SCOPE 49 - Methods to Assess Adverse Effects of Pesticides on Non-target Organisms  

The general approach for evaluating the potential adverse effects of pesticides in non-target species was established over four decades ago by Lehman, Fitzhugh and Nelson at the Food and Drug Administration which was responsible at that time for regulating pesticides in the US. This approach, which was similar to that used for drugs, involved a case-by-case comparison of the riskbenefit ratio for each pesticide.

1. Acute studies which included LD₅₀ determinations coupled with a description of the symptoms or adverse effects 
2. Chronic studies which were multidose one- to two-year rodent studies. 

The intent of this approach was to identify all of the potential adverse effects that could be produced by either acute or chronic exposure to the pesticide and to determine the exposure dose of the pesticide required to produce all such effects. 

Subchronic (90-day) studies were utilized to establish the dose levels for the chronic studies; additional kinetic, metabolism, and mechanism-of-action studies were carried out, if needed, to characterize the toxicity of the pesticide. 

Both the acute and chronic data were then used to establish a threshold for the adverse effects of the pesticide; and, in most cases, the tolerances and other regulatory positions were established by dividing the lowest threshold by an appropriate safety factor. This approach, which was subsequently adopted as the acceptable daily intake (ADI) method for food additives, was and is used both in the US and abroad for regulating pesticides as well as other classes of chemicals.

During the subsequent two decades preceding the creation of the EPA and the transfer of the major responsibility for regulating pesticides from FDA to EPA, several additions to the toxicity testing requirements were made to evaluate the hazard of pesticides. The discovery of organophosphate (OP) insecticide potentiation by Frawley at the FDA led to a requirement for testing of all new OP insecticides for such synergistic effects, and a leghorn chicken test was also required for all new organophosphate insecticides to detect any delayed neuromuscular effects of these agents. The thalidomide episode stimulated the development of teratology as a part of the toxicologic testing procedures; and growing concern over other types of feto-toxicity stimulated the development of multi-generation protocols for evaluating fertility, postnatal and other reproductive effects of pesticides. Methods to detect and evaluate the genotoxic effects of pesticides were stimulated by the studies of Bruce Ames and his associates. Significant developments occurred in the areas of neurotoxicology, behavioural toxicology, and in immunotoxicology; but these have not yet resulted in validated protocols, and thus they have not been incorporated into regulatory requirements for pesticides. Although these changes in toxicological methodology provided a broader database for predicting potential adverse effects in non-target species, the basic approach in pesticide regulation continues to rely on both acute and chronic data for hazard evaluation and for the setting of pesticide tolerances.

The current approach to pesticide regulation differs from this earlier approach in two ways. First, it is focused on the chronic effects of pesticides rather than on the acute effects, and, second, it is focused on a single chronic effectcancer. Emergency room physicians who treat pesticide poisoning have pointed out that it is the acute effects of pesticides that are responsible for most of the actual morbidity and mortality associated with most types of pesticide poisoning, and they argue that we should be more concerned about the acute effects which are 'real-world' than the cancer risks which are largely hypothetical. Further, the current protocol used for chronic studies on pesticides is based on the National Cancer Institute/National Toxicology Program Oncogenicity Bioassay rather than on the more holistic procedure which was previously required by FDA. The NCI/NTP protocol was designed to serve as a screening test for carcinogenicity, and, consequently, is neither adequate nor appropriate as a bioassay for chronic toxic effects other than cancer. 

Another major difference between the current approach and the previous `FDA' approach for evaluating the adverse effects of pesticide exposure involves the way in which toxicity data are interpreted. In the current system, the carcinogenicity data from the chronic rodent studies is extrapolated using mathematical models which provide a numerical estimate of the upper bound of the cancer risk, and these numbers (Q<sub>l</sub>*values are then used for a variety of regulatory purposes. In essence, this approach substitutes mathematical guidelines for the scientific judgement that was the key element in the `FDA' approach.

Since the basic purpose of conducting toxicity studies is to provide an accurate prediction of the potential adverse effects of a chemical in non-target as well as in the test species, we need to ask whether the methodology that we use both for testing and for interpreting the test results is yielding answers that protect both the environment and public health. Based on the arguments presented in the preceding discussion, it is likely that our predictions regarding the adverse effects of pesticides on humans would be improved if they included a greater emphasis on the acute effects and on the non-cancer chronic effects of pesticide exposure.

Protocols for detecting both acute and chronic neurotoxicity and the behavioural effects of pesticides (including specific tests for neuronal dysfunction) should be a part of the requirements for pesticide registration. Recent workshops devoted to these areas (LSRO report to FDA on `Predicting Neuro-toxicity and Behavioral Dysfunction from Preclinical Toxicologic Data' and the `Workshop on the Effects of Pesticides on Human Health which was organized by the Task Force on Environmental Cancer, Heart and Lung Disease) have recommended methods and test batteries that would detect sensory, motor, autonomic, cognitive and behavioural dysfunction.

Another area that would be of direct benefit to the medical personnel responsible for treating pesticide exposure in humans is the development of antidotal and symptomatic treatment for the various classes of pesticides. Although we have more specific antidotes available to treat pesticide poisoning than for any other single class of acute poisons, the number of antidotes is small for non-insecticidal pesticides, and these cases must be treated symptomatically. The current acute toxicity protocols do not currently provide the kind of clinical information needed in such cases.

Endocrine- and immunologic-related adverse effects of pesticides are another example of an area where we need toxicologic testing methods that will detect both acute and chronic toxicity and provide a reliable basis for predicting human effects.

Before we add new protocols to the current methodology, however, we need to evaluate the answers provided by our current methodology. The two scientific disciplines which contribute most directly to the evaluation of cancer and all other human health hazards that may result from exposure to pesticides and all other toxic chemicals are epidemiology and toxicology. When these disciplines function in this 'watch-dog' role, both have the same goal: to gather sufficient information about the potential adverse effects of exposure to the chemical so that we can reliably predict both the type of adverse effects that may occur and the exposure conditions likely to produce each of these anticipated adverse effects.

Although the primary difference between these two disciplines is the subjects (i.e., epidemiologists use humans, whereas toxicologists use rodents), there are also major differences in how we approach the problem, the techniques we use to generate the data and how we use these data to predict the hazards. Each starts and ends at the same place, but each discipline gets there by a different path.

It should be pointed out, however, that there is no disagreement between epidemiologists and toxicologists as to the priority of human data over rodent data. Toxicologists recognize that in those situations where we have sufficient human data, there is no need to do any type of toxicity studies unless we are predicting for environmental or non-human effects. The main reason why toxicology rather than epidemiology serves as the basis for most regulatory decisions involving cancer and other adverse health effects is that human data are insufficient for most chemicals for which estimates of safe exposures are needed. In addition, there are ethical constraints, particularly with chemicals which are not drugs, that often preclude the type of epidemiologic studies we need to make reliable predictions about risks to humans.

Since it is obvious that we will need to continue to rely on the toxicity database for making hazard or risk predictions, it is important that in addition to searching for new methods for obtaining and interpreting the toxicology data, we also ask whether our current approach is giving us correct answers.

One approach to the question of whether risk assessment methods based on the results of oncogenicity studies in rodents are giving us the right answers is to compare the predictions from the epidemiologic studies with those from the toxicity database. When we do this for those agents for which we have predictions from both sources, there are only a few discrepancies. We do not have animal data demonstrating that asbestos causes mesothelioma or lung tumours and only limited evidence that arsenic and benzene cause cancer in rodents, although the human evidence for these chemicals is clearly sufficient to establish causation. More recently, epidemiologists have concluded that the herbicide 2,4-D causes non-Hodgkins lymphoma in exposed farmers and that alcohol ingestion causes cancer of the oesophagus, liver, and possibly breast; but neither of these chemicals produces cancer in rodents. Although it has been argued that the epidemiologic evidence in these cases is weak since it is based on an association rather than on a causeeffect link, these findings are of particular concern to toxicologists since they are not like asbestos, where our failure to produce lung tumours can be attributed to the lack of a good animal model and to inadequate testing. We now have 18-negative rodent studies for ethanol, and the 2,4-D oncogenicity study was considered by the EPA FIFRA Science Advisory Panel to be an excellent study. Neither 2,4-D nor ethanol is mutagenic; and acetaldehyde which is the primary metabolite of ethanol in both rodents and man is also not mutagenic, although it did produce lung tumours in an inhalation study in rats. While it is reassuring to find that, in most cases, the toxicologic data supports the epidemiologic predictions, it is also clear that the exceptions must be investigated and resolved to maintain the credibility of the animal tests as reliable predictors of cancer and other adverse effects of chemicals in humans.

We also need to look at the exceptions in the reverse situation in which the toxicity data are positive, but the epidemiology findings are negative. Such exceptions can occur if the test species is more susceptible to the tumourigenic effects of a chemical than the target species; e.g., it has been suggested that the liver tumours produced by the halogenated hydrocarbons are illustrative of this situation. Many of the pesticides and solvents in this group, such as chlordane, heptachlor, trichloroethylene, and methylene chloride, produce liver tumours in rodents, but epidemilogic studies generally do not support these findings.

It can be argued, of course, that the epidemiologic studies are inadequate, and that is certainly true for some of these agents; it can also be argued that, in these cases, the rodent studies are over-predicting the cancer risk for man. The B₆C₃F₁ mouse which is widely used as a rodent test species does have a high incidence of liver tumours, and it has been suggested by the International Agency for Research on Cancer (IARC, Lyon, France) that when positive results are found only in mice and not in other species that such results should not be used as a basis for a regulatory action. Similar arguments have been made regarding the increased incidence of thyroid tumours in rats exposed to goitrogens such as the sulfonamides. In this case the mechanism by which the thyroid tumours are produced appears to have a clear threshold, and it has been recommended by an EPA Working Group that such agents be regulated on the basis of a no-observed-effect level (NOEL) approach rather than with the conventional cancer models.

Another example of a case where rodent tests appear to be giving incorrect answers, or at least over-predicting the human risk, is the production of kidney tumours in the male rat by gasoline, p-dichlorobenzene, limonene, and tetrachloroethylene. The mechanism by which these agents cause kidney tumours in male rats appears to involve conjugation with an -2u-globulin and the formation of hyaline droplets in the P2 segment of tubule. Since these droplets are not found in female rats or other species including humans, it is argued that male rat kidney tumours produced through this mechanism are not predictive for human carcinogenesis. On the basis of this argument and the lack of positive genotoxicity evidence for p-dichlorobenzene, the EPA halogenated organics subcommittee of the SAB has recommended that the classification for this agent be downgraded from B2 to C in the new drinking water regulations for p-dichlorobenzene. During the past few months, the EPA has held workshops on the predictive value of mouse liver tumours, rat kidney tumours, and thyroid tumours induced by various goitrogens, and on the use of pharmacokinetic parameters in risk assessment. Some of the recommendations from these workshops are likely to be incorporated in the new Cancer Assessment Guidelines being developed by EPA.

One of the most common complaints about the use of mathematical models in risk assessment is that these models ignore the relevant biology. Partly in response to this criticism, a NAS/NRC committee has issued a report which presents methods for incorporating pharmacokinetic data in risk assessment. When recommendations of this committee were applied to the risk assessment of methylene chloride, the predicted carcinogenic risk was reduced by a factor of 7. Another approach that is receiving considerable current attention is to identify and quantify all uncertainties that exist in the mathematical modelling process, so that we can design and carry out the research needed to substitute hard data for these uncertainties. The procedures should improve both the relevance and the accuracy of the model-based approach. We must also consider the more basic issue, which is whether there is sufficient scientific or biologic justification for extrapolating the results of our high-dose rodent studies to the low doses needed for prediction.

As was pointed out previously, the current protocol for conducting chronic toxicity studies is based on the NCI oncogenicity bioassay, which was a screening test designed primarily to answer two qualitative questions: Is the chemical oncogenic? What kind of tumours are produced? When this program was transferred to NTP, some additional features were added to make it more like the conventional chronic toxicity study, and an effort was made to obtain an indication of doseresponse relationsips. Basically, however, it is still a screening test for oncogenesis, and the manipulation of these data with sophisticated mathematical models which generate finite risk estimates does not alter the quality of the input. The idea that these models somehow generate quantitative data which is valid far beyond the range of the actual data may well be an illusion; in any event, it is an idea that needs to be rigorously tested. At the same time, we need to re-examine the protocols that we use to generate the cancer data in animals to determine whether we can develop procedures that will provide quantitative information that can be used either on a stand-alone basis or as input for the extrapolation models.

Many of these issues are being considered by various scientific and regulatory groups both in the US and in other countries, and some of the issues may be addressed in legislative proposals to revise the laws under which pesticides are regulated (e.g., HR 4737, HR 4739, and the Pesticide Reform Bill). In this process of seeking remedies for actual or perceived deficiencies in the scientific methodology used to identify and quantify the hazards of pesticide exposure, we need to remember that the major problem is not methodological, but is rather the lack of data on which to base the prediction.