Mechanical worrying drives cell migration in crowded environments 1

1Department of Bioinformatics, University of Texas Southwestern Medical Center, Dallas, TX 5 2Department of Cell Biology, University of Texas Southwestern Medical Center, Dallas, TX 6 3Institut Curie, PSL Research University, CNRS, Paris, France. 7 4Department of Biochemistry, School of Medicine, University of Utah, Salt Lake City, UT 8 *contributed equally 9 §Corresponding author, Gaudenz.Danuser@UTSouthwestern.edu, Reto.Fiolka@UTSouthwestern.edu 10 11


26
Cell migration is critical to processes ranging from embryogenesis and wound healing to cancer 27 metastasis. 1 When spatially confined, animal and non-animal cells alike exhibit bleb-based motility, a type 28 of amoeboid migration characterized by weak adhesion and minimal proteolytic destruction of the 29 surrounding matrix. 7,8 Amoeboid cells can migrate through tight spaces by deforming their body and 30 nucleus, even to the point of nucleus rupture. 6,9,10 Studies of invasive cancer cells, particularly of 31 melanoma, have reported that rounded amoeboid cells are enriched at the tumor edge, 11 and can move 32 through Matrigel and tumor xenografts. 2,12,13 Indeed, the pro-migratory effect of intracellular contractility, 33 which is associated with rounding and surface blebbing, is well known. 11,13-17 The pronounced rounded 34 morphology of these cells seems to be at odds with a migration mode that relies on shape deformation. 35 Round, amoeboid cells have been observed to move in tunnels. 4 Although under some conditions 36 amoeboid cells secrete matrix metalloproteinases (MMPs), 18 amoeboid migration is not usually 37 considered a proteolytic migration mode, and so it has been generally assumed that these tunnels were 38 pre-formed by 'helper' cells moving in a mesenchymal manner, a migration mode in which the 39 extracellular matrix (ECM) is degraded via MMPs. 4 Despite the notion that amoeboid cells cannot remodel 40 their environment to generate their own paths and thus would be forced to immobility in a very dense 41 microenvironment, blebbing cells in soft extracellular matrices have been observed to physically 42 manipulate fibers, pushing them out of their way 19 and tugging on them with adhesions, 20 suggesting the 43 possibility of an amoeboid migration mode enabled by matrix remodeling. 44 To investigate the mechanism of this potentially new path generating migration strategy, we created 45 dense, yet pliable in vivo mimetics by encapsulating cells in fibrous Collagen 1 gels with bulk stiffnesses 46 on the order of 1 kPa (Advanced biomatrix.com). 21, 22 We imaged cells in 3D with high-resolution light-47 sheet microscopy at near isotropic ~350 nm resolution, using specially designed sample chambers that do 48 not interfere with the mechanical properties of the mimetics (Fig. 1a, Movie 1). 23,24 Adsorption of collagen 49 to hard surfaces increases collagen stiffness near the surface, 25-28 which shifts and eventually diminishes 50 the rounded, blebbing morphotype. To observe migration of these cells without mechanical interference, 51 our chambers enabled cell imaging at greater than 1 mm away from any stiff surfaces. 52 To determine whether our in vivo mimetics prohibited deformation-based migration through pores, we 53 fluorescently labeled collagen and then measured collagen pore size. 24,29,30 Although blebs were small 54 enough to fit inside the pores in the collagen network, the nucleus and cell body of amoeboid melanoma 55 cells were too large to fit through the existing pores ( Fig. 1a,b, Movie 2), thus rendering a deformation-56 based mode of migration unlikely. Long-term time-lapse imaging of melanoma cells confirmed that 57 amoeboid cells were nevertheless able to move through the mimetics while maintaining their largely 58 spherical shape (Fig. 1c, Extended Data Fig. 1a,b). 59 As a model cell for bleb-based migration through soft crowded environments, we chose metastatic 60 melanoma cells. In vivo, melanoma metastasizes to soft environment such as the brain. 31 Consistent with 61 these clinical observations, we noted that melanoma in the soft environment of the zebrafish hindbrain 62 not only have an amoeboid morphology but bleb extensively (Extended Data Fig. 1c). We next tested 63 whether the amoeboid morphology was associated with melanoma metastatic potential within our in vivo 64 mimetic. We imaged populations of primary melanoma cells that were harvested from patients and then 65 passaged in a mouse xenotransplantation system. 32,33 Comparing three mechanically distinct collagen 66 microenvironments, including the mimetic, we found that samples with higher metastatic efficiency were 67 enriched in the amoeboid morphology compared to the stretched mesenchymal morphology (Figure 1d, 68 Extended Data Figure 1d,e). Furthermore, using unbiased cell shape motif detection, 34 we discovered that 69 a parental melanoma cell line (A375P) had a lower average bleb count than a subpopulation of the cell 70 line that had been enriched for metastatic potential in mouse xenografts (A375M2) (Fig. 1e,f). 35 Together, 71 these results establish the significance of our experimental system as a model of metastatic cell migration 72 in soft, ECM-dense tissues. 73 Migrating amoeboid cells can carve a path without the need for extracellular proteolytic degradation. 74 To determine how blebbing melanoma cells migrate through soft, dense collagen, we imaged them 24 75 hours after seeding. Over this time frame, many cells had created tunnels (Fig. 1g). We observed a similar 76 tunneling phenomenon with a different melanoma cell line (Extended Data Fig. 2a), as well as with 77 pediatric Ewing sarcoma cells (Extended Data Fig. 2b). Tunnel creation in dense matrices is usually ascribed 78 to matrix metalloproteinase (MMP)-dependent mesenchymal migration. 3,36 Thus, our finding of an 79 amoeboid cell morphology associated with a clearly demarcated, cell-generated path seemed paradoxical 80 in view of the current paradigms of cancer cell motility. To begin to solve this puzzle, we applied a broad 81 spectrum MMP inhibitor, GM6001, 37,38 and found no effect on the ability of melanoma cells to tunnel (Fig.  82 1h). The MMP-independence of the migration mode also held when collagen was not first solubilized 83 using pepsin, which removes a collagen crosslinking site potentially rendering the collagen easier to 84 digest. 39 Although tunneling is somewhat less frequent in unpepsinized collagen, the pore size is greater 85

107
as is the extent of directional collagen realignment by the cells, suggesting a reduced need to tunnel 108 (Extended Data Fig. 3). We confirmed that GM6001 inhibited MMP activity by measuring the abundance 109 of an antibody that recognizes the MMP-cleaved collagen site in tunnels (Fig. 1i,j) and by direct 110 measurement of MMP enzymatic activity on a synthetic substrate (Extended Data Fig. 2c). Thus, tunneling 111 is not mediated by the enzymatic activity of MMPs. 112

Path generation is mediated by bleb-driven ablation of the extracellular matrix (ECM). 113
To address alternative path generation mechanisms to protease-activity, we analyzed the interactions of 114 cells with collagen in greater detail. Cells inside tunnels were often highly polarized, with many large blebs 115 at the cell front facing the enclosed end of the tunnel (Fig. 2a, Movies 3,4). Measuring the difference in 116 frequencies of protrusive and retractive motion on and off blebs, we found that blebs were on average 117 protrusive and non-blebs retractive (Fig. 2b), suggesting that blebs are responsible for cell protrusion 118 through collagen. We next measured the colocalization of blebs with collagen, finding that collagen was 119 enriched in regions near blebs, but not directly on blebs (Fig. 2c). This is explained by bleb interdigitation 120

148
into pores in the collagen network, resulting in high collagen fiber density at the base of blebs (Fig. 2d). 149 Then we examined collagen motion, which showed movement of individual collagen fibers at the front of 150 tunneling cells (Fig. 2e,f). Using a 3D optical flow algorithm designed to capture multi-scale motion both 151 near and away from cells (Fig. 2g), 40 we compared the collagen speed near blebs with the bleb speed for 152 both protruding and retracting blebs (Fig. 2h). For protrusive blebs, at low bleb speeds collagen speed 153 increased linearly with bleb speed, whereas at high bleb speeds collagen motion plateaued, consistent 154 with bleb interdigitation into collagen pores. For retracting blebs we found that even at high bleb speed, 155 collagen was pulled in concert with the blebs, meaning that retracting blebs pull collagen towards the cell 156 surface. Indeed, at the fronts of highly polarized cells, collagen was often enriched into a shell at the cell 157 periphery alongside extensive internalization of labeled collagen (Fig. 2i,j). Over long periods of time, cells 158 slowly agitated the collagen shell, breaking off fragments of the collagen and pulling them into the cell 159 (Fig. 2k). 160 To determine the mechanism of collagen internalization, we incubated cells with high molecular weight 161 dextran, finding that it was ingested alongside labeled collagen (Fig. 2l). Internalization of large liquid-162 phase molecules is indicative of macropinocytosis, since such molecules are excluded from smaller 163 endocytic vesicles. 41 Treatment with the sodium hydrogen exchange inhibitor 5-(N-ethyl-N-164 isopropyl)amiloride (EIPA), an inhibitor of macropinocytosis, 42 decreased the number of dextran-labeled 165 vesicles and internalized collagen fragments, with the number of dextran vesicles and collagen fragments 166 highly correlated across cells ( Fig. 2m-o). Testing the hypothesis that collagen localization at detected 167 dextran vesicles was random, we found that the distribution of p-values was heavily tilted towards small 168 values, indicating likely collagen enrichment within dextran vesicles (Fig. 2p). Furthermore, we did not 169 observe intracellular vesicles containing fluorescently-labeled clathrin light chain (CLC) that were 170 associated with internalized collagen (Extended Data Fig. 4), indicating that clathrin-mediated endocytosis 171 does not contribute to internalization. Thus, we conclude that macropinocytosis is the dominant form of 172 collagen internalization in this form of path generation for cell migration. The centrality of 173 macropinocytosis to this form of amoeboid cancer cell migration, combined with its known role in 174 enabling nutrient uptake in depleted cancer microenvironments, 43 highlights the importance of 175 macropinocytosis to metastasis. 176 Phosphoinositide 3-kinase (PI3K) establishes bleb polarity in feedback with collagen remodeling. 177 For productive path generation, the slow destruction of dense extracellular matrix at the cell front 178 critically depends on persistence factors that promote the highly polarized and continuous formation of 179 large blebs, which abrade and internalize matrix material in a directed fashion. To measure polarization 180 on the 3D cell surface, we used an approximation of a spherical normal distribution, which has fit 181 parameters that intuitively correspond to the direction of the peak and the peak's inverse width, here 182 termed the polarization magnitude (Fig. 3a). Using these statistics, we first confirmed that the distribution 183 of large blebs was more polarized than the overall bleb distribution (Fig. 3b). Measuring the directional 184 correlation of large bleb polarization and collagen localization, we next found that large blebs were 185 systematically biased towards areas of high collagen density (Fig. 3c). Hypothesizing that adhesions might 186 couple collagen and bleb localization, we found that paxillin-containing adhesion complexes indeed 187 formed ( signaling. 47 Moreover, PI3K signaling was more directionally aligned with large blebs than with blebs of all 193 sizes (Fig. 3g), suggesting that PI3K signaling is involved specifically in the polarization of large blebs. 194 Despite their small size, adhesions in the cortical area persisted for several minutes (Fig. 3h), in contrast 195 to the ~1 min lifetime of similarly-sized nascent adhesions formed in cells on a coverslip (Fig. 3i,j). 48 The 196 localization and persistence of cortical adhesions at the front of tunneling cells may enable the 197 recruitment of PI3K to the cell front. 198 To test this hypothesis, we acutely inhibited FAK signaling using a small molecule inhibitor of FAK-kinase 199 activity. FAK inhibition resulted in a decrease in bleb volume, even though bleb polarization and number 200 were unaffected (Fig. 3k). PI3K polarization and mean cell surface motion were also decreased ( Fig. 3k, 201 Extended Data Fig. 6). Measuring the full-width half-maxima of the FAK inhibition response times, we 202 found that PI3K polarization fell first, followed by bleb volume and then cell surface motion. This led us to 203 conclude that PI3K polarity is upstream of large bleb formation at the cell front. Indeed, stratification of 204 blebs by volume revealed that large blebs in particular were enriched for high PI3K signaling and also 205 associated with increased collagen motion (Extended Data Fig. 7). 206 To determine if the relationship between PI3K and bleb size was causative, we used photoactivation to 207 increase PI3K signaling locally in blebbing cells, resulting in a striking increase in proximal but not distal 208 bleb size (Fig. 3l, Extended Data Fig. 8, Movie 6). We also pharmacologically inhibited PI3K signaling by 209 acute addition of a low dose of an inhibitor specific for PI3Ka. In the region of former high PI3K activity, 210 PI3K biosensor intensity and bleb size rapidly decreased, even though de novo bleb formation was not 211 inhibited (Extended Data Fig. 9, Movie 7). Aggregating over multiple cells, we found that both PI3K 212 polarization and bleb size were decreased by PI3K inhibition, whereas the number of blebs and bleb 213 polarization were not affected (Fig. 3m). Altogether, these results indicate that PI3K is responsible for 214 generating large blebs but does not govern the frequency or location of bleb initiation. 215  Extended Data Fig. 10), supporting the notion of a sustained feedback between blebbing and local PI3K 244 activity. 245 We therefore sought to uncover the mechanism that couples PI3K activity and bleb size. Bleb growth is 246 thought to be driven by pressure-based cytoplasmic flows 51 with actin filament formation only at very late 247 growth stages to reform the actin cortex. 52 In contrast, PI3K is usually associated with actin-driven 248 protrusion. 53 Thus, the involvement of PI3K in bleb expansion seemed paradoxical. We noted, however, 249 that PI3K and actin were directionally correlated (Fig. 4b,c) and that regions of the cell with higher PI3K 250 activity protruded faster (Fig. 4d). Indeed, in agreement with previous findings, 51 we found that blebs that 251 ultimately reached a larger size did so by growing faster (Fig. 4e). Inhibition of PI3K activity dramatically 252 decreased the growth rate and final bleb size ( Fig. 4e), confirming that fast bleb growth and large size are 253 due to PI3K signaling. 254 Given the known association of PI3K with actin-based migration, we wondered if actin polymerization 255 could be playing a role in bleb growth despite previous reports otherwise. We found that the F-tractin 256 construct, which localizes to filamentous actin, 54 was absent during bleb expansion and localized only to 257 the bleb cortex, as previously reported. 52 However, expressing low levels of HALO-tagged actin showed 258 that actin is present during bleb expansion (Extended Data Fig. 11a). Calculating the ratio of actin to F-259 tractin revealed that actin was enriched in blebs relative to filamentous actin (Extended Data Fig. 11b), 260 raising the possibility that nascent actin filaments formed in blebs, but are not recognized by the F-tractin 261 probe. To further explore this possibility, we used a TIRF microscope with a high numerical aperture (NA) 262 objective to image actin in the growing blebs of HeLa cells gene-edited to express a copy of the beta actin 263 gene fused to GFP. We found that actin filaments are indeed formed in the bleb during expansion (Fig. 4f,  264 Movie 8). Furthermore, stratification of blebs by their final size revealed that large blebs contained a 265 significantly higher concentration of actin than small blebs, in particular later in their life (Fig. 4g). 266

282
The putative link between PI3K signaling and actin polymerization may rest on the Rac1 -WAVE -Arp2/3 283 pathway, 44,55 which promotes lamellipodia expansion. To test the involvement of this pathway also in bleb 284 expansion, we first blocked Arp2/3 activity by the small molecule inhibitor CK666, 56,57 which decreased 285 bleb volume but did not alter bleb number (Fig. 4h). This is consistent with previous findings that both 286 genetic and acute inhibition of the Arp2/3 complex reduces bleb size. 58 Coupled with previous findings 287 that CK666 treatment does not decrease intracellular pressure 59 and thus is not expected to globally 288 decrease the driving force for bleb expansion, our results suggest that active actin polymerization 289 contributes to bleb growth, especially in large blebs. To add to this conclusion, we employed a 290 photoactivatable Rac1 construct, 60 which allowed us to increase Rac1 activity locally and acutely, avoiding 291 the more global and pleiotropic effects Arp2/3 inhibition may exert. Local photoactivation resulted in a 292 dramatic, reproducible increase in local bleb size that was immediately reversible upon light cessation 293 ( Fig. 4i,j, Movie 9). Our results thus show that the mechanism by which PI3K localization promotes bleb 294 growth at the front of tunneling cells is via Rac1-mediated actin polymerization. 295

Discussion. 296
Our data uncover a mode of cell migration that is effective in dense, soft environments. We refer to this 297 mode as worrying, which means wearing down or tearing repeatedly, like a dog worrying a bone, a 298 meaning that predates the more figurative use as a term related to anxiety. 61 In the context of cell 299 migration worrying denotes that the core element of the mechanism is the sustained agitation and tearing 300 of the extracellular matrix at the cell front by persistently polarized and dynamic cell surface blebs. The 301 persistence is mounted by a mechanochemical feedback between actin-enforced large bleb formation, 302 matrix ablation, adhesion signaling, and PI3K/Rac1 triggered activation of actin filament assembly inside 303 the bleb. The discovery of this self-reinforcing machinery depended on the development of 3D imaging 304 assays to capture cell dynamic behaviors without mechanical interference from the microscope optics, 305 computer vision to extract the relations between cell blebbing and signaling, and optogenetic approaches 306 to acutely interfere with the feedback loop. 307 Our custom-built technology enabled the study of migration in dense, yet soft environments. Not only are 308 such tissues common in vivo, with melanoma in particular known to prefer soft environments, 62 but many 309 tissues throughout the body are likely mechanical composites with pockets of mechanically soft 310 microenvironments. 63,64 It is well established that mesenchymal migration, which is especially common in 311 stiff environments, is facilitated by ECM remodeling leading to tunnel generation and directional fiber 312 alignment. 65 From the recent discovery that amoeboid immune cells build specialized actin structures to 313 push fibers out of the way, 66 combined with the results of this study, we now conclude that all major 314 modes of migration are remodeling modes, in which the environment is at least transiently reorganized. 315 Hence, alongside the core processes of cell migration established by Abercrombie 67 decades ago, i.e. of 316 protrusion, adhesion and contraction, we add environmental remodeling as the fourth process. Moving 317 forward, the use of advanced technologies to dissect migration in ever more complex environments will 318 be critical to enhancing our understanding of the mechanisms of environmental remodeling and its 319 integration with the other core processes. 320 Our results additionally underscore how much remains to be understood about blebs. Blebs are known to 321 play a role in diverse processes beyond cancer cell migration, including apoptosis, cytokinesis, cell 322 spreading, and virus entry. 68 Although the role of blebs as spatial compartments has been previously 323 understood, we show here as an additional function for blebs the pushing and pulling of the environment 324 in order to remodel it. In so doing, we found that bleb expansion is locally controlled by PI3K signaling and 325 mediated by branched actin polymerization. This raises the possibility that the branched actin machinery 326 governing bleb expansion in cell migration, may also be critical to other processes involving blebs, for 327 example organelle segregation into blebs in apoptosis, demanding the need for future studies to 328 understand how cellular control of blebbing enables critical cell functions. 329

330
We The data that support the findings of this study are available from the corresponding authors upon 347 reasonable request. 348 349 Code Availability 350 Much of the code used in this study was associated with previously published methods papers and is 351 available at https://github.com/DanuserLab. All remaining code will be made available at that repository 352 upon publication.

MMP activity assay 563
We used the MMP activity assay kit by Abcam used for the assay. The control MMPs were dissolved in assay buffer and a 2mM AMPA working solution 568 was prepared with assay buffer. The MMP and the test samples were mixed 1:1 vol/vol with the AMPA 569 working solution and incubated for 1h at 37 o C. The MMP green substrate working solution was prepared 570 in assay buffer and then mixed 1:1 vol/vol in the black walled 96 well plate and further incubated for 1h. 571 The samples were then read on a Biotek, Synergy H1 hybrid plate reader at Ex/Em = 490/525 nm. 572

Recombinant DNA Constructs 573
The GFP-AktPH construct was obtained from the laboratory of Jason Haugh (North Carolina State  574 University, Raleigh NC) 2 and cloned into the pLVX-IRES-puro vector (Clontech). The GFP-tractin construct 575 was a gift from Dyche Mullins (Addgene plasmid # 58473; http://n2t.net/addgene:58473; 576 RRID:Addgene_58473) 3 and was cloned into the pLVX-IRES-puro vector (Clontech). Paxillin-pEGFP was a 577 gift from Rick Horwitz (Addgene plasmid # 15233 ; http://n2t.net/addgene:15233; 578 RRID:Addgene_15233) 4 . mRuby2-CLC was a gift from the laboratory of Sandra Schmid (UT Southwestern 579 Medical Center). Cells expressing lentiviral vectors were created by following the manufacturer's 580 instructions for virus preparation and cell infection (Clontech). Cells were selected for expression by 581 treatment with puromycin, G418, or by fluorescence activated cell sorting. 582 The photoactivatable PI3K construct (Idevall-Hagren et al., 2012) was created by cloning mCherry-CRY2-583 iSH2 (Addgene Plasmid #66839) into the pLVX-neo vector (Clontech). The CIBN-CAAX plasmid was 584 obtained from Addgene (Plasmid #79574) and cloned into the pLVX-puro vector. Cells expressing both the 585 mCherry-CRY2-iSH2 and the CIBN-CAAX constructs were selected by treatment with 10 mg/mL puromycin 586 and fluorescence activated cell sorting. It is critical for the two part cry2 photoactivation system that cells 587 express sufficient concentration of the CIBN-CAAX construct or the cry2 construct will aggregate in the 588 cytosol instead of being recruited to the membrane. Thus, the optimal ratio of CIBN:cry2 is greater than 589 one; cells expressing insufficient CIBN-CAAX will not respond to light. We also noted through the course 590 of our experiments that cells will stop expressing one or both of these constructs if not kept constantly 591 under selective pressure. Such a loss of expression will result in non-responsive cells. The PA-Rac1 592 construct was obtained from Yi I. Wu (University of Connecticut Health Center, Farmington, CT). 593 Overexpression of fluorescently tagged monomeric actin can perturb cell cytoskeletal dynamics. To avoid 594 this artifact while imaging tagged actin, we expressed HALO-tagged actin under the control of a truncated 595 CMV promotor, which results in lower expression of tagged actin than the full length promoter. The 596 original actin construct features an 18 amino acid linker between mNeonGreen and actin in a pLVX-597 shRNA2 vector and was obtained from Allele Biotech. We truncated the CMV promoter, and replaced the 598 mNeonGreen fluorophore with the HALO tag sequence. The sequence of the CMV100 promoter region is 599 as follows, with the CMV sequence highlighted and the start codon in bold: 600 TTACGGTAAATGGCCCGCCTGGCTGACCGCCGCTAGCGCTAACTAGTGCCACCATG 602

Phase-contrast imaging 603
Live-cell phase-contrast imaging was performed on a Nikon Ti microscope equipped with an 604 environmental chamber held at 37 o C and 5% CO2 and imaged with 20x magnification. 605

Cells on top of gels 606
Collagen slabs were made from rat tail collagen Type 1 (Corning; 354249) at a final concentration of 3 607 mg/mL, created by mixing with the appropriate volume of 10x PBS and water and neutralized with 1N 608 NaOH dish. The dish was then placed in a 37°C incubator for 4 hours. Following incubation, 1 mL of medium was 613 gently added to the dish. The medium was gently stirred to suspend debris and unattached cells. The 614 medium was then drawn off and gently replaced with 2 mL of fresh medium. 615

Cells embedded in 3D collagen 616
Collagen gels were created by mixing bovine collagen I (Advanced Biomatrix 5005 and 5026) with 617 concentrated phosphate buffered saline (PBS) and water for a final concentration of 2 mg/mL collagen. 618 This collagen solution was then brought to pH 7 with 1N NaOH and mixed with cells just prior to incubation 619 at 37 o C to induce collagen polymerization. Cells were suspended using trypsin/EDTA (Gibco), centrifuged 620 to remove media, and then mixed with collagen just prior to incubation at 37 o C to initiate collagen 621 polymerization. To image collagen fibers, a small amount of collagen was conjugated directly to AlexaFluor 622 568 dye and mixed with the collagen sample just prior to polymerization. FITC-conjugated collagen was 623 purchased from Sigma (C4361). 624

3D confocal imaging 625
The cell/collagen mixture described in the previous section was added to Nunc Lab-Tek II Chambered 626 Coverglass samples holders with a No. 1.5 borosilicate glass bottom (Thermo Scientific). Cells were fixed 627 with paraformaldehyde and stained with Hoechst and FITC-phalloidin. Images were acquired on a Zeiss 628 LSM 880 using a Plan-Apochromat 63x/1.4 Oil objective. 629

Classification of cell morphology 630
For Figure 1d, we classified the fraction of cells as "rounded" as follows. For cells embedded in 3D collagen, 631 we labeled as rounded cells with extensive blebbing as well as round cells with few protrusions of any 632 sort. For cells placed on top of collagen and imaged using phase contrast microscopy, we manually scored 633 each cell as either rounded or stretched, with the rounded morphology indicating the amoeboid 634 phenotype. 635

Zebrafish injection and imaging 636
B16F10 melanoma cells expressing Lifeact-eGFP were injected into the hindbrain ventricle of 2 days post-637 fertilization wildtype zebrafish larvae using previously described protocols. 5 Briefly, B16F10 melanoma 638 cells were suspended in HBSS. 25-50 cancer cells were transplanted into the hindbrain ventricle of 639 anesthetized larvae. Injected zebrafish larvae were incubated at 31 o C with 0.2 mM PTU to prevent 640 pigment formation. Live-cell in vivo imaging was performed using a Zeiss spinning disc microscope with a 641 QuantEM EMCCD camera. 642 High NA TIRF microscopy 643 Human cervical adenocarcinoma cells HeLa-Kyoto with TALEN-edited ActB fused with GFP (Cellectis, 644 France) were maintained in DMEM/F12 supplemented with 10% FBS (Invitrogen) at 37°C and 5% CO2, and 645 imaged using CO2-independent medium (Invitrogen) supplemented with 10% FBS. Cells were confined by 646 a PDMS stamp in a non-adhesive, PLL-g-PEG (0.5mg/ml) coated chamber of ~3µm height, as described 647 previously, 6 and imaged 20 minutes after initiating the confinement to observe actin dynamics within 648 blebs. The high NA TIRF consisted of a standard setup equipped with a 473nm laser 500mW 649 (Laserquantum), an objective TIRF NA=1.49 (Olympus), and a camera (Andor Zyla 4.2). A single notch filter 650 was used in the emission light path to block the laser line at 473 nm (Chroma). Acquisition was controlled 651 by the Andor SOLIS software. 652

3D cell tracking from phase-contrast movies 653
Cells were embedded in 2.0 mg/mL pepsinized bovine collagen in Nunc Lab-Tek II Chambered Coverglass 654 samples holders as described above. Live-cell phase-contrast imaging was performed on a Nikon Ti 655 microscope as described above. Cells were outlined manually using ImageJ, and position and shape data 656 were exported for analysis using Matlab. Cell shape was calculated using roundness, given by 657 4*area/(π*major_axis^2), and cells were classified as either round (roundness > 0. Matlab (Mathworks). To reduce deconvolution artifacts, images were apodized, as previously described. 8 678 Following deconvolution, we used our previously published u-shape3D analysis framework. 9 to segment 679 cells, detect blebs, map fluorescence intensity to the cell surface, measure surface motion, and calculate 680 polarization statistics. Briefly, images of cells were segmented to create a cell surface represented as a 3D 681 triangle mesh. We used u-shape3D's twoLevelSurface segmentation mode, which combines a blurred 682 image of the cell interior with an automatically thresholded image of the cell surface. Blebs were detected 683 by decomposing the surface into convex patches, and using a machine learning algorithm to classify the 684 patches as a bleb or not a bleb. For each patch classified as a bleb, the bleb neck was defined as the 685 boundary between that patch and neighboring patches. Distance from a bleb neck was calculated at every 686 face on the mesh as the geodesic distance to the closest bleb neck. To determine the fluorescence 687 intensity at each mesh face, we used the raw, non-deconvolved, fluorescence image. At each mesh face, 688 a kd-tree was used to identify the cell-interior voxels within a sampling radius of 1 or 2 µm of the mesh 689 face. Before averaging the intensity values in these voxels, the intensity values were depth-normalized to 690 correct for surface-curvature dependent artifacts. 10 The u-shape3D software, as well as the trained 691 machine learning models used here, are available with the previously published manuscript. 9 692 Polarization statistics were calculated by mapping data defined on the cell surface to a sphere, and fitting 693 the mapped data to a 3D von Mises distribution, which is akin to a spherical normal distribution. We 694 calculated bleb polarization by representing each bleb by the location on the bleb surface farthest from 695 the bleb neck, with distances measured on the cell surface. Additionally, since the adhesion images had 696 substantial fluorescence background, to measure adhesion polarization, we bandpass filtered the raw 697 images via a difference of Gaussians procedure, selecting for objects between 1 and 6 pixels in radius. 698

3D collagen image analysis 699
To enhance linear image features, such as collagen fibers, the 3D collagen images were processed with a 700 steerable filter of width 2 pixels, as previously described. 8 To emphasize collagen fiber location, some 701 figure panels, as indicated in the figure legends, show steerable-filter enhanced collagen. Other collagen 702 images, especially those related to endocytosis, were neither filtered nor deconvolved to avoid the 703 creation of artifacts. Collagen polarization near the cell surface was measured after mapping image 704 intensity values from steerable-filtered images onto the cell surface. Following steerable filtering and 705 automatic thresholding, the nematic order parameter of collagen networks was calculated as described 706 previously 8 , except that the average fiber directionality in each 3D image was used as the reference 707 direction. The fiber directionality was calculated at each voxel via a steerable filter. Collagen pore size 708 analysis was also performed as described previously. 11 Images were filtered and then thresholded at 2.5 709 times the intensity threshold calculated by Otsu's algorithm. 12 To measure pore sizes, for each image, we 710 first fitted the largest possible sphere into the collagen pores. We then iteratively fitted the next largest 711 sphere into the pores space minus the volume of previously fitted spheres until no remaining spheres 712 above a size threshold would fit. We defined the distribution of collagen pore sizes as the distribution of 713 fitted sphere diameters. Collagen motion was measured using a previously published 3D optical flow 714 algorithm. 13 This algorithm combines a matching framework for large displacements across frames with a 715 variational framework for small displacements. We mapped the magnitude of the collagen motion 716 calculated via optical flow onto the cell surface using the framework for mapping fluorescence intensity 717 onto mesh faces described above. To create panel 3H, we separated cell surface motion, aggregated 718 across multiple cells, into bins by magnitude and found the mean collagen motion magnitude associated 719 with each bin. The collagen sample moves during imaging, and although the average collagen motion in 720 each frame was subtracted from the measured collagen velocities, residual local motions contribute to 721 create a non-zero background collagen motion. 722

3D dextran assay image analysis 723
To measure the uptake of collagen fragments alongside 70 kDa dextran, we first segmented the cell using 724 the dextran channel. To do so, we inverted each 3D image, subtracted the median intensity, normalized 725 by the 99 th intensity percentile, subtracted the image background, thresholded, morphologically dilated 726 by 1 pixel, morphologically eroded by 8 pixels, filled holes, and finally selected the largest image 727 component. Since the cell is morphologically eroded to a greater extent than it is dilated, the cell 728 segmentation is effectively shrunk, reducing the effect of segmentation errors on later analysis. To detect 729 dots of endocytosed collagen and dextran, we employed a previously published multiscale stochastic 730 filter. 14 For this filter, we used scales of 1.5 to 4 pixels, an a = 0.01, and detected dots only inside the 731 segmented cell. The p-value distribution shown in Fig. 2o results from testing, for each cell, the hypothesis 732 that collagen fluorescence intensity is greater at the location of detected dextran dots than elsewhere in 733 the cell. To calculate the p value for each cell, we randomly picked n collagen intensity values within the 734 cell 100,000 times, where n is the number of detected dextran dots, and calculated the probability that 735 the mean of the randomly picked values was greater than the mean of the collagen intensity values at the 736 true detected dextran dots. 737

738
Photoactivation of subcellular regions was performed using a 488 nm laser at 10% power via the FRAP 739 module of a Zeiss LSM780 outfitted with temperature and CO2 control. Cells for the PI3K optogenetics 740 were treated with 200uM of PI3K inhibitor IV just prior to photo activation. To assess bleb size change in 741 phase contrast movies, we analyzed multiple blebs within the stimulated region by manually outlining 742 individual blebs at their largest size using ImageJ. Bleb size was measured prior to activation and during 743 activation in the same sub-region of the cell. 744

745
3D surface renderings were made in ChimeraX. 15 Colored triangle meshes representing the cell surface 746 were imported into ChimeraX from u-shape3D as Collada dae files, as previously described. 9 To render 747 collagen, steerable-filtered images were opened directly in ChimeraX and thresholded. To create the 748 rendering of adhesions shown in Figure 3d, the raw paxillin images were bandpassed, admitting objects 749 between 0.5 and 3 pixels in radius, and then median filtered. 750