Address: World Health Organization European Centre for Environment and Health, A. van Leewenhoeklaan 9, PO Box 10, 3730 AA De Biltoven
Past position(s): Director, Toxicology, National Institute for Public Health and Environment
Areas of interest: Environmental health
Whereas the principles and methods of risk assessment for non-carcinogenic chemicals are similar in different parts of the world, it is striking that approaches for risk assessment of carcinogenic chemicals vary greatly. There are not only marked differences between countries, but even within a country different approaches are applied or advocated by various regulatory agencies, committees and scientists in the field of risk assessment. Risk assessment for non-carcinogens is rather consistent and pretty well established partly because of the long history and better understanding of the nature of toxic effects in comparison with carcinogens and a high degree of consensus and confidence by both scientists and the general public on methods used and their outcome.
For non-carcinogenic chemicals, safety factors were introduced to compensate for uncertainties in the toxicology data (which are derived mostly from animal experiments) and in their applicability to large, heterogeneous human populations. In doing so, recommended or required limits on safe human exposures were usually set at a fraction (the safety or uncertainty factor approach) of the exposure levels in animals that could be clearly documented as the no observed adverse effects level (NOAEL) or the lowest observed adverse effects level (LOAEL). It was then assumed that as long as human exposure did not exceed the recommended limits, the hazardous properties of chemical substances would not be manifest. For many types of chemicals, this practice, in somewhat refined form, continues to this day in toxicological risk assessment.
During the late 1960s and early 1970s regulatory bodies, starting in the United States, were confronted with an increasingly important problem for which many scientists considered the safety factor approach to be inappropriate, and even dangerous. This was the problem with chemicals that under certain conditions had been shown to increase the risk of cancers in humans or experimental animals. These substances were operationally referred to as carcinogens. There is still debate and controversy on the definition of a carcinogen, and there is a wide range of opinion about techniques to identify and classify carcinogens and the process of cancer induction by chemicals as well.
The initial discussion started much earlier, when scientists in the 1940s discovered that chemical carcinogens caused damage by a biological mechanism that was of a totally different kind from those that produced other forms of toxicity. These scientists, using principles from the biology of radiation-induced cancers, put forth what is referred to as the “non-threshold” hypothesis, which was considered applicable to both radiation and carcinogenic chemicals. It was hypothesized that any exposure to a carcinogen that reaches its critical biological target, especially the genetic material, and interacts with it, can increase the probability (the risk) of cancer development.
Parallel to the ongoing scientific discussion on thresholds, there was a growing public concern on the adverse role of chemical carcinogens and the urgent need to protect the people from a set of diseases collectively called cancer. Cancer, with its insidious character and long latency period together with data showing that cancer incidences in the general population were increasing, was regarded by the general public and politicians as a matter of concern that warranted optimal protection. Regulators were faced with the problem of situations in which large numbers of people, sometimes nearly the entire population, were or could be exposed to relatively low levels of chemical substances (in consumer products and medicines, at the workplace as well as in air, water, food and soils) that had been identified as carcinogenic in humans or experimental animals under conditions of relatively intense exposures.
Those regulatory officials were confronted with two fundamental questions which, in most cases, could not be fully answered using available scientific methods:
Regulators recognized the need for assumptions, sometimes scientifically based but often also unsupported by experimental evidence. In order to achieve consistency, definitions and specific sets of assumptions were adapted that would be generically applied to all carcinogens.
Carcinogenesis Is a Multistage Process
Several lines of evidence support the conclusion that chemical carcinogenesis is a multistage process driven by genetic damage and epigenetic changes, and this theory is widely accepted in the scientific community all over the world (Barrett 1993). Although the process of chemical carcinogenesis is often separated into three stages—initiation, promotion and progression—the number of relevant genetic changes is not known.
Initiation involves the induction of an irreversibly altered cell and is for genotoxic carcinogens always equated with a mutational event. Mutagenesis as a mechanism of carcinogenesis was already hypothesized by Theodor Boveri in 1914, and many of his assumptions and predictions have subsequently been proven to be true. Because irreversible and self-replicating mutagenic effects can be caused by the smallest amount of a DNA-modifying carcinogen, no threshold is assumed. Promotion is the process by which the initiated cell expands (clonally) by a series of divisions, and forms (pre)neoplastic lesions. There is considerable debate as to whether during this promotion phase initiated cells undergo additional genetic changes.
Finally in the progression stage “immortality” is obtained and full malignant tumours can develop by influencing angiogenesis, escaping the reaction of the host control systems. It is characterized by invasive growth and frequently metastatic spread of the tumour. Progression is accompanied by additional genetic changes due to the instability of proliferating cells and selection.
Therefore, there are three general mechanisms by which a substance can influence the multistep carcinogenic process. A chemical can induce a relevant genetic alteration, promote or facilitate clonal expansion of an initiated cell or stimulate progression to malignancy by somatic and/or genetic changes.
Risk Assessment Process
Risk can be defined as the predicted or actual frequency of occurrence of an adverse effect on humans or the environment, from a given exposure to a hazard. Risk assessment is a method of systematically organizing the scientific information and its attached uncertainties for description and qualification of the health risks associated with hazardous substances, processes, actions or events. It requires evaluation of relevant information and selection of the models to be used in drawing inferences from that information. Further, it requires explicit recognition of uncertainties and appropriate acknowledgement that alternative interpretation of the available data may be scientifically plausible. The current terminology used in risk assessment was proposed in 1984 by the US National Academy of Sciences. Qualitative risk assessment changed into hazard characterization/identification and quantitative risk assessment was divided into the components dose-response, exposure assessment and risk characterization.
In the following section these components will be briefly discussed in view of our current knowledge of the process of (chemical) carcinogenesis. It will become clear that the dominant uncertainty in the risk assessment of carcinogens is the dose-response pattern at low dose levels characteristic for environmental exposure.
This process identifies which compounds have the potential to cause cancer in humans—in other words it identifies their intrinsic genotoxic properties. Combining information from various sources and on different properties serves as a basis for classification of carcinogenic compounds. In general the following information will be used:
Classification of chemicals into groups based on the assessment of the adequacy of the evidence of carcinogenesis in animals or in man, if epidemiological data are available, is a key process in hazard identification. The best known schemes for categorizing carcinogenic chemicals are those of IARC (1987), EU (1991) and the EPA (1986). An overview of their criteria for classification (e.g., low-dose extrapolation methods) is given in table 1.
Table 1. Comparison of low-dose extrapolations procedures
|Current US EPA||Denmark||EEC||UK||Netherlands||Norway|
|Genotoxic carcinogen||Linearized multistage procedure using most appropriate low-dose model||MLE from 1- and 2-hit models plus judgement of best outcome||No procedure specified||No model, scientific expertise and judgement from all available data||Linear model using TD50 (Peto method) or “Simple Dutch Method” if no TD50||No procedure specified|
|Non-genotoxic carcinogen||Same as above||Biologically-based model of Thorslund or multistage or Mantel-Bryan model, based on tumour origin and dose-response||Use NOAEL and safety factors||Use NOEL and safety factors to set ADI||Use NOEL and safety factors to set ADI|
One important issue in classifying carcinogens, with sometimes far-reaching consequences for their regulation, is the distinction between genotoxic and non-genotoxic mechanisms of action. The US Environmental Protection Agency (EPA) default assumption for all substances showing carcinogenic activity in animal experiments is that no threshold exists (or at least none can be demonstrated), so there is some risk with any exposure. This is com- monly referred to as the non-threshold assumption for genotoxic (DNA-damaging) compounds. The EU and many of its members, such as the United Kingdom, the Netherlands and Denmark, make a distinction between carcinogens that are genotoxic and those believed to produce tumours by non-genotoxic mechanisms. For genotoxic carcinogens quantitative dose-response estimation procedures are followed that assume no threshold, although the procedures might differ from those used by the EPA. For non-genotoxic substances it is assumed that a threshold exists, and dose-response procedures are used that assume a threshold. In the latter case, the risk assessment is generally based on a safety factor approach, similar to the approach for non-carcinogens.
It is important to keep in mind that these different schemes were developed to deal with risk assessments in different contexts and settings. The IARC scheme was not produced for regulatory purposes, although it has been used as a basis for developing regulatory guidelines. The EPA scheme was designed to serve as a decision point for entering quantitative risk assessment, whereas the EU scheme is currently used to assign a hazard (classification) symbol and risk phrases to the chemical's label. A more extended discussion on this subject is presented in a recent review (Moolenaar 1994) covering procedures used by eight governmental agencies and two often-cited independent organizations, the Inter- national Agency for Research on Cancer (IARC) and the American Conference of Governmental Industrial Hygienists (ACGIH).
The classification schemes generally do not take into account the extensive negative evidence that may be available. Also, in recent years a greater understanding of the mechanism of action of carcinogens has emerged. Evidence has accumulated that some mechanisms of carcinogenicity are species-specific and are not relevant for man. The following examples will illustrate this important phenomenon. First, it has been recently demonstrated in studies on the carcinogenicity of diesel particles, that rats respond with lung tumours to a heavy loading of the lung with particles. However, lung cancer is not seen in coal miners with very heavy lung burdens of particles. Secondly, there is the assertion of the nonrelevance of renal tumours in the male rat on the basis that the key element in the tumourgenic response is the accumulation in the kidney of α-2 microglobulin, a protein that does not exist in humans (Borghoff, Short and Swenberg 1990). Disturbances of rodent thyroid function and peroxisome proliferation or mitogenesis in the mouse liver have also to be mentioned in this respect.
This knowledge allows a more sophisticated interpretation of the results of a carcinogenicity bioassay. Research towards a better understanding of the mechanisms of action of carcinogenicity is encouraged because it may lead to an altered classification and to the addition of a category in which chemicals are classified as not carcinogenic to humans.
Exposure assessment is often thought to be the component of risk assessment with the least inherent uncertainty because of the ability to monitor exposures in some cases and the availability of relatively well-validated exposure models. This is only partially true, however, because most exposure assessments are not conducted in ways that take full advantage of the range of available information. For that reason there is a great deal of room for improving exposure distribution estimates. This holds for both external as well as for internal exposure assessments. Especially for carcinogens, the use of target tissue doses rather than external exposure levels in modelling dose-response relationships would lead to more relevant predictions of risk, although many assumptions on default values are involved. Physiologically based pharmacokinetic (PBPK) models to determine the amount of reactive metabolites that reaches the target tissue are potentially of great value to estimate these tissue doses.
The dose level or exposure level that causes an effect in an animal study and the likely dose causing a similar effect in humans is a key consideration in risk characterization. This includes both dose-response assessment from high to low dose and interspecies extrapolation. The extrapolation presents a logical problem, namely that data are being extrapolated many orders of magnitude below the experimental exposure levels by empirical models that do not reflect the underlying mechanisms for carcinogenicity. This violates a basic principle in fitting of empirical models, namely not to extrapolate outside the range of the observable data. Therefore, this empirical extrapolation results in large uncertainties, both from a statistical and from a biological point of view. At present no single mathematical procedure is recognized as the most appropriate one for low-dose extrapolation in carcinogenesis. The mathematical models that have been used to describe the relation between the administered external dose, the time and the tumour incidence are based on either tolerance-distribution or mechanistic assumptions, and sometimes based on both. A summary of the most frequently cited models (Kramer et al. 1995) is listed in table 2.
Table 2. Frequently cited models in carcinogen risk characterization
|Tolerance distribution models||Mechanistic models|
|Hit-models||Biologically based models|
|Probit||Multihit||Cohen and Ellwein|
|Gamma Multihit||Linearized Multistage,|
1 Time-to-tumour models.
These dose-response models are usually applied to tumour-incidence data corresponding to only a limited number of experimental doses. This is due to the standard design of the applied bioassay. Instead of determining the complete dose-response curve, a carcinogenicity study is in general limited to three (or two) relatively high doses, using the maximum tolerated dose (MTD) as highest dose. These high doses are used to overcome the inherent low statistical sensitivity (10 to 15% over background) of such bioassays, which is due to the fact that (for practical and other reasons) a relatively small number of animals is used. Because data for the low-dose region are not available (i.e., cannot be determined experimentally), extrapolation outside the range of observation is required. For almost all data sets, most of the above-listed models fit equally well in the observed dose range, due to the limited number of doses and animals. However, in the low-dose region these models diverge several orders of magnitude, thereby introducing large uncertainties to the risk estimated for these low exposure levels.
Because the actual form of the dose-response curve in the low-dose range cannot be generated experimentally, mechanistic insight into the process of carcinogenicity is crucial to be able to discriminate on this aspect between the various models. Comprehensive reviews discussing the various aspects of the different mathematical extrapolation models are presented in Kramer et al. (1995) and Park and Hawkins (1993).
Besides the current practice of mathematical modelling several alternative approaches have been proposed recently.
Biologically motivated models
Currently, the biologically based models such as the Moolgavkar-Venzon-Knudson (MVK) models are very promising, but at present these are not sufficiently well advanced for routine use and require much more specific information than currently is obtained in bioassays. Large studies (4,000 rats) such as those carried out on N-nitrosoalkylamines indicate the size of the study which is required for the collection of such data, although it is still not possible to extrapolate to low doses. Until these models are further developed they can be used only on a case-by-case basis.
Assessment factor approach
The use of mathematical models for extrapolation below the experimental dose range is in effect equivalent to a safety factor approach with a large and ill-defined uncertainty factor. The simplest alternative would be to apply an assessment factor to the apparent “no effect level”, or the “lowest level tested”. The level used for this assessment factor should be determined on a case-by-case basis considering the nature of the chemical and the population being exposed.
Benchmark dose (BMD)
The basis of this approach is a mathematical model fitted to the experimental data within the observable range to estimate or interpolate a dose corresponding to a defined level of effect, such as one, five or ten per cent increase in tumour incidence (ED01, ED05, ED10). As a ten per cent increase is about the smallest change that statistically can be determined in a standard bioassay, the ED10 is appropriate for cancer data. Using a BMD that is within the observable range of the experiment avoids the problems associated with dose extrapolation. Estimates of the BMD or its lower confidence limit reflect the doses at which changes in tumour incidence occurred, but are quite insensitive to the mathematical model used. A benchmark dose can be used in risk assessment as a measure of tumour potency and combined with appropriate assessment factors to set acceptable levels for human exposure.
Threshold of regulation
Krewski et al. (1990) have reviewed the concept of a “threshold of regulation” for chemical carcinogens. Based on data obtained from the carcinogen potency database (CPDB) for 585 experiments, the dose corresponding to 10-6 risk was roughly log-normally distributed around a median of 70 to 90 ng/kg/d. Exposure to dose levels greater than this range would be considered unacceptable. The dose was estimated by linear extrapolation from the TD50 (the dose inducing toxicity is 50% of the animals tested) and was within a factor of five to ten of the figure obtained from the linearized multistage model. Unfortunately, the TD50 values will be related to the MTD, which again casts doubt on the validity of the measurement. However the TD50 will often be within or very close to the experimental data range.
Such an approach as using a threshold of regulation would require much more consideration of biological, analytical and mathematical issues and a much wider database before it could be considered. Further investigation into the potencies of various carcinogens may throw further light onto this area.
Objectives and Future of CarcinogenRisk Assessment
Looking back to the original expectations on the regulation of (environmental) carcinogens, namely to achieve a major reduction in cancer, it appears that the results at present are disappointing. Over the years it became apparent that the number of cancer cases estimated to be produced by regulatable carcinogens was disconcertingly small. Considering the high expectations that launched the regulatory efforts in the 1970s, a major anticipated reduction in the cancer death rate has not been achieved in terms of the estimated effects of environmental carcinogens, not even with ultraconservative quantitative assessment procedures. The main characteristic of the EPA procedures is that low-dose extrapolations are made in the same way for each chemical regardless of the mechanism of tumour formation in experimental studies. It should be noted, however, that this approach stands in sharp contrast to approaches taken by other governmental agencies. As indicated above, the EU and several European governments—Denmark, France, Germany, Italy, the Netherlands, Sweden, Switzerland, UK—distinguish between genotoxic and non-genotoxic carcinogens, and approach risk estimation differently for the two categories. In general, non-genotoxic carcinogens are treated as threshold toxicants. No effect levels are determined, and uncertainty factors are used to provide an ample margin of safety. To determine whether or not a chemical should be regarded as non-genotoxic is a matter of scientific debate and requires clear expert judgement.
The fundamental issue is: What is the cause of cancer in humans and what is the role of environmental carcinogens in that causation? The hereditary aspects of cancer in humans are much more important than previously anticipated. The key to signifi- cant advancement in the risk assessment of carcinogens is a better understanding of the causes and mechanisms of cancer. The field of cancer research is entering a very exciting area. Molecular research may radically alter the way we view the impact of environmental carcinogens and the approaches to control and prevent cancer, both for the general public and the workplace. Risk assessment of carcinogens needs to be based on concepts of the mechanisms of action that are, in fact, just emerging. One of the important aspects is the mechanism of heritable cancer and the interaction of carcinogens with this process. This knowledge will have to be incorporated into the systematic and consistent methodology that already exists for the risk assessment of carcinogens.