Epithelial-mesenchymal transition spectrum quantification and its efficacy in deciphering survival and drug responses of cancer patients

Epithelial-mesenchymal transition (EMT) is a reversible and dynamic process hypothesized to be co-opted by carcinoma during invasion and metastasis. Yet, there is still no quantitative measure to assess the interplay between EMT and cancer progression. Here, we derived a method for universal EMT scoring from cancer-specific transcriptomic EMT signatures of ovarian, breast, bladder, lung, colorectal and gastric cancers. We show that EMT scoring exhibits good correlation with previously published, cancer-specific EMT signatures. This universal and quantitative EMT scoring was used to establish an EMT spectrum across various cancers, with good correlation noted between cell lines and tumours. We show correlations between EMT and poorer disease-free survival in ovarian and colorectal, but not breast, carcinomas, despite previous notions. Importantly, we found distinct responses between epithelial- and mesenchymal-like ovarian cancers to therapeutic regimes administered with or without paclitaxelin vivo and demonstrated that mesenchymal-like tumours do not always show resistance to chemotherapy. EMT scoring is thus a promising, versatile tool for the objective and systematic investigation of EMT roles and dynamics in cancer progression, treatment response and survival.

Thank you for the submission of your manuscript to EMBO Molecular Medicine. We have now received comments from the Reviewers whom we asked to evaluate your manuscript.
We apologise that it has taken more time than we would have liked to return a decision, but unfortunately the Reviewers delivered their evaluations with some delay. I trust that the inevitable frustration due to the delay will be somewhat tempered by the fact that the Reviewers are all quite supportive and, in my opinion, offer valuable suggestions to improve your manuscript.
Reviewer 1 specifically acknowledges the significant usefulness of the scoring tool and in this respect would like you to provide an algorithm to allow others to readily apply the approach to their favourite system. S/he also would like you to verify whether the signatures perform well with a reduced set of genes. Finally, this Reviewer suggests that a parallel should be drawn with previously reported cancer stemness signatures.
Reviewer 2 raises interesting points with respect to pancreatic cancer (with reference to the alleged negative correlation between EMT and survival) and also suggests further analysis comparing mesenchymal vs. epithelial cancer types. These points appear very interesting and while to specifically address them might not be required, I believe that the information gained would be potentially very valuable and would increase the impact of your work. Reviewer 2 notes that some discrepancies between certain clinical samples and theoretically corresponding cell lines were not discussed. S/he also challenges the contention that EMT reversal leads to less metastases, mentions other points of interest and cautions against overstating the clinical applications of your tool.
Reviewer 3 would like more information on how the EMT signatures other than the breast cancer one were derived and validated. S/he also suggests verifying EMT status after patient stratification based on different subtypes of a given cancer. This Reviewer, similarly to Reviewer 2, also mentions the growing evidence that EMT reversal actually increases metastatic colonisation. Finally, Reviewer 3 suggests that overall and disease-free survival as well should be considered in your evaluations.
In conclusion, while publication of the paper cannot be considered at this stage, we would be pleased to consider a suitably revised submission, provided that the Reviewers' concerns are addressed as outlined above.
I look forward to receiving you revised manuscript as soon as possible.
***** Reviewer's comments ***** Referee #1 (Comments on Novelty/Model System): The presented study provides a potentially useful tool to evaluate and objectify EMT signatures in primary cancers as well as cancer-derived tumor cells. Most importantly, it would substantially increase the impact of the study if the authors provide an easy to handle on line tool allowing researchers to rapidly assess EMT signatures in their biological samples of interest or primary patient material. This would also increase the impact of the study and improve signatures to be used in evaluating therapy regimes.

Referee #1 (Remarks):
Tan et al. developed a method to quantify the EMT status of tumors and tumor-derived cell lines by developing a generic EMT gene expression signature. Based on transcriptomics of ovarian, breast, bladder, lung, colorectal and gastric cancers Tan et al. defined a gene expression signature that discriminates between epithelial (low, negative EMT score), mesenchymal (high, positive EMT score), or intermediate phenotypes, which display a certain degree of plasticity. Consistent with the hypothesis that this generic EMT signature monitors the cellular phenotype, functional studies demonstrated increased EMT scores upon e.g. knockdown of CDH1 or overexpression of TWIST in epithelial cancer cells. By 'EMT scoring' clinical samples and cell lines from various cancers, the authors identified cancer type-specific EMT states with e.g. predominant epithelial phenotypes in colorectal cancers. Most strikingly, however, the correlation of primary signatures with those observed in cancer-derived cells revealed a striking correlation and appeared barely biased by stromal content in primary material. In contrast to previous suggestions, poor patient overall (OS) and disease-free survival DFS were not exclusively associated with mesenchymal, but also with the epithelial cancer phenotypes. Moreover, by correlating the EMT score with chemotherapeutic responses of cancer cell lines Tan et al. suggest that the EMT-score indicates prevalence in the chemo-sensitivity of certain tumors instead of a general resistance or reduced sensitivity of MC-like tumors. This suggestion was further evaluated by analyzing Paclitaxel sensitivity of ovarian cancer patients. Supporting the hypothesis of phenotype-specific responses, Paclitaxel increased OS and DFS of patients suffering from mesenchymal ovarian carcinomas, but decreased the prognosis for patients suffering from epithelial ovarian carcinomas. The manuscript presents a highly interesting tool, that could substantially contribute to objectifying the evaluation of EMT phenotypes in vitro as well as malignancies. However, in its present form the scoring remains difficult to apply, involves too many genes whereas other potential markers are not considered and lacks an option to include the estimation and correlation with stemness phenotypes.
Major aspects to be addressed: 1) The authors should at least provide a easy to use algorithm allowing other researchers to evaluate the EMT-status of their biological system of choice by gene expression signatures derived by microarrays.
2) The authors should evaluate if their signatures performs equally well with a reduced set of genes analyzed. This would potentially allow evaluating EMT signatures based on smaller qRT-PCR arrays. At least the authors should identify the 'best performing' genes.
3) Along these lines it remains elusive why the authors did not include miRNAs as additional markers, since these were shown to be substantially deregulated in various cancer according to the EMT-status, for instance the miR-200 family. 4) The authors should attempt to assess the expression of putative CSC-markers reported for various cancers and try to correlate this with their EMT signature. In addition the authors have to correlate their signature with previously reported stemness signatures (e.g. Ben-Porath et al., 2008.) Minor aspects: 1) Wherefrom were the data derived used for analyses in Figure E1? Where are the references? The same applies for Figure 1B, C. The data processing and experimental procedures are not or barely described in the methods section or in the figure legend. 2) Figure E2B should be included in the main figure section. I would prefer one main figure together with Figure 3B. Accordingly, Figure 3A should be one main figure.
Referee #2 (Comments on Novelty/Model System): The reviewers wish to state that they are not able to evaluate in detail the soundness of the bioinformatics/statistics in the manuscript as they are not experts in that area. They were, however able to evaluate the importance and context of the question addressed, novelty and the biological background.
Referee #2 (Remarks): The authors provide a universal EMT-signature that can be applied to several cancer types. They show that this tool allows classification of tumor samples or cell lines due to their epithelial or mesenchymal genetic profile. In doing so, the authors very well appreciate the limitations of in vitro cell line models and distinguish between gene signatures of primary tumors and cell lines. In the end, the generated tumor-specific EMT-signature is correlated to clinical outcome. The authors address one of the most pressing questions in the field, i.e., is there a general EMT-signature and how does it affect clinical behaviour in different entities. They do so in an unbiased and careful manner, thereby putting out a tool (the generic EMT-signature) that should be of broad interest and utility to a lot of researchers from different fields of inquiry.
A few minor issues might be of interest to address.
For the creation of the generic, cancer-specific EMT-signature, the authors use gene sets of different tissues (bladder, breast, colorectal, gastric, lung and ovarian) ( Figure 2). Given the implications of EMT in pancreatic cancer (PC) and the emerging evidence for a negative correlation between EMT and the survival rates of patients with PC (Ref. [1][2][3], inclusion of PC-specific EMT-signatures should be taken into consideration. In Figure 3A, the generic EMT score is applied to several cancer entities, including data belonging to primary tumors and cell lines. Although many clinical samples and cell lines of the same cancer type overlap in their Epi/Mes classification, there are also cases of discrepancy (bladder, ovarian, lung and prostate carcinomas). Unfortunately, the authors do not discuss these differences. One possibility could be that the EMT-score for cell lines is less reliable due to smaller cohorts. Alternatively, it would mean that cell lines indeed transform during in vitro culturing and do not always reflect the phenotype of the correspondent primary tumor.
In the text belonging to Figure 3A, the authors draw the conclusion that the presented analysis might help to predict the origin of the tumor. One should be careful with this presumption, since at least in breast cancer it has been shown that basal-like tumors arise from luminal/epithelial, rather than from basal/mesenchymal populations (Ref. 4,and 5). Next, the authors investigate whether the generic EMT-signature has clinical implications ( Figure  4). They find that the EMT status does not entirely correlate with clinical response in patients suffering from breast or ovarian cancer. It is most surprising that the distribution of Epi and Mes phenotypes is very similar between patients that show a good clinical response and those suffering from a progressive disease. Together with the drug sensitivity data, this analysis shows that EMT should not be the only factor taken into consideration when deciding over treatment regimens. Nevertheless, the authors were able to show that mesenchymal ovarian cancers respond better to chemotherapy than their epithelial counterparts. To consolidate the ability of the EMT score to predict sensitivity to chemotherapy it would be necessary to assess additional data from several cancer types and maybe even compare cancer entities that tend to be more mesenchymal (e.g. colon) to the more epithelial ones (e.g.: liver or renal). In the discussion, the authors mention the EMT score as a method to assess effectiveness of EMT reversion therapies, aiming to inhibit the metastatic tumor potential. While the idea that the EMT score can detect phenotypic transitions is uncontestable, stating that the reversion of the EMT process leads to less metastasis is not entirely correct. It was recently shown that reversion of EMT even promotes metastatic colonization (Ref. 6). Taken altogether, EMT scoring can be very useful when classifying tumor entities, but one should be aware of its limitations when it comes to the choice of therapy and the prediction of clinical outcome. The authors therefore provide a versatile tool for tumor classification based on their EMT status, but its power in clinical applications should not be overrated. Future cancer subtype analysis and metastasis-free survival analysis are strongly suggested.

Referee #3 (Remarks):
In this manuscript the authors report a novel EMT scoring method to quantitatively estimate the degree of EMT in cell lines and tumors. The authors first developed cancer-specific EMT signatures using transcriptomics of a large collection of cancer cell lines and clinical samples, and demonstrated that they exhibit good correlation with published cancer-specific EMT signatures. From the established cancer-specific EMT signatures, they further derived a generic EMT signature and a method for universal EMT scoring. The authors also applied this EMT scoring to examine its efficacy in correlating EMT status with patient survival rates and responses to treatment. This is an interesting study that is aims at developing methods to quantitatively assess the correlation between EMT status and cancer characteristics. The authors developed and applied a valid methodology, and the quality and explanation of data is adequate, making the conclusion robust. The paper is therefore of high scientific interest.
I suggest the following possible revisions to improve the quality of this manuscript: 1. Figure 1 and E1. The breast cancer-specific EMT signature is well explained and validated. However, it is not sure how some of the other cancer-specific EMT signatures are derived and verified, for example the ovarian and bladder cancers, as depicted in Figure 2A.
2. Figure 3 and E2. It is claimed by the authors that EMT status does not necessarily correlate with poorer survival. Did the authors try performing the analyses after stratifying patients based on different subtypes of the same cancer? For example, will performing Kaplan-Meier analysis by subtypes of tumors (e.g. luminal, basal, ERBB2+, triple-negative subtypes of breast cancer) give a different result? 3. Figure 3 and E2. Given the critical roles of EMT in tumor cell dissemination and cancer metastasis, it is suggested that the authors also check metastasis-free survival, not only OS and DFS.
4. There is growing evidence suggesting that the reversal of EMT, the mesenchymal-epithelial transition (MET), may be necessary for efficient metastatic colonization. The authors should discuss this point and the potential complications in the interpretation of their results.
Minor concerns: 1. What are the values of the y axis in Figure 1C  Remark: The manuscript presents a highly interesting tool, that could substantially contribute to objectifying the evaluation of EMT phenotypes in vitro as well as malignancies. However, in its present form the scoring remains difficult to apply, involves too many genes whereas other potential markers are not considered and lacks an option to include the estimation and correlation with stemness phenotypes.
We appreciate the reviewer's positive comments regarding our tool. We agree with the Referee#1 that the tool in its current form is relatively costly and difficult to apply, especially in the clinical setting. We have carefully considered the three limitations of our scoring system outlaid by Referee#1 and have provided a detailed response to each in the respective comments: i) involves too many genes (Comment 2); ii) Other potential markers are not considered (Comment 3); iii) Lacks an option to include the estimation and correlation with stemness (Comment 4).
Major aspects to be addressed: The authors should at least provide an easy to use algorithm allowing other researchers to evaluate the EMT-status of their biological system of choice by gene expression signatures derived by microarrays.
Response: We agree with the Referee that it would be more useful if an online tool for EMT scoring is available. However, as we are in the process of patent filing (patent currently preparing a follow-up paper in requirement of patent filing, in which we will make the online tool available once we are granted such clearance.

Comment 2
The authors should evaluate if their signatures perform equally well with a reduced set of genes analyzed. This would potentially allow evaluating EMT signatures based on smaller qRT-PCR arrays. At least the authors should identify the 'best performing' genes.
Response: We agree with Referee#1 that the tool will be more attractive in terms of cost and utility, especially in the clinical setting, if the number of genes could be reduced.
Indeed, we are in the process of patent filing and are working on a follow-up paper using a reduced number of genes for diagnostic purposes (publication number: WO2013043132 A1). One possible way to identify the best-performing genes is based on the weights of differentially expressed genes between epithelial and mesenchymal states given in Table   E1A. However, as pointed out by the reviewer, it is important that the reduced set will perform equally well. Thus, we have studied the effect of reducing the number of genes in the generic EMT signature in terms of correlating EMT scores between a reduced and the full EMT signature in the manuscript. We performed the analysis for both tumors and cell lines. We compared the overall concordance in tumor EMT phenotype estimated by the full and reduced EMT signatures, and re-validated the EMT scoring on the reduced set using the same set of functional intervention dataset as in Fig. E3 and Table E3 of the revised manuscript. The EMT score computed from the reduced set has been added to Table E3 and E4A. The analysis and result are given in the Expanded View section 'Effect of reducing genes in generic EMT signature' and in Fig. E5.

Comment 3
Along these lines it remains elusive why the authors did not include miRNAs as additional markers, since these were shown to be substantially deregulated in various cancer according to the EMTstatus, for instance the miR-200 family.
Response: We agree with Referee#1 that including miRNAs in the signature might allow for the more precise quantification of EMT. However, as we seek to quantify EMT in various cancers, and because the development of our signature and scoring system was based on a relatively large number of cell lines and tumors, the amount of miRNA data is not as abundantly available as mRNA gene expression data. This prompted us instead to focus only on mRNA gene expression data. Furthermore, since our EMT scoring system is based on a generic method two-sample Kolmogorov-Smirnov test, the EMT signature is therefore customizable. The present generic EMT signature could be further refined or modified. In order to identify potential miRNA markers, we gathered 6 datasets where miRNA and mRNA expression data are available from GEO (bladder, GSE40355; pancreas, GSE32688; prostate, GSE21034; breast, ovarian, TCGA; multiple myeloma, GSE17498) (The Cancer Genome Atlas, 2012; The Cancer Genome Atlas Research, 2011; Donahue et al, 2012;Hecker et al, 2013;Lionetti et al, 2009;Taylor et al, 2010), and correlated generic EMT score with miRNA expression. Looking at miRNA implicated in EMT (Hao et al, 2014;Zhang & Ma, 2012), we found that the miR-200 family members (miR-200a, miR-200b, miR-200c, miR-141, miR-429) negatively correlate with the generic EMT score (Fig. E2). However, we also noted that the miRNA involved in promoting or suppressing EMT might be cancer-dependent, as the correlation or anti-correlation were not consistent in all cancers or with previous reports (Hao et al, 2014;Zhang & Ma, 2012).
This discrepancy may also stem from the platform-specific, cross-hybridization problem, as miRNA are short, closely related sequences (between family members), and hence measuring their expression is technically challenging (Mestdagh et al, 2014). Nevertheless, the expression of miR-200 and miR-34 (miR-34a, miR-34b, miR-34c) families were relatively consistent across bladder, breast, pancreas, prostate, and ovarian cancers; hence, scoring EMT using an miRNA-based EMT signature would complement and may allow for a more precise estimate of EMT when used in concert with an mRNA-based EMT signature. It is important to note that this assessment is preliminary, as only small cohorts were analysed. An in-depth investigation will be only possible when more data with both miRNA and mRNA gene expression data become available. We have included and discussed the result of the analysis in Fig. E2 and in the Expanded View section 'Generic EMT signature and miRNA' in the revised manuscript.

Comment 4
The authors should attempt to assess the expression of putative CSC-markers reported for various cancers and try to correlate this with their EMT signature. In addition the authors have to correlate their signature with previously reported stemness signatures (e.g. Ben-Porath et al., 2008.) Response: Because EMT is associated with the acquisition of stemness (Frisch et al, 2013;Huang et al, 2012;Tam & Weinberg, 2013;Thiery et al, 2009), as suggested by the Referee, we have assessed the correlation of the generic EMT score and stemness using 21 published stem cell gene sets found in the Molecular Signature Database (Msigdb v4.0) (Subramanian et al, 2005), as well as those published stem cell markers (Medema, 2013). The enrichment score from ssGSEA (Verhaak et al, 2013)  as CD44 and CXCR4 were observed in the majority of the cancer types. However, this analysis is preliminary due to certain limitations: the stem cell signatures were derived from different cell types, and no consideration was given to the different types of stem cells. For example, in breast cancer, there exist at least two types of stem cells (Liu et al, 2014). Nevertheless, we added this data into the Results section in the Expanded View, Fig.   E11 and commented on it in the Discussion section of the main text.
Minor aspects: 1) Wherefrom were the data derived used for analyses in Figure E1? Where are the references? The same applies for Figure  processing procedure and software used were given in 'Section Data preprocessing for Affymetrix microarray expression data'. Since the experimental procedures of these public datasets were described in their publications, and many datasets were involved, we did not duplicate these descriptions in the Materials and Methods. We instead provided the references in Tables E6 and E8.
2) Figure E2B should be included in the main figure section. I would prefer one main figure together with Figure 3B. Accordingly, Figure 3A should be one main figure.
Response: We have re-arranged the figures such that Figure E2B and Figure 3B form the new Figure 4 in the revised manuscript, and Figure 3A has been changed to the new Figure   3 in the revised manuscript, as suggested by Referee#1.

Referee #2 (Comments on Novelty/Model System):
The reviewers wish to state that they are not able to evaluate in detail the soundness of the bioinformatics/statistics in the manuscript as they are not experts in that area. They were, however able to evaluate the importance and context of the question addressed, novelty and the biological background.
Referee #2 (Remarks): The authors provide a universal EMT-signature that can be applied to several cancer types. They Response: We thank Referee#2 for supporting the importance of the tool and for raising several issues critical to the improvement of our manuscript. We have provided detailed responses to each of the issues raised in the following sections.
A few minor issues might be of interest to address.

Comment 1:
For the creation of the generic, cancer-specific EMT-signature, the authors use gene sets of different tissues (bladder, breast, colorectal, gastric, lung and ovarian) ( Figure 2 Fig. 3 of the revised manuscript, we observed an EMTed pancreatic cancer population in both tumors and cell lines. As we are seeking a generic EMT signature that is common amongst the different cancer types and since Figure   E3 in the revised manuscript shows that EMT scoring is capable of accurately predicting EMT status of pancreatic cancer cell lines, we believe the present EMT scoring is also applicable to pancreatic cancer. In addition, we showed in Table E1C that this generic EMT scoring is highly correlated with cancer-specific EMT. Hence, we opine that it is highly likely that incorporating a pancreatic-specific EMT signature would not change significantly the present generic EMT signature. Thus, we did not repeat the procedure to generate a pancreatic cancer-specific EMT signature or generic EMT signature, but rather, we performed additional analyses validating the applicability of this generic EMT signature in pancreatic cancer and added the data to the Results section, Expanded View and Fig. E4 of the revised manuscript. Furthermore, we have included two recent pancreatic datasets that have overall survival information, and have studied the correlation of EMT and survival in pancreatic cancer as well (Fig. 4A in revised manuscript). Even though it is not significant, there is a trend that EMT correlates with poorer overall survival, consistent with previously published results (Arumugam et al, 2009;Hotz et al, 2007;Nakajima et al, 2004).
Comment 2.1: In Figure 3A, the generic EMT score is applied to several cancer entities, including data belonging to primary tumours and cell lines. Although many clinical samples and cell lines of the same cancer type overlap in their Epi/Mes classification, there are also cases of discrepancy (bladder, ovarian, lung and prostate carcinomas). Unfortunately, the authors do not discuss these differences. One possibility could be that the EMT-score for cell lines is less reliable due to smaller cohorts.
Alternatively, it would mean that cell lines indeed transform during in vitro culturing and do not always reflect the phenotype of the correspondent primary tumour.
Response: We thank Referee#2 for pointing out this oversight. We agree with Referee#2 that the EMT score distribution of clinical samples and cell lines of the same cancer type were largely similar in most cases but that there were also discrepancies in terms of EMT score mean and range in some cases such as ovarian and prostate cancers. Although it is true that cell lines acquire additional transformations in culture, we believe the discrepancies stemmed from the differences in the number of samples, and that the data of cell lines and tumours were not paired. Hence, the different compositions for each cancer type such as histology, grade, and stage may have caused the difference in the distribution of the EMT scores. For example, in prostate cancer, the tumours were made up of primary and metastatic, recurrent and non-recurrent prostate carcinoma, whereas the 8 prostate cancer cell lines were derived mainly from metastatic sites (Table R1). We have provided this explanation of the discrepancies in the Results section in the revised manuscript. In the text belonging to Figure 3A, the authors draw the conclusion that the presented analysis might help to predict the origin of the tumour. One should be careful with this presumption, since at least in breast cancer it has been shown that basal-like tumours arise from luminal/epithelial, rather than Response: Our intention was to say that the wide range of the EMT score observed within a cancer type might reflect the difference in cell of origin in the cancer. For example, in breast cancer, subtypes Luminal-A and -B originated from mature luminal cells (hence more Epi), whereas the Basal subtype originated from luminal progenitors (hence more Mes) (Lim et al, 2009). We did not mean to imply that EMT is predictive of the cell of origin. We have removed these sentences from the Results section to avoid this confusion in the revised manuscript.
Comment 3: Next, the authors investigate whether the generic EMT-signature has clinical implications ( Figure   4). They find that the EMT status does not entirely correlate with clinical response in patients suffering from breast or ovarian cancer. It is most surprising that the distribution of Epi and Mes phenotypes is very similar between patients that show a good clinical response and those suffering from a progressive disease. Together with the drug sensitivity data, this analysis shows that EMT should not be the only factor taken into consideration when deciding over treatment regimens.
Nevertheless, the authors were able to show that mesenchymal ovarian cancers respond better to chemotherapy than their epithelial counterparts. To consolidate the ability of the EMT score to predict sensitivity to chemotherapy it would be necessary to assess additional data from several cancer types and maybe even compare cancer entities that tend to be more mesenchymal (e.g. colon) to the more epithelial ones (e.g.: liver or renal). profiles of responders and non-responders in these cancers (Fig. E8B). We found no significant difference between responders and non-responders in terms of EMT score, albeit there is a very slight trend that responders tend to have higher EMT score in predominantly Epi cancers (colorectal cancer), and a slight trend that responders tend to have a lower EMT score in predominantly Mes cancers (melanoma). We were not able to compare the results with that of the cell lines (Table E5) because the majority of these cohorts were treated with a combination of chemotherapeutics. Furthermore, as these data are from a relatively small cohort, further study is required to validate the current observation. These analyses have been added to the Results section, and the Expanded View of the revised manuscript.

Response
On the other hand, we were not able to verify the findings of either the benefit of Mes tumour to paclitaxel in other cancer types because the data were not available. Instead, we found three datasets with survival and treatment information (glioma, GSE43388; ER+ breast cancer, GSE16391, and multiple myeloma, GSE9782) (Desmedt et al, 2009;Erdem-Eraslan et al, 2013;Mulligan et al, 2007). Interestingly, we observed a differential response in Epi and Mes tumours (Fig. E10). Even though, in general, patients with glioma receiving radiotherapy and chemotherapy have a better overall survival, the benefit is greater in patients with a Mes glioma (p=0.0117). In contrast, patients with Epi multiple myeloma have a better disease-free survival rate when administered bortezomib instead of dexamethasone (p=0.0349). We observed no difference in ER+ tumours in patients administered with letrozole or tamoxifen in terms of EMT stratification. These data provide preliminary evidence of the potential benefit in stratifying patients by EMT status using generic EMT scoring. We concede that the cohorts were relatively small. However, we believe that our analysis would open the door to opportunities for incorporating the EMT scoring system into translational research protocols of clinical trials in the future when more data is available to assess the benefit of stratification by EMT.
Comment 4: In the discussion, the authors mention the EMT score as a method to assess effectiveness of EMT EMT may promote metastatic colonization. This paradigm, to an extent, fits our previously proposed model where mesenchymal micro-metastases must re-acquire an epithelial phenotype to proliferate at the metastatic site (Thiery, 2002). The original text in the manuscript was not accurate ("the EMT reversion therapy as a means to reduce the metastatic potential") in that we are not postulating that EMT reversal therapy could cure metastasis, but instead we meant that EMT reversal therapy may sensitize cancer cells to specific drugs. Furthermore, our intent was not to achieve a full reversal of EMT-which, as pointed out by Referees#2, #3 and Tsai et al. could be detrimental-rather, we were referring to modifying the EMT status along the spectrum to improve patient responses to treatment. In our experience, the use of a single agent does not fully reverse EMT but it is sufficient to reduce invasion (unpublished), anoikis, spheroidogenesis and clonogenicity in vitro (Huang et al, 2013), and in vivo using orthotopic grafting of a human cancer line (Sim & Thiery, 2014;manuscript in preparation From the established cancer-specific EMT signatures, they further derived a generic EMT signature and a method for universal EMT scoring. The authors also applied this EMT scoring to examine its efficacy in correlating EMT status with patient survival rates and responses to treatment.
This is an interesting study that is aims at developing methods to quantitatively assess the correlation between EMT status and cancer characteristics. The authors developed and applied a valid methodology, and the quality and explanation of data is adequate, making the conclusion robust. The paper is therefore of high scientific interest.
I suggest the following possible revisions to improve the quality of this manuscript: Comment 1: Figure 1 and E1. The breast cancer-specific EMT signature is well explained and validated.
However, it is not sure how some of the other cancer-specific EMT signatures are derived and verified, for example the ovarian and bladder cancers, as depicted in Figure 2A.
Response: Figure 1 depicted the procedure for deriving the cancer-specific EMT signature for the bladder, colorectal, gastric, and lung cancer-specific EMT signatures. Breast and ovarian cancer-specific EMT signatures were derived using a similar procedure (apart from Steps 1 and 2 depicted in Figure 1A) and were published elsewhere (Akalay et al, 2013;Miow et al, 2014); these were derived from comparing the profile of immunofluorescence staining of CDH1 and CDH2 (Materials and Methods). Since the cancer-specific EMT signatures were derived using the same procedure and have good correlation with the generic EMT signature (Table E1C), we performed additional analyses to verify only the validity of bladder cancer-specific EMT signature. The result is added in Fig. E1 (Prat et al, 2012). When stratified by molecular subtype, we observed a better disease-free survival (DFS) in patients with Basal and Claudin-Low breast cancers of a low EMT score than in those with a high EMT score (hazard ratio=0.6549, p=0.0089). There was no significant difference in other breast cancer subtypes when correlating EMT and DFS or overall survival. However, the correlation of EMT and DFS in Basal and Claudin-Low subtypes was not coherent in all breast cancer cohorts, probably due to the smaller sample sizes. Therefore, we did not report this finding initially. As pointed out by Referee#3, it might be of great interest to the reader, and we have added these results to Response: We thank the Referee for pointing this out, as this made us aware that we were not clear in the definition of disease-free survival (DFS). The definition of DFS we adopted in the manuscript broadly encompassed (local) recurrence-free survival, progression-free survival, and distant metastasis-free survival (DMFS). The majority of the data were in fact distant metastasis-free survival. We have indicated in Fig. 3B which cohort data are DMFS, and inserted the definition of DFS in the text.

Comment 4:
There is growing evidence suggesting that the reversal of EMT, the mesenchymal-epithelial transition (MET), may be necessary for efficient metastatic colonization. The authors should discuss this point and the potential complications in the interpretation of their results.
Response: We thank the reviewer for raising this important point. Tsai et al. is a landmark paper and we fully agree with Tsai et al. as well as with Referees#2 and #3 that reversing EMT may promote metastatic colonization. This paradigm, to an extent, fits our previously proposed model where mesenchymal micro-metastases must re-acquire an epithelial phenotype to proliferate at the metastatic site (Thiery, 2002). The original text in the manuscript was not accurate ("the EMT reversion therapy as a means to reduce the metastatic potential") in that we are not postulating that EMT reversal therapy could cure metastasis, but instead we meant that EMT reversal therapy may sensitize cancer cells to specific drugs. Furthermore, our intent was not to achieve a full reversal of EMT-which, as pointed out by Referees#2, #3 and Tsai et al. could be detrimental-rather, we were referring to modifying the EMT status along the spectrum to improve patient responses to treatment. In our experience, the use of a single agent does not fully reverse EMT but it is sufficient to reduce invasion (unpublished), anoikis resistance, spheroidogenesis and clonogenicity in vitro (Huang et al, 2013), and in vivo using orthotopic grafting of a human cancer line (Sim & Thiery, 2014;manuscript in preparation