Beneficial compaction of spinal cord lesion by migrating astrocytes through glycogen synthase kinase-3 inhibition

The migratory response of astrocytes is essential for restricting inflammation and preserving tissue function after spinal cord injury (SCI), but the mechanisms involved are poorly understood. Here, we observed stimulation of in vitro astrocyte migration by the new potent glycogen synthase kinase-3 (GSK-3) inhibitor Ro3303544 and investigated the effect of Ro3303544 administration for 5 days following SCI in mice. This treatment resulted in accelerated migration of reactive astrocytes to sequester inflammatory cells that spared myelinated fibres and significantly promoted functional recovery. Moreover, the decreased extent of chondroitin sulphate proteoglycans and collagen IV demonstrated that scarring was reduced in Ro3303544-treated mice. A variety of in vitro and in vivo experiments further suggested that GSK-3 inhibition stimulated astrocyte migration by decreasing adhesive activity via reduced surface expression of β1-integrin. Our results reveal a novel benefit of GSK-3 inhibition for SCI and suggest that the stimulation of astrocyte migration is a feasible therapeutic strategy for traumatic injury in the central nervous system.

The findings build upon previous studies which have reached a similar conclusion regarding the role of astrocytes in spinal cord injury and repair, as reviewed by many of the authors of this ms (Renault-Mihara et al., 2008; cited in manuscript) as well as White and Jakeman (Restor Neurol Neurosci. 2008;26(2-3): 197-214). However, previous studies used either genetically modified mice or transplantation approaches, in contrast to the pharmacologic enhancement of glial cell migration in the present study. A previous study (Dill et al., 2008, cited in ms) reported that GSK-3 inhibition improved neurite and axon growth in vitro and in vivo and improved functional outcomes in rats following contusion injury to the thoracic spinal cord, although this paper did not comment on the contribution of astrocytes to the functional recovery. The role of GSK-3 in regulating the Wnt-betacatenin pathway and the role of this pathway in cell migration is well-characterized, although effects of GSK-3 inhibition on astrocyte migration do not appear to have been examined previously. The finding that GSK3 inhibition decreases beta1-integrin expression is novel. Comments: 1. For figure 1C, it is unclear if the authors measured axon length of neurite length. Axon length is referred to in the results section, while neurite length is mentioned in the figure and methods. 2. Supplemental Fig 2 demonstrates that sustained inhibition of GSK-3 by Ro330354 decreased astrocytic recolonization following a scratch wound injury in vitro. This contrasts with the increased migration observed using the Boyden chamber/transwell assay. The authors explanation for this discrepancy is that intercellular contacts are required for the effective recolonization of wounded astrocyte monolayers (page 6) and that glial cells migrate as single cells during development (Klambt, 2009) (Page 15). This explanation seems unsatisfactory, as the disruption of intercellular contacts should allow the astrocytes to migrate as single cells. Moreover, the scratch wound assay seems more relevant to SCI than does the developmental migration of astrocytes. Additional discussion of these contrasting results is warranted. 3. Another discrepancy, as noted by the authors, is that the 5 days postinjury administration of Ro3303544 contrasts with the increased compaction of inflammatory cells at 14 DPI, but not at 7 DPI. The explanation is the time required for the upregulation of ECM proteins, such as beta1integrin, which is demonstrated to occur at 10 DPI. However, levels of beta1-integrin were not examined at earlier time points. To support the authors explanation, additional data is necessary to demonstrate the time course of beta1-integrin upregulation. 4. It is unclear whether the vehicle for Ro3303544 contained DMSO for both in vitro and in vivo studies (pages 18 and 19). This should be clarified and the DMSO concentration specified. 5. For the in vivo studies, it would be helpful to demonstrate drug efficacy by showing beta-catenin accumulation in the nucleus of spinal cord neurons. 6. The sample size for some studies is unclear. In Fig 3B, the results are the mean +/-SEM of one experiment performed in quadruplicate. It is unclear if the results are from four separate animals or from one animal. What is the sample size for Fig 3C and 3E? 7. Methods for measurement of Luxol Fast Blue area should be described, including how the ROI was determined and controls for the plane and level of the longitudinal sections, as sections through dorsal cord would give very different results from sections through ventral cord. LFB area is typically measured is cross sections and not longitudinal sections. The cross section measurement at the lesion epicenter is subject to artifacts associated with tissue shrinkage (see the Scheff et al., 2003 paper cited). Total volume of injured or spared tissue (total, gray matter, white matter) is preferred and correlates better with functional deficits than do area measurements at the injury epicenter. 8. The Basso Mouse Scale should be used for evaluation of locomotor function in mice following SCI, not the BBB scale which was designed for rats. 9. Data regarding the injury parameters (force, displacement, velocity) obtained using the impactor device should be reported. This could be included in the supplementary information. 10. The decreased collagen IV immunoreactivity following GSK3 inhibition observed in the present study contrasts with the increased collagen production previously observed in GSK3beta deficient mice (page 9). This discrepancy should be discussed. 11. The glial response to SCI varies widely among strains (see White and Jakeman review cited above). Thus, the authors should comment on the relevance of these results to other strains and species, as well as to human SCI.
Referee #2 (Novelty/Model system Comments): see remarks below Referee #2 (Other remarks): The role of reactive astrocytes in the process of walling off inflammatory cells and preserving function following spinal cord injury has been well established. Following injuries to the CNS that open the blood brain barrier astrocytes migrate away from the core of the lesion to encircle the zone of inflammation, protecting the remaining tissue. Taking advantage of a new specific inhibitor of GSK-3, Ro3303544, the authors have investigated the effects of GSK-3 inhibition on astrocyte migration. The authors report that sustained inhibition of GSK-3 stimulated astrocyte migration in vitro led them to administer Ro3303544 after contusive SCI in mice and examine the consequences of this treatment in vivo. While the positive in vitro data was quite nice, I didn't think there was a need to report so much of the negative data. This seemed superfluous to me. The authors report an interesting enhancement in the compaction of inflammatory cells by reactive astrocytes in vivo after treatment with this drug. The results were clear cut. BRDu experiments showed that the enhanced compaction phenomenon was not due to increased mitosis but rather to a change in migratory behavior. Also, the effect of the drug in reducing CSPGs and preserving myelinated fibers in the vicinity of the scar was especially interesting. In addition, a variety of experiments were performed suggesting, but not proving that decreasing integrin expression and reduced adhesion was part of an underlying mechanism of how the drug resulted in enhanced migration in vivo. I wondered if the authors have access to a TIRF microscope so that they could observe the close contacts of astrocytes with the substrate in treated versus control cultures during there scratch assay. Also, the authors might consider doing some kind of shaker or forced flow assay in an attempt to show that it is easier to disrupt astrocytes from the culture dish in the presence of the drug. These kinds of assays would give a direct assessment of the strength of adhesion rather than an indirect one. Finally, the report of improved behavior after drug treatment makes this paper especially worthy. The authors have done a rather poor job of discussing the results thoroughly and citing some of the relevant literature pertaining to this work. The authors report that certain inhibitory ECM molecules such as CSPGs are reduced in treated animals but there is no discussion of work leading up to why the authors examined these molecules or how such reduction might be therapeutic. Papers from the Silver lab such as those by Fitch et al and Busch et al showing the effects of inflammatory cells on astrocyte migration and axonal dieback are appropriate.

Referee #3 (Other remarks):
Renault-Mihara et al. This is a generally well written article of significance in the field of brain and spinal cord neurotrauma. The authors examined effects of Ro3303544, a specific GSK-3B inhibitor, on astrocyte migration in vitro and on functional recovery after contusive spinal cord injury in vivo. The studies are carefully designed and the results demonstrate that prolonged exposure to this compound induces appropriate signaling pathways and increases astrocyte migration in vitro. When it is administered daily for 5 days by i.p. injection beginning immediately after injury, the compound reduced the size of a spinal cord contusion lesion resulting in improved functional recovery.
The results are novel because they demonstrate for the first time that GSK-3B inhibition might contribute to recovery after trauma in part through actions on astrocytes. Previous cited studies have suggested that relevant targets of this pathway might include fibroblasts, endothelial cells and neurons, and inhibition of GSK-3B has anti-inflammatory effects, including reduced activation of NFKappaB, reduced release of TNFalpha and other cytokines, and attenuation of reactive oxygen species. In addition, Adachi et al have demonstrated previously that i.p. or i.c.v. administration of Ro3303544 stimulates neuroprogenitor cell proliferation in the adult mouse brain (2007). The present study supports the addition of astrocyte migration to the long list of possible targets of GSK-3 inhibition as a therapy for repair. However, the possible action of this compound on other sensitive cells in the complex in vivo SCI environment is only superficially addressed in the discussion. In addition, examination of the dosing regimens required for effects on astrocyte migration in vitro suggest that astrocyte migration this is not the sole target of action in the animal.
Major concerns: 1. The specificity and dose response study of Ro3303544 illustrated in Figures 1 and 2 show effects on neuronal and astrocyte nuclear translocation of beta-catenin after several days exposure at 500 nM and 1 micromolar, respectively. These are surprisingly high doses for a compound with a reported IC50 on kinase inhibition of 0.6 nM. The paper should include some comparison of these effective doses with effects on other cellular processes that are relevant for wound healing events. 2. Supplemental figure 2 illustrates a dose-depended inhibition of astrocyte migration in a wound scratch assay. The prepared astrocytes exhibit morphological characteristics of reactive astrocytes, and they are maintained in the presence of 10% FBS, which maintains them in a non-migratory state. In fact, there is virtually no migration of astrocytes into the wound under these conditions as is illustrated by the counts of <100 cells/mm2 in control conditions. Thus, documenting the inhibitory effects in this assay is not informative. To evaluate the potential inhibitory effects of the compound on astrocyte migration in vitro, it would be more useful to add factors that induce migration in these cells as a positive control and then examine the effect of the inhibitor on a measurable wound closure rate. 3. There are no controls for the in vitro studies such as total cell counts or BrDU incorporation to confirm that the effects on cell density in the scratch are not due to decreased cell survival or proliferation. 4. The neuroanatomical analysis of lesion size from Figure 4 is based on a single saggital section per animal. The lesion site is irregular and variable in shape and the treatment is expected to affect morphology and migration of border cells. A wide range of sample sizes could be illustrated in any one animal by taking sections from different medial-lateral sites. Precise mid-saggital sections are not always possible (see ctrl 14 DPI), and even when they are available, there is no guarantee that the area of the lesion in this section is precisely correlated to lesion volume. Therefore, a 3 dimensional measure of lesion volume is not only expected for state of the art in this field, but is also essential in order to evaluate the effects of the treatment on the lesion. 5. The description of methods used for unbiased counting BrDU+/GFAP+ cells in vivo is inadequate. Based on prior studies and the functional role of b-catenin signaling in cell proliferation, this is an important thing to understand. There are no examples of how double stained cells were identified and defined, what the positive cells look like at high power, or how the sampled regions were chosen to obtain double cell counts that were not biased by the amount of tissue compaction or size of the cells. 6. White matter sparing in figure 5 also suffers in accuracy of reporting because it is impossible to standardize and provide unbiased sampling of a complex 3 dimensional lesion using a single section. The area of positive staining is taken from a sample box of unknown size and location. 7. The behavioral results reported on page 9 paragraph 3 are not consistent with the graph in figure  3C. Specifically, the text indicates a difference in motor scores beginning at 7 dpi, this is not indicated by * on the graph and appears to have not reached significance and should not be reported as "superior motor recovery". Results should specify repeated measures ANOVA main effects in addition to post-hoc corrected test results. 8. The discussion should include other plausible targets of systemic GSK-3 inhibition on the complex cellular environment at the lesion site, especially including effects on progenitor cells and the activation state and/or migration of inflammatory cells. Only a cell-specific blockade of GSK-3 in astrocytes would specifically address whether this cell type is primarily responsible for the reduced lesion size in this complex lesion or whether other effects leading to reduced secondary injury contribute to a smaller lesion. The argument that there is no difference in area occupied by CD11b+ staining at 7 dpi indicates that the effects are not on inflammatory cells is inadequate. The distribution of these cells at one time point does not provide sufficient information regarding their activation state or functional characteristics. b. Assuming that there is no effect of the compound on cell proliferation and survival (see item 3 above), the graph in Figure 2D would be even more compelling if the cell area was expressed as area/cell, by dividing by the # of nuclei in the sample region. There is not sufficient description of the "semi-automatic quantitative analysis" in the methods or supplementary methods. The description of these effects should indicate that actin distribution appears unchanged.
c. Figure 2A illustrates the absence of an effect of 30 minutes of exposure of astrocytes to Ro3303544. None of the related experiments would suggest that this short exposure would have an effect. This figure adds nothing to the manuscript. d. The images in Figure 3C and 3D are badly out of focus. In addition, high power images of the lesion border of the time point most different between the two treatments (14 dpi) would help illustrate effects on astrocyte compaction, process morphology, and interactions between macrophages and astrocytes at the border. e. The authors are encouraged to read and consider use of the mouse locomotor scale for future studies (Basso et al., 2006). The BBB scale is designed to assess locomotor recovery following contusion injury in rats. Evaluation of hip and knee joint movement after spinal cord contusion in mice is considered to be unreliable; these differences affect the spread of scores at the lower end of the scale used for this study. Based on the scatterplot, it appears that a Chi squared analysis of the proportion of animals capable of weight supported stepping would be helpful as it might reveal an important functional milestone in the treated mice. We are now submitting the revised version of our manuscript, "Beneficial compaction of spinal cord lesion by migrating astrocytes through glycogen synthase kinase-3 inhibition", to EMBO Molecular Medicine. Firstly, we would like to thank the three referees for their detailed examination of our article, which greatly contributed to improving the impact of our study. You will see that we have positively addressed most, if not all, the numerous points raised by the referees. In particular, we performed extensive in vivo experiments that allowed us to provide immunohistochemical verification that GSK-3 inhibition is effective in vivo, 3-dimensional measurements of lesion volume, cross-sectional analysis of demyelination at various levels, as well as an evaluation of motor function based on the Basso Mouse Scale. Moreover, we have also analyzed the β1-integrin expression in the spinal cord at an additional time-point following spinal cord injury in both groups. Finally, we now provide additional, consistent in vitro data that better characterize the effects of the drug. Accordingly, we have modified the manuscript which now includes nine main figures and three supplemental figures, plus one supplemental movie. To address the stimulating issues raised by the referees, we have added an extended version of the Discussion to the Supporting Information instead of to the manuscript because of space limitations All these modifications and our detailed point-by-point answers to the referees' comments are included in the enclosed document. We hope that the resulting online review process file will be found interesting by readers of EMBO Molecular Medicine.
We look forward to hearing from you soon. Thank you very much for your consideration.
Point-by-point answers to the comments of the referees Referee #1:

Comments: 1. For figure 1C, it is unclear if the authors measured axon length of neurite length. Axon length is referred to in the results section, while neurite length is mentioned in the figure and methods.
We now state "neurite" in the Results section.

Supplemental Fig 2 demonstrates that sustained inhibition of GSK-3 by Ro330354 decreased astrocytic recolonization following a scratch wound injury in vitro. This contrasts with the increased migration observed using the Boyden chamber/transwell assay. The authors explanation for this discrepancy is that intercellular contacts are required for the effective recolonization of wounded astrocyte monolayers (page 6) and that glial cells migrate as single cells during development (Klambt, 2009) (Page 15
). This explanation seems unsatisfactory, as the disruption of intercellular contacts should allow the astrocytes to migrate as single cells. Moreover, the scratch wound assay seems more relevant to SCI than does the developmental migration of astrocytes. Additional discussion of these contrasting results is warranted. This is a very important point. The definition of migration as movement from one place to another led us to distinguish migration from movements such as membrane ruffling or protrusion. Accordingly, one can say that astrocytes are fairly immobile cells in vitro in the absence of any promigratory stimulus. In the wound scratch model, the wound initiates and imposes directionality to the movement of astrocytes with intercellular contacts with their neighbors. Inhibition of GSK-3 is known to drastically perturb the polarization of astrocytes, which is the first step of the recolonization process. In the present revised manuscript, we provided the supplemental time lapse movie which now clearly illustrates that the extension and directionality of the protrusions are both altered upon GSK-3 inhibition by Ro3303544. In a transwell assay, the migration of single cells is induced by the gradient of soluble molecules (chemotaxis: gradient of serum in our case) and nonsoluble substrate (haptotaxis: transwell membranes are coated by ECM proteins). When intercellular junctions are disrupted in the monolayers, no stimulus for migration remains, and the cells cannot exhibit the migratory potential that they would express if transferred to an ad hoc system such as the transwell in our case. In this revised manuscript, these points are addressed in the extended version of the Discussion in the Supporting Information (page 10; line13): 'Current data about the role of GSK-3 in cell migration are contradictory (Etienne-Manneville & Hall, 2003;Kapoor et al, 2008). Our observation that the acute inhibition of GSK-3 with Ro3303544 results in drastically compromised recolonization of wounded astrocytic monolayers (Fig S1A and supplemental movie 1), or alternatively in normal migration of astrocytes as single cells in transwell assay ( Fig S2E), demonstrates first that the effect of GSK-3 depends on the migration assay used for evaluation, and thereby very likely on the migration mode involved. In the wound scratch model, the wound initiates and imposes directionality on the movement of astrocytes, which have intercellular contacts with their neighbors. Inhibition of GSK-3 is known to drastically perturb the polarization of astrocytes in this model (Etienne-Manneville & Hall, 2003), as illustrated in Supplemental Movie 1.
In the transwell assay, the migration of single cells is induced by the gradient of soluble molecules (chemotaxis: serum gradient in our case) and non-soluble substrate (haptotaxis: transwell membranes are coated by ECM proteins). Unexpectedly, in this model the activity of GSK-3 was not required for a normal migration rate. A second major parameter for predicting the effect of GSK-3 inhibition on migration is the duration of the inhibition. The sustained inhibition of GSK-3 before beginning the assay indeed dramatically enhanced the migration of astrocytes in the transwell assay ( Fig 2B). The disruption of intercellular contacts ( Fig S1B) that are required for the initial polarization of cells in this assay (Dupin et al, 2009) prohibited de facto the evaluation of the outcome of sustained GSK-3 inhibition in the wound scratch assay. It remains to be determined whether the loss of adherens junctions between astrocytes upon GSK-3 inhibition results from changes in the stoichiometric ratio within N-cadherin/β-catenin complexes due to the transfer of β-catenin into another cellular pool (Kam & Quaranta, 2009), or is a consequence of β1-integrin down-regulation (Chattopadhyay et al, 2003. Although the in vivo pattern of endogenous astrocyte migration following CNS injury is unknown, glial cells migrate as single cells during development (Klambt, 2009). Moreover, the observed increased compaction of inflammatory cells in vivo after administration of Ro3303544 (Fig 4) suggests that the migration of astrocytes observed in the single-cell transwell assay more closely resembles the situation of in vivo reactive astrocytes, and reveals that in vitro wound scratch assay is not always relevant to in vivo CNS injury.'

Another discrepancy, as noted by the authors, is that the 5 days post injury administration of Ro3303544 contrasts with the increased compaction of inflammatory cells at 14 DPI, but not at 7 DPI. The explanation is the time required for the upregulation of ECM proteins, such as beta1-
integrin, which is demonstrated to occur at 10 DPI. However, levels of beta1-integrin were not examined at earlier time points. To support the authors explanation, additional data is necessary to demonstrate the time course of beta1-integrin upregulation. We accordingly examined the expression level of β1-integrin upon Ro3303544 treatment by immunoblot analysis and confirmed its significant downregulation at 5 DPI (new Fig 9B). This data rules out the possibility that the delayed compaction of the inflammatory cells, relative to the administration period of the drug, is related to a delayed pharmacokinetic effect of the drug. Thus, these observations strengthen the proposed molecular mechanism for the enhanced migration, which involves simultaneous Ro3303544-induced downregulation of β1-integrin and spontaneous upregulation of ECM proteins after injury. Please note that we have also corrected a previous mistake: the error bars on the histogram at 10 DPI in Fig 9B now Fig 3C) now shows the upregulation and nuclear accumulation of active β-catenin in the spinal cords of Ro3303544-treated mice at 4 DPI.
6. The sample size for some studies is unclear. In Fig 3B, Scheff et al., 2003 paper cited). Total volume of injured or spared tissue (total, gray matter, white matter) is preferred and correlates better with functional deficits than do area measurements at the injury epicenter.
As mentioned above (point #6), we now include a 3-dimensional measurement of the lesion volume ( Fig 4B) in the present revised manuscript. In addition, we also provided a new analysis of demyelination at 42 DPI based on eriochrome cyanine staining of transverse sections at various levels of the lesion (see new Fig 6A) (Fitch et al, 1999). While it is speculated that the physical contraction of the fibrous scar may also contribute to cavity formation in some previous reports (Klapka & Muller, 2006), two recent studies using different experimental approaches have observed that reduction of the scarring was associated with reduced cystic cavities in rat (Iannotti et al, 2006;Xia et al, 2008).

However, further studies in relevant models are needed to examine whether the stimulation of reactive astrocyte migration through GSK-3 inhibition does indeed limit the development of these cystic cavities.'
Referee #2

The role of reactive astrocytes in the process of walling off inflammatory cells and preserving function following spinal cord injury has been well established. Following injuries to the CNS that open the blood brain barrier astrocytes migrate away from the core of the lesion to encircle the zone of inflammation, protecting the remaining tissue. Taking advantage of a new specific inhibitor of GSK-3, Ro3303544, the authors have investigated the effects of GSK-3 inhibition on astrocyte migration. The authors report that sustained inhibition of GSK-3 stimulated astrocyte migration in vitro led them to administer Ro3303544 after contusive SCI in mice and examine the consequences of this treatment in vivo.
While the positive in vitro data was quite nice, I didn't think there was a need to report so much of the negative data. This seemed superfluous to me. We appreciate for the reviewer's constructive comments. To address this advice, we removed previous Supplemental Figure 1 that demonstrated similar migration capabilities of spinal and cortico-striatal astrocytes. Accordingly, we have now moved the following text from the Results section to the Supporting Information (cell culture sub-section, (Page 2, Line 17)): 'Primary cultures of spinal astrocytes require far more meticulous preparation than cultures of cortico-striatal astrocytes. Because astrocytes demonstrate regional heterogeneity, especially in regard to their adhesion properties (Denis-Donini et al, 1984), the relative migratory capabilities of these two types of astrocytes were compared. Both types projected indistinguishable, characteristic polarized protrusions in response to a monolayer scratch and recolonized the wounded area at similar speeds (not shown). Therefore, the effect of Ro3303544 was evaluated on cortico-striatal astrocytic monolayers for the in vitro studies related to the in vivo context of the spinal cord.' We understand that the referee was concerned that non-specialists on cell migration would be confused by the apparent contradiction between the inhibition of recolonization upon Ro3303544 treatment in the wound scratch assay and its stimulation in the transwell assay. However, given that the wound scratch assay is widely considered the most relevant in vitro model for mimicking in vivo injury (see remark #2 of Referee #1), we think that it is important to bring to the attention of the community that, in some cases, this assay is not informative and may even provide misleading results if the in vivo effect of Ro3303544 is considered. Also, in consideration of remark #2 of Referee #3, the revised version of the manuscript contains a new version of the Fig S2 and a time-lapse movie that shows the dynamics of astrocyte recolonization in a control setting and the altered polarization of the astrocytes in the Ro3303544 condition, as well as the absence of drug toxicity. Concerning the former Fig 2A, please see our response to Referee #3, minor concern c. We believe that Fig S3 is also informative because it demonstrates that different signaling pathways are activated following the inactivation/inhibition of GSK-3 in fibroblasts and astrocytes, which may explain the contrasting responses observed in vivo: fibrosis in the skin as opposed to reduced scarring in neural tissue (please also see our answer to point #10 of Referee #1).

The abstract has been edited as follows: 'A variety of in vitro and in vivo experiments further suggested that GSK-3 inhibition stimulated astrocyte migration by decreasing adhesive activity via reduced surface expression of β1-integrin.'
The following sentence has been added to the Discussion (Page 14, Line 3): 'Given the complexity of the lesion environment, as well as the number of molecules that potentially modulate the migration of reactive astrocytes, it seems plausible that the actual mechanism for the in vivo enhancement of migration by Ro3303544 is more complex than our proposed model.' Finally, the report of improved behaviour after drug treatment makes this paper especially worthy. The authors have done a rather poor job of discussing the results thoroughly and citing some of the relevant literature pertaining to this work. The authors report that certain inhibitory ECM molecules such as CSPGs are reduced in treated animals but there is no discussion of work leading up to why the authors examined these molecules or how such reduction might be therapeutic.

Papers from the Silver lab such as those by Fitch et al and Busch et al showing the effects of inflammatory cells on astrocyte migration and axonal dieback are appropriate.
Thanks to the comments of the referee, we have improved the Discussion. We invite the referee to read the extended version of the Discussion in the Supporting Information.

Referee #3
Major concerns: 1. The specificity and dose response study of Ro3303544 illustrated in Figures 1 and 2 show effects on neuronal and astrocyte nuclear translocation of beta-catenin after several days exposure at 500 nM and 1 micromolar, respectively. These are surprisingly high doses for a compound with a reported IC50 on kinase inhibition of 0.6 nM. The paper should include some comparison of these effective doses with effects on other cellular processes that are relevant for wound healing events.
We provide the requested dose-response studies in the new Supplemental Fig 2 (Fig S2) in the revised version. Fig S2A and S2B demonstrate statistically significant dose-dependent nuclear accumulation of β-catenin in astrocytes. The reviewer will see that statistical significance is reached at 100 nM Ro3303544. Although all migration assays were performed in the presence of aphidicolin, a potent anti-mitotic agent for astrocytes that does not interfere with astrocyte migration, as we previously confirmed (Renault-Mihara et al, Mol Biol Cell, 2006), we now include in vitro BrdU incorporation experiments in Fig S2C that show Ro3303544 dose-dependently stimulates the proliferation of astrocytes in vitro. Fig S2D shows that 48-h treatment with 100 nM Ro3303544 exerts a significant pro-migratory effect in transwell assay.

Supplemental figure 2 illustrates a dose-depended inhibition of astrocyte migration in a wound scratch assay. The prepared astrocytes exhibit morphological characteristics of reactive astrocytes, and they are maintained in the presence of 10% FBS, which maintains them in a non-migratory state. In fact, there is virtually no migration of astrocytes into the wound under these conditions as is illustrated by the counts of <100 cells/mm2 in control conditions. Thus, documenting the inhibitory effects in this assay is not informative. To evaluate the potential inhibitory effects of the compound on astrocyte migration in vitro, it would be more useful to add factors that induce migration in these cells as a positive control and then examine the effect of the inhibitor on a measurable wound closure rate.
Although it is true that astrocytes in 10% FBS do not exhibit their maximal migration potential in the wound scratch assay, their correct migration in the transwell assay may contradict the assertion that astrocytes in this condition are maintained in a non-migratory state. Nevertheless, we now provide a supplemental movie of the "fast migration condition" (based on the seminal article by Faber-Elman, J Clin Invest. 1996 Jan 1;97(1):162-71.) that clearly illustrates: 1) the normal dynamics of wound recolonization by astrocytes in control conditions, 2) the alteration of polarization events (reduced protrusions that exhibit defective directionality) in Ro3303544-treated astrocytes, and 3) the absence of drug toxicity during this period (no swelling, condensation, or vacuolization). Selected frames from this time-lapse movie are presented in the new Fig S1A. 3. There are no controls for the in vitro studies such as total cell counts or BrdU incorporation to confirm that the effects on cell density in the scratch are not due to decreased cell survival or proliferation.
Both the reported anti-apoptotic effect of GSK-3 inhibition after SCI in vivo (Cuzzocrea et al, 2006, cited) and our observation of reduced lesion volume in the spinal cords of Ro3303544-treated mice are not consistent with the decreased survival of astrocytes in the presence of Ro3303544. In our case, a reduction in the survival of astrocytes in the Ro3303544 condition is very unlikely for the following reasons: 1) treatment of hippocampal neurons, recognized as very sensitive cells, for 72 h with 1 µM Ro3303544 resulted in enhanced neurite outgrowth (Fig 1C), which is not a sign of toxicity; 2) time-lapse recordings of astrocytes, known as robust cells, have not revealed any cell exhibiting any sign of toxicity; 3) 48-h treatment of astrocytes promoted BrdU incorporation ( Fig  S2C); 4) images of nuclei in Fig S1B and C show that no nuclear abnormality (morphology, condensation, etc.) appeared upon Ro3303544 treatment; and 5) it is difficult to conceive that the same treatment with 1 µM Ro3303544 for 48 h would be toxic in a wound scratch assay and stimulate migration in the transwell assay. As mentioned in the methods, all migration experiments presented (both wound scratch assay and transwell) were performed in the presence of aphidicolin, which is a potent inhibitor of astrocyte proliferation (Renault-Mihara, Mol Biol Cell, 2006), thereby excluding any influence of proliferation on the observed effects. However, considering that Ro3303544 dose-dependently stimulated incorporation of BrdU in astrocytes in vitro (Fig S2C), we evaluated the influence of this proliferative effect of Ro3303544. Comparison of wound scratch recolonization in control and Ro3303544 conditions, in the presence and absence of aphidicolin, revealed that the inhibition of migration in this assay by Ro330354 (observed in the presence of aphidicolin) was not compensated for by the increase in proliferation mediated by Ro3303544 (not shown). In other words, in the absence of aphidicolin, Ro3303544 still inhibited the recolonization of the wound compared to the control condition.

The neuroanatomical analysis of lesion size from Figure 4 is based on a single saggital section per animal. The lesion site is irregular and variable in shape and the treatment is expected to affect morphology and migration of border cells. A wide range of sample sizes could be illustrated in any one animal by taking sections from different medial-lateral sites. Precise mid-saggital sections are not always possible (see ctrl 14 DPI), and even when they are available, there is no guarantee that the area of the lesion in this section is precisely correlated to lesion volume. Therefore, a 3 dimensional measure of lesion volume is not only expected for state of the art in this field, but is also essential in order to evaluate the effects of the treatment on the lesion.
This is a very good point. To address this issue, we have performed 3-dimensional measurements. Figure 4A and B show the new data. The methodology for both lesion volume and scar analysis is described in the Supporting Information (Page 5, Line 24): 'Three-dimensional analysis of lesion volume was performed as follows: five sagittal sections centred on the mid-sagittal axis and separated from each other by 200 µm were doubleimmunostained with anti-GFAP (1/500) and anti-CD11b (1/200) antibodies. Images were acquired with a Keyence Bz-9000 epifluorescence microscope and tiled with Keyence analysis software. GFAP-negative areas were manually traced using Image J software version 1.44e. The calculation of the lesion volume, approximated as the sum of conical frustra, was based on the following formula: , with B1 and B2 the areas of two consecutive sections, and h the distance between them (200 µm). At 7 DPI and 42 DPI, five mice were used per group, and at 14 DPI, 12 and 10 mice were used in the control and Ro3303544 groups, respectively. The extent of the scar was evaluated at 14 DPI on midsagittal sections stained with anti-collagen IV (1/100) and anti-CSPG antibodies (1/100) (six mice per group). Immunohistochemical detection was achieved using the avidin-biotin complex immunoperoxidase technique and visualized with diaminobenzidine chromogen (Vector Laboratories). The collagen IV-and CSPG-positive areas were quantified using ImageJ software version 1.44e. Previous analysis of lesion volume confirmed that, in these animals, the lesion areas in mid-sagittal sections were correlated with the lesion volume.' Please note that following this new analysis we have revised our data and now report that the lesion volume at 42 DPI is not affected by the treatment. This finding likely highlights that the sub-acute period following SCI is critical for recovery.  Figure 4D now shows the appearance of BrdU/GFAP cells at high magnification. In our specific case, given that the treatment promotes the compaction of the lesion, we think that the stereological procedures for controlling the amount of tissue compaction would, on the contrary, introduce a bias. The counting of double-positive cells in the ring of reactive astrocytes bordering the lesion is in our opinion the most realistic procedure. This point is addressed in the extended version of the Discussion (Supporting Information): (Page 13, Line 14): 'The major defects observed after ganciclovir-targeted death revealed the contribution of the dividing reactive astrocyte pool to the process of walling off leukocytes after brain and spinal cord injuries (Bush et al, 1999;Faulkner et al, 2004;Myer et al, 2006;Voskuhl et al, 2009). The mitogenic effect of Ro3303544 observed in vivo in brain progenitors (Adachi et al, 2007) and in vitro in astrocytes led us to investigate whether proliferation was involved in its in vivo effect. Since BrdU incorporation in vivo was not significantly increased, the accelerated compaction of the lesion by Ro3303544 seems to result from the migration of reactive astrocytes, rather than from astrocyte proliferation. Such a discrepancy between the mitogenic effect observed in vitro and in the brain, as well as the absence of any effect in the spinal cord, has already been observed in the case of transforming growth factor alpha (TGFα) and astrocytes (Rabchevsky et al, 1998;Reynolds et al, 1992;Sharif et al, 2006;White et al, 2008). The responses of the brain and the spinal cord to injury are known to differ in intensity and nature: the acute inflammatory response is more important in the spinal cord than the brain (Schnell et al, 1999), while cells with β-catenin signaling increase in the cortex and the sub-cortical zone following brain injury but not following spinal cord injury (White et al, 2010). Consistent data showing that the phenotype of microglia critically affects their ability to support or impair cell renewal from adult stem cells (Butovsky et al, 2006;Ekdahl et al, 2003) may additionally take on added significance when considering reactive astrocytes as a variety of stem/progenitor cells (Buffo et al, 2008;Robel et al, 2011). 6. White matter sparing in figure 5 also suffers in accuracy of reporting because it is impossible to standardize and provide unbiased sampling of a complex 3 dimensional lesion using a single section. The area of positive staining is taken from a sample box of unknown size and location. The previous analysis was based on serial sagittal sections. However, we have replicated this analysis using cross-section analysis at various levels. Please refer to the response to Referee #1, point #7.

The description of methods used for unbiased counting BrdU+/GFAP+ cells in vivo is
7. The behavioural results reported on page 9 paragraph 3 are not consistent with the graph in figure 3C. Specifically, the text indicates a difference in motor scores beginning at 7 dpi, this is not indicated by * on the graph and appears to have not reached significance and should not be reported as "superior motor recovery". Results should specify repeated measures ANOVA main effects in addition to post-hoc corrected test results.
To address this comment, we have performed a new series of mice and evaluated their recovery using BMS scores. Please refer to the response to Referee #1, point #8. The corresponding results are as follows (page9, line12): 'The recovery of motor function was then monitored over 42 days using the Basso Mouse Scale open-field score (BMS) (Basso et al, 2006). The mice in the Ro3303544 group exhibited a tendency for greater motor function recovery compared to the control group as early as 7 DPI (1.08 ± 0.97 vs. 1.85 ± 1.12 in the control and Ro3303544 groups, respectively). Control mice, with a mean BMS score of 2.95 ± 1.21 at 42 DPI, could not support their weight on their hind limbs. By contrast, mice in the Ro3303544 group had a mean BMS score of 5.00 ± 2.05 at 42 DPI, and many mice in this group were able to walk with forelimb-hindlimb coordination. The BMS score of the Ro3303544 group was statistically better than that of the control group (2-way repeated measures ANOVA: Pvalue related to an effect of the treatment = 0.0151), and Bonferroni's multiple comparisons test at each time-point demonstrated statistical significance from 21 DPI until the end of the observation period, i.e., 42 DPI (Fig 6B). '

The discussion should include other plausible targets of systemic GSK-3 inhibition on the complex cellular environment at the lesion site, especially including effects on progenitor cells and the activation state and/or migration of inflammatory cells. Only a cell-specific blockade of GSK-3
in astrocytes would specifically address whether this cell type is primarily responsible for the reduced lesion size in this complex lesion or whether other effects leading to reduced secondary injury contribute to a smaller lesion. The argument that there is no difference in area occupied by CD11b+ staining at 7 dpi indicates that the effects are not on inflammatory cells is inadequate. The distribution of these cells at one time point does not provide sufficient information regarding their activation state or functional characteristics. We have accordingly addressed these issues in the Discussion. We invite the referee to read the extended version of the Discussion (Supporting Information): (Page 13, Line 3): 'After CNS injury, the mechanisms underlying the compaction of leukocytes involve both the migration of astrocytes (Okada et al, 2006) and the repulsion of inflammatory cells (Herrmann et al, 2008). Stimulation of astrocyte migration without efficient restriction of inflammatory cells is indeed ineffective for improving recovery after SCI (White et al, 2008). These mechanisms are still poorly understood. Notably, nucleotide release by astrocytes has recently been shown to induce the retraction of microglial processes (Orr et al, 2009). In the course of our investigation of the molecular mechanisms controlling astrocyte migration towards the lesion centre, we have confirmed that astrocyte chemotaxis towards inflammatory cells was unaffected by previous treatment with Ro3303544 (Renault-Mihara et al, in preparation), suggesting that the reduced lesion size does not result from a reduced secondary injury.' Minor concerns: a. The introduction should reference Herrmann et al., 2008 in the citations of a role of STAT3 activation in astrocyte protective function after SCI (paragraph 2). In Okada et al.,including progenitors and astrocytes. In Herrmann et al., were targeted for gancyclovirmediated death. Both studies report similar effects on astrocytic scar compaction and functional recovery. We do not understand the reference to ganciclovir-mediated death, which was not used in Herrmann's study and is probably a reference to the seminal studies of the same group (Bush et al, 1999;Faulkner et al, 2004; cited in our manuscript). We did not mean any disrespect by failing to cite this reference in the Introduction and apologize if it has been considered as such. We intended to introduce the notion that the beneficial compaction requires the migration of astrocytes ("Recently, using several conditional knock-out mice targeting STAT3 signaling in reactive astrocytes (Okada et al, 2006), we observed that the compaction and seclusion of infiltrating inflammatory cells in the lesion centre by migrating reactive astrocytes during the sub-acute phase of SCI is associated with improved locomotor recovery, suggesting that the migration of reactive astrocytes might constitute a new therapeutic target for the early phase of SCI (Renault-Mihara et al, 2008)."), which was not studied in Herrmann et al. However, this important contribution, which was already acknowledged in our previous version, is now cited in the Introduction, as requested by the referee, as well as in the Discussion. ) strongly suggest that some adult progenitors are also targeted in the spinal cords of 73.12 GFAP-cre mice. The comparison of the pattern of reporter expression in these two mouse strains suggests that the targeting in our nestin-cre mice is probably even more selective for reactive astrocytes than that of the 73.12 GFAP-cre line. In our nestin-cre mice, expression of GFP is indeed limited to the reactive astrocytes surrounding the lesion, whereas β-gal staining is observed throughout the spinal cords of uninjured GFAP-cre mice.
b. Assuming that there is no effect of the compound on cell proliferation and survival (see item 3 above), the graph in Figure 2D would be even more compelling if the cell area was expressed as area/cell, by dividing by the # of nuclei in the sample region. There is not sufficient description of the "semi-automatic quantitative analysis" in the methods or supplementary methods. The description of these effects should indicate that actin distribution appears unchanged. These experiments were performed using cell suspensions that were seeded onto a transwell, implying that the medium contained aphidicolin and that the survival of astrocytes in the Ro3303544 group was not decreased (higher migration capabilities). Accordingly, we did not observe any instance of mitosis. The procedure for quantitative analysis is now described in detail in the Supporting Information: (Page 5, Line 6): 'After 48-h treatment with 1 µM Ro3303544 or control, astrocytes were seeded at a low density onto glass coverslips coated with 10 µg/ml laminin in the medium used for transwell experiments, which includes aphidicolin, and, when indicated, 1 µM Ro3303544. Cells were fixed and stained for Factin and α-tubulin 15 h after seeding. Images were randomly captured with an epifluorescence microscope (Zeiss, Axioplan 2). Semi-automatic quantitative morphometric analysis was performed on α-tubulin-stained astrocytes using ImageJ software version 1.44e. The scheme of the analysis, performed with a macro, was as follows: after thresholding, the binary images were processed using the "close" and "fill holes" functions. If necessary, cells were then manually separated using the eraser tool. Cells with more than two neighbors were excluded from the data resulting from "particle analysis." The new illustration of our results in Fig 2C better reflects the actual analysis, and the y-axis legend has been corrected accordingly. Since the morphometric analysis has been performed on tubulin signal, we do not understand why the actin distribution should be unchanged. Obviously, we could accurately measure the outlines of each cell using tubulin staining.
c. Figure 2A illustrates the absence of an effect of 30 minutes of exposure of astrocytes to Ro3303544. None of the related experiments would suggest that this short exposure would have an effect. This figure adds nothing to the manuscript. The exposure time was actually 30 min + 15 h for the migration assay, during which time astrocytes were still exposed to Ro3303544. In our opinion, these negative data confirm that, in itself, the inhibition of GSK-3 is not anti-migratory and demonstrate that not only the duration of GSK-3 inhibition but also the assay used, and thereby probably the migration mode involved, are important for the effect of GSK-3 on migration. Nevertheless, we agree that these data are not essential to the main discussion, so we moved the former Fig 2A to Supporting Information Fig S2E. d. The images in Figure 3C and 3D are badly out of focus. In addition, high power images of the lesion border of the time point most different between the two treatments (14 dpi) would help illustrate effects on astrocyte compaction, process morphology, and interactions between macrophages and astrocytes at the border. Figure 4A (formerly 3C) in the present revised manuscript now displays both merged and separate layers at 14 DPI: CD11b staining and Hoechst layers consistently contribute to illustrate the compaction of the lesion by reactive astrocytes. Figure 4C shows that reactive astrocytes potently wall off the lesion at 14 DPI.
e. The authors are encouraged to read and consider use of the mouse locomotor scale for future studies (Basso et al., 2006). The BBB scale is designed to assess locomotor recovery following contusion injury in rats. Evaluation of hip and knee joint movement after spinal cord contusion in mice is considered to be unreliable; these differences affect the spread of scores at the lower end of the scale used for this study. Based on the scatterplot, it appears that a Chi squared analysis of the proportion of animals capable of weight supported stepping would be helpful as it might reveal an important functional milestone in the treated mice. As described above in response to point #7, we have performed a whole new series of mice and have evaluated their recovery using Basso Mouse Scale (Fig. 6B). Please find enclosed the final reports on your manuscript. We are pleased to inform you that your manuscript is accepted for publication and will be sent to our publisher to be included in the next available issue of EMBO Molecular Medicine if or once we have received your licenses (see below).
Please see below for additional IMPORTANT information and instructions regarding your article, its publication, and the production process.
Congratulations on your interesting work.