Author response:
[Editors' note: The authors included an author response to reviews from another journal]
Reviewer #1 (Comments to the Authors):
In this manuscript the authors describe that cells in collective movements adopt a superdiffusive behavior to out pace individual cells. This behavior is regulated by cell-cell junctional stability and force transmission. The authors state that speed is regulated by vinculin through mechanosensitivity.
While is makes intuitive sense that cells may move more efficiently collectively as it reduces their exploratory space and therefore increases their efficiency of movement,
We agree that this is an intuitive explanation. However, previous literature had shown that confluent cells may or may not migrate depending on conditions that do not solely depend on the space available per cell, but also involve the intrinsic activity of the cell, its cortical tension, and its adhesion with its neighbors, with sometimes counterintuitive effects (doi: 10.1016/J.CEB.2021.07.011). This was the reason that motivated us to investigate how these various ingredients affected space exploration efficiency on different time scales.
Our results indeed refute the intuition that cells move more efficiently when their exploratory space is reduced by showing that the outcome depends on the time scale considered (Fig. S3B). Specifically, on short time scales (less than 3 hours), the area explored by individual MDCK cells is larger than that explored by MDCK cells at confluence. On a longer time scale (greater than 3 hours), however, the area explored by confluent MDCK cells is larger. This switch is a direct consequence of the change in migratory behavior from persistent random walk to superdiffusion, Moreover, its position in time depends on the cell line: extrapolation of our results on RPE-1 cells suggests that it should theoretically occur after approximately 300hrs, if this time scale was experimentally accessible (Fig. S3F).
…the role of junctions specifically is less clear.
We are sorry that we were not able to clearly convey the roles of junctions. We have substantially rewritten our text to address this and all the changes are highlighted in orange. As summarized in Fig. 6F, junctions have three roles. The first role is on persistence, through velocity coordination between neighbors, the second is on speed, through the stability of junctions, and the third role is on directionality, through the sensitivity of the monolayer to the wound edge.
The first role is evidenced thanks to the comparison of the MSD between single cell and confluent migration assays and the use of the alpha-catenin KD cell line. Alpha-catenin depletion is known to be the most potent disruptor of adherens junctions (DOI:10.1091/mbc.e06-05-0471, , DOI:10.1126/science.aaf7119, (DOI:10.1073/pnas.1002662107, DOI:10.1073/pnas.1119313109), and we show that it significantly alters the superdiffusive behavior that emerges in the confluent migration assay (Fig. 3E,F, 5C). Therefore, junction integrity is critical for the control of cell persistence.
Moreover, alpha-catenin depletion induces a loss of velocity coordination between neighbors (Fig. S3E), which we show through numerical simulations to induce superdiffusion (Fig. 3G). By contrast, E-cadherin KO and vinculin mutants have no effect on the superdiffusion of confluent cells (Fig. 3E, 4A). Therefore, the critical molecular ingredient is the link provided by alpha-catenin to the cytoskeleton that provides junction integrity.
The second role of junctions is evidenced thanks to the comparison of cell speeds between single and confluent migration assays with the vinculin mutants (Fig. S4A). Results show that cell speed is reduced of about 10µm/h by confluence, regardless of the mutant except for YE, whose only difference with other mutants is its lower stability (Fig. 4F). This supports that junction stability, and not the other effects of mutants, controls cell speed (we provide a detailed demonstration in the response to the following question). As expected, junction integrity is required as well, as seen from the higher cell speed of the alpha-catenin KD cell line compared to WT (first MSD point in Fig. 3B, E).
The third role of junctions is evidenced thanks to the comparison between confluent and directed migration assays (Fig. 6A). Results show that the wound healing rate is proportional to cell speed at confluence, regardless of the mutant except for YE, which displays no tension gradient in junctions from front to back cells (Fig. 6C). This supports that such gradient is required for cells to identify on which side is the wound edge. As expected, junction integrity is required as well, as seen from the loss of directional bias of the alpha-catenin KD cell line (Fig. 5F).
The authors chose vinculin as the basis by which to manipulate tensions at cell-cell junctions, but this comes with considerable drawbacks. Namely, since vinculin appears at both cell-cell and cell-matrix junctions, its role and the role of its mutations is not clear here. The authors state that the collective migration speed is related to junctional stability, but because vinculin is also at FA, how can this be concluded?
We apologize for the lack of clarity. We hope that the highlighted changes in the revised manuscript will improve this point. As exemplified above, comparing cell migration between isolated cells and confluent cells is essential to enable us to distinguish between the contributions of AJs and FAs. Indeed, since isolated cells lack AJs, the impact of vinculin mutants on single cell migration can only be explained by their effects on FAs. This is how we first determine the effects of vinculin mutants on migration that depend on FAs. Because confluent cells also have FAs, we expect that the effects of vinculin mutants on the migration of isolated cells will still be present in confluent cells, to which will be added the effects of these mutants on AJs and their consequences on migration, if any.
Therefore, when compared to WT cells, if a given mutant decreases or increases migration speed in individual cells, and does so in confluent cells in the same proportion, then its effects at confluence can be entirely explained by its effects in individual cells, and there are no additional effects of that mutant from AJs. This is indeed what we observe for all mutants except the YE mutant (Fig. S4C), leading us to conclude that none of the vinculin mutants, except the YE mutant, have an effect on migration at confluence that results from AJs. In contrast, the YE mutant has effects on migration at confluence that cannot be explained by its effect on individual cell migration. Therefore, the effects of YE at confluence depend on AJs, whether they result from alterations in AJs, FAs, or both. To distinguish between these scenarios, we proceed by elimination, comparing the effects of YE to those of other mutants on force transmission and adhesion stability, and how these two factors associate with migration speed, as explained below. In FAs, YE alters force transmission differently in individual cells and at confluence, but we already know from Fig. 2 that force transmission in FAs cannot alone explain the speed of migration. This result rules out an indirect effect of AJs on cell migration at confluence through FAs. Furthermore, in AJs, YE affects stability and force transmission, but TL has the same effect on force transmission as YE and we already know that none of the effects of TL on migration depend on AJs (Fig. 3, S4C). This result rules out an effect of force transmission in AJs on migration speed at confluence. We conclude that stability at the AJ level, which is the remaining property specifically impaired by YE, is what regulates migration speed at confluence.
The manuscript's logic and flow are not clear in some places, making the story hard to follow. As one example, the FRAP data, which the authors suggest is used to investigate vinculin's combined role does not help in this capacity as the interpretation and its connection to the bigger story are not clear.
We are sorry again for the lack of clarity. We used FRAP data to evaluate the effects of vinculin mutants on adhesion stability. Indeed, mutants have different effects on adhesion stability (Fig. 2E, 4F). In addition, they also have different effects on force transmission (Fig. 2D, 4D,E). The partial overlap in functional alterations caused by the mutants allows us to test the involvement of the overlapping function (here stability) in the overall migration outcome. For example, if two mutants have a similar effect on adhesion stability but different effects on migration speed (such as TL and T12), we can then rule out that speed results from adhesion stability. Similarly, if two mutants have different effects on stability but a similar effect on speed (such as TL and YE), we can also rule out that speed results from stability. We applied the same reasoning to force transmission to conclude that neither adhesion stability nor force transmission alone is sufficient for cells to migrate rapidly. However, the combination of the two enables rapid migration.
As another example, the information derived from the use of the mutants is not clear in the context of the message in the manuscript since they affect cell-cell and cell-matrix junctions and in some places show results that are counter intuitive and not well-explained, to which the authors admit they are surprising but then do not explain their meaning.
As such, it is very hard to follow the logic with regard to the information resulting from the mutant experiments.
We provide above a detailed break-down of our strategy to analyze the results. We regret that our manuscript did not adequately convey our conclusions and we hope that the new version of the manuscript improves this point.
Proliferation has been shown to play a role in wound healing. Does proliferation change in the various conditions?
This is an important point. The average speed of cells at confluence is approximately 20 µm/h (Fig. 4B), which means that each cell moves approximately its own size in one hour. During this time, assuming a 16-hour cell cycle, 6% of the cells would have divided, each of them likely pushing one of its neighbors a distance equivalent to the size of a cell. Therefore, cell proliferation accounts for at most a few percent of the total cell movement. For this reason, we can assume that growth does not account for a large part of the movement we observe. This is consistent with previous work showing that proliferation does not contribute significantly to wound healing (DOI: 10.1073/pnas.0705062104, DOI: 10.1083/jcb.201207148).
Minor comments:
The authors should provide a better description of the mutants: what does a tailless mutant not bind, or bind differently? More context is needed to help interpret the results. While the mutants have all been published on before, it would be helpful to have more information here so that the manuscript is easier to follow.
We are sorry that the information we provided was insufficient. We have now detailed the mutations to help the reader understand how interactions are altered.
Figure 1A is not necessary. Figure 1 overall is fairly predictable as there have been many papers using the persistent random walk as the best model to single cell migration (dating back to the early 1990's). The authors define a new term, angular memory, which they show decreases with increasing delta t as one would predict.
We acknowledge that persistent random walks have already been observed for individual cells, as in references 3-4 cited in the introduction. Nevertheless, we believe that Figure 1 is important because not all cells migrate as persistent random walkers when isolated. Some migrate in a more exotic manner, resulting in superdiffusive behavior, as in references 5-8 cited in the introduction. Since we observe superdiffusive behavior at confluence (Figure 2), it was therefore necessary to show whether or not single cells were superdiffusive too. We also use this figure to introduce angular memory, a measure that, to our knowledge, has never been used before. According to intuition, it decreases to 0 for persistent random walkers, just as another resembling measure, velocity autocorrelation, would do. However, the angular memory of fractional Brownian walkers does not vanish with increasing delta t (Fig. 3D), while velocity correlation would, just as that of persistent random walkers. This difference makes angular memory much more appropriate for distinguishing between the two migration behaviors, and prompted us to introduce it in the first figure as a reference.
In the wound healing assay, which cells were measured? Leading edge or interior, and does it matter?
Figure 5A shows that cells behave differently depending on their distance from the wound. This is because the traces shown correspond to the first few hours of the movie, during which the cells at the front begin to move first. Figure S5A shows the speed of the cells over time after the wound and indicates that the cells reach a stable speed after approximately 3 to 4 hours. Figure S5B shows the speed of the cells as a function of distance from the wound at steady state. These results show that the speed of the cells no longer depends on the distance from the wound at this stage. As indicated in the “Materials and Methods” section, we only considered time points beyond this stage for subsequent analyses of population-averaged MSD and velocity presented in Figure 5, so the location of cells at the front or rear was irrelevant.
Reviewer #2 (Comments to the Authors):
To migrate cells must spatially explore their environments, a process that is guided by intrinsic signals (adhesive and mechanical properties, etc) and extrinsic (gradient cues) signals. This exploration can occur on the single or multicellular level. In this study, the authors examine the effect of cell-cell interactions, guidance cues, and cell mechanics in the exploratory capacity of MDCK cells. The authors show that cell-cell adhesion provides a "infinite directional memory for migration" and cell speed is dependent upon the focal adhesion stability, cell mechanics, and the mobility of adherens junctions-these processes are modulated by vinculin.
My three major concerns with the manuscript are as follows:
(1) While there is potential new information about the role cell-cell junctions and guidance cues play in cell migration, there is not enough NEW insight presented. Rather the role of vinculin in these processes is expected given what is already known about its ability to control focal adhesion stability, mechanics, and adherens junctions.
We agree that our cell migration results make sense based on the effects of vinculin mutants on the stability and force transmission of adhesions. Nevertheless, we argue that this was not the only possible scenario. Indeed, we find that none of the effects of vinculin mutants on AJs (except YE) have an impact on cell migration (Fig. S4C). One might have expected that the increased stability provided by the TL and T12 mutants would reduce the speed of collective cell migration, just as the YE mutant increased cell speed due to its altered stability. This is not what we found, and this reveals a nonlinear relationship between AJ stability and migration speed that could be investigated more thoroughly in future studies. Another example is that the effects of the mutants on force transmission in AJs do not impact migration speed at confluence but do impact directed collective migration (Fig. 6). One might have expected that vinculin-mediated force transmission in AJs would impact collective migration, whether directed or not.
More importantly, we show that the role of intercellular adhesion in cell migration is more complex than expected. Indeed, it depends on the timescale considered: intercellular adhesion is detrimental to short-term spatial exploration and beneficial in the long term (Fig. S3B). Such a timescale-dependent behavior is impossible to predict from previously known effects of the mutants or other molecular considerations. Furthermore, we show that this behavior can be fully explained by the coordination of velocities between neighbors, which depends on intact connections between AJs and the cytoskeleton via alpha-catenin, but is independent of vinculin mutants that connect AJs to the cytoskeleton in parallel with alpha-catenin. One might have expected these connections to also have an impact on velocity coordination, and thus on spatial exploration, but we show that this is not the case (Fig. 3). Finally, we show that directed collective migration has a negligible impact on cell exploration at our experimental timescale (Fig. 5), whereas we initially expected the wound to make migration more ballistic. This reveals that such a directional signal affects spatial exploration at much longer timescales than expected.
Overall, our results quantify the outcome of competing effects and provide timescales at which one effect outweighs the other in influencing cell migration. We believe this is an original approach that provides substantial new insights into collective cell migration.
(2) The phenotypes of the cells expressing the mutant vinculins varying greatly. These phenotypes are not addressed despite the fact that they could potentially complicate the analyses. For example, there are dramatic differences between focal adhesion numbers and sizes in the cells expressing the different vinculin mutants; cell spreading is also dramatically altered. Likewise, the T12 mutant vinculin has previously been reported to have increased adhesive strength, increased traction forces, and cell spreading. How does this knowledge change the interpretation?
We agree that vinculin mutants may have effects on the size and number of FAs, cell spreading, and traction forces that we do not examine here. These consequences can be explained by the effects of these mutants on force transmission in FAs and on their stability, which we report in our work. They do not affect our interpretations. Here, we provide a predictive model of migration speed based on the combination of two consequences of vinculin function, namely stability and force transmission. An interesting avenue for future research would be to assess whether these combinations can be reduced to a single higherlevel effect of vinculin on the cellular phenotype that would be sufficient to predict migration speed. This work remains to be done, as neither FA size and number, cell spreading, adhesion force, nor traction forces alone are sufficient to predict migration speed.
Along the same lines, it has previously been established that tagged version of vinculin do not efficiently integrate into adherens junctions. Published work from the Nelson laboratory suggests that GFP-vinculins do not localize to cell-cell junctions and work from other laboratories suggests localization occurs only when the endogenous vinculin is silenced.
We are aware that some GFP-vinculin constructs may not localize as well as the endogenous protein at AJs. This is due to the localization of the GFP tag on the head of vinculin and depends on the length of the linker between GFP and the head of vinculin. The longer the linker, the easier the interaction with AJ partners. Unlike these constructs, the vinculinTSMod sensors we use in our work do not carry a GFP on the head and do not suffer from the same limitations.
Furthermore, vinculin recruitment to AJs depends on force, with little or no recruitment when tension on the AJs is relaxed (DOI: 10.1038/ncb2055). Vinculin recruitment has in fact already been used as an indicator of AJ tension in Drosophila (DOI: 10.1038/s41467-01807448-8). Consequently, the amount of vinculin visible at the AJs varies depending on the tension exerted on the AJs, which our results confirm: vinculin is more difficult to detect at the AJs in cells located at the front of a wound than in those located at the back (Fig. 6B), which is consistent with the difference in vinculin tension between front and back cells (Fig. 6C) and to the E-cadherin tension gradient between front and back cells (DOI: 10.1083/jcb.201706013). Overall, these results show that vinculin is not always easy to detect at AJs, but this is due to the properties of vinculin, which the constructs we use reproduce better than previous constructs (see also below).
The images in figure S2 and the prebleach images in figure S4 do not show convincing localization of the mutant vinculins to cell-cell adhesions. This is of special concern given that YE mutant protein hardly has any discernable localization to cell-cell junctions; additionally, none of the mutant proteins were tested for their ability to co-localize with adherens junction components. This raises the question if the parameters being examined and the conclusions drawn from them are affected by a difference in localization.
We agree that the recruitment of vinculin at intercellular contacts may be difficult to see.
Besides force-dependent effects mentioned above, other factors are involved. The images shown in Figures S2 and S4 are from live cells in which cytoplasmic vinculin is still present, and its level proportional to the mobility of vinculin. Indeed, the TL and T12 mutants show a more marked contrast between intercellular contacts and the cytoplasm, which is consistent with their greater stability at AJs (Fig. 4F). Conversely, YE shows lower contrast, which is consistent with the lower stability of this construct at AJs (Fig. 4F). The FL construct lies between the two. As a result, the cytoplasmic content can variably mask vinculin recruitment at the AJs depending on the mutant.
We have now performed additional quantifications of mutant recruitment at intercellular contacts as a function of distance from the basal surface of the cells and relative to their recruitment in FAs, in live cells. Results are shown in the new Fig. S4F. We find that all the constructs are recruited to intercellular contacts with a density that is at most half of that in FAs and that varies along the height. FL shows the highest density, localized more apically, consistent with the localization of an AJ-bound actin belt. The mutants appear to be more homogenously distributed along the height of the lateral surface, which may be explained by their impaired autoinhibition (TL, T12), or mechanosensitivity (YE). This variability also contributes to the difficulty in seeing vinculin recruitment in all cells in a single z-slice.
To confirm the proper recruitment of vinculin constructs to AJs we have performed immunofluorescence against alpha-catenin and phalloidin on each of the stable cell lines. Results are shown in the new Fig. S4D and E. In these experiments, cell permeabilization allows for the release of some of the cytoplasmic pool of vinculin, which highlights the recruitment of all vinculin constructs to intercellular contacts. There, all vinculin constructs colocalize with alpha-catenin and F-actin, as expected. Additionally, images displayed are maximum intensity projections to mitigate recruitment variability along the height.
Overall, our results clearly support the localization of vinculin at intercellular contacts, and the differences between the constructs are consistent with the effects of their mutations.
(3) There is a lack of new mechanistic insight. Conclusions are made about a role of vinculin dimerization. This conclusion appears to be based upon the usage of the mutant version of vinculin Y1065. Did the authors directly measure the ability of this mutant protein to dimerize? Is actin binding also affected.
The binding properties of the Y1065E mutant, including its dimerization and binding to actin, have already been characterized by other researchers (ref. 40 in our manuscript, as well as DOI:10.1111/j.1432-1033. 1997.01136.x or DOI: 10.1016/j.febslet.2013.02.042). We assumed that these properties are now well established and can be used to explain higher-level phenotypes that we show for the first time, to our knowledge.
Reviewer #3 (Comments to the Authors):
Canever et al. tracked two epithelial cell lines on collagen coated glass and showed that isolated cells (non confluent) move as persistent random walkers, whereas confluent monolayers migrate super diffusive, with long range directional memory. By systematically perturbing adhesion machinery they found that focal adhesion mutations mainly tune the speed of single cell tracks, but cannot create long range memory, while force bearing adherens junctions are essential for the super diffusive regime-genetically perturbing them collapses collective memory. These interesting results identify junctional tension as important to switch epithelial cells/sheets between individual and collective search modes - an important quantitative insight that is of clear relevance to cell biologists.
- The presented data is nicely quantitative and convincing, but I have subtle concerns about the generality of the findings. While the authors show that the differential behavior, they describe is not cell-line specific (MDCK, RPE), there are no experiments evaluating the generality of their conclusions across different matrix conditions. How are the measured migration parameters affected by matrix stiffness? Cell migration on collagen coated glass coverslips is a relatively narrow and artificial condition. How is the collective directional memory expected to behave on softer substrates? The generality of the conclusions could be strengthened by repeating measurements using hydrogels of varying stiffness. Further, it should be discussed to which tissues in the body the selected matrix conditions and migration modes plausibly apply.
We agree that the generality of our results and the relevance of glass-rigid substrates is an important point. In vivo, epithelial cells rest on a basement membrane with a typical stiffness of approximately 10 MPa, as demonstrated by experimental evaluations on various tissue explants, including renal glomeruli and Bruch's membrane, which are relevant to MDCK and RPE-1 cells (DOI: 10.1111/j.1742-4658.2007.05823.x, DOI: 10.1172/JCI106898, DOI:10.1038/eye.1987.35), we have added these references in the manuscript to support our experimental strategy. In vitro, the most significant effects of substrate stiffness on FAs and cell migration generally occur at much lower stiffnesses, between 0.2 and 100 kPa, and cell phenotypes generally plateau at levels comparable to those observed on glass, even below 100 kPa (DOI: 10.1242/jcs.133645, DOI: 10.1038/ncb3268, DOI:10.1039/c5ib00307e, DOI: 10.1039/c9sm01893j). Furthermore, substrate stiffness has much more moderate effects on confluent cells than on isolated cells. For example, it has been previously demonstrated that confluent layers of MCF10A epithelium showed no change in velocity coordination in the range of 3 to 65 kPa (DOI: 10.1083/jcb.201207148). Therefore, collagen-coated glass appears to be a reasonable model for the basement membrane. Overall, we believe that we have conducted our experiments under physiological conditions, and that our results apply to a wide range of substrate stiffnesses.
- It would be nice to see how long it takes confluent cell layers to close rectangular wounds of defined size when cells migrate as individual (adherens junctions perturbation) versus collective (wt) (on substrates of different stiffness). Presumably, there should be faster wound closure under the collective regime, at least for simple shaped wounds.
This is an interesting question, which our results indirectly address. In our study, we measured the wound healing speed of the WT MDCK cell line as well as lines expressing mutant vinculin constructs (Fig. 6A). These results show that this speed ranges from 5 to 15 µm/h depending on the construct expressed (and for reasons that we explain in the manuscript). These values make it easy to estimate the time required to close a wound based on its width. For example, it would take 5 hours to close a 100 µm wide wound for the WT cell line, which has a rate of 10 µm/h (on both sides of the wound).
Wound closure for cells with disrupted adhesive junctions has already been documented (DOI: 10.1083/jcb.200910041). The results show that wound closure is indeed slower than with WT cells. Although this previous study does not reveal the underlying causes, our work now shows that there are two factors: weaker directional memory due to impaired intercellular coordination and, in the longer term, an additional lack of sensitivity to the guidance signal provided by the wound.
- Akin to substrate stiffness variation, I am missing experiments that test the effect of cytoskeletal tension on these migration modes. Experiments with Rho kinase or myosin inhibitors could meaningfully broaden the scope of this study.
Rho kinase or myosin inhibitors applied to cells during the time required to assess migration patterns (a movie recorded overnight is necessary to obtain a statistically reliable calculation of MSD over 3 to 4 hours) are likely to affect many other cellular processes in addition to the cytoskeletal tension directly involved in migration. We believe that the accumulation of these effects will make interpretation of the results very difficult. For example, it has been shown that inhibition of ROCK by Y27 promotes healing of corneal endothelial lesions by affecting proliferation through cyclin D and p27 (DOI: 10.1167/iovs.13-12225), or by improving respiration, which would provide the energy necessary for migration (DOI: 10.1096/fj.202101442RR). Consistently, another study on HaCaT epidermal cells confirms that myosin phosphatase accelerates wound healing through proliferation (DOI: 10.1016/j.bbadis.2018.07.013). In contrast, in HUVEC cells, ROCK inhibition significantly impaired the proliferation and migration of vascular endothelial cells in vitro in a dose-dependent manner (DOI: 10.1097/ICO.0000000000000493).
Furthermore, previous studies have highlighted that differential contractility at the subcellular level is important for collective migration (DOI: 10.1038/ncb2133, DOI: 10.1083/jcb.201706013), which is not possible to examine with global activation or inhibition of contractility. This prompts the development of more refined and specific measurement and disruption strategies to assess the respective impact of cytoskeletal tension on cell-cell and cell-matrix adhesion mechanisms. Our work, which uses biosensors to assess how this tension differentially affects cell-cell and cell-matrix adhesions, is a step in this direction. The localized spatio-temporal activation or inhibition of myosin subtypes or Rho GTPase regulators specific to these adhesion structures will likely answer these questions in the future, but we believe that the development and application of these approaches will require a substantial amount of work that goes beyond the scope of our study.
