Comments Submitted by
James D. Wilson on EPA Report:
Atrazine: Carcinogenicity Hazard Assessment
and Characterizaton
EPA Docket Number OPP-00637
January 17, 2000
THE NEW THEORY PRESENTED TO EXPLAIN INCREASED RATES OF MAMMARY TUMOR FORMATION IN CERTAIN RAT STRAINS TREATED WITH ATRAZINE IS COGENT, ELEGANT, PLAUSIBLE, AND WELL SUPPORTED BY DATA. IT SHOULD BE ACCEPTED AS VALID FOR REGULATORY PURPOSES. Treating female Sprague-Dawley rats with atrazine at or near maximum tolerated intakes leads to an increased age-specific incidence of mammary tumors in those animals. This observation puzzled toxicologists: atrazine neither increases the mutation rate in various tissues, nor interacts with the estrogen receptor. Generally, chemicals exposure to which is associated with mammary tumor development are either mutagens (e.g., dimethylbenzanthrene, diethylnitrosamine), or affect estrogen economy in some way (e.g., single-component birth-control agents, diethylstilbestrol). This uncommon result has spawned research designed to unravel the mystery. Results now suggest a theory which accounts for the increase in mammary tumors. Sprague-Dawley rats uniquely tend toward early onset of senility in sex-related parts of their endocrine system. In this rat strain, atrazine diminishes the release of two hormones from the hypothalamus, luteinizing hormone and norepinephrine. Through a complex cascade of effects on other endocrine organs, this effect disrupts the estrous cycle and leads to a sustained increase in circulating prolactin. Prolactin stimulates mitosis of cells in the breast. Increasing the rate of mitosis increases the probability that normal background DNA damage will lead to formation of mutant cells and thus tumors.
Two things may make EPA reluctant to take regulatory action based on this theory. First, it is novel. Although the basic endocrine physiology on which it is based is well-established, the effect of atrazine on the hypothalamus is something not frequently noted in toxicology. It takes some getting used to for the scientific community (who must at least tacitly accept the validity of this theory). Second, not all the details of the interactions are understood as well as one might wish. The planned review of this theory by the Science Advisory Panel represents a part of the process of testing the theory’s acceptability by the broader scientific community. As to the missing details, these fall into the “nice to know” category: the chain of causality, from effect on luteinizing hormone release to effect on mammary tumor rates, is well established by the evidence that has been presented. No experimental evidence that contradicts the theory has been developed and presented. Thus it seems that EPA should be prepared to accept this theory as the best available, and act upon it when an appropriate occasion arises.
THE THEORY OF ATRAZINE ACTION MAY INFORM EITHER OR BOTH OF TWO EXPERT JUDGMENTS MADE IN THE PROCESS OF A PESTICIDE-RESIDUE RISK ASSESSMENT. The theory usefully informs two judgments: the relevance of observing the increased mammary-tumor rate in Sprague-Dawley rats, and the choice of which algorithm to use in quantitative risk assessment of atrazine. The document being reviewed includes a Conclusion section that describes a tentative decision regarding the relevance issue, and states that the planned Science Advisory Panel meeting will not address quantitative risk assessment. Yet it seems not advisable to leave unremarked what the implications of accepting this theory may be. In fact it seems particularly worthwhile to describe these, since, as I argue below, the decision on relevance is not adequately supported by evidence presented and described in this report.
Recall that the question of “relevance” arises early in the logic used to judge safety of pesticide residues. (Putzrath and Wilson, 1999) Safety is judged by comparing two numbers, one representing a “reasonable worst case” exposure (mean daily intake) of the residue, the other a measure of toxicity, representing a daily intake judged to pose only a negligible risk of harm, even if the entire U.S. population consumed the residue at that level throughout its lifetime. If the worst-case mean daily intake is smaller than this negligible-risk lifetime intake, then the pesticide is considered safe. If not, not.
This means of judging safety was invented by FDA and sister agencies worldwide around 1950. It has been used ever since. The 1958 amendments to the Federal Food, Drug, and Cosmetic Act (FDCA) explicitly assume that FDA will use this comparison to judge safety of food additives and pesticides. The same statutory standard is implicit in the 1996 Food Quality Protection Act, since it makes explicit the implicit criterion for safety from the 1958 amendments, “reasonable certainty of no harm when used as intended.” (Putzrath and Wilson, 1999)
(EPA uses various terms to identify this negligible-risk intake level, including “RfD,” “RsD,” and “PAD;” FDA uses the term “ADI.” Putzrath and Wilson (1999) introduced a generic term, “Safety Index” to denote any of these measures. We did so because the procedures used to identify RfD, RsD, etc. values differ only slightly in the logic used.)
Note that the procedures used to identify the worst-case intake number and the RfD / RsD number for the pesticide residue are highly policy-constrained. Very specific ways to interpret very specific animal-toxicity information are used to identify the RfD (or RsD or other Safety Index) number. We in the risk analysis profession are very confident that food additives and pesticide residues are safe, as long as their safety is judged using these procedures. Our confidence is based largely on the fact that in the fifty years of use, about 2000 substances have been evaluated and approved for use, and in only one instance has harm to consumers been observed. In that case - use of a cobalt soap as a foam stabilizer in beer - the toxicologists doing the evaluation here and in Canada underestimated the worst-case intake. They did not believe that alcoholic, hard-working stevedores and brewery workers could consume a case of beer (~5 liters) a day. Several of these workers suffered liver and kidney damage. (The approval of that additive was rescinded.) This failure rate is very low, less than one per thousand evaluations. Very few human endeavors can claim to do as well.
The conclusions stated in this report concern a decision that concludes one step in the safety assessment process. The logic of this process is conveniently described as having three steps, each of which ends in a decision usually based on application of expert judgment. (Putzrath and Wilson, 1999)
Step 1: Identify any studies that do not meet accepted standards for quality, and any in which responses observed are considered not to be relevant to human risk. (A “relevant” study or response is one in which the animal responses observed are held to predict how humans will respond.) Any studies judged not of sufficient quality or not relevant are set aside and not considered further in identifying a Safety Index value.
Step 2: From among the studies judged both relevant and of sufficient quality, select one (the “critical study”) to serve as the basis for the RfD. Criteria exist by which selection is made.
Step 3: Using an appropriate algorithm, calculate the Safety Index value for the substance. The algorithm used is specified by contingent instructions, depending on the nature of the adverse effect observed in the critical study, the quality of the data base, and other factors. That part of EPA’s tumor guidelines called “quantitative risk assessment” consists of a dissertation on the contingent instructions for the algorithm used if tumor is the adverse effect observed in the critical study. If the adverse effect is not tumor, then the default algorithm is, “Divide the largest NOAEL from the critical study by 100.”
Note that the current draft of the cancer guidelines does not explain this logic at all clearly. By this logic, long accepted in practice, once a response is judged to be “not relevant,” that response can not be used as the basis for judging safety. The current draft guidelines seem to suggest that the “relevance” judgment is independent of the choice of study to use as the basis for the RfD / RsD value. The logical relationship between these two judgments needs to be emphasized.
In the case of atrazine, the choice of critical study is crucial to identifying a Safety Index value. The decision concerning the relevance to humans of mammary tumors in Sprague-Dawley rats is crucial: If this response is not relevant, if it is judged not to predict how humans will respond, given similar exposure, then some “not tumor” response will be taken as the basis for the Safety Index, and the third-step algorithm will be the “NOAEL/100” default (unless some contingency implies a different procedure). If, however, this response is judged to be relevant, as the Conclusions state, then the mammary-tumor response will be chosen as the basis for the Index value, and one of the “tumor” algorithms will be used.
EPA’s present policies regarding the algorithm used to identify an Index value for a tumor end-point allow for a choice to be made. One choice remains the traditional “linear, no-threshold extrapolation.” The other choice, in effect, treats the occurrence of tumor as a secondary consequence of some other toxicologic effect and thus uses the “NOAEL/100” method (or something very like it). (In this developing policy, EPA also reserves the right to do something intermediate between these two extremes, although no examples of an intermediate position have yet appeared.) So far, the few examples of substances that have trodden the “carcinogen, but use NOAEL/100” branch of this decision pathway are those that act through an endocrine-related pathway, notably some that affect thyroid hormone economy. If the “it’s relevant” conclusion stands, atrazine seems to resemble these thyroid-affecting substances very closely. Clearly there exists some intake of atrazine that will not exceed the ability of the hypothalamus to respond to stress, and thus will not affect mammary tumor rates. In this circumstance, a “threshold” assumption seems appropriate. Using the “linear, no-threshold” assumption would violate the balance between risk and benefit that exists in this safety-assessment process.
THE CONCLUSIONS DESCRIBED IN THIS DOCUMENT DO NOT FOLLOW FROM THE TECHNICAL ANALYSIS. In an earlier memo (Wilson, 2000), I pointed out that a discrepancy exists in this document between the purpose stated in the Introduction and the material in the Conclusions section (§3.5). It is almost as though whomever wrote the Conclusions did not read the preceding pages, or, at least, did not review them to be sure that the evidence needed to reach those conclusions was present. That evidence is not present. The observations and inferences described in loving detail in this report do not support the conclusions stated.
The problem is that the issue addressed in the Conclusions, whether or not the observation of increased rate of mammary tumors in atrazine-treated Sprague-Dawley rats should be considered to predict human response, is raised only because treatment of rats of a different standard strain does not lead to a similar increase. In atrazine-treated Fisher 344 rats, no increase in mammary tumors is observed. Since these two sets of observations conflict, only one can be used as the basis for judging safety of atrazine residues in food. But the report hardly mentions the difference. It focuses instead on the extensive research done to understand the biologic mechanism by which atrazine treatment leads to tumors in Sprague-Dawley rats. For the regulatory purpose inferred from the stated conclusions, the key inference concerns the behaviors of these two rat strains. That inference is not discussed.
The evidence discussed in this paper is necessary to the “relevance” decision, but not sufficient. Key to this decision is this inference: Is the Sprague-Dawley response particular to that strain? Does some peculiarity of that strain’s physiology or anatomy or genetic makeup cause it to respond in this way? Are humans known to share that feature? If the answer to these questions is “no,” then this response is not relevant. If the answer is “yes” or “not proved,” then the tumor response is taken as relevant, for regulatory purposes.
(Note that if this response is judged “relevant,” then a second mechanism-based inference will be important for the third step. If the observed difference in response between Sprague-Dawley and Fisher rats is judged to stem from quantitative differences, then the magnitude of this difference will be important in deciding which algorithm to use.)
The judgment of relevance is usually the one requiring the most subtle exercise of expert judgment in the entire risk assessment. A host of observations integrated by experience are important to the judgment. (One, for instance, is which of the two rat strains is generally a better predictor of human response.) Unfortunately, the observations and inferences needed and used to reach this judgment have not been described in this report.
The discussion of mechanistic experiments concludes that in Sprague-Dawley rats treatment with atrazine affects the hypothalamus, and that growth of mammary tumors is sped up by the sequelae of hypothalamus effects. The critical step in the mechanism is concluded to be effects on the excretion of one or more hormones. It may well be that this effect is unique to the Sprague-Dawley rat. It may be that it is general, only that the Sprague-Dawley rats are more responsive to atrazine. The Conclusions must be based on judging that this hypothalamus effect is not unique to the Sprague-Dawley rat; otherwise the tumors must be judged “not relevant.” However, there is not the smallest mention of this key inference.
Thus the public must conclude that this report’s conclusions are not supported by the evidence and reasoning that appears therein.
Thank you very much for the opportunity to comment.
REFERENCES TO LITERATURE CITED
R. M. Putzrath and J. D. Wilson, 1999. “Fundamentals of health risk assessment: Use, derivation, validity and limitations of safety indices” Risk Analysis, 19: 231-248.
J. D. Wilson, 2000. Memo to Docket OPP-00637, 1/5/00