Almost 7 years after our research showed Roundup and Roundup-tolerant GM corn to be toxic, we welcomed the intention of a detailed study inspired by half of our work, which is quoted in the first few words of the summary of Steinberg et al. . “Half of our work”, because the testing of long-term effects of Roundup alone at environmental levels (0.1 ppb) on a diet without pesticides was not accomplished nor tested in Steinberg et al.’s work, in contrast with ours, as indicated by these authors. GMOs treated with Roundup were tested as feed in comparison to controls, which we also did, but with the opposite results. The goal of this discussion is to analyse the reasons for the discrepancies in results, especially because these are the only two long-term feeding studies on this matter. Steinberg et al. used a different rat strain called Wistar Han RCC versus Sprague-Dawley used by Seralini et al.  and, notably, by Monsanto in their 90-day rat feeding study which first allowed the authorization of consumption in the world . The Wistar Han rat is considered less sensitive than the Sprague-Dawley strain, without real proof. Steinberg et al.  also used a larger number of animals, as is suited for a carcinogenicity study. In contrast, we followed a long-term toxicity protocol. Our retracted and republished rat study [9, 16] was originally not even meant to be a cancer study; this was indicated in the paper. Therefore, the experimental design and number of rats was consistent to OECD 452 when it began in 2008; at this time the 1981 norms were routinely used; although there was no guidelines for GMOs, we had to adapt and improve these . However, as tumours were developing at higher rates in the treatment groups, we were required to report this as an indicator for other serious effects that must be followed up. Steinberg et al.  also benefited from far more funding than our pioneering study.
The serious deficiencies of Steinberg’s protocol: glyphosate-based residues at high levels everywhere, even in controls when studying a glyphosate-tolerant GMO
With regard to GM maize, there was a control group with supposedly no GM feed in the Steinberg study, another group fed GMO corn unsprayed with Roundup, and a group fed GMO corn sprayed with Roundup, but the feed made from these plants, as well as the control group feed, were all contaminated with 30–140 ppb of this pesticide (i.e. µg/kg, corresponding to 300–1400 times more of glyphosate and thus the other Roundup residues than the levels we found to be toxic in our study . This is explained in Steinberg et al.’s paper (in  discussion, 1st column, 2nd paragraph), as well as in another publication from the same group, specifically on the contamination of their feed [23, 24]. This increases the background level of serious diseases in the controls, preventing any specific observable effect of the GMO treatment on animals.
Given such neglect of the contamination issue, we would have stopped there instead of drawing scientifically inadequate conclusions. The glyphosate contamination was due probably to a mix with another transgenic plant tolerant to Roundup, such as soybean or rapeseed. But Steinberg et al.  continued with their long-term toxicity study with that contaminated diet. They consider that the glyphosate detected “has no harmful effect” (discussion, 1st column, 2nd paragraph) since “It shows no genotoxic or carcinogenic potential in mice and rats”, according to EFSA. This could be considered as biased and selective citing of non-experimental opinions from EFSA. However, this agency of the European Union was compromised on this issue, according to several investigations that led to resignations . The debate is still ongoing between supporters of the World Health Organization’s International Agency for Research on Cancer (IARC), which declared glyphosate a possible carcinogen , and supporters of other regulatory agencies, including EFSA and the European Chemicals Agency (ECHA), which stated that it does not pose a carcinogenic risk.
A second serious mistake: many other carcinogenic compounds in controls masking any long-term observable effect of the treatment
In addition, the oxidized petroleum residues  and arsenic and other heavy metals formulants found in Roundup  must be taken into consideration, because these are by themselves carcinogenic and endocrine disruptors. This is a serious and inexcusable gap in the scientific assessment, which gives rise to a major methodological and interpretive bias.
Steinberg et al.  also confuse the terms glyphosate and Roundup, while we have identified from 2013 to 2018 the presence of toxic residues alongside glyphosate [7, 8]. Moreover, the rat feed in Steinberg et al.’s experiment contained 70–290 ppb arsenic/fresh weight arsenic [23, 24], suggesting that all animal groups were subjected to its chronic toxicity. This by itself may mask the possible effects of the studied GMO over the long term.
Steinberg et al.’s control diets, as well as the treatment diets, were contaminated with many other pollutants (results, end of 1st column ): dioxins and furans, PCBs, PAHs, mycotoxins and nitrosamines in “low and similar” amounts . All diets also contain pesticides: 2-phenylphenol, cypermethrin, deltamethrin, tetramethrin, ethoxyquin, piperonyl butoxide, pirimiphosmethyl, N-desethyl-pirimiphos-methyl and propiconazole, and glyphosate, as well as the glyphosate metabolite AMPA. As they are supposed to be below the regulatory limits (which we have repeatedly indicated to be inadequate  since these were never tested in long-term studies with the toxic formulations of pesticides), “they could not affect the health of rats in any way”, according to the authors (results, 2nd column, 1st paragraph ). But this assertion can be ruled out by experts on this domain since combined effects of mixtures, as well as endocrine and carcinogenic effects, are all possible from these contaminants , which are effectively observed in their controls.
This contamination and the justification given in the paper  are therefore, scientifically not acceptable. In order to avoid such problems, in our study, we fed the control animals with plants specially grown without pesticides, using organic farming methods. This is the correct way to study and compare the effect of a pesticide or a GMO that contains a pesticide in the diet of the treated group. These organic methods have been proven to yield crops that test negative for pesticides in comparison to pesticide-treated neighbouring ones .
In 2015, we published a study  showing the heavy contamination of commercial laboratory animal feeds with pesticides and many other pollutants even in regulatory experiments. We concluded that tumours that arise in animals fed such diets cannot be assumed to be “spontaneous” or naturally occurring. We had 5–8 times fewer tumours and diseases in our pesticide-free control rats than did Steinberg et al. . This key finding could be explained by the lack of contaminants in our diets, which were therefore able to highlight any effects from the GMO and/or the Roundup. It is also important to use a rat strain susceptible to tumours because Homo sapiens is also highly susceptible to cancer. However, the differential sensitivity between the different rat strains is not so obvious if all the chemical contaminants are not carefully measured in their environment and diets.
A third serious mistake: many other glyphosate-tolerant GMOs and others in controls
The contamination of the foods of Steinberg’s control diets does not stop there. They contain GM corn NK603, albeit in “non-quantifiable” traces (Results , “Feed composition analysis”), which are supposed to “not influence the feeding trials in any way”. However, no evidence is presented to support this assumption. Steinberg et al.’s diets also contain many other types of GMO, as they admit in their raw data (various GMOs and their associated pesticides contaminate commercial laboratory animal feeds, as explained above).
This highlights the negligence in the production of the diets. The presence in the diets of numerous contaminants biases the experiment towards finding “no effect” from the test substances.
The results from Steinberg et al.: major biases and subjective interpretations
In fact, many rats, including in the control and treated groups in Steinberg et al.’s experiment had tumours and cancers in various organs after 2 years. According to the authors, they arose spontaneously, and comparably in all groups, but for us they could be explained by the contamination of their feed. Their very high level may mask any treatment-related effect.
No research was done on the chronology of the appearance of the tumours or their size compared with controls, contrary to our experiment. The chronology of mortality was not studied. Steinberg et al. studied two different doses of the GMO, whereas we studied three. When only two doses are studied, no dose-related effect can be investigated.
The authors considered a priori that all the feed contaminations would have no effect. This is only their subjective opinion, and many indications that we have cited can prove the contrary.
High mortality rates in males fed GM NK603 corn
In spite of the many weaknesses of the study design, Steinberg et al.  still found significant differences, most notably in male mortality, which was higher in the animals fed the GM corn sprayed with Roundup for 2 years (Table 12).
In addition, increased incidence of pituitary neoplasia, and disorders of the sex hormones estradiol and thyroid in females were also noticed by them (discussion ). We also showed such differences many years ago in Seralini et al. [12, 16]. Steinberg et al. dismissed these effects as “not… adverse” due to the lack of histopathological alterations in the estrogen-sensitive tissues and organs. However, lesions can be missed in the histopathological sectioning, and/or some functional alterations that have biological effects on the organism may not result in histopathological changes. It is not the place of Steinberg et al. to dismiss such changes based on assumptions, like EFSA or industry conclude, particularly in a research study conducted with the aim of revealing any health risk to humans. Moreover, the histopathological sections are not shown even in supplementary data , and thus cannot be analysed by others to confirm or refute the interpretation of Steinberg et al.
Steinberg et al. dismiss some statistically significant differences in treatment groups as not biologically relevant since they are “small” or “not dose-related”, the latter meaning there should be an effect proportional to the dose of the GMO. However, such assertions are not scientifically justifiable . A dose-related observation begins with three doses and not two according to OECD. Moreover, an effect that is statistically significant should not be dismissed as “small” and the effects of hormone disruptors are often not proportional to the dose.
In order to dismiss the differences, the authors compare the effects observed in this experiment with the “historical control data” obtained from the previous feeding trials conducted under the related GRACE program or others. This use of unrelated historical control data violates the Test Guidelines of the Organisation for Economic Cooperation and Development (OECD) on the conduct and design of chronic toxicity and carcinogenicity studies —guidelines that Steinberg et al. cite in their paper. The OECD states, “the concurrent control group is always the most important consideration” when considering the effects of the treatment under test.
We also oppose such lax approaches, which can only lead to the underestimation of potential toxicological effects. All differences must be carefully considered, and a precautionary approach taken, which favours the investigation rather than neglect of risks, since this is the last stage of biosafety research before mass consumption by animal and human populations. No clinical trials are made for food.
Surprisingly, the authors finally conclude from all this that we should no longer bother to conduct long-term studies on agricultural GMOs in general, which is contrary to the spirit of scientific inquiry and (more importantly) is not supported by the concerning results that were found in spite of the methodological weakness of the study. We strongly disagree, even if it is the same reasoning for Monsanto, industry and EFSA that has allowed commercial release on this basis.
Serious undeclared conflicts of interests at many levels
Pablo Steinberg declares that he has no conflict of interests in the article he publishes on these results and their interpretations, which we have revealed as very lax, in accordance with Monsanto’s practices that we had counter-appraised and published .
However, his published paper mentions that he subcontracted some of the analyses to the Covance Laboratory in Madison, Wisconsin, USA, in which Monsanto has invested and performed many studies with the same GMO and others. These studies  had previously shown bias, which we previously documented in detail [32,33,34,35].
In addition, Steinberg noted that he was an expert for the International Life Science Institute (ILSI) in his declaration of interest on another public document for G-TwYST, the acronym for this experiment . ILSI was chaired by Monsanto and it is well known as an industry-funded lobbying organization  which has worked to weaken regulation and testing, including of GMOs and pesticides, and is supportive of their use .
Pablo Steinberg also sat on the BfR Committee; i.e. the German risk assessment agency responsible for the evaluation of glyphosate. This agency copy-pasted whole passages of the evaluation written by Monsanto and has come under severe scrutiny because of this practice .
We also note that Pablo Steinberg has been (and is still) an adviser, since 2005, to the Danone Institute. The group of the same name is a member of ILSI. The publication of our 2012 study motivated Steinberg’s work because we found many risks for this GMO and Roundup. But is now proven, thanks to the Monsanto Papers, that as soon as our study was published, a senior member of the Danone group coordinated the contacts between Monsanto and the French government  that led the latter not to take our research into account at the regulatory level. If our research had been considered, it would have provoked a ban on GM NK603 corn and Roundup. Danone has molecular biology laboratories that produce genetically modified microorganisms for yogurts and other products. Monsanto has subsequently carried out an until then unprecedented defamation campaign against us, through a “scientific community” constructed by and consisting of ILSI members and other long-time industry-aligned scientists or influenced by these ones. These facts were revealed in the seven libel suits that we brought and won against these lobbies from 2011 to 2017 .
For all these reasons, the findings and conclusions of Steinberg et al.’s study  are unreliable, and the paper should be retracted, and the results deleted from regulatory appraisals and risk assessments.