Little difference in overall Rho sensor levels was observed at the leading edge of engrailed-positive and -negative regions of wild-type and RhoGAP19D-depleted embryos (Fig. 8F). The lifetimes of Rho sensor puncta were analyzed and binned for all individual foci that appeared, disappeared or persisted during 5 min of image capture. This analysis revealed a population of stable Rho sensor puncta that occurred in engrailed-positive, RhoGAP19D-depleted regions (Fig. 8G,H). Together, these data indicate that RhoGAP19D plays a role in Rac, Rho1 and myosin II regulation at the leading edge of epidermal cells during dorsal closure.

As the ELMO-MBC complex inhibits Rho1 (Fig. 3) and the GAP activity of RhoGAP19D is predicted to stimulate Rho1-GTP hydrolysis (Barcellos et al., 2013; Lazarini et al., 2013), a possible explanation for the synthetic defect between the two proteins (Fig. S3) is that Rho1-GTP is hyper-stabilized when both regulators are lost. To overcome previous attempts to analyze loss of function of both Rho family GTPase regulators, RhoGAP19D RNAi was expressed in engrailed stripes in mbc mutant embryos. Compared with wild type, these embryos displayed a severe disruption of the dorsal closure process, which included altered DE-cadherin signal and highly disorganized epidermal cells (Fig. 9). Similar perturbations were observed in engrailed-positive (LifeAct-GFP) and engrailed-negative regions (Fig. 9), which suggests possible non-cell-autonomous defects. However, an unexpected early disruption of the amnioserosa layer prevented specific analysis of the epidermal leading edges.

To analyze the effects of Rho1 hyperactivation, constitutively active Rho1 (UAS-Rho1^V14) was expressed in engrailed stripes. A gradient of perturbations were observed: 54% (n=24) of embryos displayed mild dorsal closure defects with a constriction of engrailed-positive leading edges (Fig. 9), as previously described (Jacinto et al., 2002); 46% (n=24) of embryos had severe defects that included highly disorganized epidermal cells and an early collapse of amnioserosa cells (Fig. 9). These data indicate that mbc, arhgap21-depleted mutants phenocopy the most severe aspects of Rho1 hyperactivation.

To test the possibility of Rho1 hyperactivation upon loss of mbc and RhoGAP19D function a dominant negative Rho1 mutant (UAS-Rho1^N19) was co-expressed in the double mutants. Expression of Rho1^N19 resulted in embryos that displayed a far more mild dorsal closure phenotype with incomplete closures (Fig. 9), which were reminiscent of earlier Rho GTPase studies (Harden et al., 1999). This result revealed a partial suppression of mbc RhoGAP19D-depleted mutants with dominant negative Rho. The use of UAS-β4GalNAcTB, an unrelated secretory pathway processing protein, or LacZ did not rescue mbc RhoGAP19D-depleted dorsal closure defects (Fig. 9) and showed that the rescue was not a result of Gal4 dilution effects. These results suggest that both the elmo-mbc complex and RhoGAP19D likely regulate Rho1 activity through independent pathways, but the mechanism remains to be defined and the use of complete nulls is necessary to definitively rule out a dependence.

This study defines precisely dorsal closure Rac and Rho activities in time and space that occur at the epidermal leading edge and initial cadherin contacts. Moreover, it identifies roles for the atypical Rac GEF, the ELMO-MBC complex and the Rho GAP RhoGAP19D in the coordination of these Rac and Rho cycles during in vivo MET-like transitions (Fig. 10A).

ELMO-MBC complex mutants were defective in the number of epidermal leading edge lamellipodia, but only showed a late closure defect (Fig. 7). This suggests that lamellipodia are dispensable for the majority of the dorsal closure process and agrees with recent models where amnioserosa provide the bulk force for dorsal closure rather than epidermal migration (Blanchard et al., 2010; Pasakarnis et al., 2016; Wells et al., 2014). Additionally, Rho1, myosin II and tension regulation at new epidermal cadherin contacts were perturbed and cadherin contacts were flattened in ELMO-MBC complex mutants (Figs 2, 3). Similarly, an ELMO-DOCK complex drives Rac activation, Rho inactivation and actin rearrangements upon E-cadherin engagement in mammalian cells, but the actin reorganization role and consequences could not be addressed (Toret et al., 2014a). Lamellipodial ELMO-DOCK-mediated Rac activation drives membrane extension (Côté and Vuori, 2007), and an analogous role at new cadherin contacts would drive contact heightening (Fig. 10B). A membrane extension role accounts for the cadherin flattening observed at new DE-cadherin contacts in elmo mbc mutants, and may explain why initial E-cadherin contacts collapse in Elmo2-depleted MDCK cells (Toret et al., 2014a). The conserved functions reveal that the ELMO-MBC activities identified here likely generally apply to MET-like processes.

This study identifies a new crucial physical state of late dorsal closure after epidermal DE-cadherin contacts form. In this region, a loss of tension at new DE-cadherin contacts is coordinated with an ELMO-MBC-dependent decrease in Rho activity and myosin localization. The novel cadherin contact that experiences little to no tension and has major implications for force-dependent cadherin interactions, such as vinculin (Hoffman and Yap, 2015; Lecuit and Yap, 2015). The formation of this tension-free zone has a major impact on embryo development and its absence results in a lateral expansion of new contacts. As new contacts form in elmo mbc mutants, the epidermis elongates and results in a leading edge that is progressively squeezed and creates the gaps observed in elmo mbc mutants (Fig. 10A).

Depletion of RhoGAP19D resulted in embryos that complete epidermal closure faster than wild type (Fig. 10A). RhoGAP19D-depleted embryos displayed complex epidermal cell phenotypes (a fragmented actomyosin cable, bimodal leading-edge tensions, transient Rac and lamellipodia states, and cadherin height defects). The fragmented actomyosin cable can explain the binary tensions. The RhoGAP19D-depleted actomyosin cable and tension behaviors resembled Rac over Rho biosensor data, which suggests a myosin-Rac link (Fig. 10A). The increased lamellipodia protrusions are consistent with the new epidermis migration (Fig. 10A,B). In wild-type embryos, the speed of the epidermal leading edge and the reduction of the amnioserosa were equal. This suggests that these two processes are normally coupled, whereas in RhoGAP19D mutants they were decoupled and the epidermal cells migrate faster over the unaffected amnioserosa. Lateral filopodial dynamics were decreased in elmo-mbc mutants and increased in RhoGAP19D mutants, which could be indirectly due to the associated dynamic lamellipodia changes or due to unexplored links with Cdc42. A transient regulation of Rac activity at new contacts, could also account for the over-heightened cadherin contacts upon RhoGAP19D depletion (Fig. 10B). Notably, mammalian Arhgap21 depletion results in faster cell migration (Barcellos et al., 2013; Bigarella et al., 2009; Lazarini et al., 2013). Additionally, mammalian Arhgap21 has an undefined cadherin role and localizes at E-cadherin contacts with kinetics that resemble Elmo2 and Dock1 (Barcellos et al., 2013; Sousa et al., 2005; Toret et al., 2014a). In dorsal closure, RhoGAP19D has a MBC-dependent enrichment in the MET-like region (Fig. 4). Together, these results suggest that, like the ELMO-DOCK complex, Arhgap21 also has conserved roles in metazoan MET-related processes.

Mammalian Arhgap21 was first reported to activate predominately Cdc42 in vitro, but later tissue culture studies favored RhoA and RhoC activation roles over Cdc42 (Barcellos et al., 2013; Dubois et al., 2005; Lazarini et al., 2013). The in vivo phenotype (faster closure) and Rac biosensor data favor a transient Rac GAP role for RhoGAP19D. In contrast, RhoGAP19D depletion also stabilized Rho sensor foci, and supports a Rho GAP function (Fig. 10A,B). Moreover, in the epidermis, loss of function of both RhoGAP19D and the ELMO-MBC complex mirrors the severest constitutively active Rho1 expression and is suppressed by dominant-negative Rho1. This argues that both proteins function in dual pathways that inactivate Rho1 (Fig. 10B). The complex Rac and Rho regulation identified here is accounted for by transferring the inhibitory relationship between Rho and Rac (Burridge and Wennerberg, 2004) to the Rho GAP and Rac GEF (Fig. 10C). RhoGAP19D loss results in an inability to stimulate Rho1 GTPase activity directly (persistent Rho activation), and also a failure to inhibit ELMO-MBC-mediated Rac activation. Improper transient Rac activity would directly or indirectly inhibit Rho processes like actomyosin-generated tension. The ELMO-MBC complex loss would prevent Rac activity, and thus not inhibit Rho until the Rho GAP, or other secondary mechanisms compensates. This explains the tapering off of Rho, myosin, and tension levels at new cadherin contacts in elmo-mbc mutants. Loss of both ELMO-MBC and RhoGAP19D would prevent Rac activation, but also all Rho inactivation (Fig. 10A). The mechanisms that underlie RhoGAP19D-mediated regulation of the ELMO-MBC complex may be direct or indirect, but the ELMO-MBC complex recruiting its negative regulator as the contact matures is a possibility (Fig. 4C).

This study identifies a new GEF-GAP protein partnership as a major regulator of the inverse relationship between the Rac and Rho cycles during integrin-to-cadherin transitions. Curiously, mammalian Arhgap21 is an EMT protein (Barcellos et al., 2013), but how the MET-related functions are linked to cadherin contact disruption and leading edge establishment remains unclear. Other cadherin-associated GEF and GAP proteins may act during non-MET-related processes such as mature junction or other cadherin contact expansions (McCormack et al., 2013). The parallels between Drosophila and mammalian systems, despite the mechanistic differences between dorsal closure, wound healing and cell pairs that form cadherin contacts, demonstrate that dorsal closure is a powerful model for MET-like processes.

The following lines were generously shared in this study: ced-12^19f3 and mbc^D11.2 (S. M. Abmayr, Stowers Institute, MO, USA), sqh-sqh::mCherry and ubi-scra^RBD-GFP (T. Lecuit, Institut de Biologie du Développement de Marseille, France) and rhogap19^Trojan-GAL4 (H. Bellen, Baylor College of Medicine, TX, USA). MIMIC-based RhoGAP19D::GFP was generated by BestGene using pBS-KS-attB1-2-PT-SA-SD-0-EGFP-FLAsH-StrepII-TEV-3xFlag (DGRC). Drosophila obtained from the Bloomington stock center were RhoGAP19D-MiMIC (51150), shg::GFP (60584), shg::mTomato (58789) (Huang et al., 2009), Gal4-69B (1774) (Brand and Perrimon, 1993), UAS-LifeAct::GFP (58718 and 58719) (Huelsmann et al., 2013), UAS-LifeAct::RFP (58362) (Cai et al., 2014), UAS-RhoGAP19D^RNAi (6435, 6436 and 32361) (Billuart et al., 2001), en-GAL4 (8165) (Lawrence et al., 1995), UAS-Rac1^V12 (6291), UAS-Rac1^N17 (6292) (Luo et al., 1994), UAS-Rho1^V14 (8144), UAS-Rho1^N19 (7327) (Fanto et al., 2000), UAS-mCD8::mRFP (27400) (Brechbiel and Gavis, 2008), mbc^Trojan-GAL4 (66840), sqh-Pak1^RBD-GFP (56549 and 56550), sqh-Pak3^RBD-GFP (52303 and 52304), Dll-UAS-lacZ (8410) and UAS-β4GalNAcTB (9896).

All Drosophila work was carried out at 25°C. To visualize dorsal closure, embryos were collected 12-14 h after egg laying, dechorionated for 1 min in bleach, aligned and mounted on a coverslip in Halocarbon 200 oil (25073, Polysciences).

All genotypes in figures are listed in Table S1. Embryos were imaged on a Nikon Ti-E inverted microscope equipped with a Yokogawa CSU-X1 spinning disk and an EM-CCD Camera (Photometrics Evolve 512). A 20× air 40×1.25 N.A. water-immersion and 100×1.4 N.A. oil-immersion objectives were used. Image acquisition was performed using MetaMorph. For 20× or 40× images, 60 planes (1.2 or 0.2 µm, respectively) were captured that spanned the dorsal side of the embryo. For 100× images, 60 planes (0.1 µm) were captured that spanned the epidermis leading edge or dorsal closure seam. GFP and RFP channels were captured sequentially. Laser power was kept constant between experiments.

Laser ablations were performed on a 1 W average power, femtosecond, near-infrared laser equipped with a spinning disk microscope for imaging. To ablate cell-cell contacts, point ablations were carried out with an average power of 350-450 mW (sample dependent) and the duration of the exposure was 200 ms.

All images were analyzed using ImageJ (http://rsb.info.nih.gov/ij/). All movie rates are listed in the legends corresponding to Movies 1-5. Cell dimensions were measured as follows. Cell widths before cadherin contact formation (wl) were measured as the distance between terminal cadherin foci at the leading edge in the 25 cells preceding the newest cadherin contact. Cell width of the new contact (w2) was measured as the distance between the two vertices adjacent to the new contact in the 10 cells following the newest cadherin contact. Cell lengths (l) were measured from the midpoint of the two vertices adjacent to the newest cadherin contact to the maximal distance obtainable within a cell in the 10 cells following the newest cadherin contact. Cadherin contact heights (h) were measured from kymographs of the 10 newest cadherin contacts of 10 µm z-stacks. Ten to 20 images collected from eight to15 embryos were analyzed for each condition.

Amnioserosa extrusion rates were calculated by imaging early stage 15 embryos for 1 h (1 min/frame). The number of extruding cells (a shrinking, collapse and loss of a DE-cadherin-positive cells) over the course of the movie was divided by the total number of initial amnioserosa cells.

Actomyosin pulse quantifications were calculated from the time of appearance to the disappearance of a medial population of LifeAct-RFP in amnioserosa cells during mid/late stage 15. Data were collected from 30 s/frame movies.

Laser ablation analysis was carried out by measuring the distance between the vertices of the ablated cell-cell contact over the time. A linear fit for the first four points was used to plot the measured distance over time, which gave the initial velocity value.

Closure rates were calculated by measuring the change in linear distance that separates the cadherin-labeled leading edge or LifeAct-labeled aminoserosa borders during dorsal closure over 5 min intervals. Thirty to 40 measurements were pooled from three embryos for each analysis. Data were collected from 1 min/frame movies.

Filopodia and lamellipodia counts were obtained from single plane still images captured as seen in Fig. 5A. The numbers of filopodia or lammellipodia webs were counted and divided by the length of the actin cable/segment length. For filopodia tip tracking, the x,y position of the filopodia tip was recorded at 2 s intervals and plotted on an x, y projection. Tracks were oriented so that the actin cable is downwards.

For all box and whisker plots, the ends of the boxes mark the upper and lower quartiles, the central horizontal line indicates the median, and the whiskers indicate the maximum and minimum values. **P<0.05 and ***P<0.0001 (unpaired t-test).

We thank W. James Nelson, Elsa Bazellieres, Qiyan Mao and Alain Garcia De Las Bayonas for critical reading of this manuscript. We thank the TRiP at Harvard Medical School for providing transgenic RhoGAP19D-RNAi fly stock. We acknowledge the IBDM imaging facility and France-BioImaging infrastructure supported by the Agence Nationale de la Recherche.