GMO’s Cause Cancer, Part I: The Curious Case of Séralini’s Rats

Heads up! Part II has been published. Check it out after this post at the link below. 

GMO’s Cause Cancer, Part II: Much Ado About Monsanto

Gilles-Éric Séralini had some rats. He fed them some GMO corn and some Roundup herbicide and they got cancer and died. Therefore, GMO’s cause cancer.

Just kidding. But this is a conclusion I've seen parroted around the internet as of late. However, I think you'll find it’s a bit more complicated than that.

What follows is an attempt to thoroughly dissect the Séralini paper and subsequent republication. This will be a gloriously meticulous exercise in demonstrating how bad science can muddy the public’s scientific understanding.


Séralini is a professor of molecular biology at the University of Caen, France, and president of the scientific board of CRIIGEN (Committee of Independent Research and Information on Genetic Engineering). He published a study in 2012 in the journal of Food and Chemical Toxicology (FCT) that was supposed to be a replication of a 2004 study conducted by Monsanto looking into the safety of Monsanto’s Roundup Ready resistant corn (NK603 R-tolerant maize) [1,2]. Séralini’s study has since been retracted by FCT and republished in another journal but before we dive into the meat of the paper let’s take a brief look at Séralini’s history and potential conflicts of interest.

In 2006, Monsanto published a study looking at the safety of YieldGard Rootworm Corn (MON 863) which designed to protect against feeding damage caused by corn rootworm larvae [3]. The findings showed that MON 863 was as “safe and nutritious as existing conventional corn varieties” [3]. After being compelled by appellate court action in Germany, Monsanto released the raw data of its study, which was reanalyzed by Séralini [4]. This reanalysis showed that MON 863 allegedly caused kidney and liver damage in rats. This reanalysis was roundly criticized for its poor statistical methods in a 2007 paper published by a panel of experts. This paper found the following [5]:

In each case, statistical findings reported by both Monsanto and Séralini et al. were considered to be unrelated to treatment or of no biological or clinical importance because they failed to demonstrate a dose–response relationship, reproducibility over time, association with other relevant changes (e.g., histopathology), occurrence in both sexes, difference outside the normal range of variation, or biological plausibility with respect to cause-and-effect. The Séralini et al. reanalysis does not advance any new scientific data to indicate that MON 863 caused adverse effects in the 90-day rat study.

In 2009, Séralini attempted the same tactic and published a paper that reanalyzed the toxicity data for three Monsanto strains of genetically modified (GM) maize: NK603, MON 810, and MON 863. NK603 is tolerant of the herbicide Roundup and MON810 and MON863 synthesize Bacillus thuringiensis (Bt) toxins which are insecticides [6]. Séralini's re-analysis also found that these strains caused damage to the liver, kidney, heart, adrenal glands, spleen, and haematopoietic system. Once again, these findings were disputed by the European Food Safety Authority (EFSA) and Food Standards Australia New Zealand [7,8].

Conflicts Of Interest

Since then, Séralini has moved on from re-analyzing papers and decided to conduct a study himself. In 2012, he published the study “Long term toxicity of a Roundup herbicide and a Roundup-tolerant genetically modified maize” [1]. Curiously, he did not list any conflicts of interest even though there were plenty he could have mentioned. For one, the Foundation for Human Progress (FHP) gave €1 million euros to help fund this €3.2 million euro study. FHP is very active in funding anti-GMO groups [9]. He also published a book and documentary right after the study was released, both entitled ‘Tous cobayes?’, which translates into ‘We Are All Guinea Pigs!’ [10,11]. He also has a few other books, such as ‘Genetically Incorrect’ and ‘We Can Clean Up’, both of which talk about how genetic manipulation is dangerous [21].

Séralini is also a consultant with a small company called Sevene Pharma, a homeopathic pharma company. He conducted research on one of their medications and was able to ‘prove’ that it can protect against pollutants, including Roundup [12,13]. These trials were conducted in vitro on human liver cells. No trials were ever conducted in humans but that did not stop Sevene Pharma from selling their Digeodren line of pills, claiming they can protect “against cell death caused by the Roundup” based off of the research Séralini provided [14]. 

The idea that this homeopathic remedy, let alone any homeopathic remedy could work is absurd. Homeopathic pills take active ingredients and dilute them to the point where there is no trace of the original substance left. Hundreds of randomized clinical trials have found over and over that they do not work [40,41,42,43,44,45,46].

Oh, and remember CRIIGEN? The organization where Séralini is the president of the scientific board? It’s an anti-GMO lobbying group that helped fund his study [22].

The Embargo System

Then there is the problem of Séralini abusing the embargo system. Usually, one week before a paper is released, journalists get a sneak peek at the paper. During this week they can fact check the paper by talking it over with their peers and seeking out expert opinion on the research. This is an important and useful process to help control media hysteria that often occurs after a paper is released. Journalist’s conduct these activities under the agreement that they will not publish anything until after the journal says they can; typically after the paper is published.

This was not the case for the Séralini paper. Journalists that wanted access early access to the paper had to sign a non-disclosure agreement stating that they were prohibited from sharing the results with any outside experts before the embargo was lifted [15]. So any journalist that saw the paper early could not seek out expert opinion on the study. Journalists were left to their own devices to fact check the paper. The result was that the first round of news on this paper was largely unskeptical and did not address any scientific concerns [17]. Anyone who broke the non-disclosure agreement was threatened with a harsh fine: “A refund of the cost of the study of several million euros would be considered damages if the premature disclosure questioned the release of the study” [11].

When the abuse of the embargo system was discovered, the “French National Centre for Scientific Research (CNRS)…decried the public-relations offensive as inappropriate for a high-quality and objective scientific debate, and reminded researchers working on controversial topics of the need to report results responsibly to the public” [11].

All that and we haven't even reached the paper yet!

The Study

Let’s do a rundown of the paper. In the introduction, the following claims are made [1]:

  • There is international debate as to the necessary length of mammalian toxicity studies in relation to the consumption of genetically modified (GM) plants including regular metabolic analyses
  • Significant disturbances have been found [in regard to GMO subchronic toxicity] and may be interpreted differently
  • Detailed analyses have revealed alterations in kidney and liver functions that may be the signs of early chronic diet intoxication, possibly explained at least in part by pesticide residues in the GM feed

For each of these claims, the only citations he provides are back to his own published papers. He does not cite any other sources or include the fact that his previously published papers were heavily criticized for methodological and statistical issues. But I digress.

The rats were either fed the Roundup-tolerant maize (GMO), maize treated with Roundup (GMO+R), water mixed with Roundup (R), or a control non-GMO maize (non-GMO). The maize was then mixed in with the standard rat chow at three different levels. The result was 10 different groups, 10 males and 10 females in each, which were broken down as follows:

  1. Controls (20) – Non-GMO corn
  2. 11% GMO (20) – 11% GMO corn + rat chow
  3. 22% GMO (20) – 22% GMO corn + rat chow
  4. 33% GMO (20) – 33% GMO corn + rat chow
  5. 11% GMO + R (20) – 11% GMO corn treated with Roundup + rat chow
  6. 22% GMO + R (20) – 22% GMO corn treated with Roundup + rat chow
  7. 33% GMO + R (20) – 33% GMO corn treated with Roundup + rat chow
  8. R(A) (20) – 0.00000005g/L or 0.000000011% Roundup in water
  9. R(B) (20) – 0.4g/L or 0.09% Roundup in water
  10. R(C) (20) – 2.25g/L or 0.5% Roundup in water

Blood samples were taken at 1, 2, 3, 6, 9, 12, 15, 18, 21 and 24 months: 11 measurements were obtained for each animal alive at 2-years. Over 56 parameters were measured from various blood, urine, and liver samples. 34 tissue and organ samples were taken after the rats died for further analysis. The following methods were used for statistical analysis: Principal component analysis (PCA), partial least-squares to latent structures (PLS), and orthogonal PLS (OPLS) [1].

The next part is very important and is an excellent example of why it is crucial that researchers release their raw data. In the results section, Séralini states that not “all data [can] be shown in one report, and [only] the most relevant are described here” [1]. Now this is certainly true, that is can be difficult to show all of your finding in a paper, but it also leaves a window open for cherry picking data that is in line with your beliefs. This is why it’s imperative that raw data be released, so methods can and findings can be independently verified.  

The paper goes on to detail many of the findings. You can read them all in the original paper, which is free online, but the gist of the findings show that the control rats fared the best and the intervention groups had a greater incidence of mortality, tumors, liver and kidney problems to name a few.

Methodological Issues

Below I've broken down the relevant areas and discussed the issues in each.


What exactly was he trying to measure? He was testing three different hypothesis in his trial:

  1. Gm corn
  2. Gm corn + Roundup
  3. Roundup

Reference data was not provided, such as the levels of Roundup normally found in maize (GM or non-GM), the levels of Roundup metabolites found in products made from the maize, “the stability of Roundup after food processing, and the circulating or tissue levels of Roundup or any surrogate metabolites…[this data] is critical for any study based on adsorption, distribution, metabolism and excretion (ADME)” [18].


The gap between the high doses (2.25g/L or 0.5% Roundup in water) and the low dose (0.00000005g/L or 0.000000011% Roundup in water) is too large to be able to determine a dose-response relationship [30,31].

Roundup is highly unlikely to be found in the drinking water supply. In France, they conducted 43,741 tests that screened for glyphosate in 21,864 stations. Only 95 tests (0.2%) detected glyphosate and at levels below 5ng/L (0.000000005 g/L). The French study noted that it only tested for glyphosate and not the other chemicals in Roundup because the “co-formulants [are] not mobile in the soil”. The French findings concluded that “The likelihood of finding the tested quantities in groundwater appears negligible. Therefore, the route of administration described in the study protocol (oral exposure) is not the most appropriate for assessing the risks related to the product's application” [30,31]. 

These findings show that all three doses used by Séralini are unlikely to be found in drinking water and have no real world applicability [30,31].

To further compound things, water consumption was recorded but not reported. We have no way of knowing how much of the water these rats actually ingested over the study.


Food consumption was recorded, but not reported.

Non-GMO corn was never verified that it was non-GMO through lab tests.

No analysis was done to see if any of the non-Roundup treated food was cross contaminated.

Rat Strain and Number of Rats

Here is where one of the biggest methodological issues of this study manifests. Séralini had 200 rats in total. 20 were included in the control group and 180 were in the various intervention groups. That gives you an intervention to control ratio of 1:9. That is far too few control rats to leverage any statistical power. There was no power analysis reported to justify the number of mice chosen in the study.

The other big issue was the duration the study ran with the Sprague Dawley (SD) strain of mice; a strain that is susceptible to cancer. SD rats are typically used in short-term experiments of 90-days to ascertain tumorigenicity or toxicity. If tumors form before 90-days, whatever is being tested is considered to be tumorigenic.

The Séralini study ran for two years. This is an issue because SD rats have been shown to spontaneously grow tumors when left to live out their lives in undisturbed conditions [33-37]. This makes the SD strain inappropriate for long-term studies of this nature, as it is difficult to determine if the tumors were caused by the intervention or by chance. Cancers also affect many types of metabolism in the body, which will add noise to the toxicity test data collected. This is why SD rats are only used for 90-day toxicity studies.

The Results of two, 2-year long studies where the rats lived undisturbed

The high tumor incidence of these rats is further compounded by the low number of rats chosen for a study of 2 years. 20 per group is only justified in a short 90-day test as it provides enough statistical test power to detect differences between groups. Because the study lasted for two years, the sample size was too small [30,31].

The Organization for Economic Co-operation and Development (OECD) recommends “20 rats per group for a 12-month chronic toxicity study (Test Guideline 452 (OECD, 2009b)) and 50 rats per group for a 24-month carcinogenicity study (Test Guideline 451 (OECD, 2009a)) or a combined chronic toxicity/carcinogenicity study (Test Guideline 453 (OECD, 2009c)). With 10 rats per group, this study falls short of the recommended number considered necessary to infer statistically significant effects of long-term treatment for the two types of analysis undertaken (chronic toxicity, carcinogenesis)” [11,30,31].

Furthermore, SD rats show “high mortality rates and high incidence rates for mammary tumors in control groups, which were the main abnormalities observed by Séralini et al. (2012). These phenotypic characteristics should have been taken into account when calculating the required number of animals” [30,31].

An article in Nature notes that “data provided…by Harlan Laboratories, which supplied the rats in the study, show that only one-third of males, and less than one-half of females, live to 104 weeks. By comparison, its Han Wistar rats have greater than 70% survival at 104 weeks, and fewer tumors” [11].

Survival Analysis – Lifespan

The following are observations made in the EFSA’s Final review of the Séralini et al. publication [30,31].

"Séralini explains that “Control male animals survived on average 624 ± 21 days, whilst females lived for 701 ± 20”. These values are the result of an incorrect calculation because the data is censored (we do not know when the animals still alive at the end of the study would have died naturally since they were euthanized). Instead, what has been calculated is the empirical mean and standard deviation of the uncensored observed values joined with the censored values for still-alive rats (as if a rat still alive at T=720 days is considered dead at T=720 days). The results given are therefore inexact because this procedure introduces a bias by underestimating the average date of death and clearly also the standard error of the estimator. To have chosen not to consider everything that occurs after 624-21=603 days is therefore not justified because this value comes from an incorrect calculation" [30,31].

"Correct calculation of the survival distribution for different groups requires the introduction of a parametric model, but the use of such an approach is constrained, given the limited amount of data per group. For example, if we fit a Gaussian model for the survival time of the males, the 24 estimated mean and standard deviation are respectively 626 and 68 days. For the females, it is 892 and 206 days. It is not correct to proceed as the authors have done and calculate the standard error for the mean merely by dividing the standard deviation by √10, as would be done for uncensored Gaussian variables. Due to censoring, the distribution of the estimator of the mean is much more spread out and asymmetric, meaning that the use of the standard error for calculating confidence intervals is not meaningful" [30,31].

No confidence intervals or p values are reported for mortality data and it appears the mortality data was not statistically analyzed. When analyzed, “there is no evidence that mortality to the end of the experiment was significantly different in the groups of rats fed either the GMO, GMO + R, or when R was administered in the drinking water at three different levels. The claim by the authors that ‘‘In females, all treated groups died 2–3 times more than controls, and more rapidly’’ is not true” [20].

Table A1 shows the mortality to the end of the experiment in each group [20].

Results of Log-Rank tests on reduced life expectancies with and without correction for multiple testing [30,31].

Naturally occurring mortality rates were also not taken into consideration. Below is a table of studies showing natural mortality rates.


The study was neither single nor double-blinded.

Mechanisms and Other Issues

Effects of feeding Roundup Ready maize to the SD rats and the effects of feeding them Roundup were identical; a highly suspicious result for two completely separate substances. No plausible mechanism for this was provided.

Growth data was not provided.

Organ weight was not provided.

No dose-response curve was observed. You would expect to see things like mortality increase as the rats ate higher concentrations of roundup or GM corn but no such curve happened.

Statistical Issues

No statistical tests on treatment differences for mortality and pathology incidence [18].

No estimation of dose/sex/ROUNDUP GMO effects or calculation of confidence intervals for these effects [18].

No adjustment for survival [18].

No analysis of cumulative tumor risks relative to survival duration [18].

No analysis of time to tumor formation [18].

No discussion or presentation of test facility historical control tumor incidence data [18].

Kaplan–Meier statistics (or similar methodologies) are absolutely required to analyze tumor prevalence and survival across time and populations. The use of mean centering and unit-variance could also artificially reduce the level of background variability, therefore, causing the significance of the observed variation to be exaggerated [18].

The results on mortality and tumor incidence are presented descriptively and are not statistically analyzed, no p-value was given [30,31].

54 comparisons were made in the study but only five were significant (P < 0.05) before False Data Rate (FDR) correction. The following were significant before the correction:

  • Hepatic pathologies‟ described by the author as liver congestions, macroscopic spots and microscopic necrotic foci [30,31]
    • for the males in the group fed 22% GMO,
    • for the males in the RB group.
    • Mammary tumors
      • for the females in the RB group.
      • Pathological signs in the mammary glands (other than tumors described by the authors as galactoceles and mammary hyperplasias) [30,31]
        • for the females in the RA group,
        • for the females in the RB group.

After FDR correction for multiple testing, there are no significant differences at the 5% level [30,31].

OPLS-DA analysis was used to discriminate between the control group and experimental groups. The application of this method is a bit peculiar. As noted in the EFSA report [30,31]:

  • Whatever the method used, it is important to validate the model obtained (i.e. ensure that it possesses good predictive properties) by:
    • an independent test set, which helps to ensure that the model fitted to the training set retains good predictive properties for new data not previously used to fit the model;
    • cross-validation methods, where different subsets of the data are used alternately as training set and test set.
    • The study’s authors have not validated the models obtained, which cannot, therefore, be used for predictive purposes.
    • Use of this method assumes a symmetric distribution of predictor variables. Biochemical parameters may have an asymmetric distribution; pre-transformation is therefore necessary. There is nothing to suggest that this was done.
    • Calculating confidence intervals for each parameter is not relevant when many parameters are used, since potential correlations between parameters are totally ignored.

For a more detailed statistical analysis with programming notes, please read these two blog entries: 

Why I Think The Séralini GM Feeding Trial Is Bogus

Séralini Rat Study Revisited

Conclusions on the Study Results

During the EFSA hearing, the study’s authors “admitted that this study was not conclusive by itself and that, though subject to improvement, it had the merit of opening up an interesting line of research…The team’s members firmly believe that, having used all techniques available, what they observed was not random. The study could certainly be improved but the team simply opened up a path and we must now collectively do better…These experiments need to be repeated since this was the first time that tests were undertaken with a pesticide as a whole at a low dose (Extracted verbatim from the report of the hearing with the study’s authors)” [30,31].

Other Long Term Studies

Séralini claimed that long-term experiments had never been conducted outside of his study [18]. This is not true. Two such studies (Malatesta et al. 2008; Sakamoto et al. 2008) have been conducted using a more appropriate rat strain (F344 for Sakamoto) and a larger number of rats. Neither found evidence of GMO related effects. However, these two studies cannot be fully compared to Séralini’s study. Both were conducted with glyphosate-tolerant soybeans but this tolerance was obtained through the synthesis of a CP4 EPSPS protein, just like the NK603 maize Séralini and Monsanto tested [30,31]. It should be noted that in  Malatesta et al. the study was conducted on a limited number of rats and only on female mice [30,31].

Séralini is correct that there should be more publications looking at the long-term toxicological effect of herbicides/pesticides and GMO’s. The EFSA is currently updating their guidelines to ensure that this can happen [30,31,32].

Raw Data Release/Selection of Data Reported

Here are some issues that could be cleared up if Séralini would release his raw data to the public, as noted in the EFSA report:

Data required to interpret the toxicity study, such as data on composition and contaminants in diets, dietary intakes and weight gain of animals, are not reported in the paper. Consumption data and the energy balance of the various diets are all the more important as tumor incidence can vary according to dietary intake (Keenan et al., 1997) [30,31].
· · ·
Without any justification, the authors have chosen to report the results of the four biochemical parameters and two hormonal parameters that they consider to exhibit 9 the greatest variation from the control group (Fig. 5B). This choice was made after the results were obtained. It is obviously to be expected that there will be differences between the 864 (18 experimental groups x 48 biochemical parameters = 864) comparisons made by the authors for the 48 biochemical parameters in the 15th month of the study [30,31].

When his paper was republished, he bizarrely opted to only release part of his data.

His team took blood samples at months 1, 2, 3, 6, 9, 12, 15, 18, 21, and 24 but only released the data for month 15.

They released the tumor and mortality data for each group of rats, but not for the individual rats — which makes it impossible to test for in-group variation [16]

When asked by the EFSA to release his raw data, Séralini said he would not release his data until “the EFSA makes public all the data underpinning its 2003 approval of NK603 maize for human consumption and animal feed” [11]. This move is somewhat ironic considering he took Monsanto to court to have their raw data released and won. His argument basically boils down to “I will if they will”, as if his ability to practice good science is somehow dependent on the actions of others.

Other Concerns


I do not know if the graphs he used in his paper are standard, but they certainly were confusing as hell

Males that drank the most Roundup had decreased mortality compared to those that drank untainted water or water with less Roundup in it.

Male Rats That Drank Roundup. Y-axis is mortality (# rats/group)

0, A, B, C = dotted, thin, medium, and bold lines, respectively. Lifespan during the experiment for the control group is represented by the vertical bar ± SEM (grey area). In bar histograms, the causes of mortality before the grey area are detailed in comparison to the controls (0). In black are represented the necessary euthanasia because of suffering in accordance with ethical rules (tumors over 25% body weight, more than 25% weight loss, hemorrhagic bleeding, etc.); and in hatched areas, spontaneous mortality.

Males that ate the highest concentration of GM corn died less.

0, 11, 22, 33 = dotted, thin, medium, and bold lines, respectively

Is any of this statistically significant? Probably not, but note that Séralini drew conclusions from this untested data the same way I just did.

Pictures of Rats

It was very odd that he included pictures of the tumor-ridden rats in his paper. The pictures were sensationalist and did not add any scientific understanding. He only included pictures of the intervention rats and not the control rats, even though both groups had tumors.

The most curious part of including these pictures is that at the very beginning of his paper he stated this: “All data cannot be shown in one report, and the most relevant are described here” [1]. 

If space was so limited why not leave out the pictures? By excluding them, he gains ½ a page more to discuss findings.

European Hearing

During the EFSA hearing, Séralini acknowledged that the study design was not suitable to assess long-term carcinogenicity (cancer incidence) and his paper was only designed to measure long-term toxicity [30,31]. I get that the words tumor and cancer are not synonymous, but the paper repeatedly referred to tumors and spent a lot of time discussing them. A report in Nature notes that Séralini, “…has promoted the cancer results as the study’s major finding, through a tightly orchestrated media offensive that began last month and included the release of a book and a film about the work” [11]. If this was truly a toxicity study it makes the pictures of the rats with tumors all the more irrelevant.

From the EFSA hearing:

Séralini et al. (2012b) do not address any of the open issues for the statistical methods as raised in EFSA’s first Statement (EFSA 2012). They state that statistical methods for the analysis of tumors endpoints cannot allow to conclude on a mortality linked or not to the treatment groups. EFSA notes that this is inconsistent with the conclusions with respect to the tumors and mortality as drawn by Séralini et al. (2012a). [30,31]
· · ·
Séralini et al. (2012b) mentioned that a scientific publication is limited with respect to space and can therefore only show the data necessary to understand and discuss the conclusions, and refer to future publications that will provide more data. It is unclear how the authors have selected the endpoints for reporting and why, for reported endpoints, the complete analysis was not provided (e.g. biochemical data were reported only for selected treatment groups, and only at one time point”) [30,31]. At this time no further papers discussing the data have been released by Séralini.

Answers To Critics

Before and after his paper was retracted, Séralini published letters defending his paper. Below is the summary arguments of those two papers and the counter-arguments I provide. I have extracted the arguments where he tries to defend his paper. Other arguments were made that do not pertain to the papers quality and are not included below but can be read in full at these two citations > [26,27]. I have also included arguments from the website http://www.gmoSé [29]. Many arguments we have already covered earlier in the paper but I will summarize them here.

Argument 1 – Selection Of Rats

Sprague Dawley rats are used routinely in such studies for toxicological and tumor-inducing effects, including those 90-day studies by Monsanto as a basis for the approval of NK603 maize and other GM crops.

Other studies have used Sprague Dawley rats in 36-month studies by (Voss et al., 2005) or in 24-month studies by (Hack et al., 1995), (Minardi et al., 2002), (Klimisch et al., 1997), (Gamez et al., 2007)

Counter-Argument 1

Yes, these studies use the Sprague Dawley line of rats but for the appropriate 90-day interval as set out by The Organization for Economic Co-operation and Development (OECD) guidelines [28]. The 90-day interval is advised because this strain of rats are susceptible to spontaneously generating tumors under normal conditions [33-37].

Just because other studies used these rats for longer periods than advised does not make the Séralini study good, it makes all of these studies equally bad.

Argument 2 – Number Of Rats

OECD guidelines (408 for 90-day study, 452 chronic toxicity and 453 combined carcinogenicity/chronic toxicity study) always asked for 20 animals per group although the measurement of biochemical parameters can be performed on 10 rats. Monsanto itself measured only 10 rats of the same strain per group on 20 to conclude that the same GM maize was safe after 3 months (Hammond et al., 2004).

Counter-Argument 2

Again, the OECD guidelines are meant for rat studies lasting 90-days, not two years. The OECD guideline state that in a carcinogenesis study there should be 50 rats per group [28]. The authors claim that they “did not perform a carcinogenesis study…but a long-term chronic full study”. Even if that is true they would still need a larger number of rats to compensate for potential rat die off during the study. They would also receive larger statistical power from the larger sample. The second issue is that they did not exactly follow the OECD guidelines anyway, as they had 20 control rats and 180 intervention rats. As for measuring only 10 rats from each group, this is in line with the OECD guidelines but I believe the selection of 10 rats has to be randomized and preferably blinded. There was no mention of this in the paper.

Argument 3 – Number Of Control Rats

Séralini’s control groups were the same size as each treatment dose group, in line with standard scientific practice.

Counter-Argument 3

Yes, each group was the same size (20 each), but there were 180 rats in 9 intervention groups and only 1 group of 20 acted as a control.

Argument 4 – No Food Intake Data Is Presented, So We Don’t Know The Dose Of Toxins Ingested

Séralini measured food intake more often than industry studies on GM foods (2x a week) and the absence of data in his published paper does not invalidate the findings observed. This was a more frequent and detailed measurement than industry tests which measured only food intake, and only on a weekly basis. Monsanto’s 90-day study on NK603 maize also does not present individual food intake data but only the mean amount consumed for each group.

Counter-Argument 4

If he took the food measurements why not release that data? And while Monsanto recorded food consumption once a week rather than twice a week that is not going to change the amount of food consumed. Neither study provided individual food intake in their papers, although Monsanto’s raw data including food intake has since been released via court order. Monsanto’s paper did provide mean amount consumed per group which is more than Séralinis paper, which reported nothing. To date, Séralini has still not released his full data including food and water consumption.

Argument 5 – No Mechanism For The Effects Observed Has Been Established

There is no requirement in any regulatory system to establish a mechanism of action for a toxin before regulatory action can be taken and there is no burden of proof on scientists who find toxic effects to establish a mechanism before they report their findings. This is fortunate because it can take decades to establish a mechanism, and sometimes a mechanism is never found.

Counter-Argument 5

Even if the mechanism of action is not established, it is highly improbable that two completely different treatments, rats fed GMO corn and rats fed Roundup, would experience the exact same outcomes.


In January 2014, the editor of Food and Chemical Toxicology A. Wallace Hays sent Séralini a letter asking him to voluntarily retract his study [24]. Séralini refused, so the editor retracted the paper [25].

Should the paper have been retracted?

The Committee on Publication Ethics (COPE) states that the below are the only reason a paper should be retracted:

  • Clear evidence that the findings are unreliable due to misconduct (e.g. data fabrication) or honest error.
  • Plagiarism or redundant publication.
  • Unethical research.

I think you could make the strongest argument for number 1; the data presented is completely unreliable and could be classified under an ‘honest error’, although I'm not so sure the misrepresentation of the data was that honest.

The paper was officially retracted because it was ‘inconclusive’. That is not a good enough reason. Lots of inconclusive papers contribute to the scientific literature. On the other hand, leaving this paper unretracted would have given it credibility. 

The publication of this paper stands as a shining example of the limitations of the peer-review process or perhaps a lapse in editorial judgment from the Journal of Food and Chemical Toxicology.

There are some that argue that the paper should not have been retracted. An article at the Environmental Health Perspectives notes the following:

“Equally troublesome is that this retraction does not really impact how the science will be viewed by scientists, but only how it is viewed by others outside of the scientific community. We feel the decision to retract a published scientific work by an editor, against the desires of the authors, because it is “inconclusive” based on a post hoc analysis represents a dangerous erosion of the underpinnings of the peer-review process, and Elsevier should carefully reconsider this decision” [19].


The paper was re-published June 24, 2014, in the Environmental Sciences Europe journal (ESEU), an open-access journal in which you pay to have your study published [38]. It has been reported that the paper underwent a second peer-review process for this journal. This is not true. Henner Hollert, the editor of ESEU, stated that the “role of the three reviewers hired by ESEU was to check that there had been no change in the scientific content of the paper” [23]. He went onto state that because Food and Chemical Toxicology had already conducted a scientific peer-review, he did not feel the need to conduct a second [23].


1 - There should be more long-term feeding trials to thoroughly evaluate the safety of pesticides and herbicides. The EFSA is currently working on this [32].

2 - Anybody doing research into these areas should be required to release the raw data along with the study. This should not just apply to the areas of food and chemical safety but to all scientific disciplines. More about this issue can be read over at

3 - Don't get your science news from gossip magazines. Or The Food Babe.

“Most people are not natural-born statisticians. Left to our own devices, we are not very good at picking out patterns from a sea of noisy data. To put it another way, we are all too good at picking out non-existent patterns that happen to suit our purposes” [39].

But wait, there's more!

GMO’s Cause Cancer, Part II: Much Ado About Monsanto


1 –

2 –

3 –

4 –

5 –

6 –

7 –

8 –

9 –

10 –

11 –

12 –

13 –

14 –

15 –

16 –

17 –

18 –

19 –

20 –

21 –

22 –

23 –

24 –

25 –

26 – http://www.gmoSééralini-replies-to-fct-journal-over-study-retraction/

27 –

28 –

29 – http://www.gmoSé

30 –

31 –

32 –

33 –

34 –

35 –

36 –

37 –

38 –

39 –

40 -

41 -

42 -

43 -

44 -

45 -

46 -