TOXICOLOGICAL SCIENCES 68, 275–279 (2002) Copyright © 2002 by the Society of Toxicology
FORUM Thresholds of Carcinogenicity of Flavors William J. Waddell 1 Department of Pharmacology and Toxicology, School of Medicine, University of Louisville, Louisville, Kentucky 40292 Received February 26, 2002; accepted April 12, 2002
Fifteen compounds approved by the FEMA (Flavor and Extract Manufacturers Association) expert panel as GRAS (Generally Recognized As Safe) and structurally related compounds have been reported to be carcinogenic in rodent studies. The dose response of the 15 compounds in these studies was scrutinized by attempting to plot the percentage of animals with tumors against the dose of the compound on a logarithmic scale in molecules of compound per kg per day (the Rozman scale). Four compounds had either no or an inverse dose response: benzaldehyde, furfural, 3,4-dihydroxycoumarin, and ␥-buterolactone. Three had a response at one dose only: anethole, estragole (2 studies), and isophorone. Obviously, a dose-response curve could not be generated for these 7 compounds. Four compounds had an increasing response at two doses (benzyl acetate, cinnamyl anthranilate, ethyl acrylate, and estragole); three compounds had increasing responses at three doses (citral, 2,4-hexadienal, and pyridine); one compound had increasing responses at four doses (methyl eugenol). The three compounds with three doses fit a linear plot with a correlation coefficient of at least 0.9; the four doses in male rats of methyl eugenol fit a linear plot with a correlation coefficient of 0.999983. The intercept at zero percentage tumors of these linear fits was at least several orders of magnitude greater than the estimated daily dose of these flavoring agents to individuals in the United States. This is interpreted to indicate that these flavoring agents have a clear threshold for carcinogenicity in animals that is well above the levels currently approved for use in foods; consequently, these animal studies should not be a cause for concern for carcinogenicity of these compounds in humans. Rather, the animal studies should be viewed as providing evidence for the safety of these compounds at current levels of human exposure. Key Words: threshold; carcinogenicity; flavors; foods; FEMA; GRAS; methyleugenol.
Extrapolation from studies for carcinogenicity in animals to the possibility of carcinogenicity in humans has been, and continues to be, a perplexing problem, e.g., see Klaassen (2001) or Hayes (2001) for summaries and discussions. The 1 To whom correspondence should be addressed at 14300 Rose Wycombe Lane, Prospect, KY 40059. Fax: (502) 228-6779. E-mail: [email protected]
debate of threshold versus nonthreshold for carcinogens is one of the most contentious in toxicology. Animal studies usually are done at or near the maximum tolerated dose (MTD) and human exposures are typically much lower. In the nonthreshold approach, extrapolation of the dose-response data in animals is usually done to zero dose assuming either that only zero dose produces no tumors or that the tumors in the vehicle control are at zero dose. In the threshold approach, extensions of the dose-response data on tumors in animals frequently are made to zero tumors with mathematical projections based on a variety of speculations. None of these assumptions in either the nonthreshold or threshold approach is without potential flaws, which, of course, is why the controversy continues. The actual shape of the dose-response curve for an adverse effect of an agent in living beings can be characterized correctly only with a sufficient number of data points. Most of the carcinogenicity studies in animals have too few data points to properly define that shape, especially at the low doses to which humans are exposed. However, it has been well established, since the original paper by Gaddum (1945), that the dose should be on a logarithmic scale to effect a linear response. The scale proposed by Rozman et al. (1996), hereinafter referred to as the Rozman scale, is ideally suited for carcinogenicity studies for several reasons, one of which is that it displays all doses on the abscissa down to a single molecule, which is the smallest dose that can be administered. This allows the proper perspective for comparison between doses used in the animal studies, those to which humans are exposed, and the lowest dose that can be administered. The compounds with carcinogenicity studies in rodents that have been approved by the FEMA (Flavor and Extract Manufacturers Association) expert panel as GRAS (Generally Recognized as Safe) and some structurally related compounds (FEMA, personal communication) were examined for dose response on the Rozman scale. It became apparent that as more data points were available from a carcinogenicity study in animals, the better the fit for a straight line showing a clear threshold. It is proposed that this empirical approach should be used in the absence of any convincing alternative.
MATERIALS AND METHODS The FEMA GRAS compounds that have been approved for use in foods, and some structurally related compounds, were reviewed for those that have had studies on carcinogenicity in animals (FEMA, personal communication). There were a total of 38 studies; 18 of these were interpreted as negative. There were two studies with allyl isovalerate; one of which demonstrated a clear doseresponse inhibition of control tumors; the other showed bladder tumors only. There were two studies (with d-limonene and methylbenzyl alcohol) with renal tubular cell tumors only in male rats; the mechanism of these tumors was considered irrelevant to human cancer and not considered further. Of the remaining 16 positive studies, there were four with either no or an inverse dose response (benzaldehyde, furfural, 3,4-dihydroxycoumarin, and ␥-buterolactone); these were not considered further; confounding by other factors is considered likely to account for either no or an inverse dose response. There were four studies with an increase in tumors at one dose only (anethole, two with estragole, and isophorone); since a line with any shape can be drawn through a single point, these were not considered further. The remaining 8 compounds were plotted on the Rozman scale. Carcinogenicity data were obtained from the cited original references. The number of tumors in the control group was subtracted from the number of tumors in each of the dosed groups. Without this correction, the shape of the curve would not change, but would only be displaced parallel to the right. Doses were converted from weight/kg/day or ppm to molecules/kg/day. The doses were those used for long-term carcinogenicity studies, typically 2-year or lifetime administration orally; in most cases they could be considered lifetime average daily doses. There were four compounds with increasing responses at two doses each (benzyl acetate, cinnamyl anthranilate, ethyl acrylate, and estragole); three compounds had increasing responses at three doses (citral, 2,4-hexadienal, and pyridine); one compound had increasing responses at four doses (methyl eugenol). Shown on each graph is the dose calculated for “eaters only” of that flavor in the United States (FEMA, personal communication). The use of cinnamyl anthranilate was discontinued in 1982, but the level consumed by humans prior to that time is shown. Also shown for methyl eugenol is the dose to eaters of this flavor from that naturally occurring in foods and for pesto eaters (Smith et al., in press). For those with three or four data points the best fit for a linear response was calculated and drawn by SlideWrite software (Advanced Graphics Software, Inc., Encinitas, CA); the correlation coefficient was calculated by the SlideWrite program and is shown for each linear fit. A linear scale of percent tumors on the ordinate provided a better fit than either a logarithmic or probit scale. A straight line was arbitrarily drawn through those with only two data points.
malignant lymphomas are considered “equivocal” evidence for carcinogenicity by the NTP primarily because of the high incidence of this tumor in historical controls. The threshold dose for each of these compounds ranges from 2 ⫻ 10 19 to 2 ⫻ 10 20 and the correlation coefficients for a linear fit range from 0.90 to 0.999991. The correlation coefficient of 0.90 is for mononuclear cell leukemia in the rat; the NTP report states that evidence from this type neoplasm is “equivocal.” The remaining correlation coefficients are closer to unity. Figure 8 shows the results for hepatocellular carcinomas in male rats administered methyl eugenol. The range of percentage tumors, after subtracting those in the control, was from 2 to 68%. The correlation coefficient for a linear fit is 0.999983 and the threshold for zero tumors is 1 ⫻ 10 20, which is almost 10,000 times the estimated dose for pesto eaters, who are considered to be the highest consumers of the compound. There were also liver carcinomas in female rats and adenomas in both sexes; these data are not shown because the lines for each of these plots were essentially superimposable with the one shown in Figure 8. The correlation coefficients for the methyl eugenol plots ranged from 0.953 to 0.999983 and the intercepts ranged from 7 ⫻ 10 19 to 2 ⫻ 10 20. DISCUSSION
The debate over how to extrapolate from the high doses used in rodent carcinogenicity studies to the low doses usually encountered by humans cannot be resolved by speculation. Speculative extrapolations below the data obtained in a specific experiment to a hypothetical one in a million, or whatever risk one wishes to assume, can of course be done readily from these data. However, these data tell us that, under the conditions of the specific experiment, zero tumors are to be produced by the dose at which this linear extrapolation crosses the abscissa. Any further extrapolation exceeds the limits of information
Figures 1– 4 show the results for the compounds that have only two data points for an increase in tumors. Note that the dose for “eaters only” for each of the compounds is at least 1000-fold less than the intercept of the linear fit with zero tumors. Furthermore, the intercept for each of these lines with the dose for zero tumors is within one order of magnitude of about 1 ⫻ 10 21 molecules/kg/day. Even more remarkable is that the intercepts for tumors from ethyl acrylate in two species and considered as either papillomas or carcinomas all crossed zero tumors at 3 ⫻ 10 20 molecules/kg/day. Figures 5–7 show the results for the compounds that have three data points for an increase in tumors. Remarkably, the intercepts for tumors from 2,4-hexadienal in two species and considered as either papillomas or carcinomas all crossed zero tumors at 2 ⫻ 10 20 molecules/kg/day. The dose for “eaters only” is closest (still more than 400⫻) for citral, but these
FIG. 1. Percentage of male rats and mice with adenomas after gavage with benzyl acetate plotted against molecules of benzyl acetate/kg/day. Data are from NTP TR 250 (NTP, 1986a).
FIG. 2. Percentage of male mice with adenomas plus carcinomas after receiving cinnamyl anthranilate in food, plotted against molecules of cinnamyl anthranilate/kg/day. Data are from NTP TR 196 (NTP, 1980).
FIG. 4. Percentage of female CD-1 mice with hepatomas after receiving estragole in food, plotted against molecules of estragole/kg/day. Data are from Miller et al. (1983).
available from the experiment and is completely unwarranted. If, however, enough doses are available in the carcinogenicity studies in animals to convince an observer of the shape of that response, and where it crosses the abscissa, under the conditions of that experiment in those animals, some pragmatic and useful interpretations may be possible. This article is such an attempt. The Rozman scale was chosen for plotting the dose response for several reasons. Dose on a logarithmic scale is obvious because it is well established that most biological responses are related to the logarithm of the dose. This emanates ultimately from the fact that the chemical potential of a substance is related to the logarithm of the activity or concentration of a substance (Waddell and Bates, 1969). Gaddum (1945), much earlier, empirically recognized the practical significance of this relationship in pharmacology. The Rozman scale uses molecules instead of weight as a measure of the dose of the chemical; this is again an obvious advantage when comparing
compounds with different molecular weights. Lastly, and most importantly, the scale is continuous to one molecule (10 0); this places all possible doses in perspective, which is essential when one wishes to consider the possibility that a single molecule might cause cancer. When Avogadro proposed his constant for the number of molecules in one gram-mole, Lord Kelvin was so impressed by the enormity of the number that he made an interesting calculation to illustrate it. He said, suppose one could mark each of the molecules in a glass of water; then pour the contents of that glass into the ocean and mix that water uniformly in all the water on earth. If then one took a one-glass sample of that mixed water, there would be about 100 molecules from the original glass in water everywhere on earth (see Schro¨ dinger, 1944). The message from this is that it is most unlikely that there is a zero concentration of any chemical in the body of any living organism. A femtogram (which is near the limit of detection with current technology) of a compound with a
FIG. 3. Percentage of male rats and mice with tumors after gavage with ethyl acrylate, plotted against molecules of ethyl acrylate/kg/day. Data are from NTP TR 259 (NTP, 1986b).
FIG. 5. Percentage of female mice with malignant lymphomas after receiving citral in food, plotted against molecules of citral/kg/day. Data are from NTP TR 505 (NTP, 2001a).
FIG. 6. Percentage of rats and mice with forestomach papillomas and carcinomas after gavage with 2,4-hexadienal, plotted against molecules of 2,4-hexadienal/kg/day. Data are from NTP TR 509 (NTP, 2001b).
molecular weight of 100 still contains more than 6 million molecules. Most of the older carcinogenicity studies in animals were designed merely to attempt to detect whether a substance is carcinogenic at any dose; indeed, many substances were carcinogenic only at or near the MTD. Furthermore, even if more than one dose produced tumors, the range of the percentage tumors from those doses usually was narrow, thus limiting confidence in the shape of the full dose-response curve. These have led to the speculation and confusion over how to extrapolate to much lower doses. Now that more doses are being used in the range that produces tumors, there is an opportunity to evaluate the shape of that curve. It is quite striking that the NTP study of methyl eugenol found four doses ranging from 2 to 68% tumors and that linearity unquestionably is the best fit for these points (r ⫽ 0.999983). Neither logarithmic nor probit scales gave nearly as good a fit. This gives one much more confidence in extrapo-
FIG. 7. Percentage of female rats and mice with tumors after receiving pyridine in drinking water, plotted against molecules of pyridine/kg/day. Data are from NTP TR 470 (NTP, 2000a).
FIG. 8. Percentage of F344/N male rats with hepatocellular carcinomas after gavage with methyl eugenol plotted against molecules of methyl eugenol/ kg/day. Data are from NTP TR 491 (NTP, 2000b).
lating this straight line to its intersection with the abscissa. Further, the fact that all the other studies with three data points have r values very near unity, similar slopes, and intersect in the same dose range lends additional confidence to the linear interpretation. There is no apparent evidence from these plots to suggest that any other extrapolation to zero dose should be attempted; indeed, such an extrapolation from the 2% tumors for methyl eugenol to zero dose (which is not even on the abscissa) would look ridiculous. The studies with benzyl acetate, cinnamyl anthranilate, ethyl acrylate, and estragole have only two data points and therefore the straight line drawn through these points is completely arbitrary. A line of any shape, of course, can be drawn through any two points. These studies are included for several reasons. One is that the slopes and intercepts are so similar to the studies with three or four points. Also, some of the data points are in the low percentage of tumors (below 16%), which is where many conventional extrapolations deviate exponentially to the left. These did not and therefore are in agreement with and support the studies with more data points. The fact that all of these curves have similar slopes and intersections with the abscissa in the range of 1 ⫻ 10 20 molecules might suggest that somewhere near this concentration is necessary to displace the system from normal equilibrium and toward neoplasia. Confirmation of this, of course, awaits a wider and more complete analysis of animal carcinogenicity studies. Such an analysis is being planned. In conclusion, this analysis is interpreted to indicate that these flavoring agents have a clear threshold for carcinogenicity in animals that is well above the levels currently approved for use in foods; consequently, these animal studies should not be a cause for concern for carcinogenicity of these compounds in humans. Rather, the animal studies should be viewed as providing evidence for the safety of these compounds at current levels of human exposure.
ACKNOWLEDGMENTS The author gratefully acknowledges the assistance and helpful comments of Drs. Tim Adams, Fred Benz, John Doull, Karl Rozman, and members of the FEMA Expert Panel. This work was supported, in part, by Pharmacon Research Foundation, Inc.
REFERENCES Gaddum, J. H. (1945). Lognormal distributions. Nature 156, 463. Hayes, A. W., Ed. (2001). Principles and Methods of Toxicology, 4th ed. Taylor & Francis, Philadelphia, PA. Klaassen, C. D., Ed. (2001). Casarett and Doull’s Toxicology; The Basic Science of Poisons, 6th ed. McGraw-Hill, New York. Miller, E. C., Swanson, A. B., Phillips, D. H., Fletcher, T. L., Liem, A., and Miller, J. A. (1983). Structure-activity studies of the carcinogenicities in the mouse and rat of some naturally occurring and synthetic alkenylbenzene derivatives related to safrole and estragole. Cancer Res. 43, 1124 –1134. NTP (1980). Toxicology and carcinogenesis studies of cinnamyl anthranilate in F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 196. Available online at http://ntp-server.niehs.nih.gov. NTP (1986a). Toxicology and carcinogenesis studies of benzyl acetate in F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 250. Available online at http://ntp-server.niehs.nih.gov. NTP (1986b). Toxicology and carcinogenesis studies of ethyl acrylate in
F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 259. Available online at http://ntp-server.niehs.nih.gov. NTP (2000a). Toxicology and carcinogenesis studies of pyridine in F344/N rats, Wistar rats, and B6C3F 1 mice. National Toxicology Program, NTP TR 470. Available online at http://ntp-server.niehs.nih.gov. NTP (2000b). Toxicology and carcinogenesis studies of methyleugenol in F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 491. Available online at http://ntp-server.niehs.nih.gov. NTP (2001a). Toxicology and carcinogenesis studies of citral in F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 505. Draft available online at http://ntp-server.niehs.nih.gov. NTP (2001b). Toxicology and carcinogenesis studies of 2,4-hexadienal in F344/N rats and B6C3F 1 mice. National Toxicology Program, NTP TR 509. Draft available online at http://ntp-server.niehs.nih.gov. ¨ sterle, D., Deml, E., VilukRozman, K. K., Kerecsen, L., Viluksela, M. K., O sela, M., Stahl, B. U., Greim, H., and Doull, J. (1996). A toxicologist’s view of cancer risk assessment. Drug Metab. Rev. 28, 29 –52. Schro¨ dinger, E. (1944). What is Life?: The Physical Aspect of the Living Cell. Cambridge University Press, Cambridge. Smith, R. L., Adams, T. B., Doull, J., Feron, V. J., Goodman, J. I., Marnett, L. J., Portoghese, P. S., Waddell, W. J., Wagner, B. M., Rogers, A. E., Caldwell, J., and Sipes, I. G. (in press). Safety assessment of allylalkoxybenzene derivatives used as flavouring substances—methyleugenol and estragole. Food Chem. Toxicol. Waddell, W. J., and Bates, R. G. (1969). Intracellular pH. Physiol. Rev. 49, 285–329.