Rho activation is apically restricted by Arhgap1 in neural crest cells and drives epithelial-to-mesenchymal transition

Epithelial-to-mesenchymal transitions (EMTs) are extremely important for tissue remodeling during embryonic development, and are central events in several pathologies, including cancer in humans (Hay, 1995; López-Novoa and Nieto, 2009; Shook and Keller, 2003; Thiery et al., 2009). EMT is a distinctive cell motility process in which an epithelial cell converts to a migratory cell, and which involves complex changes in cell morphology, cytoskeletal dynamics and cadherin-based cell adhesions (Duband et al., 1995; Hay, 1995; Shook and Keller, 2003). However, the mechanisms regulating these changes within epithelial tissues, especially in vivo, remain poorly understood, in part because few studies have examined the effects of molecular manipulation on the motile cell behaviors underlying EMT in intact systems.

Small GTPases of the Rho family play important roles in regulating cytoskeletal dynamics, cell adhesions and motility during many cellular processes (Jaffe and Hall, 2005). The GTPase Rho has known functions in cell migration, but its role in the unique process of EMT and the initiation of motility within an epithelium remains unclear (Clay and Halloran, 2011; Nieto, 2002; Thiery et al., 2009). Conflicting results of Rho function have come from previous studies, with some concluding that Rho stimulates EMT (Bakin et al., 2000; Berndt et al., 2008; Bhowmick et al., 2001; Cho and Yoo, 2007; Liu and Jessell, 1998; Masszi et al., 2003) while others conclude that it inhibits EMT (Groysman et al., 2008; Nakaya et al., 2008; Ozdamar et al., 2005; Shoval and Kalcheim, 2012). These conflicting results likely reflect the fact that Rho signaling is highly context dependent. Indeed, in several cases tumor cell invasion, an outcome of EMT, can be stimulated by Rho or its effector Rho Kinase (ROCK) (Croft et al., 2004; Peng et al., 2012; Wilkinson et al., 2005), but cells can switch between ROCK-dependent and -independent invasion depending on context (Sahai and Marshall, 2003). Furthermore, Rho can regulate different processes in distinct subcellular locations (Guilluy et al., 2011; Jaffe and Hall, 2005; Samarin and Nusrat, 2009), and is strongly influenced by extracellular cues (Lessey et al., 2012; Provenzano and Keely, 2011), which can be disrupted in in vitro EMT models. To define precise functions of Rho in a complex process such as EMT, it is crucial to determine where and when Rho is activated within the cell, how these patterns are established and how Rho mediates specific behaviors in intact systems.

The specific subcellular localization and level of Rho activity are dictated by numerous regulators. RhoGTPases can be activated by guanine nucleotide exchange factors (GEFs) and inactivated by GTPase-activating proteins (GAPs). Interestingly, the GEFs and GAPs greatly outnumber the Rho proteins. For example, more than 70 RhoGAPs have been identified in various species (Tcherkezian and Lamarche-Vane, 2007). The existence of so many GAPs is itself evidence that RhoGTPase activity and spatiotemporal distribution are tightly regulated and crucial to function. However, very little is known about which GAP family members regulate particular cellular processes and how they sculpt localized GTPase activity during cell motility. The founding member of the RhoGAP family, Arhgap1 (aka RhoGAP, p50RhoGAP, Cdc42GAP), has biochemical GAP activity towards Cdc42 and Rho (Barfod et al., 1993; Ridley et al., 1993; Zhang and Zheng, 1998), and loss-of-function studies suggest diverse functions. Arhgap1 knockout mice have increased apoptosis (Wang et al., 2005), decreased wound healing and decreased migration of fibroblasts and hematopoietic cells (Wang et al., 2006; Yang et al., 2006). In cancer cells, Arhgap1 is a potential target of both oncogenic and tumor suppressive microRNAs (Ahn et al., 2012; Ouchida et al., 2012) and is necessary for TGF-β-induced invasion (Ahn et al., 2012). Similar to the RhoGTPases themselves, their GAP modulators may regulate multiple processes and are likely influenced by cellular context. To date, specific functions for Arhgap1 in cell motility are poorly understood, and roles in EMT have not been described. Moreover, how Arhgap1 patterns localized Rho activity during motility processes is unknown.

Zebrafish neural crest cells (NCCs) are an excellent model with which to study EMT in vivo. EMT is a crucial step in NCC development and is required for migration (Le Douarin and Kalcheim, 1999). Here, we image the subcellular localization of endogenous active Rho and F-actin as NCCs undergo EMT in intact embryos. We show that Rho is activated and F-actin is enriched apically in premigratory NCCs during apical detachment, an initial step required for EMT. We find that Rho, ROCK and myosin II are required for detachment and EMT. Furthermore, we show that Arhgap1 restricts Rho activation during apical detachment. Arhgap1 knockdown results in broader Rho activation and prevents apical detachment and EMT. Together, these data reveal a previously unknown mechanism in which Arhgap1 focuses Rho activation to a specific subcellular area, which in turn promotes localized ROCK-driven actomyosin contraction that triggers specific cell motility behaviors essential for EMT.

Zebrafish (Danio rerio) embryos were obtained from wild-type (AB) matings. Embryos were raised in E3 embryo medium at 23-28.5°C and staged as described (Kimmel et al., 1995). Animal work was carried out in compliance with United States Department of Health and Human Services guidelines, and the animal use protocol was approved by the University of Wisconsin Institutional Animal Care and Use Committee.

DNA plasmids were generated using Gateway (Invitrogen) or In-Fusion (Clontech) cloning into tol2 (Kwan et al., 2007) or pCS2 vectors, respectively. The Rho biosensor (GFPrGBD) (Benink and Bement, 2005), utrophin F-actin biosensor (mCherry-UtrCH) (Burkel et al., 2007), mCherry, prenylated GFP (GFP-CAAX) or prenylated mCherry (mCherry-CAAX) were Gateway cloned behind a 4.9 kb fragment from the regulatory element of the NCC specific gene sox10 (-4.9sox10) (Wada et al., 2005). Arhgap1 was cloned behind GFP into pCS2. To mosaically label NCCs, 12.5-25 pg DNA was injected at the one-cell stage as described previously (Andersen et al., 2010) with 25 pg tol2 mRNA in a volume of 1 nl. mRNA encoding Pard3-GFP or GFP-Arhgap1 was synthesized with mMessage (Life Technologies) and 70-150 pg was injected at the one- to two-cell stage. Morpholinos were synthesized by Gene Tools (Corvalis, OR) and injected at the one-cell stage. Arhgap1MO1 (5′-AGCAAGAGTTATGGTGCCTACATGA-3′) or Arhgap1MO2 (5′-GTCACCTGTAAAGGAGACAGATTAA-3′) were injected at 500-750 μM in a 1 nl volume. For live imaging, a p53 morpholino (5′-GCGCCATTGCTTTGCAAGAATTG-3′) was co-injected at 250 μM. A standard scrambled morpholino was used as a control: 5′-CCTCTTACCTCAGTTACAATTTATA-3′.

Embryos showing intermediate levels of probes or markers were selected for imaging and mounted in 1% low-melting point agarose in E3 with 10 mM HEPES in imaging chambers as described previously (Andersen et al., 2010). Imaging was carried out on an Olympus FV1000 laser-scanning confocal microscope using a 60× oil immersion objective (NA 1.35), and began at 14 hpf when hindbrain NCCs are at varying stages of EMT. Cells were selected from rhombomeres 1-4 based on the following criteria: for apical detachment, we chose NCCs in contact with both the apical midline and the basal edge of the neuroepithelium, and acquired images every 30 seconds; for blebbing, we chose NCCs that had lost their apical attachments and begun rounding, and acquired images every 10 seconds. For ratiometric imaging, acquisition parameters were optimized to ensure GFP and mCherry were below saturation. To label apical regions, we injected pard3-GFP mRNA into one- to two-cell embryos.

Volocity software (Perkin Elmer) was used for ratiometric analysis of the Rho biosensor. Z-stacks were merged using a brightest point accumulation and noise was removed from both channels using a fine filter. Background was subtracted and the software calculated each channel threshold. GFP/mCherry intensity was created as a new channel based on a rainbow look-up table.

In blebs, we measured active Rho in a region of interest (ROI) drawn around the membrane of a retracting bleb, and an ROI-containing adjacent, non-blebbing membrane. The GFP/mCherry intensity in the ROIs was averaged over time during bleb retraction. For individual blebs, average GFP/mCherry intensity was compared between the two ROIs using paired, one-tailed t-tests. Owing to subtle differences in expression levels and acquisition parameters between embryos, we did not use GFP/mCherry for comparisons between embryos. Instead, we created a normalized measure of active Rho we termed R-value by dividing GFP/mCherry intensity in the bleb ROI by that in the non-blebbing membrane ROI. Values above one indicate higher active Rho in the bleb. For each bleb we averaged R-values over time, made pairwise comparisons between blebs using Tukey’s pairwise multiple comparisons test, and calculated an average R-value for the population of NCCs.

During apical detachment, R-values were calculated by dividing the GFP/mCherry intensity of a five-pixel diameter ROI in the NCC tail by that in an ROI at the NCC leading edge. In individual NCCs, we calculated R-values for each time point and defined three periods for analysis. ‘Before EMT’ included time points from 60 minutes to 40 minutes before apical detachment. ‘During EMT’ included 10 minutes before apical detachment until the NCC completely segregated from the neuroepithelium. ‘After EMT’ included all times after the NCC delaminated and began to migrate away from the neuroepithelium. We calculated the population average R-value during these periods and compared ‘during’ versus ‘after’ EMT using a paired, two-tailed t-test and compared ‘before’, ‘during’ and ‘after’ EMT using Tukey’s multiple comparisons test. Comparisons between R-values from NCCs in wild-type and Arhgap1 morpholino-injected embryos were made using an unpaired, two-tailed t-test.

To determine the proportion of cell area in which Rho was activated, we calculated the average maximum GFP/mCherry intensity in the cell over the time. For wild-type NCCs this was 20 minutes, beginning 10 minutes before detachment. For NCCs in Arhgap1 knockdown embryos, this was 20 minutes beginning when Rho activation was first observed. We set a threshold at one-third the average maximum, determined the proportion of the cell area with GFP/mCherry above this threshold, and averaged this value over the 20 minutes. Comparison of active Rho area between wild-type and Arhgap1 knockdown was carried out using an unpaired, two-tailed, t-test.

All kymographs were generated in Volocity by drawing a line spanning the area of interest in confocal z-projections. To quantify active Rho during apical detachment, a line was drawn on the kymograph at the cell border for the NCC tail and leading edge. GFP/mCherry along this line was normalized to the highest observed value, and averaged. In individual NCCs, these values were compared using paired, two-tailed t-tests.

In NCCs labeled with GFP-CAAX, we calculated the proportion of NCCs undergoing EMT. Only cells in the neuroepithelium at the onset of imaging were counted, and cells were considered to complete EMT after the entire cell had delaminated from the neuroepithelium. Comparisons between groups were made with Fisher’s exact tests.

Embryos were dechorionated and incubated in pharmacological inhibitors diluted in E3 buffered with 10 mM HEPES at the following concentrations: 2 μg/ml C3 (Cytoskeleton), 50 μM ROCKout (EMD Millipore) with 4% DMSO and 50 μM (-)-blebbistatin (Sigma) with 4% DMSO. For live imaging, embryos were pre-treated for 2 hours (C3) or 30 minutes (ROCKout) prior to mounting, and treated throughout imaging.

For cell morphology analysis, embryos were injected with -4.9sox10:GFP-CAAX DNA at the one-cell stage. Embryos were treated with ROCKout or blebbistatin from 14-16 hpf or C3 from 12-16 hpf, and fixed with 4% paraformaldehyde at 16 hpf. For Arhgap1 knockdown, embryos were injected with -4.9sox10:GFP-CAAX DNA and control or Arhgap1 morpholino and fixed at 16 hpf. Embryos were incubated with rabbit anti-GFP primary antibody (1:1000, Life Technologies) followed by Alexa-488 goat anti-rabbit IgG (1:1000, Life Technologies) secondary. Embryos were mounted in 70% glycerol. Images of the hindbrain were acquired on a Nikon TE3000 microscope with a 40× objective using MetaMorph software (Molecular Devices). Cell morphology in rhombomeres 1-4 was scored by experimenters blinded to treatments. Labeled cells were counted as elongated if they contacted both the apical midline and the basal surface, determined from overlays of fluorescent and DIC images. After scoring, cells from like treatments were pooled and the proportion of NCCs with an elongated morphology was determined. Comparisons between treatments were made using Fisher’s exact tests.

All statistics were calculated with Prism 5.0 (GraphPad Software). For box plots: line is median, + is mean, box covers upper and lower quartiles, and bars encompass minimum and maximum.

Premigratory NCCs span the neuroepithelium from apical to basal surfaces before undergoing a series of stereotyped EMT behaviors and then beginning directed migration away from the neuroepithelium (Fig. 1A). One of the first discernible behaviors of NCC EMT is the detachment of the cell ‘tail’ from the apical midline, which is preceded by or coincides with loss of apical cell junction components (Ahlstrom and Erickson, 2009). After detachment the trailing tail continues to retract as the cell rounds at the basal side of the neuroepithelium where it begins membrane blebbing and subsequently delaminates (Berndt et al., 2008). We previously showed that blebbing and exit from the neuroepithelium depend on ROCK and myosin II (Berndt et al., 2008). However the molecular mechanisms that control apical detachment and initiate EMT have been poorly understood.

F-actin and myosin II (actomyosin) contractile forces are important for many cell motility processes (Martin, 2010). To determine whether actomyosin is involved in apical detachment, we imaged F-actin during early stages of EMT. We mosaically labeled NCCs with membrane-targeted GFP (GFP-CAAX) and an F-actin biosensor (mCherry-UtrCH) (Berndt et al., 2008; Burkel et al., 2007) that allows visualization of individual cells. We began imaging before apical detachment, when NCCs spanned the apical-basal extent of the neuroepithelium. We used the embryonic midline, visible with DIC optics, to identify the apical surface. In all NCCs analyzed (n=7), we found transient accumulations of F-actin beneath the membrane prior to apical detachment that move forward as the tail translocates and dissipate near the end of retraction (Fig. 1B,B′, white arrowheads; Fig. 1C; supplementary material Movie 1). Following apical detachment, before the next tail retraction within the neuroepithelium, F-actin again accumulates in the tail and cycles similarly to initial detachment (Fig. 1B,B′, yellow arrowheads; Fig. 1D; supplementary material Movie 1). Because we see F-actin in tails not only during initial detachment, but also during later retractions, we are likely visualizing F-actin specifically associated with tail detachment and retraction, rather than with adherens junctions prior to initial detachment.

To test our method to identify the apical surface and thus corroborate that we are imaging apical detachment, we labeled embryos with the apical marker Pard3-GFP. We also mosaically labeled NCCs and asked whether NCC tails with midline contact, based on DIC, were indeed apical (supplementary material Fig. S1). Most NCCs classified as having midline contact showed Pard3-GFP near the midline (Table 1). Importantly, in all cases where premigratory NCCs lacked Pard3-GFP, a neighboring cell had Pard3-GFP in the same region as the NCC tail, indicating it was indeed in the apical neuroepithelium. Overall, these results show that we can image the early steps of EMT in vivo and that F-actin distribution is tightly regulated at the apical side of NCCs during detachment in a manner consistent with actomyosin contraction promoting this process.

Rho activation is often upstream of actomyosin contraction (Jaffe and Hall, 2005). To better define Rho functions in EMT, we asked where Rho is activated during EMT behaviors. We imaged active Rho with a biosensor consisting of the Rho-binding domain of Rhotekin (rGBD), fused to GFP (Benink and Bement, 2005) (GFP-rGBD), which has the advantage of labeling endogenous Rho and accurately reports active Rho localization in multiple systems (Bement et al., 2005; Benink and Bement, 2005; Godsel et al., 2010; Miller and Bement, 2009; Nakaya et al., 2008; Rohani et al., 2011; Vaughan et al., 2011; von Dassow et al., 2009; and see below). We labeled NCCs with GFP-rGBD and cytosolic mCherry as a volume marker, and performed ratiometric imaging. NCCs expressing the biosensor exhibited characteristic behaviors of EMT, exited the neuroepithelium and initiated migration; thus, the biosensor did not appear to affect normal NCC behaviors. We first tested the Rho biosensor during NCC blebbing, a behavior known to require ROCK (Berndt et al., 2008). Indeed, we found that Rho is activated as blebs approach maximal extension and remains elevated as they retract (Fig. 2B-D; supplementary material Movie 2). In most cases, the ratio of GFP-rGBD to mCherry (GFP/mCherry) in retracting bleb membrane was significantly higher than adjacent non-blebbing membrane. (supplementary material Fig. S2; Table S1). Across the NCC population, the average GFP/mCherry intensity in retracting blebs normalized to non-blebbing membrane (referred to as R-value, see Materials and methods; supplementary material Fig. S2) showed active Rho was 52% higher in blebs during retraction (Fig. 2E). Comparisons between blebs showed no differences, suggesting Rho activation in blebs is tightly controlled. This pattern of active Rho is similar to the dynamic pattern of F-actin we previously reported in blebs (Berndt et al., 2008), suggesting the result of Rho activation in NCC blebs is to promote actomyosin and bleb retraction.

To test whether Rho is activated during initial apical detachment, a behavior where F-actin is also concentrated (see Fig. 1), we imaged active Rho during early EMT. We found that Rho is activated in NCC apical tails prior to detachment, and remains activated as EMT proceeds (Fig. 3B; supplementary material Movie 3). Comparisons of GFP/mCherry intensity from kymographs during retraction (Fig. 3C,C′,D,D′; supplementary material Fig. S3) showed that in all cases active Rho was higher in the tail relative to the leading edge (supplementary material Table S2). Analysis of active Rho level in the tail normalized to the leading edge (R-value, see Materials and methods; supplementary material Fig. S3) showed that in individual NCCs, Rho activation was highest just prior to and during apical detachment, and lowest after NCCs had delaminated from the neuroepithelium (Fig. 3E,G; supplementary material Fig. S3). We analyzed active Rho levels across the NCC population during two time periods: ‘during EMT’ and ‘after EMT’ (see Materials and methods). During EMT, active Rho was 62% higher in tails relative to the leading edge (Fig. 3F). After EMT, the average active Rho signal was reduced, yet still 35% higher in the tail relative to the leading edge.

To determine whether Rho is activated specifically during EMT, we measured active Rho in premigratory NCCs at an earlier stage, from 60 to 40 minutes before apical detachment (Fig. 3G). The average active Rho level in apical tails during this period was significantly lower than during EMT, and was similar to levels after EMT (Fig. 3H). These data show the higher level of active Rho in apical tails is specific to detachment and EMT, and support the idea that Rho promotes these processes.

To confirm the specificity of the Rho biosensor and test the function of Rho in NCC apical detachment, we treated embryos with C3 exotransferase to inhibit Rho activation. C3 treatment eliminated all specific subcellular localization of biosensor signal (Fig. 4B,C). The population average of active Rho level was significantly lower than in untreated embryos (Fig. 4D). These data indicate that the biosensor provides a specific readout of active Rho in NCCs. To test whether Rho activation is required for apical detachment and EMT, we mosaically labeled NCCs with GFP-CAAX and imaged during C3 treatment. NCCs maintained an elongated morphology similar to premigratory NCCs, and most did not undergo detachment during C3 treatment (Fig. 5B; supplementary material Movie 4). Only 25.6% of labeled NCCs completed EMT (average imaging time 140.2 minutes), compared with 77.2% in untreated embryos (average imaging time 112.5 minutes) (Fig. 5D). Only NCCs that began in the neuroepithelium were scored and EMT was considered complete when the entire cell was outside of the neuroepithelium. To assess detachment, we scored cell morphology in embryos treated with C3 and fixed at 16 hpf. Rho inhibition caused a substantial increase in the proportion of NCCs that spanned the entire apicobasal width of the neuroepithelium, i.e. cells maintaining a premigratory morphology (Table 2). Together, these data show that Rho activity is necessary for apical tail detachment.

To identify molecules downstream of Rho that promote detachment and EMT, we tested the roles of ROCK and myosin II. We mosaically labeled NCCs with GFP-CAAX and treated embryos with a ROCK inhibitor (ROCKout). Most of the labeled NCCs located in the neuroepithelium at imaging onset maintained apical contact (Fig. 5C; supplementary material Movie 5) and did not undergo EMT (86.2% failed to complete EMT; Fig. 5D). During ROCKout treatment, NCCs exhibited some protrusive behavior near the midline (Fig. 5C′; supplementary material Movie 5), which resembled behaviors reported in other neuroepithelial cells (Hong and Brewster, 2006), but did not display EMT-associated blebbing at the basal surface. We further quantified the effects of ROCK or myosin II inhibition on apical detachment by scoring cell morphology in embryos fixed at 16 hpf. Inhibition of ROCK or myosin II significantly increased the proportion of NCCs spanning the neuroepithelium (Table 2), indicative of failure to initiate apical detachment. Importantly, NCCs that were outside the neuroepithelium at imaging onset during ROCKout treatment were capable of directed migration (Fig. 5C, arrowheads; supplementary material Movie 5), a behavior associated with lower active Rho levels (Fig. 3F,H). In addition, premigratory NCCs could undergo cell divisions during ROCKout treatment, although both cells remained near the apical midline throughout cytokinesis and did not detach following division (supplementary material Movie 5). Thus, this degree of ROCK inhibition disrupts only a subset of NCC behaviors, detachment and blebbing (Berndt et al., 2008), associated with the highest relative levels of active Rho, suggesting EMT is dependent on higher levels of Rho-ROCK signals than cell division or directed NCC migration. Together, these data show that ROCK and myosin II are required for apical detachment and EMT, and suggest Rho-ROCK-mediated actomyosin contraction drives these behaviors.

To determine whether Rho-ROCK signaling is required for apical F-actin accumulation, we imaged with the F-actin biosensor during C3 or ROCKout treatment. We found that F-actin still accumulated in the apical tail of NCCs after inhibition of Rho (9/10 NCCs in three embryos) or ROCK (8/8 NCCs in four embryos) (Fig. 5E,F). These results suggest that Rho signaling is not necessary for initial F-actin polymerization, but only for the ROCK-mediated actomyosin contractile forces that act on F-actin to drive detachment.

We next sought to determine what mechanisms control active Rho localization during EMT. Arhgap1 was selected as a candidate because it is expressed in the central nervous system at NCC EMT stages (Thisse and Thisse, 2004). We tested two splice-blocking morpholinos designed to knock down Arhgap1. Both morpholinos caused deletion of all or part of exon 4, which led to a reading frame shift and premature stop (supplementary material Fig. S4), and both showed similar effects on NCCs (see below). We selected the morpholino with more efficient exon exclusion (MO1, supplementary material Fig. S4) for live imaging experiments.

To test whether Arhgap1 affects the level, timing or pattern of Rho activation in NCCs, we imaged the Rho biosensor in Arhgap1 morpholino-injected embryos. We found that Rho was activated to similar levels as in wild-type embryos (R-value=1.50 in Arhgap1 knockdown versus R-value=1.62 in wild type); however, the area of Rho activation appeared broader in Arhgap1 knockdown NCCs (Fig. 6B; supplementary material Movie 6). We calculated the proportion of the total NCC area occupied by the highest active Rho signal (see Materials and methods) and found that Arhgap1 knockdown caused expansion in the area of Rho activation (Fig. 6C). To examine the effects of Arhgap1 knockdown and expanded Rho activation on apical detachment and EMT, we labeled NCCs with GFP-CAAX and performed live imaging after Arhgap1 knockdown. Most NCCs remained elongated, did not detach, and failed to undergo EMT in Arhgap1 knockdown embryos (Fig. 6D), with only 38.9% of NCCs completing EMT compared with 71.2% in control morpholino injected embryos (Fig. 6E). We scored the proportion of NCCs spanning the neuroepithelium in embryos fixed at 16 hpf, and found a significant increase in NCCs with premigratory morphology after Arhgap1 knockdown (Table 2). We observed this phenotype with two separate morpholinos (Table 2), suggesting the phenotype was not due to non-specific effects.

Interestingly, we found that the timing of Rho activation was similar between Arhgap1 knockdown and wild-type embryos. Arhgap1 knockdown NCCs showed an early phase of lower active Rho followed by higher Rho activation. On average, the increase occurred 53.1 minutes into imaging, similar to when Rho activity increased during EMT in wild-type NCCs (average 41.8 minutes into imaging). Together, these data indicate that Arhgap1 does not affect the timing of Rho activation, but is instead required to inhibit Rho activity in basolateral cell areas and thus focus Rho activity to the most apical cell region. Our finding that Rho inhibition and Arhgap1 knockdown cause similar phenotypes also suggests that the proper distribution of Rho signal is crucial, and that focusing Rho activation is necessary for apical detachment and NCC EMT.

To assess the spatial distribution of Arhgap1 protein, we expressed a GFP-Arhgap1 fusion protein in all cells. Hindbrain neuroepithelial cells showed GFP-Arhgap1 throughout the cytoplasm and in puncta resembling vesicles (Fig. 6F), consistent with a report of Arhgap1 in endosomes (Sirokmány et al., 2006). In a minority of premigratory NCCs (21.7%), GFP-Arhgap showed lower levels in the apical region (Fig. 6G), whereas most premigratory NCCs showed a more uniform distribution of GFP-Arhgap1. Thus, although it is possible that Arhgap1 is transiently reduced or trafficked differently in apical regions of NCCs, localized Rho activation is likely generated through a combination of broad inhibition by Arhgap1 and localized activation in the apical region by as yet unidentified GEFs.

Here, we provide the first live imaging of active Rho during EMT in an in vivo system. Our experiments have revealed new insight into mechanisms of Rho function in EMT. We find that high Rho levels in a discrete apical region of pre-migratory NCCs generate actomyosin contractile forces to promote apical detachment and drive EMT. Moreover, we have identified Arhgap1 as an essential regulator of Rho activity in NCC EMT.

The mechanisms by which motility is initially generated within an epithelium in the early stages of EMT have been poorly understood. Disruption or modulation of cadherin adhesions is an important step in initiating EMT (Clay and Halloran, 2011; Gumbiner, 2005; Nieto, 2011; Thiery et al., 2009; Wheelock et al., 2008). However, changes in cadherins are not always sufficient to induce EMT, indicating additional signals are required (Derksen et al., 2006; Knudsen et al., 2005; Maeda et al., 2005; Nelson et al., 2008; Park and Gumbiner, 2010). Here, we show that localized apical Rho-ROCK signaling in NCCs drives detachment of cell contacts, initiating motility within the neuroepithelium and EMT. Together with our F-actin localization results, these findings suggest actomyosin contractility acts in concert with disruption of adhesions to drive EMT. This idea is supported by a previous study showing that NCC apical tails can shear during EMT (Ahlstrom and Erickson, 2009), suggesting that active contractile forces are involved in detachment. Our data reveal a molecular basis for this force, namely Rho-ROCK-dependent contraction. Rho-ROCK forces could act in parallel to cadherin modulation or could directly affect cadherin-based adhesions. Multiple avenues of crosstalk exist between Rho GTPases and cadherins (Braga et al., 1997; Niessen et al., 2011; Samarin and Nusrat, 2009; Yap and Kovacs, 2003), and effects of Rho signaling and contractility on cell junctions can be disparate. Actomyosin forces can be necessary for junctional integrity (le Duc et al., 2010; Nelson and Chen, 2003; Warner and Longmore, 2009a; Warner and Longmore, 2009b; Yonemura et al., 2010), or can disrupt junctions (de Rooij et al., 2005; Samarin et al., 2007; Utech et al., 2005). Complex interactions between Rho, adhesions and tension likely orchestrate EMT, and may create multiple roles for Rho in NCCs and EMT. By showing where and when Rho is active within cells during EMT, and by analyzing how manipulation of Rho signaling affects motile behaviors in vivo, we have uncovered a role for Rho in promoting detachment of apical NCC tails.

We also found that Rho is activated in blebs at the basal side of the NCCs during EMT. The pattern of Rho activation is similar to dynamic F-actin in blebs (Berndt et al., 2008), consistent with Rho-generated actomyosin driving NCC bleb retraction. How does actomyosin contractility in different cell locations result in directionality of movement? Contraction at the apical side is coupled with loss of apical adhesions, which results in tail detachment and retraction (Fig. 7). By contrast, blebs contact a very different extracellular environment, including the basal lamina and the ECM of the peripheral mesenchyme, to which cells may adhere and thereby generate traction and forward movement. In primordial germ cells, blebs protrude forward and, as they retract, actomyosin coupled with adhesion pulls the cell forward (Goudarzi et al., 2012; Kardash et al., 2010). Our data are consistent with NCC blebs driving forward movement in a similar manner. Thus, although the direction of contractile force in NCC apical and basal regions may be opposing, the result is forward motility.

The specific outcome of Rho signaling in NCCs likely depends in part on the level of Rho activation. We show that Rho is activated to the highest levels in the neuroepithelium during EMT and that different levels of activation are associated with different cell behaviors. The lower levels of apical active Rho we see in premigratory NCCs may mediate a function separate from EMT, such as maintenance of epithelial adhesions. Indeed this was suggested as a function for Rho in a previous study that concluded Rho-ROCK signals suppress EMT (Groysman et al., 2008). One potential explanation for the differing findings between that study and ours is that experimental manipulation of Rho does not disrupt all Rho functions equally. Strong inhibition could block all Rho-ROCK functions, including epithelial maintenance, causing epithelial disruption that may resemble EMT. Weaker inhibition may block only the EMT behaviors requiring higher levels of active Rho, mainly affecting EMT. Several differences in experimental conditions between our study and that of Groysman et al. could affect the degree of Rho inhibition and/or experimental outcome, including specific inhibitors used (ROCKout versus Y27632), different time scales of treatments and analyses, different labeling methods (mosaic versus broad), and different experimental preparations. Moreover, distinct populations of NCCs (cranial versus trunk) and NCCs from different species (chick versus zebrafish) can use different mechanisms to complete EMT (Théveneau and Mayor, 2012), and these mechanisms could have different Rho requirements. Thus, although the results of these studies differ, each may describe true biological functions of Rho in NCCs.

The outcome of GTPase signaling in NCCs may be further influenced by the fact that multiple family members are expressed in NCCs at different stages. In chick, RhoB is expressed specifically in NCCs, whereas RhoA is expressed more broadly in the neuroepithelium (Groysman et al., 2008; Liu and Jessell, 1998). In addition, RhoV and RhoU, which do not signal through ROCK, are expressed in Xenopus and chick NCCs, and have functions in NCC induction and migration (Fort et al., 2011; Guémar et al., 2007; Notarnicola et al., 2008). In addition, Rac1 promotes chick NCC migration and Rac activation is influenced by Rho (Shoval and Kalcheim, 2012). It will be important to determine how Rho family members function together in NCC EMT and to further uncover the diverse molecular processes that may be used by NCCs in different species and populations to ultimately carry out EMT.

Interestingly, we also found that Rho is activated in trailing edges of migrating NCCs after EMT, but at lower levels than during EMT. This lower Rho activation may reflect previously described functions for Rho in suppressing inappropriate cell protrusions during directed NCC migration after EMT (Carmona-Fontaine et al., 2008; Matthews et al., 2008; Rupp and Kulesa, 2007). The level of active Rho in migrating NCCs is similar to those we observe in apical regions of premigratory NCCs. Thus, even at the same level, Rho activation may regulate multiple cellular functions depending on timing or context.

Not only is the level of active Rho important, but the specific subcellular localization is also a crucial determinant of Rho signaling outcome. In avian gastrulation EMT, Rho is activated in the basal region of epiblast cells prior to EMT by the Rho GEF Net1 (Nakaya et al., 2008). Before EMT, this basally localized Rho maintains the basement membrane, and during EMT it is downregulated, allowing basement membrane degradation. By contrast, we found that, in NCCs, apical localization of Rho activation promotes EMT. Our data fit a model (Fig. 7) in which localized Rho activation focuses the force of actomyosin contraction and drives apical detachment and EMT. We show that Arhgap1 has an essential function in NCC EMT, acting to restrict the localization of Rho activation, presumably by inactivating Rho in non-apical regions. Our finding that Arhgap1 knockdown does not inhibit EMT to the same degree as Rho inhibition suggests that dispersed Rho activation does not have as severe an effect as loss of Rho activity. Indeed, there may be other signals in apical region of NCCs that are unaffected by Arhgap1 knockdown that act to promote detachment. Little is known about how Arhgap1 functions in cell motility processes. In 3D cultures of adenocarcinoma, where Rho GTPase activation is associated with invasion, loss of Arhgap1 decreases invasion (Ahn et al., 2012). This finding was paradoxical because Arhgap1 inactivates RhoGTPases. However, this is consistent with our model of NCC EMT where Arhgap1 does not completely eliminate Rho activation, but rather maintains a balance of Rho activation and focuses signaling to specific subcellular regions. Loss of Arhgap1 also affects migration of mouse embryonic fibroblasts, although this effect is thought to be due to its GAP activity towards Cdc42 because total active Rho levels were unchanged (Yang et al., 2006). However, these studies did not examine whether Rho distribution was altered in Arhgap1 knockout cells. Through live imaging, we have revealed a new function for Arhgap1 in restricting the area of Rho activation within a cell during the crucial process of EMT. Overall, the mechanisms we define here may be used broadly to generate motility in epithelial tissues.

We are grateful to Bill Bement for the Rho and F-actin biosensors and for many helpful discussions and comments on the manuscript. We also thank Julie Canman for comments on the manuscript. We thank Hannah Vakilizadeh and Kevin Grunewald for technical assistance, and John Irwin and Kyle Swiggum for fish care.