Uncovering Microbial Composition in Human Breast Cancer Primary Tumour Tissue Using Transcriptomic RNA-seq

Round 1

Reviewer 1 Report

In this manuscript the authors studied microbiome of patient breast cancer tissues or normal breast tissues using RNA-seq data (rRNA depleted) instead of DNA sequencing of 16S rRNA or shot gun approach, then tried to link the microbiome changes with breast cancer clinical characteristics. The finding is very preliminary, and I have many concerns:

Major concerns:

1)Lacking novelty.

It seems that the authors want to report a methodology study. However, using human genome unmapped RNA data from RNA-seq to study breast cancer tumor microbiome patients has been reported (PMID 29190829). This study approach is not new. If not considered as a methodology study, the simple testing of the microbiome composition of breast tumor can not justify reporting since the breast tumor microbiome is already well known.

2)Missing biological confirmation.

The accuracy of the approach to investigating microbiome approach should be verified by standard microbial detection methods.

3) The meaning of the observed associations between microbes and clinical characteristics is questionable. Especially the associations were based on a very limited patient cohort without ruling out confounding factors.

Specific comments:

1)Of the unmapped RNA reads (4.7-7.3% of total), only a tiny fraction was assigned to microbial features (~0.03% of total). Where did most of the unmapped RNA reads come from? Are they background noise? Since the approach does not prioritize finding the bacterial RNA signal, the detection specificity and influence of noise is very concerning. Using positive and negative tissue samples (not only controls of data analysis) is critical to verify the specificity and sensitivity of the method.

2)Data of individual samples instead of means should be shown.

Author Response

In this manuscript the authors studied microbiome of patient breast cancer tissues or normal breast tissues using RNA-seq data (rRNA depleted) instead of DNA sequencing of 16S rRNA or shot gun approach, then tried to link the microbiome changes with breast cancer clinical characteristics. The finding is very preliminary, and I have many concerns:

Major concerns:

Point 1:

It seems that the authors want to report a methodology study. However, using human genome unmapped RNA data from RNA-seq to study breast cancer tumor microbiome patients has been reported (PMID 29190829). This study approach is not new. If not considered as a methodology study, the simple testing of the microbiome composition of breast tumor can not justify reporting since the breast tumor microbiome is already well known.

Response 1: We are aware of the fact, that similar methods have been used in the other study already and that study was cited in our manuscript. We see significance in some other traits of our study: we provided first such a study on patients from Slovakia or region of central Europe. In the light of the previous studies findings, that microbiome can differ depending on geographical location, we presume findings can contribute to the global picture of breast microbiome. Also, we provided comparison of cancer phenotypes microbiome, which haven’t been compared yet (although we are aware some of groups possess only few samples). Furthermore, there are only very few studies done on breast tumour microbiomes and we believe new validations are needed to in this area of research. Also, our study shows how data can be used for multiple purposes, since experimental approach was originally designed for differentially expressed genes analysis using standard RNA-seq.

Point 2:

Missing biological confirmation.

The accuracy of the approach to investigating microbiome approach should be verified by standard microbial detection methods.

Response 2: On this point, we are not able to provide improvement in a given time. Our samples exist in the form of isolated RNAs (rRNA-depleted). We won’t be able to use standard approach of microbial identification. However, performance of methods we used were validated by various other studies in the past (for example PMID: 32558637 34201359), and also in the study PMID 29190829 mentioned by reviewer. We believe, method was already validated enough by comparison to another more traditional approaches or by computational methods. Our study was meant as additional bioinformatics investigation of RNA-seq data. It was supposed to show possibility of using data for multiple analysis (not just for differentially expressed gene analysis).

Point 3:

 The meaning of the observed associations between microbes and clinical characteristics is questionable. Especially the associations were based on a very limited patient cohort without ruling out confounding factors.

Response 3: In some cases, there were limited number of patients, this must be noted as a limitation and we provided that information. However, experimenting with the bigger dataset (Chinese samples from SRA) showed, that raising number of patients dramatically, did not improve results much (also mentioned in manuscript). It is not possible to provide higher number of samples for clinical characteristics since they are not available at the time. We believe, results from fewer samples can be useful too, when limitations are clearly stated.

Specific comments:

Point4:

Of the unmapped RNA reads (4.7-7.3% of total), only a tiny fraction was assigned to microbial features (~0.03% of total). Where did most of the unmapped RNA reads come from? Are they background noise? Since the approach does not prioritize finding the bacterial RNA signal, the detection specificity and influence of noise is very concerning. Using positive and negative tissue samples (not only controls of data analysis) is critical to verify the specificity and sensitivity of the method.

Response 4: These numbers are relative to the parameters used for mapping and classification of the reads. Vast majority of them are human reads that did not have enough quality to map on human genome or did not map for some other reason, but more sensitive Kraken2 algorithm has identified them as human. The number of these reads depend also on sequencing, since in Slovak datasets there were more of them than in Chinese dataset. Minority of this data (labelled as “other”) could be microbial reads that were not assigned to its taxon by software algorithm or microbial transcripts of organisms not present in a database (this is probably really small number if any), although it also can be some noise that somehow passed through quality filtering. Only 0.07-0.5% of filtered sequences were classified as “other”.

Identified human reads were discarded from further statistical evaluation by mapping to human genome. Dropping human-based reads (as many as possible) was done to lower the risk of false positive and improve speed of microbial classification, however, used software (Kraken2) should be able to identify bacterial transcripts correctly in the mix of other species transcripts.

It is worth mentioning, the overall quality of our sequencing was sufficiently high, 93-95% of all RNA-seq reads mapped to the human genome.

Point 5:

Data of individual samples instead of means should be shown.

Response 5: We tried to improve this, although it was not clear to us which “means” are mentioned here. We provided numerical values of classified reads for every sample in new supplementary Table S1. If the reviewer was mentioning any other data, these could be still added to our supplementary website, if required

Reviewer 2 Report

General comment
This manuscript focuses on the meaning of microbiome/microbiota in human body other than well known enteric one.
They aim to obtain further understanding and significant developments in therapies on the disease by targeting the relationship between breast cancer and microbiome/microbiota, through the comparison of tendencies between different ethnic groups.

Each point
The Introduction section was too long, because it includes the contents which should be included in the Discussion section.
Furthermore, there were some duplications in the Discussion section by which the section was made lengthy, thus the manuscript can be more concise by careful effort for the description.

As the authors say, there are not so much strong novelty on the theme, the relationship between breast cancer and the breast microbiome/microbiota, and no strong tendency/correlation was found as the result.
However, since this kind of research has a strong difficulty including ethical and/or number of samples and so on, it is almost impossible to avoid bearing a propensity of preliminary trial.
In this situation, the deposition of this kind of studies might be a help for finding out significant correlations and/or principles for next innovation, thus, this work has a substantial value as a component of a big data to be constructed.

There is other difficulty in definition of the "healthy" control, cause there are few disease-free old peoples who should be suitable counter part for patients bearing cancers. 
In the aspect of microbiome/microbiota, "non-cancerous young" versus "non-cancerous" old should be investigated for basic microbiome/microbiota study.
If the authors already have this kind of data, discussion about this point by presenting this data on to current data set could be an one way for the improvement of this manuscript, although the theme should be much more carefully discussed as a different research subject. 
This point of view might be coming up to correlate with biological significance of microbiome/microbiota in various tissues even being separated from the correlation with diseases, and the reviewer personally is being attracted by this topic in basic science. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

My previous major concerns were not sufficiently addressed.

For a methodology study, it lacks (1) novelty (2)biological prove of accuracy. The authors cited three publications to argue the validity of the approach using RNA-seq to identify microbiome, however, this further decreased the novelty as a methodology study. The whole study is just applying an established method to study breast cancer microbiome (which even has been studied using the similar approach).

The results are very primary and purely descriptive for the breast cancer microbiome study part. For large data set analysis, there are always some statistically significant correlations be found. Without biological confirmation, and further support from different cohorts, how meaningful the findings generated from a few samples is in question. The tissue microbiome is expected to have high varieties among individuals even heterogenous in different regions of the same tumor. It is difficult to believe the microbiome findings from a limited clinical samples can reflect the complected tumor biology especially the role of microbiome in breast cancer is uncertain.  

To clarify my previous question about show individual samples instead of means, the authors should show the diversity using individual data instead of bar graphs based on average or mean value.       

Author Response

Please see the attachment.

Author Response File: Author Response.docx