Igor Linkov#, Richard Wilson# and George M. Gray*

* To whom correspondence should be addressed

Harvard Center for Risk Analysis

Harvard School of Public Health

718 Huntington Ave.

Boston, MA 02115

phone: 617-432-4341

fax: 617-432-0190

ggray@hsph.harvard.edu

#Department of Physics

Harvard University

Jefferson Laboratories

Cambridge, MA 02138

Running Title: Anticarcinogenicity and weight or survival depression

Linkov, I., Wilson, R., and Gray, G.M. (1998)

It has been suggested that the decreased tumor rates (anticarcinogenicity) commonly observed in the National Toxicology Program (NTP) rodent bioassays may be caused by compound-induced decreases in body weight or decreases in survival of treated animals. In this study, weight decrement and survival depression following chemical treatment was studied for those chemicals which induce site specific decreases in tumor rates (anticarcinogens) and those which do not (non-anticarcinogens) in a database of 312 chemicals tested in the NTP bioassay program prior to 1983. There is an evident difference in weight depression and animal survival between anticarcinogens and non-anticarcinogens but it is small relative to the variability between chemicals in the two groups. We argue that weight or survival depression cannot, in a simple way, explain the difference between anticarcinogenic and non-anticarcinogenic chemicals. In fact, there are a number of chemicals that are anticarcinogenic with no evidence of weight or survival depression and many chemicals that cause significant weight or survival decreases with no apparent anticarcinogenic effects. There is a small but statistically insignificant relationship between the degree of weight depression and the number of tumor sites found to have lower tumor rates in treated animals. These analyses suggest that biological factors other than weight and survival depression are involved in decreased tumor rates in rodent bioassays. These results, and those of the companion paper (Linkov et al., 1998), suggest that the anticarcinogenic responses observed in rodent cancer bioassays should be carefully considered in evaluations of the overall carcinogenic potential of chemicals.

Long-term treatment of rodents with high doses of chemicals clearly has led to decreases in tumor rates at specific sites (e.g., Weinberg and Storer, 1985; Davies and Monro, 1994; Haseman and Johnson, 1996; Dunnick et al., 1996; Elwell et al., 1996; Linkov, et al., 1998). Random effects in the long-term rodent bioassays (random variations in the background rate and multiple statistical comparisons) seem to account for only a small portion of the observed anticarcinogenic effects if stringent levels of statistical significance are chosen (Linkov, et al., 1998).

It is well established that animals raised on calorically restricted diets show fewer tumors and other diseases as well as later onset of tumors (Hart et al., 1995a; Hart et al., 1995b). Several studies of National Toxicology Program (NTP) bioassays demonstrate a correlation between tumor incidence and animal body weight in control groups (Seilkop, 1995; Haseman et al., 1994). Although reduced body weight in the bioassay may not always be due to reduced caloric intake, these observations have led to the suggestion that the weight decreases induced by chemical treatment in long-term rodent bioassays may be responsible for the observed anticarcinogenic effects (Haseman and Johnson, 1996). Haseman and Johnson (1996) analyzed 218 recent two-year rodent bioassays conducted by the NTP. They concluded that "the studies with the most reduction in tumor incidence were generally those studies in which body weight decrement in the top-dose group relative to control was the greatest."

Chemically treated animals could lose weight because of changes in food consumption or changes in nutrient absorption or utilization. It is not clear whether the weight changes induced by high-dose chemical treatment are functionally equivalent to those caused by restricted caloric intake. The influence of body weight decreases on tumor responses is critical to interpretation of rodent bioassays since a weight decrement, usually of the order of 10%, is often considered necessary for proper conduct of a bioassay. This means, of course, that weight decrements are almost universal in rodent cancer bioassays.

Another possible explanation for the apparent decreases in tumor rates with chemical treatment is a reduced survival rate of treated animals. If chemical treatment leads to fewer animals surviving long enough to develop tumors, then the lower rates of specific tumor types in treated animals when compared to controls could lead to false findings of anticarcinogenicity.

One of the factors which complicates such analyses using the NTP database are historical changes in the background tumor rate and animal survival. Increased body weight and decreased survival of rodents over the last 20 years has been observed in many laboratories including those conducting NTP studies (Keenan et al., 1996, Haseman et al., 1994, Turturro et al., 1996). The average body weight for B6C3F1 mice was reported to increase from about 40g in 1981 to more than 50 g in 1988 (Haseman, 1994). The average background liver tumor rates for B6C3F1 mice increased from about 10% (female) and 35% (male) in 1981 to about 50% and 60% respectively by 1989 (Turturro et al., 1996). Clearly, these changes over time can lead to difficulties in interpretation of observed anticarcinogenic responses.

However, it has been found that body weight and survival show relatively little change over time in the early NTP studies (initiated in 1976-1981) (Rao et al, 1990a, Rao et al., 1990b). Furthermore, these studies show relatively constant, although variable, tumor rates for most of the sites in B6C3F1 mice (Rao et al., 1990a). Early NTP studies, therefore, might provide a better database to test association of anticarcinogenic effects and changes in body weight or survival than later NTP studies.

This study uses a database of all NTP studies finished by 1983 (CBDS database). The animals in this database had significantly smaller changes in weight and tumor rates over time as compared to that observed in the recent studies. The study concludes that the observed anticarcinogenic effects in the database can not be fully explained simply by weight or survival depression in the dosed animals. This conclusion is supported by comparison of the weight and survival decrements in treated animals for anticarcinogens and non-anticarcinogens and comparison of body weight decreases induced by chemicals that reduce tumor rates of one, two, and three or more types

The analyses were conducted following the methods discussed in Gray et al. (1995) and presented in the complimentary paper (Linkov, et al., 1998). A few key points are reiterated below.

The carcinogenesis bioassays performed under the NCI/NTP program are the primary source of data for this analysis. The Carcinogenesis Bioassay Database System (CBDS) is used and includes only early experiments (finished by 1983) at which point the CBDS format was abandoned by NTP. Only those CBDS experiments that: 1) have a control group, 2) last more that 70 weeks, and 3) have food, water or gavage as exposure routes are used. Our modified CBDS database contained a total of 312 chemicals that met the above criteria and have experiments on all four sex/species combinations for mice and rats (all mouse and rat strains were used).

The same tumor categorization system used in Byrd et al. (1991), Gray et al. (1995) and Linkov et al. (1998) was implemented in this work. There are a total of 102 categories (sites) which include both tumor types and tumor locations (Linkov et al., 1998 - Table 1). Among these 102 categories, 66 were classified as malignant primary neoplasms and 36 as benign primary neoplasms. The authors will be happy to share the tumor classification system with interested scientists.

The statistical significance for each tumor type in each experiment was calculated by a one-sided Fisher's Exact test and Cochran-Armitage mortality adjusted dose trend test for dosed and control groups of animals. The numbers of tumors in each group were those at the end of the experiment. The numbers of animals at risk were taken to be the initial numbers of animals in a group ("unadjusted" analysis). Survival adjustment was not done to avoid a bias in our analysis of survival depression. Although there are many suggestions for adjusting individual tumor rates to account for survival in a specific experiment (e.g., Dinse and Lagakos, 1982; Gart et al., 1986), we believe that an unadjusted analysis is appropriate for comparison of the distribution of chemically-induced survival depression in a large number of studies. The conclusions of this paper could not be altered by using an adjusted analysis.

A positive anticarcinogenic response was defined if the treated animals show a statistically significant decrease of tumor rates compared to the control group in at least one site/ strain/sex/species. Otherwise the response was classified as negative.

Throughout the analysis a chemical was counted as positive for anticarcinogenicity if it caused a tumor decrease in at least one site of at least one sex of mice or rats in any of the 102 tumor classifications. Otherwise, the chemical is classified as a non-anticarcinogen. This division is exhaustive, i.e. every one of the 312 compounds is either an anticarcinogen or a non-anticarcinogen. For each sex/species combination, the 312 chemicals can be therefore divided into 4 non-overlapping groups: anticarcinogens in both sex/species, non-anticarcinogens in both sex/species and two groups of anticarcinogens only in one sex/species. The classifications of anticarcinogenicity and carcinogenicity are not mutually exclusive; that is a chemical can simultaneously be a carcinogen at one site and an anticarcinogen at another site (Linkov et al., 1998).

The weight and survival depression due to chemical administration was established for each tested group of animals according to the following procedure. The mean body weight of animals at a specific time point after the beginning of chemical administration was calculated for each group. The change in body weight in a dosed group compared to the control group was then recorded as a percentage change. If there were several dose groups and/or control groups, the value corresponding to the average depression was recorded. Weight change for a specific chemical was calculated by averaging recorded weight changes over all dosed groups for male and female mice and rats. Similarly, the average probability of survival was calculated using number of animals alive in each dose group at the specified time after the beginning of dosing.

The weight and survival depression was usually calculated at one year following the beginning of the experiment. According to Turturro et al., (1996) and Haseman and Johnson (1996) this point in time is late enough for the test chemicals to have an impact on body weight but early enough so that most animals were still alive and tumor free. The body weight at 12 months was also found to correlate well with animal survival at 2 years (Turturro et al. 1995, Laroque et al. 1996). Survival depression was also calculated at 70 weeks.

Although the small difference in weight and survival depression between anticarcinogens and non-anticarcinogens was obvious, Kolmogorov-Smirnov tests were performed to give more statistical precision (Rosner, 1995). The Wilcoxon signed rank test was used to compare the mean weight and survival depressions in these two groups of chemicals. All statistical analyses were conducted with StatView software (Abacus Concepts, Berkeley, CA).

It is well known that the rodents recently used in long-term bioassays at the NTP are larger and heavier than at earlier times in the program (Rao et al., 1990a, Rao et al., 1990b, Keenan et al., 1996, Haseman et al., 1994, Turturro et al., 1996). Figure 1 shows the average weight of mice in the modified CBDS database as a function of the year of the end of the experiment. The dashed line is the regression line for the more recent NTP experiments (103 control groups from Turturro et al., 1996). The average weight for the control male (MM) and female (FM) B6C3F1 mice increased slightly (about 0.5%/yr) from 1973 to 1983, although there is a considerable variation among studies. A more rapid increase (about 3%/yr) in body weight of mice is indicated by analysis of a group of chemicals which included some of the later years of the CBDS database plus some more recent bioassays (Turturro et al., 1996, Haseman, 1994). The weight increase for the rats shown in Fig. 2, is more pronounced than that for mice. The Pearson correlation coefficient, R, which express the scatter of points about the regression line, is 0.24 for FM, 0.4 for MM, 0.73 for FR and MR. While the combined data of the CBDS database and the more recent data could obviously be fit by a simple straight line (with a slope less than that of Turturro et al., 1996) the data are also consistent with the observation of others that the animal body weight increases more rapidly in later years.

Increases in background tumor rates of test animals over time have been linked to increases in body weight (Seilkop, 1995, Turturro et al., 1995). To illustrate this relationship, Fig. 3 and Fig. 4 plot background tumor rates for the most common tumors in mice and rats (liver and leukemia, respectively) against the average body weight of control animals. Data from the modified CBDS database are presented as points with fitted lines and the later data (from Turturro et al., 1996) only as a fitted line. There is obviously a weaker association between liver tumor rate and body weight for the early than for the late experiments. Clearly, there is considerable variability in the background tumor rate between experiments with similar body weights. The correlation coefficient, R, is 0.33 for FM, 0.57 for MM, 0.6 for FR and 0.56 for MR. All of these are much lower than that reported by Turturro et al. (1996) for the later NTP studies.

Smaller changes in animal weight and background tumor rates in the modified CBDS database, compared to later studies, highlight its utility for this study.

Haseman and Johnson (1996) observed that experiments which demonstrate decreased tumor rates also often demonstrate decreased body weights in treated animals. They suggested that the observed anticarcinogenic effect might be a consequence of, or an artifact of, weight change. This weight change might have been caused by a reduction in appetite with a consequent reduction in food (and chemical) intake (Masoro, 1996). This suggestion, in its strongest form, would lead to the supposition that all chemicals in the database which demonstrate weight decreases in treated animals would display anticarcinogenicity. This issue was addressed by plotting the cumulative distribution function (CDF) of percent weight decrease for chemicals in the CBDS database which caused at least one significant (p≤0.01) tumor incidence decrease and those which show no decreases when tested in mice or rats (Figure 5). The first important feature that we note is that the CDFs are similar showing that weight depression is common in the CBDS database. Both the group of anticarcinogens and that of non-anticarcinogens include many chemicals that depress weight and many chemicals that do not depress weight. Nonetheless, the CDF for anticarcinogens is clearly to the right of that for non-anticarcinogens showing that the average weight depression among anticarcinogens is slightly higher in accordance with the suggestion of Haseman and Johnson (1996). For the female rats there is a statistically significant (p=0.005) difference in the distributions between anticarcinogens and non-anticarcinogens by the Kolmogorov-Smirnov test. The more powerful Wilcoxon signed rank test indicates a statistically significant difference in the mean weight depression for all four groups. The fractional difference in mean weight depression between anticarcinogens and non-anticarcinogens was 3.2% for female mice (p=0.017), 1.5% for male mice (p=0.045), 5.1% for female rats (p=0.0001), and 2.8% for male rats (p=0.008). More than 15% of mouse and rat anticarcinogens do not depress weight. About 50% of mouse anticarcinogens depress weight by more than 10%, while 40% of mouse non-anticarcinogens depress weight by more than 10%.

In an earlier paper (Gray et al., 1995) it was suggested that chemicals that show tumor increases at multiple sites in one species are more likely to show tumor increases in other species, and therefore may be of greater concern for humans. Similarly, it has been suggested (Haseman and Johnson, 1996, J.K Haseman, personal communication) that there might be a relationship between weight depression and the number of sites at which tumor decreases are detected. To investigate this we constructed weight depression CDFs for chemicals causing decreases in one, two or three or more sites in a specific sex and species (Figure 6). Although there is a small difference in weight depression between single site and multiple site anticarcinogens, it is non-significant by the Kolmogorov-Smirnov test at p≤0.05.

Chemicals could also lead to apparent, but spurious, anticarcinogenic effects if treated animals die of toxicity before the development of tumors. To examine this, survival decreases at one year were analyzed in the same manner as weight depression. Figure 7 shows CDFs for the percentage of animals dead before 1 year after the start of the experiment for chemicals which caused at least one significant tumor type decrease at the end of the experiment and those which show no decreases when tested in mice and rats. There were differences in survival decreases at one year for anticarcinogens and non-anticarcinogens, but they are small and not statistically significant by the Kolmogorov-Smirnov test. The lowest p-value is 0.08 for female rats. There was no significant difference in the means of the distributions by the Wilcoxon signed rank test. The CDFs for anticarcinogens and non-anticarcinogens were also found to be similar when the survival was evaluated at 70 weeks rather than at one year (data not shown).

The similarity of the distributions for anticarcinogens and non-anticarcinogens suggest that survival depression may not be important in distinguishing non-anticarcinogens and anticarcinogens.

Observations of decreased tumor rates associated with decreased body mass in rodents on calorically restricted diets (e.g., Hart et al., 1995a; Hart et al., 1995b) has prompted the suggestion that anticarcinogenic effects in long-term carcinogenesis bioassays might be due to the chemically-induced decreases in body weight (Haseman and Johnson, 1996). These observations were based upon a limited studies of some chemicals in the later part of CBDS database and in more recent NTP studies. In the present analysis of the full CBDS database there appears to be only a small difference in the pattern of weight decrements induced by chemicals causing decreases in tumor rates and those that show no decreases (but may show increases). Weight depression in the bioassay is so common that it likely will be associated with almost any response. It is quite interesting that we find no significant association between the "strength" of an anticarcinogenic effect (using number of sites in the same animals demonstrating anticarcinogenic effects as a crude measure of strength) with the degree of weight depression induced by a chemical.

A possible explanation for the difference between anticarcinogens and non-anticarcinogens in weight depression is that reduced food intake reduces both body weight and tumor rates. This cannot be rigorously proved with the present dataset where animals are fed ad libitum and individual food intake is not measured. But the difference in weight depression is small, only 5% or less of the average of the two groups and is smaller than the variation between chemicals in a group. The weight depression for non-anticarcinogens and the variation within groups cannot be explained in any simple way by reduced food intake. The existence of some anticarcinogens with no weight depression can be explained simply by true anticarcinogenicity. On the other hand if one insists that reduced caloric intake causes the tumor reduction, one must assume that for some unexplained reason, the reduced food intake did not lead to reduced body weight in these cases. This seems a torturous extension of the model. Again, the structure of the bioassay where weight and food intake was not measured for each animal precludes rigorous proof.

Others have found individual compounds or groups of compounds that decrease the spontaneous incidence of specific tumor types in a way that cannot be explained by decreases in body weight (or survival) (Dunnick et al, 1996; Elwell et al., 1996).

It is possible that chemicals with significant weight reductions but no statistically significant decrease in tumor rates in treated animals compared to controls have either (1) tumor rates reduced but not enough to be statistically significant; or (2) chemically-induced increases in tumors offset by weight related decreases at the same tumor type. The latter possibility is impossible to evaluate with current experimental data.

The question of the similarity, or difference, in pathologic responses from weight depression induced by chemical treatment as compared to that from caloric restriction deserves further investigation. Caloric restriction clearly reduces a number of pathologies (e.g., Hart et al., 1995a; Hart et al., 1995b) but it is not clear if weight depression is a cause of the reductions or just another sign of caloric restriction. Ideally, the influence on tumor rates of chemically-induced or caloric restriction-induced weight depression could be addressed by using data from bioassays where dietary control (Turturro et al., 1996) is implemented rather than ad libitum feeding. Few of these data now exist. A model is required to proceed further with currently available data. We have begun attempts to model body weight-induced changes in tumor rates to establish how large this effect would have to be to explain our findings of anticarcinogenicity in the CBDS database.

The small difference in survival depression observed between those chemicals which cause anticarcinogenic effects and those that do not indicate that anticarcinogenicity is unlikely to be due to the idea that treated animals do not survive long enough to develop tumors. An interesting complication is that rodents on calorically restricted diets, and therefore with reduced body weight, actually live considerably longer than controls (e.g., Hart et al., 1995a; Hart et al., 1995b) so in whole life studies a body weight induced increase of survival may offset chemically-induced survival depression.

To illustrate the type of results observed in the NTP rodent bioassays, information on anticarcinogenicity, carcinogenicity, weight decrements and survival decrements was compiled for a group of anticarcinogens. Chemicals found by the NTP to have either no or equivocal evidence of carcinogenicity were selected (Table 1). A purely statistical definition of carcinogenicity (Gray, et al., 1995) would classify four of these compounds as carcinogens in one species but not in both species. This small set illustrates the difficulties in finding a patterns of association between anticarcinogenicity and weight or survival depression. Included is a compound which shows significant weight reduction coupled with significant survival increase (rotenone) while another, butylated hydroxytoluene, demonstrates reduced body weight while survival is unaltered.

The simplest explanation for all of these observations is true anticarcinogenicity for some chemicals.

We are also exploring a positive approach to the interpretation of tumor decreases in rodent bioassays. We have been working with Ms. Mei Liu and Dr. Herbert Rosenkranz (University of Pittsburgh) on structure-activity relationship analyses of chemicals demonstrating anticarcinogenic effects (Liu et al., 1998). If a structural basis to anticarcinogenicity can be identified, then large numbers of chemicals could be screened for possible chemopreventive activity.

Observations such as those in this paper should be an important part of the ongoing discussion about the purpose and utility of the bioassay. Clearly, interpretation of rodent chemical carcinogenesis bioassays is difficult. This and the companion paper (Linkov, et al., 1998) highlight the difficulty of even defining a carcinogen. If compounds are biologically active, causing increases at some sites and decreases at others, how are the dangerous and beneficial aspects to be balanced? Lack of a clear description of its purpose invites criticism of the rodent bioassay. On the other hand, it is possible that many, or all, of the results, both carcinogenic and anticarcinogenic, are high-dose phenomena with little relevance to much lower exposures. In this case, the physiologic perturbations induced by chemical treatment are so great as to render interpretation almost impossible. A great deal of further thought about the interpretation of the rodent bioassay is in order. It is our hope that significant attention will be given to both carcinogenic and anticarcinogenic responses in the evaluation of long-term rodent chemical bioassays in the future.

This work was supported in part by EPA Cooperative Agreement No. 822917-01-0. Dr. Edmund Crouch wrote the original computer programs and gave us invaluable help in using them. Authors benefited greatly from discussions with Drs. Richard Day, John S. Evans, Frank Pompei, Lorenz Rhomberg, Herbert Rosenkranz, Nancy Sussman, Alexander Shlyakhter and Rita Schoeny. We thank Ilya Shlyakhter, Kirat Singh and Azat Shagiakhmetov for excellent technical assistance.

Byrd, D., Crouch, E.A.C. and Wilson, R. (1991) Organ Specificity of Rodent Tumor Induction: A Test of Statistical and Graphical Methods by Examination of Tumor Induction from Ingestion of Selected Substances, in: Risk Analysis, Zervos, C., ed. Plenum Press, New York

Davies, T.S., and Monro, A. (1994). The rodent carcinogenicity bioassay produces a similar frequency of tumor increases and decreases: Implications for risk assessment. Reg. Toxicol. Pharmacol. 20, 281-301.

Dinse, G.E., and Lagakos, S.W. (1982). Nonparametric estimation of lifetime and decease onset distributions from incomplete observations. Biometrics 38, 921-932.

Dunnick, J.K., Elwell, M.R., and Haseman, J.K. (1996) Decreased incidence of spontaneous mammary gland neoplasms in female F344 rats treated with amphetamine, methylphenidate, or codeine. Cancer Letters 102, 77-83.

Elwell, M.R., Dunnick, J.K., Hailey, J.R., and Haseman, J.K. (1996) Chemicals associated with decreases of mononuclear cell leukemia in the Fischer rat. Toxicol. Path 24, 238-245.

Gart, J.J., Krewski, D., Lee, P.N., Tarone, R.E., Wahrendorf, J. (1986). Statistical Methods in Cancer Research. International Agency for Research on Cancer, Lyon.

Gray, G.M., Li, P., Shlyakhter, I., and Wilson, R. (1995). An empirical examination of factors influencing prediction of carcinogenic hazard across species. Reg. Toxicol. Pharmacol. 22, 283-291.

Hart, R.W., Keenan, K., Turturro, A., Abdo, K.M., Leakey, J., and Lyn-Cook, B. (1995a). Caloric restriction and toxicity. Fundam. Appl. Toxicol. 25, 184-195.

Hart, R.W., Turturro, A., Leakey, J., and Allaben, W.T. (1995b) Diet and test animals. Science 270, 1419-1421.

Haseman, J.K., Bourbina, J., and Eustis, S.L. (1994). Effects of individual housing and other experimental design factors on tumor incidence in B6C3F1 mice. Fundam. Appl. Toxicol. 23, 44-52.

Haseman, J.K., and Johnson, F.M. (1996). Analysis of national toxicology program rodent bioassay data for anticarcinogenic effects. Mutation Res. 350, 131-141.

Keenan, K.P., Laroque, P., Ballam, G.C., Soper, K.A., Dixit, R., Mattson, B.A., Adams, S.P., and Coleman, J.B. (1996). The effect of diet, ad libitum overfeeding, and moderate dietary restriction on the rodent bioassay: the uncontrolled variable in safety assessment. Toxicol. Pathol. 24, 757-768.

Laroque, P., Keenan, K., Soper, K., Dorian, C., Hubert, M-F., and Duprat, P. (1996). Effect of initial body weight and moderate dietary restriction on survival in Sprague-Dawley (SD) rats. Fund. Appl. Toxicol 30, 206.

Linkov, I., Wilson, R., and Gray, G.M. (1998). Anticarcinogenic responses in rodent cancer bioassays are not explained by random effects. Fund. Appl. Toxicol.

Rao, G.N., Haseman, J.K., Grumbein, S., Crawford, D.D., and Eustis, S.L. (1990a). Growth, body weight, survival and tumor trends in F344/N rats during an eleven-year period. Toxicol. Pathol. 18, 61-70.

Rao, G.N., Haseman, J.K., Grumbein, S., Crawford, D.D., and Eustis, S.L. (1990b). Growth, body weight, survival and tumor trends in (C57BI6xC3H/HeN) F1 (B6C3F1) mice during a nine year period. Toxicol. Pathol. 18, 71-77.

Rosner, B. (1995). Fundamentals of Biostatistics. Duxbury Press, Wadsworth Publishing Company, Belmont MA.

Seilkop, S.K. (1995). The effects of body weight on tumor incidence and carcinogenicity testing in B6C3F1 mice and F344 rats. Fundam. Appl. Toxicol. 24, 247-259.

Turturro, A., Duffy, P., Hart, R.W (1995). The effects of caloric modulation on toxicity studies. In: Dietary Restriction: Implications for the design and interpretation of Toxicity and Carcinogenicity Studies, R.W. Hart, D.A. Neuman, and R.T. Robertson (eds). ILSI Press, Washington, DC.

Turturro, A., Duffy, P., Hart, R.W., and Allaben, W.T. (1996). Rationale for the use of dietary control in toxicity studies - B6C3F1 mouse. Toxicol. Pathol. 24, 769-775.

Weinberg, A.M., and Storer, J.B. (1985). Ambiguous carcinogens and their regulation. Risk Anal. 5, 151-156.

Figure 1 - Average weight at one year for control groups of B6C3F1 Mice in the 312 chemicals from the modified CBDS database used in this study. The average weight in the untreated or vehicle control group was determined by averaging three closest observations within one year after the beginning of experiment. Solid line is least-squares fit to all points. Dashed line is the regression line for later studies reported by Turturro et al., 1996

Figure 2 - Average weight at one year for control groups of F344 Rats in the 312 chemicals from the modified CBDS database used in this study. The average weight in untreated or vehicle control group was determined by averaging three closest observations within one year after the beginning of experiment. Solid line is least-squares fit to all points.

Figure 3 - Rates of liver tumors (sites 7+64 (Linkov et al., 1998)) for control groups of B6C3F1 mice with different average weights in the database of 312 chemicals. The average weight in untreated or vehicle control group was determined by averaging three closest observations within one year after the beginning of experiment. Line is least-squares fit to all points. Dashed line is the regression line for later studies reported by Turturro et al., 1996

Figure 4 - Rates of leukemia (site 100 (Linkov et al., 1998)) for control groups of F344 rats with different average weights in the database of 312 chemicals. The average weight in untreated or vehicle control group was determined by averaging three closest observations within one year after the beginning of experiment. Line is least-squares fit to all points

Figure 5 – Smoothed cumulative distribution functions of weight decreases at one year after the start of the bioassay for chemicals with and without anticarcinogenic effects at p≤0.01 level of statistical significance.

Figure 6 – Smoothed cumulative distribution functions of weight decreases at one year after the start of the bioassay for chemicals found to have anticarcinogenic effects at p≤0.01 level in one site (light dotted line), 2 sites (heavy dotted line) and 3 and more sites (solid line) within a specific sex-species group in the database of 312 chemicals.

Figure 7 – Smoothed cumulative distribution functions of survival depression at one year after the start of the bioassay for chemicals with and without anticarcinogenic effects at p≤0.01 level of statistical significance