A RhoA and Rnd3 cycle regulates actin reassembly during membrane blebbing

Significance

The plasma membrane and the underlying actin cortex show dynamic interactions. When the plasma membrane detaches from the actin cortex, the plasma membrane protrudes. The protruded membrane is called a membrane bleb and is often observed during cell migration or cytokinesis. In the present study, we determined the molecular mechanisms involved in the reassembly of the actin cortex in membrane blebs using live-cell imaging.

Abstract

The actin cytoskeleton usually lies beneath the plasma membrane. When the membrane-associated actin cytoskeleton is transiently disrupted or the intracellular pressure is increased, the plasma membrane detaches from the cortex and protrudes. Such protruded membrane regions are called blebs. However, the molecular mechanisms underlying membrane blebbing are poorly understood. This study revealed that epidermal growth factor receptor kinase substrate 8 (Eps8) and ezrin are important regulators of rapid actin reassembly for the initiation and retraction of protruded blebs. Live-cell imaging of membrane blebbing revealed that local reassembly of actin filaments occurred at Eps8- and activated ezrin-positive foci of membrane blebs. Furthermore, we found that a RhoA–ROCK–Rnd3 feedback loop determined the local reassembly sites of the actin cortex during membrane blebbing.

Actin filaments usually lie beneath the plasma membrane. When the plasma membrane detaches from actin filaments, spherical protrusions of the membrane, termed blebs, form. Several lines of recent evidence suggest membrane blebs are used for cell migration under both physiological and pathological conditions. For example, migrating primordial germ cells (PGCs) use membrane blebs to migrate in zebrafish (1). Similarly, actively migrating PGCs exhibit membrane blebbing in Drosophila melanogaster embryos (2). Dictyostelium use membrane blebbing during chemotaxis (3, 4). Thus, membrane blebbing is widely used as a driving force of motility across species (4, 5). In addition, cancer cells use a membrane blebbing-associated mode of motility in metastasis (6). Cancer cells migrate without degrading the matrix by protruding membrane blebs in 3D extracellular matrixes (5, 7). Recently, cell physical confinement, down-regulation of cell adhesion to the extracellular matrix, and up-regulation of intrinsic cortical contraction forces were identified as key conditions for the induction of membrane blebbing-associated cell migration in many cell types including invasive cancer cells (8–10).

However, the molecular mechanisms underlying membrane blebbing remain to be elucidated. Membrane blebs are initiated as a rapid protrusion of the plasma membrane, which is driven either by a change in intracellular hydrostatic pressure after local disruption of membrane–actin cortex interactions or by a local breakdown of actin filaments. Thereafter, actin filaments polymerize beneath the protruded membrane to halt bleb expansion. When actin filaments cover the protruded membrane, myosins are recruited to these filaments (11). It is unknown how actin cortex reassembly is triggered during these processes (11, 12), although there are at least two possibilities. One possibility is that actin filaments constitutively form beneath the plasma membrane and the plasma membrane stops extending when actin filaments are sufficiently reconstructed. The other possibility is that actin cortex reassembly is triggered by the activation of unidentified signaling when the cell senses a cytoskeleton-free plasma membrane region.

In the present study, we addressed this issue using live-cell imaging and revealed that the rapid recovery of actin filaments at membrane blebs is promoted by epidermal growth factor receptor kinase substrate 8 (Eps8) and ezrin, and regulated by a RhoA–Rho-associated protein kinase (ROCK)–Rnd3 feedback loop.

Results and Discussion

Membrane Blebs Retract from Multiple Sites.

We used the human colon carcinoma cell line DLD1 to observe membrane blebbing. When cultured in 2D conditions, DLD1 cells did not exhibit membrane blebbing (Fig. 1A, Left). However, DLD1 cells actively formed membrane blebs when embedded in a type I collagen gel (Fig. 1A, Right). To visualize the process of actin cortex reassembly in protruded membrane blebs, we established DLD1 cells stably expressing both a filamentous actin marker, Lifeact–RFP, and a membrane marker, the pleckstrin homology domain of phospholipase Cδ (PLCδ–PH) tagged with green fluorescent protein (GFP) (Fig. 1B).

Fig. 1.

Eps8 locally accumulates at the initial phase of membrane retraction in the membrane bleb. (A) DLD1 cells exhibit membrane blebbing when cultured in a type I collagen matrix (Right). (Scale bar, 20 μm.) (B and C) Membrane blebbing of DLD1 cells transfected with Lifeact–RFP and GFP-tagged PLCδ–PH. Timing relative to the first image is indicated in white text. Actin cortex reassembly started from multiple sites of the protruded membrane (arrowheads). (Scale bar: B, 5 μm; C, 2 μm.) (D) Localization of GFP–MRLC1 during membrane bleb expansion and retraction. MRLC1 accumulates at multiple regions of membrane blebs (arrowheads). (Scale bar, 1 μm.) (E) Localization of GFP-tagged Eps8 in membrane blebs of DLD1 cells. Eps8 accumulates in multiple foci at the protruded membrane (arrowheads). (Scale bar, 2 μm.) (F and G) Kymographs showing actin localization (red) with respect to actin cytoskeleton-related proteins (green) during bleb retraction. Bleb extension is shown on the horizontal axis, and time is shown on the vertical axis. Eps-8 (F, green) localizes to the protruded membrane before actin filaments. MRLC1 (G, green) is recruited after actin filaments. (H) Timing of arrival of Eps8 and MRLC1 relative to that of actin filaments (t = 0 s). Data are the mean ± SD.

After expansion of a cytoskeleton-free membrane, actin is progressively recruited to the membrane during the retraction phase. By repeated and careful observations, we observed that the recovery of actin filaments occurred locally from multiple independent regions of membrane blebs (Fig. 1C, arrowheads) and actin filaments grew from these sites to cover the entire blebbing membrane (Movie S1). We did not detect continuous expansion of the actin cortex from the edge of the membrane detachment site. These observations are in good agreement with those of a previous report, which showed that mDia1, an essential actin nucleator for the regrowth of actin filaments at membrane blebs, localizes to the cortex of blebs in a speckle pattern (13). We next examined the localization of myosin regulatory light chain 1 (MRLC1) tagged with GFP (GFP–MRLC1) (Fig. 1D). GFP–MRLC1 also accumulated at the initiation sites of actin cortex reassembly. These data suggest that the retraction of blebbing membranes begins from multiple regions of the protruded membrane.

Eps8 Localizes at the Initiation Sites of Actin Cortex Reassembly.

Considering that actomyosin complexes locally reassembled at multiple sites during retraction of membrane blebs, we next sought to identify molecules that are involved in the regulation of the local accumulation of actin filaments. We examined the distributions of actin cytoskeleton-related proteins during membrane blebbing. Among the proteins tested, Eps8 was recruited to the blebbing membrane and showed a unique distribution (Fig. 1E and Movie S2). Eps8 is an actin barbed-end capping protein with actin bundling activity (14, 15); however, its roles and dynamic behavior in membrane blebbing have not been well characterized. Therefore, we focused on the functional analysis of Eps8 in membrane blebbing in the following experiments. Eps8 accumulated in multiple foci at the blebbing membrane (Fig. 1E and Movie S2). Eps8-positive domains accumulated earlier than the actin cytoskeleton. The delay between the recruitment of actin and that of Eps8 was visualized on two-color kymographs (Fig. 1 F and H). On the other hand, myosin was recruited to the plasma membrane after actin cortex reassembly (Fig. 1 G and H). Thus, endogenous Eps8 marked regions where the recovery of actin filaments was initiated during the retraction of membrane blebs.

Eps8 Is Required for Continuous Membrane Blebbing.

Next, to elucidate the functional importance of Eps8 in membrane blebbing, we knocked down Eps8 expression using short hairpin RNA in DLD1 cells (Fig. 2A). Given that Eps8 was recruited to retracting membrane blebs, we hypothesized that depletion of Eps8 would lead to defects in membrane bleb retraction. Unexpectedly, Eps8-knockdown (KD) cells had a spherical morphology and did not exhibit membrane blebbing (Fig. 2B). This phenotype was reversed by exogenous expression of Eps8 (Fig. 2C and Fig. S1A). This finding suggests that Eps8 is required for the initiation of membrane blebs. As the up-regulation of intracellular pressure and generation of membrane blebs is dependent on the contractile actomyosin cortex in the cytoplasm, we next examined whether formation of the actomyosin cortex covering the entire cell periphery is impaired in Eps8-KD cells (Fig. S1B). In Eps8-KD cells, the amount of actomyosin cortex covering the entire cell periphery was greatly reduced, as judged by phalloidin staining (Fig. S1 B–D). Furthermore, the level of the phosphorylated myosin light chain was also significantly decreased in Eps8-KD cells (Fig. S1 F and G). Therefore, cytoplasmic actomyosin cortex depletion in the cytoplasm may decrease the contractile activity required for the generation of membrane blebs in Eps8 KD cells.

Fig. 2.

The end-capping and actin-bundling activities of Eps8 are required for continuous membrane blebbing. (A) Expression of Eps8 is greatly reduced in Eps8-KD DLD1 cells. (B) Eps8-KD DLD1 cells are spherical and do not exhibit membrane blebbing when cultured in a type I collagen matrix. (Scale bar, 10 μm.) (C) Exogenous expression of GFP-tagged mouse Eps8 restores membrane blebbing (arrow) in Eps8-KD DLD1 cells. (Scale bar, 10 μm.) (D) Schematic drawings of mutant Eps8 constructs. The number of amino acid residues of Eps8 is shown. (E) Total cell lysates of DLD1 cells expressing each construct separated by SDS/PAGE and immunoblotted with an anti-GFP mAb. (F) The percentages of GFP-positive cells showing membrane blebbing relative to the total number of GFP-positive cells in a given field were calculated. In each experiment, the total cell number was 100 (n  =  3). Data are the mean ± SD. *P < 0.05 (Student's t test). (G) Localization of GFP-tagged mutant Eps8 in Eps8-KD DLD1 cells. Expression of the GFP-tagged Eps8 mutant lacking the proline-rich region (GFP–Eps8ΔPR) or the GFP-tagged Eps8 mutant lacking the SH3 domain (GFP–Eps8ΔSH3) restores membrane blebbing in Eps8-KD DLD1 cells (arrows). (Scale bar, 10 μm.)

Fig. S1.

Eps8 is required for continuous membrane blebbing. (A) Exogenous expression of GFP-tagged mouse Eps8 significantly restores membrane blebbing in Eps8-KD DLD1 cells. Data are mean ± SD. *P < 0.05 (Student's t test) (n = 3). (B) Parental and Eps8-KD DLD1 cells were mixed and stained for Eps8 (red) and with phalloidin (green). Eps8-KD DLD1 cells are marked with asterisks. (Scale bar, 10 μm.) (C and D) Quantification of the fluorescence intensity of actin filaments along the white line shown in B. The fluorescence intensity of phalloidin was measured at six positions per cell. The SD was calculated based on the values from four independent experiments (Student's t test, **P < 0.01). (E) The fluorescence intensity of phalloidin was quantified in Eps8-KD DLD1 cells expressing wild-type Eps8, GFP–Eps8ΔPR, GFP–Eps8ΔSH3, GFP–Eps8ΔBundle, or GFP–Eps8ΔCap. SD was calculated based on the values from four independent experiments (Student's t test, **P < 0.01, *P < 0.05). (F) Parental and Eps8-KD DLD1 cells were mixed and stained for Eps8 (red) and for phosphorylated myosin light chain (p-MLC) (green). Eps8-KD DLD1 cells are marked with asterisks. (Scale bar, 10 μm.) (G) The fluorescence intensity of the p-MLC signal was quantified in wild-type DLD1 cells and Eps8-KD DLD1 cells. (H) The fluorescence intensity of p-MLC signal was quantified in Eps8-KD DLD1 cells expressing wild-type Eps8, GFP–Eps8ΔPR, GFP–Eps8ΔSH3, GFP–Eps8ΔBundle, or GFP–Eps8ΔCap. SD was calculated based on the values from four independent experiments (Student's t test, *P < 0.05). (I) The fluorescence intensity of phalloidin was quantified in wild-type DLD1 cells and ezrin-KO DLD1 cells. (J) The fluorescence intensity of the p-MLC signal was quantified in wild-type DLD1 cells and ezrin-KO DLD1 cells.

Eps8 regulates actin barbed-end capping activity and mediates actin bundling (16–19). In addition, Eps8 interacts with a number of proteins, including Sos (16), IRSp53 (17), Src (20), Abi (21), and EGFR (22), through its proline-rich (PR) region and SH3 domain. Therefore, we examined which function of Eps8 underlies its involvement in the regulation of membrane blebbing. We performed a mutational analysis of Eps8 (Fig. 2 D and E) (23) and tested whether these mutants could rescue membrane blebbing when transiently expressed in Eps8-KD DLD1 cells (Fig. 2 F and G). Among the tested mutants, a GFP-tagged Eps8 mutant lacking the PR region (GFP–Eps8ΔPR) and a GFP-tagged Eps8 mutant lacking the SH3 domain (GFP–Eps8ΔSH3) rescued membrane blebbing in Eps8-KD DLD1 cells (Fig. 2 F and G). On the other hand, expression of an actin filament-bundling–deficient mutant (GFP–Eps8ΔBundle) and an actin-capping–deficient mutant of Eps8 (GFP–Eps8ΔCap) did not reverse the phenotype of Eps8-KD DLD1 cells (Fig. 2 F and G). Furthermore, in Eps8-KD DLD1 cells, expression of GFP–Eps8ΔPR or GFP–Eps8ΔSH3 reestablished the formation of contractile actin cables and the level of myosin light chain (MLC) phosphorylation (Fig. S1 E and H). On the other hand, expression of GFP–Eps8ΔBundle or GFP–Eps8ΔCap failed to rescue the reduction in the amount of actin filaments covering the entire cell periphery and the level of MLC phosphorylation in Eps8-KD DLD1 cells (Fig. S1 E and H). These data suggest that both the actin-bundling and actin-capping activities of Eps8 are essential for the formation and/or stabilization of cytoplasmic actomyosin cables and for the generation of continuous membrane blebbing.

Ezrin Is Activated in Multiple Regions of the Protruded Membrane.

What recruits Eps8 to specific foci of membrane blebs? Ezrin/radixin/moesin (ERM) family proteins were recently reported to bind to Eps8 directly and regulate its functions (24). Furthermore, Eps8 and ERM proteins were recently reported to colocalize at the plasma membrane of membrane blebs (25). Therefore, we analyzed the functional relationship between ERM family proteins and Eps8 in membrane blebbing in detail.

As already reported (11), GFP–ezrin did not show a restricted distribution pattern but localized uniformly at the blebbing membrane (Fig. 3A). However, ezrin adopts an active open conformation when phosphorylated at Thr567. Therefore, we next examined the distribution of phosphorylated ERM during blebbing. Because GFP–ezrin uniformly localized at all membrane blebs, an anti-total ERM antibody stained all protruded membrane blebs (Fig. 3B). Interestingly, an anti-phosphorylated ERM antibody stained some membrane blebs more strongly, suggesting that ERM is activated only in expanding or retracting blebs (Fig. 3B). Phosphorylated ERM colocalized with Eps8 and actin filaments, indicating that ERM was activated only during membrane bleb retraction (Fig. 3 C and D).

Fig. 3.

Activation of ezrin occurs at retracting membranes and is required for the rapid retraction of membrane blebs. (A) Membrane blebbing in DLD1 cells transfected with GFP–ezrin. GFP–ezrin localizes uniformly at the protruded membrane. (Scale bar, 2 μm.) (B) DLD1 cells were fixed and stained with an anti–phospho-ERM antibody (red) and an anti-total ERM antibody (green). Nuclei were stained with DAPI (blue). The asterisk indicates a membrane bleb in which ERM proteins were not activated. (Scale bar, 2 μm.) (C) DLD1 cells were fixed and stained with an anti–phospho-ERM antibody (green) and an anti-Eps8 antibody (red). The arrowheads indicate the colocalization of Eps8 and phosphorylated ERM proteins. (Scale bar, 2 μm.) (D) DLD1 cells were fixed and stained with an anti–phospho-ERM antibody (green) and Alexa 594–phalloidin (red). The boxed area shows the membrane blebs with regrowing actin filaments. High-magnification image of the boxed area is shown in the right panels. The asterisks indicate a membrane bleb covered with actin cortex. (Scale bar, 10 μm.) (E) Total cell lysates of wild-type DLD1 cells and ezrin-KO DLD1 cells separated by SDS/PAGE and immunoblotted with an anti-total ERM antibody, an anti-ezrin antibody, an anti-Eps8 antibody, and an anti–α-tubulin antibody. (F) Membrane blebbing of wild-type DLD1 cells and ezrin-KO DLD1 cells transfected with Lifeact–RFP and GFP-tagged Eps8. (Scale bar, 10 μm.) (G) Tricolor map of membrane blebs in wild-type DLD1 cells and ezrin-KO DLD1 cells. Angular coordinates are shown on the horizontal axis, and time is shown on the vertical axis. Red zones represent expansion, blue zones represent retraction, and white zones represent no movement. (H) Histogram of bleb expansion and retraction velocities in wild-type DLD1 cells and ezrin-KO DLD1 cells. (I) The frequencies of membrane blebs in wild-type DLD1 cells and ezrin-KO DLD1 cells during 10 min were quantified. **P < 0.01 (Student's t test). (J) The sizes of membrane blebs in wild-type DLD1 cell and in ezrin-KO DLD1 cells during 10 min were quantified. **P < 0.01 (Student's t test).

Next, we examined the effect of ERM depletion on the localization of Eps8 in membrane blebs. In DLD1 cells, only ezrin was expressed among ERM family proteins (Fig. 3E). Therefore, we knocked out the ezrin gene using the CRISPR/Cas9 system (Fig. 3E). In ezrin-knockout (KO) DLD1 cells, Eps8 was no longer recruited to retracting membrane blebs and remained in the cytoplasm (Fig. 3F). Furthermore, actin reassembly at the protruding membrane was severely retarded in ezrin-KO DLD1 cells in comparison with wild-type DLD1 cells (Movies S3 and S4). We then performed histogram analyses of bleb expansion and retraction velocities in wild-type DLD1 cells and ezrin-KO cells (Fig. 3H). The speed of the retraction phase of membrane blebs in ezrin-KO cells was slower than that in wild-type DLD1 cells (Fig. 3H). Slow retraction of blebs decreased the frequency of blebs and the formation of larger blebs in ezrin-KO cells (Fig. 3 I and J). These analyses strongly support the notion that the retraction of membrane blebs was severely retarded in ezrin-KO cells.

On the other hand, the expanding speed of membrane blebs was not significantly different between ezrin-KO cells and wild-type cells (Fig. 3H). Although Eps8 KD cells did not show any membrane blebbing, ezrin-KO cells showed continuous slow membrane blebbing. This difference in phenotype between Eps8 KD and ezrin-KO cells may be partly because the cytoplasmic pool of Eps8 remained intact in ezrin-KO cells. In good agreement with this assumption, the cytoplasmic actomyosin cortex of ezrin-KO cells was similar to that of wild-type cells (Fig. S1 I and J). Therefore, we conclude that loss of ezrin only affected the retraction phase of membrane blebbing, not the expanding phase.

Exogenous expression of wild-type ezrin, but not the phosphorylation-defective mutant (T567A) of ezrin, rescued the localization of Eps8 at retracting membrane blebs in ezrin-KO DLD1 cells (Fig. S2 A and B). As previously reported (11), the phosphor-mimic mutant of ezrin (T567E) suppressed the formation of membrane blebs. Eps8 accumulated uniformly along the plasma membrane in ezrin-KO cells overexpressing the T567E mutant (Fig. S2A). These findings also support the notion that activation of ezrin is required for the recruitment of Eps8 to the retracting blebs and for the reassembly of actin filaments at the protruded membranes.

Fig. S2.

Ezrin is required for the recruitment of Eps8 to protruded membrane blebs and the rapid retraction of membrane blebs. (A) Localization of GFP–Eps8 in ezrin-KO DLD1 cells transfected with wild-type ezrin, the T567A ezrin mutant, and the T567E ezrin mutant. (Scale bar, 10 μm.) (B) The percentage of cells with blebs in which GFP–Eps8 was recruited to the plasma membrane relative to the total number of GFP–Eps8–positive cells in a given field was calculated. In each experiment, the total cell number was 100 (n  =  3). Data are the mean ± SD. *P < 0.05 (Student's t test). (C) Transwell migration assays were performed using Transwell filters (8-μm pore diameter; Corning Costar) coated with 1.0 mg/dL type I collagen. Membrane filter inserts were precoated with 100 μL of 1.0 mg/dL type I collagen solution in each well before seeding cells. Then, 2 × 10⁵ cells were seeded into the upper chamber, which contained the same medium as the lower chamber. After migration for 6 h, cells on the lower side were stained with DAPI, and the number of cells per high-power field was counted. Data are the mean ± SD. *P < 0.05 (Student's t test).

We next examined whether the retardation of actin reassembly affected the speed of cell migration using the Boyden chamber assay. In a Transwell migration assay, ezrin-KO DLD1 cells showed reduced migratory activity in comparison with wild-type DLD1 cells (Fig. S2C). Reexpression of ezrin significantly rescued the phenotype of ezrin-KO DLD1 cells; however, expression of the phosphorylation-defective mutant (T567A) mutant did not enhance the migration of ezrin-KO DLD1 cells (Fig. S2C). Taken together, these findings show that activation of ERM is required for recruitment of Eps8 to retracting membrane blebs, rapid reassembly of the actin cortex in membrane blebs, and efficient cell migration.

Activation of Ezrin Is Regulated by the RhoA–ROCK–Rnd3 Feedback Loop.

Finally, we examined the molecular mechanisms that regulate activation of ezrin only at retracting membranes during membrane blebbing. The RhoA–ROCK pathway is reportedly involved in the activation of ERM family proteins (26–30); therefore, we examined the distribution of ROCK in membrane blebs. When GFP-tagged ROCK-1 was expressed in DLD1 cells, it was recruited specifically to the retracting membrane (Fig. 4A). GFP–ROCK-1 was recruited to the retracting membrane blebs even in ezrin-KO cells, indicating that the recruitment of ROCK is epistatic to the activation of ezrin (Fig. 4B).

Fig. 4.

The RhoA–ROCK–Rnd3 feedback loop determines the actin reassembly sites of retracting membranes. (A) GFP–ROCK-1 is recruited to the retracting protruded membrane in DLD1 cells. (Scale bar, 2 μm.) (B) GFP–ROCK is recruited to the retracting protruded membrane in ezrin-KO DLD1 cells (arrowheads). (Scale bar, 2 μm.) (C) Membrane blebbing of DLD1 cells transfected with Lifeact–RFP and GFP-tagged AHD. (Scale bar, 2 μm.) (D) Membrane blebbing of DLD1 cells transfected with Lifeact–RFP and GFP-tagged Rnd3. Rnd3 accumulation gradually disappears upon the initiation of membrane blebbing retraction (t = 25 s). (Scale bar, 2 μm.) (E) Kymographs showing the localization of actin (red) with respect to that of Rnd3 (green) during bleb retraction. Bleb extension is shown on the horizontal axis, and time is shown on the vertical axis. (F) Kymographs showing the localization of Rnd3 (red) with respect to that of Eps8 (green) during bleb retraction. Bleb extension is shown on the horizontal axis, and time is shown on the vertical axis. (G) GFP–p190B–Rho–GAP localizes only at expanding blebs that lack the actin cortex. The membrane localization of p190B Rho–GAP is gradually lost upon the initiation of actin cortex recovery. (Scale bar, 2 μm.)

Then, we next examined the distribution of active RhoA in membrane blebs. It was recently reported that the well-conserved C-terminal domain of anillin [anillin homology domain (AHD)] selectively binds to the GTP-bound form of RhoA and that GFP-tagged AHD is a useful biosensor for the detection of active RhoA (31). When GFP–AHD was expressed in DLD1 cells, it was recruited to retracting membrane blebs where reassembly of the actin cortex occurred (Fig. 4C and Movie S5).

As for the molecular mechanisms involved in the selective activation of RhoA in membrane blebbing, there are at least two possibilities. One possibility is that certain Rho–GEFs are recruited to retracting membranes. The other possibility is that Rho–GAPs are recruited to expanding blebs to inhibit RhoA activation. We tested these possibilities and found that Rnd3 and p190B–Rho–GAP were recruited to expanding membrane blebs where reassembly of the actin cortex did not occur (Fig. 4 D–F). Rnd3 (also named RhoE) reportedly antagonizes RhoA signaling by activating p190–Rho–GAP (32). Interestingly, Rnd3 was recruited to expanding membrane blebs; however, this membrane localization was gradually lost during the retraction of membrane blebs (Fig. 4 D and E, and Movie S6). When mCherry–Rnd3 and GFP–Eps8 were simultaneously expressed in DLD1 cells, Rnd3 and Eps8 were only detected in the expanding membrane and retracting membrane, respectively (Fig. 4F and Movie S7). GFP–p190–Rho–GAP also had a distribution pattern similar to that of Rnd3 during membrane blebbing (Fig. 4G).

Rnd3 disappeared as the reassembly of the actin cortex proceeded (Fig. 4E). However, when the actin cytoskeleton was disrupted by Latrunculin B treatment, Rnd3 persisted at the protruded membrane, indicating that Rnd3 preferentially localizes to the cortex-free plasma membrane (Fig. S3).

Fig. S3.

Membrane localization of Rnd3 persists at the protruded plasma membranes in DLD1 cells treated with Latrunculin-B. Membrane blebbing of DLD1 cells transfected with Lifeact–RFP and GFP-tagged Rnd3. Cells were treated with 1 µM Latrunculin-B (Lat B) at t = 0. The accumulation of Rnd3 persisted at the actin cytoskeleton-free protruded membranes (arrowheads). (Scale bar, 10 μm.)

Rnd3 is a constitutively active GTP-binding Rho family protein, and its membrane localization is inhibited when Rnd3 is phosphorylated by ROCK (33). Membrane retention of Rnd3 was recently reported to be regulated by ROCK phosphorylation of serine-240, which leads to the sequestration of Rnd3 in the cytoplasm via its binding to 14-3-3 protein (34). Interestingly, overexpression of the Rnd3 S240A mutant, which is resistant to inhibition by ROCK, phenocopied the ezrin-KO bleb phenotype. (Fig. 5 A and B, and Movie S8). When Rnd3 S240A was overexpressed, the membrane recruitment of Eps8 was retarded (Fig. 5A), resulting in a slower membrane bleb retraction phase and the formation of larger blebs than in wild-type DLD1 cells (Fig. 5 C–E).

Fig. 5.

A model of Rnd3- and RhoA-mediated regulation of actin cytoskeleton during membrane-blebbing cycle. (A) Localization of Eps8 in DLD1 cells expressing GFP–wild type Rnd3 (Upper) and GFP–Rnd3 S240A mutant. (Scale bar, 10 μm.) (B) Tricolor map of membrane blebs in DLD1 cells expressing GFP–wild type Rnd3 or GFP–Rnd3 S240A mutant. Angular coordinates are shown on the horizontal axis, and time is shown on the vertical axis. Red zones represent expansion, blue zones represent retraction, and white zones represent no movement. (C) Histogram of bleb expansion and retraction velocities in DLD1 cells expressing wild-type Rnd3 or the Rnd3 S240A mutant. (D) The frequencies of membrane blebs in DLD1 cells expressing GFP–wild type Rnd3 or GFP–Rnd3 S240A mutant during 10 min were quantified. **P < 0.01 (Student's t test). (E) The sizes of membrane blebs in DLD1 cells expressing GFP–wild type Rnd3 or GFP–Rnd3 S240A mutant during 10 min were quantified. *P < 0.05 (Student's t test). (F) In the expansion phase of membrane blebbing, Rnd3 and p190B–Rho–GAP inhibit the activation of RhoA. As the protruded membrane areas become enlarged, the relative concentration of Rnd3 decreases. Sporadic activation of RhoA leads to ROCK phosphorylation of Rnd3 and removal of p190B–Rho–GAP from the membrane. Thus, RhoA activation is amplified and sustained by the positive-feedback loop. ROCK also phosphorylates ezrin and activated ezrin recruits, which leads to reassembly of the actin cortex.

Taken together, these findings show that, although Rnd3 and p190–Rho–GAP are predominant at expanding membrane blebs, RhoA and ROCK are activated at retracting membrane blebs. We propose that this positive-feedback mechanism of RhoA underlies the switch between the expansion and retraction phases of membrane blebbing (Fig. 5F). When the intracellular pressure increases, the plasma membrane protrudes. Rnd3 and p190–Rho–GAP are present at the plasma membrane and inhibit RhoA activation at the expanding plasma membrane. When the protruded membrane area increases and the concentration of Rnd3 per surface area of the membrane decreases, sporadic activation of RhoA may be amplified and stabilized by RhoA–ROCK phosphorylation of Rnd3, subsequent sequestration of Rnd3 in the cytoplasm, and inactivation of p190–Rho–GAP. Increased ROCK activity leads to the phosphorylation of ERM and the recruitment of Eps8 to the protruded membrane. The activation of ERM and recruitment of Eps8 at the plasma membrane induces the rapid retraction of the protruded membrane by promoting the reassembly of the actin cortex. In conjunction with activation of ROCK, active RhoA may also activate another RhoA effector protein, mDia1, which was recently reported to be essential for the regrowth of actin filaments at the retracting membrane (13).

Finally, we would like to discuss some unsolved issues in this study. In ezrin-KO cells or in cells overexpressing Rnd3 S240A, the retraction of membrane blebs was severely retarded. However, it should be noted that, at the plasma membrane in ezrin-KO cells or in cells overexpressing Rnd3 S240A, actin filaments continued to cover the entire protruded membrane, albeit more slowly, even in the absence of Eps8. Therefore, another mechanism may be present that enables ezrin- and Eps8-independent reconstruction of the actin cytoskeleton at the protruded membrane. This molecular mechanism will require investigation in future studies.

In the present study, we proposed that the local positive-feedback loop of RhoA activation and global suppression by Rnd3 determines the actin reassembly foci in membrane blebs. There may be unidentified molecules involved in these processes. In terms of candidates, the STRIPAK complex components FAM40A, FAM40B, and STRN3 were recently reported to be involved in the amoeboid migration of cancer cells by regulating the activation state of ERM proteins (35). Therefore, it will be interesting to examine how the STRIPAK complex is involved in the regulation of membrane blebbing in future studies.

In addition to understanding the molecular components involved in the regulation of membrane blebs, it is also important to clarify whether changes in the physical properties of the protruded membrane are involved in the regulation of membrane blebs. Membrane tension reportedly changes during bleb expansion (36, 37), suggesting that it may lead to the gating of stretch-activated ion channels and evoke active signaling (38). On the other hand, membrane tension was recently reported to modulate the activity of small GTPases (39). We showed that Rnd3 preferentially localizes to the cortex-free plasma membrane of expanding membrane blebs (Fig. S3), but the molecular mechanism remains to be clarified. In future studies, it will be very interesting to determine how cells sense the actin cortex-free membrane via mechanical and chemical signals.

Materials and Methods

Reagents.

DLD1 cells were grown in DMEM supplemented with 10% (vol/vol) FCS. The following primary antibodies were used for immunofluorescence microscopy and immunoblotting: mouse anti-Eps8 monoclonal antibody (mAb) (Becton Dickinson), mouse anti–α-tubulin mAb (Sigma), rabbit anti-ezrin antibody, rabbit anti-ERM antibody, and rabbit anti–phospho-ezrin (Thr567)/radixin (Thr564)/moesin (Thr558) mAb (41A3). cDNAs encoding full-length ezrin, MRLC1, Eps8, anillin, p190–Rho–GAP-B, Rnd3, and RhoA were amplified by RT-PCR, fused to the sequence encoding enhanced GFP, and ligated into the pCAGGS-neo vector. Expression vectors of GFP–Rnd3 S240A mutant or GFP–ROCK-1 was kindly provided by Dr. A. S. Yap (University of Queensland, St. Lucia, Queensland, Australia) or Dr. M. Takeichi (CDB, Kobe, Japan), respectively. Stable ezrin-KO clones were produced using the CRISPR–Cas9 system (40). Oligonucleotides were phosphorylated, annealed, and cloned into the BbsI site of PX330 according to protocols of the Feng Zhang laboratory (Massachusetts Institute of Technology, Cambridge, MA).

Immunofluorescence Microscopy, Live Imaging, Tricolor Map, and Histogram Analysis.

Details about immunofluorescence microscopy, live imaging, and the quantitative analyses of live-imaging data are provided in Supporting Information and Figs. S4 and S5.

Fig. S4.

Extraction of cell contour and its smoothing for spatiotemporal visualization of membrane blebs. (A) Original grayscale frame image. (B) The cell contour is extracted by Otsu’s binarization. From the binarization image, the center of gravity, $p$ (green point), and the distance between $p$ and the cell contour in the direction $\theta$, $d_{\theta ,t}$ (orange line), are determined. In the lower image, the orange line intersects with the contour three times (as indicated by orange arrows), and thus there is ambiguity to determine the distance $d_{\theta ,t}$. (C) A smoothed contour (red) is derived through an optimization algorithm based on dynamic programming. By the smoothing, any line emerged from $p$ will intersect with the contour only once, and thus the distance $d_{\theta ,t}$ is determined uniquely. (D) The process of the optimal smoothing is illustrated. The value $d_{\theta }$ is the original distance given by the binarization. The aim of the optimization is to determine $a=\widehat{d_{\theta }}$ and $b=\widehat{d_{359-\theta }}$ for $\theta =0,\dots 179$ so that $a\sim d_{\theta }$ and $b\sim d_{359-\theta }$ and $a$ and $b$ will not violate the conditions for smoothness.

Fig. S5.

Spatiotemporal visualizations of membrane blebs and their application for counting blebs. (A) An example of the distance map shows white “blobs” on it. Because a whiter part in the distance map indicates that the part is further from the center, a consecutive white region, i.e., blob, will correspond to a bleb. The horizontal width of a blob is relative to the spatial width of a bleb, and the vertical height is relative to the lifetime of a bleb. (B) A tricolor map visualizes the membrane bleb dynamics more directly. An expanding bleb is represented as a red blob, and a retracting bleb is represented as a blue blob. (C) The number of the membrane blebs emerged in a video can be measured by detecting the local peaks (green) in the distance map, because each white blob in the distance map corresponds to a bleb. (D) By using the process of the peak detection, a segmentation result of the distance map is obtained. Each segment clarifies the area of a blob and, equivalently, spatiotemporal area of a bleb. (E) The color segmentation result of D is used for showing each bleb area on the contour of a frame. A bleb is represented in the same color from the beginning of its expansion to the end of the retraction.

Immunofluorescence Microscopy

In brief, cells cultured on coverslips were fixed with 8% paraformaldehyde prepared in PBS for 10 min at room temperature (RT), treated with 50 μg/mL digitonin prepared in PBS for 5 min, and washed with PBS three times. Fixed cells were blocked by incubation with 5% BSA prepared in PBS for 30 min at RT. Cells were incubated at RT for 1 h with the primary antibody and for 30 min with the secondary antibody. For actin staining, Alexa Fluor 488 or 594–phalloidin (Life Technologies) was included with the secondary antibody. Specimens were observed at RT with a confocal microscope (LSM700; Carl Zeiss MicroImaging) equipped with a heating stage heated to 37 °C and a Plan-APO (63/1.40 N.A., oil immersion) objective with appropriate binning of pixels and exposure time. Images and movies were analyzed with ZEN 2012 (Carl Zeiss MicroImaging).

Live Imaging

All fluorescence imaging was performed using a 63× oil-immersion objective on an inverted microscope (LSM700; Carl Zeiss MicroImaging) interfaced to a laser-scanning confocal microscope equipped with a heating stage heated to 37 °C. Images were captured on a device camera and acquired on a PC using ZEN2012 software (LSM700; Carl Zeiss MicroImaging). Images were acquired at 488 nm for GFP-tagged proteins or at 555 nm for RFP- or mCherry-tagged proteins. Each imaging video frame is a 8-bit grayscale image, and the frame interval is indicated in the supplementary movie legends. The movie captures a single cell.

Cell Contour Extraction

Because the movie captures only a single cell, its contour can be extracted easily by Otsu’s binarization method (Fig. S4 A and B). Let $I_t$ denote the binarized image of the $t$ th video frame. Its pixel value $I_t\left(x,y\right)$ becomes 1 (Fig. S4B, white region) if the pixel $\left(x,y\right)$ is inside the cell region and $0$ (black) if not. The cell contour at the $t$ th frame is roughly represented as the boundary of the largest connected component of pixels with value 1.

Visualization of Membrane Bleb Dynamics

For analyzing membrane bleb dynamics, we first generate a “distance map” $D\left(\theta ,t\right)$, which is a 2D visualization of spatiotemporal activity of membrane blebs. Its value $d_{\theta ,t}=D\left(\theta ,t\right)$ is the distance between the contour (i.e., membrane) and the center of gravity of the cell region, $p$, in the direction $\theta$ at the frame $t$ (Fig. S4B). The center of gravity $p$ is fixed for all frames and determined as follows:

$$p=\left(\frac{{\displaystyle {\sum }_{x,y,t}x\cdot I_t\left(x,y\right)}}{{\displaystyle {\sum }_{x,y,t}{I}_t\left(x,y\right)}},\frac{{\displaystyle {\sum }_{x,y,t}y\cdot I_t\left(x,y\right)}}{{\displaystyle {\sum }_{x,y,t}{I}_t\left(x,y\right)}}\right).$$

Because contour shape tends to be complex due to blebs, there are sometimes multiple intersections with the contour in the direction $\theta$ (Fig. S4B, Lower). In this case, there is ambiguity to determine the distance $d_{\theta }$ for the direction $\theta$. To solve those problems, a contour smoothing operation is applied at individual frames before generating the distance map. Note that the aim of our smoothing is not just to make the curvature of the contour lower but to make any line emerged from p intersect with the contour only once. This means that the distance $d_{\theta }$ is determined uniquely (as $\widehat{d_{\theta }}$ in the later notation) for arbitrary direction $\theta$. For this aim, typical smoothing operations (such as spline approximation and low-pass filtering) are not applicable. We need to design a new smoothing operation for our aim.

Specifically, we treat the contour smoothing operation as the following optimization problem (hereafter, $d_{\theta }$ and $\widehat{d_{\theta }}$ are used instead of $d_{\theta ,t}$ and $\widehat{d_{\theta ,t}}$ for simplicity):

$$\text{minimize}J={\displaystyle \sum }_{\theta =0}^{359}\left|d_{\theta }-\widehat{d_{\theta }}\right|,$$

$$\text{with respect to}\left\{\widehat{d_{\theta }},|\theta \in \left[\mathrm{0,359}\right]\right\},$$

subject to the following:

• (condition 1) $\widehat{d_{\theta }}\in \left[d^{min},d^{max}\right],$

• (condition 2) $-\epsilon \le \widehat{d_{\theta }}-\widehat{d_{\theta +1}}\le \epsilon \hspace{0.17em}\hspace{0.17em}\text{for}\hspace{0.17em}\theta \in \left[\mathrm{0,358}\right]\hspace{0.17em}\hspace{0.17em}\text{and}\hspace{0.17em}-\epsilon \le \widehat{d_{359}}-\widehat{d_0}\le \epsilon ,$

where the value $d_{\theta }$ is the distance between $p$ and the original contour in the direction $\theta$ and the variable $\widehat{d_{\theta }}$ is the distance after smoothing. Accordingly, the minimization of the objective function $J$ means the minimization of the change by smoothing (i.e., change from $d_{\theta }$ to $\widehat{d_{\theta }}$). Note that, if we have multiple intersections in the direction $\theta$, that is, if we have multiple distances as $d_{\theta }$, we will choose $d_{\theta }$, which is closest to $\widehat{d_{\theta }}$ for the calculation of $\left|d_{\theta }-\widehat{d_{\theta }}\right|$. One, therefore, might think that $\widehat{d_{\theta }}=d_{\theta }$ for all $\theta$ is the optimal because $J=0$; however, this might be wrong in some case because the following conditions are violated at some $\theta$.

Condition 1 restricts the range of $\widehat{d_{\theta }}$. The simplest example of the range is $\left[d^{min},d^{max}\right]=\left[0,W\right]$ or $\left[0,H\right]$ when the frame image size is $W\times H$. (This example is an extreme. Generally, it is possible to set a far narrower range, and the narrower range will save the computation time drastically.) Condition 2 is more important because it regulates the smoothness of the contour after smoothing. Specifically speaking, the absolute difference between the neighboring contour points (specified by $\widehat{d_{\theta }}$ and $\widehat{d_{\theta +1}}$) should be smaller than $\epsilon$. The smaller the parameter $\epsilon$ is, the smoother the resulting contour is. Especially, if $\epsilon =0$, the resulting contour becomes a circle. A minor but important point is that condition 2 also restricts the smoothness between $\widehat{d_{359}}$ and $\widehat{d_0}$. In other words, without this additional constraint, the resulting contour might show some discontinuity between $\theta =359$ and 0.

The globally optimal solution of the above optimization problem for contour smoothing is derived by dynamic programming (DP). The DP-based algorithm for the derivation is shown as (Fig. S4D):

$$\begin{array}{ll}\text{Input}\hspace{0.17em}\mathrm{:}\hfill & \begin{array}{l}\text{Initial}\hspace{0.17em}\text{contour}\hspace{0.17em}\text{by}\hspace{0.17em}\text{binarization},\hspace{0.17em}\left\{d_{\theta },|\theta \in \left[0,359\right]\right\}\\ \text{Valid}\hspace{0.17em}\text{distance}\hspace{0.17em}\text{range}\hspace{0.17em}U=\left[d^{min},d^{max}\right]\\ \text{Valid}\hspace{0.17em}\text{distance}\hspace{0.17em}\text{change}\hspace{0.17em}\text{range}\hspace{0.17em}V=\left[-\epsilon ,\epsilon \right]\end{array}\hfill \\ \text{Step}\hspace{0.17em}\text{1:}\hfill & \text{for}\hspace{0.17em}\text{all}\hspace{0.17em}a\in U,\hspace{0.17em}b\in U\hspace{0.17em}\text{satisfying}\hspace{0.17em}a-b\in V\hfill \\ \hspace{1em}\hspace{1em}2\mathrm{:}\hfill & g\left(a,b;0\right)=\left|d_0-a\right|+\left|d_{359}-b\right|\hfill \\ \hspace{1em}\hspace{1em}3\mathrm{:}\hfill & \text{for}\hspace{0.17em}\theta =1,\cdots ,179\hfill \\ \hspace{1em}\hspace{1em}4\mathrm{:}\hfill & \text{for}\hspace{0.17em}\text{all}\hspace{0.17em}a\in U,\hspace{0.17em}b\in U\hfill \\ \hspace{1em}\hspace{1em}5\mathrm{:}\hfill & g\left(a,b;\theta \right)=\left|d_{\theta }-a\right|+\left|d_{359-\theta }-b\right|+\underset{\left\{\left(a\prime ,b\prime \right)|a-a\prime \in V,b-b\prime \in V\right\}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}}{\min }g\left(a\prime ,b\prime ;\theta -1\right)\hfill \\ \hspace{1em}\hspace{1em}6\mathrm{:}\hfill & \left(\widehat{d_{179}},\widehat{d_{180}}\right)=\underset{\left\{\left(a\prime ,b\prime \right)|a\prime -b\prime \in V\right\}}{\mathrm{argmin}}g\left(a\prime ,b\prime ;179\right)\hfill \\ \hspace{1em}\hspace{1em}7\mathrm{:}\hfill & \text{for}\hspace{0.17em}\theta =178,\cdots \hspace{0.17em}0\hfill \\ \hspace{1em}\hspace{1em}\text{8}\mathrm{:}\hfill & \left(\widehat{d_0,}\widehat{d_{359-\theta }}\right)=\underset{\left\{\left(a\prime ,b\prime \right)|\widehat{d_{\theta +1}}-a\prime \in V,\widehat{d_{359-\theta -1}}-b\prime \in V\right\}\hspace{0.17em}}{\mathrm{argmin}}g\left(a\prime ,b\prime ;\theta \right)\hfill \\ \text{Output}\mathrm{:}\hfill & \hspace{0.17em}\text{The}\hspace{0.17em}\text{globally}\hspace{0.17em}\text{optimal}\hspace{0.17em}\text{smoothed}\hspace{0.17em}\text{contour}\left\{\widehat{d_0,}|\theta \in \left[0,359\right]\right\}\hfill \end{array}$$

Although we will not describe the details of the algorithm, it might be noteworthy that it uses a technique to apply DP for the optimization of “cyclic” variables, where the first and the last variables are related to each other. For our case, $\widehat{d_0}$ and $\widehat{d_{359}}$ are those variables and related by condition 2. Usually, DP is used for the optimization of sequential variables, where those variables are not related. In the proposed algorithm, we therefore convert cyclic variables into sequential variables by coupling two variables; that is, we convert $\widehat{d_0,}\widehat{d_1,}\dots ,\widehat{d_{\theta },}\dots ,\widehat{d_{359}}$ into $\left(\widehat{d_0,}\widehat{d_{359}}\right),\left(\widehat{d_1,}\widehat{d_{358}}\right),\dots ,\left(\widehat{d_{\theta },}\widehat{d_{359-\theta }}\right),\dots ,\left(\widehat{d_{179},}\widehat{d_{180}}\right)$. In the resulting new sequence, the first (2D) variable $\left(\widehat{d_0,}\widehat{d_{359}}\right)$ and the last variable $\left(\widehat{d_{179},}\widehat{d_{180}}\right)$ are not related anymore. Another noteworthy point is that DP examines all possible values for each variable $\widehat{d_{\theta }}$ for guaranteeing the global optimality of its solution. This point can be confirmed by watching steps 1 and 4, where all possible values of $\widehat{d_{\theta }}$ and $\widehat{d_{359-\theta }}$ are examined as $a$ and $b$, respectively, and the results are stored into $g\left(a,b;\theta \right)$ in steps 2 and 5.

By applying the above contour-smoothing operation for each frame $t\in \left[1,T\right]$, the distance map $D\left(\theta ,t\right)$ is generated. Because this map is 2D, it is observed as a $360\times T$ grayscale image and its intensity value is relative to the (smoothed) distance $\widehat{d_{\theta ,t}}$. From an example of the distance map (Fig. S5A), it is possible to guess the expansion and the retraction of blebs. Specifically, in the distance map, there are many white “blobs” and each blob suggests the spatiotemporal region occupied by a bleb. The horizontal width of a blob is relative to the spatial width of a bleb, and the vertical height is relative to the lifetime of a bleb. Moreover, the whitest peak inside a blob will corresponds to the direction $\theta$ and time $t$ of the most expanded contour point of a bleb.

Hereafter, we will introduce several modifications of the distance map for clarifying the bleb activity. The first modification is a (temporal) “difference map” $\mathrm{\Delta }\left(\theta ,t\right)$. The difference map is used for understanding the speed of contour shape motion, i.e., the speed of the expansion and the retraction of blebs. Using the distance map $D\left(\theta ,t\right)$, the difference map is generated immediately by $\mathrm{\Delta }\left(\theta ,t\right)=D\left(\theta ,t\right)-D\left(\theta ,t-1\right)$. This means that the motion speed is derived for each direction while assuming no rotational distortion in cell shape dynamics. From the difference map $\mathrm{\Delta }\left(\theta ,t\right)$, it is easy to derive a motion speed histogram of the cell just by counting how many elements of $\mathrm{\Delta }\left(\theta ,t\right)$ has a specific value $v.$

The difference map $\mathrm{\Delta }\left(\theta ,t\right)$ is further modified as a “tricolor map” $\overline{\mathrm{\Delta }}\left(\theta ,t\right)$ for emphasizing the expansion and the retraction of blebs (Fig. S5B). The tricolor map $\overline{\mathrm{\Delta }}\left(\theta ,t\right)$ is generated as follows:

$$\overline{\mathrm{\Delta }}\left(\theta ,t\right)=\{\begin{array}{cc}\text{Red},& \mathrm{\Delta }\left(\theta ,t\right)>\delta \\ \text{Blue},& \mathrm{\Delta }\left(\theta ,t\right)<-\delta \\ \text{White},& \text{otherwise}.\end{array}$$

In the tricolor map, an expanding bleb is represented as a red blob and a retracting bleb as a blue blob. If a part of the cell contour is almost stationary, the corresponding area will be pained as white in the tricolor map. In the example (Fig. S5B), preprocessing on $\mathrm{\Delta }\left(\theta ,t\right)$ and postprocessing on $\overline{\mathrm{\Delta }}\left(\theta ,t\right)$ were introduced for better visibility. The preprocessing is Gaussian smoothing (to suppress trivial speed fluctuation that makes jaggy tricolor boundary). The postprocessing is removal of very small red-colored and blue-colored connected components.

Counting Blebs

It is possible to count the number of blebs emerged during the video by using the distance map $D\left(\theta ,t\right)$. As noted above, a blob in the distance map will correspond to a bleb. We thus can count the number of blebs by counting the number of the local whitest peaks on $D\left(\theta ,t\right)$. We use “findmaxima” implemented in ImageJ for detecting the peaks (Fig. S5C). Note that a bleb around $\theta =0$ is observed as two blobs around the left border ($\theta =0$) and the right border ($\theta =359$), respectively, in $D\left(\theta ,t\right)$ and will give two peaks wrongly. A simple remedy to avoid this artifact is to concatenate two copies of the distance map $D\left(\theta ,t\right)$ horizontally into a larger distance map $D\prime \left(\theta ,t\right)$ ($\text{size}:\hspace{0.17em}720\times T$) and apply findmaxima to it. Then the correct number is given by counting the peaks between $\theta =180$ to 539.

It is also possible to segment the cell contour into individual blebs by using the segmentation function of findmaxima (Fig. S5D). The segmentation function divides the distance map $D\left(\theta ,t\right)$ into $N$ regions, where $N$ is the number of peaks. Roughly speaking, each region is the set of pixels belong to a peak and thus corresponds to a blob around the peak. If a unique color is assigned to each segment on $D\left(\theta ,t\right)$, we immediately can create a colored cell contour, where each bleb is outlined by a different color (Fig. S5E).