Scientific requirements for the assessment of food safety
Ad van Dommelen
Keywords:  Organisation for Economic Co-operation and Development (OECD); Biosafety/Foodsafety; Genetic engineering.
Correct citation: Dommelen, A. van (1999), "Scientific Requirements for the Assessment of Food Safety." Biotechnology and Development Monitor, No. 38, p. 3-7.

Recently, the unpredicted outcomes of several risk assessment studies have further raised concerns about the effects of practical applications of genetically engineered organisms (GEOs). Since the relevance of these findings is still controversial, this article argues for the systematic development of relevant research questions for food safety assessment.

The biological safety of genetically engineered products concerns possible unwanted effects on living organisms and the environment in general. For the impact on humans, the assessment of products intended for dietary purposes is of prior importance. Although safety assessment of GEOs has been a contentious issue over the past 25 years, all concerned would probably agree that a satisfactory testing procedure is necessary. The issue then becomes: what is a satisfactory procedure?
The knowledge and insights required for the process of genetic engineering does not automatically provide a sufficient basis for the safety assessment of its products. In fact, developing satisfactory safety tests may be even more time-consuming than the technological feat of producing a viable GEO. From a scientific point of view, the most reliable method would probably be to proceed with using the new products and see what happens over time. It may well be that most of the transgenic food and feed products have no detrimental effect on their consumers. However, the point of testing is to detect in advance the small percentage of genetically engineered products that may pose a problem.
Expensive time and resources have been wasted by not sufficiently taking into account the demands of scientific research methodology. In many cases confined testing such as in field experiments or laboratory tests does not answer relevant questions, because these tests have not been designed to do so.
For instance, a safety test was performed to identify possible hazards of the use of transgenic Klebsiella planticola, a soil bacterium that was genetically modified for an enhanced ethanol production. In a microcosm of wheat plants grown on sterile soil, no effect was observed, indicating that transgenic Klebsiella planticola does not directly affect wheat plants. However, in an alternative experimental setting that used natural soil, it was demonstrated that the transgenic bacteria will affect the microbial environment of the plants’ roots, which in turn will affect the wheat. This alternative test included more potentially relevant research questions and thereby identified this hazardous impact of the transgenic bacteria.
As this example makes clear, if a hazard has not been identified all other steps in the process of risk assessment (see box) will become useless. Therefore, generating the relevant research questions is important for hazard identification. As it is, the debate is riddled with conveniently vague terms that mix science and policy such as ‘analogy’, ‘similarity’, ‘comparison’, ‘acceptable evidence’, ‘reasonable certainty’, ‘familiarity’, and ‘substantial equivalence’. In practice, these terms may turn out to bring into the debate more meaning than is justified by their scientific basis.

The sequence of risk assessment
  • identification of hazards: to recognize and characterize possible unwanted impacts;
  • estimation of their magnitude: to assess the scale and severity of a possible unwanted impact;
  • estimation of their likelyhood: to assess the chance of a specific unwanted impact to occur;
  • quantification of risk: risk is defined as ‘hazard times the likelyhood of its occurrence’;
  • evaluation of risk: to assess the importance of some estimated risk against the background of possible costs and benefits.

Finding relevant questions for safety testing
One example that has recently received broad media attention was the study conducted by Arpad Pusztai, who was a senior scientist of the Rowett Research Institute (RRI) in Aberdeen, Scotland. Pusztai’s study raised concern about the effects of lectins produced in genetically modified (GM) potatoes (see box). The story illustrates the complexity of studying food safety; something may be wrong, but it is not as yet clear what it is. Furthermore, these experiments do not by themselves uncover the biological mechanisms that are responsible for the effects found.
Although Pusztai’s results are debatable from a scientific point of view, in terms of the general methodology for testing food safety, an important lesson from this episode is that the scientific community is still in the process of finding relevant questions for the hazard identification of genetically engineered food products.
A research problem can be defined by specifying a set of relevant questions. Experiments that produce unpredicted effects, as in the case of the RRI research on genetically modified potatoes, are more interesting from a methodological perspective than experiments that produce no observed effect. The former give an outlook on relevant research questions, while the latter may represent the failure to address relevant research questions.
One set of relevant questions for future research derived from the RRI experiments should address the susceptibility of the gut ecology of mammals to lectins. For hazard identification, it might be of importance that lectins survive the passage through the digestive tract in an intact form. If this were the case, it would have to be investigated whether this is the result of the molecular structure of the lectins.

Methodological basis of scientific evidence
A test is a tool for answering questions. The results of a test will always depend on the specific questions that were included in the testing procedure. Thus, the challenge of designing satisfactory tests for food safety lies in finding relevant research questions for the hazard identification of genetically engineered food products. Any question that is not explicitly incorporated into the test will thereby implicitly be excluded from consideration. The design of a satisfactory safety test thus comes with a methodological burden of proof. Since testing is costly and time consuming, any research question that makes up part of a specific test must be included on the basis of its relevance.
On the other hand, when a potentially relevant research question is excluded, this decision also bears a burden of proof regarding its irrelevance to the research purpose. Excluding a potentially relevant question from the test, and thus from investigation, could lead to approval for the application of a genetically engineered food product that may prove to present a hazard if it were tested using an alternative procedure.
The burden of proof for claims on safety is preceded by the burden of proof for claims on sufficient knowledge, which again can be discussed in terms of the relevant research questions. For example, if a conventional food product receives the predicate generally recognized as safe (GRAS), this implies that a list of research questions has been addressed to arrive at this conclusion. Empirical findings of a certain test can only be interpreted against the background of the questions the test was meant to address.
There are no ready-made theories that could provide a basis for safety testing of genetically modified food. Previous safety testing of chemical substances and drugs can help but might not be sufficient in this new field of safety assessment. An important methodological concern in experiments for food safety of GEOs is related to the model organism that is used for testing. In using model organisms, a translation is being made from one set of relevant questions to another one. This transition too comes with a burden of proof. The inference from one species to another must be supported by arguments about the respective relevance of biological mechanisms and the associated relevant research questions for the purpose of investigation.For instance, results from experiments with rats may be more conclusive of the toxicity of substances or food products to human consumers than they are for the assessment of allergenicity. The latter may be hard to detect due to the difficulties to address the relevant effects in rats.
Unwanted effects such as toxicity may depend on a number of factors. A preliminary definition of toxicity would be: substance A is toxic to organism B in circumstance C. This implies that one cannot assess the toxicity of a substance without reference to the recipient and to the circumstances. A substance that is not toxic to adults may be toxic to children. It may also be non-toxic to healthy individuals, but toxic to individuals with a decreased resistance. Additional concerns may be in the time scale (it may be non-toxic over shorter periods of time, but toxic in the long run), or in frequency or quantity of use (incidental use in smaller quantities may be quite harmless). The same general methodological considerations apply to the study of allergenicity.
Scientific models and the food safety tests based upon these models are our ‘scientific sensory organs’, which would allow for the timely detection of a possible adverse reaction. However, our knowledge of the structure of food allergens and toxicants and the mechanisms involved is still poor. Adverse reactions are a research problem in their own right, from which a satisfactory test for food safety will develop with increasing insight into the underlying mechanisms.

Transgenic potatoes and lectins
Lectins are a specific kind of protein that occur in all plant nuts, seeds and bulbs, including many foodstuffs. Some lectins, such as those in red kidney beans, are toxic to humans and need to be destroyed by heat before consumption. Others, such as tomato lectin, are apparently harmless when eaten raw.
Several lectins are considered to be toxic to the gut flora of insects and could therefore potentially be used as natural insecticides. However, their potential harm to human consumers has first to be ruled out. Research at the RRI focused on the snowdrop (Galanthus nivalis) bulb lectin (GNA) which is considered to be toxic to insects, but not to humans. 
For this investigation potatoes were genetically modified to express the gene of GNA. These transgenic potatoes were developed for research purposes and not as a food crop. The task of the RRI was to analyse whether the parent and transgene line were of the same chemical composition. Furthermore, rats were fed with transgenic and non-transgenic tubers to determine whether the genetically modified potatoes affected the mammalian gut and metabolism.
Rats that were fed in the experiments with high doses of genetically engineered potatoes expressing GNA showed deviations in their immune systems and internal organs. After publicly raising concern about these findings in August 1998, Pusztai was urged to retire. An internal Audit Committee at the RRI doubted the significance of the results due to methodological inconsistencies of the research. 
The debate was revived in February 1999 when an international group of researchers defended Pusztai’s findings. In May 1999, however, the UK Royal Society published another critical review of the data on possible toxicity of the GM potatoes. The review claimed that Pusztai’s findings lacked significance mainly because of: 
  • uncertainty about the differences in chemical composition between strains of non-GM and GM potatoes;
  • possible dietary differences due to non-systematic dietary enrichment. GM potatoes contained 20 per cent less proteins than unmodified potatoes. Rats that were fed GM potatoes therefore had to be given additional proteins to avoid starvation; 
  • the small quantity of animals tested with several diets, all of which were non-standard diets for rats. This resulted in multiple comparisons with little statistical significance;
  • poor experimental design, lacking ‘blind’ measurements. It is general practice for feeding trials of this kind that the scientists making the measurements are not aware of how the animals have been treated to prevent unintentionally biased results; 
  • application of inappropriate statistical techniques in the analysis of results;
  • lack of consistency of findings within and between experiments. For instance, some results showed differences in body weights and weights of organs between rats that were fed with GM-potatoes and those that were fed with non-GM potatoes. However, these differences did not show any discernible pattern.
The review concluded that "Although we have no evidence of harmful effects from genetic modification, this of course does not mean that harmful effects can be categorically ruled out." Pusztai himself emphasized that his findings were preliminary and should be understood as a stimulus for further testing. 

Difficulties in the interpretation of ‘substantial equivalence’
To bring some unity and clarity to food safety regulation and policy, the Organisation for Economic Co-operation and Development (OECD) has introduced the notion of ‘substantial equivalence’: "The concept of substantial equivalence embodies the idea that existing organisms used as food, or as a source of food, can be used as the basis for comparison when assessing the safety of human consumption of a food or food component that has been modified or is new." This is a pivotal concept, since: "If the new or modified food or food component is determined to be substantially equivalent to an existing food, then further safety or nutritional concerns are expected to be insignificant." (OECD 1993)
However, the OECD report itself demonstrates the likelihood of scientific controversy in the use of the notion of ‘substantial equivalence’. In the first case study that is presented to demonstrate the applicability of the suggested approach, the concerned researcher, studying chymosin derived from transgenic Escherichia coli K-12 bacteria, immediately raises the problem: "There is a good deal of scientific consensus on how to assess the safety of an enzyme preparation. However, there is less consensus regarding criteria by which one decides at what point an enzyme preparation is different enough from an accepted one that formal review is required to establish safety. For example, at what point do manufacturing changes or strain modifications become significant enough to warrant review? At what point is the substantial equivalence of two enzyme preparations no longer self-evident? This is as much a regulatory question as a scientific one."
The OECD report tries to elaborate this for the assessment of enzymes prepared from genetically modified microbes: they "can be considered substantially equivalent to each other if three conditions are met: the enzymes themselves are substantially equivalent, for example having similar intended uses and functional properties; the microbes from which they are derived are substantially equivalent, for example being safe strains of species with a safe history of use as sources of food-enzymes; and the manufacturing and purification processes are substantially equivalent." However, the practical use of this specification remains limited since there are no agreed-upon criteria by which substantial equivalence is determined for each of these parameters, and the OECD report concluded that "in the safety evaluations of the two enzyme preparations (...), the term ‘substantial equivalence’ was nowhere used." The criterion of ‘substantial equivalence’ is scientifically void unless it is specified in terms of the sets of relevant questions that are required to assess any comparison or equivalence. It is impossible to assess equivalence without a clear view on the possible differences. On the other hand, if the evaluations were made on the basis of specified sets of relevant research questions, then there is not much use for the notion of ‘substantial equivalence’.
Pusztai concluded in a report of October 1998 that, in terms of their chemical composition, "the GNA-GM-potato lines investigated (...) were not ‘substantially equivalent’ to the appropriate parent tubers." The challenge for future research would be to investigate whether this lack of ‘substantial equivalence’ in chemical composition relates to any harmful dietary effects the potato may have.
The same challenge applies to the general assessment of food safety. There is an ongoing debate between those who propagate safety regulations based on the products of genetic engineering as in the USA on the one hand and those who propagate safety regulations based on the processes of genetic engineering as it is done in Europe. In fact, this conflict boils down to an artificial controversy, because both the products and the processes of genetic engineering may raise relevant safety concerns that should be specified in explicit sets of relevant questions. Given the present limited experiences with research on the safety of foods derived by genetic engineering, it would be premature to restrict the decision on either the product or the process only.

Towards an international research programme
To make ‘substantial equivalence’ a useful concept in a regulatory context, it must be embedded in a developing body of knowledge about the relevant biological concerns. ‘Substantial equivalence’ can only be made specific by adaptation of the relevant research questions to specific cases. To facilitate this learning process, different categories of possibly relevant research questions may be distinguished by a provisional definition of adverse reactions to food: ‘Food (component) F’ may cause ‘adverse reaction G’ if an ‘agent K’ can cause an ‘effect P’ to a ‘consumer Q’ via ‘mechanism R’ under ‘conditions S’ in ‘environment T’. The elements of this provisional definition constitute categories of possibly relevant research questions for the assessment of food safety.
The process of identifying relevant research questions for food safety testing and monitoring could be greatly enhanced by an accumulating inventory of possible research questions with arguments for their relevance. Such an inventory could use an interactive homepage on the internet for experts and consumers from all over the world to put forward their concerns and findings. A number of internet sites is now dedicated to the challenges of food safety. However, since they lack a uniform format in which individual research questions and their context of relevance is clarified, the interactive use of these sites and the comparison of arguments is limited.
A more unified format for an integrating internet site on food safety would be to distinguish categories of possibly relevant research questions. These categories could be found and be proposed on this internet site, together with arguments for their relevance or irrelevance in a specified context. This would also be a convenient format for countries that have fewer resources to dedicate to this important issue.
Ad van Dommelen

Houtsma & van der Schot, Office for research, consultancy and journalism, Keizersgracht 8, 1015 CN Amsterdam, the Netherlands.
Phone (+31) 20 623 94 80; Fax (+31) 20 638 87 71; E-mail Hout.Schot@inter.nl.net

Dommelen, A. van (1998), "Useful Models for Biotechnology Hazard Identification: What is this Thing called ‘Familiarity’?" In: P. Wheale et al. (eds.), The Social Management of Genetic Engineering, Aldershot, UK: Ashgate, pp. 219-236.

Dommelen, A. van (1999), Hazard Identification of Agricultural Biotechnology: Finding relevant questions, Utrecht, the Netherlands: International Books.

Metcalfe, D. et al. (1996), "Assessment of the Allergenic Potential of Foods Derived from Genetically Engineered Crop Plants", Critical Reviews in Food Science and Nutrition, 36 (S), pp. 165-186.

OECD (1993), Safety Evaluation of foods derived by modern biotechnology: concepts and principles, Paris, France: OECD.

Pusztai, A. and Bardocz, S. (1996), "Biological Effects of Plant Lectins on the Gastrointestinal Tract: Metabolic consequences and Applications", Trends in Glycoscience and Glycotechnology 8, pp. 149-165.

Contributions to the Biotechnology and Development Monitor are not covered by any copyright. Exerpts may be translated or reproduced without prior permission (with exception of parts reproduced from third sources), with acknowledgement of source.


back to top
monitor homepage
index of this issue
contact us