Proposed criteria for the evaluation of the scientific quality of mandatory rat and mouse feeding trials with whole food/feed derived from genetically modified plants

There are a number of internationally recognized test guidelines, depending on the purpose and duration of the feeding trial, which can be recommended as starting points for the evaluation of whole food/feed. These are, among others, the OECD Test Guidelines 408 (90-day subchronic toxicity study), 451 (2-year carcinogenicity study), 452 (1-year chronic toxicity study) or 453 (combined chronic toxicity/carcinogenicity study), complemented by the EFSA guidance on conducting repeated-dose 90-day oral toxicity studies in rodents on whole food/feed, the EFSA considerations on the applicability of OECD TG 453 to whole food/feed testing and/or the EFSA explanatory statement on the above-mentioned EFSA guidance document. In the OECD documents, standardized testing protocols are described. However, since the OECD Test Guidelines were intended to evaluate chemicals, adaptations of these protocols are made when wanting to test whole food/feed.

The risk assessment scenarios 1 and 2 were originally described by EFSA for 90-day studies. In the case of scenario 2, in which no hypothesis to be tested has been identified, the study design based on the relevant OECD Test Guideline 408 and the corresponding EFSA recommendations is followed and the full range of observations/parameters described in the above-mentioned OECD Test Guideline is recorded. In the case of scenario 1, a study design of the feeding trial considering appropriate endpoints on a case-by-case basis will be required to test a specific hypothesis. In case of planning an extended feeding trial to test the potential chronic toxicity and/or carcinogenicity of whole food/feed derived from a GM plant, the protocols described in the corresponding OECD Test Guidelines are adapted, e.g. as described by EFSA.

A quantitative compositional analysis of the plant materials and diets is necessary to identify factors (e.g. the presence of a mycotoxin) other than the transgenic modification that could influence the outcome of the feeding trial. In case that such a factor is identified, the probability that it could lead to unintended effects in the laboratory animals is considered when evaluating the outcome of the feeding trial. The analysis of the plant material and diets includes nutrients [proximates (ash, total carbohydrates, fat, protein), starch, fibres, fatty acids, amino acids, sugars, minerals, vitamins], and plant-specific secondary metabolites, in particular antinutrients (e.g. phytic acid, trypsin inhibitor, lectins) and toxicants as specified in the OECD Consensus documents for the work on the Safety of Novel Foods and Feeds: Plants (http://www.oecd.org/science/biotrack/consensusdocumentsfortheworkonthesafetyofnovelfoodsandfeedsplants.htm). In addition, the plant material and diets are tested for the presence of genetically modified organisms, chemical contaminants (heavy metals, nitrosamines, polychlorinated dioxins, polychlorinated biphenyls, polycyclic aromatic hydrocarbons, pesticides, mycotoxins) and the microbial contamination. Representative plant materials and diet samples are taken for the analyses. Based on the analysis of the plant material, nutritionally balanced diets adjusted to the dietary requirements of the rat or mouse strain used are formulated.

The plant material is fed orally and ad libitum. In principle, it is not necessary to use two dose levels in addition to the control (zero dose level), but in practice two dose levels may help in the interpretation/evaluation of results in order to distinguish between treatment- and non-treatment-related effects.

EFSA suggested the following incorporation rates as reference values for high doses in 90-day feeding trials in rodents: 60 % for rice (dehulled), 50 % for maize, 30 % for soybean meal, 25 % for rapeseed meal and 20 % for full fat soybean as well as potatoes (heated, dried). In this regard, a preliminary “maximal inclusion rate finding study” to determine the maximal level of material from a specific plant species that could be incorporated in the animal diet without leading to a nutritional imbalance is conducted before performing a 90-day feeding trial, particularly in case of an incorporation rate or a plant species never having been tested before in a feeding trial.

Body weight and feed consumption are monitored regularly. Although cage mean values are used for the statistical analysis (since the cage is the experimental unit), the individual animal body weight values are reported if a cage houses two animals.

The fasting period prior to blood sampling is the same for all animals. This is an important point to decrease variability in the haematology and clinical biochemical analyses. Rats are fasted overnight (i.e. about 16 h), in the case of mice the period should only comprise 5–6 h. Blood is taken from the same site in all animals, and possible deviations are justified. Blood is stored under appropriate conditions, either without or with an anticoagulant as specified.

The clinical chemistry analyses of the blood samples are completed as quickly as possible, ideally within 1 day, to minimize potential variability between the samples. An analysis within 1 day might not be possible when handling a high number of samples, as occurs in the case of feeding trials being performed according to internationally accepted guidelines. In such a case, the analyses are performed according to a randomized sampling design (see criterion no. 6), where all analyses within one “sampling block” are done closely in time. The date of the analyses is recorded, and the sampling structure is considered when performing the statistical analyses.

The advantage of a blind study is that it avoids the influence of any bias that experimenters and analysts may have with regard to possible outcomes. Without blinding, prior expectations may lead to an unconscious biasing of actions and observations. This is particularly relevant for those endpoints in which subjective elements might influence the observations. The blinding is applied to all staff involved, such as the personnel working with the animals, those that weigh the animals, those involved in the necropsy, and all analyses of material stemming from the experiments.

A preliminary power analysis is performed to estimate the probability to detect effect sizes of potential relevance. Effect sizes of potential relevance are based on an expert opinion, but may also be inspired by the analysis of datasets from non-GM groups in previous studies with the same plant species and with sufficient similarity to the intended animal study. The power analysis requires for all parameters the specification of: (a) the significance level of the test (this will be assumed to be 5 %, unless regulatory requirements justify another value); (b) the sample size (which will be that of the intended design); (c) the effect size (this will be the effect size of potential relevance); and (d) the variability, e.g. the standard deviation (within each dose group) of the measurements (this will have to be derived from previous data or be specified based on an expert opinion). In a scenario 2, standard sample sizes (e.g. mentioned in OECD Test Guidelines) are used. The role of a power analysis is then to check the statistical power of tests for the different parameters to be measured. The power analysis may focus on a subset of endpoints of primary interest.

In the frame of the statistical analysis, group differences (effect sizes) of interest (e.g. GM group vs. control group) are tested for effects different from zero (difference tests). Finally, toxicologists assess whether the observed effects are relevant from a toxicological point of view by taking into account the complete set of measured parameters, whether or not the differences are significantly different.

All statistical methods are reported and comprehensively described. It is important to differentiate between the “classical” (analysis of variance + post hoc test) and novel approaches (e.g. standard effect size analyses) until the risk assessment community has accepted the latter approaches. The statistical analysis plan is explicit regarding what mistakes/shortcomings should be avoided. Furthermore, raw data are made available to third parties to allow for an independent analysis.

Taken together, nine criteria to evaluate the scientific quality of rat and mouse feeding trials with whole food/feed derived from genetically modified plants are proposed and a number of specific aspects to be taken into account in conjunction with these individual quality criteria are addressed. It is recommended that the quality assessment of a feeding trial in the frame of a regulatory decision process is made on a case-by-case basis considering all relevant quality criteria proposed in this letter. It is important to note that a feeding trial does not automatically provide useful information simply because it meets the nine proposed criteria.

Only in case a trigger is available from the initial molecular, compositional, phenotypic and/or agronomic analyses and therefore the rationale of the study prior to testing is formulated in form of hypotheses regarding specific endpoints, feeding trials with whole food/feed may provide an added scientific value for the risk assessment of GM crops.

It is expected that this letter will trigger a broader scientific debate on the quality of rodent feeding trials with whole food/feed, and for this purpose, contributions are welcome in the Discussion Forum established by the scientific coordinators of the European research projects GRACE, G-TwYST and GMO90+ in Archives of Toxicology.