Title: Genome-wide analysis reveals down-regulation of miR-379/miR-656 cluster in human cancers
Versions: 1 & 2 11 November 2012/ 21 March 2013
Reviewer number: 1
Reviewer: Prof Gregory J Goodall (nominated by Prof Mark Ragan)
Laddha et al. have profiled publicly available cancer microRNA expression profiles, finding miRNAs of the miR-379/miR-656 cluster to be downregulated in several cancer types, including glioblastoma. Several of these miRs were measured in a small number of glioblastomas by the authors, finding results consistent with the downregulation. Bioinformatic prediction of the targets of members of the cluster indicated an enrichment for genes involved in glioblastoma. Assessment of publicly available gene methylation data indicated the locus is hypermethylated in glioblastomas, consistent with its downregulation. MEF2A was measured in several glioblastomas and found to be reduced relative to controls, consistent with previous reports that MEF2 regulates expression of the locus. Surveying public data for several other cancers suggested downrugulation of the locus may occur in some other cancers.
Comment: The manuscript contains interesting observations regarding miR-379/miR-656 locus in glioblastoma and its likely regulation by MEF2A. The Results section is rather porly structured and explained, and the Figure legends lack detail, requiring constant reference to the Abstract and the Materials and Methods sections to follow what was done and assess the claims. The manuscript would benefit from a rewrite, inserting more information into the Results and Figure legends.
Response: We have incorporated more information in the results section and the figure legends as suggested by the reviewer.
Comment: Figures 2 and 3 shows the downregulation of many members of the cluster in glioblastomas relative to controls, but there is no mention of what the controls are, making interpretation difficult.
Response: In Figure
the normalized (level 3) expression data from TCGA repository has been used to analyze the miRNA expression difference. As mentioned in the ‘miRNA expression data and analysis’ section under ‘Materials and Methods’ the TCGA server provided brain miRNA expression from 10 control samples. In the revised manuscript we have added a supplemental file (supplement 7) with individual samples IDs (TCGA barcodes) for easy reference. As information provided by TCGA these 10 control samples were brain tissues from 10 unrelated individuals who did not suffer from GBM.
we have reported data of real-time PCR for samples collected by us. Here the three non-GBM controls were patients of transitional meningioma for whom supratentorial parasagittal sections of the brain were used. This has now been modified in the revised manuscript in the section ‘Sample collection and histopathological analysis’ under ‘Materials and Methods’.
Comment: Page 5 “PI3K and AKT genes are targeted by seven different miRNAs from the cluster” but are these seven miRNAs among the ones that are down-regulated in GBM? Figure 1 would be more useful if it included an indication of which of the C14 cluster miRNAs shown on the Fig are actually down-regulated in GBM.
Response: We have now modified Figure
as suggested by the reviewer. The miRNAs found to be downregulated in GBM have been colored in green. We have also modified relevant sections in the results and the figure legend. As seen in the modified figure, not all seven miRNAs are downregulated as reveled by the statistical analysis of the microarray data. The actual biological cross-talk will be clear after extensive functional studies for mRNA:miRNA interactions.
Comment: On Page 9 it is stated that cluster miRs were downregulated in several cancers at significantly higher proportion than expected by random chance. The statisitical test used and the P value should be given.
Response: In all the cancers tested we have found the C14 miRNAs to be downregulated significantly more than expected by random chance (Figure
). The p values were: p<10
for GBM, BRCA and KIRC and p<0.002 for OV. These were obtained by chi-squared tests. The details are in supplemental file 6. We have also added a new figure in the main manuscript with this data (Figure
Comment: Page 14. “Flanking sections measuring 5 mm were then taken and stained
by H&E for histological analysis.” Was this not 5 um rather than 5 mM?
Response: We thank the reviewer for pointing out this mistake. It has now been corrected in the revised manuscript.
Quality of written English: Acceptable
Reviewer’s response: My comments have been adequately addressed.
Reviewer’s response: The numbering of Figs in the published version will need to match the numbering in the text (currently it does not for Figures 4 onwards).
Author’s Response: We think the confusion about figure numbering arose from the fact that we have uploaded two separate files for two panels of Figure
. So, although the total number of figures are 6, but 7 files have been uploaded for figures. We have checked the text in the manuscript and there are no errors. We hope the two separate panels can be merged at the publication stage.
Quality of written English: Acceptable
Title: Genome-wide analysis reveals down-regulation of miR-379/miR-656 cluster in human cancers
Versions: 1 & 2 10 December 2012/ 5 March 2013
Reviewer number: 2
Reviewer: Dr Alexander Max Burroughs (nominated by Dr L Aravind)
Laddha and colleagues mine TCGA data to investigate the role of a specific miRNA-rich genome region in different cancer types. In addition to uncovering a likely role for this genome region in GBM, the authors supply independent experimental data supporting their claims.
Targeted computational analysis of large, publicly-available datasets can provide fruitful avenues of investigation for researchers as the large consortiums generating these datasets often lack sufficient manpower to thoroughly comb through the data. Along these lines, I find the manuscript of general interest particularly since this specific miRNA cluster has not been the subject of extensive research in the past. However, I have a few points for the authors to consider:
Comment: The examined genomic locus covers many miRNA, and several of these belong to the same miRNA family (e.g. miR-379, miR-380, miR-411, miR-758, and miR-1197 all belong to the same family according to miRBase). Is it possible the observed downregulation is largely driven by one or two families being instead of a locus as a whole? I specifically bring this up because of another paper implicating members of the above family in GBM, see Skalsky RL and Cullen BR in published in September 2011 in Plos One.
Response: We thank the reviewer for his concern. Although this may be a concern for other regions of the genome but for the C14 miRNA cluster this issue does not confound our data. We checked the reference suggested by the reviewer and found that they have talked about only one family within the cluster of miR-376 (miR-376a, -b or-c). The other miRs of this cluster are actually not part of the same family. Also upon multiple alignment of the pre-miRNA sequences of the miRNAs from C14 one finds very low similarity between them pointing to the above fact. Although theoretically it is still “possible that the observed downregulation is largely driven by one or two families being instead of a locus as a whole”; but given all the evidences provided by our study this is a remote possibility.
Comment: At several points in the paper the authors compare activity at the C14 locus to randomly-selected genome regions, but explanations of the criteria determining selection of these random regions are relatively scant. In each case are the authors filtering out other miRNA-rich genomic regions or instead selecting for miRNA-rich regions? Are they selecting genome regions with similar characteristics (e.g. similar ratio of coding or non-coding transcripts, similar ratio of repetitive regions, etc.)? If instead the authors are simply taking similarly-sized genomic sequences, I would think that 10 random selections would not be enough to amass a viable background set since localized attributes can substantially influence several of the characteristics being investigated. Since this selection underlies some of the more crucial findings of the manuscript, this could be considered more carefully or at least better-described in the text.
Response: We have used 10 random sets in two different scenarios in this study. Both are separately described below:
(i) To check that the observed downregulation of C14 miRNAs in cancers is not by chance, we have selected 10 random sets of miRNAs to analyze from the miRNA expression data. This datasets are not from contiguous stretches of genome but consists of similar number of miRNAs as C14 (38 miRNAs randomly selected from genome excluding C14). As the observed downregulation is much more than expected by random chance and we observed significant downregulation in multiple cancers, this finding is unlikely to be a false positive. In addition, as the total number of miRNAs for which expression data was available (for GBM) was only 534, more random sets will actually repeat the same miRNAs across multiple sets making the analyses redundant. The following text in the Results section has more specific information: “
To exclude the possibility of this being a chance finding, the analyses were repeated with ten random sets of miRNAs (38 in each set) for the same samples. These miRNAs were chosen excluding the C14 miRNAs. The number of downregulated miRNAs ranged between 3 to 9 out of 38 (95% CI, 4.56-7.03), which was significantly lower than the observation in C14 cluster (26 out of 38, p<10
(i) To check the possible altered methylation pattern of C14 in GBM, we have compared it with the methylation pattern of another imprinted large miRNA cluster on chromosome 19 (C19). The other 10 regions are randomly selected from the genome. The genomic sizes are not comparable, but the number of methylation probes is comparable (~200 probes per region) in a contiguous genomic stretch. The specific coordinates of each region have been given in the supplementary information. As the distribution of the probes on the microarray is not uniform throughout the genome, it is practically improbable to select enough number of regions with different distributions of mRNAs, miRNAs, repeats etc.). This section has now been modified in the revised manuscript.
Comment: The authors use quite a conservative method for identifying significant miRNA up/down regulation. Did the authors consider a different method, for example edgeR in Bioconductor, and how did these results compare?
Response: edgeR is a tool for analyzing digital gene expression data generated from NGS platforms, taking read counts and library size as inputs. The data we have primarily focused is the GBM dataset, which has microarray data for which edgeR is unsuitable. The other cancer types in TCGA, for which we have used small RNA sequencing data for expression analysis, do not have level 1 data which would have the information required by edgeR for analysis. The different expression data analysis tools primarily differ in their method of normalization of the raw data. As we have used the normalized expression data from TCGA, we have directly applied statistic to test differential expression.
Comment: As currently written, the manuscript is unclear on which cell types were analyzed when examining methylation patterns. Are these the same as the samples used for miRNA profiling? Do they match the same cancer type? Perhaps this is evident to those well-versed with the TGCA dataset but making this clearer would improve readability.
Response: We thank the reviewer for this constructive suggestion. The methylation experiment was performed on the DNA samples isolated from the tumor specimen of each patient. The data were from 76 GBM samples which were also included in the miRNA expression analysis. We have now added a sentence in the relevant section under ‘Materials and Methods’.
Comment: The authors state “the results compel us to say that majority [sic] of the miRNAs in C14 regulate the glioma pathway and a coordinated downregulation of the miRNAs can cause a major systemic perturbation.” I’m not sure the analysis presented by the authors determines causative relationship. It seems quite possible that the opposite could be true: some systemic perturbation is causing the downregulation of the miRNAs in this genomic location.
Response: The statement has been modified as follows: “The results indicate that majority of the miRNAs in C14 regulate the glioma pathway and a coordinated downregulation of the miRNAs might cause a major systemic perturbation. Whether the observed downregulation is actually a cause of the systemic perturbation or it is an effect of another global perturbation will be revealed by further studies.”
Comment: While the authors specifically address upregulation of C14 miRNAs in GBM, how much upregulation is observed in the other cancer lines?
Response: We actually observed a downregulation (not ‘upregulation’) of C14 miRNAs in GBM and other cancers and all the data presented in the manuscript are from patient tumor samples and not from cancer lines. The proportion of downregulated miRNAs varied from 12-30% of all downregulated miRNAs and all cases were statistically significant. A new figure has been added in the revised manuscript (Figure
) to explain the results and the detailed analysis is in Additional file
Comment: In the conclusions, the authors state “…the required transcription factor is significantly downregulated”. Are the authors certain MEF2 is the only transcription factor active in this location?
Response: Given the complex nature of human biology, it is almost certain that MEF2 is not the only molecule transcriptionally regulating this cluster. However, it has been reported in the literature that regulating Mef2 levels one can regulate the expression levels of the miRNAs in this cluster implicating Mef2 as the most important transcription factor for the cluster. Details of this findings can be found in Fiore R et al, EMBO J.2009; 28:697-710.
Quality of written English: Needs some language corrections before being published
Response: We have put in sincere efforts to make the language better and more accurate in the revised manuscript.
Reviewer’s response: The comment was not intended to suggest that all of the miRNAs from C14 belong to the same miRNA family; this is certainly not the case. I was instead making the observation that mir-379 family members (see miRNA gene family mir-379: http://www.mirbase.org/cgi-bin/mirna_summary.pl?fam=MIPF0000126) are present in C14 and appear to largely belong to the set of significantly down-regulated miRNAs; the supplied reference also found members of this family involved in GBM. I was wondering if the authors checked the list of down regulated miRNAs, within and outside of C14, for the presence of complete (or nearly complete) families of miRNAs. While I very much appreciate the author’s detailed responses to the other comments, particularly the extensive clarifications relating to methodology, I think this remains a point of interest: if specific miRNA families are making key contributions this could aid further studies into target identification, throwing more light on GBM.
Author’s Response: We thank the reviewer for clarifying his earlier concern and for providing us information about the family members. In light of his comment we now have included a statement in the re-revised manuscript in the relevant section. The statement is appended below.
“It is interesting to note that miR-379 from C14 has four other family members located in the same cluster (
). Majority of these members were found amongst the downregulated C14 miRNAs in GBM. It remains to be seen whether studying the miRNAs belonging to the same family (may or may not be in the same cluster) can give us more insight into their biology, especially when an additive effect of many miRNAs are investigated.”
Reviewer’s response: I apologize as my initial comment was not very clear: as the authors are surely aware, several different methods exist for addressing the multiple testing problem in both microarray and next-generation sequencing. Use of the Bonferroni correction is a starkly conservative choice; I was interested in why the authors would choose this corrective method over other, more frequently used methods like FDR. Does a more inclusive method increase the total number of genes from other regions in the genome, lessening the “signal” observed from the C14 region?
Author’s Response: As we had used data generated by a third party (TCGA), which is also normalized by them, we wanted to reduce the false positives by using the more stringent Bonferroni correction. However, we understand and appreciate reviewer’s concern and have performed FDR corrections in two of our datasets (namely, GBM & BRCA). We do not observe a significant alteration of the results. Specifically, upon application of FDR the data changes as detailed below:
In case of GBM (microarray platform), the fraction of C14 downregulated miRs vs. miRNome downregulated were 0.763 vs. 0.159 (by Bonferroni) and 0.815 vs. 0.204 (by FDR). In BRCA (NGS platform), the C14 downregulated miRs vs. miRNome downregulated was 0.585 vs. 0.101 (by Bonferroni) and 0.634 vs. 0.154 (by FDR).
For calculating FDR adjusted p values we have run [p.adjust(<file name>, method=”fdr”)] command in R on the p-values obtained from Mann-Whitney test and then taken adjusted p values <0.05.
Since, the change in the data was not altering the conclusion we did not change the dataset presented in the manuscript. This analysis was done to take care of the reviewer’s concern.
Quality of written English: Acceptable
Reviewer’s name: Dr. Yuriy Gusev (reviewer 3)
This reviewer provided no comments for publication.