The three NK603 publications discussed by us were written - and hence, experiments were planned - before EFSA published its 2011 Guidance on 90-day feeding trials. While the retrospective application of new standards (i.e. study methods) as undertaken by EFSA  is debatable, it is broadly accepted that risk assessment informing regulatory decision-making should be based on the latest state of knowledge and standards. A re-evaluation of the applied methods of relevant, older studies is justified if applied to all such relevant studies - but meaningless or even counterproductive when applied selectively to one study or a subset of studies only triggered by their reported (here adverse) effects. We consider such a selective re-evaluation as inappropriate. We note that the main responsibility for EFSA's recent work lies with the European Commission that in its mandate to EFSA singles out the Séralini study for review .
EFSA criterion: study objectives
] notes with regard to Séralini et al. [3
] that the ‘study objectives are unclear’. EFSA further clarifies that ‘If a specific guideline is chosen and followed, then the objectives are inherently defined in the guideline.’ To clarify what ‘inherently defined’ objectives entail, we analyzed the four Guidelines of the Organization for Economic Co-operation and Development [18
] discussed in this context (OECD nos. 408, 451, 452, 453, respectively). While the three latter Guidelines revised in 2009 give clear objectives, Guideline 408 on 90-day oral toxicity originally adopted in 1981 and revised in 1998 lacks such clarity. Guideline 408 is the guideline of choice by Monsanto Company for their 90-day rat feeding trial.
Open issue according to EFSA: ‘The study objectives need to be clearly stated a priori in the study protocol.’
Séralini et al.  stated that ‘Guideline 408 of the Organization for Economic Co-operation and Development (OECD) was followed by some manufacturers for GMOs even if it was not designed for that purpose. We have explored more parameters and more frequently than recommended in this standard in a long-term experiment.’ The authors explained that they added biochemical and haematological measurements using 10 rats per group as described in the OECD Guideline 453. This apparent combination of elements of two OECD guidelines and the addition of more parameters did not qualify as proper description of the objectives of the work by EFSA.
In its technical 90-day study, Monsanto tested a low-dose group (LD 11% NK603) and a high-dose group (HD 33% NK603) versus two parental control groups and six non-parental, so-called reference controls for 13 weeks. Blood and urine biochemical parameters, amongst others, were measured at an interim time point after 4 weeks and at termination of the study. The statements by Monsanto  and later by Hammond et al.  to have adapted the OECD Guideline 408 were considered by EFSA to be sufficient. Neither the nature nor the consequences of the adaptations made to the requirements of OECD Guideline 408 were explained in these two publications or in the scientific opinions of EFSA Panel on Genetically Modified Organisms [16, 17].
Since OECD Guideline 408  in contrast to the more recent OECD guidelines does not provide clear objectives, it is doubtful whether a mere reference to this guideline as provided by Monsanto suffices EFSA's new standards on study objectives. We conclude that regarding this criterion, the Séralini publication provides at least the same quality in terms of clarity and explanation as the Monsanto and Hammond studies.
EFSA criterion: study design
The study design applied by Séralini et al.  triggered two main criticisms by many commentators including EFSA : (a) the number of rats per group and (b) the choice of the strain of rats.
A. EFSA  states: ‘[g]iven that Séralini et al. (2012) conducted a two-year study, it is unclear why an OECD guideline suitable for a 2-year chronic toxicity or carcinogenicity study (i.e. OECD 451, OECD 452, or OECD 453) was not adhered to.’ Séralini et al.  used 10 rats per group as recommended by OECD Guideline 453  for the chronic toxicity phase. But OECD Guideline 453 also clearly states that this lower number is only justified when a carcinogenicity phase with 50 rats per group is run in parallel to support the interpretation of the toxicity data. Later, Séralini et al.  argued that the OECD Guideline 452 in its 1981 version , which was applicable for the 2-year toxicology study conducted from 2008 to 2010, advised to only use 10 rats out of the 20 per group for blood and urine sampling and analysis. The recommendation to use at least 10 rats per group for biochemistry analysis at the chosen time points is maintained in the current version of Guideline 452.
Monsanto  used 20 rats per sex and group, while the actual values for blood chemistry, haematology and qualitative, and quantitative urine analyses were based on samples from 10 rats per sex and group, thus, comparable to the approach of Séralini et al. . Nowhere in the two versions of Guideline 452 or in the Monsanto study, a method for how to select the 10 animals out of the 20 animals per group for the biochemistry tests is specified.
B. Regarding the choice of SD rats, EFSA notes that this “strain of rats chosen is known to be prone to development of tumours over their life […] This is neither taken into account nor discussed in the Séralini et al. (2012) publication.” A discussion on the choice of rats with regard to their tendency to develop tumours is not relevant for 90-day studies. Nevertheless, EFSA recommends stating reasons for the choice of the strain in its new guidance on 90-day feeding trials.
Open issue according to EFSA: ‘The biological relevance of the rat strain used should be justified with respect to the design choices.’
Neither Séralini et al.  justify the choice of SD rats nor do Hammond et al. . Monsanto  explains that ‘[t]he rat was selected for the study since this species has been traditionally used to assess the safety and wholesomeness of food.’ This pragmatic explanation underlines the customary use of SD rats in such trials which of course extends also to the Séralini study but does not give a scientifically sound justification for its use with regard to the design choice as implied by EFSA's wording .
In (Additional file 1
: Table S1), we present information on the actual use of SD rats in peer-reviewed, long-term studies and the types of information researchers usually provide and scientific journals require on the choice of animals. According to the information available, SD rats were chosen as standard test organisms by the two largest toxicity/carcinogenicity research projects worldwide and at least 21 long-term studies:
The National Toxicology Program of the US Department of Health and Human Services uses this strain in its 2-year studies, after in-depth discussions on the suitability and advantages of the SD rat over previously used strains .
The European Ramazzini Foundation for Oncology and Environmental Sciences (Italy) uses SD rats in its Ramazzini Foundation Cancer Program since more than 40 years .
Contrary to many media reports based to a large extent on scientists' personal opinions posted by the Science Media Centre , SD rats are used routinely in long-term toxicology and carcinogenicity studies. The rational to do so is that (additional) carcinogenic effects of test substances can be detected more effectively in rats that develop tumours at a relevant rate within the time span of a 2-year trial. Actually, the prime issue of an informed debate would not be so much the choice of the strain of rats but the choice of the numbers of rats depending on the strain used. The OECD Guideline 116  on the conduct of long-term carcinogenicity studies states that ‘[f]or strains with poor survival such as Sprague–Dawley rats, higher numbers of animals per group may be needed in order to maximize the duration of treatment (typically at least 65/sex/group).’ The US Food and Drug Administration (US FDA)  recommends for carcinogenicity tests with strains with known problems in survivorship that the researchers need to ensure that at least 25 rats from those used initially remain alive at the end of the 2-year trial. For long-term toxicity trials as conducted by Séralini et al. , no specific recommendations are given by the US FDA. With regard to the recommended number of 10 rats per toxicological test in OECD Guideline 452 , the above recommendation of FDA should be applied to ensure that in average, more than 10 rats would be available for terminal testing. The reported survival rates of the control rats in Séralini et al.  (70% for males, 80% for females) lie at or above the upper range presented in available literature on survival rates of SD rats in 2-year studies [28–31].
We conclude that with regard to the choice of rats, a pragmatic approach was followed by Monsanto, Hammond et al. [14
] and also Séralini et al. [3
]. Since the Séralini group had set out to repeat the 90-day feeding trial by Monsanto but to differ with regard to the extended testing time and parameters measured, the choice of rat was appropriate in order to allow for comparability with regard to this factor. We also conclude that the SD rat strain used is a standard organism for toxicity and carcinogenicity tests. A rejection of the validity of the Séralini study based on the choice of rats would therefore either be based on a lack of familiarity with the scientific field. Or these criteria would have to apply to all carcinogenicity trials conducted to date with this rat strain and, thus, leave us with hardly any valid carcinogenicity study. Criticism of the Séralini study based on the lack of a scientific explanation of the choice of the rat strain would similarly apply to the Monsanto and Hammond publications.
Open issue according to EFSA: ‘Suitable controls for all treatment groups are not present.’
None of the discussed OECD guidelines was developed for the testing of whole foods but rather for testing the toxicity of isolated chemicals when feeding rats with standardized, nutritionally balanced feed. In most cases, the test substance is administered separately from the normal feed or mixed with the feed in a minor volume. As long as the addition of the test substance to the feed does not alter significantly the relative concentration of the feed components, a single control with zero test substance is sufficient. If the test substance is an inherent component of the feed as in the NK603 feeding trial and the different test concentrations are produced through varying percentages of GM and non-GM maize varieties in the feed, appropriate controls are necessary according to EFSA . Hammond et al.  and Monsanto  report that all test groups received feed containing 33% maize with a different percentage of GM versus non-GM maize. Séralini et al.  did not report whether they had adjusted the total maize content to 33% in those batches containing 11% or 22% GM maize. To clarify, Séralini et al.  informed that all feed batches were mixed to contain 33% maize in total.
We conclude that based on the information given by Séralini et al. [15
], this open issue is closed.
Open issue according to EFSA: ‘Measures taken to reduce the risk of bias (e.g. blinding) are not reported.’
While EFSA  has noted that Séralini et al.  did not report on measures taken to reduce bias in their experiments, we note that also neither Hammond et al.  nor Monsanto  reported such measures. Blinding measures are not requested by any of the above-mentioned OECD guidelines. However, Séralini et al.  explained that anatomopathology and biochemical measures were indeed performed in a blinded manner.
The issue of blinded evaluation was promoted by EFSA when it presented its draft  of the above-mentioned Guidance on a 90-day oral toxicity study for public consultation. EFSA's approach was rejected by many commentators from pathology sciences and industry  with reference to the ‘Best Practices Guideline: Toxicologic Histopathology’ of the Society of Toxicologic Pathology. This guideline explains that unblinded evaluation is regarded as the appropriate approach in the first round of pathological observations because it ‘allows the pathologist to intensely focus the histopathologic evaluation and to find important, and sometimes subtle, differences between the tissues of treated and untreated animals’. The guideline recommends that blinded evaluations should be applied in the re-evaluation of findings in specific tissues and when samples with a defined spectrum of lesions due to a known toxic syndrome are investigated. These recommendations to not apply blinding as a general measure were reflected in EFSA's report on their public consultations  and in the final EFSA Guidance on 90-day oral toxicity study. It is unclear why EFSA  requested blinding measures in the context of the Séralini study but did take no issue of this in its positive opinions on the NK603 dossiers including Monsanto's rat feeding studies.
We conclude that a double standard was applied with regard to the criterion of blinding. We also conclude that EFSA’s approach does not reflect the previously published opinions of various experts in the field.
EFSA criterion: feed and treatment formulation
Open issue according to EFSA: ‘The appropriateness and comparability of the diets cannot be assessed as critical information about their composition is not reported.’
Séralini et al.  did not present their data on the composition of the feed but informed that ‘all feed formulations consisted in balanced diets, chemically measured as substantially equivalent except for the transgene’. Similarly, Hammond et al.  did not present such data but mentioned that all diet preparations were analyzed to confirm that they met the supplier's specifications for certified 5002 rodent diet. Monsanto  presented detailed data on the composition of the feed, as it is standard procedure for technical studies for regulatory feeding trials.
We conclude that the two publications by the Hammond and Séralini groups do not fully comply with this criterion while both studies mention that the composition of the different feed was measured and found to be comparable. A rejection of only one study due to the lack of reporting of data would constitute a double standard.
Open issue according to EFSA: ‘The stability of the diets cannot be assessed as details of their storage conditions are not provided.’
] noted that Séralini et al. [3
] did not inform about the storage conditions of the feed - likewise, neither Hammond et al. [14
] nor Monsanto [12
] informed about this. Consequently, the stability of the diets cannot be assessed in any of the studies submitted by the developers to EFSA, which went uncommented in EFSA’s positive opinions on NK603 maize [16
Open issue according to EFSA: ‘It is impossible to evaluate whether or not there was any contamination of the diets, e.g. by mycotoxins, as it is not reported.’
Although Séralini et al. [3
] did not report about mycotoxin measurements in their original publication, later, the authors [15
] stated that all measured mycotoxins showed concentrations below the recommended threshold values for food and feed. The EU maximum level is 4 ppb for e.g. the sum of Aflatoxines B1, B2, G1, and G2 in processed maize for human consumption and 20 ppb or lower for Aflatoxine B1 in feed materials. Similarly, Hammond et al. [14
] informed that the levels of aflatoxins were below detection limits but did not provide the actual individual measurements or raw data. Monsanto [12
] did not refer to aflatoxins alone but to the measurement of a total of 19 different mycotoxins. The author noted that there was no contamination with mycotoxins that might interfere with the results. Of the 19 analyzed mycotoxins, 4 (Deoxynivalenol, Fumonisin B1, B2, B3) could be detected in individual concentrations up 1.6 ppm in the various maize samples. The concentration did not exceed 3 ppm in the formulated diet, which Monsanto [12
] regards as the No Observed Effect Level for Fumonisin B1 in rats according to Voss et al. [36
]. The EU maximum level for Fumonisin B1 and B2 lies at 60 ppm for maize as feed material while the EU maximum level for maize products for human consumption is 1 ppm or lower.
Open issue according to EFSA: ‘The amount of residues of glyphosate and its metabolites in treated maize NK603 is not reported.’
Séralini et al. [3
] did not present the results of their glyphosate measurements but stated that pesticide content did not exceed standard limits. The current EU maximum residue level for glyphosate in maize is 1 ppm. Hammond et al. [14
] noted that the levels of glyphosate were below detection limits but also did not give any data. In contrast, Monsanto [12
] informed that glyphosate was indeed measurable: ‘Glyphosate residue in the test grain (0.09 ppm) was slightly above the analytical detection limit of 0.05 ppm.’ It is very likely that the NK603 maize used as feed was treated with a glyphosate-containing herbicide during its growth. In order to address the presence of two factors relevant for risk assessment - the transgene and the complementarily used herbicide - information on the use of glyphosate and its residues should be presented in research relevant for risk assessment [37
Open issue according to EFSA: ‘The exposure to GMO, GMO + R, and R cannot be evaluated since the food and water intakes of the GM- and R-treated groups, respectively, are not clearly reported.’
Séralini et al.  did not present the values of feed and water intakes (both containing the test substances) - but also Hammond et al.  did not present feed intake data (only feed contained the test substance). Monsanto  presented feed intake data, as this is standard procedure for technical studies for regulatory feeding trials. However, this data still cannot be used to quantify the exposure to the CP4 EPSPS protein because Monsanto  did not quantify the concentration of CP4 EPSPS protein in the maize kernels used to mix into the feed. Monsanto  only conducted qualitative ELISA tests to confirm the presence or absence of the CP4 EPSPS protein. The impossibility to calculate exposure data based on the information given by Monsanto  was not flagged by EFSA in its two positive opinions on the NK603 dossier.
While the presentation of data on concentrations of contaminants is certainly useful, a rejection of the validity of the Séralini study based on the fact that only statements on such concentrations are given is debatable. And again, we conclude that applying a criterion to only one study would constitute a double standard in all four open issues discussed above.