Keep up with the latest news and comment on genetically modified foods          

GMWatch responds to criticisms of Seralini's study

We've compiled a list of the most common criticisms of Seralini's study, which found increased tumours, mortality, and organ damage in rats fed GM maize NK603 over a 2-year period and similar results in rats fed Roundup in tiny amounts, less than levels permitted in food, feed, and drinking water. Our answers follow.

We'll update this document as and when needed.

1. Seralini's experiments do not conform to internationally accepted protocols.

No such protocols exist for GM food safety testing and industry and regulators have opposed attempts to establish them. So industry is free to design its own tests on its own products – or even to leave out inconvenient data from the dossier of tests it submits to regulators.

2. Groups of rats (sample size) are too small.

Seralini used the same number of rats as Monsanto analysed for blood and urine chemistry in its 90-day tests on GM foods and the same number as is recommended in OECD chronic toxicity protocol that Seralini mentioned in his paper. More rats are needed for a carcinogenicity study, but Seralini had no reason to embark on a carcinogenicity study, as no existing data from Monsanto or independent researchers indicated that NK603 GM maize or tiny amounts of Roundup might cause tumours or cancer.

3. Seralini used an insufficient number of controls.

Seralini's control groups were the same size as each treatment dose group, in line with standard scientific practice. It is not good scientific practice to introduce extra irrelevant "reference" control groups, though Monsanto has routinely done this in its tests on GM foods. This practice only introduces data "noise" which hides any toxic effects of the genetic modification.

4. Seralini used a type of rat naturally prone to tumours.

The Sprague-Dawley (SD) rat is the standard for Monsanto's 90-day tests on GMOs and for industry and independent 2-year chronic toxicity and carcinogenicity tests on chemicals. The "spontaneous" rate of tumours does not matter in a controlled experiment – what matters is the increase in tumours in treatment groups. In Seralini's study, all treatments in both sexes increased large tumour incidence by 2 3-fold in comparison to controls.

However, Seralini's study was for chronic toxicity, not carcinogenicity. In order to draw conclusions about the cancer-causing ability of NK603 GM maize and Roundup, proper carcinogenicity studies would have to be carried out with a 50 rats per sex per group (as compared with Seralini's 10 rats per sex per group) over 2 years to test specifically for this effect. Nevertheless, the findings of tumours (increased number, lower age of onset, and greater aggressiveness) are so striking in the treatment groups that they demand further investigation through a full-scale carcinogenicity study.

5. The fact that 30% of controls got tumours shows this rat is an unreliable model.

Control groups developed some tumours, but treated rats developed more. By the end of the study, 50-80% of female animals had developed tumours in all treated groups, with up to 3 tumors per animal, as against only 30% of controls. Peer-reviewed data show that the SD rat is an excellent human-equivalent model for predicting cancer in humans in long-term (2-year) studies: it gets around the same number of tumours as humans do over its lifespan. However, it must be remembered that Seralini's study was not a carcinogenicity study, not because of the type of rat used, but because of the relatively few numbers of rats per group.

6. SD rats get tumours when food intake is unrestricted.

The rats in Seralini's study had unrestricted access to food and water, but so did the rats in Monsanto's 90-day studies on GM foods, and so do most humans in real life, so this aspect of Seralini's study reflects standard industry testing practices as well as realistic human exposures.

7. The effects seen in Seralini's study are within the historical norms for this type of rat.

The only scientifically valid controls with which treatment groups of rats should be compared are the concurrent controls within the experiment. "Historical control data" drawn from a variety of sources should not be used, because in scientific experiments we only test one variable at a time.

In Seralini's experiment, the variables, each of which was tested separately, were exposure to NK603 maize, Roundup, and a combination of the two. This excludes the possibility that effects could be due to irrelevant factors such as different environmental conditions in which crops used in the diets were grown (which can produce differences in toxins or nutritional content). Industry studies on GMOs and chemicals often invoke historical control data to mask significant effects of harm found from exposure to the tested substance, but the practice is unscientific and places public health at risk

8. No food intake data is presented, so we don't know the dose of toxins ingested.

Seralini measured food intake more often than industry studies on GM foods and the absence of data in his published paper does not invalidate the findings observed.

9. The outcomes including tumour incidence did not follow a linear dose response pattern (where the toxic effect increases as the dose increases).

Many toxins, especially those that affect the hormonal system, have nonlinear dose-response patterns Roundup is one. Scientists have published papers about nonlinear dose-response patterns since the 1990s, but industry and some risk assessment bodies cling to the outdated toxicological model of linear dose-response.

10. Outcomes were sex-specific, e.g. the majority of tumours were found in females.

Sex-specific toxic effects are well documented in the scientific literature, including in a study on Roundup toxicity in rats and in animal studies on GM foods. The sex-specificity of certain toxic effects found in animal feeding trials on GM foods is routinely used by industry and EFSA as an excuse to dismiss them, but this is scientifically unjustifiable, as sex-specific effects are to be expected when the hormonal system is involved.

11. No mechanism for the effects observed has been established.

There is no requirement in any regulatory system to establish 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 mechanism, and sometimes a mechanism is never found.

12. Seralini's study is flawed and should be dismissed.

No study is perfect. But Seralini's is far stronger, in terms of study length, parameters examined, and carefulness of design (which enabled distinction between effects of the genetic modification and the herbicide it is grown with), than the 90-day studies carried out by industry for regulatory approval of GM foods.