Phenotypic Plasticity of Fibroblasts during Mammary Carcinoma Development

Round 1

Reviewer 1 Report

All flow and design are much influencing and informative. 

However, some question and advice are required for further understanding. 

1) First of all conceptually, I'm wondering whether CAFs genotype transforms according to time relapse or geographic proportion. 

2) cCAF, vCAF, and dCAF gene expression is much higher in late stage. Basically, what does it mean, early vs late? Pathologically DCIS (preinvasive) or invasive is easily classified. However, in the same invasiveness stage, early and late mean stage? or size? If so. what is the criteria between early vs late, clinically or pathologically ? 

3) dCAF signifying EMT is confusing in Fig 1. Why SMA and dCAF is not correlated? 

4) Fig 3 pathological image is also confusing, why early? Looking into both figures, all looks like invasive ca. 

5) Fig 5/6 demonstated mRNA and protein correlation in tumor stroma. If so, microRNA or epigenetic silencing or posttranscriptional modification and so on never happens in CAF? 

Those questions could be a curiosity in less expert than you authors, I hope to answer properly.  

Author Response

We are grateful for the reviewers’ constructive comments. Dealing with them has improved our manuscript considerably. We addressed all questions, whenever possible/required by new experiments or otherwise by discussion. Please find our detailed point-by-point reply below. Changes in the manuscript addressing the reviewers’ queries are underlined. Primary literature cited within this letter is displayed in a reference list at the end of the text. 

 

 

Reviewer 1

All flow and design are much influencing and informative. However, some question and advice are required for further understanding. 

 

1) First of all conceptually, I'm wondering whether CAFs genotype transforms according to time relapse or geographic proportion. 

Reply: This is an important question we unfortunately do not have a definite answer for. It is important to note that tumor development in the PyMT model is heterogeneous [1]. Tumors develop in each of the ten mammary glands at different times. Moreover, within a single mammary gland, late and early tumors can be found at the same time [1]. When looking at a single mammary gland in animals that are around 18-20 weeks old, we always observed late carcinomas, but at the same time hyperplasia, adenoma and early carcinoma lesions. In these lesions, fibroblast phenotypes were similar to those observed in mammary glands of younger mice (8-12 weeks) that did not contain late carcinomas (phenotypes as shown in Figures 1, 5, and 6 for early carcinoma). Thus, we conclude that rather the proximity to late-stage tumor cells or a corresponding microenvironment rather than the age of the mice was responsible for the accumulation of the specific fibroblast phenotypes we observed. This is supported by GSEA analysis where a tendency (significant p-value but insignificant FDR) of an increased hypoxia response in late-stage CAFs was apparent, which is more prevalent in late-stage carcinomas that have large hypoxic/necrotic areas. Of course tumor, increased tumor cell necrosis may be a driver of the inflammatory signature in late-stage through the release of danger-associated molecular patterns CAFs [2]. We now discuss this issue at the beginning of the discussion section on p.17 of the revised manuscript.  

 

2) cCAF, vCAF, and dCAF gene expression is much higher in late stage. Basically, what does it mean, early vs late? Pathologically DCIS (preinvasive) or invasive is easily classified. However, in the same invasiveness stage, early and late mean stage? or size? If so. what is the criteria between early vs late, clinically or pathologically? 

Reply: With the phrase ‘early tumors’ we summarize hyperplasia, adenoma/MIN and early carcinoma (comparable to human ductal carcinoma in situ with early invasion (DCIS+EI)) as defined by [1]. ‘Late tumors’ are only late stage carcinomas with invasion also as defined by Lin et al. However, the limitations as outlined above apply. At least we are sure that early tumors did not contain late carcinomas (defined pathologically). We have now added this information to the Results section on p.7. We hope that this explanation is sufficient.  

3) dCAF signifying EMT is confusing in Fig 1. Why SMA and dCAF is not correlated? 

Reply: In cancer, EMT (which is then called type 3 EMT) is usually not associated with αSMA expression by tumor cells [3]. However, EMT-induced expression of αSMA can be found in basal type carcinomas. Since the PyMT model represents luminal type carcinoma [4], the expression of αSMA by tumor cells undergoing EMT (which likely are dCAF) is not expected. Moreover, the PyMT model does not represent a tumor model with a high frequency of EMT [4].

 

 

4) Fig 3 pathological image is also confusing, why early? Looking into both figures, all looks like invasive ca. 

Reply: Indeed, in Figure 3 we show an example of early carcinoma (comparable to human ductal carcinoma in situ with early invasion (DCIS+EI) as defined by Lin et al. 2003 [4]. This represents the most advanced stage we found in mice at an age of 8-12 weeks. We hope that the explanation given above is sufficient to understand the choice of time points in our study. We also hope the reviewer can appreciate the difference in histology of early and late carcinoma shown in Figure 3.

 

 

5) Fig 5/6 demonstated mRNA and protein correlation in tumor stroma. If so, microRNA or epigenetic silencing or posttranscriptional modification and so on never happens in CAF? 

Reply: We thank the reviewer for pointing this out. Certainly, regulation of mRNA stability, translation, as well as epigenetic mechanisms operate in CAF. Also post-translational modifications are expected to affect protein function. For instance, we now analyzed nuclear localization of p65, which depends on post-translational phosphorylation. Moreover, we observed altered expression of Ep300 and Sirt1, which can affect histone acetylation status. Our study focused on mRNA and protein correlation and indeed we did not observe that mRNA and protein did not correlate in any of our targets. However, since this does not mean that all genes in our gene signatures follow this pattern of regulation, we now mention this limitation of our approach in the discussion section on p. 20 and 21.

 

Those questions could be a curiosity in less expert than you authors, I hope to answer properly. 

 

References

Lin, E.Y.; Jones, J.G.; Li, P.; Zhu, L.; Whitney, K.D.; Muller, W.J.; Pollard, J.W. Progression to malignancy in the polyoma middle T oncoprotein mouse breast cancer model provides a reliable model for human diseases. Am J Pathol 2003, 163, 2113-2126. Kuraishy, A.; Karin, M.; Grivennikov, S.I. Tumor promotion via injury- and death-induced inflammation. Immunity 2011, 35, 467-477. Zeisberg, M.; Neilson, E.G. Biomarkers for epithelial-mesenchymal transitions. J Clin Invest 2009, 119, 1429-1437, doi:10.1172/JCI36183. Hollern, D.P.; Andrechek, E.R. A genomic analysis of mouse models of breast cancer reveals molecular features of mouse models and relationships to human breast cancer. Breast Cancer Res 2014, 16, R59, doi:10.1186/bcr3672.

 

Reviewer 2 Report

In this manuscript, Elwakeel and co-workers have used the MMTV-PyMT breast cancer model to characterise the stromal fibroblasts during three stages of tumour progression: normal breast, early and late carcinoma. In particular, they have used markers of four CAF sub-populations that have been recently characterised by the Pietras team in late stage MMTV-PyMT tumours. First the authors performed immunofluorescence staining of tissues and then gene expression analysis of FACS sorted bulk fibroblast population to evaluate changes in CAF subpopulations during breast cancer progression. The authors then confirmed changes in CAF subpopulation during tumour progression using FACS analysis. Furthermore, they use their gene expression data to identify a gene signature of 906 genes that discriminates CAFs in early and late carcinoma; they show that the signature reflects changes in CAF subpopulations and they validate the regulation of two genes at the protein levels. A GSEA analysis suggests that there is increased inflammatory signalling and NFKB activation in late stage carcinoma. Finally, they use their CAF signatures of early and late carcinoma for predictive analysis (tumour stage and survival) using publicly available gene expression data of breast cancer.

This study is potentially interesting because it addresses an important question in field, since only recently it has become clear that there exist subpopulations of CAFs with distinct functions. Overall the study is well designed and the experiments seem properly performed and analysed (I cannot comment on the quality of the random forest analysis, since I have no experience). What is somehow disappointing is that the authors spent a vast part of the manuscript to show that their data match with what has been previously found in the Pietras’ work. While this is important, it also limits the novelty of their study. Some more data analysis to better describe differences between early and late stage would help to improve the impact of their work. The authors should also discuss their results in the context of other relevant works that have addressed similar questions. Additionally, a more focused validation of the data could help with strengthening their findings.

Specific comments

Can the authors use their three stage gene expression data for fuzzy c-means clustering to better group genes based on their temporal profile?

Increased inflammation is a quite expected change associated to tumour development. While it is important that the authors showed that with their gene expression data, have the authors tried other type of enrichment analyses (e.g. using Reactome or GO) to identify more categories that describe the regulated genes?

Since the authors have found increases inflammation in later stage of carcinogenesis, I would suggest to perform some validation of their results based on this finding, instead of on two proteins, Otx1 and Hexim1, which are not clearly related to inflammation. Moreover, why were those two chosen for validation? How many hits in total were chosen for validation, how many were successful and how many not?

References to relevant works that have addressed similar questions are missing. In particular:

a. Calvo F. et al Nature Cell Biology 2013 (Erik Sahai Lab): gene expression and functional analyses were performed on fibroblasts isolated from different stages of breast cancer progression in the MMTV-PyMT model;

b. Raz Y. et al Journal Experimental Medicine 2019 (Neta Erez Lab): gene expression analysis, as well as tissue immunostaining analyses and transplantation experiments were performed to characterise two subpopulations of CAFs identified in the MMTV-PyMT model.

The authors should discuss their results in the context of these two studies.

Figure 7. I am not a big fan of using gene expression data of total tumour tissue to assess predictive value of stromal signatures, because I find results difficult to interpret. The fact that the predictive ability of the top 20 gene signature is very week, as also mentioned by the authors in the discussion and clearly shown in Figure 7, also suggests that this may not be the best way to use available gene expression data. For example, does the amount of stroma in the analysed tumour tissues influence the results? One suggestion is to use only tumours that contain similar amounts of stroma (e.g. >40%. This information should be available for the TCGA datasets). Another option may be to use gene expression data from laser-captured microdissected tumour stroma (e.g. works from Morag Park’s lab).

Minor

Page 6, line 196: since there are batch effects, the authors should clearly state what samples were included in each batch. In the Methods, the authors also mention the batch effect (page 14, line 442) and refer to supplementary Figure 1A. However, in that figure there is no indication of what component indicates such batch effect. Please add this information, too.

Page 7, Table 1: add what ES, NOM and NES mean. How is it possible that an adjusted p-value of 0.15 passed the significance test?

The authors should explain why they decided to compare their work to the 4 subpopulations identified by Kristian Pietras work and not to those identified by other groups (e.g. Neta Erez).

Reference to specific works (instead of one review) should be included in the Introduction, for example when speaking about the epigenetic control in CAFs, page 2, line 79.

The first paragraph of the discussion is not very clear. I found difficult to understand what the take home message is. Also for other parts of the discussion it is not very clear what the message that the authors want to give is.

Page 11, line 328: the authors show that their signature of early stage carcinoma CAFs correlates with stage 0-1 breast cancers. Could that be the reason why that signature is indicative of better survival? Would the results be the same if the authors would plot the survival curves based on tumour stage? This possibility should be discussed in the manuscript.

Author Response

We are grateful for the reviewers’ constructive comments. Dealing with them has improved our manuscript considerably. We addressed all questions, whenever possible/required by new experiments or otherwise by discussion. Please find our detailed point-by-point reply below. Changes in the manuscript addressing the reviewers’ queries are underlined. Primary literature cited within this letter is displayed in a reference list at the end of the text. 

 

 

Reviewer 2

 

In this manuscript, Elwakeel and co-workers have used the MMTV-PyMT breast cancer model to characterise the stromal fibroblasts during three stages of tumour progression: normal breast, early and late carcinoma. In particular, they have used markers of four CAF sub-populations that have been recently characterised by the Pietras team in late stage MMTV-PyMT tumours. First the authors performed immunofluorescence staining of tissues and then gene expression analysis of FACS sorted bulk fibroblast population to evaluate changes in CAF subpopulations during breast cancer progression. The authors then confirmed changes in CAF subpopulation during tumour progression using FACS analysis. Furthermore, they use their gene expression data to identify a gene signature of 906 genes that discriminates CAFs in early and late carcinoma; they show that the signature reflects changes in CAF subpopulations and they validate the regulation of two genes at the protein levels. A GSEA analysis suggests that there is increased inflammatory signalling and NFKB activation in late stage carcinoma. Finally, they use their CAF signatures of early and late carcinoma for predictive analysis (tumour stage and survival) using publicly available gene expression data of breast cancer.

This study is potentially interesting because it addresses an important question in field, since only recently it has become clear that there exist subpopulations of CAFs with distinct functions. Overall the study is well designed and the experiments seem properly performed and analysed (I cannot comment on the quality of the random forest analysis, since I have no experience). What is somehow disappointing is that the authors spent a vast part of the manuscript to show that their data match with what has been previously found in the Pietras’ work. While this is important, it also limits the novelty of their study. Some more data analysis to better describe differences between early and late stage would help to improve the impact of their work. The authors should also discuss their results in the context of other relevant works that have addressed similar questions. Additionally, a more focused validation of the data could help with strengthening their findings.

 

Specific comments

 

1) Can the authors use their three stage gene expression data for fuzzy c-means clustering to better group genes based on their temporal profile?

Reply: In the present case, we performed RNA-seq on only two stages (early vs. late carcinoma), while the untransformed mammary gland was only considered in immunohistological stainings. We agree with the Reviewer that fuzzy c-means clustering is a versatile tool to visualize temporal gene expression profiles. Such a soft clustering technique would benefit most if more than two clusters should be formed (k/c > 2). However, as shown in Supplementary Figure 1F, the elbow heuristic indicated that k=2 results in the optimal number of clusters for our data. This gives a clear partitioning of the genes into up- and down-regulation between early and late carcinoma samples (Figure 4).

We tested the suggested approach and compared our previous k-means clustering to fuzzy c-means clustering. As shown in Response Figure 1 below, clustering with c=2 gave an identical partitioning of the data, supporting the robust changes across replicates between the two stages. Since no difference was observed between both approaches, we kept the original figure in the manuscript.

 

Response Figure 1: C-means clustering compared to k-means clustering. Genes were clustered with the indicated method and projected on a two-dimensional scale for visualization via PCA.

 

2) Increased inflammation is a quite expected change associated to tumour development. While it is important that the authors showed that with their gene expression data, have the authors tried other type of enrichment analyses (e.g. using Reactome or GO) to identify more categories that describe the regulated genes?

Reply: We used PANTHER V14 for Reactome and GO term analysis (Table 1). These analyses mainly confirmed increased inflammation in late-stage CAFs. Apart from that, an enrichment of protease inhibitors was identified. For early-stage CAFs, GO analysis revealed an increase in a varying set of transcriptional regulators, which includes transcription factors as well as epigenetic regulators. These findings are now shown in Table 1 and integrated into the discussion in the revised version of our manuscript. We would like to point out that, although increased inflammation is certainly associated with tumor development, it depends on the stage of the tumor and individual cell types which type of inflammation is apparent. NF-κB-driven (anti-microbial) inflammation is involved in oncogenic transformation, but is also associated with tumor rejection at least in lymphocytes [5,6]. Therefore, our finding that an NF-κB signature is induced in late-stage compared to early stage CAFs is not necessarily trivial. Reasons why NF-κB activity may be induced in CAFs rather late during tumor development may be, for instance, increased tumor cell necrosis, which is apparent only in late carcinomas in the PyMT model (see also response to comment 1 of reviewer 1). 

3) Since the authors have found increases inflammation in later stage of carcinogenesis, I would suggest to perform some validation of their results based on this finding, instead of on two proteins, Otx1 and Hexim1, which are not clearly related to inflammation. Moreover, why were those two chosen for validation? How many hits in total were chosen for validation, how many were successful and how many not?

Reply: When selecting our initial targets, we systemically followed the gene list in terms of significance (highest significance first) and did not consider genes with a low base mean (< 1000, likely not highly expressed) as well as genes encoding secreted proteins (usually not detectable). We then tested the remaining genes for which antibodies existed that were previously tested. Of these, Otx1 and Hexim1 produced good staining results. We did not neglect targets that did not follow our gene expression profile. Targets that we tested, but for which the antibodies did not yield sufficient staining quality were, for instance, Mmp13, Ep300, Cav1, Glul, Ifitm1, and Slpi.

The genes related to increased inflammation were mostly secreted proteins, which are usually unsuitable for histological analysis. Therefore, to follow the reviewer’s suggestion, we analyzed the expression of the NF-κB subunit p65 in the nucleus, which is required for NF-κB signaling. We unfortunately did not find an antibody against phosphorylated p65 of sufficient quality. The data related to nuclear p65 indicate an increase in α-SMA+ late-stage CAFs, but no difference (in total stroma) or even an increase (in α-SMA- stroma) in early-stage tumors. The data are shown in Figure 5. IN early-stage tumors, nuclear p65 was mainly observed in cells with a lymphocyte morphology, which likely affected histology analysis. While NF-κB activity, e.g. in CAFs and epithelial cells, may promote tumorigenesis, it is required for the activity of anti-tumor lymphocytes [5,6] as indicated above. Thus, NF-κB activity in individual cell types likely affects tumor development in different manners. Besides p65, we also tested antibodies against the transcriptional regulators identified by GO term analysis of early-stage tumors. Of those tested, Foxo1 and Sirt1 gave sufficient staining results within the time frame of this revision (Figure 5). We interestingly observed co-expression of these two markers in a subset of early-stage CAFs as well as late-stage CAFs, which were of higher relative abundance in early stages as suggested by our transcriptome data. Our new findings are described on p. 12/13 and discussed on p. 19 and 20 of the revised manuscript.       

 

 

4) References to relevant works that have addressed similar questions are missing. In particular:

Calvo F. et al Nature Cell Biology 2013 (Erik Sahai Lab): gene expression and functional analyses were performed on fibroblasts isolated from different stages of breast cancer progression in the MMTV-PyMT model; Raz Y. et al Journal Experimental Medicine 2019 (Neta Erez Lab): gene expression analysis, as well as tissue immunostaining analyses and transplantation experiments were performed to characterise two subpopulations of CAFs identified in the MMTV-PyMT model.

The authors should discuss their results in the context of these two studies.

Reply: We are grateful for this suggestion and now discuss the studies pointed out by the reviewer on p. 19 and 21.

 

 

5) Figure 7. I am not a big fan of using gene expression data of total tumour tissue to assess predictive value of stromal signatures, because I find results difficult to interpret. The fact that the predictive ability of the top 20 gene signature is very week, as also mentioned by the authors in the discussion and clearly shown in Figure 7, also suggests that this may not be the best way to use available gene expression data. For example, does the amount of stroma in the analysed tumour tissues influence the results? One suggestion is to use only tumours that contain similar amounts of stroma (e.g. >40%. This information should be available for the TCGA datasets). Another option may be to use gene expression data from laser-captured microdissected tumour stroma (e.g. works from Morag Park’s lab).

Reply: We thank the Reviewer for pointing out this potential source of bias. To our knowledge, the stroma content of the tissue samples is not directly available in TCGA. We therefore obtained stromal estimates from published data ([7]; downloaded from https://bioinformatics.mdanderson.org/estimate/disease.html), in which a ‘stromal score’ based on gene expression profiles is given to each sample to define the proportion of stroma within the sampled tissue.

Based on these data, we performed the following analyses:

We compared the stromal content of the tumor samples that were used in the context of our model. Importantly, we did not observe any systematic shift or bias of the stromal score between the tumor stages. Moreover, the distribution of stromal scores did not differ between predicted and annotated tumor stages, supporting that the random forest predictions are not strongly influenced by the amount of stroma in the bulk tumor samples. In the new Supplementary Figure 2C,D, we now show the stromal scores and compare their distribution between the groups predicted by our model or the stage annotations given by TCGA. In addition, we added the stromal score per sample to the gene expression heatmap in Figure 8C. Following the Reviewer’s suggestion, we trained the random forest model on subsets of samples with high or low stroma content. To this end, we split the training data based on their stromal score and then trained two random forest models on the 25% samples with the highest or lowest scores. As expected, due to the reduced number of training samples, both models performed worse than the original model based on the full dataset (see Response Figure 2 below). Importantly, however, there was no difference in the error rate between both models, indicating that tumor samples with higher stroma content do not perform better per se. Since these models are training on very few observations, we decided not to include this analysis in the revised manuscript.

Response Figure 2: Random forest model trained on subsets defined from the sample stroma content. The full training data set was split by 25% quantiles. Quantiles with the highest (N = 43) and lowest (N = 42) stroma content were used for model training. The out-of-bag (OOB) error rate is shown for each model to evaluate performance. For comparison the RF classifier was trained on 20 randomly picked subsets.

 

 

Minor

1) Page 6, line 196: since there are batch effects, the authors should clearly state what samples were included in each batch. In the Methods, the authors also mention the batch effect (page 14, line 442) and refer to supplementary Figure 1A. However, in that figure there is no indication of what component indicates such batch effect. Please add this information, too.

Reply: We apologize for the lack of clarity. The samples in this study were prepared in two subsequent rounds. The resulting batch effect can be seen in Supplementary Figure S1A (PC1). The term batch is now added to the samples instead of type in Figure S1A and the samples are now labeled with batch1 and batch2. Moreover, we have modified the figure legend accordingly.

 

 

2) Page 7, Table 1: add what ES, NOM and NES mean. How is it possible that an adjusted p-value of 0.15 passed the significance test?

Reply: We apologize for not including the abbreviations in the first version of our manuscript. They are now spelled out in the legend to table 1. GSEA uses an FDR of 0.25 to indicate significance. The reason for this is that ‘An FDR of 25% indicates that the result is likely to be valid 3 out of 4 times, which is reasonable in the setting of exploratory discovery where one is interested in finding candidate hypothesis to be further validated as a results of future research. Given the lack of coherence in most expression datasets and the relatively small number of gene sets being analyzed, using a more stringent FDR cutoff may lead you to overlook potentially significant results.’ For more information, please see:

http://software.broadinstitute.org/cancer/software/gsea/wiki/index.php/FAQ#Why_does_GSEA_use_a_false_discovery_rate_.28FDR.29_of_0.25_rather_than_the_more_classic_0.05.3F)

It is important to note that most papers using GSEA show the nominal p-value alone and omit the FDR, although the nominal p-value is not adjusted for gene set size and multiple hypothesis testing. If we would have followed this example a large number of potentially interesting gene sets would have been identified, including modulation of TGF-β signaling and hypoxia. However, this approach certainly would not have been correct.  

 

3) The authors should explain why they decided to compare their work to the 4 subpopulations identified by Kristian Pietras work and not to those identified by other groups (e.g. Neta Erez).

Reply: We have chosen to compare our data with the Pietras work based on comparability of the system and method (isolation of fresh fibroblasts, analyzed by deep sequencing). The study by Raz et al. used a Nanostring assay of only 561 genes rather than the unbiased approach we used. Calvo et al. used cultured and immortalized cells, which are not comparable with the cells we analyzed.   

 

 

4) Reference to specific works (instead of one review) should be included in the Introduction, for example when speaking about the epigenetic control in CAFs, page 2, line 79.

Reply: We replaced the reference as suggested by the reviewer

 

 

5) The first paragraph of the discussion is not very clear. I found difficult to understand what the take home message is. Also for other parts of the discussion it is not very clear what the message that the authors want to give is.

Reply: We rephrased parts of the discussion in order to better indicate the main messages of our data.

 

 

6) Page 11, line 328: the authors show that their signature of early stage carcinoma CAFs correlates with stage 0-1 breast cancers. Could that be the reason why that signature is indicative of better survival? Would the results be the same if the authors would plot the survival curves based on tumour stage? This possibility should be discussed in the manuscript.

Reply: Following this important point, we compared our early-stage CAF signature with survival only in stage 0-1 breast tumors, which revealed also better survival based on the signature within this group of patients (shown in Fig. 7E,F). These data suggest that survival is at least partially independent of stage.  

 

References

Lin, E.Y.; Jones, J.G.; Li, P.; Zhu, L.; Whitney, K.D.; Muller, W.J.; Pollard, J.W. Progression to malignancy in the polyoma middle T oncoprotein mouse breast cancer model provides a reliable model for human diseases. Am J Pathol 2003, 163, 2113-2126. Kuraishy, A.; Karin, M.; Grivennikov, S.I. Tumor promotion via injury- and death-induced inflammation. Immunity 2011, 35, 467-477. Zeisberg, M.; Neilson, E.G. Biomarkers for epithelial-mesenchymal transitions. J Clin Invest 2009, 119, 1429-1437, doi:10.1172/JCI36183. Hollern, D.P.; Andrechek, E.R. A genomic analysis of mouse models of breast cancer reveals molecular features of mouse models and relationships to human breast cancer. Breast Cancer Res 2014, 16, R59, doi:10.1186/bcr3672. Pires, B.R.B.; Silva, R.; Ferreira, G.M.; Abdelhay, E. NF-kappaB: Two Sides of the Same Coin. Genes (Basel) 2018, 9, doi:10.3390/genes9010024. Erez, N.; Truitt, M.; Olson, P.; Arron, S.T.; Hanahan, D. Cancer-Associated Fibroblasts Are Activated in Incipient Neoplasia to Orchestrate Tumor-Promoting Inflammation in an NF-kappaB-Dependent Manner. Cancer Cell 2010, 17, 135-147, doi:10.1016/j.ccr.2009.12.041. Yoshihara, K.; Shahmoradgoli, M.; Martinez, E.; Vegesna, R.; Kim, H.; Torres-Garcia, W.; Trevino, V.; Shen, H.; Laird, P.W.; Levine, D.A., et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nature communications 2013, 4, 2612, doi:10.1038/ncomms3612.

Round 2

Reviewer 2 Report

The authors have provided a revised version of the manuscript where they have also included additional data. The quality of the work has improved and they have addressed previous concerns.