Excitable Rho dynamics drive cell contractions by sequentially inducing ERM protein-mediated actin-membrane attachment and actomyosin contractility

Migration of endothelial and many other cells requires spatiotemporal regulation of protrusive and contractile cytoskeletal rearrangements that drive local cell shape changes. Unexpectedly, the small GTPase Rho, a crucial regulator of cell movement, has been reported to be active in both local cell protrusions and retractions, raising the question of how Rho activity can coordinate cell migration. Here we show that Rho activity is absent in local protrusions and active during retractions. During retractions, Rho rapidly activated ezrin-radixin-moesin proteins (ERMs) to increase actin-membrane attachment, and, with a delay, non-muscle myosin II (NMII). Rho activity was excitable, with NMII acting as a slow negative feedback regulator. Strikingly, inhibition of SLK/LOK kinases, through which Rho activates ERMs, caused elongated cell morphologies, impaired Rhoinduced cell contractions, and reverted Rho-induced blebbing. Together, our study demonstrates that Rho activity drives retractions by sequentially enhancing ERM-mediated actin-membrane attachment for force transmission and NMII-dependent contractility.


Introduction
In response to external stimuli, cells can polarize for directed migration, forming distinct cytoskeletal structures specific to the cell front and back.In the absence of directional cues, many adherent cell types lose their front-back polarization, but remain motile and undergo cycles of random protrusion and retraction, often occurring in a wave-like manner [1][2][3][4] .This intrinsic cellular behavior occurs on the timescale of minutes and is also widely observed in migrating cells, both in 2D and in 3D environments [5][6][7][8][9] .The small GTPases of the Rho family (RhoGTPases), in particular the Rac, Cdc42, and Rho subfamilies (referred to as Rac, Cdc42, and Rho hereafter), regulate actin cytoskeletal dynamics in motile cells by driving protrusive and contractile cell shape changes.Across many cell types and contexts, Rac and Cdc42 have well-defined roles in cellular protrusions, where they promote actin polymerization-driven cell edge extensions [10][11][12][13][14][15] .Although widely accepted as the master regulator of cellular contractility through its effectors Rho associated kinase (ROCK) and non-muscle myosin II (NMII) 16,17 , Rho's role in regulating cell motility remains incompletely understood.Elevated Rho activity has been observed in cell retractions [18][19][20][21][22] , but also near the cell edge in protrusions and ruffles, where it showed high correlations with local outward cell edge movements 14,[23][24][25][26][27] .A plausible effector downstream of Rho that drives protrusions is the formin mDia1 14,28 .However, exogenous activation of Rho generally causes cells to contract and subcellular activation of Rho using optogenetics has not yet been observed to induce protrusions [29][30][31][32][33] .In addition to the established Rho-activated targets ROCK and mDia1, Rho can activate the ezrin-radixin-moesin family proteins (ERMs), through the kinases SLK (Ste20-like kinase) and LOK (lymphocyte-oriented kinase), which have recently emerged as Rho effectors [34][35][36][37] .SLK/LOK activate ERMs 35,38 , which, when activated, link actin to the plasma membrane, to form the actin cortex and therefore control shape and mechanical properties of cells 39 .The actin cortex is involved in cell motility in multiple ways.A contractile cortex can drive cell rear retraction or generate pressure during amoeboid cell migration 40,41 .In protrusions, the actin cortex forms a barrier that has to locally weaken or detach from the plasma membrane before protrusions can be initiated 42,43 .
We set out to determine whether Rho activity drives protrusions or retractions and to test for a possible involvement of ERM activation downstream of Rho.We visualized Rho activity along with the Rho effectors NMII or ERMs using fluorescencebased reporters in single unpolarized endothelial cells that exhibit random cycles of minute and µm-scale protrusions and retractions.We found Rho was enriched in retractions, but not in protrusions, and identified that edge-proximal Rho dynamics were pulsatile, displaying hallmark characteristics of excitability.Rho activation led to the sequential (i) SLK/LOK mediated activation of ERMs to enhance actin-membrane attachment and (ii) ROCKdependent NMII activation to generate contraction.Activated ERMs coincided with Rho in retractions, whereas NMII accumulated with a delay.The inhibition of SLK/LOK kinases resulted in elongated cell phenotypes and impaired Rho-induced cell contractility, demonstrating that both ERMs and NMII are required for contractile cell shape changes downstream of Rho.

Results
Sparsely plated HT-HUVEC exhibit, random, minute-scale protrusion-retraction cycles In the absence of directional cues, HT-HUVEC (hTERT immortalized human umbilical vein endothelial cells 31 ) exhibit random motile behavior when plated on uniform substrates at low density.Such random motility is characterized by cycles of protrusions and retractions that occur on a timescale of minutes (Fig. 1a, Video 1).We found that such behavior was particularly suitable for quantitative analysis when observed within a few hours after cell plating, when cells had completed spreading and assumed relatively homogenous and compact adhesion surfaces while showing active protrusion/retraction behavior.We quantified cell shape changes based on time-lapse sequences of cells stably expressing fluorescence-based reporters, automated cell segmentation, followed by edge velocity tracking, using established methods 14,19 .For cell edge velocity tracking, segmented cell outlines were divided into 180 equally spaced peripheral coordinate windows spanning the entire cell perimeter.Each window had an adjustable depth towards the cell interior in which fluorescencebased reporter activity could be measured (Fig. 1b).Window velocities were calculated using per frame displacements, resulting in 2D spatiotemporal edge velocity maps (Fig. 1b,c).From such maps, we identified protrusions and retractions as events exceeding defined minima of outward and inward velocity and spatiotemporal scale (Fig. 1c, Materials and Methods).This analysis revealed that over a 1h duration, randomly sampled cells (n=55) had on average 27.63 ± 10.27 (mean±SD) individual protrusion events, each with an average mean velocity of 6.39 ± 0.61µm/min, lasting for 4.21 (± 0.75) min, and spanning 8.26% (± 1.32%) of the cell perimeter.In comparison, the same cells had 20.79 ± 12.17 retraction events with a mean velocity of -5.93 ± 0.63µm/min, lasting for 5.03 (± 1.10) min and spanning 6.69% (± 1.10%) of the cell perimeter (Fig. 1d).

Spatiotemporal analysis of Rho, Rac, and Cdc42 using FRETbased activity probes in randomly motile HT-HUVEC
To enable spatiotemporal analysis of Rac, Cdc42, and Rho activities in thresholded protrusions and retractions, we first tested the suitability of existing FRET-based activity probes in HT-HU-VEC.For Rho, we tested RhoA-based RhoA2G 44 , DORA-RhoA 25 , and a RhoB-based RhoB sensor 20 .The probes were stably expressed in HT-HUVEC via lentiviral transduction and their responses to acute, pathway-selective perturbations were assessed (Materials and Methods).Activation of Rho using thrombin resulted in increased Rho activity as reported by all three probes, with the RhoB sensor showing by far the strongest response (Fig. 1e,f).RhoA and RhoB share >85% sequence identity, and functionally many GEFs, GAPs, and downstream effectors 34,45 .RhoA and RhoB differ in their C-terminal domains, resulting in distinct lipid modifications and localization patterns, with RhoA and RhoB being mainly cytoplasmic and membrane-localized, respectively 46,47 .Consistent with this, the RhoA-based probes RhoA2G and DORA-RhoA primarily localized to the cytoplasm, whereas the RhoB sensor was mainly plasma membrane-localized (Supplementary Fig. 1a).We found that local Rho activities reported by RhoA2G and DORA-RhoA appeared to be affected by the local cell geometry, when viewed by epifluorescence microscopy.The highest activities were generally enriched in the cell periphery and lowest activities in the cell center, corresponding to high and low membrane/cytoplasm ratios, respectively (Supplementary Fig. 1a (i,ii)).To assess a possible localization bias of reported Rho activities, we performed pixel-by-pixel correlations between normalized Rho activities and their localizations for each of the three probes.RhoA2G and DORA-RhoA both yielded significantly stronger anticorrelations compared to the RhoB sensor (Supplementary Fig. 1a (iii),b), indicating that the activities reported by RhoA-based probes were affected by local cell geometry.Thus, ratiometric FRET analysis based on epifluorescence imaging data insufficiently corrects for cell geometry when most of the inactive probe resides in the cytoplasm, as in the case of RhoA-based probes.To further explore the suitability of the RhoB sensor for spatiotemporal analysis of Rho activity, we tested its response to plasma membrane translocation of mCherry-FKBP-ARHGEF1(GEF) and mCherry-FKBP-ARHGAP29(GAP), through rapamycin-induced heterodimerization with plasma membrane-localized Lyn11-FRB, to acutely (in)activate Rho activity in cells (Fig. 1g,h,i).ARHGEF1 and ARHGAP29 are a Rho-specific GEF and GAP, respectively, and the RhoB sensor showed bidirectional, expected responses to plasma membrane translocation of their catalytic domains.Thus, due to the superior response to thrombin, the absence of a noticeable localization bias, and the robust responses to both activating and inactivating Rho-regulators, we considered the RhoB sensor as the preferred Rho probe for spatiotemporal analysis in HT-HUVEC and our imaging systems for subsequent analysis.
As Rac and Cdc42 reporters, we chose RaichuEV-Rac and RaichuEV-Cdc42 19,42,48,49 .They reported expected responses to pathway-stimulating signals.Increased activities occurred upon plasma membrane translocation of mCherry-FKBP-TIAM(GEF) (Fig. 1j) and mCherry-FKBP-ITSN1(GEF) (Fig. 1k) with TIAM being a Rac-specific and ITSN1 a Cdc42-specific RhoGEF.We noted that RaichuEV-Rac and RaichuEV-Cdc42 co-activated in response to the GEF translocations, consistent with positive feedback mechanisms that engage during cell protrusions 50 .RaichuEV-Rac and RaichuEV-Cdc42 were inactivated and activated upon Rho activation and inactivation, respectively (Fig. 1f,i) (with the exception for ARHGEF1-activated Rho, Fig. 1h), consistent with an antagonism between protrusive and contractile RhoGTPase activities 51 .In addition to expected bidirectional responses of all RhoGTPase probes tested, their temporal responses were on the order of seconds, further validating their suitability for spatiotemporal analysis in motile cells.
We then imaged randomly motile HT-HUVEC stably expressing RaichuEV-Rac, RaichuEV-Cdc42, RhoB sensor, DORA-RhoA, or RhoA2G or at 25s intervals to identify spatiotemporal activation patterns of Rho GTPases during protrusionretraction cycles (Fig. 2, Videos 2-6).For each probe, panels (i) show RhoGTPase activity differences between thresholded  protrusions and retractions, normalized to non-moving membrane segments in the same time period.Panels (ii) show single cell spatiotemporal maps visualizing edge velocities and RhoGTPase activities within 1.95µm from the cell edge, and cross-correlation between the two maps.Panels (iii) show cross correlations between edge velocity and RhoGTPase activities compiled from multiple cells.As expected, relative to membrane segments classified as non-moving, Rac1 and Cdc42 were elevated in protrusions and reduced in local retractions (Fig. 2a,b (i)).Cross-correlation analysis between edge velocity and Rac1 (Fig. 2a (iii)) or Cdc42 (Fig. 2b (iii)) showed strong positive correlations with lags of negative 25-50s (i.e.peak RhoGTPase activities lagging peak edge velocities), confirming previous results 14,15 .Interestingly, the RhoB sensor showed an opposite pattern of activity -uniformly inactive in protrusions, and variably but consistently activated in retractions (Fig. 2c (i)).Cross-correlation between edge velocity and RhoB sensor activity yielded a sharply negative peak lagging edge velocity by 25-50s (Fig. 2c (iii)).Similar analyses using DORA-RhoA or RhoA2G confirmed enrichment of Rho activity in retractions (Fig. 2d,e (i)).Cross-correlations between edge velocity and RhoA2G or DORA-RhoA signals were less uniform between cells, but similarly consistently negative (Fig. 2d,e (iii)), confirming that edge-proximal Rho activity is correlated with local edge retractions and anticorrelated with protrusions.Together, our analyses demonstrate that µm and minute-scale cell shape changes in randomly motile HT-HUVEC are associated with distinct RhoGTPase activity patterns, with elevated Rac and Cdc42 activities protrusions and elevated Rho activity in retractions.

Pulsatile activation of Rho during membrane retractions
We subsequently focused on Rho activity in membrane retractions, since Rac and Cdc42 have well-documented roles in driving membrane protrusions.Visually, RhoB sensor activity propagated from the cell interior towards the cell edge before retraction onset (Fig. 3a,b, Video 7).Our cross-correlation analyses between edge velocities and Rho activity yielded a negative correlation and time-lag, however, these parameters result from averaged relationships in protrusions and retractions combined.To specifically determine the kinetics of Rho activation relative to membrane retractions (but not Rho inactivation in protrusions), we analyzed RhoB sensor activity in parts of spatiotemporal edge velocity maps where protrusions transitioned into retractions.(Supplementary Fig. 2, Materials and Methods).To compare RhoB sensor activation across multiple retraction events, we aligned Rho kinetics and edge velocity time courses by setting time = 0 to the transition point from protrusion to retraction (edge velocity = 0).The resulting averaged Rho activity time-course showed a striking, pulse-like activation during retractions (Fig. 3c).RhoB sensor activity began to rise well before retraction onset, while edge velocity remained positive.Upon reaching the transition point, RhoB sensor levels were already well above basal activation levels and increased in a non-linear fashion.Surprisingly, RhoB sensor activity did not remain statically elevated throughout the retraction.Peak RhoB sensor activity 2 min after retraction onset was followed by a marked reduction and return to baseline levels, indicating the presence of an inactivation mechanism triggered before retraction completion.
Such pulse-like patterns of activity are generated by activator-inhibitor coupled excitable systems, comprised of fast auto-activation paired with delayed self-inhibition [52][53][54] .Plotting the rate of change of RhoB sensor activity as a function of RhoB sensor activity during the buildup phase (approximately 6min before retraction onset to 1min post retraction onset, shaded rectangle in Fig. 3c) revealed a linear positive correlation (Fig. 3d), i.e., Rho activity increase accelerating with increasing Rho activity.This is consistent with active Rho engaging in a positive feedback loop to further increase its own activation, a phenomenon noted in other instances of excitable Rho behavior 55,56 .In addition, we observed pulsatile behavior (Fig. 3i, Video 8) and propagating waves (Fig. 3j, Video 9) of both RhoB sensor activity and edge retraction, patterns consistent with an excitable Rho signalling network.To confirm this finding, we repeated Rho buildup analyses using the DORA-RhoA and RhoA2G probes and observed similar pulsatile activation patterns and accelerating positive feedback phases, albeit on a smaller scale (Fig. 3e-h).This was likely due to RhoA2G and DORA-RhoA's reduced dynamic range, and cell geometric bias artificially increasing Rho activity near the cell edge (Supplementary Fig. 1).Buildup profiles of Rac and Cdc42 activity in protrusion-retraction transitions showed a decrease in activity aligned with retraction onset but no pulsatile patterning (Supplementary Fig. 3).Thus, Rho activity increases before retraction onset and is pulsatile, with rapid activation during and inactivation before the retraction process is completed.

Rho, ROCK, NMII are part of an excitable system during membrane retractions
Biological excitability is characterized by positive self-feedback which generates a rapid increase in activity, followed by delayed inhibition by a downstream effector that returns activity to basal levels 53,54 .Excitable signalling modules can drive cell protrusions during chemotaxis or propagating waves of actin polymerization 19,53,[57][58][59] .Moreover, previous studies have found that Rho and its cytoskeletal effectors are excitable, resulting in pulsatile or wave-like propagation of Rho activity 55,56,60,61 .Interestingly, several studies identified that accumulation of the downstream target NMII with a delay relative to Rho can act as a negative regulator of Rho through recruitment of RhoGAPs 55,56,[60][61][62] .To test whether this was the case during membrane retractions, we first coexpressed fluorescently tagged myosin regulatory light chain (mRuby3-MLC) and RhoB sensor in HT-HUVEC to analyze myosin activation alongside edge velocity and Rho activity.The localization-based myosin activity reporter mRuby3-MLC is cytoplasmic when inactive but forms distinct puncta when incorporated into myosin filaments upon activation.Spatiotemporal analyses revealed that edge-proximal mRuby3-MLC signal was low during protrusions and early retractions (Fig. 4a, Video 10).Levels increased in intensity beginning at retraction onset, peaking several minutes after maximal Rho activity levels (Fig. 4b).Cross-correlation analysis between RhoB sensor and MLC-mRuby3 signal showed a delayed positive correlation (Supplementary Fig. 4a), significantly highest at edge depths of 5-8µm (Supplementary Fig. 4b,c), reflecting the absence of edge-proximal myosin and localization to inward moving actin filaments.The slow, delayed accumulation of myosin following Rho activation in protrusion-retraction events closely matched the   localization of negative regulators in excitable system models 53 , consistent with NMII activation downstream of Rho activity being part a negative regulator of Rho itself.
To test this further, we measured RhoB sensor activity in response to acute treatments of cells with inhibitors of ROCK (Y27632, 20µM) and myosin light chain kinase (MLCK, ML7, 20µM), i.e., kinases acting upstream of NMII activation.We were unable to use blebbistatin as a direct inhibitor of myosin motor activity, as the color of solutions containing blebbistatin interfered with FRET-based Rho activity measurements.Consistent with NMII acting as a negative regulator of Rho, both drug additions resulted in increased RhoB sensor activity, although with distinct kinetics.ML7 addition resulted in a sharp, transient increase (Fig. 4c), whereas Y27632 addition caused a slower, but persistent increase of Rho activity (Fig. 4d).Within MLCK-inhibited cells, stress fibers were maintained or increased, and RhoB sensor activity transiently increased in peripheral regions (Fig. 4e, Video 11), whereas ROCK inhibition caused sustained accumulation of RhoB sensor activity throughout the cells, particularly where stress fibers had dissolved, and except for peripheral ruffling regions (Fig. 4f, Video 12).Treating cells with both drugs resulted in a combined response reflective of the two independent pathways of MLC phosphorylation (Fig. 4g).Consistent with ROCK and MLCK preferentially activating NMII in the cell interior and the cell periphery, respectively, ROCK inhibition increased and MLCK inhibition decreased the cells' adhesion surface 63,64 (Fig. 4h).Together, both inhibitors increased Rho activity, demonstrating generally that activated NMII negatively regulates Rho.Specifically, the sustained increase in Rho activity following ROCK inhibition and stress fiber dissolution suggested that the Rho-ROCK-NMII pathway plays a critical role as a negative regulator of excitable Rho activity (Fig. 4i).

Rho activity and ERM activation are spatiotemporally coupled
Given that mRuby3-MLC signal was at its lowest near the cell edge at retraction onset and accumulated with a substantial delay relative to Rho (Fig. 4b), we questioned whether additional Rho effectors besides NMII were involved in driving early phases of membrane retractions.Rho is also known to activate ERMs through the kinases SLK and LOK 34,35,38 .Since release of cortical actin from the plasma membrane is necessary for the initiation of membrane protrusions 42,43 , we hypothesized that re-establishment of membrane-cortex attachment through activation of ERMs could be involved in retraction initiation.To test this, we first performed live-cell imaging of RhoB sensor and mRuby3-MLC-expressing HT-HUVEC, then fixed the cells and stained them using anti-pERM antibody that detects activated (phosphorylated) ERMs.Matching membrane retractions from time-lapse sequences to pERM signals revealed a striking enrichment of pERM in retractions, consistent with Rho locally activating ERMs (Fig. 5a, Video 13).We also found that in ROCK and MLCK inhibited cells, locally increased Rho activity colocalized with locally increased pERM signals, further supporting Rho activity being upstream of ERM activation (Fig. 5b).
We next sought to monitor the spatiotemporal relationship between Rho and ERM activation during membrane retractions in live cells.We generated a localization-based ERM activation reporter, composed of mRuby3-tagged ezrin and cytoplasmic miRFP680 that we expressed stoichiometrically from the same mRNA through the use of a P2A ribosomal skipping sequence 65 (Fig. 5c).Since ERMs are recruited to the plasma membrane upon activation 39 , the pixel-by-pixel ratio of ezrin-mRuby3/miRFP680 fluorescence is an indication of local ezrin activation.Analysis of cell-edge dynamics, RhoB sensor activity, and ezrin-mRuby3/miRFP680 together revealed a striking spatiotemporal correlation of Rho activity and ezrin-mRuby3/miRFP680 (Fig. 5d,e, Video 14).Activity buildup plots of both individual and compiled (Fig. 5f,g) protrusion-retraction events demonstrated no detectable time lag between Rho activation and ezrin signal in retractions.This tight spatiotemporal correlation between Rho activity and active ezrin suggested that ERMs are immediate-early effectors of Rho, which serve to enhance the force transmission between actin and the plasma membrane that is necessary during retractions.

Rho rapidly activates ERMs via SLK/LOK
To test whether SLK/LOK kinases are involved in ERM activation downstream of Rho in HT-HUVEC (Fig. 6a), we first characterized a recently developed SLK/LOK inhibitor (Cpd31) 66 .Treatment of HT-HUVEC with Cpd31 caused an almost complete loss of pERM signal, as assessed by quantitative immunofluorescence, with only some detectable pERM signal remaining in peripheral structures resembling retraction fibers (Fig. 6b).No major changes in the actin cytoskeleton of treated cells were apparent (Fig. 6b).pERM signal decreased dose-dependently (Fig. 6c,d), and occurred within 2.5min of Cpd31 addition at 5µM (Fig. 6e,f).This surprisingly rapid loss of pERM upon SLK/LOK inhibition indicated a high turnover rate of ERM phosphorylation 38 .We next used acute treatment of cells with thrombin or nocodazole to assess Rho-dependent changes in pERM signals (Fig. 6a).Thrombin and nocodazole both rapidly activate Rho: Thrombin through the GPCR PAR-1 67 ; nocodazole through release of microtubulesequestered GEF-H1, a Rho-specific GEF, from depolymerizing microtubules 68,69 .Both treatments caused rapid, concurrent increases in RhoB sensor activity (Fig. 1e, Supplementary Fig. 5a) and pERM signals (Fig. 6g,h).Strikingly, the increases in pERM signal were potently suppressed when Cpd31 was present (Fig. 6h,i).Treating cells with ROCK inhibitor (Y27632) caused no significant reduction in pERM signals, indicating that ROCK is not required for ERM activation in HT-HUVEC (Fig. 6g,h).Our results therefore demonstrate an essential role of SLK/LOK in rapid ERM activation downstream of Rho and argue for the Rho-SLK/LOK-ERM signaling axis being a highly responsive signaling module for the regulation of actin-membrane attachment and force transmission.

SLK/LOK dependent ERM activation regulates cell morphology and is required for Rho-driven cell contractions
Given the importance of Rho activity for cell shape homeostasis and cell contractions, we took advantage of the ability to selectively inhibit the Rho downstream targets ROCK/NMII (using Y27632) and SLK/LOK/ERM (using Cpd31) to identify their individual or combined roles in these processes.
For cell shape homeostasis, we plated HT-HUVEC without, or in presence of Cpd31, Y27632, or Cpd31 and Y27632 combined, fixed the cells after 2h, and stained them with the AF647-conjugated surface label wheat germ agglutinin (WGA647) (Fig. 7a).The advantage of this protocol is that cells have not yet assumed heterogenous cell morphologies prior to exposure to the inhibitors, facilitating the detection of morphological phenotypes.Importantly, for all of the conditions tested, cells attached and spread to similar adhesion areas (Fig. 7c).Y27632 treatment is known to cause elongated cell morphologies and retraction defects [70][71][72] .We found that inhibition of either ROCK or SLK/LOK had similar effects, with cells having extended cellular processes (Fig. 7a).To quantify drug-induced cell shape changes, we chose eccentricity and solidity as metrics.Eccentricity quantifies cell elongation and solidity is the ratio of cell area and the area of a cell's convex hull (cell compactness) (Fig. 7b, Materials and Methods).Both Cpd31 and Y27632 yielded significantly increased eccentricity and significantly reduced solidity (Fig. 7d,e), and combined treatment had additive effects on both metrics (Fig. 7d,e).These results indicated that the Rho effectors ROCK and SLK/LOK both contribute to cell shape homeostasis by restricting extended cellular processes.
To directly assess the role of Rho-driven ERM activation in membrane retractions, we treated fully adhered cells with nocodazole, as it activates Rho and causes a reduced adhesion surface of cells (Supplementary Fig. 5a,b).Upon nocodazole addition, cells rapidly contracted and many started blebbing (Fig. 7f,g, Video 15).Strikingly, when Cpd31 was added along with nocodazole, both the fraction of cells blebbing and cell contractions were significantly reduced, demonstrating that SLK/LOK are required for cells to exhibit a contractile response downstream of Rho (Fig. 7f,g,h,i).In samples where cells were first treated with nocodazole for 30min to induce Rho-dependent blebbing, the addition of Cpd31 increased cell spreading and a significant fraction of cells recovered from blebbing (Fig. 7f,g,h,i, Video 15).This suggested that the process of bleb retraction not only resolves blebs, but also maintains blebbing cells in a blebbing state.Intriguingly, cells treated with combined nocodazole and Cpd31 often had "runaway" protrusions resulting in cell fragments that detached from the main cell body (Fig. 7g).Importantly, Cpd31 treatment did not prevent nocodazole-induced Rho activation (Supplementary Fig. 5c).
Together, our results demonstrate that the Rho effectors ROCK and SLK/LOK are both critical for Rho-induced cell contractions.Moreover, SLK/LOK activation of ERM downstream of Rho appears to be critical to safeguard cell integrity.

Discussion
By analyzing the spatiotemporal dynamics of Rho activity in randomly motile HT-HUVEC, we found that Rho was consistently elevated in µm and minute-scale cell edge retractions and consistently absent from protrusions.That Rho was elevated in retractions was confirmed by the use of three different FRET probes and is consistent with both recent studies using FRET-based and localization-based reporters for Rho 21,22 and the generally accepted role of Rho in activating NMII-dependent cell contractions through ROCK.
Intriguingly, our spatiotemporal analysis revealed that Rho activity in retraction events displays hallmark characteristics of excitability, consistent with fast positive and slow/delayed negative inhibition.This finding is supported by the multiple instances of excitable Rho activity generating contractile pulses in non-migratory contexts 55,56,60,61 .By including MLC or the ERM protein ezrin in our spatiotemporal analyses of Rho dynamics, we found that MLC accumulated with a minute-scale delay at retracting cell edges relative to Rho activity, whereas ezrin showed high co-localization without detectable delay.In previous studies on excitable Rho dynamics, actomyosin-associated RhoGAPs have been identified as negative regulators of Rho, in line with our findings that myosin inhibition increased Rho activity 56,60,73 .Therefore, it is likely that an actomyosin-associated RhoGAP is recruited to retractions as they evolve, progressively shutting off Rho and ensuring contractions stop.
While positive feedback to Rho most likely involves the recruitment of a RhoGEF 55,60,61,73,74 , it is less clear how this occurs during retractions.The tight spatiotemporal coupling between Rho and ezrin demonstrates that ERMs are early Rho effectors during retractions and indicates they could be components of a positive feedback mechanism.Both positive and negative feedback from ERMs to Rho activity have been observed 35,75,76 .However, negative feedback through ERMs is unlikely in our case, given the absence of delay between Rho and ezrin activation.Focal adhesion dynamics may also contribute to regulating Rho during protrusion-retraction cycles.Focal adhesions have been shown to recruit RhoGEFs and GAPs in a maturation-dependent manner, with RacGEFs associated with newly formed adhesions near the cell edge and RacGAPs and RhoGEFs with mature adhesion sites deeper in the cell interior 77 .Furthermore, increased tension on focal adhesions can recruit the RhoGEFs LARG and GEF-H1 78 , meaning greater Rho-induced contractility leads to higher Rho activity.Finally, local microtubule disassembly induces retractions 79 , and the interaction between focal adhesions and microtubules can control Rho activity through the sequestration and release of GEF-H1 80 , further supporting a plausible role for GEF-H1 in Rho's positive feedback.
What are the specific roles of ERMs and NMII during Rho-dependent cell edge retractions?The initial absence of NMII in early phases of retractions, despite elevated Rho, can be explained by the absence of NMII filaments in branched lamellipodial actin networks 81 .Our results are consistent with a model in which ERMs are the main immediate effectors of Rho through SLK/LOK during early phases of cell edge retractions, by re-establishing a link between plasma membrane and peripheral actin networks (Fig. 7j).In adherent cells, these networks typically move radially inwards, due to NMII contractility and/or polymerization-driven treadmilling [82][83][84] , which in our model drags the membrane inwards once Rho-activated ERMs enable force transmission, with ERMs acting as the actin-PM clutch.Actin network remodeling occurs during early phases of edge retraction, progressively allowing for accumulation of NMII filaments 8,85 , which in turn exert contractile forces to further accelerate edge retraction.Increasingly remodeled and contractile actomyosin then recruits a putative RhoGAP, which shuts down Rho, with the observed delay, to resolve the retraction.
Rho/ROCK-dependent NMII activation is important for cell migration and inhibiting ROCK causes a tail-retraction defect [70][71][72] .Treating cells with the SLK/LOK kinase inhibitor induced morphological phenotypes similar to those observed during ROCK inhibition, with extended cellular processes that failed to retract.The model that both NMII and ERMs are required for cell contractions was further supported by our results showing that Rho-induced cell contractions were impaired in the presence of the SLK/LOK inhibitor.Furthermore, cells that were induced to bleb in response to acute Rho activation stopped blebbing and started spreading when SLK/LOK inhibitor was added.Cells treated this way had aberrant "runaway" protrusions that occasionally detached from cells, highlighting the critical importance of Rho-dependent ERM activation through SLK/LOK for the maintenance of cellular integrity.
In summary, our results show that in endothelial cells, Rho drives membrane retractions through two mechanisms that act sequentially during the cell edge retraction process: SLK/LOK-activated ERMs enhance actin-membrane attachment and force transmission, which enable ROCK-activated NMII to then effectively pull the membrane inwards, akin to ERMs acting as a clutch and NMII as a motor.Previous studies found that initiation of cell protrusions is preceded by a local reduction in membrane-proximal actin or ezrin, and that protrusions stalled when actin-membrane attachment was enhanced through synthetic activation of ezrin 42,43 .Therefore, in cells in which the Rho-SLK/LOK-ERM module exists, edge-proximal Rho activity appears to be incompatible with driving membrane protrusions.Since RhoA has been proposed to drive cell edge protrusions by activating mDia1 14 , it will be important to investigate whether edge-proximal RhoA and ERMs are spatiotemporally correlated in these contexts.
Our findings raise several important questions to be addressed in future research.Which are the RhoGEFs and RhoGAPs that mediate Rho excitability in cell type and context-dependent manner, and is Rho excitability a general phenomenon in motile cells?We found that Rho activity is self-limiting in randomly motile cells, with activated NMII being part of a slow negative feedback mechanism.If Rho activity is excitable during directed cell migration as well, this implies Rho activity is pulsatile rather than forming a stable rear-front gradient.Beyond cell motility and given accumulating evidence for actin-membrane attachment in regulating diverse cell morphogenetic processes, including membrane tension propagation 86 , we anticipate that our identification of ERMs as highly responsive Rho effectors will have far-reaching implications for our understanding of cell morphogenesis and migration.

Cell culture
HT-HUVEC, generated by stable transduction of primary HU-VEC from mixed (male, female) donors with hTERT have been described 31 .They were cultured in either EGM2 (Lonza CC-3162), Endothelial Cell Growth Medium 2 (PromoCell, C-22011) or in EndoGRO VEGF (Millipore Sigma, SCME002), supplemented with 50µg/ml Hygromycin.HT-HUVEC stably expressing reporter constructs were generated by lentiviral transduction, followed by antibiotic selection (10µg/ml blasticidin, or 0.5mg/ml G418).Cells expressing multiple fluorescence-based reporters were generated by sequential lentiviral transductions and antibiotics selections.Stably transduced cells were maintained in the presence of blasticidin and/or G418 as applicable.HEK293FT cells (not authenticated), used for lentivirus production, were grown in DMEM (Thermo Fisher Scientific 11965092), supplemented with 10% FBS (Corning 35-077-CV), 5% GlutaMAX (Thermo Fisher Scientific 35050061).All cells were grown in the absence of antimicrobial agents and the absence of mycoplasma in cell cultures was routinely verified using a PCR test
Plasmid constructs generated during this study will be available from Addgene following publication.

Cell plating, live-cell, and fixed cell microscopy
Optical 96-well glass-bottom plates (Cellvis P96-1.5H-N) were coated with bovine collagen type I, 31µg/ml in PBS, for 2-4h at 37°C.Cells were plated at densities and for durations prior to fixation or live-cell imaging as specified.For live-cell imaging, cells were overlaid with a CO2-independent live-cell imaging solution (LIS) with low autofluorescence, composed of 125mM NaCl, 5mM KCl, 1.5mM MgCl2, 1.5mM CaCl2, 10mM D-glucose, 20mM HEPES pH 7.4, 1% FBS and 5ng/ml bFGF, and plates were sealed during imaging using aluminum microplate seals (Po-larSeal, Thomas Scientific 1152A34).Cells were fixed by adding fixation solution (4% formaldehyde in PBS) at a ratio of 1:1 to culture medium or LIS (final 2% formaldehyde) and incubated for 15min at RT.Following two PBS washes, cells were either stained with WGA-AF647 (2.5µg/ml in PBS, 10min at RT), or permeabilized for immunofluorescence staining using ASBB, a permeabilization/blocking solution (10% FBS, 1% BSA, 0.1% Triton X-100, 0.01% NaN3, in PBS) for 30min.Incubation with primary antibodies, diluted in ASBB 1:400, was done either at RT for 2h or at 4°C overnight.Incubation with secondary antibodies, diluted 1:1000 in ASBB, was done at RT for 1h.Fixed cells were imaged overlaid with PBS.
Imaging data were acquired using one of the four livecell imaging systems described below, as indicated in the following table : Live-cell imaging system 1 Fig. 1f, h-k Live-cell imaging system 1.Fully automated widefield/Yokogawa spinning disc confocal fluorescence microscope system (Intelligent Imaging Innovations, 3i), built around a Nikon Ti-E stand, equipped with Nikon's Perfect Focus System, a 40x 1.3NA Plan Fluor oil objective, a 3i laser stack (405, 442, 488, 514, 561, 640nm), a broad-range fluorescence light source with integrated excitation filter wheel, (Lambda XL, Sutter), a Yokogawa CSU-W1 scanning head with dual camera port and emission filter wheels, two sCMOS cameras (Andor Zyla 4.2), enclosed by an environmental chamber (Haison), and controlled by SlideBook software (3i).Live-cell imaging system 2. Fully automated fluorescence microscope system (ImageXpress Micro XL, Molecular Devices), equipped with white light LED light source (SOLA, Lumencor), a Zyla 5.5 sCMOS camera (Andor), a 20x 0.75 Plan Apo objective (Nikon), and controlled by MetaXpress software.Live-cell imaging system 3. Fully automated widefield fluorescence microscope system (Nikon), built around a Nikon TI-E stand, equipped with Nikon's Perfect Focus System, a 10x 0.45 NA Plan Apo air, a 20x 0.75 NA Plan Apo air, and a 40x 1.3 NA Plan Fluor oil immersion objective, a liquid light guide-coupled white light LED light source (SOLA SE UV-nIR, Lumencor), excitation and emission filter wheels (Lambda 10-3, Sutter), a sCMOS camera (Prime-BSI, Photometrics), enclosed by a custom-built environmental chamber (Digital Pixel), and controlled using Nikon NIS Elements AR software.Live-cell imaging system 4. Fully automated widefield fluorescence microscope system (Nikon), built around a Nikon TI2-E stand, equipped with Nikon's Perfect Focus System, a 10x 0.45 NA Plan Apo air, a 20x 0.75 NA Plan Apo air, and a 40x 1.3 NA Plan Fluor oil immersion objective (Nikon), a liquid light guidecoupled multispectral LED light source (SpectraX, Lumencor), a dual-camera image splitter (TwinCam, Cairn) with a custom-integrated high-speed emission filter wheel (HS-1025, FLI), two sCMOS cameras (Orca Fusion-BT, Hamamatsu), a triggered device hub (NI-BB, Nikon), enclosed by a custom-built environmental chamber (Digital Pixel), and controlled using Nikon NIS Elements AR software.
Transient transfection of cDNA, synthetic Rho, Rac, Ccd42 (in)activation using rapamycin-induced heterodimerization HT-HUVEC expressing RhoB sensor, RaichuEV-Rac, or RaichuEV-Cdc42 were plated in collagen-coated wells of 96-well glass-bottom plates at 15,000 cells/well the day before the transfection.The day of the transfection, the culture medium was replaced with antibiotics-free full growth medium, 80µl per well.Then, 0.2µg of DNA (LynFRB : mCherry-FKBP constructs 5:1 w/w) and 0.25µl of Lipofectamine 2000 (Thermo Fisher Scientific, 11668019), diluted in 20µl OptiMEM (Thermo Fisher Scientific, 31985070), was added as per manufacturer's recommendation.The transfection mix was replaced after 2h with full growth medium.16-18h later, cells were overlaid with LIS and images captured using live cell imaging system 1, using a 40x 1.3 NA oil objective lens, at 0.33µm/pixel resolution (2x2 binning).FRET, CFP, mCherry channels were acquired sequentially.60 images were captured at 15s intervals, Rapamycin was added after timepoint 10 at 0.5µM final.

Image analysisbackground subtraction, cell segmentation and cell tracking
Background images for each channel were generated by imaging wells without cells and filled with LIS.Multiple images from multiple wells were averaged and subjected to circular image filtering using MATLAB functions imfilter and fspecial.To adjust raw images of cells, the median intensity of neighboring pixels from the average background image was subtracted.Cell masks were generated using histogram-based thresholding based on pixel intensity distribution of the YFP-FRET channel.To smooth the cell edge and refine the mask, Gaussian circular averaging filters were applied.To remove small debris trails attached to/touching cells, MATLAB functions imopen and strel were used to remove edge elements with a radius less than 1 pixel.Subsequent cell trajectories were created using a nearest-neighbor consecutive pairing algorithm as described previously 19 .

YFP-FRET and CFP channel alignment, FRET/CFP ratio, and fluorophore bleaching correction
Methods for YFP-FRET and CFP channel registration have been described previously 19 .FRET/CFP ratios were computed as perpixel ratios from background-subtracted, registered, and noise-filtered YFP-FRET and CFP images.By plotting FRET/CFP averaged per field of view over time from untreated samples, a bleaching correction curve was generated, displaying exponential-like decay.Assuming FRET/CFP across multiple untreated cells remained constant throughout an experiment, the FRET ratio array for each time frame was divided against the curve, normalizing it.

Image analysismapping cell-edge velocity and RhoGTPase activities
Cell edge velocities and RhoGTPase activities in peripheral coordinate windows were analyzed as previously 19 .Briefly, the cell edge of segmented cells was divided into 900 equally spaced coordinate windows.For each coordinate window, velocity vectors were formed by the dot product between a unit vector normal to the cell edge and the displacement vector from frame to frame.Coordinate windows were then binned into groups of 5, forming 180 averaged coordinate windows, each with their own velocity vector.Coordinate windows had an adjustable depth parameter, including 0.98, 2.0, 3.3, 4.9, 6.5, and 8.1µm.FRET/CFP or any expressed protein's value for each coordinate window was the average value of all pixels assigned to the window.To remove stochastic noise from edge velocity and FRET activity data, the MATLAB function ndnanfilter was applied.

Image analysisthresholding protrusions and retractions
Cell edge velocity maps were first thresholded either above 3.9µm/min, creating a map of only protrusive activity, or below -3.9µm/min to depict retractive activity.Using MATLAB's bwareaopen function, we then applied a minimum size threshold of 25 pixels to each map, leaving distinct protrusion and retraction events above the size threshold.For each event, data such as area size, pixel IDs, time range, cell edge coordinate range, average edge velocity, average FRET activity were recorded.

Cross-correlation analysis
All cross-correlation analyses were performed using MATLAB's xcorr function.Cell edge velocity and protein activation arrays were shifted to a Standard Normal  (0,1) distribution, using the Central Limit Theorem.To avoid a built-in time lag when comparing FRET/CFP to edge velocity at a given frame x, edge velocities from frame x-1 → x and x →x+1 were averaged.Then, cross-correlation was computed for each window coordinate vector of all time points, yielding 180 correlation vectors.These were averaged to form one cross-correlation vector per cell.The overall cross correlation for a given experimental condition was obtained by averaging the vectors from all cells analyzed.

Edge velocity vs. RhoGTPase activity buildup plots
Previously generated edge velocity accompanying FRET/CFP arrays from sample cells were loaded and processed using MATLAB.To avoid bias, only edge velocity maps were visualized when choosing protrusion-retraction transition events.Rectangular regions of interest undergoing protrusion retraction cycling were identified.A region of interest spanning 15 coordinate windows or 8.3% of the cell perimeter was created.Only events with > 20 frames or 8.3min of averaged positive edge velocity preceding the transition event were used.The exact transition point from protrusion to retraction was identified as the first frame x with (i) negative acceleration, (ii) positive edge velocity at frame x-1, and negative edge velocity at frame x. Between frame x-1 and x, the frame with lowest absolute-valued edge velocity was used as the transition point.The length of the rectangular ROI was 41 frames, using ± 20 frames before and after the transition frame (Supplementary Fig. 2 a,b).Edge velocity and FRET/CFP at a chosen edge depth were plotted on the same graph (Supplementary Fig. 2c).

Image analysisezrin-mRuby3/miRFP680 ratio calculation
Ezrin-mRuby3 and miRFP680 image channels were background corrected and segmented using the YFP-FRET derived cell mask.Circular averaging image filters were applied to each channel (ndanfilter and fspecial in MATLAB).Each channel was normalized to its bleaching curve.Then, the per-pixel ezrin-mRuby3/miRFP680 ratio was computed.The 1 st and 99 th percentile values of the ratio were calculated, and used as the lower and upper limits, i.e., any ratio value less than the 1st percentile was rounded up to the first percentile, and any values above the 99 th percentile was rounded down to the 99 th percentile.This was to avoid outlier extreme ratio values due to stochastic noise.

Quantitative immunofluorescence
Fixed cells in 96-well glass bottom dishes were stained with 1:400 diluted anti-pERM antibody, (1:1000 diluted AF568-conugated secondary antibody), 1:200 diluted AF488-conjugated phalloidin, and 1:10,000 diluted Hoechst.16 images were captured per well, using a 20x 0.75 NA air objective at 0.33µm/pixel resolution.Images with visible staining or imaging artifacts were discarded, the remaining analysis was fully automated.Cell nuclei were detected based on the Hoechst images using a previously described MATLAB routine 93 .The nuclei in the resulting mask were dilated by 7 pixels (2.3µm) and the resulting image regions used to determine per-cell pERM signals, using background-subtracted pERM images.

RhoGTPase activity response to rapamycin-induced Rho, Rac and Cdc42 (in)activation
Images were processed and raw FRET/CFP ratios were computed as described above.An interactive custom MATLAB routine was then used to manually identify cells co-expressing mCherry-FKBP-tagged RhoGTPase regulatory domains and Lyn-FRB, based on the presence of mCherry fluorescence.ROIs were drawn within cells identified this way and the corresponding raw FRET/CFP time-courses automatically computed.For controls, ROIs were drawn in cells without mCherry expression.All timecourses were first individually normalized to the average FRET/CFP of the timepoints prior to rapamycin addition.Then, the time-courses of mCherry-expressing cells were normalized by averaged control time-courses, before averaging per condition, across two biological replicates.

RhoGTPase activity responses to drug additions
Cells were plated 3-4h before imaging in collagen-coated glassbottom 96 well plates at a density of 2,000 cells/well.Full growth medium was replaced with LIS 1h prior to imaging.Time lapse sequences were acquired at 20x magnification, 2x2 binning, 0.65µm/pixel, at 1min intervals for 60min.Drugs diluted in LIS and LIS only (control) were added to wells after indicated times.Image data was processed as described above and a pixel-wise YFP-FRET/CFP calculation was performed.For each field of view, average Rho-FRET activity per frame was calculated and normalized (1) to its own FRET average in the frames pre-drug addition, and (2) to the mean control FRET from all trol FOVs.These normalizations scaled each FRET response to 1 during the pre-drug addition period and accounted for any increase in FRET activity due to the addition of control LIS.

Cell shape analysis
HT-HUVEC were seeded at 750 cells per well on collagen-coated 96-well glass-bottom plates in either full growth medium (control) or full growth medium supplemented with either Cpd31 (5µM), Y27632 (10µM), or both Cpd31 (5µM) and Y27632 (10µM).Cells were incubated for 2h at 37°C, before fixation with 2% formaldehyde in PBS and staining with AF647-conjugated wheat germ agglutinin (WGA, 2.5µg/ml in PBS, 10min at RT) as a membrane marker and Hoechst (1:10,000) to stain the nuclei.Images were then acquired using live-cell imaging system 4, acquiring 16 sites per well across 30-40 wells at 20x magnification with a pixel size of 0.33µm.Image processing was performed using custom MATLAB code.Objects (cells) were automatically detected and cropped from each image.To exclude cell debris and other particles, only objects with an area greater than 3,000 pixels were retained.Object properties (centroid coordinates, area, eccentricity, solidity, major axis length, minor axis length, perimeter, major axis orientation and mean intensity) were then extracted from each cropped image, as defined by MATLAB's built-in regionprops function.Solidity is defined as the ratio of the area occupied by the cell over the area of the convex hull of the cell shape.Eccentricity is defined as the ratio between the inter-focal distance and the major axis of an ellipsoid.The number of objects in the nuclear and cell masks were used to discard objects without a nucleus (i.e., large cell debris), and cases with more than one nucleus but only one object (i.e., cells touching each other and binucleate cells).

Cell area and blebbing quantification
RhoB sensor-expressing HT-HUVEC were seeded at 1,000 cells per well in collagen-coated 96-well glass-bottom plates and incubated in full growth medium for 3-4h before replacing the full growth medium by LIS.At specified times during image acquisition, LIS (control), nocodazole (15µM), both nocodazole (15µM) and Cpd31 (5µM) or Cpd31 (5µM) were administered to the corresponding wells.Image series were acquired using live-cell imaging system 3 or 4 at 1min intervals over 135min in the YFP channel using 10x magnification and 0.65µm/pixel resolution.Image analysis was performed using custom MATLAB code.The image series were registered using the hardware stage position data and the background illumination profile was corrected using images from wells left without cells.Cell segmentation was then performed using a local minima histogram-based segmentation algorithm to extract cell occupancy area per field of view per frame.Then, cells were manually counted and classified as either blebbing or spread from images taken from one frame before drug addition and at the end of the time series.

Statistical testing and reproducibility
For comparison of two groups, the non-parametric Whitney-Mann U rank sum test was used.For comparison of multiple groups, one-way ANOVA followed by the Tukey-Cramer's pairwise comparisons were performed.All data shown are from multiple biological and technical replicates, as specified in the figure legends.Statistical analyses were performed using MATLAB or Graphpad Prism.

Figure 1 .
Figure 1.Random motility of HT-HUVEC, cell edge velocity tracking, validation of FRET probes for Rho, Rac, and Cdc42 in HT-HUVEC (a) HT-HUVEC cell expressing the RhoB sensor, overlayed with temporally color-coded cell outlines to illustrate protrusion retraction cycles over 60min.Scale bar, 10µm.(Video 1).(b) Illustration of cell edge velocity analysis (Materials & Methods).(Left) Masked HT-HUVEC cell expressing the RhoB sensor.Scale bar 10 µm.(Right) close-up showing equally spaced cell edge windows (red) used for analysis and their movements (blue, yellow arrows).(c) (Left) Corresponding cell edge velocity map from (a), (top right) automatically segmented protrusions, and (bottom right) retractions, based on defined spatiotemporal parameters (Materials & Methods).(d) Spatiotemporal parameters of individual protrusions and retractionsas identified in (c).Data points are individual events, violin plot displays means ± 25th/75th percentile of events as bolded black lines from n=54 cells, from 3 biological replicates.(e) FRET probes for Rho (RhoB sensor, DORA-RhoA, RhoA2G) were stably expressed in HT-HUVEC and their responses to 1U/ml thrombin were measured over 55min in sub-confluent cultures.The RhoB sensor showed by far the strongest response.Responses were normalized to control-treated cells, means of n= number of fields of view analyzed as indicated ± 95% CI, from 3 biological replicates.(f) The responses of the RhoB sensor and FRET probes for Rac and Cdc42 (RaichuEV-Rac and RaichuEV-Cdc42) to thrombin stimulation (0.5U/ml) were assessed, similar to (e).Means of n=20 fields of view per condition ± 95% CI, from 2 biological replicates.(g) Strategy to acutely (in)activate RhoGT-Pases by plasma membrane translocation of RhoGEF and RhoGAP domains through addition of rapamycin.(h-k) HT-HUVEC expressing RhoB sensor, RaichuEV-Rac, or RaichuEV-Cdc42 were transiently contransfected with Lyn-FRB and mCherry-FKBP-RhoGEF/GAP as indicated, and RhoGTPase FRET/CFP signals were measured in response to rapamycin addition (0.5µM).Means ± 95% CI, normalized to untransfected cells, from n=number cells as indicated, from 2 biological replicates.

Figure 3 .
Figure 3. Pulsatile activation of Rho in cell edge retractions (a) (Top) HT-HUVEC, expressing the RhoB-sensor, with region of interest denoted by white box.Scale bar, 10µm.(Bottom) local cell-edge displacements over 10min of the protrusion retraction event shown in (b).(b) Time-lapse illustrating RhoB sensor activation during a protrusion retraction event during a 10min interval.Time=00:00 denotes retraction onset.Scale bar, 10µm.(Video 7).(c, e, g) Activity buildup plots displaying averaged RhoB sensor, DORA-RhoA, and RhoA2G activities, respectively, compared during protrusion-retraction transitions.Time = 0 denotes the protrusion-retraction transition.Grey shaded region shows timepoints included in the rate-of-change analysis in d,f,h.Means ± 95 CI.RhoB sensor: n=23 events, DORA-RhoA: n=33 events, RhoA2G: n=25 events, from 3 biological replicates.(d, f, h) Plot of the rate-of-change in Rho activation as a function of Rho activity from t= -7min to t = 1.5min (18 25s timepoints) from the buildup plots for RhoB sensor, DORA-RhoA, and RhoA2G.Black boxes denote timepoint means, error bars show ± 95% CI.Linear line of best fit shown by dotted blue line.(i, j) (Left) Thresholded edge velocity and (Right) RhoB sensor activity maps (window depth 1.95µm) of representative cells illustrating (i) pulsatile activity and (j) a propagating wave of RhoB sensor activity.Region of interest denoted by black arrows on both sets of maps (Videos 8, 9).