This article is based on a subset of the same rats (females only) used in the article by Seralini et al. which was published in Food and Chemical Toxicology in 2012, was subject to considerable criticism such that it was retracted by that journal and was subsequently republished in Environmental Sciences Europe in 2014. Many of the criticisms were statistical in nature which is the issue that concerns me here. I am not convinced that Mesnage et al. have avoided those statistical errors.
An important aspect of the design of the experiment reported by Seralini et al. was that there were three experimental groups of 10 female rats that were given different doses of Roundup in drinking water. A surprising aspect of the results in the 2012 paper was that there was no evidence of a dose-response relationship; if anything, there were fewer reported pathologies in female rats subject to the stronger doses of Roundup (Table 2). Consequently, it is important to know whether Mesnage et al. analysed all three of the relevant experimental groups (plus controls) to test for a dose-response association and if not, why not.
It is also unclear whether the authors have addressed the criticism about the unit of analysis - which should be the rat and not the gene. There is likely to be clustering of gene disturbances in rats and ignoring this gives comparisons which do not properly allow for chance. Their Fig. 2(b) suggests that this has not been recognised.
It is important to note that the results in Fig. 1 are for liver only and are based on only a small part of the total variability. We are not told how many variables contributed to the Principal Components Analyses for liver and kidney (the videos in the additional material are singularly unenlightening). More generally, it is not at all easy to work out what the relevant statistics are for the treatment and control groups across all the analyses.
Competing interests
None
Response to comment posted by Prof Ian Plewis
Michael N Antoniou, King's College London
24 September 2015
We thank Prof Plewis for his interest and comments on our article. Our response to the issues he raises is as follows.
The controversy surrounding the retraction of the Seralini et al. 2012 study from Food and Chemical Toxicology and its republication in Environmental Sciences Europe re-establishing its rightful place within the scientific literature, has been dealt with extensively in numerous articles, which the interested reader can consult. There is therefore no need to re-visit this topic here.
The reasons as to why the particular subset of animals was selected for our investigation are clearly described in the appropriate sections (Introduction, Results) of our paper. In terms of how the transcriptomic data we obtained was analysed starting with the principle component analysis, is again described in detail in the relevant sections (Materials and Methods, Results) of our paper. Thus there is no need to reiterate material pertaining to matters of protocol here and suggest that those interested carefully consult the appropriate sections of our paper.
As to the statistical analysis of the data, we employed methods that are standard in the field dealing with large scale microarray datasets and thus perfectly valid. However, if Prof Plewis wishes to use alternative methods, in accordance with internationally accepted procedure the raw data obtained from the microarray transcriptome analysis we obtained has been made publicly available (accession number: GSE66060) allowing independent reanalysis of our results by others.
Finally, Prof Plewis may wish to take advantage of the open peer review policy of Environmental Health to see pre-publication versions of our article and our comments to reviewers. Here can be seen that issues were raised by the reviewers regarding analysis of our data and we responded accordingly to their satisfaction.
Environmental Health
Design and analysis
24 September 2015
This article is based on a subset of the same rats (females only) used in the article by Seralini et al. which was published in Food and Chemical Toxicology in 2012, was subject to considerable criticism such that it was retracted by that journal and was subsequently republished in Environmental Sciences Europe in 2014. Many of the criticisms were statistical in nature which is the issue that concerns me here. I am not convinced that Mesnage et al. have avoided those statistical errors.
An important aspect of the design of the experiment reported by Seralini et al. was that there were three experimental groups of 10 female rats that were given different doses of Roundup in drinking water. A surprising aspect of the results in the 2012 paper was that there was no evidence of a dose-response relationship; if anything, there were fewer reported pathologies in female rats subject to the stronger doses of Roundup (Table 2). Consequently, it is important to know whether Mesnage et al. analysed all three of the relevant experimental groups (plus controls) to test for a dose-response association and if not, why not.
It is also unclear whether the authors have addressed the criticism about the unit of analysis - which should be the rat and not the gene. There is likely to be clustering of gene disturbances in rats and ignoring this gives comparisons which do not properly allow for chance. Their Fig. 2(b) suggests that this has not been recognised.
It is important to note that the results in Fig. 1 are for liver only and are based on only a small part of the total variability. We are not told how many variables contributed to the Principal Components Analyses for liver and kidney (the videos in the additional material are singularly unenlightening). More generally, it is not at all easy to work out what the relevant statistics are for the treatment and control groups across all the analyses.
Competing interests
NoneResponse to comment posted by Prof Ian Plewis
24 September 2015
We thank Prof Plewis for his interest and comments on our article. Our response to the issues he raises is as follows.
The controversy surrounding the retraction of the Seralini et al. 2012 study from Food and Chemical Toxicology and its republication in Environmental Sciences Europe re-establishing its rightful place within the scientific literature, has been dealt with extensively in numerous articles, which the interested reader can consult. There is therefore no need to re-visit this topic here.
The reasons as to why the particular subset of animals was selected for our investigation are clearly described in the appropriate sections (Introduction, Results) of our paper. In terms of how the transcriptomic data we obtained was analysed starting with the principle component analysis, is again described in detail in the relevant sections (Materials and Methods, Results) of our paper. Thus there is no need to reiterate material pertaining to matters of protocol here and suggest that those interested carefully consult the appropriate sections of our paper.
As to the statistical analysis of the data, we employed methods that are standard in the field dealing with large scale microarray datasets and thus perfectly valid. However, if Prof Plewis wishes to use alternative methods, in accordance with internationally accepted procedure the raw data obtained from the microarray transcriptome analysis we obtained has been made publicly available (accession number: GSE66060) allowing independent reanalysis of our results by others.
Finally, Prof Plewis may wish to take advantage of the open peer review policy of Environmental Health to see pre-publication versions of our article and our comments to reviewers. Here can be seen that issues were raised by the reviewers regarding analysis of our data and we responded accordingly to their satisfaction.Competing interests
We declare that we have no competing interests.