Tobacco Documents Online

addi.tives as methods improved to the point of being capable of detecting trace carcinogenic contaminants. This was the scientific and regulatory setting when in 1969 the FDA gathered together a group of scientists to consider how food additives and pesticides should be tested and evaluated for possible carcino- genicity. The Report of that FDA Advisory Committee on Protocols for Safety Evaluation and in particular the Panel on Carcinogenesis was publashed in 1971. This report stated among other things that: 1) Testing should be done at doses and under experimental conditions likely to yield maximum tumor incidence; 2) And that at least two species should be used for all carcinogenicity tests, and that, 3) For compounds judged carcinogenic at test levels, a virtually safe dose could, in principle, be estimated by downward extrapolation using some arbitrarily selected but conservative dose-response ;.urve. . These recommendations initiated the regulatory use of the MTD bioassay and IjLiiear risk assessment. Details on both the bioassay and the method 0 of ext:rapolation would evolve, but this 1971 Panel Report gave their

-12- deveP.opment official sanction. The Panel's recommendations were moti- vated essentially by two concerns - 1) the fears over persistent low doses of carcinogens in food the limitations of negative studies on small numbers of animals It was clear that no unqualified negative answer is ever possible. That all a negative study can do is to supply an upper limit to the possible carcinogenicity. It was pointed out that these upper limits are uncom- fortably large. Even with as many as 1000 test animals and using only 90% confidence limits, the upper limit yielded by a negative experiment is 2.3 cancers per 1000 test animals. The report contains the following statement: "'No one would wish to introduce an agent into a human population for which no more could be said that it would probably produce no more than 2 tumors per 1000." So how does one increase the sensitivity of the bioassay with a limited number of animals? The answer - increase the dose well beyond the anticipated use level and extrapolate the results down to these low doses. If the study is positive, the fidelity of the extrapolation depends on the dose-response curve - a small part of which is accessible . in the experimental range. And if the study is negative, it has a theoretical positive upper bound.

-13- rfant:el and Bryan had already introduced the concept of using this upper limit of the negative study to base the downward extrapolation on, using a I probit per 10 fold dilution slope. And so, in principle, quantita- tive extrapolation as a regulatory method (QRA) was approved. But it was never assumed that cancer necessarily occurred at low doses, but oinly that if it did, it would be safely bounded by the extrapolation procedure. Now extrapolating a positive response instead of an upper limit to a negative response makes no essential difference as Starr and othe•ts> have shown unless we know what the dose response curve is. Since our e);trapolation models are quite simplistic and without an adequate biological basis - there need not be any cancer at all at low doses - and if there is, we have no idea about its actual dose response. The notion that one can calculate expected values of an actual risk from such an analysis is really quite bizarre. Extrapolation using bioassay data 3-4 orders of magnitude removed is not a procedure for estimating risk - it's regulatory standard setting - It's not science, it's Po__~licX. And it's conservative or, if you will, prudent policy - and I'm just describing it, not advocating it. The scientists who wrote the 1971 Protocols Report had to face two problems. The QRA procedure they were recommending was extremely conservative, it would ban any carcinogenic food additive, because the amount of additive is usually substantial. Remember the additive has to be used in amounts sufficient to accomplish its intended effect. But this didn't really bother anyone at the time - carcinogenic additives had no place in food anyway. And they pointed out that this application of

F -14- QRA would be consistent with the Delaney Clause. They didn't think back then r:n terms of impurities and low level environmental contaminants - and if they had, I don't think they would have recommended high to low dose extrapolation. I believe this is true because of how they handled the second problem. If you tested a food additive in a carcinogen bioassay and the result was negative, the logical result of their analysis would require a downward extrapolation from the upper 90% confidence limit ala Mantel and Bryan. But again, because of the substantial amount of food additive required for a functional effect - typically at least several ppm, this extrapo- lation would ordinarily result in the ban of the additive at effective and useful doses. What did they do? They ignored their discussion on QRA, they ignored Mantel and Bryan and said that the sensible thing to do was to use a 100-fold safety factor! Their statement was that for agents not judged carcinogenic the use of QRA to estimate a safe dose would be logical, but would give a level so low as virtually to exclude from use agents for which there was no ~ positive evidence of carcinogenicity. And they wouldn't do it. This c:cmmonsense approach to the cancer problem was soon to be challenged by two related difficulties that defied easy solution. The first was how to deal with animal drug residues - those in food-producing animals as the result of ingestion of added drugs for prophylaxis, for the treatment of disease or for growth promotion. In 1962, Congress-had put yet another Delaney Clause in the Act with the Animal Drug Amendments - this

I -15- time though with a legal loophole called the DES Proviso. It said in effect that you could use effective animal drugs even if they were carcinogenic so long as none remained in the edible tissue of the animal after slaughter - no residue would be permitted and the FDA was given the task of approving the analytical, method to assure it. This ushered in r the era that some industry groups characterized as "chasing zero." The ot:her related difficulty occurred in food and color additives. Analytical methods were becoming so sensitive that traces of carcinogenic contaminants were being found, particularly in colors. It was hard not to firid carcinogenic derivatives of aniline, a carcinogen, in aniline- based colors. And so the question was how do you regulate a substance which does not test out as a carcinogen itself, but which contains a chemical at low but detectable levels which is a known carcinogen? Both of these problems would challenge FDA for many years, the first culminat- ing in the final SOM document in 1985, the other in the constituents policy in 1983. The need for both a procedure for risk assessment and a level of acceptable risk were common to both issues and of course they are interrelated. If you use the Probit model and a IO-8 acceptable risk level, you come out about the same place as if you used a linear model and a 10-6 acceptable risk level. Extrapolation Models and Background Additivity In the late 70's and early 80's, there was a good deal of debate over the best form of the extrapolation model. The original Probit model of 0 Mantel and Bryan was considered too arbitrary and not conservative enough - with the developing trend toward a 10-6 or one in a million

-16- acceptable risk level. The One-hit model came from the one molecule theory and was easier to explain, but it wasn't a good fit for most data. The,re were other models. You may recall the Logit, the Weibull, the Mult.-ihit and the Gamma Multi Hit - all of them competing with the Mu1t:i-S_tage. None of these were based on biology. The critical steps and mechanisms in the development of tumours were and are still unknown. But: the multistage model had the best biological credentials, having first been used to explain the steep increase in the age adjusted rates of some cancers in humans by Armitage and Doll in 1961. And most people believed cancer was a multistage process, so there was a simulacrum of a biological basis. Since the early 80's, the strongly curvilinear models have virtually been abandoned. What was that? Well, first the low level risks that emerged from these different models were embarrassingly divergent. When the various models were applied to the risk of saccharin by t:b,e NAS in 1978, the risk estimates ranged over 5 million. (Slides 7 and 8)- Saccharin Risk Estimates (1978). If OM:B had been paying attention back then, they would have been exultant - this risk assessment certainly made the uncertainties in modelling crystal clear! Today our risk assessments don't differ very much. EPA ordinarily uses the MultistaRe with an algorithm that constrains it to be linear and FDA uses the Gaylor_Kode11 procedure for most carcinogens, which is designed to be linear. The other models could not be easily linearized and were abandoned. Since then our risk assessments have been 0 more nearly in agreement, more uniformly conservative and much less revealing of still unresolved uncertainties. The linear multistage

-17- yielded greater conservatism and had the right name, but what really clinched linear risk assessment was an idea published in 2aor 3 papers in 1977 and 1978 by Crump, Hoel, Peto and their co-workers. These papers contained a notion which today is unfortunately part of quantitative risk assessment mythology, namely that there are sound biological reasons for belj:eving that every carcinogen response curve is linear at low dose rates, as far as humans are concerned. This proposition rests on the presence of background carcinogens and the way they interact with the carcinogen in question. The reasoning is part biological and part mat:hematical. The_biological_)art Approximately one of five Americans develops a cancer, and every person is exposed to thousands of carcinogens in food, in the environment and even endogenously. This 20% background rate, from these many different chemj'.cals must surely provide some significant mechanisms that are shared by t:he carcinogen in question. In other words, the carcinogen being adde>.d' and some of the background carcinogens must share a common pathway to carcinogenesis - and thus produce cancer through an identical mechanism. Their effects are functionally indistinguishable. Now the mathematical part As the slide shows, (Slide 9) - Background Additivity, the cancer inci- dence I(d) will be a linear function of the dose rate A at low dose rates provided that the slope (F1 (Do))is positive. They defend the assump- tion that the slope is positive by arguing that even if there were a threshold, it would be a threshold for each cell; there would be a

distribution of these and at least one of which would be below the critical value. Since cancer is believed to be of single cell origin, this one activated cell would initiate the cancer, Fi(Do) would be positive and the probability of response would be linear at low doses. They conclude "... in environments already containing appreciable amounts of carcino- genic processes, the effects of any slight addition to these processes will be proportional to the amount added. ... its implications are that much previous investigation of the form of the dose-response relationship at infinitesimal doses is irrelevant to the interpretation of animal studies for the formulation of social policy." It's hard to know what to say in the face of such confidence - for which there is no experimental data at all. These excathedra pronouncements are not believed by everyone but they continue to haunt some people including some in the regulatory agencies. The implications are, if this is trite, that dose-response curves become approximately linear just below the observable range - so long as they are roughly linear in the observ- able range. Not everyone believes this - I certainly don't. Alice Whittemore, Mel Andersen and other pharmacokineticists still believe that tumor probabilities are proportional to effective doses and these generate very non-linear dose-response curves. And I suppose that the folks in Kurt Harris's lab at NCI still feel they have accomplished somethang by finding mutations in p53, a putative tumor suppressor gene d in hua:~n hepatocellular carcinomas in China, despite the theory that says

-19- these mutations are expected to be produced by "background carcinogens," not ,just aflatoxin. MTD The use of MTDs (Maximum Tolerated Doses) had been challenged throughout the period. Perry Gehring and Phil Watanabe had shown in 1976 that large doses could exceed metabolic and physiological thresholds, leading to prolonged retention in the body, formation of different metabolites and in some cases disproportionate increases in reactions between reactive electrophilic metabolites and macromolecules. They concluded that dose-dependent alterations in the fate of chemicals must be considered or at high doses you risk the likelihood of disproportionate increases in toxicity including carcinogenesis. They reported evidence of possible dose.-dependent effects-in styrene, ethylene glycol, aniline, carbon disu:li`ide, 2-naphthylamine, benzopyrene, bis-hydroxycoumarin, salicyl- amide, amphetamine and sulfobromophthalin (Gehring, et al) (1976). By the: late 1970's, enough bioassay data had accumulated largely owing to NCi and later NTP studies, to provide a sufficient basis to examine the results of the studies for correlations between the responses in rats and mice. In 1979, Crouch and Wilson examined the carcinogenic potencies for 70 chemicals in the two species. They demonstrated empirically that good correlations existed for.the potencies between the different species. This was an important finding, because if there were good interspecies correlations between potency estimates for rats and mice, then it was reasonable to believe that humans and animals might also be similar in their carcinogenic responses. But in 1985, Berstein, et al, reported

-20- that the MTD's used in 186 NCI experiments were also highly correlated witpa potency. (Sl:tde 10) - MTD - Potency Correlation. Correlations between MTDs in rats and mice are not surprising because both species could respond similarly to high doses of different chemi- cals. However, the strong correlation between these MTDs and the derived carcinogenic potencies is startling. The correlation is surprising because MTDs are determined in a 90 day study and this time period has been regarded as too small a f raction of a rodent's lifetime to reflect the presence of a carcinogenic process - much less predict the strength of the carcinogen. Berstein and co-workers showed (Slide-11) that potency estimates from NCI Bioassays were restricted to an approximately 30-fold range surrounding ln(2). The TDsn is TD50 virtually the same as the MTD. They used a one-hit model to show this and an idealized 2-Group experimental design, but they and others have shown that this high correlation is not sensitive to "reasonable" depar- tures from either the experimental design or the extrapolation model used. Riethh and Starr (1989), and others since, have investigated these corre- lations in detail. It's clear now that: . o - The correlations between the MTDs and the estimated potencies are real. They do not depend on a "selected" data base. I