The ZEB1 pathway links glioblastoma initiation, invasion and chemoresistance

Transcription

1 The ZEB1 pathway links glioblastoma initiation, invasion and chemoresistance Florian A. Siebzehnrubl, Daniel J. Silver, Bugra Tugertimur, Loic P. Deleyrolle, Dorit Siebzehnrubl, Matthew R. Sarkisian, Kelly G. Devers, Antony T. Yachnis, Marius D. Kuepper, Daniel Neal, Nancy H. Nabilsi, Michael P. Kladde, Oleg Suslov, Simone Brabletz, Thomas Brabletz, Brent A. Reynolds and Dennis A. Steindler Corresponding author: Florian A. Siebzehnrubl, Department of Neurosurgery, University of Florida College of Medicine Review timeline: Submission date: 02 April 2013 Editorial Decision: 19 June 2012 Additional correspondence (author): 02 December 2012 Additional correspondence (editor) : 04 December 2012 Additional correspondence (author) : 11 December 2012 Additional correspondence (editor) : 13 December 2012 Submission to EMM: 19 December 2012 Editorial Decision: 22 January 2013 Resubmission: 02 April 2013 Editorial Decision: 02 May 2013 Revision: 02 May 2013 Accepted: 06 May 2013 Transaction Report: (Note: With the exception of the correction of typographical or spelling errors that could be a source of ambiguity, letters and reports are not edited. The original formatting of letters and referee reports may not be reflected in this compilation.) (Please note: This manuscript was originally submitted to the EMBO Journal. The correspondence and referee comments from the EMBO Journal are included in this Peer Review Process File.) Editors: Thomas Schwarz-Romond, Natascha Bushati, Céline Carret 1st Editorial Decision 19 June 2012 Thank you very much for transferring your paper on ZEB1's role in invasion and chemoresistance in glioblastomas for consideration to The EMBO Journal editorial office. Having received two rather consistent sets of comments, I am in a position to reach a final decision to not unnecessarily delay further proceedings of your study. As you will see from the enclosed comments, both scientists recognize potential clinical value of the paper, but take issues with the amount of actual novel and definitive molecular characterizations based on the scope of our relatively broad and very competitive title. Given the associated uncertainties, as explained in detail by their thorough and knowledgeable assessments, I hope you understand that there is not much that I can do than to return the paper to you at this point with the message that we are unable to offer further proceedings. EMBO 1

2 I still hope that relatively timely communication together with the explicit comments of our referees might facilitate necessary improvements for efficient submission to a more appropriate title. REFEREE COMMENTS Referee #1 The authors describe the role of ZEB1 and progression of Glioblastomas. They have identified ZEB1 expressing cells at the invasive front of xenografted glioblastomas. Separation and re-grafting of a LP and a HP tumour cell fraction gives rise to invasive and demarcated tumours, respectively. ZEB knockdown leads to reduced stemness of GBM CSC. There are several issues that need to be addressed A) ZEB1 expression: " ZEB1 expression is predictive of shorter survival": The fact that ZEB expression correlates with tumour grade could simply be a function of proliferation, and could be as meaningful as Ki67 labelling, and other cell cycle markers in general. In fact the influence of ZEB1 on stemness (negative regulation of) indicates exactly that. IDH1/2 status is an established prognostic marker. To validate the role of ZEB1 it would be important to correlate ZEB1 expression with IDH1 mutation status. How does ZEB1 expression relate to other, established and robust parameters, such as age of the patients: (which goes with IDH1 mutation status as well?). ZEB1 and MGMT: Does the expression of these two markers colocalise in a tissue section? The histology sections show increased nuclear expression with increasing grade. On how may samples was this tested? Does it correlate with Ki67 Labelling index? A co-localisation study of MYB and MGMT would be useful to determine if the expression is mutually exclusive, as suggested in the molecular data. On the other hand, the authors state "These findings suggest ZEB1 may mediate glioma chemoresistance through its inhibitory actions on mir-200c and c-myb"; followed by " c-myb knockdown reduced expression of MGMT in hgbm cells," and finally "and found no significant differences in methylation between shgfp and shzeb1 knockdown cells". How can this apparent contradiction be clarified? ZEB1, LP and HP cell separation: Cells were separated based on the hypothesis that slow and fast proliferating cells are differentially invasive. The authors re-implanted populations sorted for slow and fast proliferating fractions into host mice. The authors have not tested alternative hypotheses, for example that a slow proliferating population could be a cancer stem cell (CSC) population and the fast proliferating cells represent transient amplifying cells (TAC). This separation could equally explain the difference in behaviours after grafting. The authors need to test (and falsify) alternative hypotheses, what slow and fast dividing population may represent. It is possible that the differences in migration and invasive behaviour can be entirely explained by presence of CSC and TAC. Characterising both fractions further is essential to make further assumptions what these two populations may represent. In fact the authors have previously reported an enrichment of stemness markers in LP fractions. A useful control experiment would be the fractionation of a stemness positive population and compare to the LP enriched population in a xenograft. IN fact, the possibility that LP cells are indeed CSC, is underlined by their relative resistance to TMZ treatment, a feature that has been reported before and is also cited by the authors. Glioblastoma samples and patients: EMBO 2

3 a. Table S1 shows incomplete data for a large set of patients. It does not become clear from the table legend, what the information in the IDH1 column means ("un" vs. "-". Does it mean untested or unmutated or unknown? To make valid assumptions about the role of ZEB1, it is paramount that complete data (sequencing for IDH1 and IDH2) are obtained. It is conceivable that that the better outcome is related to IDH1 and not a function of ZEB1 expression. b. A large number of patients are of relatively young age. In particular those patients from whom the cell lines are derived, were exceptionally young. Do these tumours have any particular features, did they show an IDH1 mutation, which glioma subtype do they belong to? c. The cell lines used in this study are not well characterised. It would be essential to know the MGMT methylation status of all samples, IDH1 mutation status, and it would be highly desirable to know more about other molecular parameters such as 10qLOH, EGFR amplification, NF1 LOH, and other established markers. Also 1p19q LOH would be useful to be tested to exclude GBM with oligodendroglial differentiation/ anaplastic oligoastrocytoma. Other points Referencing the Methods: Vimentin is not in the list antibodies. This is important because the authors claim (and the data suggest) that it is human specific.?., Overall the Immunofluorescence images are acceptable but not excellent. Have they been taken on a confocal microscope? It seems they lack definition, and it would be helpful to have insets or a separate row of high magnification areas, for example of figure 1C. Referencing of methods is poor: The histological methods are not described properly and the references given are not describing the pre-treatment and other methods of ZEB1 on formalin fixed tissues. In fact the references are only related to frozen sectioning. The staining of the tumours is not DAB, but nickel enhanced DAB which is not described in the methods In conclusion, at present, there are many missing data that may in fact explain some of the effects. The roles of ZEB1 may not be attributes to the pathways described here and in fact may rather represent phenomena related to stemness and proliferation. Some of the referencing is superficial. Referee #2 In this manuscript the authors argue for a role for Zeb1 in the development of invasive and chemoresistant glioblastome multiforme (GBM) tumor cells. They find that Zeb1 expression is associated with a low proliferation subgroup of tumor cells and that RNAi-mediated removal of Zeb1 in this population reverts them to chemosensitive, less invasive tumors in xenografts. They then go on to propose a model and show experimental support for pathways involving Zeb1/mir200/Myb/MGMT in the regulation of chemosensitivity and Zeb1/miR200/Robo1 in the regulation of adhesion. The human association studies presented are provocative. This is an interesting story. Chemo-resistance or sensitivity could also, in part, be a reflection of their prior work showing Zeb1 represses stemness-inhibiting micrornas. Prior work has shown an association between EMT, cancer "stemness", and chemoresistance. The new information here is the specific role of Myb and MGMT in this process. The experimental support for this pathway is reasonably strong but unlikely to be the whole story. Some of the data in Figure 3D, in particular, are not that compelling. For example, I don't appreciate the change in MGMT level in the presence of anti-mir200c. With respect to the Robo-1 mediated regulation of adhesion. The data supporting this contention is not as strong as for Myb/MGMT-mediated chemosensitivity. More experimental work is required to support this claim. Also I should note that the experimental support for altered adhesion is Figure 1B - an IHC for adhesion molecules. This hardly constitutes a functional assay for adhesion. In general the authors often over-interpret results to emphasize differences. For example, p6 they state: "the majority of the Zeb1 negative tumors were characterized by NF1 loss". The actual number was 52% - while factually correct I would not make much out of this. Many Figure legends EMBO 3

4 lack sufficient detail to fully appreciate what manipulations are done in each figure panel. Despite the Zeb1 and mir200s connection they should also look for the presence of other EMT inducers and whether their level changes with various experimental manipulations. Many tumors express multiple EMT inducers and it is possible that they serve specific functions during EMT. The present work argues that Zeb1 is important for GBM invasion and chemosensitivity, but others may also contribute. Minor points 1. Supplemental Figure 2A - need corroborating Western blot for Zeb1 KD. 2. Supplemental Figure 2 - the image shown doesn't seem to jive with the graph data. Maybe you could try another type of migration assay and see if similar. Additional correspondence (author) 02 December 2012 I hope this finds you well. With this I wish to ask whether you would consider a new submission entitled "The ZEB1 pathway regulates tumor formation, invasion and chemoresistance in glioblastoma". This manuscript is an extensive revision of a paper that has been transferred to your journal previously (EMBOJ T), and includes novel data on tumor initiation, as well as deeper mechanistic insights into the respective pathways responsible for invasion and chemoresistance. We are certain that our revisions address all of the previous reviewers' criticisms from the EMBO Journal to the fullest. Our study now provides a clear dissection of the molecular functions of ZEB1 in glioblastoma, and the remarkable differences between cancers of the brain and other solid tissues. We find that ZEB1 induces tumor invasion in a fashion that is notably distinct from its known function in other solid tissue cancers. We have further validated and extended our data on the mechanism for induction of chemoresistance, and this mechanism is described here for the first time. Finally, we have dramatically improved our analyses of patient samples, and further strengthened the predictive value of ZEB1 in glioblastoma. We therefore feel that our manuscript is considerably stronger and includes further novel findings, and hope that you are willing to consider it. I understand the difficulties of appraising a work without actually seeing it, but the fact that previous reviewers attested to the relevance of our work encourages me to seek your advice on whether to submit this new manuscript to your excellent journal. Additional correspondence (editor) 04 December 2012 Thank you very much for kind enquiry about possible resubmission of a revised paper to The EMBO Journal editorial office. Better diagnostics, respective specific therapeutic intervention(s) for glioblastoma are certainly a timely and relevant issue. Thus, establishing novel molecular mechanisms with regard to ZEB1- function might be in principle of interest to The EMBO Journal. However, all original referees ranked the overall novelty and general interest of your previous submission as medium, and thus the paper is not of sufficient priority for further consideration. (I attach the comments of a third referee that responded after I sent the decision simply for your information). In light of this, it would need significant further reaching results and amendments to change such a very consistent assessment on your study. With this in mind, I would not be opposed to assess a new version of the study at the sole editorial level. Pending on the outcome of this, and given the rather evident translational implications of your work, we could at the same time inquire with our sister journal EMBO Molecular Medicine about potential interest to pursue your study. EMBO 4

5 I hope that this letter has clarified our relatively strong and rather more general demands. I still hope that this response might facilitate a rapid decision on how to proceed with the study. In case you were to pursue The EMBO Journal, respective EMBO Molecular Medicine for potential presentation of your results, please simply attach a finalized PDF to a return message. (Please notice that I will be absent from the office until Monday next week, 10th of December, but promise to get back to you as soon as I can). Referee #3 remarks: In this manuscript the authors argue for a role for Zeb1 in the development of invasive and chemoresistant glioblastome multiforme (GBM) tumor cells. They find that Zeb1 expression is associated with a low proliferation subgroup of tumor cells and that RNAi-mediated removal of Zeb1 in this population reverts them to chemosensitive, less invasive tumors in xenografts. They then go on to propose a model and show experimental support for pathways involving Zeb1/mir200/Myb/MGMT in the regulation of chemosensitivity and Zeb1/miR200/Robo1 in the regulation of adhesion. The human association studies presented are provocative. This is an interesting story. Chemo-resistance or sensitivity could also, in part, be a reflection of their prior work showing Zeb1 represses stemness-inhibiting micrornas. Prior work has shown an association between EMT, cancer "stemness", and chemoresistance. The new information here is the specific role of Myb and MGMT in this process. The experimental support for this pathway is reasonably strong but unlikely to be the whole story. Some of the data in Figure 3D, in particular, are not that compelling. For example, I don't appreciate the change in MGMT level in the presence of anti-mir200c. With respect to the Robo-1 mediated regulation of adhesion. The data supporting this contention is not as strong as for Myb/MGMT-mediated chemosensitivity. More experimental work is required to support this claim. Also I should note that the experimental support for altered adhesion is Figure 1B - an IHC for adhesion molecules. This hardly constitutes a functional assay for adhesion. In general the authors often over-interpret results to emphasize differences. For example, p6 they state: "the majority of the Zeb1 negative tumors were characterized by NF1 loss". The actual number was 52% - while factually correct I would not make much out of this. Many Figure legends lack sufficient detail to fully appreciate what manipulations are done in each figure panel. Despite the Zeb1 and mir200s connection they should also look for the presence of other EMT inducers and whether their level changes with various experimental manipulations. Many tumors express multiple EMT inducers and it is possible that they serve specific functions during EMT. The present work argues that Zeb1 is important for GBM invasion and chemosensitivity, but others may also contribute. Minor points 1. Supplemental Figure 2A - need corroborating Western blot for Zeb1 KD. 2. Supplemental Figure 2 - the image shown doesn't seem to jive with the graph data. Maybe you could try another type of migration assay and see if similar. Additional correspondence (author) 11 December 2012 Thank you very much for your kind offer. I would like to ask you for your editorial opinion on the attached new manuscript. As you will see, we have chosen to remove our previous findings on low and high proliferating cells, as we felt these would make our increasingly complex story even more complicated. In brief, this new version contains now data on overexpression of ZEB1 (mainly Figs. 3-5), mechanistic data on invasion and ROBO (Fig. 3), more data on chemoresistance in vitro and in vivo (Fig. 4), experiments on stemness and tumor formation (Fig. 5), better validation of our findings in patient specimens and a multivariate analysis on the effects of ZEB1 on patient outcome. EMBO 5

6 All of these findings have not been reported previously. I hope these new data will find your approval, but if you do not deem this manuscript sufficiently advanced over our previously submitted paper, please feel free to share it with your colleagues at EMBO Molecular Medicine. I would like to thank you again very much for your time and consideration and look forward to hearing from you. Additional correspondence (editor) 13 December 2012 I now found the time to check your modified version and appreciate the amendments and consolidation with respect to the previous version that was reviewed at The EMBO Journal. Despite these, and having discussed the case also with my editorial colleagues, we are not convinced that the highly relevant translational outline of ZEB-function in glioblastoma would go conceptually much beyond reported notions in the current literature. Thus, I am afraid we had to conclude that we are not able to offer further consideration. As promised however, I also discussed the paper and the existing communication with my colleague from our sister journal EMBO Molecular Medicine (EMM) ( I am thus happy to relay that their scientific editor Natascha Bushati would be prepared to solicit the necessary peer-review input to facilitate hopefully timely presentation of your results in EMBO Molecular Medicine. I thus kindly ask you to submit/contact Natasha directly (Natascha.Bushati@embo.org) to enable efficient proceedings. Once more, I am sorry that I am unable to reach a more positive conclusion from the perspective of The EMBO Journal, but I hope that EMM might turn out to be a suitable alternative for presentation of your results. Submission to EMM 192 December 2012 (Point by point response to the Referee comments at EMBO Journal): We would like to thank the reviewers for their helpful comments. When addressing the criticisms on this manuscript, we were facing a study that was fairly intricate at the start evolving into an evermore complex story. Therefore, we chose to omit our previous data on low and high proliferating cells from this revised manuscript, in order to develop this particular angle into a study that can stand on its own, with the necessary additional data. The present revision focuses on the functions of ZEB1 in glioblastoma, specifically dissecting its roles in invasion, chemoresistance and tumorigenesis. We strove to improve our story by adding immunoblot data on a second cell line, more data on the mechanism of invasion, as well as experiments on tumor formation capacity of ZEB1 knockdown and control cells. We also succeeded in overexpressing ZEB1 in our glioblastoma cell lines, and show increased invasion and tumorigenicity after targeted expression of ZEB1. Referee #1 The authors describe the role of ZEB1 and progression of Glioblastomas. They have identified ZEB1 expressing cells at the invasive front of xenografted glioblastomas. Separation and re-grafting of a LP and a HP tumour cell fraction gives rise to invasive and demarcated tumours, respectively. ZEB knockdown leads to reduced stemness of GBM CSC. There are several issues that need to be addressed A) ZEB1 expression: " ZEB1 expression is predictive of shorter survival": The fact that ZEB expression correlates with EMBO 6

7 tumour grade could simply be a function of proliferation, and could be as meaningful as Ki67 labelling, and other cell cycle markers in general. In fact the influence of ZEB1 on stemness (negative regulation of) indicates exactly that. IDH1/2 status is an established prognostic marker. To validate the role of ZEB1 it would be important to correlate ZEB1 expression with IDH1 mutation status. How does ZEB1 expression relate to other, established and robust parameters, such as age of the patients: (which goes with IDH1 mutation status as well?). We thank this reviewer for their comments, and we have clarified our manuscript to better reflect the interpretations of our findings. In particular, we find a negative association of ZEB1 with proliferation (presented in Figure S2), rendering association of our survival data with proliferation index highly unlikely. This is supported by the fact that ZEB1 is expressed mainly in low proliferating cells, but absent in the high proliferating fraction. Immunofluorescence studies in patient specimens have confirmed that Ki67 and ZEB1 are mutually exclusive. Since the new manuscript is focusing on other functions of ZEB1, we have chosen not to present this data, but to incorporate it into a study that will exclusively deal with ZEB1 s functions on proliferation and on low and high proliferating cells. However, if the reviewers felt that this data is crucial to the current study, we can include these studies. Regarding the IDH1/2 mutation status, we agree with the reviewer that this is an established prognostic marker. However, these mutations are present only in about 5% of primary glioblastoma, while our analysis indicates presence of ZEB1 in 40-45% of specimens (n.b. that all our specimens are from primary glioblastoma). IDH1/2 mutations are much more prevalent (and important) in secondary glioblastoma (60-90%), and we are currently undertaking a study on ZEB1 and IDH1 in secondary tumors. Nevertheless, we have analyzed IDH1 mutation status in those samples that were available for histological assessment (a total of 34, findings are presented in Supplementary Table 1; see also comment on IDH1 below), and found only two samples bearing the IDH1 mutation. Since the prevalence of this mutation is very low compared to ZEB1 expression status, we concluded that IDH1 mutation has no relevance to our current study. We have performed a multivariate analysis of ZEB1 and other parameters, including age, survival, KPS scores, MGMT expression, and different subclass markers. We found significant correlations of ZEB1 expression with survival, treatment response and MGMT expression, but not with age. All these data are presented in Figs. 6 and S5. ZEB1 and MGMT: Does the expression of these two markers colocalise in a tissue section? The histology sections show increased nuclear expression with increasing grade. On how may samples was this tested? Does it correlate with Ki67 Labelling index? A co-localisation study of MYB and MGMT would be useful to determine if the expression is mutually exclusive, as suggested in the molecular data. On the other hand, the authors state "These findings suggest ZEB1 may mediate glioma chemoresistance through its inhibitory actions on mir-200c and c-myb"; followed by " c-myb knockdown reduced expression of MGMT in hgbm cells," and finally "and found no significant differences in methylation between shgfp and shzeb1 knockdown cells". How can this apparent contradiction be clarified? We thank this reviewer again for the helpful comments, and we have added additional data regarding ZEB1 and MGMT expression in protein and tissue samples (Fig. 6), as well as in our xenograft animals (Figs. 2 and 4). We have further clarified the manuscript to better reflect the role of c-myb in the induction of MGMT (c-myb directly induces MGMT). We have also added promoter methylation analysis for MGMT in all our cell lines, showing that the regulation of MGMT expression is indeed on the transcriptional level, and unrelated to epigenetic methylation of the MGMT. ZEB1, LP and HP cell separation: Cells were separated based on the hypothesis that slow and fast proliferating cells are differentially invasive. The authors re-implanted populations sorted for slow and fast proliferating fractions into host mice. The authors have not tested alternative hypotheses, for example that a slow proliferating population EMBO 7

8 could be a cancer stem cell (CSC) population and the fast proliferating cells represent transient amplifying cells (TAC). This separation could equally explain the difference in behaviours after grafting. The authors need to test (and falsify) alternative hypotheses, what slow and fast dividing population may represent. It is possible that the differences in migration and invasive behaviour can be entirely explained by presence of CSC and TAC. Characterising both fractions further is essential to make further assumptions what these two populations may represent. In fact the authors have previously reported an enrichment of stemness markers in LP fractions. A useful control experiment would be the fractionation of a stemness positive population and compare to the LP enriched population in a xenograft. IN fact, the possibility that LP cells are indeed CSC, is underlined by their relative resistance to TMZ treatment, a feature that has been reported before and is also cited by the authors. The new version of our manuscript includes a better analysis of stemness with regards to ZEB1 expression. In particular, ZEB1 expressing cells are more tumorigenic, and express higher levels of typical stemness markers SOX2, OLIG2 and CD133, than ZEB1 knockdown cells. We agree with the caveats this reviewer is raising, but given the lack of definitive markers for transit-amplifying progenitors in cancer, an unambiguous characterization of HP cells would be exceedingly difficult. In order to not make this story more complex, we have chosen to focus more on the specific functions of ZEB1 in invasion, chemoresistance and stemness, and will develop our findings in LP and HP cells into a story of their own. This will afford more of a focus on characterizing these fractions better without detracting too much from our goal on the functions of ZEB1. In the current manuscript, we present clearly that ZEB1 can induce typical stemness-associated genes in glioblastoma, and that ZEB1 expression is tied to sphere and tumor formation. Therefore, we conclude that ZEB1-positive cells are a cancer stem cell population (which fits with our previous findings in LP cells). Glioblastoma samples and patients: a. Table S1 shows incomplete data for a large set of patients. It does not become clear from the table legend, what the information in the IDH1 column means ("un" vs. "-". Does it mean untested or unmutated or unknown? To make valid assumptions about the role of ZEB1, it is paramount that complete data (sequencing for IDH1 and IDH2) are obtained. It is conceivable that that the better outcome is related to IDH1 and not a function of ZEB1 expression. b. A large number of patients are of relatively young age. In particular those patients from whom the cell lines are derived, were exceptionally young. Do these tumours have any particular features, did they show an IDH1 mutation, which glioma subtype do they belong to? c. The cell lines used in this study are not well characterised. It would be essential to know the MGMT methylation status of all samples, IDH1 mutation status, and it would be highly desirable to know more about other molecular parameters such as 10qLOH, EGFR amplification, NF1 LOH, and other established markers. Also 1p19q LOH would be useful to be tested to exclude GBM with oligodendroglial differentiation/ anaplastic oligoastrocytoma. Regarding again these helpful comments: (a) we have improved the presentation of patient data in Table S1 ( un in the IDH1 column refers to an unknown status, while - means unmutated). As in our response to the above comment, we would like to point out that the specimens in our study are derived from primary glioblastoma, an entity where IDH mutations are found in about 5-10% of cases, while we find ZEB1 expression in 40-45% of cases. We would further like to point out that we have studied the IDH1 mutation status in 37/50 of our samples (74%) and found no significant relationship between IDH1 mutation and ZEB1 expression. Out of the 37 samples tested, only 2 (5.4%) were positive for the IDH1 mutation. One of these had a survival of 48 months, the other of 10 months. It is difficult to conceive how these two samples could force the rather strong correlation between ZEB1 expression and survival and therapy response. (b) The median age of our sample cohort is 64 years, which fits perfectly with typically reported age at diagnosis for glioblastoma (e.g. see Schwartzbaum et al. 2006, and the CBTRUS statistical report, While we agree that the cell lines used are from younger patients, two out of three lines are derived from patients in their fourth decade, which is still within the typical range of glioblastoma. We have included a more thorough characterization of their specific subclass and find that these lines fit into the proliferative type (Table S1), as do most of the patient specimens analyzed. (c) We have included an MGMT promoter methylation analysis for all three cell lines in the new manuscript (Figure S4), and we find that ZEB1 expression has no relation to MGMT promoter EMBO 8

9 methylation. We have further included analysis of EGFR expression, NF1 LOH and other subclass markers as reported by Phillips et al and Brennan et al These data are presented in Table S1. Our findings are validated in a large cohort of primary glioblastoma, where we find that EGFR amplification is the only one of these factors associated with the ZEB1-related pathology. Notably, PTEN is located on chromosome 10q, and we found no significant differences in PTEN LOH between our respective groups. Di Stefano et al. (2012) recently reported that EGFR amplification is largely independent of the genetic profile (10q and/or IDH mutation) in a dataset of >1300 glioma samples. We would further like to point out that the histopathology of the cell lines used for this study was reported in Deleyrolle et al. 2011, where we found no oligodendroglial component in these tumors. Other points Referencing the Methods: Vimentin is not in the list antibodies. This is important because the authors claim (and the data suggest) that it is human specific.?., We apologize for the confusion, but Vimentin was never part of our study. We used an antibody against the human Nestin protein, which is listed in the table of antibodies. Overall the Immunofluorescence images are acceptable but not excellent. Have they been taken on a confocal microscope? It seems they lack definition, and it would be helpful to have insets or a separate row of high magnification areas, for example of figure 1C. We have improved all images presented in the manuscript. All of the in vivo images are single-plane projections of confocal z-stacks, as is specified in the Methods section. We have also included higher magnification images for all data that was presented in the previous Figure 1C. Referencing of methods is poor: The histological methods are not described properly and the references given are not describing the pre-treatment and other methods of ZEB1 on formalin fixed tissues. In fact the references are only related to frozen sectioning. The staining of the tumours is not DAB, but nickel enhanced DAB which is not described in the methods. We thank the reviewer for bringing this to our attention, and we have improved the histology methods section to include the paraffin-embedded tissue. The DAB-staining is fairly strong and may therefore be mistaken for Nickel-enhanced DAB, but in fact is from a regular DAB kit (Vector Elite). We have included this information in the methods section. Referee #2 In this manuscript the authors argue for a role for Zeb1 in the development of invasive and chemoresistant glioblastome multiforme (GBM) tumor cells. They find that Zeb1 expression is associated with a low proliferation subgroup of tumor cells and that RNAi-mediated removal of Zeb1 in this population reverts them to chemosensitive, less invasive tumors in xenografts. They then go on to propose a model and show experimental support for pathways involving Zeb1/mir200/Myb/MGMT in the regulation of chemosensitivity and Zeb1/miR200/Robo1 in the regulation of adhesion. The human association studies presented are provocative. This is an interesting story. Chemo-resistance or sensitivity could also, in part, be a reflection of their prior work showing Zeb1 represses stemness-inhibiting micrornas. Prior work has shown an association between EMT, cancer "stemness", and chemoresistance. The new information here is the specific role of Myb and MGMT in this process. The experimental support for this pathway is reasonably strong but unlikely to be the whole story. Some of the data in Figure 3D, in particular, are not that compelling. For example, I don't appreciate the change in MGMT level in the presence of anti-mir200c. The new version of this manuscript contains improved western data for anti-mir-200c. In addition, we have performed all immunoblots on lysates from an additional cell line in order to validate our findings. Additionally, we have added knockdown and rescue experiments for MGMT in Figure 4, EMBO 9

10 as well as a more thorough validation in patient specimens in Figure 6. With respect to the Robo-1 mediated regulation of adhesion. The data supporting this contention is not as strong as for Myb/MGMT-mediated chemosensitivity. More experimental work is required to support this claim. Also I should note that the experimental support for altered adhesion is Figure 1B - an IHC for adhesion molecules. This hardly constitutes a functional assay for adhesion. We agree with the reviewer that the proposed mechanism for invasion was a particular weakness of the previous version, and have now included much more compelling data in Figure 3. These new data include increased invasion after overexpression of ZEB1, association of ROBO expression with the invasion front in vivo, and reduced invasion after knockdown of ROBO. In addition, we have included immunofluorescence images for ROBO in invasive cells of patient tumor specimens (Figure 6C). Our method of quantifying invasion is an unbiased assessment based on stereological principles, and thus allows obtaining quantitative information of in vivo tumor invasion. In general the authors often over-interpret results to emphasize differences. For example, p6 they state: "the majority of the Zeb1 negative tumors were characterized by NF1 loss". The actual number was 52% - while factually correct I would not make much out of this. Many Figure legends lack sufficient detail to fully appreciate what manipulations are done in each figure panel. Despite the Zeb1 and mir200s connection they should also look for the presence of other EMT inducers and whether their level changes with various experimental manipulations. Many tumors express multiple EMT inducers and it is possible that they serve specific functions during EMT. The present work argues that Zeb1 is important for GBM invasion and chemosensitivity, but others may also contribute. We thank the reviewer for these helpful comments, and we have revised our manuscript for more clarity. All figure legends and figures have been re-worked, and we hope it is now easier to follow what has been done in each individual panel. We agree that ZEB1 is unlikely to be the only EMT-associated factor in glioma, and our additional data in Figures 1, 6, S1 and S5 support this notion. We chose to focus on ZEB1 since we found a correlation between ZEB1 and survival in the TCGA dataset, but observed no correlation with other EMT factors. Since EMT is a highly regionalized process it is possible that the TCGA dataset underreports the presence of these factors in glioma, but since our intention was to study the functions of ZEB1, and not necessarily EMT in general, we made this protein the core of our article. We think that the new manuscript does justice to the potential roles of other EMT proteins and hope this reviewer will agree. Minor points 1. Supplemental Figure 2A - need corroborating Western blot for Zeb1 KD. We have included a western blot for ZEB1 knockdown in Figure Supplemental Figure 2 - the image shown doesn't seem to jive with the graph data. Maybe you could try another type of migration assay and see if similar. The new manuscript much more emphasizes invasion experiments carried out in vivo. We are presenting the data on invasion in Figure 3 and Figure S3. 2nd Editorial Decision 22 January 2013 Thank you for the submission of your manuscript "The ZEB1 pathway links glioblastoma initiation, invasion and chemoresistance". I approached the previous reviewers from The EMBO Journal, but unfortunately neither of them was available to comment on your revised manuscript. I therefore asked an alternative expert to assess the suitability of the study for EMBO Molecular Medicine, based on your point-by-point response and the new version of the manuscript. EMBO 10

11 As you will see, the reviewer acknowledges the potential interest of the study, but overall does not support publication of the manuscript. The reviewer highlights that the mechanistic insight provided remains too limited to support the major claims. I am afraid that we are therefore unable to offer publication of the current manuscript. You will recognize that the reviewer provides a rather detailed report and has a number of suggestions on how to improve the manuscript. Given the continuous interest in your findings, we would thus have no objection to consider a new manuscript at some time in the future, conditioned you would extend the study in a way that would convince the reviewer, particularly her/his explicit concerns about necessary mechanistic advance. However, we realize that addressing these criticisms might require a lot of additional time and effort and be technically challenging. I would thus understand if you chose to rather seek rapid publication elsewhere. I apologize for the slight delay reaching such a decision, presumably also caused be the recent festive season. I am sorry to have to disappoint you on this occasion. ***** Reviewer's comments ***** Referee #1 (Comments on Novelty/Model System): Novelty might be increased after elucidation of the mechanisms mediating ZEB1 actions. Referee #1 (Remarks): Response to the Editor's request for arbitration. General comments on the new version. I agree with Referees' opinion that the manuscript describes an interesting story. The authors moved from their previous observation on a ZEB1/miRNA axis regulating stemness and EMT in pancreatic and colon cancer and further go on to propose two main aspects: i) a similar role for ZEB1 in glioblastoma and ii) the mechanistic involvement of mir-200/c-myb/mgmt and of mir- 200/Robo1/N-cadherin in regulating chemoresistance and adhesion, respectively. The study provides original and convincing data on the ability of ZEB1 to promote initiation, invasion and chemoresistance of GBM: this is of particular importance since GBM is a very aggressive tumor still requiring the identification of novel therapeutic targets. However, the mechanistic role of the mir- 200/c-MYB/MGMT and of mir-200/robo1/n-cadherin are not strongly supported by the experimental findings (i.e. the model shown in Fig. 5E). The major points that still need to be addressed, some of which have been raised by Referees (see below), are: - The association between Zeb1 and cell proliferation in xenografts vs primary tumors. - The role of ROBO1 in mediating the effect of ZEB1 upon invasion. - The role of mir-200 and c-myb in mediating the effect of ZEB1. Overall, the manuscript would be suitable for EMBO Mol. Med., because of its topic and since some of the Referees' concerns on the above issues have been addressed. However, I believe that, although the manuscript has been improved, all of the authors' main claims are not fully supported in order to justify publication of the manuscript as is. Point-by-point response to previous review. Referee #1 The referee raised the question of whether ZEB1 is predictive of poor prognosis as a consequence of its correlation with other prognostic markers (e.g. proliferation index, IDH1/2 mutations, age). The authors claim a negative association of ZEB1 with proliferation (Figure S2) but also show a positive correlation between proliferation markers and ZEB1 expression in human primary tumors (Fig. 6G and Fig. S5). Therefore, it appears that these findings are conflictual: is ZEB1 associated EMBO 11

12 with high-proliferating or low-proliferating cells? This appears to be critical given the claimed role of ZEB1 in prognosis and stemness. The responses to the other questions of the Referee are satisfying. The number of samples tested has been added, as requested. The issue of co-staining of ZEB1 and MGMT has beed addressed by saying "MGMT and ZEB1 were frequently co-localized in ZEB1+ tumor specimens (Fig. 6E, arrowheads)", but these findings should be quantified. No co-staining of ZEB1 and c-myb has been provided. As for the clarification of the apparent contradiction about MGMT regulation by ZEB1/c-MYB and methylation, the authors provide experimental data suggesting that ZEB1 upregulates c-myb (western blot in Fig. 4C) that in turn transcriptionally activates MGMT promoter (ChIP fig. 4D) without affecting its methylation (Fig. S4). The Referee raised the hypothesis that low-proliferating cells (ZEB1-positive) are tumorigenic because they are cancer stem cells (and not simply by expression of ZEB1) while high-proliferating cells (ZEB1-negative) are non tumorigenic transient amplifying cells. In the revised manuscript, the authors have removed the low- and high-proliferating cell story and focused on the demonstration that ZEB1 can induce stemness-associated genes in glioblastoma, promote sphere formation in vitro and tumor formation in vivo, thus convincigly concluding that ZEB1-positive cells are a cancer stem cell-like population. The authors replied convincingly to the last major points raised by the Referee and also to the other clarifications requested. Referee #2. The Referee believes that the specific role of Myb and MGMT in mediating ZEB1-induced chemoresistance is reasonably supported by experimental evidence but unlikely to be the whole story and would require further work. The revised manuscript includes new data on western blots after anti-mir-200, additional cell lines as well as knockdown and rescue experiments for MGMT in Figure 4, and a more thorough validation in patient specimens in Figure 6. However, in agreement with Referee's opinion the authors' claim is still not strongly supported. For instance, why, in contrast with c-myb levels, MGMT levels of anta.control-treated cells are similar to those of anta-mir-200? Similar observations for TMZ-sensitivity (Fig. 4C). Does depletion of mir-200 or c-myb influence the ZEB1-modulated (by knockdown or overexpression) TMZ sensitivity, reflecting the levels of the ZEB1/miR/c-myb/MGMT axis members? The sentence "we found that ZEB1 reduced expression levels of ROBO1 (Fig. 3A) while overexpression of ZEB1 had the opposite effect (Fig. 3B)" on page 8 is confusing and I believe incorrect. The Referee required more experimental work to support the claimed modulation of invasion by ROBO1. The new data including increased invasion after overexpression of ZEB1, association of ROBO expression with the invasion front in vivo, and reduced invasion after knockdown of ROBO, all address this point. However, a few points remain to be explained. i) The lack of overall reduced N-cadherin protein levels after ZEB1 knockdown (Fig. 3A), is contrasting the reduced expression in ZEB1-positive cells in tumor xenografts (Fig. 2C). The supposed regulation of N-cadherin localization, rather than expression, is not proven. ii) The role of mir-200 in mediating the effect of ZEB1 upon ROBO. iii) The role of ROBO in mediating the effect of ZEB1 upon invasion. The Referee says that the work argues that Zeb1 is important for GBM invasion and chemosensitivity, but others may also contribute. The authors addressed this point in the revised manuscript, accordingly. EMBO 12

13 Resubmission 02 April 2013 We would like to thank the reviewer for their very helpful comments. The revised manuscript takes all of these concerns into account, and we are specifying below how each of the criticisms has been addressed. Referee #1 The referee raised the question of whether ZEB1 is predictive of poor prognosis as a consequence of its correlation with other prognostic markers (e.g. proliferation index, IDH1/2 mutations, age). The authors claim a negative association of ZEB1 with proliferation (Figure S2) but also show a positive correlation between proliferation markers and ZEB1 expression in human primary tumors (Fig. 6G and Fig. S5). Therefore, it appears that these findings are conflictual: is ZEB1 associated with high-proliferating or low-proliferating cells? This appears to be critical given the claimed role of ZEB1 in prognosis and stemness. To address the apparent disparity between ZEB1-positive tumors expressing proliferative markers and ZEB1 being associated with lower proliferation rates in vitro, we have included immunofluorescence staining for ZEB1 and Ki-67 of patient specimens. This demonstrates that ZEB1 and Ki-67 are almost mutually exclusive in patient tumors. As shown in Figure S6C, we find a 3% overlap between ZEB1 and Ki-67, supporting the lower proliferation rates of ZEB1+ cells reported in Figure S2. As the ZEB1+ population is a stem cell population, it is conceivable that these cells are more quiescent (as we have demonstrated in Deleyrolle et al. Brain 2011) and give rise to rapidly proliferating cells, which in turn influence the subclass analysis. Indeed, the Phillips lab has found such a hierarchy recently (Chen et al. Cancer Cell 2010). We have added this to the discussion on page 18. The responses to the other questions of the Referee are satisfying. The number of samples tested has been added, as requested. The issue of co-staining of ZEB1 and MGMT has beed addressed by saying "MGMT and ZEB1 were frequently co-localized in ZEB1+ tumor specimens (Fig. 6E, arrowheads)", but these findings should be quantified. No co-staining of ZEB1 and c-myb has been provided. We have quantified the ZEB1/MGMT coexpressing population, and find an approximately 50% overlap between the two markers. This data has been added to the text on page 14, and a graphic representation can be found in Figure S6C. Regarding the co-localization of ZEB1 and c-myb in human patient specimens, we attempted immunostaining with three different commercially available c-myb antibodies, but none produced a strong enough signal for a meaningful analysis. Based on our immunoblot experience, c-myb is expressed at low levels in the tested cell lines and patient specimens, and therefore it may be unfeasible to obtain co-localization data. Even without c-myb coexpression data from patient specimens, we still present a significant overlap between ZEB1 and MGMT in these specimens, as well as strong data for this pathway in cells derived from primary tumors. As for the clarification of the apparent contradiction about MGMT regulation by ZEB1/c-MYB and methylation, the authors provide experimental data suggesting that ZEB1 upregulates c-myb (western blot in Fig. 4C) that in turn transcriptionally activates MGMT promoter (ChIP fig. 4D) without affecting its methylation (Fig. S4). The Referee raised the hypothesis that low-proliferating cells (ZEB1-positive) are tumorigenic because they are cancer stem cells (and not simply by expression of ZEB1) while high-proliferating cells (ZEB1-negative) are non tumorigenic transient amplifying cells. In the revised manuscript, the authors have removed the low- and high-proliferating cell story and focused on the demonstration that ZEB1 can induce stemness-associated genes in glioblastoma, promote sphere formation in vitro and tumor formation in vivo, thus convincigly concluding that ZEB1-positive cells are a cancer stem cell-like population. The authors replied convincingly to the last major points raised by the Referee and also to the other EMBO 13

14 clarifications requested. Referee #2. The Referee believes that the specific role of Myb and MGMT in mediating ZEB1-induced chemoresistance is reasonably supported by experimental evidence but unlikely to be the whole story and would require further work. The revised manuscript includes new data on western blots after anti-mir-200, additional cell lines as well as knockdown and rescue experiments for MGMT in Figure 4, and a more thorough validation in patient specimens in Figure 6. However, in agreement with Referee's opinion the authors' claim is still not strongly supported. For instance, why, in contrast with c-myb levels, MGMT levels of anta.control-treated cells are similar to those of anta-mir-200? Similar observations for TMZ-sensitivity (Fig. 4C). We have performed additional knockdown experiments for anta-control and anta-mir-200, and have replaced the MGMT immunoblot images in Figure 4C. New MTT experiments also resulted in more pronounced changes in TMZ sensitivity that are more in line with c-myb and MGMT expression levels. Does depletion of mir-200 or c-myb influence the ZEB1-modulated (by knockdown or overexpression) TMZ sensitivity, reflecting the levels of the ZEB1/miR/c-myb/MGMT axis members? The revised manuscript now includes additional rescue experiments where we antagonized mir-200 in ZEB1 knockdown cells, knocked down c-myb in ZEB1 overexpressing cells, and overexpressed mir-200 in ZEB1 overexpressing cells. These experiments, which affect TMZ sensitivity as expected further strengthen our data on the ZEB1/miR/c-MYB/MGMT pathway and are presented in Figure S4A, B and E. The sentence "we found that ZEB1 reduced expression levels of ROBO1 (Fig. 3A) while overexpression of ZEB1 had the opposite effect (Fig. 3B)" on page 8 is confusing and I believe incorrect. The reviewer is absolutely correct the sentence should be we found that ZEB1 knockdown reduced expression levels and has been changed accordingly. The Referee required more experimental work to support the claimed modulation of invasion by ROBO1. The new data including increased invasion after overexpression of ZEB1, association of ROBO expression with the invasion front in vivo, and reduced invasion after knockdown of ROBO, all address this point. However, a few points remain to be explained. i) The lack of overall reduced N-cadherin protein levels after ZEB1 knockdown (Fig. 3A), is contrasting the reduced expression in ZEB1-positive cells in tumor xenografts (Fig. 2C). The supposed regulation of N-cadherin localization, rather than expression, is not proven. We thank the reviewer for these helpful suggestions. The revised manuscript contains additional data supporting the changes in N-cadherin localization in Figure 3B and Figure S3. New immunofluorescence images demonstrate that N-cadherin is distributed more evenly across the cellmembrane in control cells, but highly localized to cell-cell contact membranes in ZEB1 knockdown cells. Corroborating our immunoblot data, we find no significant differences in overall fluorescence intensity for N-cadherin between the two conditions. Immunofluorescence imaging demonstrates that ROBO1 is present at the cell membrane in controls, but not ZEB1 knockdown cells. Finally, we performed flow cytometry using an antibody against the extracellular portion of N-cadherin and show that the protein is present in equal amounts on both cell populations. We interpret these data that N-cadherin is tethered to the cytoskeleton in non-invasive ZEB1 knockdown cells, resulting in its greater localization to cell-cell contacts, while ROBO1 detaches N-cadherin from its cytoskeletal anchor in invasive controls, resulting in greater diffusion of N-cadherin across the cell membrane, which contributes to higher cell motility. The results and discussion sections have been updated accordingly. EMBO 14