Threshold of toxicological concern for chemical substances present in the diet: A practical tool for assessing the need for toxicity testing

Abstract

The present paper was extensively discussed during a Workshop on “Threshold of Toxicological Concern for Chemical Substances Present in the Diet”, held in Paris on 5–6 October 1999. A report of this meeting will be published in the very near future. For more detailed information, please contact the corresponding author.The de minimis concept acknowledges a human exposure threshold value for chemicals below which there is no significant risk to human health. It is the underlying principle for the US Food and Drug Administration (FDA) regulation on substances used in food-contact articles. Further to this, the principle of Threshold of Toxicological Concern (TTC) has been developed and is now used by the Joint FAO/WHO Expert Committee on Food Additives (JECFA) in their evaluations. Establishing an accepted TTC would benefit consumers, industry and regulators, since it would preclude extensive toxicity evaluations when human intakes are below such threshold, and direct considerable time and cost resources towards testing substances with the highest potential risk to human health. It was questioned, however, whether specific endpoints that may potentially give rise to low-dose effects would be covered by such threshold.

In this review, the possibility of defining a TTC for chemical substances present in the diet was examined for general toxicity endpoints (including carcinogenicity), as well as for specific endpoints, namely neurotoxicity and developmental neurotoxicity, immunotoxicity and developmental toxicity. For each of these endpoints, a database of specific no-observed-effect levels (NOELs) was compiled by screening oral toxicity studies. The substances recorded in each specific database were selected on the basis of their demonstrated adverse effects. For the neurotoxicity and developmental neurotoxicity databases, it was intended to cover all classes of compounds reported to have either a demonstrated neurotoxic or developmentally neurotoxic effect, or at least, on a biochemical or pharmacological basis were considered to have a potential for displaying such effects. For the immunotoxicity endpoint, it was ensured that only immunotoxicants were included in the database by selecting most of the substances from the Luster et al. database, provided that they satisfied the criteria for immunotoxicity defined by Luster. For the developmental toxicity database, substances were selected from the Munro et al. database that contained the lowest NOELs retrieved from the literature for more than 600 compounds. After screening these, substances showing any effect which could point to developmental toxicity as broadly defined by the US EPA (1986) were recorded in the database.

Additionally, endocrine toxicity and allergenicity were addressed as two separate cases, using different approaches and methodology.

The distributions of NOELs for the neurotoxicity, developmental neurotoxicity and developmental toxicity endpoints were compared with the distribution of NOELs for non-specific carcinogenic endpoints.

As the immunotoxicity database was too limited to draw such a distribution of immune NOELs, the immunotoxicity endpoint was evaluated by comparing immune NOELs (or LOELs—lowest-observed-effect levels—when NOELs were not available) with non-immune NOELs (or LOELs), in order to compare the sensitivity of this endpoint with non-specific endpoints.

A different methodology was adopted for the evaluation of the endocrine toxicity endpoint since data currently available do not permit the establishment of a clear causal link between endocrine active chemicals and adverse effects in humans. Therefore, this endpoint was analysed by estimating the human exposure to oestrogenic environmental chemicals and evaluating their potential impact on human health, based on their contribution to the overall exposure, and their estrogenic potency relative to endogenous hormones.

`The allergenicity endpoint was not analysed as such. It was addressed in a separate section because this issue is not relevant to the overall population but rather to subsets of susceptible individuals, and allergic risks are usually controlled by other means (i.e. labelling) than the Threshold of Toxicological Concern approach. However, as several researchers are currently examining the existence of a threshold in allergy, the possibility of determining threshold doses for food allergens was put into perspective, and the likelihood for chemical substances to induce allergy at dietary relevant doses was discussed.

The analysis indicated that, within the limitation of the databases, developmental neurotoxicity and developmental toxicity were not more sensitive than other non-specific endpoints.

Although the cumulative distribution of NOELs for neurotoxic compounds was significantly lower than those for other non-cancer endpoints, these substances were accommodated within the TTC of 1.5 μg/person/day. Furthermore, the analysis demonstrated that none of the specific non-cancer endpoints evaluated in the present study was more sensitive than cancer and, that a TTC of 1.5 μg/person/day based on cancer endpoints provides an adequate margin of safety.

Analysis of the immunotoxicity database showed that for the group of immunotoxicants examined here, the specific immunotoxic endpoint was not more sensitive than other endpoints. In other words, the distribution of immunotoxic NOELs for these compounds did not appear to differ from the distribution of non-specific endpoints NOELs for the same compounds.

The dietary intakes of environmental oestrogenic chemicals were estimated and their oestrogenic potencies were compared with that of endogenous hormones, in order to assess their impact on human health. The results are in line with scientific data obtained so far, suggesting that estrogenic compounds of anthropogenic origin, in comparison with endogenous hormones, possess only little hormonal activity like phytoestrogens. Results of animal studies do not suggest that hormonal effects are to be expected from the rather low concentrations found in foods.

More data are necessary to determine threshold doses for food allergens. However, provided that numerous criteria need to be satisfied before sensitization occurs, it is unlikely that small molecules used in little amounts in foods would induce such reactions.

On the basis of the present analysis, which was conducted using conservative assumptions at each step of the procedure (i.e. in data compilation and data analysis), and continually adopting a “worst case” perspective, it can be concluded that a Threshold of Toxicological Concern of 1.5 μg/person/day provides adequate safety assurance. Chemical substances present in the diet that are consumed at levels below this threshold pose no appreciable risk.

Moreover, for compounds which do not possess structural alerts for genotoxicity and carcinogenicity, further analysis may indicate that a higher Threshold of Toxicological Concern may be appropriate.

Introduction

Humans are exposed to thousands of chemicals, either naturally occurring or man-made. This review has been prepared with the aim of assessing whether a generic threshold value or range of values can be established, which may preclude the need for extensive and normally expensive toxicity studies and safety evaluations when human intakes are below such established thresholds. Some general remarks are made concerning the world of chemicals, toxicity testing and safety evaluation, the principle of a threshold and differing risk assessment philosophies in different countries or regions. Special consideration is given to the possibility of defining a threshold of toxicological concern for general toxicity endpoints (including carcinogenicity) as well as for the specific endpoints neurotoxicity and developmental neurotoxicity, immunotoxicity and developmental toxicity. Additionally, attention is given to endocrine toxicity and allergenicity, which are also matters of concern when evaluating possible adverse reactions of substances.

It has been estimated that there are over five million man-made chemicals known, of which only a limited number (approximately 70,000) are in commercial use today (Beck et al., 1989). Furthermore, it has been reported that there are more than 100,000 naturally occurring substances of known structure, but it can be assumed that many more exist, the structures of which are not yet elucidated.

Humans are exposed to thousands of chemicals every day by different routes, each having their own barrier system. Effects through exposure by inhalation are determined by the properties of a chemical, for example its molecular weight, relative volatility and/or particle size. The chemicals that reach the alveolar space still have to pass the alveolar lining before entering the circulatory system of the body. Similarly, effects through dermal exposure to substances may—apart from chemicals reacting directly with the dermal epithelium—depend on the properties of a chemical to penetrate the skin. Finally, the entry of chemicals into the human body by oral exposure is dependent on the ability of the chemical or its breakdown products to pass the gastrointestinal lining and to enter the body via the lymph or portal venous system.

When considering substances migrating into food from packaging materials or other dietary sources, apart from the normal food constituents and the naturally occurring substances in food, particularly vegetables and fruits (NRC, 1996), a large number of food additives or many other substances are in use. Materials coming into contact with food may add another 3000 substances (Hayes and Campbell, 1986). A similar number of substances are used in flavourings. Evidently not all substances are present at the same time and, similarly, the levels of use and subsequently the levels of exposure for humans is extremely low for many substances. For example, the vast majority of food additives are present in small amounts, especially in regard to so-called “indirect food additives”. One does not have to argue that it is impossible to subject all the chemical substances to which humans are exposed to extensive toxicological testing. Furthermore, if sufficient facilities to perform such testing within a reasonable time were available, it still can be questioned whether testing of all these substances would be a rational and practical approach. Therefore, the establishment of a scientifically based generic threshold will be a useful tool to discern which substances of concern should be subjected to elaborate testing, when human intake is higher than the generic threshold.

In the past four decades toxicity testing has grown to maturity and today a systematic, usually tiered approach is used to establish whether adverse effects occur and if so, to investigate at what levels of exposure such adverse effects remain absent, and whether a dose–response relationship can be established. On the basis of these findings, a safety evaluation may be performed to assess at what levels of exposure humans may not experience any risk. The basis of such an evaluation is usually the established no-observed-effect level (NOEL) in animal testing (WHO, 1987) also referred to as an experimental threshold.

Although the word “threshold” in classical pharmacology is used to define a level above which a desired effect is seen, in toxicology a threshold is defined as a dose at, or below which, a response is not seen in an experimental setting. Establishing proof of absence of an effect at such a dose in absolute terms is scientifically and practically not feasible since only a limited number of experimental animals will be used for practical and economic reasons.

Additional means, which are used sometimes to establish toxicological thresholds, are mechanistic and biotransformation information. The threshold principle is based on the assumption that at or below that threshold, homeostasis is maintained. This is, in essence, true for almost all toxicological endpoints, with the exception of genotoxic carcinogens where, for regulatory purposes, it is often assumed that the threshold does not exist.

Although in many cases current risk assessment (RA) is based on relatively simple safety assessments, more advanced systems have been proposed and are also in use for specific endpoints. These new RA methodologies are directed to assess the quantitative risks rather than a qualitative assessment of risk, the latter leading to the provision of absence of an appreciable health risk.

For example, when the Joint FAO/WHO Expert Committee on Food Additives (JECFA) estimates ADIs (acceptable daily intakes) or TDIs (tolerable daily intakes), their evaluation is based on toxicological concern, which means considering the type of effects, the potency, and the relevance for humans. The ADI¹ is calculated from the NOEL² in animal studies, adjusted with a safety factor.

More than 30 years of practice with this qualitative approach has shown its usefulness for regulatory purposes. Safety evaluation according to the ADI concept can be applied for all kinds of toxicological effects for which a threshold is assumed. It does not provide a quantification of the risk. However, the period of time over which this method has been used without any manifestation of toxic effects arising from the use of compounds managed in accordance with the ADI concept suggests that the resulting actual risk is probably very low.

As stated earlier, quantitative RA has been developed in recent years and is used particularly in the RA of carcinogenic substances and methods have been recognized internationally. However, classification of carcinogens into two main categories—genotoxic and non-genotoxic substances—for the purpose of assessing risk is not harmonized internationally. In general, genotoxic carcinogens are regarded as non-threshold toxicants with risk at all dose levels, whereas non-genotoxic carcinogens are regarded as threshold toxicants with no risk at sufficiently low doses.

In quantitative RA of carcinogens which are genotoxic, several mathematical models have been proposed. Such models are usually based on the assumption that a linear relationship exists between the quantity of exposure and the response for the particular endpoint. The use of such quantitative cancer RA models has given cause for a number of generic and non-generic factors, which may increase the calculated cancer risk considerably due to using the worst case approach. For example, the policies concerned include among others factors weight vs surface area; maximum or average likelihood vs upper 95% confidence; malignant vs malignant plus benign tumours; average animal sensitivity vs most sensitive animal; pharmacodynamics vs effective dose. A combination of all the above default options might in the worst case increase the estimated cancer risk up to 10,000 times (Barnard, 1994). Thus, the estimated risk may be unrealistically higher by orders of magnitude and this may need further consideration.

Moolenaar (1994) has compared the methods and approaches used internationally in RA of carcinogens. He did not find any country in which estimated cancer risks systematically were placed in context with other risks. He concluded “The US EPA has established the estimation of an upper limit to carcinogen risk as a goal for risk assessment, while European counterparts have established the estimation of the likely incidence of cancer in the human population as the goal for risk assessment. The US approach depends heavily on conservative generic default assumptions to bridge areas of uncertainty and derive estimates of the upper limit to risk. The European approach favours case-by-case application of scientific judgements to resolve uncertainties and derive estimates of the expected incidence of cancer in the human population”.

The concept that there is a level of exposure to a given substance below which no significant risk is expected to exist has been widely accepted, and as depicted earlier, the establishment of ADIs is based on that concept (SCF (Scientific Committee for Food), 1996; WHO, 1987).

Frawley (1967) was the first to present an analysis to establish a generic threshold value (threshold of regulation) or range of values with the aim to reduce extensive toxicity studies and safety evaluations, and to address, within the available capacity, those substances for which the potential or actual intake is substantial. In 1986, Rulis conducted a similar analysis of the FDA's Priority-Based Assessment of Food Additives (PAFA) database containing 159 compounds with subchronic or chronic toxicity data, LD₅₀ values from 18,000 oral rodent studies contained in the Registry of Toxic Effects of Chemical Substances (RTECS), and TD₅₀ values for 130 compounds found in the carcinogen potency database of Gold et al. (1984). Both scientists concluded that an intake for humans between 1 and 10 μg/kg body weight/day of various chemical substances might not pose a risk to humans.

Munro et al. (1996) compiled a large database of reference substances from which a distribution of NOELs could be derived. This database includes structure and toxicity endpoints for a wide variety of organic chemicals, selected according to strict criteria. Three classes according to Cramer et al. (1978) were identified:

• Class I: Substances of simple chemical structure and efficient modes of metabolism which would suggest a low order of oral toxicity.

• Class II: “Intermediate” substances which possess structures that are less innocuous than class I substances but do not contain structural features suggestive of toxicity like those substances in class III.

• Class III: Substances of a chemical structure that permits no strong initial presumption of safety or may even suggest significant toxicity or have reactive functional groups.

The database includes the usual types of toxicity studies (subchronic, chronic, reproductive and teratology studies) used for evaluation. The database mainly consists of studies in rodents and rabbits. Dog and monkey studies were not included since many had too few animals per group to derive statistically valid NOELs, were too short in duration or were hampered by insufficiencies originating from palatability problems of the diet. For each substance with a defined NOEL, an LOEL was also included. When chronic NOEL data were not available, they were compiled from the NOEL data from subchronic studies. These were divided by a factor of three, based on a retrospective analysis performed by others (Beck et al., 1993; Lewis et al., 1990; Weil and McCollister, 1963). NOELs selected were the NOELs suggested by the authors, even though they were in a number of cases over-interpretations of their data. However, obvious misjudgements were corrected.

The database described above was used to calculate human exposure thresholds for the three structural classes identified, using the 5^th centile of the distributions of NOELs divided by an uncertainty factor of 100. Human exposure levels were respectively 1800, 540 and 88 μg/person/day for classes I, II and III.

The procedure described above was also the basis for a procedure for safety evaluation of flavouring substances described by Munro and Kroes (1998) and Munro et al. (1999). In this procedure, in addition to the derivation of human exposure limits by class, other endpoints including cancer were evaluated. For the cancer endpoint, the carcinogenic potency database (CPD) as originally compiled by Gold et al. (1984) and further updated (Gold et al., 1989) was used. In a workshop organized by Munro (1990), factors that influence the selection of an appropriate threshold value for carcinogens were evaluated. It was concluded that when using a threshold of 1.5 μg/person/day, the probability of exceeding a risk of 10⁻⁶ for new chemicals entering the database was low, especially when it was assumed that less than 50% of such new chemicals would be potential carcinogens. Eliminating substances with structural alerts for carcinogenicity would reduce the probability considerably (Ashby and Tennant, 1988, Tennant and Ashby, 1991).

In the papers of Munro and Kroes (1998) and Munro et al. (1999), other selected endpoints such as reproductive effects and neurotoxicity were assessed as well in the Munro database, and immunotoxicity was assessed by evaluating a limited number of immunotoxic chemicals from the Luster database (Luster et al., 1992, Luster et al., 1993). For developmental abnormalities and neurotoxic compounds, human exposure threshold values were again calculated using the 5^th centile NOEL divided by an uncertainty factor of 100, and assuming an average individual weight of 60 kg per person. The various human exposure threshold values are given below:

Various human exposure threshold values (Munro and Kroes, 1998)

For immunotoxicity, an insufficient number of substances was available to calculate a reliable 5^th centile NOEL. Therefore, a comparison was made of non-immunotoxic NOEL vs immunotoxic NOEL, and similarly for the LOEL. It was noticed that usually the non-immunotoxic NOEL and LOEL were lower but that in the cases when immunotoxic NOELs (and LOELs) were lower, this was less than 10-fold lower than the non-immunotoxic counterpart.

The concern was raised by others that a derivation of the 5^th centile NOEL based on the currently used databases drawn from acute, subchronic and chronic studies in rodents may not adequately cover endpoints which might give rise to important low dose effects such as neurotoxic, immunotoxic, endocrinologic and developmentally toxic events (SCF (Scientific Committee for Food, 1996). Therefore, it was decided to undertake the present study to further assess human exposure threshold values for the specific endpoints neurotoxicity and developmental neurotoxicity, immunotoxicity and developmental toxicity, and to examine these endpoints in order to determine whether changes in the different parameters of these specific systems would occur at particularly low levels of exposure, and what these levels would be. The approach involved the collection of data pertaining to NOELs for these various endpoints, and the assessment as to whether the distributions for the various NOELs differed significantly from the distributions for general toxicity as described by Munro et al. (1996, 1999).

In addition, special attention has been given to endocrine effects and allergenicity which are presently major subjects of public concern. Both issues are matters to be considered when evaluating possible adverse reactions to chemical substances, but it was decided to address them as separate cases.

Usually, traditional reproductive studies cover endocrine substances, but since endocrine toxicity has become a public concern and scientific claims have been raised about the potential oestrogenic activity of certain environmental chemicals, a specific evaluation of this endpoint was undertaken. A different methodology was adopted for evaluating the endocrine toxicity endpoint since currently available data do not permit the establishment of a clear causal link between endocrine disrupting chemicals and adverse effects in humans. No specific database on endocrine toxicity was built up because it was not possible to compile substances on the basis of their demonstrated specific adverse effect, as was done for the other endpoints. Therefore, the approach adopted in this case was to estimate the possible human exposure to environmental oestrogenic chemicals, to relate their potencies to that of endogenous hormones and assess the potential impact of oestrogenic environmental chemicals on human health.

Allergenicity was confined in a separate section of the paper because this issue is not relevant to the overall population but rather to subsets of susceptible individuals within the population. As a consequence, allergic risks are usually controlled by other means (i.e. labelling) than the Threshold of Toxicological Concern approach. However, as several researchers are currently examining the existence of a threshold in allergy, it was felt important to put this case into the context of our evaluation.