Food Chemical Risk Analysis, David R. Tennant, ed., New York: Chapman and Hall, 1997, pp. 267-295
Lois Swirsky Gold1,2, Thomas H. Slone1,2, and Bruce N. Ames2
1Life Sciences Division, E.O. Lawrence Berkeley National Laboratory, Berkeley, California 94720 (U.S.A.)
2Division of Biochemistry and Molecular Biology, University of California, Berkeley, California 94720 (U.S.A.)
Epidemiological studies have identified several factors that are likely to have a major effect on reducing rates of cancer: reduction of smoking, increased consumption of fruits and vegetables, and control of infections. Other factors include avoidance of intense sun exposure, increased physical activity, reduction of high occupational exposures, and reduced consumption of alcohol and possibly red meat. Risks of many forms of cancer can already be lowered, and the potential for further risk reduction is great. In the United States cancer death rates for all cancers combined are decreasing if lung cancer -- 90% of which is due to smoking -- is excluded from the analysis (Ames, Gold and Willet, 1995). The focus of this chapter is prioritization of possible cancer hazards in the diet.
Doll and Peto (1981) estimated that 35% of cancer was due to dietary factors, and the plausible contribution ranged from 10-70%. We have reviewed the more recent epidemiological literature (Ames et al., 1995), which generally supports the earlier estimate with a slightly narrower estimated range of 20-40% (Ames, Gold, and Willett, 1995). Current research on diet and cancer is slowly clarifying specific risk factors. New data have most strongly emphasized the inadequate consumption of protective factors rather than the excessive intake of harmful factors. The estimate for the contribution of dietary factors has been narrowed slightly downward largely because the large international contrasts in colon cancer rates are probably due, in addition to diet, to differences in physical activity, which is inversely related to colon cancer risk in many studies (Gerhardsson, Floderus, and Norell, 1988; Slattery et al., 1988; Thun et al., 1992). For breast cancer, the Doll and Peto estimate for the dietary contribution of 50% is still plausible, although that may not be avoidable in a practical sense if rapid growth rates are the most important underlying nutritional factor.
Adequate consumption of fruits and vegetables is associated with a lowered risk of degenerative diseases such as cancer (Ames, Shigenaga, and Hagen, 1993). A review of nearly 200 studies in the epidemiological literature showed that the lack of adequate consumption of fruits and vegetables is consistently related to cancer (Block, Patterson, and Subar, 1992; Hill, Giacosa, and Caygill, 1994; Steinmetz and Potter, 1991). The quarter of the population with the lowest dietary intake of fruits and vegetables has roughly twice the cancer rate for many types of cancer (lung, larynx, oral cavity, esophagus, stomach, colon and rectum, bladder, pancreas, cervix, and ovary) compared with the quarter with the highest consumption of those foods. The protective effect of consuming fruits and vegetables is weaker and less consistent for hormonally-related cancers, such as breast cancer. Laboratory studies suggest that antioxidants such as vitamins C and E and carotenoids in fruits and vegetables account for a good part of their beneficial effect (Ames, Shigenaga, and Hagen, 1993). Present epidemiological evidence regarding the role of greater antioxidant consumption in human cancer prevention is inconsistent. Nevertheless, biochemical data indicate the need for further investigation of the wide variety of potentially effective antioxidants, both natural and synthetic. Evidence supporting this need includes the enormous oxidative damage to DNA, proteins, and lipids (Ames, Shigenaga, and Hagen, 1993), as well as indirect evidence such as heightened oxidative damage to human sperm DNA when dietary ascorbate is insufficient (Fraga et al., 1991).
A wide array of micronutrients and other compounds in fruits and vegetables, in addition to antioxidants, may contribute to the reduction of cancer. Folic acid may be particularly important. Low folic acid intake causes chromosome breaks in rodents (MacGregor et al., 1990) and in humans (Blount et al., 1997; Everson et al., 1988), and increases tumor incidence in some rodent models (Bendich and Butterworth Jr., 1991). Folic acid is essential for the synthesis of DNA.
In rodents a calorie-restricted diet compared to ad libitum feeding markedly decreases tumor incidence and increases lifespan (Hart, Neumann, an Robertson, 1995; Pariza and Boutwell, 1987; Roe, 1989; Roe et al., 1991). Protein restriction appears to have a similar effect on rodents as calorie restriction, although research is less extensive on protein restriction (Youngman, Park, and Ames, 1992). An understanding of mechanisms for the marked effect of dietary restriction on aging and cancer is becoming clearer and may, in good part, be due to reduced oxidative damage and reduced rates of cell division. Although epidemiological evidence on restriction in humans is sparse, two types of epidemiological evidence support the possible importance of growth in the incidence of human cancer: studies indicating higher rates of breast and other cancers among taller persons (Hunter and Willett, 1993; Swanson et al., 1988) and studies of Japanese women (who are now taller and menstruate earlier) indicating increased breast cancer rates. Also, many of the variations in breast cancer rates among countries and trends over time within countries are compatible with changes in growth rates and attained adult height (Willett and Stampfer, 1990).
Although epidemiological studies most clearly support the benefits of fruits and vegetables in the prevention of cancer, strong international correlations suggest that animal (but not vegetable) fat and red meat may increase the incidence of cancers of the breast, colon, and prostate (Armstrong and Doll, 1975). However, large prospective studies have consistently shown either a weak association or a lack of association between fat intake and breast cancer (Hunter and Willett, 1993). Consumption of animal fat and red meat have been associated with risk of colon cancer in many case-control and cohort studies; the association with meat consumption appears more consistent (Giovannucci et al., 1994; Goldbohm et al., 1994; Willett and Stampfer, 1990). Consumption of animal fat and red meat (Hunter and Willett, 1993; Swanson et al., 1988) has also been associated with risk of prostate cancer (Giovannucci et al., 1994; Le Marchand et al., 1994). Mechanisms for those associations are not clear, but they may include the effects of dietary fats on endogenous hormone levels (Henderson, Ross, and Pike, 1991), the local effects of bile acids on the colonic mucosa, the effects of carcinogens produced in the cooking of meat, and excessive iron intake.
Alcoholic beverages cause inflammation and cirrhosis of the liver, and liver cancer (IARC, 1988). Alcohol is an important cause of oral and esophageal cancer, is synergistic with smoking (IARC, 1988), and possibly contributes to colorectal cancer (Freudenheim et al., 1991; Giovannucci et al., 1995).
Epidemiological studies do not support the idea that synthetic industrial chemicals are causing a significant amount of human cancer. Although some epidemiological studies find an association between cancer and low levels of industrial pollutants, the associations are usually weak, the results are usually conflicting, and the studies do not correct for diet, which is a potentially large confounding factor. Outside the workplace, the levels of exposure to synthetic pollutants are low and rarely seem plausible as a causal factor when compared to the wide variety of naturally occurring chemicals to which all people are exposed. (See below) (Gold et al., 1992).
Mechanistic studies of carcinogenesis indicate an important role of endogenous oxidative damage to DNA that is balanced by elaborate repair and defense processes, some of which are dietary protective agents. Also key is the rate of cell division (which is influenced by hormones, growth, cytotoxicity, and inflammation) since this determines the probability of converting DNA lesions to mutations. These mechanisms may underlie many epidemiological observations.
Current regulatory policy to reduce cancer risk, is based on the idea that chemicals which induce tumors in rodent cancer tests are potential human carcinogens; however, the chemicals tested for carcinogenicity in rodents have been primarily synthetic (Ames and Gold, 1990; Gold et al., 1984, 1986, 1987, 1990, 1993, 1995, 1997). The enormous background of human exposures to natural chemicals has not been systematically examined. This has led to an imbalance in both data and perception about possible carcinogenic hazards to humans from chemical exposures. The regulatory process does not take into account: 1) that natural chemicals make up the vast bulk of chemicals humans are exposed to; 2) that the toxicology of synthetic and natural toxins is not fundamentally different; 3) that about half of the chemicals tested, whether natural or synthetic, are carcinogens when tested using current experimental protocols; 4) that testing for carcinogenicity at near-toxic doses in rodents does not provide enough information to predict the excess number of human cancers that might occur at low-dose exposures; 5) that testing at the maximum tolerated dose (MTD) frequently can cause chronic cell killing and consequent cell replacement (a risk factor for cancer that can be limited to high doses), and that ignoring this effect in risk assessment greatly exaggerates risks.
The vast proportion of chemicals to which humans are exposed are naturally-occurring. Yet public perceptions tend to identify chemicals as being only synthetic and only synthetic chemicals as being toxic; however, every natural chemical is also toxic at some dose. We estimate that the daily average American exposure to burnt material in the diet is about 2000 mg, and to natural pesticides (the chemicals that plants produce to defend themselves against fungi, insects, and animal predators) about 1500 mg (Ames, Profet, and Gold, 1990b). In comparison, the total daily exposure to all synthetic pesticide residues combined is about 0.09 mg based on the sum of residues reported by the US Food and Drug Administration (FDA) in their study of the 200 synthetic pesticide residues thought to be of greatest concern (US Food and Drug Administration, 1993). We estimate that humans ingest roughly 5,000 to 10,000 different natural pesticides and their breakdown products (Ames, Profet, and Gold, 1990b). Despite this enormously greater exposure to natural chemicals, among the chemicals tested for carcinogenicity, 78% (1007/1298) are synthetic (i.e. do not occur naturally).
It has often been assumed that humans have evolved defenses against natural chemicals that will not protect against synthetic chemicals. However, humans, like other animals, are extremely well protected by defenses that are mostly general rather than specific for particular chemicals (e.g. continuous shedding of surface cells that are exposed). Additionally, most defense enzymes are inducible, and are effective against both natural and synthetic chemicals including potentially mutagenic reactive chemicals (Ames, Profet, and Gold, 1990b).
Since the toxicology of natural and synthetic chemicals is similar, one expects, and finds, a similar positivity-rate for carcinogenicity among synthetic and natural chemicals. Among chemicals tested in rats and mice in our Carcinogenic Potency Database (CPDB) (Gold et al., 1984, 1986, 1987, 1990, 1993, 1995, 1997), about half of the natural chemicals are positive as are half of all chemicals tested. Cooking food produces numerous by-products. Concentrations of natural pesticides in plants are usually measured in parts per thousand or million rather than parts per billion, which is the usual concentration of synthetic pesticide residues or water pollutants. Therefore, since humans are exposed to so many more natural than synthetic chemicals (by weight and by number), human exposure to natural rodent carcinogens, as defined by high dose rodent tests, is ubiquitous (Ames, Profet, and Gold, 1990a).
It is probable that almost every fruit and vegetable in the supermarket contains natural pesticides that are rodent carcinogens, and no diet can be free of chemicals identified as carcinogens in high-dose rodent tests. Even though only a tiny proportion of natural pesticides have been tested for carcinogenicity, 35 of 64 that have been tested are rodent carcinogens (Table 1) and occur in the following 79 common plant foods and spices: alcoholic beverages, allspice, anise, apple, apricot, banana, basil, beet, broccoli, Brussels sprouts, cabbage, cantaloupe, caraway, cardamom, carrot, cauliflower, celery, cherries, chili pepper, chocolate, cinnamon, cloves, cocoa, coffee, collard greens, comfrey herb tea, coriander, currants, dill, eggplant, endive, fennel, garlic, grapefruit, grapes, guava, honey, honeydew melon, horseradish, kale, lemon, lentils, lettuce, licorice, lime, mace, mango, marjoram, mushrooms, mustard, nutmeg, onion, orange, paprika, parsley, parsnip, peach, pear, peas, black pepper, pineapple, plum, potato, radish, raspberries, rhubarb, rosemary, rutabaga, sage, savory, sesame seeds, soybean, star anise, tarragon, tea, thyme, tomato, turmeric, and turnip.
Humans also ingest large numbers of natural chemicals from cooking food. For example, more than 1000 chemicals have been identified in roasted coffee. Only 28 have been tested for carcinogenicity according to the most recent results in our CPDB, and 19 of these are positive in at least one test (Table 2) totaling at least 10 mg of rodent carcinogens per cup (Clarke and Macrae, 1988; Fujita et al., 1985; Kikugawa, Kato, and Takahashi, 1989; Maarse et al., 1994). Among the rodent carcinogens in coffee are the plant pesticides caffeic acid (present at 1800 ppm) (Clarke and Macrae, 1988) and catechol (present at 100 ppm) (Rahn and König, 1978; Tressl et al., 1978). Two other plant pesticides, chlorogenic acid and neochlorogenic acid (present at 21,600 ppm and 11,600 ppm respectively) (Clarke and Macrae, 1988) have not been tested for carcinogenicity. Chlorogenic acid and caffeic acid are mutagenic (Ariza et al., 1988; Fung et al., 1988; Hanham, Dunn, and Stich, 1983), and clastogenic (Ishidate, Jr., Harnois, and Sofuni, 1988; Stich et al., 1981). For another plant pesticide in coffee, d-limonene, data are available on mechanism of carcinogenicity that suggest the rodent results are not relevant to humans because carcinogenicity in the male rat kidney is associated with a urinary protein that humans do not excrete (Dietrich and Swenberg, 1991). Some other rodent carcinogens in coffee are products of cooking, e.g. furfural and benzo(a)pyrene. The point here is not to indicate that rodent data necessarily implicate coffee as a risk factor for human cancer, but rather to illustrate that there is an enormous background of chemicals in the diet that are natural and that have not been a focus of attention for carcinogenicity testing. A diet free of naturally-occurring chemicals that are rodent carcinogens, is impossible.
Since the results of high-dose rodent tests are routinely used to identify a chemical as a possible cancer hazard to humans, it is important to try to understand how representative the 50% positivity rate might be of all the untested chemicals. If half of all chemicals (both natural and synthetic) to which humans are exposed would be positive if tested, then the utility of a test to identify a chemical as a "potential human carcinogen" is questionable. To determine the true proportion of rodent carcinogens among chemicals would require a comparison of a random group of synthetic chemicals to a random group of natural chemicals. Such an analysis has not been done. We have found that the high positivity rate is consistent for several datasets: among chemicals tested in rats and mice, 59% (330/559) are positive in at least one experiment, 59% of synthetic chemicals (257/432), and 57% of naturally occurring chemicals (73/127). Among chemicals tested in at least one species, 55% of natural pesticides (35/64) are positive, 61% of fungal toxins (14/23) and 68% of the chemicals in roasted coffee (19/28) [Table 2]. Additionally, in the Physician's Desk Reference, 49% (117/241) of the drugs for which animal cancer tests are reported are carcinogenic (Davies and Monro, 1995).
It has been argued that the high positivity rate is due to selecting more suspicious chemicals to test for carcinogenicity. For example, chemicals may be selected that are structurally similar to known carcinogens or chemicals. That is a likely bias, since cancer testing is both expensive and time-consuming, and it is prudent to test suspicious compounds. On the other hand, chemicals are selected for testing for several reasons, including the extent of human exposure, level of production, and scientific questions about carcinogenesis. Although mutagens are positive in rodent bioassays more frequently than non-mutagens (79% of mutagens versus 49% of non-mutagens are positive), among the chemicals tested in rats and mice, 55% are non-mutagens; this suggests that the prediction of positivity may often not be the basis for selecting a chemical to test. Moreover, while some chemical classes are more often carcinogenic in rodent bioassays than others, e.g., nitroso compounds, aromatic amines, nitroaromatics and chlorinated compounds -- prediction is still imperfect (Omenn, Stuebbe, and Lave, 1995).
One large series of mouse experiments by Innes et al. (1969) has been frequently cited (US National Cancer Institute, 1986) as evidence that the true proportion of rodent carcinogens is actually low among tested substances. In the Innes study, among 119 chemicals tested -- primarily the most widely used pesticides at that time and some industrial chemicals -- only 11 (9%) were judged as carcinogens. We note that those early experiments lacked power to detect an effect because they were conducted only in mice (not in rats), they included only 18 animals in a group (compared with the usual 50), the animals were tested for only 18 months (compared with the usual 24 months), and the Innes dose was usually lower than the highest dose in subsequent mouse tests of the same chemical (Gold, Slone, and Ames, 1997).
To assess whether the low positivity rate in the Innes study may have been due to the design of the experiments, we used results in our CPDB to examine subsequent bioassays on the Innes chemicals that had not been evaluated as positive. Among 34 such chemicals that were subsequently retested, 16 had a subsequent positive evaluation of carcinogenicity (47%), which is similar to the proportion among all chemicals in our database. Of the 16 new positives, 6 were carcinogenic in mice and 12 in rats. Innes had recommended further evaluation of some chemicals that had inconclusive results in their study. If those were the chemicals subsequently retested, then one might argue that they would be the most likely to be positive. Our analysis does not support that view, however. We found that the positivity rate among the chemicals that the Innes study said needed further evaluation was 6 of 16 (38%) when retested, compared to 10 of 18 (56%) among the chemicals that Innes evaluated as negative.
We have argued that mutagenesis, and thus carcinogenesis, is increased by increasing either DNA damage or cell division in cells that are not discarded. There is enormous endogenous DNA damage from normal oxidation, and the evidence suggests that oxidative damage is a major factor not only in aging, but in the degenerative diseases of aging such as cancer (Ames, Shigenaga, and Hagen, 1993). The steady-state level of oxidative damage in DNA is about one million oxidative lesions per rat cell (Ames, Shigenaga, and Hagen, 1993). Thus, this high background suggests that the cell division rate must be a factor in converting lesions to mutations and thus cancer (Ames, Shigenaga, and Gold, 1993). Raising the level of either DNA lesions or cell division will increase the probability of cancer. Just as DNA repair protects against lesions, p53 guards the cell cycle and protects against cell division if the lesion level gets too high; however, neither defense is perfect (Ames, Gold, and Willett, 1995). Cell division is also a major factor in loss of heterozygosity through non-disjunction and other mechanisms (Ames and Gold, 1990; Ames, Gold, and Willett, 1995).
A plausible explanation for the high positivity rate in rodent bioassays, which is supported by an ever increasing array of papers, is that the MTD of a chemical can cause chronic cell killing and cell replacement in the target tissue, a risk factor for cancer that can be limited to high doses. Thus it seems likely that the high positivity rate in standard rodent bioassays at the MTD will be primarily due to the effects of high doses for the non-mutagens, and to a synergistic effect of cell division at high doses with DNA damage for the mutagens. Ad libitum feeding in the standard bioassay can also contribute to the high positivity rate (Hart, Neumann, and Robertson, 1995), plausibly by increased cell division due to high caloric intake (Ames, Shigenaga, and Gold, 1993; Hart, Neumann, and Robertson, 1995).
Although cell division is not measured in routine cancer tests, many studies on rodent carcinogenicity show a correlation between cell division at the MTD and cancer. Cunningham and co-workers have analyzed 15 chemicals at the MTD, 8 mutagens and 7 non-mutagens, including several pairs of mutagenic isomers, one of which is a rodent carcinogen and one of which is not (Cunningham et al., 1995; Hayward et al., 1995). A perfect correlation was observed: the 9 chemicals causing cancer caused cell division in the target tissue and the 6 chemicals not causing cancer did not. A similar result has been found in the analyses of Mirsalis et al. (1993), e.g. both dimethyl nitrosamine (DMN) and methyl methane sulfonate (MMS) methylate liver DNA and cause unscheduled DNA synthesis, but DMN causes both cell division and liver tumors, while MMS does neither. At high doses, chloroform induces liver cancer (Larson, Wolf, and Butterworth, 1994) and sodium saccharin induces bladder cancer by chronic cell division (Cohen and Lawson, 1995). Extensive reviews on rodent studies (Gold, Slone, and Ames, 1997; Ames and Gold, 1990; Ames, Shigenaga, and Hagen, 1993; Cohen and Ellwein, 1991; (Cohen, 1995; (Cohen and Lawson, 1995; Counts and Goodman, 1995) document that chronic cell division can induce cancer. A large epidemiological literature reviewed by Preston-Martin et al. (1990, 1995) indicates that increased cell division by hormones and other agents can increase human cancer.
Several of our findings in large-scale analyses of the results of animal cancer tests (Gold, Manley, and Ames, 1992), are consistent with the idea that cell division increases the carcinogenic effect in high dose bioassays, including: the high proportion of chemicals that are positive; the high proportion of rodent carcinogens that are not mutagenic; the fact that mutagens, which can both damage DNA and increase cell division at high doses, are more likely than non-mutagens to be positive, to induce tumors in both rats and mice, and to induce tumors at multiple sites. Analyses of the limited data on dose-response in bioassays are consistent with the idea that cell division from cell-killing and cell replacement is important. In the usual experimental design of dosing at the MTD and half MTD, both doses are high and may result in cell division. Even at these two high doses, about half of the positive sites in NTP bioassays are statistically significant at the MTD but not at half the MTD (Gold, Manley, and Ames, 1992).
To the extent that increases in tumor incidence in rodent studies are due to the secondary effects of inducing cell division at the MTD, then any chemical is a likely rodent carcinogen, and carcinogenic effects can be limited to high doses. Thus, true risks at the low doses of most human exposures to the general population are likely to be much lower than what would be predicted by the linear model that is the default in US regulatory risk assessment. The true risk might often be zero.
We have discussed validity problems associated with the use of the limited data from animal cancer tests for human risk assessment. Adequate risk assessment from animal cancer tests requires more information about many aspects of toxicology for each chemical than the limited data now available from standard bioassays, such as effects on cell division, induction of defense and repair systems, and species differences. Standard practice in regulatory risk assessment for a given rodent carcinogen is to extrapolate from the high doses of rodent bioassays to the low doses of most human exposures by multiplying carcinogenic potency in rodents by human exposure. Strikingly, however, since potency estimates are constrained to lie within a narrow range about the MTD (Bernstein et al., 1985; Freedman, Gold, and Slone, 1993; Gold, Slone, and Ames, 1997), the dose usually estimated by regulatory agencies to give one cancer in a million, can be approximated simply by using the MTD as a surrogate for carcinogenic potency. The "virtually safe dose" (VSD) can be approximated from the MTD. Gaylor and Gold (1995) used the ratio MTD/TD50 and the relationship between q1* and TD50 found by Krewski et al. (1993) to estimate the VSD. The VSD was approximated by the MTD/740,000 for NCI/NTP rodent carcinogens. This result questions the utility of bioassay results to estimate risk, and demonstrates the limited information about risk that is provided by bioassay results. The MTD/740,000 was within a factor of 10 of the VSD for 96% of carcinogens. Without data on mechanism of carcinogenesis for a given chemical, the true risk of cancer at low dose is highly uncertain, and could be zero, even for rats or mice.
Given the limited information from rodent bioassays about mechanisms of carcinogenesis and low-dose risk, as well as the fact that there is an imbalance in bioassay data because the vast proportion of test agents are synthetic chemicals while the vast proportion of human exposures are to naturally occurring chemicals, what is the best use that can be made of bioassay results in efforts to prevent human cancer? In several papers we have emphasized that it is important to set research and regulatory priorities about cancer prevention by gaining a broad perspective about the vast number of chemicals to which humans are exposed. One reasonable strategy is to use a rough index to compare and rank possible carcinogenic hazards from a wide variety of chemical exposures at levels that humans typically receive, and then to focus on those that rank highest (Ames, Magaw and Gold, 1987; Gold et al., 1992; Gold, Slone, Manley, et al., 1994; Gold, Slone, and Ames, 1997). Ranking is a critical first step that can help to set priorities when selecting chemicals for chronic bioassay or mechanistic studies, for epidemiological research, and for regulatory policy. Although one cannot say whether the ranked chemical exposures are likely to be of major or minor importance in human cancer, it is not prudent to focus attention on the possible hazards at the bottom of a ranking if, using the same methodology, there are numerous common human exposures with much greater possible hazards. In earlier papers we ranked possible hazards from a variety of typical human exposures to rodent carcinogens. The analyses are based on the HERP index (Human Exposure/Rodent Potency), which indicates what percentage of the rodent carcinogenic potency (TD50 in mg/kg/day) a human receives from a given daily lifetime exposure (mg/kg/day). TD50 is the daily lifetime dose rate estimated to halve the proportion of tumor-free animals by the end of a standard lifetime (Peto et al., 1984). TD50 values in our Carcinogenic Potency Database (CPDB) span a 10 million-fold range across chemicals. In general, the ranking by HERP is expected to be similar to a ranking of "risk estimates" using current regulatory risk assessment methodology for the same exposures, since linear extrapolation from the TD50 generally leads to low-dose slope estimates similar to those based on the linearized multistage model (Krewski, Szyszkowicz, and Rosenkranz, 1990). As we discussed above, the VSD is approximately equivalent to the ratio of the high dose in a bioassay divided by 740,000 (Gaylor and Gold, 1995).
Our earlier analyses indicated that some historically high exposures in the workplace and some pharmaceuticals rank high, and that there is an enormous background of naturally occurring rodent carcinogens in typical portions of common foods that cast doubt on the relative importance of low-dose exposures to synthetic chemicals such as pesticide residues (Ames, Magaw and Gold, 1987; Gold et al., 1992; Gold, Slone, Manley, et al., 1994). In this chapter we address the relative ranking by HERP of average US dietary exposures to rodent carcinogens that either occur naturally in food, are products of cooking and food preparation, or are present in food as residues of synthetic pesticides, food additives, or contaminants. In order to calculate HERP, in addition to TD50, data are required on both concentration of a chemical in food and the average consumption of the food. We have tried to include as many chemicals as possible by calculating HERP for all chemicals for which we have been able to obtain reliable average dietary exposure data, for both natural and synthetic chemicals.
The average daily exposures in the ranking (Table 3) are ordered by possible carcinogenic hazard (HERP). Results are reported for average exposures to 25 natural chemicals in the diet (in boldface), and to 20 synthetic chemicals. A few convenient reference points are: The median HERP value in Table 3 of 0.0007%; the upper bound risk estimate used by regulatory agencies of one in a million (using the q1* potency value derived from the linearized multistage model), i.e. the VSD, which converts to a HERP of 0.00003% if based on a rat TD50 and 0.00001% if based on a mouse TD50; and the background HERP of 0.0003% for the average chloroform level in a liter of US tap water, which is formed as a by-product of chlorination.
The ranking maximizes the HERP values for synthetic compared to natural chemicals because we have reported historically high values for exposures that may now be much lower, e.g. DDT, PCBs, and because all exposures to synthetic chemicals are averages in the total diet, whereas for many natural chemicals the exposures are for individual foods (for which concentration data were available).
Table 3 indicates that many ordinary foods would not pass the regulatory criteria used for synthetic chemicals. For many natural chemicals the HERP values are in the top half of the table, and natural chemicals are markedly underrepresented because so few have been tested in rodent bioassays. We discuss several categories of exposure below and indicate that for some chemicals mechanistic data are available that suggest that the chemical would not be expected to be a cancer hazard at the doses to which humans are exposed, and their ranking by HERP would not be relevant in risk assessment.
These are markedly underrepresented in our analysis compared to synthetic pesticide residues because few natural chemicals have been tested for carcinogenicity. Importantly, for each plant food listed, there are about 50 additional untested natural pesticides. Although ~10,000 natural pesticides and their break-down products occur in the human diet (Ames, Profet, and Gold, 1990a), only 63 have been tested adequately in rodent bioassays (Table 1). Average exposures to many natural pesticide rodent carcinogens in common foods rank above or close to the median, ranging up to a HERP of 0.1%. These include caffeic acid (lettuce, apple, pear, coffee, plum, celery, carrot, potato); safrole (in spices), allyl isothiocyanate (mustard), dlimonene (mango, orange juice, black pepper); estragole (in spices); hydroquinone and catechol in coffee; and coumarin in cinnamon. Caffeic acid is more widespread in plant species than other natural pesticides. Some natural pesticides in the commonly eaten mushroom (Agaricus bisporus) are rodent carcinogens (glutamyl-p-hydrazinobenzoate, p-hydrazinobenzoate), and the HERP based on feeding whole mushrooms to mice is 0.02%. For d-limonene, no human risk is anticipated because tumors are induced only in male rat kidney tubules with involvement of alpha2u-globulin nephrotoxicity, which does not appear to be possible in humans (USEPA, 1991; Hard and Whysner, 1994)
Synthetic pesticides currently in use that are rodent carcinogens and quantitatively detected by the US FDA as residues in food, are all included in Table 3. Most are at the bottom of the ranking, but HERP values are just above the median for ethylene thiourea (ETU), before recetn discontinuance on some crops UDMH (from Alar) before its discontinuance, and DDT before its ban in the U.S. in 1972. These rank below the HERP values for many naturally occurring chemicals. The HERP value for ETU would be about 10 times lower if the US EPA potency value were used instead of our TD50; EPA combined rodent results from more than one experiment, including one with lower doses of ETU and administration in utero, and obtained a lower potency (US Environmental Protection Agency, 1992). DDT and similar, early pesticides, have been a concern because of their unusual lipophilicity and persistence, although there is no convincing epidemiological evidence of a carcinogenic hazard (Key and Reeves, 1994). Current exposure to DDT is in foods of animal origin, and the HERP value is 0.00008%.
In 1984 the US EPA banned the agricultural use of ethylene dibromide (EDB) the main fumigant in the U.S., because of the residue levels found in grain, HERP = 0.0004%. This HERP value is 350,000 times lower than the HERP of 140% for the high exposures that some workers received in the 1970s (Gold et al., 1992).
Three synthetic pesticides, captan, chlorothalonil, and folpet, were evaluated in 1987 by the National Research Council (NRC) as being of relatively high risk to humans (National Research Council, 1987), and were also reported by FDA in the Total Diet Study (TDS). The contrast between the extremely low HERP values for these exposures (chlorothalonil = 0.0000001%, folpet = 0.000000008%, captan = 0.000000006%) and the high risk estimates of the 1987 NRC report (which differ by a factor of 99,000 for chlorothalonil, 463,000 for folpet, and 116,000 for captan) is because the exposure estimates used by the NRC (i.e. the EPA Theoretical Maximum Residue Contribution) are hypothetical maximum exposure estimates whereas the FDA monitors the actual food supply to estimate dietary intakes of pesticides. Hence, using hypothetical maxima, results in enormously higher risk estimates than using measured residues (See chapter 14 in this volume by F.F. Busta and C. Chaisson.)
This can also produce chemicals that are rodent carcinogens. Alcoholic beverages are a human carcinogen (see "Other aspects of diet" above), and the HERP values in Table 3 for U.S. average exposure to alcohol in beer (2.1%) and wine (0.5%) are at the top of the ranking. Ethyl alcohol is one of the least potent rodent carcinogens in our Carcinogenic Potency Database, but the HERP is high because of high concentrations and high U.S. consumption (average daily consumption of ethyl alcohol in beer in the U.S. is 13 ml). Another fermentation product, urethane (ethyl carbamate), has a HERP value of 0.00001% in average beer consumption; in a daily 2 slices of whole wheat toast, the HERP would be 0.00003%).
Cooking food is plausible as a contributor to cancer. A wide variety of chemicals are formed during cooking. Rodent carcinogens formed include furfural and similar furans, nitrosamines, polycyclic hydrocarbons, and heterocyclic amines. Furfural, a chemical formed naturally when sugars are heated, is a widespread constituent of food flavor. The HERP value for furfural in average consumption of coffee is 0.02% and in white bread is 0.004%. Nitrosamines formed from nitrite or nitrogen oxides (NOx) and amines in food can give moderate HERP values, e.g. in bacon, the HERP for diethylnitrosamine is 0.0007% for and dimethylnitrosamine is 0.0004%. A variety of mutagenic and carcinogenic heterocyclic amines (HA) are formed when meat, chicken or fish are cooked, particularly when charred. Compared to other rodent carcinogens, there is strong evidence of carcinogenicity for HAs in terms of positivity rates and multiplicity of target sites; however, concordance in target sites between rats and mice is generally restricted to the liver (Gold, Slone, Manley, et al., 1994). Under usual cooking conditions, exposures to HA are in the low ppb range. HERP values for HA in pan fried hamburger range from 0.00006% for PhIP to 0.000005% for IQ (Table 3). PhIP induces colon tumors in male but not female rats. A recent study indicates that whereas the level of DNA adducts in the colonic mucosa was the same in both sexes, cell proliferation was increased only in the male, contributing to the formation of premalignant lesions of the colon (Ochiai et al., 1996). Therefore, there was no correlation between adduct formation and premalignant lesions, but there was between cell division and lesions.
These can be either naturally occurring rodent carcinogens (e.g. allyl isothiocyanate and alcohol) or synthetic rodent carcinogens (butylated hydroxyanisole [BHA] and saccharin, Table 3). The highest HERP values for average exposures to synthetic rodent carcinogens in Table 3 are for exposures in the 1970s to BHA (0.009%) and saccharin (0.005%), both nongenotoxic rodent carcinogens. For both of these additives, data on mechanism of carcinogenesis strongly suggest that there would be no risk to humans at the levels found in food.
BHA is a phenolic antioxidant that is Generally Regarded as Safe (GRAS) by the US FDA. By 1987, after BHA was shown to be a rodent carcinogen, its use declined six fold (HERP=0.001%) [USFDA, 1991]; this was due to voluntary replacement by other antioxidants, and to the fact that the use of animal fats and oils, in which BHA is primarily used as an antioxidant, has consistently declined in the USA. The mechanistic and carcinogenicity results on BHA indicate that malignant tumors were induced only at a dose above the MTD at which cell division is increased in the forestomach, which is the only site of tumorigenesis; the proliferation is only at high doses, and is dependent on continuous dosing until late in the experiment (Clayson et al., 1990). Humans do not have a forestomach. We note that the dose-response for BHA curves sharply upward, but the potency value used in HERP is based on a linear model; if the California EPA potency value (which is based on a linearized multistage model) were used in HERP instead of TD50, the HERP values for BHA would be 25 times lower (California Environmental Protection Agency, 1994).
For saccharin, which has largely been replaced by other sweeteners, there is convincing evidence that the induced bladder tumors in rats are not relevant to human dietary exposures. The carcinogenic effect requires high doses of sodium saccharin which form calculi in the bladder, and subsequent regenerative hyperplasia. Thus tumor development is due to increased cell division, and if the dose is not high enough to produce calculi then there is no increased cell division and no increased risk of tumor development (Cohen and Lawson, 1995).
Of the 23 fungal toxins tested for carcinogenicity, 14 are positive (61%). The mutagenic mold toxin, aflatoxin, which is found in moldy peanut and corn products, interacts with chronic hepatitis infection in human liver cancer development (Qian et al., 1994). There is a synergistic effect in the human liver between aflatoxin (genotoxic effect) and the hepatitis B virus (cell division effect) in the induction of liver cancer (Wu-Williams, Zeise, and Thomas, 1992). The HERP value for aflatoxin of 0.008% is based on the rodent potency; if the lower human potency value calculated by the US FDA from epidemiological data were used instead, the HERP would be about 10-fold lower (US Food and Drug Administration, 1993b). Biomarker measurements of aflatoxin on populations in Africa and China, which have high rates of both hepatitis B and C viruses and liver cancer, confirm that those populations are chronically exposed to high levels of aflatoxin (Groopman et al., 1992; Pons Jr., 1979). Liver cancer is rare in the USA. Although hepatitis B and C viruses infect less than 1% of the U.S. population, hepatitis viruses can account for half of liver cancer cases among non-Asians (Yu et al., 1991).
Polychlorinated biphenyls (PCB) and tetrachlorodibenzo-p-dioxin (TCDD), which have been a concern because of their environmental persistence and carcinogenic potency in rodents, are primarily consumed in foods of animal origin. In the USA, PCBs are no longer used, but exposure persists from industrial products. Consumption in food in the USA declined about 20-fold between 1978 and 1986 (Gartrell et al., 1986; Gunderson, 1995). The HERP value for the most recent reporting of the US FDA Total Diet Study (1984-86) is 0.00008%, towards the bottom of the ranking, and far below many values for naturally occurring chemicals in common foods. It has been reported that some countries may have higher intakes of PCBs then in the USA (World Health Organization, 1993).
TCDD, the most potent rodent carcinogen, is produced naturally by burning when chloride ion is present, e.g. in forest fires. The sources of human exposure appear to be predominantly anthropogenic, e.g. from incinerators (US Environmental Protection Agency, 1994a). TCDD has received enormous scientific and regulatory attention, most recently in an ongoing assessment by the US EPA (US Environmental Protection Agency, 1994a, 1994b, 1995). Some epidemiologic studies suggest an association with human cancer, but the evidence is not sufficient to establish causality. Estimation of average U.S. consumption is based on limited sampling data, and the EPA is currently conducting further studies of concentrations in food. The HERP value of 0.0007% is at the median of the values in Table 3. TCDD exerts many or all of its harmful effects in mammalian cells through binding to the aryl hydrocarbon (Ah) receptor. A wide variety of natural substances also bind to the Ah receptor, (e.g. tryptophan oxidation products) and insofar as they have been examined, they have similar properties to TCDD (Ames, Profet, and Gold, 1990b). For example, a variety of flavones and other plant substances in the diet, such as indole carbinol (IC), also bind to the Ah receptor. IC is the main breakdown compound of glucobrassicin, a glucosinolate that is present in large amounts in vegetables of the Brassica genus, including broccoli (Bradfield and Bjeldanes, 1987).
Caution is necessary in drawing conclusions from the occurrence in the diet of natural chemicals that are rodent carcinogens. It is not argued here that these dietary exposures are necessarily of much relevance to human cancer. In fact, the epidemiological results discussed above indicate that adequate consumption of fruits and vegetables reduces cancer risk at many sites, and that protective factors like intake of vitamin C and folic acid are important, rather than intake of individual rodent carcinogens. Our analysis does indicate that widespread exposures to naturally-occurring rodent carcinogens cast doubt on the relevance to human cancer of low-level exposures to synthetic rodent carcinogens. Our results call for a re-evaluation of the utility of animal cancer tests done at the MTD for providing information that is useful in protecting humans against low-level exposures in the diet when a high percentage of both natural and synthetic chemicals appear to be rodent carcinogens at the MTD, when the data from rodent bioassays are not adequate to assess low-dose risk, and when the ranking on an index of possible hazards demonstrates that there is an enormous background of natural chemicals in the diet that rank high, even though so few have been tested in rodent bioassays.
Our discussion of the HERP ranking indicates the importance of data on mechanism of carcinogenesis for each chemical. For several chemicals, mechanistic data have recently been generated which indicates that they would not be expected to be a risk to humans at the levels consumed in food (e.g. saccharin, BHA, chloroform, d-limonene, discussed above). Recent developments in science and regulatory policy have also emphasized the importance of evaluating mechanistic data, rather than relying exclusively on default, worst-case assessments. The NRC's recent report Science and Judgment in Risk Assessment and the EPA's draft document Proposed Guidelines for Carcinogen Risk Assessment both recommend improvements in the risk assessment process that involve incorporating consideration of dose to the target tissue, mode of action, and biologically based dose-response models, including a possible threshold of dose below which effects will not occur (National Research Council, 1994; US Environmental Protection Agency, 1996).
Our analysis in this chapter suggests several areas for further research into diet and cancer, including epidemiological, toxicological, and biochemical investigations. Further understanding of the role and mechanism of endogenous damage could lead to new prevention strategies for cancer. Present epidemiological evidence regarding the role of greater antioxidant consumption in human cancer prevention is inconsistent. Nevertheless, biochemical data indicating massive oxidative damage to DNA, proteins and lipids, as well as indirect evidence such as increased oxidative damage to human sperm DNA with insufficient dietary ascorbate, indicate the need for further investigation of the wide variety of potentially effective antioxidants, both natural and synthetic. Additionally, studies on the importance of dietary fruits and vegetables in cancer suggest the importance of further work on micronutrient deficiency as a major contributor to cancer. Studies in rodents and humans suggest further work on caloric intake and body weight, and the effects on hormonal status.
Since naturally occurring chemicals in the diet have not been a focus of cancer research, it seems reasonable to investigate some of them further as possible hazards because they often occur at high concentrations in foods. Only a small proportion of the many chemicals to which humans are exposed will ever be investigated, and there is at least some toxicological plausibility that high dose exposures may be important. In order to identify untested dietary chemicals that might be a hazard to humans if they were to be identified as rodent carcinogens, we propose an index, HERT, which is analogous to HERP: the ratio of Human Exposure/Rodent Toxicity. HERT uses readily available LD50 values rather than the TD50 values from animal cancer tests that are used in HERP. This approach to prioritizing chemicals makes assessment of human exposure levels critical at the outset. The validity of the HERT approach is supported by 3 analyses: First, we have found that for the exposures to rodent carcinogens for which we have calculated HERP values (N=68), the ranking by HERP and HERT are highly correlated (Spearman rank order correlation = 0.89). Second, we have shown that without conducting a bioassay the regulatory VSD can be approximated by dividing the MTD by 740,000 (Gaylor and Gold, 1995). Since the MTD is not known for all chemicals, and MTD and LD50 are both measures of toxicity, acute toxicity (LD50) can reasonably be used as a surrogate for chronic toxicity (MTD). Third, we and others (Zeise et al., 1984) have found that LD50 and carcinogenic potency are correlated; therefore, HERT is a reasonable surrogate index for HERP since it simply replaces TD50 with LD50.
We have calculated HERT values using LD50 values as a measure of toxicity in combination with available data on concentrations of untested natural chemicals in commonly consumed foods and data on average consumption of those foods in the US diet. We considered any chemical with available data on rodent LD50, that had a published concentration >=10 ppm in a common food, and for which estimates of average US consumption of that food were available. Among the set of 171 HERT values we were able to calculate, the HERT ranged across 7 orders of magnitude from 0.000001 to 13.5. We report in Table 4 the top ranking HERT values for average exposures in the US diet. (Because the value is so high, we also include cassava, which is a staple in some parts of Africa and South America).
It might be reasonable to investigate further the chemicals in Table 4 in chronic carcinogenicity bioassays. For example, solanine and chaconine, the main alkaloids in potatoes, are cholinesterase inhibitors that can be detected in the blood of almost all people (Ames, 1983; Ames, 1984; Harvey et al., 1985).
Chlorogenic acid was clastogenic at a concentration of 150 ppm (Ames, Profet, and Gold, 1990a), which is 100 times less than its concentration in roasted coffee beans and similar to its concentration in apples, pears, plums, peaches, cherries and apricots. Chlorogenic acid and caffeic acid are also mutagens (Ames, Profet, and Gold, 1990a). The genotoxic activity of coffee to mammalian cells has been demonstrated (Tucker et al., 1989).
Cyanogenesis, the ability to release hydrogen cyanide, is widespread in plants, including several foods, of which the most widely eaten are cassava and lima bean (Poulton, 1983). Cassava is eaten widely throughout the tropics, and is a dietary staple for over 300 million people (Bokanga et al., 1994). There are few effective means of removing the cyanogenic glycosides that produce hydrogen cyanide, and cooking is generally not effective (Bokanga et al., 1994; Poulton, 1983). In mice, no standard lifetime studies of caffeine have been conducted. In rats, cancer tests of caffeine have been negative, but one study that was inadequate because of early mortality showed an increase in pituitary adenomas (Yamagami et al., 1983).
The chemicals in Table 4 might reasonably be evaluated further by the National Toxicology Program as candidates for further testing.
This work was supported through the Lawrence Berkeley Laboratory by US Department of Energy, contract DE-AC-03-76SFO0098, and through the University of California, Berkeley by National Institute of Environmental Health Sciences Center Grant ESO1896. We thank Neela B. Manley for comments on the manuscript, and Stuart W. Krasner for providing the disinfection byproducts database.
Ames, B.N. (1984) Cancer and diet. Science 224, 668-670, 757-760.
Clarke, R.J. and Macrae, R. (eds) (1988) Coffee. Vol. 1-3, Elsevier, New York.
Doll, R. and Peto, R. (1981) The Causes of Cancer. Oxford University Press, New York.
Gold, L.S., Slone, T.H., Stern, B.R., Manley, N.B., and Ames, B.N. (1992) Rodent carcinogens: Setting priorities. Science 258, 261-265. PDF
Technical Assessment Systems (1989) Exposure 1 Software Package. TAS, Washington, DC.
U.S. Food and Drug Administration (1992) Exposure to Aflatoxins. USFDA, Washington, DC.
Return to the Carcinogenic Potency Project Home Page:
Last updated: November 24, 1998