Confined keratocytes mimic in vivo migration and reveal volume-speed relationship

Fish basal epidermal cells, known as keratocytes, are well-suited for cell migration studies. In vitro, isolated keratocytes adopt a stereotyped shape with a large fan-shaped lamellipodium and a nearly spherical cell body. However, in their native in vivo environment, these cells adopt a significantly different shape during their rapid migration towards wounds. Within the epidermis, keratocytes experience 2D confinement between the outer epidermal cell layer and the basement membrane; these two deformable surfaces constrain keratocyte cell bodies to be flatter in vivo than in isolation. In vivo keratocytes also exhibit a relative elongation of the front-to-back axis and substantially more lamellipodial ruffling, as compared to isolated cells. We have explored the effects of 2D confinement, separated from other in vivo environmental cues, by overlaying isolated cells with an agarose hydrogel with occasional spacers, or with a ceiling made of PDMS elastomer. Under these conditions, isolated keratocytes more closely resemble the in vivo migratory shape phenotype, displaying a flatter apical-basal axis and a longer front-to-back axis than unconfined keratocytes. We propose that 2D confinement contributes to multiple dimensions of in vivo keratocyte shape determination. Further analysis demonstrates that confinement causes a synchronous 20% decrease in both cell speed and volume. Interestingly, we were able to replicate the 20% decrease in speed using a sorbitol hypertonic shock to shrink the cell volume, which did not affect other aspects of cell shape. Collectively, our results suggest that environmentally imposed changes in cell volume may influence cell migration speed, potentially by perturbing physical properties of the cytoplasm.


25
After injury to an animal's skin, the wound-healing response restores tissue architecture 26 and protects against entry of pathogens. Multiple cell types orchestrate an array of responses 27 depending on the nature of the wound, including immune defense, cell proliferation, and 28 production of extracellular matrix (Martin, 1997;Singer & Clark, 1999). The restoration of 29 epithelium over the wound, or re-epithelialization, is critical for barrier function across animal 30 taxa. Mammalian re-epithelialization requires that basal keratinocytes proliferate near the wound 31 margin and migrate under the desiccated scab using actin-based lamellipodia (Singer & Clark, 32 1999). The multi-layered adult zebrafish epidermis heals using basal cell crawling and 33 intercalation events (Richardson et al., 2016). In embryonic epithelia, from insects to mammals, 34 rapid wound closure is driven by actomyosin "purse-string" contraction at the wound margin, 35 often in combination with lamellipodial motility (Abreu-Blanco et al., 2012;McCluskey & Martin, 36 1995;Radice, 1980). 37 In vivo work has suggested that re-epithelialization occurs in a dynamic mechanical 38 environment, yet many studies of the biophysical mechanisms driving wound closure have 39 occurred outside of the relevant tissue context. Depending on the species of animal and the 40 type of injury, the initial wound and subsequent healing process can lead to changes in tissue 41 tension, extracellular matrix composition, and cell crowding (Evans et al., 2013;Franco et al., 42 2019;Singer & Clark, 1999). In contrast, studies of cultured cells on plastic or glass surfaces 43 provide far simpler environments. Such studies illustrate how molecular interactions determine 44 cell-scale forces and ultimately tissue-scale stresses and motion (Mogilner et al., 2020;Trepat & 45 Fredberg, 2011). In vitro culture systems, from simple to complex, can also reveal cell-intrinsic 46 responses to environmental complexities by allowing independent control over many external 47 variables. Given the recent proliferation of methods for simulating physiological 48 microenvironments, we were particularly interested in identifying physical properties of wounds 49 and the wound environment that promote or limit efficient migration of epidermal cells. 50 Zebrafish provide an excellent means to study the same cell type migrating during 51 wound healing both in vivo and in culture. Re-epithelialization has been particularly well studied 52 at the larval stage, because of the organism's optical clarity, extremely fast healing response 53 (<30 minutes), and simple two-layered epidermal architecture (Arora et al., 2020;Kennard & 54 Theriot, 2020;Rasmussen et al., 2015;Sonawane et al., 2005). The outer periderm cell layer 55 maintains electrochemical gradients via tight junctions and forms an actomyosin purse-string 56 that contracts the wound margin, while the inner basal cells serve as epidermal stem cells and 57 crawl rapidly towards the wound on a basement membrane using actin-rich lamellipodia (Gault 58 et al., 2014;Lee et al., 2014;Sonawane et al., 2005). Physical cues for basal cell migration 59 include osmotic shock and electric fields. These develop locally when skin wounds disrupt 60 transepithelial gradients of osmotic pressure and ionic electrical potential (Gault et al., 2014;61 Kennard & Theriot, 2020). 62 In addition to contributing to the wound response in vivo, zebrafish skin cells are also 63 well-studied in culture. Isolated fish basal epidermal cells, known as keratocytes, move rapidly 64 and persistently while maintaining a remarkably constant fan-like shape (Goodrich, 1924). This 65 simple cell-scale geometry has facilitated many quantitative studies on cytoskeletal mechanisms 66 of motility and shape determination (Mogilner et al., 2020). Polymerization of a dense actin 67 network at the leading edge smoothly advances the membrane (Garner & Theriot, 2022;Keren 68 et al., 2008;Theriot & Mitchison, 1991), while contraction of non-muscle myosin II at the rear 69 powers actin retrograde flow, network compaction, and actin disassembly (Svitkina et al., 1997;70 Wilson et al., 2010). To transmit these forces to the substrate, adhesions are located throughout 71 the lamellipodium and are particularly prominent at the cell's rear corners (Fournier et al., 2010;72 Oliver et al., 1999). Physical perturbations, such as changes to substrate adhesiveness and 73 membrane area, directly affect keratocyte shape and speed (Barnhart et al., 2011;Lieber et al., 74 2013). 75 Together, in vivo and isolated keratocyte studies can be used in complementary ways to 76 examine how the physical environment of a wound affects migration. For example, in vivo 77 keratocytes are confined to a 2D plane, sandwiched between a basement membrane and the 78 periderm. While the forces exerted by these tissue layers are not known, they are notably 79 absent for isolated keratocytes under standard culture conditions where individual cells move 80 freely on a rigid substrate such as glass or polystyrene. Therefore, isolated keratocytes can be 81 examined with and without experimentally applied confinement in order to simulate, and thereby 82 better understand, in vivo skin conditions. Interestingly, a number of confinement methods that 83 restrict cells to a thin 2D plane have recently been deployed for a variety of other cell types in 84 culture, revealing surprising changes in behavior, as compared to unconfined conditions. These 85 changes include the formation of stable bleb-based protrusions, dorsal actin patches, and 86 seemingly unproductive rear blebs and dorsal adhesions (Gaertner et al., 2022;George et al., 87 2018;Ramalingam et al., 2015;Rape & Kumar, 2014;Srivastava et al., 2020). 2D confinement 88 has also been shown to induce dramatically faster migration, particularly in transformed and 89 poorly-adherent cells (Bergert et al., 2012;Liu et al., 2015;Logue et al., 2015;Ruprecht et al., 90 2015Ruprecht et al., 90 , 2015Toyjanova et al., 2015;Tozluoğlu et al., 2013). However, less is known about the 91 effect of 2D confinement on adherent, well-polarized cell types, especially cells that are already 92 adapted to a flat, thin environment, such as the fish keratocyte. 93 Comparing in vivo and isolated keratocyte cell shapes at high spatiotemporal resolution 94 during rapid migration, we find that keratocytes adopt a taller, rounder cell body in the absence 95 of the skin's physical constraints. Simply adding 2D confinement to isolated cells, via multiple 96 methods, achieved good mimicry of the overall cell shapes observed in vivo. Furthermore, our 97 well-controlled confinement system allowed a detailed quantification of keratocyte shape, size, 98 and speed during a sudden change in environment, revealing an unexpected relationship 99 between volume loss and speed decrease. 100

101
Keratocytes experience 2D physical constraints in the zebrafish epidermis 102 In order to understand the role of the physical environment in keratocyte migration, we 103 compared cells from the same lineage migrating in two settings: 1) the caudal fin epidermis after 104 wounding by laceration , and 2) primary culture after epidermal 105 dissociation (Lou et al., 2015). We conducted time-lapse confocal microscopy on in vivo cells 106 using wounded larvae expressing LifeAct-mNeonGreen in a mosaic fashion in approximately 107 5% of cells (Methods). Laceration was performed in order to induce basal cell migration at 3 108 days post fertilization (dpf), when larvae have naturally emerged from their chorion and are well-109 sized for manual wounding. In parallel, we also examined migration of isolated cells from 110 embryos at a similar developmental stage using primary culture from a stable LifeAct-EGFP fish 111 line. Keratocytes were extracted from 2 dpf larvae to obtain the optimal number of polarized 112 cells (Lou et al., 2015). 113 These movies confirmed that keratocytes in both conditions migrate rapidly with an 114 actin-rich lamellipodium (Figure 1, arrowheads). In vivo, cells migrate collectively towards the 115 wound for a relatively short duration, previously measured to be less than 15 minutes in our 116 standard tissue laceration (Kennard & Theriot, 2020). In contrast, isolated cells migrate 117 persistently for up to several hours without any directional cue, as previously observed 118 (Goodrich, 1924). Using apical-basal cross-sections, we observed that in vivo cells have a 119 uniform height of between 3-4 µm ( Figure 1A, Video S1). In contrast, isolated cells have a 120 spherical cell body situated behind the thin lamellipodium, where the spherical body can be very 121 tall (~8 µm) while the lamellipodium is typically more uniform and much flatter than its 122 counterpart in vivo (<0.5 µm) ( Figure 1B, Video S2). We also observed that cells migrating in 123 vivo are elongated parallel to the direction of migration and exhibit lamellipodial ruffling, but 124 isolated cells are wide with a smooth fan-shaped lamellipodium and their longest axis 125 perpendicular to the direction of migration ( Figure 1C). Since keratocyte cell bodies adopt a 126 taller, rounder shape without the influence of the epidermal environment, we concluded that the 127 overlying cell layer and the basement membrane place a geometric constraint on basal cells, 128 confining them two-dimensionally. We hypothesized that this physical confinement, rather than 129 specific cell-cell interactions or soluble factors, could be largely responsible for the shape 130 differences between in vivo and isolated basal cells. 131 Isolated keratocytes adopt elongated, narrow shapes under two methods of 132 2D confinement 133 We predicted that the introduction of artificial 2D confinement could induce an overall 134 shape change in isolated cells, rendering them more similar in their shape and organization to 135 cells observed in vivo. We sought to directly compare isolated keratocytes in confined and 136 unconfined environments using multiple confinement strategies. First, we confined isolated 137 keratocytes beneath an agarose gel, along with rigid polystyrene beads. Each bead held the 138 agarose gel above the substrate in its local area like a tent-pole. This geometry made it possible 139 for motile cells to move reversibly from areas of strict confinement to areas without confinement 140 over short distances in a single field of view (Figure 2A). These non-confined areas helped us 141 to control for gel effects other than physical contact. Using phase contrast imaging, we observed 142 that cells far from beads, presumably confined by the agarose, displayed unusual shapes for 143 isolated keratocytes. The confined cell shapes featured an ovoid cell body, elongated parallel to 144 the direction of migration and positioned behind the lamellipodium ( Figure 2B, top panel). For 145 the cell shown in Figure 2B, we recorded its migration as it transitioned from a confined region 146 to a region near an agarose-elevating bead, where the cell displayed a wider phase halo that 147 encircled the entire cell body, suggesting increased height as expected ( Figure 2B, yellow 148 arrowhead). As the cell became less confined, it simultaneously adopted a more circular cell 149 body that rested on top of a wider lamellipodium, similar to keratocytes in unconfined samples 150 ( Figure 2B,C, Video S3). These observations suggest that 2D confinement is directly 151 connected to an elongated, narrow cell shape, with confined cells elongating parallel to their 152 axis of migration and unconfined cells generating wider lamellipodia such that the cell's major 153 axis becomes perpendicular to its direction of migration. These observations further indicated 154 that the shape changes between these two states were rapidly reversible over 1-2 minutes and 155 depend only on the cell's immediate environment and not on its recent history. 156 To improve reproducibility, imaging, and temporal control of confinement, we 157 implemented a vacuum-controlled confinement device that lowers a polydimethylsiloxane 158 (PDMS) ceiling to a height of ~3 µm (Le Berre et al., 2012). This flat, rigid, polyethylene glycol-159 coated ceiling also provides simpler geometry, mechanics, and surface chemistry compared to 160 a viscoelastic agarose gel (Figure 2D, S1). When the PDMS ceiling is lowered onto isolated 161 keratocytes, they display rapid shape changes. keratocytes enabled a quantitative study of the interplay between environment dimensionality 171 and cell shape. Keratocytes were isolated from 2 dpf larvae stably expressing cytoplasmic 172 mCherry and then labeled with the far-red membrane dye CellMask Deep Red. We made 173 parallel measurements on keratocytes isolated from mCherry-negative sibling larvae to control 174 for mCherry phototoxicity. Each cell was imaged in the PDMS confiner with the ceiling in the 175 raised position to characterize both its three-dimensional shape and its motility behavior. Under 176 our standard imaging conditions, collection of a set of high-resolution z-stacks (240 nm slices) 177 for 10-15 fields of view in a single sample could be completed in about two minutes. After this 178 initial shape determination, all cells in the sample were re-imaged using single slices at the level 179 of the lamellipodium using two paired images at exactly 1 minute apart in order to accurately 180 measure cell speed. Next, a second round of z-stack images were collected immediately before 181 the ceiling was lowered. After lowering the ceiling and restoring proper focus, we continued 182 imaging the same fields of view, alternating collection of z-stacks and one-minute single-slice 183 speed measurements, for 20-30 minutes. Using this approach, we collected shape and speed 184 data on large populations of the same individual cells before and after confinement ( Figure 3A, 185 left). Some cells did not have adherent lamellipodia and exhibited circular or irregular shapes, 186 which we categorized as unpolarized ( Figure 3A, third column). Other cells ruptured upon 187 confinement, as evidenced by scattered puncta of previously cytoplasmic mCherry ( Figure 3A, 188 fourth column). All time frames were individually scored for these two behaviors (Methods). 189 Unpolarized or ruptured states occurred spontaneously in 12% of unconfined cells, 41% of cells 190 confined to a height greater than 2.5 µm, and 85% of cells confined to less than 2.5 µm ( Figure  191 3B). Unpolarized cells were effectively non-motile compared to their polarized counterparts 192 ( Figure S2A). For further analyses, we retained only cells which were able to continue migrating 193 with persistent polarity after confinement. Among this population, the cells had an average 194 height of 8 µm for the unconfined condition and 4 µm for the confined condition ( Figure S2B). 195 This dataset allowed a quantitative examination of the effect of confinement on multiple 196 shape variables. We used an unbiased, information-rich approach built on Principal Component 197 Analysis, or PCA, for cell shape analysis (Pincus & Theriot, 2007). PCA requires that cell 198 shapes be represented as vectors. To construct vectors that meaningfully represent cell shape, 199 we segmented the CellMask or mCherry channel for each cell, aligned the mask to the direction 200 of its instantaneous velocity, projected the mask into the XY or XZ plane, then extracted a set of 201 point coordinates that outlined the cell boundary ( Figure S3A,C, Methods). We conducted PCA 202 separately for both the XY-and XZ-plane datasets, thereby determining the axes of greatest 203 shape variation across the combined confined and unconfined populations. We refer to each 204 axis as a shape mode, and the combined set of axes as a shape space. XZ-plane 205 segmentations from mCherry-negative cells were not included in the PCA, due to 206 heterogeneous CellMask signal near the apical surface ( Figure S3A). For the XY-plane PCA, 207 mCherry labeling did not have a significant effect on any of the top three shape mode scores (p 208 > 0.05 for each mode, two-way ANOVA for label and confinement condition), so we pooled 209 mCherry-positive and mCherry-negative samples. 210 In both the XY and XZ planes, just 3 shape modes describe over 90% of shape variation 211 ( Figure 3C). These shape modes can be assigned human-interpretable meaning by generating 212 the outlines of idealized cells as they varied along each mode, while holding the coefficients of 213 other modes constant at zero. For example, XZ shape mode 1 appears to be largely correlated 214 with cell height ( Figure S2C). Additionally, a cell with a positive XZ shape mode 1 coefficient 215 has a round cell body that is much taller than the lamellipodium. A negative coefficient along XZ 216 shape mode 1 indicates a thin cell with uniform height (Figure S2D). This shape mode 217 describes 71% of the XZ shape variation. The average value of the coefficient for XZ shape 218 mode 1 was lower for confined cells as compared to unconfined ones (Table S1), confirming 219 our initial observation that confinement induced unusually flat cells, as expected given the mode 220 of confinement. 221 The axes that best distinguish confined and unconfined cell shapes are XZ shape mode 222 1 and XY shape mode 3, which appears to primarily report on the cell's length parallel to the 223 direction of migration ( Figure S2E). For both modes, the confined cells have exaggerated 224 negative scores, indicating flat, highly elongated cells ( Figure 3D). In contrast, other shape 225 modes showed much smaller effect sizes between confined and unconfined cells (Table S1, 226 Figure S2F). There was also a strong correlation between height and inverse length for 227 confined cells (Pearson r = 0.73, p < 0.0001). Therefore, cell elongation in the XY plane occurs 228 concomitantly with Z-plane confinement, even though the confinement method does not impose 229 any boundaries in the XY plane. Elongation may occur simply because the cell body and the 230 dense lamellipodium cannot be stacked vertically, and instead the cell body is carried behind 231 the lamellipodium for confined cells, while it is carried above the rear axle of the lamellipodium 232 in unconfined cells. Alternatively, the confined cell could be longer due to differential friction on 233 the cell body versus the lamellipodium. Regardless of the mechanism of shape change, 234 elongation in the direction of migration is a stereotypic response to confinement. 235 Next, we used this PCA-generated shape space, developed on measurements of 236 isolated cells only, to map the differences between in vivo and isolated keratocyte shapes. We 237 measured in vivo shapes using 3 dpf larvae mosaically expressing either LifeAct-mNeonGreen 238 or cytoplasmic mNeonGreen with a membrane-tagged mRuby. Cells were imaged during the 239 first 10 minutes after laceration and manually segmented in the XY plane using both channels 240 ( Figure S3B). We extracted only the XY shape because the XZ shape was obscured by the 241 tailfin's slight curvature and tilt relative to the imaging plane. These cells were projected into the 242 isolated cell shape space, with the in vivo wound direction aligned to the isolated cell velocity 243 direction. Despite large cell-to-cell variability, in vivo and unconfined cells were well separated in 244 the shape space, with in vivo cells scoring as narrower and longer than the isolated cells 245 ( Figure 3E, Table S2). In addition, in vivo cells are more variable in length than width, whereas 246 unconfined cells are more variable in width than length. Confined cells are variable along both 247 modes, with individual scores overlapping both unconfined and in vivo scores. The average 248 shape coefficients for all three populations are surprisingly collinear, with confined cells falling 249 between unconfined and in vivo. Therefore, 2D confinement phenocopies in vivo shape by 250 elongating cells in the direction of migration, although confinement does not fully mimic the 251 narrowness or lamellipodial ruffling of in vivo cells. These shape features may be a product of 252 cell-cell interactions or mechanical properties not recapitulated by the rigid PDMS ceiling. In vivo 253 shape similarity is best achieved by moderate confinement that substantially deforms the 254 otherwise spherical cell body, without rupturing the cell. These quantitative comparisons 255 demonstrate that 2D confinement makes a significant contribution to cell shape in the larval 256 zebrafish epidermis. 257

258
Our paired image data additionally enabled an investigation into the effect of 259 confinement on keratocyte speed, to understand how the skin environment may or may not 260 promote cell migration during wound healing. In the pairwise comparison before and during 261 confinement, isolated keratocytes showed a speed decrease of 22% ± 8% (s.d., n=10 samples, 262 Figure 4A). As a control, keratocytes were measured using the same imaging protocol without 263 confinement. Under these conditions, speed showed no overall change over a similar duration 264 ( Figure S4A, left). To explore possible influences on speed, we reconstructed the cell boundary 265 in three dimensions ( Figure 4B). Surprisingly, this analysis revealed confinement decreased 266 cell volume by 21% ± 1% (s.d., n=5 samples, Figure 4C). In contrast, unconfined control 267 samples maintained volume within 2% ± 3% (s.d., n=4 samples, Figure S4A, middle). Surface 268 area measurements were not significantly different between unconfined and confined conditions 269 ( Figure 4D). This observation is consistent with previous keratocyte surface electron 270 micrographs and tether pulling measurements, which indicated that keratocytes maintain a fully 271 extended plasma membrane under high tension (Lieber et al., 2013). These results suggest that 272 in response to extreme cell body deformation, keratocytes lose volume overall because they 273 can neither expand their membrane area nor redistribute sufficient volume from the cell body to 274 the lamellipodium. 275 Cell size, shape, and speed changes were measured within the same paired dataset, 276 which allowed us to explore the relationships among variables. Pearson correlation coefficients 277 were calculated between all possible pairs of measurements for both unconfined and confined 278 cells on a per-cell basis and the results summarized as a heat map ( Figure 4E). Surface area, 279 volume, and XY shape mode 1 (width) formed a cluster of cell size variables that were all highly 280 correlated with one another across confinement conditions. The variables that define the 281 flattening/elongation response, including height, XZ shape mode 1 (height), and XY shape 282 mode 3 (inverse length), showed high correlations between confined shape values and low 283 correlations between unconfined values, indicating that confined shape is unrelated to prior cell 284 shape properties. Overall, the heat map emphasized strong correlations between many size and 285 shape variables. 286 To generate hypotheses about causes of cell speed loss under confinement, we 287 investigated correlations involving both confined and unconfined speed measurements ( Figure  288 4E). Notably, confined volume and confined speed exhibited a correlation coefficient of 0.48 289 ( Figure 4F). Confined speed correlated similarly with unconfined volume, unconfined surface 290 area, and confined surface area, which was expected given the strong correlations within the 291 cell size cluster of the heat map. In contrast, size variables were not correlated with unconfined 292 speed, suggesting that the size/speed relationship is specific to the application of confinement. 293 XZ shape mode 1 (height), although uncorrelated with the cell size cluster, also showed 294 significant correlations with speed ( Figure 4G). The correlation coefficient was 0.48 before 295 confinement and 0.46 during confinement, indicating that height is associated with speed 296 independently of applied environmental changes. Finally, cell-to-cell speed variability persisted 297 across conditions, indicated by a good correlation between unconfined and confined speed 298 ( Figure S4B). Therefore, cell size, height, and cell-intrinsic properties are all indicative of cell 299 speed immediately after confinement. 300 Hypertonic shock causes volume and speed loss on par with confinement 301 Cell volume changes in response to confinement are likely to affect cytoplasmic 302 properties, and by extension, cytoskeletal dynamics. In order to directly test a potential volume-303 speed relationship, we altered keratocyte volume in a manner unrelated to physical 304 confinement. Specifically, we used hypertonic shock to rapidly decrease keratocyte volume. We 305 made alternating 3D shape and speed measurements for up to 35 minutes using our standard 306 imaging protocol. During the acquisition, the medium was swapped for media supplemented 307 with 20-50 mg/ml sorbitol, which does not enter animal cells (Watari et al., 2004;Wood et al., 308 1968). The hypertonically-shocked cells maintained lamellipodial migration, displaying the 309 standard unconfined keratocyte shape with a large fan-shaped lamellipodium ( Figure S4C). 310 Pooling across sorbitol concentrations, the hypertonic treatments achieved volume changes 311 comparable to those observed in confinement (Figure 5A,B). Surface area remained constant 312 during the volume decrease, again suggesting that keratocytes do not remodel their membrane 313 on the timescale of minutes ( Figure 5C). Interestingly, hypertonic shock slowed cells down by 314 20% ± 10% (s.d., n=11 samples, Figure 5D), similar to the speed decrease we had observed 315 for confined cells. 316 To extend our observations on the effect of hypertonic shock to other aspects of 317 keratocyte shape in addition to total cell volume, we generated a second shape space (using 318 the methodology described above) combining cells after treatment with hypertonic medium and 319 control cells in isotonic medium (Figure S4D,E). The principal shape modes in the XY plane 320 were strikingly similar to the modes established for the combined unconfined and confined cell 321 populations (compare Figure S4D and Figure 3C, left half). However, the modes in the XZ 322 plane were qualitatively different, reflecting the observation that cells treated with hypertonic 323 shock never achieved the completely flat profiles of confined cells (compare Figure S4E and 324 Figure 3C, right half). Within this new shape space, we compared individual cell shapes before 325 and after hypertonic shock. Overall, there was very little change in shape mode coefficients for 326 hypertonically-shocked cell shapes versus their pre-shock values ( Figure 5E). XZ shape mode 327 3 had the biggest difference between conditions, with an effect size of 1.5. However, this mode 328 was highly correlated with cell volume and therefore does not describe a distinct shape change 329 separate from the volume-dependent effect ( Figure S4F). If XZ shape mode 3 is not 330 considered, the shape mode effect sizes for hypertonic treatment are all within ±0.65 standard 331 deviations, which is much smaller than the effect sizes for confinement of >2 standard 332 deviations (Table S1,S3). Therefore, we conclude that only a change in volume, and not a 333 volume-independent change in shape, is associated with speed loss following hypertonic shock. 334 Keratocytes responded to confinement and hypertonic shock with comparable losses in 335 volume and speed, despite dramatic differences in type of perturbation and resulting cell shape 336 ( Figure 5F). In both conditions, cells lost around 20% of both their speed and volume, with a 337 similar level of sample-to-sample variability in either experiment. This suggests that speed 338 reduction from both confinement and hypertonic shock may involve a shared mechanism, 339 mediated through rapid volume loss. 340

341
Here, we examined keratocytes in vivo, in culture, and under artificial 2D confinement to 342 evaluate the effect of skin-like geometric constraints on lamellipodial migration. We 343 demonstrated that the larval skin environment deforms in vivo keratocytes relative to their 344 isolated shape, so that they are flattened in height and elongated parallel to the direction of 345 migration. A similar effect may occur in the multi-layered adult zebrafish epidermis, which also 346 displays lamellipodial migration with cells elongated substantially in the direction of migration 347 (Richardson et al., 2016). 2D confinement in culture is sufficient to recapitulate this physical 348 constraint and its effect on cell shape, without need of neighboring cells or specific adhesions. 349 However, despite having a more physiological shape, confined keratocytes are slower than their 350 unconfined counterparts. 351 It is possible that shape changes partially contribute to confined speed loss. Our cross-352 sectional shape analysis revealed that cell roundness is connected to cell speed. As measured 353 with XZ shape mode 1, we observed that flat, thin cells tended to be slower than those with tall, 354 spherical cell bodies both before and during confinement ( Figure 4G). A possible explanation 355 for this correlation is that a round cell body may promote faster cell speed, consistent with a 356 previously proposed "cell body rolling" mechanism of translocation of keratocytes (Anderson et 357 al., 1996;Okimura et al., 2018). The rolling model proposes that the rotation of cell body 358 material acts like a wheel rolling in the direction of migration, partially contributing to cell speed. 359 It seems likely that a confined cell body, which has a much less circular XZ cross-section, would 360 roll less efficiently. Cell body rotational tracking, in parallel with shape analysis, could clarify 361 whether altered rolling rates explain the relationship between roundness and speed. 362 The hypertonic shock experiments suggest that keratocyte speed is partially determined 363 by cell volume or a physical property correlated with volume, such as cytoplasmic density. A 364 20% reduction in volume from either hypertonic shock or confinement was associated with an 365 equivalent percent reduction in speed (Figure 5F), even though cells were more deformed 366 under confinement than hypertonic shock. This substantial volume change during active 367 migration is surprising, given that the cell is only 60-80% water and that many biochemical and 368 diffusive processes are sensitive to protein concentration and cytoplasmic macromolecular 369 crowding (Milo & Phillips, 2016;Molines et al., 2022;Neurohr & Amon, 2020). Nonetheless, 370 similar volume-speed trends have been reported for 1) primary human neutrophils in various 371 media tonicities (Rosengren et al., 1994), 2) cancer cells in 3D hydrogels of various stiffnesses 372 (Wang et al., 2020) and 3) Dictyostelium amoebae under 2D confinement with various applied 373 loads (Srivastava et al., 2020). All three studies reported that conditions yielding smaller cell 374 volume also caused slower migration speeds, in agreement with our findings. Using paired 375 single-cell measurements and quantitative shape analysis, this work more directly connects 376 speed and volume, by decoupling volume changes from geometric perturbations and other 377 alterations in cell shape. 378 The underlying connection between speed and volume remains unidentified. Based on a 379 lack of cell rupture or vesiculation, it seems likely that cells lose water volume and therefore 380 experience a change of cytoplasmic physical properties. Rapid water flux through aquaporins 381 has been demonstrated many times in response to osmotic shock (Preston et al., 1992). 382 Recently, volume loss due to 2D confinement was measured to also occur rapidly, with HeLa 383 cells losing 30% of cell volume within 30 milliseconds (Venkova et al., 2022). This volume flux is 384 a reasonable rate for ~100,000 aquaporin channels (Gade & Robinson, 2006). Thus, water 385 efflux is a likely cause of confinement-induced volume loss. 386 Water loss leads to substantial cytoplasmic changes, as measured by increases in cell 387 stiffness and decreases in diffusion rates (Guo et al., 2017;Joyner et al., 2016;Miermont et al., 388 2013;Zhou et al., 2009). These changes occur independently of an intact actin cytoskeleton 389 and therefore are thought to operate via general cytoplasmic molecular interactions. 390 Interestingly, osmotic-induced water flux has been recently shown to have a dramatic effect on 391 microtubule polymerization and depolymerization in fission yeast (Molines et al., 2022). Here, 392 volume loss, rather than accelerating polymerization via an increase in tubulin concentration, 393 instead slowed tubulin diffusion and ultimately polymerization and depolymerization. Actin 394 polymerization is a diffusion-limited reaction and therefore likely to be sensitive to cellular 395 properties that affect cytoplasmic diffusion rate, like volume (Drenckhahn & Pollard, 1986). 396 Additionally, keratocytes require global actin transport because network assembly is high at the 397 front of the cell, while disassembly dominates at the rear (Wilson et al., 2010). Normally, actin 398 monomers can diffuse across the cell in 3 seconds and are further aided by bulk fluid flow from 399 back to front (Keren et al., 2009;Raz-Ben Aroush et al., 2017). Water efflux could perturb these 400 transport processes and thereby additionally reduce actin availability for polymerization at the 401 cell front. Therefore, loss of cytoplasmic water may impede keratocyte migration via altered 402 actin diffusion and transport, while leaving cytoskeletal shape and organization relatively intact. 403 In tissues, osmotic shock is reported to affect wound closure rates across a number of species. 404 Axolotl skin explants heal slower in the presence of hypertonic media (Tanner et al., 2009). With 405 the opposite perturbation, hypotonic shock, zebrafish larvae and Drosophila embryos heal faster 406 (Gault et al., 2014;Kennard & Theriot, 2020;Scepanovic et al., 2021). Zebrafish, as a 407 freshwater species, are thought to use such hypotonic shock as a normal signal for migration 408 after wounding. It is tempting to speculate that hypotonic shock promotes wound healing not just 409 through signaling pathways, but also through physical effects on cells. Indeed, individual 410 keratocytes swell to 50% of their pre-wounding volume following normal wounding in freshwater 411 (Kennard & Theriot, 2020), which could significantly boost speed if the volume-speed 412 relationship holds in the swelling regime. Hypotonic shock additionally leads to a massive fluid 413 influx that increases the space between basal cells and the overlying periderm (Kennard et al., 414 2022). This expansion of the epidermis may release some of the 2D constraints on keratocytes 415 and provide additional speed benefits. The evidence presented here suggests that hypotonic 416 shock may promote wound closure by physically increasing cell migration speed, potentially by 417 both diluting the crowded cytoplasm and reducing 2D confinement. 418 Acknowledgments 420 We are grateful to Christopher Prinz for drawing the schematic in Figure 1C and Andrew 421 Kennard for sharing fish lines and plasmids, as well as advice concerning many aspects of this 422 project. We are also grateful to Emily Hatch for sharing her expertise in confinement methods. 423 We thank Jeff Rasmussen (Westerfield, 2007). Experiments were approved by University of Washington 456 Institutional Animal Care and Use Committee (protocol 4427-01). Animals were raised on a 12 457 hr light, 10 hr dark cycle at 28.5 °C. Adults were crossed through natural spawning. Embryos 458 were collected and raised in 100 mm petri dishes at 28.5 °C. Each dish contained 30-50 459 embryos in system water. 460 Transgenic zebrafish lines 461 The following previously generated transgenic zebrafish lines were used: 462 TgBAC(ΔNp63:Gal4) la213 (Rasmussen et al., 2015), Tg(UAS:LifeAct-EGFP) mu271 (Helker et al., 463 2013), Tg(UAS:mCherry) (Kennard et al., 2022). 464

465
The UAS:mNeonGreen-P2A-mRuby3-CAAX as previously published (Kennard & 466 Theriot, 2020). The UAS:LifeAct-mNeonGreen-P2A-mRuby3-CAAX plasmid was generated 467 using the Tol2kit (Kwan et al., 2007) using standard Gateway cloning methods (Invitrogen). 468 Briefly, the final plasmid was formed by LR reaction (LR Clonase II, Invitrogen) to recombine 469 plasmids p5E-UAS (Tol2kit), pME-LifeAct-mNeonGreen-P2A (this work), and p3E-mRuby3-470 CAAX (Kennard & Theriot, 2020) into destination vector pDestTol2CG2 (Tol2kit), containing the 471 cmlc2:GFP transgenic marker. The construction of the p3E-mRuby-CAAX plasmid has been 472 previously published (Kennard & Theriot, 2020). The pME-LifeAct-mNeonGreen-P2A plasmid 473 was generated from the previously published pME-mNeonGreen-P2A (Kennard & Theriot,474 2020) using Q5 mutagenesis with primers #1 and #2 (Table S4) to insert the LifeAct peptide 475 (Riedl et al., 2008) in-frame at the N terminus. The pME and p3E insert sequences were 476 confirmed by Sanger sequencing and the final destination vector was confirmed by restriction 477 digest. 478 Tol2 transposase mRNA was synthesized using the SP6 mMESSAGE mMACHINE 479 reverse transcription kit (Invitrogen), with the Tol2kit plasmid pCS2A-transposase as a template. Tissue laceration 489 Experiments were performed at 3 dpf. Mosaically-expressing larvae were prepared for 490 imaging and lacerated as previously described . Briefly, larvae were 491 anesthetized in larval imaging media, a mixture of E3 (5 mM NaCl, 0.17 mM KCl, 0.33 mM 492 CaCl2, 0.33 mM MgSO4), 160 mg/ml Tricaine (Sigma), and 0.8 mM Tris pH 7. Larvae were 493 mounted in 2% agarose (Invitrogen) onto a 25 mm coverslip. Larvae were mounted on their 494 right side, with the dorsal-ventral axis aligned parallel to the coverslip. The cooled agarose was 495 covered with excess media and a portion of the agarose around the tail fin was removed. Before 496 wounding, a 60X magnification field of view, which included multiple construct-expressing cells, 497 was selected from the tail fin either dorsal or ventral to the notochord. Laceration was performed 498 with needles pulled from solid borosilicate glass rods (Sutter BR-100-10) using a Brown-Flaming 499 type micropipette puller (Sutter P-87). The needle was dragged through the tailfin twice, from 500 near the end of the notochord towards the posterior edge of the tailfin at a 45° angle. 501 Keratocyte isolation and cell culture reagents 502 Keratocytes were isolated from 2 dpf larvae, which were either wild-type or screened at 503 1-2 dpf for transgenes of interest. Isolation was performed as previously described (Lou et al., 504 2015). Briefly, larvae were dechorionated and anesthetized with 160 mg/ml Tricaine. Larvae 505 were washed twice in PBS, incubated in cell dissociation buffer (Fisher) at 4°C for 30 minutes, 506 mechanically mashed by a 200 µL pipette tip, and incubated in 25% trypsin and 1 mM EDTA for 507 ~15 minutes at 28°C. The trypsin was quenched with fetal bovine serum supplemented with 508 antibiotic/antimycotic and the supernatant further concentrated by centrifugation for 3 minutes at 509 500 g. This cell-rich solution was plated on coverslips coated with rat tail collagen I (Gibco) and 510 incubated for 1 hr at room temperature to allow keratocytes to adhere. The samples were 511 washed three times with Leibovitz's Media (L-15), supplemented with 10% fetal bovine serum 512 and antibiotic/antimycotic, then imaged in the same media. Some samples were incubated at 513 4°C for 1-2 hours to reduce microbial growth, followed by 1 hour at room temperature before 514 imaging. Membrane staining was performed immediately before imaging each coverslip, using 515 CellMask Deep Red (Fisher) for 10 minutes. For hypertonic experiments, the imaging media 516 was supplemented with 20-50 mg/ml D-sorbitol (Sigma) and filter-sterilized. 517 2D confinement by agarose gel overlay 518 On the day of each experiment, a 2X solution of 2% agarose (w/w, UltraPure Low 519 Melting Point Agarose Powder, Invitrogen) was dissolved by heating and then kept fluid at 42°C. 520 A 2X imaging solution was prepared by mixing 20% fetal bovine serum into L-15. The final 521 solution of 1% agarose and 10% fetal bovine serum was mixed from the 2X agarose solution 522 and 2X imaging media in a 1:1 (v/v) ratio. An agarose slab was made by pipetting the final 523 agarose mixture between two microscope slides with four #1.5 coverslip spacers and allowed to 524 gel at room temperature. Polystyrene beads (25 µm diameter, Polysciences 07313) were diluted 525 1:100 in phosphate saline buffer and 10 µl of the dilution was distributed on a sample of 526 keratocytes isolated from larvae 2 dpf (see Keratocyte isolation and cell culture reagents). The 527 agarose slab was carefully draped over the keratocyte and bead sample and allowed to settle 528 for 10 minutes before imaging. 529 2D confinement by vacuum-controlled device 530 The vacuum-controlled confiner device was manufactured similarly to previous 531 descriptions and comprised a polydimethylsiloxane (PDMS) suction cup with a piston, which 532 was capped by a coverslip with a micropatterned PDMS surface (Le Berre et al., 2012Berre et al., , 2014. 533 Device production required two molds, one for the suction cup and one for the micropatterned 534 ceiling (Figure 2D, S1). The suction cup mold was designed for use on top of a 25 mm coverslip 535 and consists of 6 separate parts. The stainless steel and aluminum parts were machined by the 536 UW Physics Department Machine Shop (for exact drawings, see Figure S5). We used 537 Autodesk Fusion 360 for design and rendering. The O-ring (#117) and glass disk (1" x 0.25") are 538 off-the-shelf parts (McMaster-Carr). The ceiling mold was manufactured from SU-8 photoresist 539 by the Washington Nanofabrication Facility. A photomask protected pillars of 440 µm diameter, 540 spaced in an array with 2.5 mm by 0.7 mm center-to-center spacing ( Figure S1C). The final 541 mold height for the pillars was 3.2 µm. 542 To manufacture each suction cup, Sylgard 184 (Thermo Fisher) was thoroughly mixed in 543 a PDMS/cross-linker ratio of 10:1 (w/w) and degassed in a vacuum chamber to remove bubbles. 544 All parts of the mold were rinsed with 5% Pluronic F127 before assembly. All parts of the mold, 545 minus the glass disk, are pre-assembled with the inner ring merely sitting on the platform (see 546 SI). The PDMS was poured into the mold to just over the height of the ledge that the glass disk 547 rests on and degassed again. The glass disk was lowered into place and the assembly was 548 baked at 80°C for 1 hr and allowed to cool. The suction cup was de-molded with the aid of 549 100% isopropanol and pierced with a 0.5-mm biopsy punch to create an inlet for a vacuum line. 550 Alternative mold designs have been described previously, including an assembly of aluminum 551 rings, coverslips, and glass slides (Le Berre et al., 2014). 552 The confinement chips were manufactured as previously described (Le Berre et al., 553 2014). Briefly, Sylgard 184 was thoroughly mixed in a PDMS/cross-linker ratio of 8:1 (w/w), 554 degassed, and poured onto the micropatterned SU-8 mold (which was encased by aluminum 555 foil to prevent PDMS from flowing off the flat mold). After further degassing, plasma-cleaned 8-556 mm coverslips were placed in the PDMS and pressed onto the patterned surface. The mold was 557 baked on a preheated 95°C hot plate for 15 minutes, and coverslips were de-molded using 558 100% isopropanol and a razor blade at a flat angle, nearly parallel to the coverslip to avoid 559 damaging the mold. 560 On the day of each experiment, each confinement chip was plasma-cleaned and 561 incubated in 0.5 mg/ml pLL-PEG (Susos) for 1 hr, then equilibrated with cell culture media for 1 562 hr. The suction cup was sanitized with 70% ethanol and allowed to dry. The glass side of the 563 confinement chip was placed onto the piston, where it self-adhered. Then, the suction cup was 564 sealed onto the sample on the microscope with -8.5 kPa gauge pressure. This vacuum level 565 held the ceiling >50 µm above the cells. To apply confinement, the vacuum was slowly 566 increased, with periodic measurements of cell height using the membrane channel. The cells 567 were considered confined at <4 µm, which required between -17 to -30 kPa, depending on the 568 sample. Afterwards, suction cups were cleaned with dish soap and stored for future use; 569 confinement chips were discarded. 570 Microscopy and image acquisition 571 Imaging for the agarose confinement experiment was performed on a Nikon Ti-E 572 inverted microscope with transmitted light using phase contrast, a 100x objective (NA 1.45,573 Nikon) with 1.5x intermediate magnifier. Images were acquired onto a Andor iXon EMCCD 574 camera using µ-Manager (Edelstein et al., 2014). All other experiments were imaged using a 575 spinning disk confocal setup, consisting of a Nikon Ti2 inverted microscope, a piezo-z stage 576 (Applied Scientific Instruments PZ-2300-XY-FT), a Yokogawa CSU-W1 spinning-disk confocal 577 with Borealis attachment (Andor), and a back-thinned EMCCD camera in 16-bit imaging mode 578 (Andor DU888 iXon Ultra). Confocal illumination was supplied by a laser launch (Vortran 579 VersaLase) with 50 mW 488 nm, 50 mW 561 nm, and 110 mW 642 nm diode lasers (Vortran 580 Stradus). Either a 100X oil objective (NA 1.45, Nikon) or a 60X oil objective (NA 1.40, Nikon) 581 were used, with a 405/488/561/640/755 penta-band dichroic (Andor) and single-band GFP, 582 RFP, and far-red filters (Chroma 535/50m, 595/50m, and 700/75m respectively). Images were 583 acquired at room temperature (20-23°C) using µ-Manager. For the experiments that tracked 584 shape, size, and speed, 10-15 positions were imaged in one acquisition from each sample. The 585 frame interval was set to 60 seconds, with z-stacks only imaged every 3 frames with a z-slice 586 interval of 240 nm. Acquiring the full z-stack at each position actually took 1-3 minutes, but the 587 frame interval between two subsequent non-stack frames was always 60 seconds (useful for 588 accurate speed quantification, see Cell trajectory analysis). During vacuum-controlled 589 confinement experiments, the pressure decrease caused a significant change in the z-position 590 of the coverslip, so the focal plane of each position was reset between confinement application 591 and further imaging. 592 Cell trajectory analysis 593 Cells were tracked based on single-slice images of the membrane channel, taken near 594 the coverslip. Tracks were calculated using custom code written in Matlab R2018b (available 595 upon request). Briefly, cell trajectories were determined using the centroid of the segmented 596 membrane signal. Segmentation was performed on the entire image for the first frame, which 597 required manual gamma adjustment and thresholding. For subsequent frames, segmentation 598 was automatically performed using Otsu thresholding within a 150 x 150 or 200 x 200-pixel 599 region around the previous centroid. Subsequent positions for the same cell were assigned by 600 calculating all possible cell-to-cell displacements between consecutive time points and matching 601 cells through the minimization of total displacement across cells. The cell density from the 602 isolation procedure was low enough that individual tracks could be identified with this approach, 603 followed by manual curation to remove misidentified tracks. For the confinement experiments, a 604 significant time gap occurred between the before-confinement acquisition and the during-605 confinement acquisition. Therefore, all matching tracks were manually assigned, based on the 606 expected direction of cell movement, approximate cell size, and intracellular components visible 607 in the phase contrast channel. Cell speeds were only calculated between frames that were 608 taken at exactly 60-second intervals. 609 Cell behavior scoring 610 Each frame of each track was scored as polarized or unpolarized in a random order by a 611 researcher blind to the experimental condition. Scoring was performed using the membrane 612 slice taken close to the coverslip. Cells were considered to be unpolarized if they were ruptured, 613 blebbing, symmetric on two or more axes, or had multiple lamellipodia or a perimeter with high-614 frequency curvature. A subset of frames was scored in triplicate, which demonstrated that the 615 internal disagreement rate was 5%. All other frames were scored once. Size, speed, and shape 616 measurements were only made on individual frames that scored as polarized. 617 Cell shape analysis 618 Cell shape analysis was based on principal component analysis (PCA) of 2D cell 619 outlines, as previously described (Pincus & Theriot, 2007). Images were processed into vector 620 representation for PCA using custom code written in Matlab R2018b (available upon request, 621 Figure S3A,C). All analyses were performed on stacks cropped to 300 x 300 pixels in the XY-622 plane, centered on the membrane channel centroid identified during tracking as described 623 above. For XY-plane shape analysis, the z-stack in the cytoplasmic mCherry or membrane 624 channel (CellMask Deep Red) was converted into a maximum-intensity Z-projection, gamma-625 corrected, and segmented using an Otsu threshold. For XZ-plane shape analysis, the 626 cytoplasmic mCherry channel was deconvolved with Huygens software (SVI) using a classic 627 maximum-likelihood estimation and an empirically measured point-spread function. The 628 deconvolved stack was gamma-corrected and segmented with Otsu thresholding. All 629 segmented images were manually inspected using an overlay with the raw signal, and frames 630 with large defects were removed from further analysis. Velocity alignment was performed on 631 both the XY-plane segmentation and the 3D segmented stack. Briefly, the image or stack was 632 rotated in the XY-plane, so that the instantaneous velocity vector was aligned parallel to the 633 positive X-axis (i.e., pointing right horizontally, Figure S3C). The cells were translationally 634 aligned by their centroid in all three dimensions, so that the shape analysis was conducted in 635 the cell frame of reference. After velocity alignment, the 3D segmented stack was converted into 636 an XZ-plane cross-section by making a maximum-intensity Y-projection. This XZ-plane 637 projection may overestimate lamellipodial thickness because zebrafish keratocytes 638 characteristically form wrinkles due to periodic detachment of the thin actin sheet from the 639 substrate (Lou et al., 2021). 640 The aligned, binarized, and projected cell shapes from both the XY-and XZ-planes were 641 then converted into 300-point contours, which outlined the boundary of the cell with point 1 642 assigned to the center of the cell front (as determined by where the velocity vector crossed the 643 contour). These contours were outputted as 600-point vectors (consisting of the X-and Y-644 coordinates of each of the 300 points). Contour-vectors from all frames that scored as polarized 645 were imported into CellTool (Pincus & Theriot, 2007) for two separate PCA runs, one for the XY-646 plane and one for the XZ-plane. Before-confinement and during-confinement cell shapes were 647 pooled for their PCA runs ( Figure 3C) and separate PCA runs were conducted with images 648 pooled from before-hypertonic, during-hypertonic, and isotonic treatments (Figure S4C,D). 649 In vivo cell shapes were segmented manually using the quick selection tool in 650 Photoshop 22.1.1 from cell clusters in the skin of mosaic fish during 0-10 minutes post 651

wounding (Microinjection for mosaic expression, Tissue laceration). Basal cells (keratocytes) 652
were imaged either using LifeAct-mNeonGreen or cytoplasmic mNeonGreen and membrane-653 targeted mRuby3 (Figure S3B). Movies were converted into maximum-intensity Z-projections. 654 Although transgene-expressing cells tended to cluster together clonally, entire cell shapes, 655 including cryptic lamellipodia, were visible in the LifeAct channel because the lamellipodia were 656 brighter than neighboring cell bodies. In larvae expressing the cytoplasmic and membrane 657 markers, lamellipodia could not be visualized beneath neighboring cells, so only the leading cell 658 in each cluster was analyzed. These segmented images were naturally aligned so that the 659 direction of the wound was parallel to the isolated keratocytes' direction of instantaneous 660 velocity. Therefore, the segmentations were converted directly to contours using Matlab and 661 projected into the before/during-confinement shape space using CellTool. 662 Phase contrast images were segmented using the Directional Gradient Vector Flow 663 Procedure (Seroussi et al., 2012), implemented using Matlab R2017b. For Figure 2C, each 664 image shown in Figure 2B was registered to the masked bead using a custom Matlab script, 665 then displayed as overlaid contours using CellTool. 666 Volume and surface area measurements 667 Volume and surface area were measured from 3D reconstructions of spinning disk 668 confocal images. Deconvolved and segmented 3D stacks of the cytoplasmic mCherry channel 669 were prepared as described above (see Cell shape analysis). We tessellated the segmentations 670 into triangulated surface meshes using custom code written in Python 3.7.1, using the vtk 671 module and tools from the Allen Institute for Cell Science Spherical Harmonics Parameterization 672 package (Viana et al., 2021). The volume and surface area of the mesh are reported as the cell 673 volume and surface area. 674

675
Each larva or isolated cell sample was considered an independent biological replicate, 676 and multiple cells were measured in each larva and sample as technical replicates. Individual 677 frames were averaged to calculate cell measurements, and individual cells were averaged to 678 calculate sample measurements. Then, measurements were plotted either on a per-frame 679 ( Figure 3D-E), per-cell (dots in Figure 4, Figure 5B-E), or per-sample basis (triangles in Figure  680 4-5), using a "SuperPlot" style (Lord et al., 2020). 681 For Figure 3D and Figure Figure 4E-G were calculated as two-sided t-tests against the 690 null hypothesis that r = 0 (i.e., that the two variables are uncorrelated), using Prism 9 691 (GraphPad). 692

693
The data that support the results of this study are available from the corresponding author upon 694 reasonable request. 695     Idealized outlines along each axis were generated by varying that shape mode while holding the coefficients of the other modes at zero. The example outlines within the plot correspond to the cell from the left half of (A). Dots: observations of cell shape (before, n = 225; after, n = 119; with partial pairing and repeated cell measurements), from 8 samples. Effect sizes for all shape modes are shown in Table S1.
(E) Plot that overlays 1) Unconfined and confined isolated cell shape coefficients, and 2) Shape coefficients of in vivo cells, measured by projecting their outlines into the shape space shown in panel (C). In vivo outlines were segmented from the first 10 minutes after wounding 3 dpf larvae mosaically expressing either a LifeAct marker or cytosolic and membrane markers (Figure S3B, Methods). Dots: observations of cell shape, including repeated cell measurements (n ≥ 135 observations for each condition). Example outlines within the plot: cells that have nearly average coefficients for their respective conditions along XY shape modes 1 and 3. Error bars: mean and standard error of the mean (SEM) for the condition indicated by the nearby example cell outline. Means and SEMs were calculated from 12 unconfined samples, 11 confined samples, and 4 larvae. Statistical comparisons are shown in Table S2. Centroid speed was measured over a 60-second interval, 5-10 minutes before confinement ("unconfined") and again 6-10 minutes after the ceiling was lowered ("confined"). Paired t-test conducted on 10 sample averages (n ≥ 3 cells from each sample), with p < 0.0001.  (A) Keratocytes isolated from 2 dpf larvae expressing mCherry were imaged during a media shift to either the original isotonic medium or media supplemented with 20-50 mg/ml sorbitol.
Plot indicates sample-averaged fold-changes, with osmolarity estimated from media and sorbitol concentrations.
(B) Detailed volume comparison of cells before and after hypertonic shock. Volume was measured using 3D reconstructions of cytoplasmic mCherry. Volumes were averaged between 2-8 minutes preceding the media swap ("before"), and from immediately after the media swap until cells left the field of view or 30 minutes had elapsed ("after"). Paired t-test conducted on 11 sample averages (n ≥ 3 cells from each sample), with p < 0.0001.
For (B-D), dots indicate individual cell measurements and triangles indicate sample averages.
The lines connect the same sample across conditions. Cell and sample values are colorcoordinated, with each color representing a different sample.
(C) Pairwise surface area comparison of cells before and after hypertonic shock, measured and sampled as in (B). Paired t-test, with p > 0.05 (ns: not significant).

(D)
Pairwise speed comparison of cells before and after hypertonic shock. Centroid speed was measured over 60-second intervals, spaced out by longer z-stack acquisitions. Before speeds were averaged within 6 minutes preceding the media swap. After speeds were averaged from 3 minutes after the media swap until cells left the field of view or 30 minutes had elapsed. Paired t-test conducted on 11 sample averages (n ≥ 3 cells from each sample), with p = 0.0002.
(E) Plot of shape coefficients before and after hypertonic shock, from the top four shape modes in the XY plane (left half) or XZ plane (right half), showing little change between before and after. Effect sizes for top six shape modes are shown in Table S3. Idealized contours for each shape mode are shown in Figure S4D,E. Dots: average shape for each cell during observation before and after hypertonic shock (before, n = 145; after, n = 83; with partial pairing), from 14 samples.