The activation of INF2 by Piezo1/Ca²⁺ is required for mesenchymal-to-amoeboid transition in confined environments

Summary

To invade tissues, cells may undergo a mesenchymal-to-amoeboid transition (MAT). However, the mechanisms regulating this transition are poorly defined. In melanoma cells, we demonstrate that intracellular [Ca²⁺] increases with the degree of confinement in a Piezo1-dependent fashion. Moreover, Piezo1/Ca²⁺ is found to drive amoeboid and not mesenchymal migration in confined environments. Consistent with a model in which Piezo1 senses tension at the plasma membrane, the percentage of cells using amoeboid migration is further increased in undulating microchannels. Surprisingly, amoeboid migration was not promoted by myosin light chain kinase (MLCK), which is sensitive to intracellular [Ca²⁺]. Instead, we report that Piezo1/Ca²⁺ activates inverted formin-2 (INF2) to induce widespread actin cytoskeletal remodeling. Strikingly, the activation of INF2 promotes de-adhesion, which in turn facilitates migration across micropatterned surfaces. Thus, we reveal a novel Piezo1/Ca²⁺/INF2 signaling cascade that regulates MAT, enabling cancer cells to adapt their migration mode in response to varying mechanochemical environments.

Graphical Abstract

eTOC Blurb:

Kar et al. find that Piezo1/Ca²⁺ activates INF2, which induces actin cytoskeletal remodeling and de-adhesion. This promotes mesenchymal-to-amoeboid transition (MAT), enabling cells to migrate through confined environments with varying levels of adhesive ligands.

Introduction

To migrate within tissues, cells must adapt to diverse mechanochemical environments. For instance, immune cells must move in and out of the blood and lymphatic systems to defend against pathogens. Similarly, cancer cells also migrate through these systems before establishing metastatic sites. Both types of cells become more invasive when they undergo a phenotypic transition from mesenchymal to amoeboid migration in response to mechanochemical signals.¹ Confinement has been shown to potently trigger the phenotypic transition from mesenchymal-to-amoeboid migration; however, the molecular mechanism(s) that underlie this phenomenon are not yet fully understood.^2–4

Confinement can trigger a phenotypic transition from mesenchymal-to-amoeboid migration in melanoma and other cancers. During mesenchymal migration, actin polymerization drives protrusion of the leading edge, whereas blebs drive protrusion of the leading edge during amoeboid migration.⁵ Blebs form when a segment of the plasma membrane separates from the underlying cortical actomyosin cytoskeleton, which requires high levels of intracellular pressure.⁶ Consequently, cells with high levels of cortical actomyosin contractility are prone to blebbing.² In most cells, blebs are retracted within a minute.⁷ Under certain conditions, however, cortical actomyosin flow in stable blebs drives “fast amoeboid migration”.²

In vertical or 2-dimensional (2D) confinement, cell movement may be driven by cortical actomyosin flow in stable blebs.^2–4 As cells move in the direction of this bleb, it is termed a ‘leader bleb’.⁴ Accordingly, fast amoeboid migration has also been described as leader bleb-based migration (LBBM).⁴ Importantly, melanoma and other cancer cell types have been observed in vivo to use a phenotypically similar mode of migration by intravital microscopy.⁸ Previous work by others demonstrated that confining cells down to 3 μm and low adhesion potently triggers the phenotypic transition to fast amoeboid (leader bleb-based) migration in transformed cells.^2,9 Thus, elucidating the molecular mechanism(s) that underlie this phenomenon may provide new therapeutic avenues for preventing additional metastatic burden.

To undergo a phenotypic transition to amoeboid migration, cells must sense mechanochemical cues from the microenvironment. Work has shown that Piezo1, Phosphodiesterase 1 (PDE1), and cytosolic phospholipase A2 (cPLA2) promote migration in confined environments.^10–12 The plasma membrane tension sensor, Piezo1, was shown to promote migration through the Ca²⁺-mediated activation of PDE1.¹⁰ cPLA2 liberates arachidonic acid from the nuclear membrane, which activates myosin ATPases.^11,12 For full activity, cPLA2 requires that the nuclear membrane be stretched.^13,14 In this model, therefore, nuclear membrane stretch is key to promoting migration.^11,12 It remains unclear, however, if these factors are sufficient to induce a phenotypic switch from mesenchymal-to-amoeboid migration in confined environments.

In invasive melanoma cells subjected to confinement, we show that cells detect increasing levels of confinement through the stretch-activated cation-selective channel, Piezo1. Upon inhibiting Piezo1 or Ca²⁺ chelation, melanoma cells less frequently adopt amoeboid migration in 2D and 3D (i.e., microchannel) confined environments. In undulating channels, which are more likely to resemble the tortuous vasculature surrounding tumors, nearly all melanoma cells adopted an amoeboid migration mode in a Piezo1/Ca²⁺-dependent fashion. Importantly, Piezo1/Ca²⁺ is found to activate inverted formin-2 (INF2), which induces actin cytoskeletal remodeling and de-adhesion. By promoting de-adhesion, the activation of INF2 by Piezo1/Ca²⁺ drives mesenchymal-to-amoeboid transition (MAT) to facilitate migration through confined environments with varying levels of adhesive ligands.

Results

To promote fast amoeboid (leader bleb-based) migration, invasive melanoma (A375-M2) cells were placed in 2D confinement. To accomplish this, cells were placed under a slab of polydimethylsiloxane (PDMS), which is held at a defined height above the cover glass by micron-sized beads.¹⁵ Previously, it was shown that the switch to leader bleb-based migration (LBBM) occurs most frequently when cells are confined down to 3 μm.² Therefore, the PDMS was held above the cover glass by 3 μm beads.¹⁵ The PDMS and cover glass were coated with bovine serum albumin (BSA; 1%) to block integrin adhesion. Under these conditions, melanoma cells migrate in the direction of a leader bleb (Figure 1A & Video S1). These cells are termed leader mobile (LM). Cells forming leader blebs but not moving are called leader non-mobile (LNM; Figure S1A & Video S2). Cells without a leader bleb and that do not move are termed no leader (NL; Figure S1A). To evaluate the role of Ca²⁺ in regulating LBBM, we chelated Ca²⁺ using BAPTA (10 μM). This treatment resulted in an over 2-fold reduction in the number of leader mobile cells (Figure 1B). Morphometric analysis also found that cells less frequently adopt a shape consistent with LBBM after Ca²⁺ chelation, including a reduction in the largest bleb area (Figure 1C & Figure S1B). Therefore, we wondered if Piezo1 and/or 2 are required in melanoma cells for LBBM. Using RT-qPCR, we could only detect Piezo1 (Figure 1D). Therefore, we focused on Piezo1 for the rest of the study. To determine if Piezo1 has a role in sensing confinement, we correlated the level of intracellular Ca²⁺ with height. In melanoma cells treated with a non-targeting, i.e., control, small interfering RNA (siRNA), intracellular Ca²⁺ was found to sharply increase with confinement (Figure 1E). In contrast, intracellular Ca²⁺ levels and confinement were poorly correlated in cells lacking Piezo1 (Figure 1F–G). Upon evaluating migration in 2D confinement, we observed an over 2-fold reduction in the number of leader mobile cells (Figure 1H). Morphometric analysis also found that cells less frequently adopt a shape consistent with LBBM after Piezo1 RNAi, including a reduction in the largest bleb area (Figure 1I & Figure S1C). Similarly, a Piezo1 inhibitor significantly reduced the percentage of leader mobile cells (Figure S1D). In contrast, the percentage of leader mobile cells was significantly increased by a Piezo1 activator (Figure S1F). Morphometric analysis also found that the largest bleb area, aspect ratio, and roundness of cells were also affected by Piezo1 activity (Figure S1E–G). Thus, Piezo1/Ca²⁺ promotes LBBM in 2D confinement.

Figure 1. Piezo1/Ca²⁺ promotes fast amoeboid (leader bleb-based) migration.

(A) Montage of a leader mobile (LM) cell vertically confined down to a height of 3 μm. Cells were stained with a fluorescent membrane dye. Arrows point to the cell body (solid) and leader bleb (dashed), which is a large and stable bleb. (B) Percent of untreated (n=54) or BAPTA (10 μM; n=33) treated cells adopting a leader mobile (LM), leader non-mobile (LNM), or no leader (NL) phenotype (χ² ≤ 0.0001). (C) Compared to untreated, the largest bleb area was reduced in cells treated with BAPTA (10 μM) (Student’s t-test; mean +/− SEM). (D) RT-qPCR for Piezo1 and 2 on total mRNA (mean +/− SEM). (E and F) Compared to cells treated with a non-targeting siRNA, intracellular Ca²⁺ levels in cells treated with a Piezo1 siRNA no longer correlate with confinement height. Intracellular Ca²⁺ levels were measured by taking the ratio of a red fluorescent Ca²⁺ indicator to a green fluorescent dye. (G) 3 days after transfection, Piezo1 RNAi was confirmed by RT-qPCR on total mRNA (Student’s t-test; mean +/− SEM). (H) Percent of non-targeting (n=73) or Piezo1 siRNA (n=34) treated cells adopting a leader mobile (LM), leader non-mobile (LNM), or no leader (NL) phenotype (χ² ≤ 0.0001). (I) Compared to non-targeting, the largest bleb area was reduced in cells treated with a Piezo1 siRNA (Student’s t-test; mean +/− SEM). (J) Montage of a no leader (NL) cell vertically confined to 3 μm treated with a Piezo1 siRNA. Cells were stained with a fluorescent membrane dye. PDMS was coated with bovine serum albumin (BSA; 1%). * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Figure S1 & Video S1–2.

Cells may encounter channel-like environments within tissues, such as within micro-capillary/lymphatic vessels.¹ Therefore, we placed invasive melanoma (A375-M2) cells in microfabricated channels coated with VCAM-1 (1 μg/mL) to simulate the microvasculature. Under these conditions, cells predominantly adopt a hybrid migration mode, which has hallmarks of mesenchymal and amoeboid migration (e.g., blebs) (Figure 2A & Video S3). We have also observed melanoma cells adopting a hybrid migration mode in fibronectin-coated (10 μg/mL) channels.¹⁶ In the presence of a Piezo1 inhibitor, however, cells predominantly adopted a mesenchymal mode of migration (Figure 2B). We then wondered if we could bias cells towards an amoeboid phenotype using undulating channels, which are more likely to resemble the tortuous vasculature surrounding tumors. We speculate that Piezo1 may be gated open in undulating channels through a curvature-induced increase in plasma membrane tension. Indeed, in undulating channels, we found that cells were almost entirely amoeboid (Figure 2D & Video S3). If Ca²⁺ was chelated with BAPTA (10 μM), however, cells predominantly adopted a mesenchymal mode of migration (Figure 2F). Notably, mesenchymal cells were also frequently motile (Figure 2F). Using GCaMP, we observed more frequent but irregular spikes in intracellular [Ca²⁺] in cells adopting amoeboid features in microchannels (Figure 2C–E). When normalized to another dye, fluorescence from a Ca²⁺ indicator was also greater in cells within undulating channels (Figure 2G). Treating cells with a Piezo1 inhibitor increased the number of cells adopting a hybrid phenotype (Figure 3A–B). Piezo1 RNAi, however, led to a significant increase in the number of cells adopting a mesenchymal phenotype (Figure 3C). Although PDMS is pre-treated, many pharmacological agents are absorbed by PDMS, which may limit its effectiveness relative to RNAi. Interestingly, Piezo1 RNAi did not affect the speed and directionality of motile cells; therefore, Piezo/Ca²⁺ appears to primarily promote the appearance of amoeboid features (e.g., blebs) in microchannels (Figure 3D–E).

Figure 2. In microchannels, motile cells predominantly adopt a hybrid phenotype.

(A) In microchannels, cells predominantly adopt a hybrid phenotype (n=28), which has hallmarks of amoeboid (i.e., blebs) and mesenchymal migration. (B) In the presence of a Piezo1/2 inhibitor (GsMTx4; 10 μM) (n=165), cells predominantly adopt a mesenchymal phenotype (χ² ≤ 0.0001). (C) Compared to other phenotypes in smooth channels, amoeboid cells display the highest level of intracellular [Ca²⁺], as measured by the root mean square (RMS) output of GCaMP (multiple comparison test; mean +/− SEM). RMS encapsulates the dynamic nature of a fluorescence signal, providing a robust measure of its overall intensity across time. (D) In undulating channels, which are more likely to resemble the tortuous vasculature surrounding tumors, cells predominantly adopt an amoeboid phenotype (n=144). (E) Compared to other phenotypes in undulating channels, amoeboid cells display the highest level of intracellular [Ca²⁺], as measured by the RMS output of GCaMP (multiple comparison test; mean +/− SEM). (F) In undulating channels, cells treated with the Ca²⁺ chelator, BAPTA (10 μM; n=148), predominantly adopt a mesenchymal phenotype (χ² ≤ 0.0001). No blebs are observed in the zoomed image (D’). (G) Compared to cells in smooth channels, intracellular Ca²⁺ levels are elevated in cells within undulating channels. Intracellular Ca²⁺ levels were measured by taking the ratio of a red fluorescent Ca²⁺ indicator to a green fluorescent dye (Student’s t-test; mean +/− SEM). Microchannels are coated with VCAM-1 (1 μg/mL). * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Video S3.

Figure 3. Piezo1 promotes amoeboid migration in undulating channels.

(A and B) In undulating channels, cells treated with a Piezo1/2 inhibitor (GsMTx4; 10 μM) (n=96) more often adopt a hybrid phenotype (χ² ≤ 0.0001). For comparison, figure 2C is re-displayed. (C) Compared to cells treated with a non-targeting siRNA (n=155), cells treated with a Piezo1 siRNA (n=210) more often adopt a mesenchymal phenotype (χ² ≤ 0.05). (D) Based on manual cell tracking, the speed of non-targeting and Piezo1 siRNA-treated cells is similar (Student’s t-test; mean +/− SEM). (E) Based on manual cell tracking, the directionality ratio for non-targeting (n=51) and Piezo1 siRNA (n=35) treated cells is similar (mean +/− SEM). (F) Compared to non-targeting (n=155), the percent of motile cells or their phenotype is not significantly affected by treatment with a cPLA2 siRNA (PLA2G4A; n=135) (χ²=NS). (G) Based on manual cell tracking, the speed of non-targeting and cPLA2 siRNA-treated cells were similar (Student’s t-test; mean +/− SEM). (H) Based on manual cell tracking, the directionality ratio for non-targeting (n=51) and cPLA2 siRNA (n=69) treated cells were similar (mean +/− SEM). (I) 3 days after transfection, RNAi of cPLA2 (PLA2G4A) was confirmed by RT-qPCR on total mRNA (Student’s t-test; mean +/− SEM). Undulating channels are coated with VCAM-1 (1 μg/mL). * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Figure S2.

Intracellular Ca²⁺ regulates several signaling pathways, including cytosolic phospholipase A2 (cPLA2). However, cPLA2 requires nuclear membrane tension for full activity.^13,14 Accordingly, cPLA2 has been shown to promote migration in confined environments by liberating arachidonic acid from nuclear membranes.^11,12 Therefore, we wondered if cPLA2 may be downstream of Piezo1/Ca²⁺. In undulating channels, we could not detect any difference in the number or phenotype of motile cells after cPLA2 RNAi (PLA2G4A; Fig. 3F–I). cPLA2 RNAi also had no significant effect on the speed and directionality of motile cells (Fig. 3G–H). In 2D confinement, cPLA2 RNAi did not affect the number of leader mobile (LM) cells (Figure S2A). We then set out to determine if either myosin light chain kinase (MYLK), which is activated by Ca²⁺/calmodulin, or Rho-associated protein kinase (ROCK) may instead regulate amoeboid migration downstream of Piezo1/Ca²⁺. By RT-qPCR, we could detect the MYLK, ROCK1, and ROCK2 isoforms in cells (Figure 4A). Moreover, each isoform could be depleted by RNAi (Figure 4B). Surprisingly, in undulating channels, only RNAi of ROCK2 led to a significant change in the phenotype of motile cells. More specifically, cells predominantly adopted a mesenchymal mode of migration after ROCK2 RNAi, whereas non-targeting, i.e., control, MYLK, and ROCK1 RNAi cells predominantly adopted an amoeboid mode of migration (Figure 4C–E). Notably, MYLK, ROCK1, and ROCK2 RNAi had no significant effect on the speed or directionality of motile cells (Figure S2C–D). In agreement with these data, cells predominantly adopted a mesenchymal phenotype after treatment with a ROCK2 inhibitor (Belumosudil; 10 μM) (Figure 4F–G). As ROCK2 has been reported to have functions in the cytosol and nucleus (i.e., regulating gene transcription), we determined if upregulating myosin ATPase activity is sufficient to restore an amoeboid phenotype after RNAi. By dephosphorylating the regulatory light chain (RLC), myosin phosphatase 1 (MYPT1) decreases myosin ATPase activity in cells.¹⁷ Therefore, we determined whether MYPT1 RNAi is sufficient to restore an amoeboid phenotype. In contrast to cells with a ROCK2 siRNA alone, cells with a ROCK2 and MYPT1 siRNA (i.e., ROCK2 + MYPT1 RNAi) predominantly adopted an amoeboid phenotype (Figure 4H–J). In 2D confinement, we observed fewer leader mobile (LM) cells after treatment with a ROCK2 siRNA, whereas treatment with a ROCK2 and MYPT1 siRNA (i.e., ROCK2 + MYPT1 RNAi) had the opposite effect (Figure S2B). Morphometric analysis found that cells less frequently adopt a shape consistent with LBBM after ROCK2 RNAi (Figure S2E–G).¹⁸ Thus, ROCK2 biases cells towards an amoeboid phenotype in confined environments by upregulating actomyosin contractility.

Figure 4. ROCK2 promotes amoeboid migration in undulating channels.

(A) RT-qPCR for MYLK, MYLK2, MYLK3, ROCK1, and ROCK2 on total mRNA (mean +/− SEM). (B) 3 days after transfection, RNAi of MYLK, ROCK1, and ROCK2 was confirmed by RT-qPCR on total mRNA (Student’s t-test; mean +/− SEM). (C) Compared to non-targeting (n=135), MYLK (n=146; χ²=NS), and ROCK1 (n=161; χ² ≤ 0.0001), ROCK2 siRNA (n=170; χ² ≤ 0.0001) treated cells were predominantly mesenchymal, whereas the percent of motile cells was not significantly affected by any siRNA. (D) Cells treated with a non-targeting siRNA (n=155) predominantly adopt an amoeboid phenotype. Blebs are observed in the zoomed image (D’). (E) Cells treated with a ROCK2 siRNA (n=170) predominantly adopt a mesenchymal phenotype. No blebs are observed in the zoomed image (E’). (F) Compared to untreated (n=144), cells treated with a ROCK2 inhibitor (Belumosudil; 10 μM) predominantly adopt a mesenchymal phenotype (n=216; χ² ≤ 0.0001). (G) Compared to vehicle, the ratio of pRLC (S19) to RLC was significantly reduced by treatment with a ROCK2 inhibitor (Belumosudil; 10 μM) (2 hr), as measured by Western blotting of total cell lysates (Student’s t-test; mean +/− SEM). GAPDH is used as a loading control. (H) Percent motile and phenotype of cells treated with a non-targeting (n=94), ROCK2 (n=123; χ² ≤ 0.0001), and with both a ROCK2 and MYPT1 siRNA (n=182; χ²=NS). (I) Compared to cells treated with a ROCK2 siRNA (n=170) alone, cells treated with ROCK2 and MYPT1 siRNA (n=223) predominantly adopt an amoeboid phenotype. Blebs are observed in the zoomed image (H’). (J) 3 days after transfection, RNAi of MYPT1 was confirmed by RT-qPCR on total mRNA (Student’s t-test; mean +/− SEM). Undulating channels are coated with VCAM-1 (1 μg/mL). * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Figure S3.

So far, we have demonstrated that Piezo1/Ca²⁺ and ROCK2 bias cells towards an amoeboid phenotype in confined environments. However, we still wondered what factors are directly regulated by Piezo1/Ca²⁺ to promote migration. Using a ROCK2 biosensor (Eevee-ROCK), we could not detect any increase in ROCK2 kinase activity upon activating Piezo1 (Figure 5A & Figure S3A–B).¹⁹ In agreement with these data, Piezo1 RNAi did not have a significant effect on the level of myosin ATPase activity, as measured by the ratio of pRLC (S19) to RLC (Figure 5B). Similarly, the ratio of pRLC (S19) to RLC was unaffected by treatment with a Piezo1 inhibitor or BAPTA, as measured by Western blotting of total cell lysates (Figure S3C–D). Thus, the prototypical regulators of myosin ATPase activity (e.g., MLCK) may not be regulated by Piezo1/Ca²⁺ in this cell type. We determined, therefore, if there are actin cytoskeletal remodeling factors downstream of Piezo1/Ca²⁺ that could bias cells towards an amoeboid phenotype. A phenomenon termed calcium-mediated actin reset or CaAR occurs following the activation of inverted formin-2 (INF2).²⁰ During CaAR, actin is remodeled or “reset” by INF2, which is an atypical formin activated by Ca²⁺ that can be endoplasmic reticulum (ER) localized.^20,21 Possibly due to monomer sequestration, the activation of INF2 by Ca²⁺ leads to a rapid de- and then re-polymerization of actin (i.e., reset) throughout the cell.²⁰ By RT-qPCR, we could detect several formins in melanoma cells, including INF2 (Figure 5C). As measured by LifeAct-mEmerald, we could measure a rapid increase in F-actin near the nucleus after treatment with a Piezo1 activator (Figure 5D–E & Video S4).²² Simultaneously, we could detect a decrease in F-actin elsewhere in cells, such as near the cell edge (Figure 5D–E & Video S4). After INF2 RNAi, however, the increase and decrease in F-actin near the nucleus and cell edge, respectively, after treatment with a Piezo1 activator were blunted (Figure 5D–F & Video S5). Interestingly, the peak [Ca²⁺] elicited by a Piezo1 activator was reduced by ~25% in cells treated with an INF2 siRNA (Figure S3E). However, Ca²⁺ levels in non-targeting and INF2 siRNA-treated cells were equivalent by ~5 min (Figure S3E). These data suggest that INF2 remodels the actin cytoskeleton in response to Piezo1/Ca²⁺.

Figure 5. INF2 remodels the actin cytoskeleton downstream of Piezo1/Ca²⁺.

(A) ROCK1/2 activity is not elevated by treatment with a Piezo1/2 activator (Yoda1; 20 μM), as measured by Eevee-ROCK (n=21; mean +/− SEM). (B) Compared to non-targeting, the ratio of pRLC (S19) to RLC was unaffected by treatment with a Piezo1 siRNA, as measured by Western blotting of total cell lysates (Student’s t-test; mean +/− SEM). GAPDH is used as a loading control. (C) RT-qPCR for mammalian formins on total mRNA (mean +/− SEM). (D) Montages of cells transiently transfected with a non-targeting or INF2 siRNA and LifeAct-mEmerald treated with a Piezo1/2 activator (Yoda1; 20 μM) at 30 sec. (E) Plot of LifeAct-mEmerald fluorescence intensities over time in cells treated with a Piezo1/2 activator (Yoda1; 20 μM) at 60 sec (mean +/− SEM). Fluorescence intensities were measured within regions of interest near the nucleus or cell edge after treatment with a non-targeting or INF2 siRNA. (F) 3 days after transfection, RNAi of INF2 was confirmed by RT-qPCR on total mRNA (Student’s t-test; mean +/− SEM). Cells were plated on fibronectin-coated (10 μg/mL) glass. * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Video S4–5.

Having established that INF2 remodels the actin cytoskeleton in response to Piezo1/Ca²⁺, we wondered if INF2 could bias cells towards an amoeboid phenotype in confined environments. In undulating channels, melanoma cells treated with a non-targeting, i.e., control, siRNA predominantly adopted an amoeboid phenotype. In contrast, cells treated with an INF2 siRNA predominantly adopted a mesenchymal phenotype (Figure 6A–C & Video S6). After INF2 RNAi, cells were also faster and more directional (Figure 6D–E). Similarly, INF2 RNAi led to a significant increase in the number of A549 cells adopting a mesenchymal phenotype in undulating channels (Figure S4A–D). By RT-qPCR, A549 cells were also found to have a similar complement of formins (Figure S4E). In the absence of VCAM-1 (1 μg/mL), however, cells even without INF2 predominantly adopt an amoeboid migration mode (Figure 6F). In 2D confinement, we also did not observe a significant decrease in the number of cells adopting a leader mobile (LM) phenotype after INF2 RNAi (Figure 6G). In agreement with these data, INF2 RNAi did not significantly affect the level of myosin ATPase activity, as measured by pRLC (S19) to RLC (Figure 6H). As INF2 RNAi has no effect in non-adherent environments, INF2 may drive amoeboid migration by promoting de-adhesion (Figure 7A–B). Indeed, in cells treated with a non-targeting siRNA, we could observe a decrease in the fluorescence of Vinculin-mEmerald at focal adhesions after the addition of a Piezo1 activator (Figure 6I). After INF2 RNAi, however, Vinculin levels were unchanged after treatment with a Piezo1 activator (Figure 6I). These results were confirmed by impedance measurements, which increase with the level of cell adhesion, on a real-time cell analyzer. Cells treated with a non-targeting, i.e., control, siRNA de-adhered shortly after treatment with a Piezo1 activator (Figure 7C). In contrast, cells treated with an INF2 siRNA failed to de-adhere (Figure 7C). Freshly plated cells also displayed superior adhesion after INF2 RNAi, as measured by cell impedance (Figure S5A). A unique INF2 siRNA had similar effects; therefore, the observed adhesion phenotype is not likely to be an off-target effect (Figure 7D & Figure S5B–C). INF2 may drive mesenchymal-to-amoeboid transition (MAT) in confined environments by inducing actin cytoskeletal remodeling and de-adhesion. We combined 2D confinement with adhesive (fibronectin; 10 μg/mL) micropatterns to test this idea. When confined to 3 μm, cells treated with a non-targeting, i.e., control, siRNA frequently crossed adhesive patterns using fast amoeboid (leader bleb-based) migration (Figure 7E & Video S7). Despite being confined to 3 μm, cells were often found to adhere to the pattern after INF2 RNAi (Figure 7F & Video S8). Consistent with this observation, we more frequently observed cells treated with an INF2 siRNA randomly migrating into and then adhering to the pattern (Figure 7G). Additionally, INF2 siRNA-treated cells were less likely to de-adhere from the pattern and then migrate away (Figure 7G). In agreement with these data, cells treated with an INF2 siRNA had larger focal adhesions, spread area, and stress fibers in the presence of a uniform concentration of fibronectin (10 μg/mL) (Figure S5D–H). Collectively, these data demonstrate that cells require the activation of INF2 by Piezo1/Ca²⁺ for migration through confined environments with varying levels of an adhesive ligand. Based on transcriptomic data from The Cancer Genome Atlas (TCGA), Piezo1 and INF2 levels are significantly elevated in metastatic tumors across multiple solid cancers (Figure 7H & Figure S5I). Additionally, survival analyses revealed that high levels of Piezo1 and INF2 mRNA correlate with increased risk in patients for multiple solid cancers (Figure 7I & Figure S5J). However, a high level of INF2 mRNA correlated with the largest increase in risk with a hazard ratio (HR) of 1.378. Thus, the capacity for switching between mesenchymal and amoeboid migration modes likely promotes the dissemination of melanoma tumors.

Figure 6. INF2 promotes amoeboid migration in undulating channels downstream of Piezo1/Ca²⁺.

(A) Representative image of F-actin, marked by LifeAct-mEmerald, of a cell transiently transfected with a non-targeting siRNA. Blebs are observed in the zoomed image. (B) Representative image of F-actin, marked by LifeAct-mEmerald, of a cell transiently transfected with a INF2 siRNA. No blebs can be observed in the zoomed image. (C) Compared to non-targeting (n=155), INF2 siRNA (n=250) treated cells were predominantly mesenchymal (χ² ≤ 0.0001). (D) Compared to non-targeting, INF2 siRNA-treated cells are significantly faster (Student’s t-test; mean +/− SEM). (E) Compared to non-targeting (n=51), INF2 siRNA (n=192) treated cells migrate more directionally (mean +/− SEM). (F) In undulating channels coated with BSA (1%), which blocks adhesion, cells treated with an INF2 siRNA (n=107) predominantly adopt an amoeboid phenotype. (G) Percent of non-targeting or INF2 siRNA treated cells adopting a leader mobile (LM), leader non-mobile (LNM), or no leader (NL) phenotype (χ²=NS). Cells were vertically confined down to a height of 3 μm. PDMS was coated with bovine serum albumin (BSA; 1%). (H) Compared to non-targeting, the ratio of pRLC (S19) to RLC was unaffected by treatment with a INF2 siRNA, as measured by Western blotting of total cell lysates (Student’s t-test; mean +/− SEM). GAPDH is used as a loading control. (I) Compared to non-targeting, INF2 siRNA treated cells plated on VCAM-1 (1 μg/mL) coated glass do not display a decrease in Vinculin-mEmerald after treatment with a Piezo1/2 activator (Yoda1; 20 μM) at 60 sec (mean +/− SEM). Vinculin-mEmerald was plotted as a fluorescence ratio at focal adhesions to an uninvolved region. Undulating channels are coated with VCAM-1 (1 μg/mL). * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Figure S4 & Video S6.

Figure 7. INF2 drives amoeboid migration by promoting de-adhesion.

(A) Summary of phenotypic data for cells in undulating channels. (B) Summary of phenotypic data for cells confined down to 3 μm by a PDMS ceiling coated with BSA (1%). (C and D) Compared to non-targeting, INF2 siRNA-treated cells fail to de-adhere from VCAM-1 (1 μg/mL) coated cover glass after treatment with a Piezo1/2 activator (Yoda1; 20 μM) at 15 min (mean +/− SEM). Cell impedance was measured using a real-time cell analyzer (RTCA). (E) Temporal color-code image (8 min between color-coded images) of a non-targeting siRNA-treated cell migrating across an adhesive (fibronectin; 10 μg/mL) micropattern. Cells were confined down to 3 μm using a dynamic cell confiner. (F) INF2 siRNA (green) treated cells adhere to adhesive (fibronectin; 10 μg/mL) micropatterns (magenta). Cells were confined down to 3 μm using a dynamic cell confiner. (G) Percent of cells that migrate across patterns (circles), randomly migrate into and then adhere to the pattern (squares), and that de-adheres from the pattern and then migrate away (triangles) for non-targeting (n=247) and INF2 (n=249) siRNA treated cells. The ‘adheres to’ phenotype is shown in (F). A p-value comparing non-targeting to INF2 siRNA-treated cells was calculated using a one-way ANOVA. (H) mRNA level (transcripts per million; TPM) of INF2 in primary (n=9,623) vs. metastatic (n=394) tumors (Mann-Whitney test; median +/− quartiles). TCGA datasets for multiple solid cancers were sourced from the GDC. (I) Survival curves for patients with low (n=4,717) vs. high (n=4,704) INF2 mRNA levels (log-rank test). The median mRNA level (TPM) was used as the cut-off value. TCGA datasets for multiple solid cancers were sourced from the GDC. * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001. See also Figure S5 & Video S7–8.

Discussion

Disseminating tumor cells must adapt to various mechanochemical environments before establishing metastatic sites. Similarly, immune cells need to adjust to different mechanochemical conditions to effectively defend against pathogens. Both immune and cancer cells have been shown to adopt amoeboid modes of migration under conditions of low adhesion and confinement.^2,23 Among cancer cell types, melanoma represents the best studied in terms of amoeboid migration.²⁴ Notably, human melanoma cells have been observed migrating in the direction of a large bleb within tumors implanted in immunocompromised mice by intravital imaging.⁸ Moreover, in sectioned tumors from melanoma patients, predominantly amoeboid cells are observed at invasive fronts.²⁵ Consequently, amoeboid migrating cells likely facilitate the spread of melanoma tumors.

Although candidate factors have been identified, it remains unclear what drives amoeboid migration.^10–12 In a Piezo1-dependent fashion, we show that intracellular [Ca²⁺] increases as cells are vertically confined. Similarly, inhibiting or activating Piezo1 led to a decrease and increase in leader bleb-based migration (LBBM), respectively. Consistent with its effects on LBBM, Piezo1 activity also correlated with bleb size. In zebrafish embryos, cytosolic phospholipase A2 (cPLA2) was shown to promote a phenotypically similar migration mode.¹² Here, we could not detect any difference in the rate of LBBM after cPLA2 RNAi. In transmigration assays, however, we previously reported a defect in migration through confining pores after cPLA2 RNAi.²⁶ Thus, cPLA2 may be specifically required for chemotactic migration through confining pores. In contrast, ROCK2 was identified as essential for LBBM, which is consistent with work by others on amoeboid migrating cells at the invasive fronts of melanoma tumors.²⁷

Cells predominantly adopted a hybrid phenotype in microchannels coated with VCAM-1. In contrast to cells confined in 2D, which form leader blebs, cells in VCAM-1 coated microchannels form F-actin and intracellular pressure-driven protrusions (i.e., blebs). Like what was found for cells confined in 2D, however, the appearance of blebs was promoted by Piezo1/Ca²⁺. Thus, in 2D and 3D (i.e., microchannels) confined environments, Piezo1 is gated open by plasma membrane tension. In undulating channels, which are more likely to resemble the tortuous vasculature surrounding tumors, cells predominantly adopted an amoeboid phenotype. These data are consistent with a curvature-induced increase in plasma membrane tension. In support of this idea, more cells adopted a mesenchymal phenotype after Piezo1 RNAi. Compared to Piezo1 RNAi, however, the chelation of Ca²⁺ with BAPTA had a much more potent effect. Therefore, other Ca²⁺ selective channels may contribute to this phenomenon. Recently, a mechanosensitive transient receptor potential (TRP) channel was shown to promote melanoma metastasis.²⁸

In cells, Ca²⁺ regulates several signaling pathways. For instance, myosin light chain kinase (MYLK) is activated by Ca²⁺/calmodulin.²⁹ To our surprise, RNAi of MYLK had no effect on amoeboid migration. In contrast, we identified ROCK2 as essential for amoeboid migration. Therefore, we wondered if Piezo1/Ca²⁺ could activate ROCK2. Using a biosensor for ROCK (Eevee-ROCK), we could not detect any increase in ROCK2 kinase activity upon activating Piezo1.¹⁹ Accordingly, we looked to actin cytoskeletal remodeling proteins as candidate factors downstream of Piezo1/Ca²⁺. This led us to evaluate the role of inverted formin-2 (INF2), which was shown to promote calcium-mediated actin reset or CaAR.²⁰ In undulating channels, cells predominantly adopted a mesenchymal phenotype after INF2 RNAi. However, in the absence of VCAM-1, cells still adopted an amoeboid phenotype after INF2 RNAi. As INF2 RNAi had no effect in non-adherent environments, INF2 may drive amoeboid migration by promoting de-adhesion. Indeed, we found that Piezo1/Ca²⁺ induced de-adhesion, whereas cells failed to de-adhere after INF2 RNAi. Moreover, actin cytoskeletal remodeling in response to Piezo1/Ca²⁺ was abolished in cells after RNAi of INF2. Therefore, remodeling of the actin cytoskeleton by INF2, which redistributes myosin-generated tension, was found to promote focal adhesion disassembly. As INF2 RNAi led to a decrease in the peak level of intracellular Ca²⁺ elicited by a Piezo1/2 activator, tension redistribution within the actin cytoskeleton following INF2 activation may also promote the opening of mechanosensitive channels. The formation of blebs generates additional sites of plasma membrane tension to activate Piezo1. Thus, intracellular Ca²⁺ levels remain elevated in confined (i.e., blebbing) cells to maintain INF2 activity. Notably, INF2 RNAi cells, which adopt a mesenchymal phenotype, migrated faster and more directionally. However, the probability that a cell will switch between a mesenchymal and amoeboid phenotype will likely depend on its specific complement of formins, which compete for G-actin.

By promoting de-adhesion, INF2 may facilitate metastasis by providing cancer cells with the ability to transition between mesenchymal and amoeboid migration modes. More specifically, INF2 may promote the dissemination of tumor cells through heterogeneous tissues with low and highly adhesive microenvironments, as INF2 was found to facilitate the migration of melanoma cells across micropatterned surfaces. In agreement with this concept, survival data for patients with solid cancers showed that high levels of INF2 correlated with increased risk. In addition to INF2, it is likely that a suite of Ca²⁺ responsive factors regulate MAT in cell type and mechanochemical environment specific fashions. Thus, we plan to work towards identifying additional Ca²⁺ responsive factors required for MAT in diverse cell types.

Resource Availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jeremy S. Logue (loguej@amc.edu).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact upon request.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines

A375-M2 (CRL-3223) and A549 (CCL-185) were obtained from the American Type Culture Collection (Manassas, VA). Cells were cultured in high-glucose DMEM supplemented with fetal bovine serum (Sigma Aldrich; 12106C), L-glutamine (Thermo Fisher; 35050061), pyruvate, HEPES, and Antibiotic-Antimycotic (Thermo Fisher; 15240096).

METHOD DETAILS

Plasmids

LifeAct-mEmerald (plasmid no. 54148) and GCaMP6s (plasmid no. 40753) were obtained from Addgene (Watertown, MA). Eevee-ROCK was kindly provided by Dr. Michiyuki Matsuda (Kyoto University). 1 μg of plasmid was used to transfect 750,000 cells in each well of a 6-well plate using Lipofectamine 2000 (5 μL; Thermo Fisher) in OptiMEM (400 μL; Thermo Fisher). After 20 min at room temperature, plasmid in Lipofectamine 2000/OptiMEM was incubated with cells in complete media (2 mL) overnight.

Chemical treatments

BAPTA (2786), GsMTx4 (Piezo1/2 inhibitor; 4912), and Yoda1 (Piezo1/2 activator; 5586) were purchased from Tocris Bioscience (Bristol, UK). Before confinement, cells were treated with drug for 1 hr. In parallel, confining devices were incubated with drug in complete media for at least 1 hr before loading cells. For measurements of intracellular [Ca²⁺], cells were loaded with both a green fluorescent dye (Thermo Fisher; C7025) and a red fluorescent Ca²⁺ indicator, Calbryte (Fisher Scientific; NC2111763), for ratiometric fluorescence imaging.

RT-qPCR

Total RNA was isolated from cells using the PureLink RNA Mini Kit (Thermo Fisher; 12183018A) and was used for reverse transcription using a High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher; 4368814). qPCR was performed using PowerUp SYBR Green Master Mix (Thermo Fisher; A25742) on a real-time PCR detection system (CFX96; Bio-Rad). Relative mRNA levels were calculated using the ΔCt method.

RNA interference

Non-targeting (4390844), Piezo1 (4392420; s18891), cPLA2 (4390824; s10592), MYLK (4392420; s533772), ROCK1 (4390824; s12097), ROCK2 (4390824; s18161), MYPT1 (4390824; s9235), and INF2 (4392420; s230622) (4392420; s34736) siRNAs were purchased from Thermo Fisher (Waltham, MA). All siRNA transfections were performed using RNAiMAX (5 μL; Thermo Fisher) and OptiMEM (400 μL; Thermo Fisher). 200,000 cells were trypsinized and seeded in 6-well plates in complete media. After cells adhered (~1 hr), siRNAs in RNAiMAX/OptiMEM were added to cells in complete media (2 mL) at a final concentration of 50 nM. Cells were incubated with siRNAs for 3 days.

Western blotting

Lysates were prepared on ice using a rapid lysate preparation kit (Cytoskeleton, Denver, CO; BLR01) in the presence of protease and phosphatase inhibitors (Cell Signaling Technology; 5872S). Samples were separated on gradient gels, transferred to nitrocellulose, and proteins were immobilized by air drying overnight before blocking (LI-COR, Lincoln, NE; 927-66003). Primary antibodies for pRLC (3671), RLC (8505), and GAPDH (5174) were purchased from Cell Signaling Technology (Danvers, MA) and diluted (1:500) in Tris-buffered saline containing 0.1% Tween 20 (TBS-T). Bands were resolved using infrared (IR) dye conjugated secondary antibodies on an Odyssey imager (LI-COR).

Immunofluorescence

Cells on fibronectin-coated (10 μg/mL) glass were fixed with paraformaldehyde (PFA; 4%) in HEPES buffered saline (HBS) for 20 min at room temperature (RT). HBS containing 0.1% Triton X-100 was used for ≥5 min at RT for permeabilization. HBS containing 1% BSA, 1% fish gelatin (Sigma Aldrich; G7765), and 5 mM EDTA was used for ≥1 hr at RT for blocking. Anti-pRLC antibody (Cell Signaling Technology; 3671) was used at 1:250 in blocking buffer overnight at 4 °C. An Alexa Fluor 488 conjugated secondary antibody (1:400) (Thermo Fisher; A-21206) and Alexa Fluor 594 conjugated phalloidin (1:200) (Thermo Fisher; A12381) were used in blocking buffer for ≥2 hr at RT. Cells were rocked in HBS for ≥5 min at RT for each wash before mounting in ProLong Glass containing Hoechst 33342 for staining nuclei (Thermo Fisher; P36983).

2D confinement

This protocol has been described in detail elsewhere.¹⁵ Briefly, PDMS (24236-10) was purchased from Electron Microscopy Sciences (Hatfield, PA). 2 mL was cured overnight at 37 °C in each well of a 6-well glass bottom plate (Cellvis; P06-1.5H-N). Using a biopsy punch (World Precision Instruments; 504535), an 8 mm hole was cut, and 3 mL of serum-free media containing 1% BSA was added to each well and incubated overnight at 37 °C. After removing the serum-free media containing 1% BSA, 300 μL of complete media containing trypsinized cells (250,000 to 1 million) and 2 μL of 3.11 μm beads (Bangs Laboratories, Fishers, IN; PS05002) were then pipetted into the round opening. The vacuum created by briefly lifting one side of the hole with a 1 mL pipette tip was used to move cells and beads underneath the PDMS. Finally, 3 mL of complete media was added to each well. Cells recovered for ~60 min before imaging.

Microchannel preparation

PDMS (Electron Microscopy Sciences; 24236-10) was prepared using a 1:7 base and curing agent ratio. Uncured PDMS was poured over the wafer mold, placed in a vacuum chamber to remove bubbles, moved to a 37 °C incubator, and left to cure overnight. After curing, small PDMS slabs with microchannels were cut using a scalpel, whereas cell loading ports were cut using a 0.4 cm hole punch (Fisher Scientific; 12-460-409).

For making PDMS-coated cover glass (Fisher Scientific; 12-545-81), 30 μL of uncured PDMS was pipetted into the center of the cover glass, placed in a modified mini-centrifuge, and spun for 30 sec for even spreading. The PDMS-coated cover glass was cured for at least 1 hr on a 95 °C hot plate. To bond the slab and coated cover glass, PDMS surfaces were activated for ~1 min by plasma treatment (Harrick Plasma; PDC-32G). The apparatus was incubated at 37 °C for at least 1 hr for complete bonding.

Microchannel coating

Before microchannel coating, surfaces were first activated by plasma treatment. VCAM-1 (R&D Systems; 862-VC) or BSA (VWR; VWRV0332) was used for coating at 1 μg/mL and 1%, respectively, in PBS. Immediately after plasma treatment, VCAM-1 or BSA solution was pumped into microchannels using a modified motorized pipette. To remove any bubbles pumped into microchannels, the apparatus was left to coat in a vacuum chamber for at least 1 hr. Microchannels were then rinsed by repeatedly pumping in new PBS. Finally, microchannels were incubated overnight at 4 °C in complete media before use.

Microchannel loading

Before cells were loaded into microchannels, complete media was aspirated, and microchannels were placed into an interchangeable cover glass dish (Bioptechs; 190310-35). Freshly trypsinized cells in 300 μL of complete media, stained with 1 μL of fluorescent membrane dye (Thermo Fisher; C10046), were pumped into microchannels using a modified motorized pipette. Once at least 20 cells were observed in microchannels by low magnification brightfield microscopy, microchannels were covered with 2 mL of complete media. Before imaging, a lid was placed on the apparatus to prevent evaporation.

Live imaging

High-resolution live imaging was performed using a DeltaVision (Issaquah, WA) Elite imaging system mounted on an Olympus (Tokyo, Japan) IX71 stand with a motorized stage (XYZ), Ultimate Focus, solid state light source, fast filter wheel for DAPI, CFP, FITC, GFP, YFP, TRITC, mCherry, and Cy5 channels, critical illumination, Olympus UPlanSApo 20X/0.75 NA DIC (air), UPlanSApo 30X/1.05 NA DIC (silicone), PlanApo N 60X/1.42 NA DIC (oil), UPlanSApo 60X/1.30 NA DIC (silicone), and UPlanSApo 100X/1.40 NA DIC (oil) objectives, Photometrics (Tucson, AZ) CoolSNAP HQ2 camera, SoftWoRx (Preston, UK) software with constrained iterative deconvolution, cage incubator, and vibration isolation table.

Cell impedance

Cell impedance was measured using a real-time cell analyzer (RTCA; Agilent). Before adding 30,000 cells to each well, plates (Agilent; 300601140) were coated with VCAM-1 (1 μg/mL) (R&D Systems; 862-VC) overnight at 4 °C.

2D confinement + adhesive micropatterns

Micropatterning was performed per the manufacturer protocol (4Dcell; Paris, France). Briefly, coverslips were first coated with non-adhesive PLL-g-PEG. Lines were patterned on coverslips using a photomask (4Dcell; UM006) and deep UV treatment. The coverslips were then incubated with fibronectin (10 μg/ml) (Thermo Fisher; 33016015) for 1 hr at room temperature. A small amount of red fluorescent BSA (1 μg/mL) (Thermo Fisher; A13101) was added to the fibronectin solution to mark the adhesive patterns. Cells were confined to 3 μm using a dynamic cell confiner according to the manufacturer protocol (4Dcell). Because cells were found to be less motile in the dynamic cell confiner, all cells were treated with siRNA towards MYPT1 to increase motility rates generally.

Genomics

Transcriptomic data for Piezo1 and INF2 were obtained by The Cancer Genome Atlas (TCGA; https://www.cancer.gov/ccg/research/genome-sequencing/tcga). Data was sourced from the Genomic Data Commons (GDC; https://portal.gdc.cancer.gov/) and accessed through cBioPortal (https://www.cbioportal.org/). For survival curves, median mRNA levels (transcripts per million; TPM) for Piezo1 and INF2 were used as cut-off values. Amalgamated data from adrenocortical carcinoma, cholangiocarcinoma, bladder urothelial carcinoma, colon adenocarcinoma, rectal adenocarcinoma, invasive breast carcinoma, diffuse glioma, glioblastoma multiforme, miscellaneous neuroepithelial tumor, cervical squamous cell carcinoma, esophageal adenocarcinoma, stomach adenocarcinoma, uveal melanoma, head and neck squamous cell carcinoma, renal clear cell carcinoma, chromophobe renal cell carcinoma, papillary renal cell carcinoma, hepatocellular carcinoma, lung adenocarcinoma, high-grade serous ovarian cancer, pancreatic adenocarcinoma, pleural mesothelioma, prostate adenocarcinoma, cutaneous melanoma, soft tissue cancer, non-seminomatous germ cell tumor, thymoma, papillary thyroid cancer, endometrial cancer, and uterine carcinosarcoma was used in each survival analysis.

QUANTIFICATION AND STATISTICAL ANALYSIS

Cell classification in 2D confinement

Cells that displayed directionally persistent migration over at least 4 frames (32 min) were classified as leader mobile (LM). Any cell with a large bleb that remained stable for at least 4 frames (32 min) was considered to have a leader bleb.

Cell classification in microchannels

Cells that moved at least ½ of their original length over a 5 hr timelapse movie were considered motile. Cells that displayed only blebs were classified as amoeboid, whereas cells that displayed only actin-based protrusions were classified as mesenchymal. Cells with blebs and actin-based protrusions were classified as hybrid.

Morphometric analysis

The Fiji (https://fiji.sc/) plugin, Analyze_Blebs (https://github.com/karlvosatka/analyze_blebs), was used to measure the largest bleb area, aspect ratio, roundness, Feret’s diameter, solidity, and circularity of cells from timelapse movies.¹⁸

Cell migration

To perform cell speed and directionality ratio analyses, we used an Excel (Microsoft) plugin, DiPer, developed by Gorelik and colleagues and the Fiji plugin (https://fiji.sc/), MTrackJ, created by Erik Meijering for manual tracking.³⁰ Brightfield imaging confirmed that beads or debris were not obstructing the cell.

Calcium dynamics

Root mean square (RMS) fluorescence values for GCaMP6s were calculated from a 3 × 3 μm square region of interest (ROI) in the cytoplasm of cells in microchannels. Time-lapse images were taken every 5 sec for 30 min.

FRET

Ratio images of FRET (CFP excitation/YFP emission) to CFP (CFP excitation/CFP emission) were generated and analyzed in Fiji (https://fiji.sc/).

Statistics

Sample sizes were determined empirically and based on saturation. As noted in each figure legend, statistical significance was determined by either a two-tailed Student’s t-test or one-way ANOVA followed by a multiple comparison test in GraphPad (San Diego, CA) Prism 7. Outliers were identified using the ROUT method (Q = 1%). Normality was determined by a D’Agostino & Pearson test. For fold change, a one-sample Student’s t-test was used to determine statistical significance (hypothetical value = 1). For categorical data, χ² tests were used to determine statistical significance. * - p ≤ 0.05, ** - p ≤ 0.01, *** - p ≤ 0.001, and **** - p ≤ 0.0001

Supplementary Material

2. Video S1. Leader mobile (LM) cell. Related to Figure 1.

Confined down to 3 μm. Cells were stained with a fluorescent membrane dye. PDMS was coated with bovine serum albumin (BSA; 1%).

3. Video S2. A leader non-mobile (LNM) cell. Related to Figure 1.

Confined down to 3 μm. Cells were stained with a fluorescent membrane dye. PDMS was coated with BSA (1%).

4. Video S3. Microchannel phenotypes. Related to Figure 2.

A motile cell adopting a hybrid phenotype in a smooth channel (left). A motile cell adopting an amoeboid phenotype in an undulating channel (right). Cells were stained with a fluorescent membrane dye. Microchannels are coated with VCAM-1 (1 μg/mL).

5. Video S4. The activation of Piezo1 induces actin cytoskeletal remodeling. Related to Figure 5.

Cell transiently transfected with a non-targeting siRNA and LifeAct-mEmerald treated with a Piezo1/2 activator (Yoda1; 20 μM). Cells were plated on fibronectin-coated (10 μg/mL) glass.

6. Video S5. Actin cytoskeletal remodeling is mediated by INF2. Related to Figure 5.

Cell transiently transfected with an INF2 siRNA and LifeAct-mEmerald treated with a Piezo1/2 activator (Yoda1; 20 μM). Cells were plated on fibronectin-coated (10 μg/mL) glass.

7. Video S6. Amoeboid migration in undulating channels requires INF2. Related to Figure 6.

An INF2 siRNA-treated cell adopting a mesenchymal migration mode in an undulating channel. Cells were stained with a fluorescent membrane dye. Microchannels are coated with VCAM-1 (1 μg/mL).

8. Video S7. Amoeboid migrating cells can cross adhesive micropatterns. Related to Figure 7.

A non-targeting siRNA-treated cell (green) migrating across an adhesive (fibronectin; 10 μg/mL) micropattern (magenta). The cell was stained with a green fluorescent dye. Cells were confined down to 3 μm using a dynamic cell confiner.

9. Video S8. INF2 is required to de-adhere from adhesive micropatterns. Related to Figure 7.

INF2 siRNA-treated cells (green) adhere to the adhesive (fibronectin; 10 μg/mL) micropattern (magenta). The cells were stained with a green fluorescent dye. Cells were confined down to 3 μm using a dynamic cell confiner.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-pRLC	Cell Signaling Technology	3671
Anti-RLC	Cell Signaling Technology	8505
Anti-GAPDH	Cell Signaling Technology	5174
Chemicals, peptides, and recombinant proteins
Blocking buffer	LI-COR	927-66003
Membrane stain, deep red fluorescent	Thermo Fisher	C10046
BAPTA	Tocris	2786
Piezo1/2 inhibitor (GsMTx4)	Tocris	4917
Piezo1/2 activator (Yoda1)	Tocris	5586
PDMS	Electron Microscopy Sciences	24236-10
~3 μm polystyrene beads	Bangs Laboratories	PS05002
VCAM-1	R&D Systems	862-VC
Fibronectin	Thermo Fisher	33016015
BSA, red fluorescent	Thermo Fisher	A13101
Ca2+ indicator, deep red fluorescent	Fisher Scientific	NC2111763
Cytosolic dye, green fluorescent	Thermo Fisher	C7025
Non-targeting siRNA	Thermo Fisher	4390844
Human Piezo1 siRNA	Thermo Fisher	4392422; s18891
Human PLA2G4A siRNA	Thermo Fisher	4390824; s10592
Human MYLK siRNA	Thermo Fisher	4392420; s533772
Human ROCK1 siRNA	Thermo Fisher	4390824; s12097
Human ROCK2 siRNA	Thermo Fisher	4390824; s18161
Human MYPT1 siRNA	Thermo Fisher	4390824; s9235
Human INF2 siRNA	Thermo Fisher	4392420; s230622
Human INF2 siRNA #2	Thermo Fisher	4392420; s34736
Experimental models: Cell lines
A375-M2	ATCC	CRL-3223
A549	ATCC	CCL-185
Oligonucleotides
Human Piezo1 forward primer: CACCAACCTCATCAGCGACT	Thermo Fisher	N/A
Human Piezo2 forward primer: AACACCATCTACAGACTGGCCC	Thermo Fisher	N/A
Human PLA2G4A forward primer: GGATTCTCTGGTGTGATGAAGGC	Thermo Fisher	N/A
Human MYLK forward primer: GAGGTGCTTCAGAATGAGGACG	Thermo Fisher	N/A
Human MYLK2 forward primer: GCTGTATGCAGCCATCGAGACT	Thermo Fisher	N/A
Human MYLK3 forward primer: GCTGAAGCAGTAAGGAGGATGC	Thermo Fisher	N/A
Human ROCK1 forward primer: GAAACAGTGTTCCATGCTAGACG	Thermo Fisher	N/A
Human ROCK2 forward primer: TGCGGTCACAACTCCAAGCCTT	Thermo Fisher	N/A
Human MYPT1 forward primer: GCAGGTGTTACACGTTCAGCTTC	Thermo Fisher	N/A
Human INF2 forward primer: GCAGGTGTTACACGTTCAGCTTC	Thermo Fisher	N/A
Recombinant DNA
Plasmid: LifeAct-mEmerald	Addgene	54148
Plasmid: GCaMP6s	Addgene	40753
Plasmid: Eevee-ROCK	Dr. Michiyuki Matsuda (Kyoto University)	N/A
Plasmid: Vinculin-mEmerald	Addgene	54304
Software and algorithms
Fiji	https://imagej.net/Fiji	N/A
Prism version 7	GraphPad	N/A
cBioPortal	https://www.cbioportal.org/	N/A
BioRender	https://www.biorender.com/	N/A
Other
Odyssey imager	LI-COR	N/A
Interchangeable cover-glass dish	Bioptechs	190310-35
Basic Plasma Cleaner	Harrick Plasma	PDC-32G
DeltaVision Elite	GE	N/A
Real-time cell analyzer (RTCA)	Agilent	N/A
Micropatterning kit	4Dcell	N/A
Dynamic cell confiner kit	4Dcell	N/A

Highlights:

• Piezo1/Ca²⁺ increases in invasive melanoma cells with the degree of confinement.

• Piezo1/Ca²⁺ activates INF2 to induce actin cytoskeletal remodeling and de-adhesion.

• The activation of INF2 promotes MAT to aid migration through confined environments.

• High levels of Piezo1 and INF2 correlate with increased risk in cancer patients.