Human Carcinogenic Risk Evaluation, Part IV: Assessment of Human Risk of Cancer from Chemical Exposure Using a Global Weight-of-Evidence Approach

James S. MacDonald*

Schering-Plough Research Institute, Kenilworth, New Jersey, 07033

Received April 13, 2004; accepted May 27, 2004

Key Words: Comet assay; quantile dispersion graphs; tail moment; Olive tail moment; extent tail moment; tail DNA; tail length; lymphocytes; DNA damage.

In the early 1960s, the National Cancer Institute developed procedures to formalize the process for the evaluation of the human risk of cancer from chemical exposure (Boorman, 1994Go; Weisburger, 1983Go). This approach was based on decades of research that demonstrated that chemicals that were known to cause cancer in humans could be shown to induce cancer in laboratory animals as well. Despite the fact that this approach by the NCI was envisioned as an initial assessment of carcinogenic activity to identify those chemicals for which further study was needed, the use of two rodent species exposed for 2 years as the primary mechanism to identify potential human hazards was in widespread use by the early 1970s (Boorman, 1994Go). In 1975, this process was developed into a recommendation that has formed the basis of regulatory guidance (Sontag, 1976Go). This guidance represented the best thinking of the time about how to accomplish the critical but difficult task of prospectively identifying chemicals that might pose a carcinogenic risk to humans. Despite significant progress in our understanding of the biology of the carcinogenic response in animals and humans in the nearly 30 years since this guideline was published, these recommendations continue to form the foundation of how we do cancer risk assessment today. We have a large and extensive database today from which to develop a more rational, science-based approach to this process. A careful consideration of the available data suggests an alternative approach to this important task.

Much of the early work done in this field that led to the use of the 2-year bioassay was done with reactive chemicals, most of which interacted with DNA (Miller, 1978Go). Ample evidence accumulated that compounds in classes like the polycyclic aromatic hydrocarbons that were known or suspected human carcinogens produced tumors in rodents. It was logical to assume that rodent tumor data could predict the hazard associated with human exposure. The concept of genetic versus epigenetic or alternative mechanisms of carcinogenesis originally proposed in the late 1980s is now well accepted. While much is understood about the mechanism of tumor induction by DNA-reactive carcinogens, assessment of human risk from exposure to nongenotoxic agents is more problematic.

Careful and comprehensive evaluation of the modes of action of chemicals that produce tumors in rodents has led to the understanding that positive findings in long-term rodent studies do not necessarily predict human hazard. We now understand, for example, why urinary bladder tumors in rats exposed to high doses of saccharin or melamine do not imply a hazard for humans (Cohen et al., 1991Go; Rodent Bladder Carcinogenesis Working Group, 1995Go). Similarly, a broad and very comprehensive array of data on male rat kidney tumors induced by compounds like d-limonene that produce a characteristic male rat hydrocarbon nephropathy do not predict a carcinogenic hazard for humans following exposure to these chemicals (Swenberg and Lehman-McKeeman, 1999Go). In the pharmaceutical arena, there are numerous similar examples. Rat mammary tumors produced in response to ß-adrenergic blocking agents in the 1970s caused significant disruption to drug development programs as concern was raised over the possibility of a similar tumorigenic response in humans. It is now clear that the biology surrounding the tumor promoting activity in rodents (hyperprolactinemia) is dissimilar between the two species and does not predict human hazard (Schyve et al., 1978Go; Welsch and Nagasawa, 1977Go). A similar major concern was raised over the appearance of the unusual mesovarial leiomyoma in rats exposed to ß-agonists like salbutamol in this same time frame. It is now well accepted that this rodent tumorigenic response is secondary to a unique pharmacodynamic effect in this tissue in the rat and is not an indicator of human risk (Jack et al., 1983Go). Other endocrine tumors in rats like thyroid tumors or gastric ECL-cell carcinoid tumors are not considered to represent a risk for humans when it can be shown that these tumors arise in this species as a result of a specific endocrinologic or pharmocodynamic response that does not occur under the conditions of human use of the respective therapeutic agents (MacDonaldet al., 1994Go; McClain, 1994Go). More recently, the body of data available surrounding peroxisome proliferators and the appearance of rodent liver tumors has been assembled into a comprehensive mode of action analysis that permits a conclusion on the relevance of this finding for human risk (Klaunig, 2003Go).

It is important to emphasize at this point that these conclusions regarding the disconnection between the appearance of an increased incidence of a particular tumor type in rodents and potential human hazard did not come easily. In most of the cases cited above, years and, in some cases, decades of detailed, careful experimentation were necessary to clearly establish the relationship between the appearance of tumors and the mode of action of this effect. Over the last several years, well-developed approaches for incorporating such mode of action data into human hazard assessment have been developed that provide a strong framework for evaluation of the data derived from rodent tumorigenicity studies (Meek et al., 2003Go; Sonich-Mullen et al., 2001Go). The issue that needs to be addressed, however, in the context of our current broader understanding is the value of the data derived from the 2-year bioassay in the first place.

Despite the extensive database on modes of action of chemicals shown to produce an increased incidence of tumors in rodents accumulated over the last 20 years and the frameworks developed to evaluate these data, there is a reluctance to move away from the 2-year bioassay as the "gold standard" for human hazard identification and the use of these data in risk assessment. An important picture of what this approach has given us in the area of pharmaceuticals came out of discussions on harmonization of international pharmaceutical regulations during the International Conference on Harmonization (ICH). The data in Table 1 show the results of queries made of several (sometimes overlapping) databases stimulated by discussions on the value of rodent bioassay data. It is clear that, irrespective of the database queried, approximately 50% of the compounds tested in 2-year rodent bioassays were positive in one or both of the species. While it could be argued that, for industrial and environmental chemicals (as in the NTP database or the Carcinogen Potency Database), most of the chemicals tested early in the program were reactive molecules that were probably genotoxic, the same is not true of the pharmaceutical databases. Few of the chemicals in these databases (FDA, CPMP, PDR, JPMA) are genotoxic as few pharmaceuticals with this profile get tested in 2-year rodent bioassays.


View this table:
[in this window]
[in a new window]
 
TABLE 1 Incidence of Positive Findings in Rodent Carcinogenicity Studies

 
While it must be acknowledged that human population studies are relatively insensitive for detecting human carcinogenic activity, this high percentage of positive findings in rodents has to be contrasted to the relatively small number of pharmaceuticals known to be carcinogenic to humans (Table 2). If this list is examined, it can be seen that the known human carcinogens in the database are either genotoxic, immunosuppressive, hormonally active, or belong to a class of reactive dermatologicals (e.g., arsenicals). The large number of positive compounds in the databases in Table 1, then, suggests that either a great number of human pharmaceuticals will be proven human carcinogens over time (an unlikely possibility) or that the assay we are using over-predicts potential human risk by a substantial margin.


View this table:
[in this window]
[in a new window]
 
TABLE 2 Pharmaceutical Agents Carcinogenic to Humans

 
So how should responsible scientists expert in the field of toxicology respond to analyses such as these? We certainly all share the responsibility to assure the protection of human health irrespective of the organization or institution in which we do our work. But we also share the responsibility to utilize the best that our science has to offer to accomplish this important task in a manner that gets us to the right answer in the shortest time frame. The data available to us support moving away from reliance on the results of a single assay to identify human hazards (2-year rodent bioassay) in favor of one based on a comprehensive, weight-of-evidence approach.

The use of such an approach is illustrated by a recent publication from the NTP (Pritchard, 2003Go). In this paper, Pritchard and colleagues employ an innovative approach to assessing the utility of various strategies (aggregations of data from various assays) to accurately predict human risk. While the approach utilized by Pritchard and colleagues in this paper can be criticized because the end point for comparison of the outcome of the various strategies is human risk as identified by the 2-year rodent bioassay (based on data from the International Agency for Research on Cancer (IARC) or the National Toxicology Program Report on Carcinogens (NTP ROC)), the information in this paper can be utilized to illustrate a point for this discussion.

Pritchard and colleagues used a series of 12 different strategies to evaluate the probability of accurately predicting human cancer risk. A significant strength of this approach is that it uses human cancer risk as the outcome (with the caveat stated above). The overall accuracy (correct prediction of human risk) of this approach is illustrated in Figure 1. The NTP rodent bioassay (rat + mouse studies, Strategy 9) or the rat bioassay alone plus incorporation of genotoxicity data (Strategy 12) was predictive in only 65–69% of the cases. A significant increase in predictive capability was seen when the results of studies with some of the currently available genetically modified mouse models was substituted for the results of the rodent bioassay (e.g., strategies 1, 2, 5, 6, and 7).



View larger version (34K):
[in this window]
[in a new window]
 
FIG. 1. The data in this figure were derived from the recent publication by Pritchard et al. (2003)Go, as discussed in the text. Using various approaches (characterized as strategies), the overall accuracy for predicting human hazard was assessed by these authors. The strategies consisted of data from the studies or combinations of studies shown in the table:
Strategy number

Study(IES) incorporated into strategy

1 p53 mouse assay
2 p53 mouse assay (g)
3 TgAC mouse assay
4 rasH2
5 p53(g) + rasH2 (ng)
6 p53 (g) + rasH2 (all)
7 p53 (g) + Tg.AC (ng)
8 p53 (g) + Tg.AC (all)
9 NTP rodent bioassay (rat + mouse)
10 NTP rat bioassay + Tg.AC (ng) + p53(g)
11 NTP rat bioassay + rasH2 (ng) + p53(g)
12 NTP rat bioassay + genotoxicity

Note. The p53, rasH2, and Tg.AC models refer to genetically modified mice employed in 6-month assays of chemicals (for descriptions of models, refer to MacDonald et al., 2001Go). As these models have been shown to be differentially sensitive to genotoxic or nongenotoxic chemicals, the strategies segregated responses based on the genotoxic status of the respective chemicals (g = genotoxic; ng = nongenotoxic; all = both genotoxic and nongenotoxic chemicals).

 
A more detailed analysis of several of these strategies is shown in Table 3. When the incidence of "false" positives and "false" negatives is looked at more closely, it can be seen that the rodent bioassay is very good at preventing false negatives. Either used alone as a single strategy or in combination with genotoxicity data, this approach did not miss a single compound identified as a potential human carcinogen by IARC or the NTP ROC. This, however, is accomplished at the cost of identifying a substantial number of chemicals as positive in the strategy even though they are not considered to be human hazards.


View this table:
[in this window]
[in a new window]
 
TABLE 3 Detailed Results of 5 of 12 NTP Strategies

 
If we agree that, from a public health viewpoint, it is not acceptable to miss chemicals that pose potential human cancer hazards with our experimental paradigms, then the results of some of the alternative strategies might be of concern. For example, Strategy 6 in the current illustration (where data from a 6-month p53 hemizygous knockout mouse assay are coupled with data from a 6-month study in the rasH2 transgenic mouse) shows four such chemicals that are putative human carcinogens (based on the IARC or NTP ROC classification) but yet were "missed" by this strategy. An examination of these four chemicals and the data available with these chemicals illustrates how we might use a global weight-of-evidence approach to improve this process.

The four chemicals that were "missed" with this strategy are cyclosporine A, estradiol 17ß, phenobarbital, and chloroform. Each of these four chemicals has been extensively studied, and there are many data available to bring into the discussion of potential human risk. There is clear evidence for the human carcinogenic risk of cyclosporine A and estradiol. The former is a potent immunosuppressive agent, and while the mechanism of human carcinogenesis is not known with certainty, it is clearly related to the pharmacodynamic action of this therapeutic agent (Ryffel, 1992Go). Similarly, estradiol's potent estrogenic activity and the well-accepted trophic stimulation of endocrine responsive tissues are part of the data base available for this compound to assess potential human risk (Russo and Russo, 1998Go). In both of these cases, clinical and/or anatomical pathological changes in animals are seen within a 6-month time frame that defines the biologic response of concern. With our understanding today of the bases of neoplastic transformation and tumor progression, the availability of data such as these would have raised a concern about potential human risk even in the absence of data from the 2-year bioassay.

Similarly, it is generally well accepted that the rodent liver and kidney tumors that arise in the 2-year bioassays with chloroform are an indirect result of chronic toxicity and the responsive cellular proliferation. In the absence of this chronic toxicity, these tumors are not considered to represent a risk for humans (Golden et al., 1997Go; Hard et al., 2000Go). While a different case from the previous three chemicals, phenobarbital is also classified as a "miss" in the current example, as it is listed as a potential human carcinogen by both the IARC and NTP ROC. There is probably no therapeutic agent for which so much human data exists. Phenobarbital has been used for over 70 years to control seizures, and several databases have carefully evaluated the prevalence of human tumors in people who have been taking this drug for many years (Whysner et al., 1996Go). The preponderance of data in these evaluations show no increase in any particular tumor type despite prolonged exposure for many years to plasma levels of the drug near those causing liver tumors in mice (McLean et al., 1992Go). Again, with these two chemicals, it can be argued that observations in short-term assays would have raised signals based on our current understanding of carcinogenesis that could have been further evaluated for a better understanding of potential human risk of cancer. In this case, the preponderance of data suggests that the rodent bioassay data do not accurately predict potential human risk. For all four of these "misses" with this strategy in the current example, the added value of data from the rodent bioassay can be questioned when all the available information is viewed together.

The point of this exercise is not to suggest that data from a p53 and rasH2 alternative mouse assay are all that is necessary to identify potential human carcinogenic hazards. It is rather to illustrate that, by taking advantage of all the data available, a more rational approach to this process can be utilized that recognizes the value of data from many sources. It is suggested that the data necessary to identify potential human hazards are available without the current reliance on the 2-year bioassay in rodents to provide definitive data in this area.

A proposed alternative approach is presented in Table 4 for consideration and discussion. The central tool that would provide the bulk of the data in this process is the core battery shown in this table. Properly conducted, information from these studies will identify any of the genotoxic, proliferative, inflammatory, or hormonal effects that may constitute a signal of concern and that are associated with the known human carcinogens. The proposed 6-month study duration is based on extensive experience in the area of pharmaceuticals; the predictive value of studies of this duration is captured in the rationale behind the ICH guidance on the necessary duration of animal toxicology studies. The value of studies of this duration across a broader spectrum of chemicals can certainly be established with a careful examination of existing data. Information from this core battery would then constitute the basis for hazard identification. The absence of a signal of concern in these studies would not require any further assessment of carcinogenic potential.


View this table:
[in this window]
[in a new window]
 
TABLE 4 Proposed Prospective Protocol for Assessment of Human Carcinogenic Potential

 
Additional useful data can also be made readily available from other studies such as those suggested as ancillary assays in Table 4. While these are not proposed as "core studies," information from assays such as these may be employed in a tiered approach to further refine and clarify observations made in the core battery. It is suggested that employing a prospective strategy such as outlined here (without data from a 2-year rodent bioassay) would yield adequate information to correctly identify the relatively small number of known human carcinogens. Properly utilized, this approach should also enable us to correctly identify agents of potential concern that warrant further study. This would thus result in the same protection from false negatives seen currently with the 2-year bioassay but in a more scientifically meaningful—and timely—manner.

Implicit in this proposal is the assumption that results in any of these assays will be considered in context with other data and no assay will be definitive on its own merit. In other words, a positive in vitro genotoxic signal would not automatically supersede data from the other assays and would, instead, be considered along with all of the available data. This would require that some sort of decision-tree be developed that would guide the use of this approach and direct the user to alternative approaches that may be necessary to better interpret a given result. Some of these might include the many molecular endpoints that are currently being developed that may one day significantly enhance this process. While agreement on the best way to construct this algorithm will not be easy to achieve, it is an important part of the ability to utilize this approach effectively and will require commitment from responsible scientists from all segments of our field.

With careful consideration from scientists from academia, regulatory authorities, and industry, an appropriately rigorous process for decision making can be envisioned that would provide a significant enhancement over the current process. Retrospective analysis of the results of the rodent bioassay to assess relevance to potential human risk is not the most effective manner to accomplish our task. A prospective, comprehensive evaluation of the biologic effects of chemicals better utilizes our current understanding of the carcinogenic process and would yield a more efficient and effective assessment of potential human risk. The proposal offered is a step in this direction.


    NOTES
 

* To whom correspondence should be addressed: Schering-Plough Research Institute, 2015 Galloping Hill Rd., Kenilworth, NJ 07033. Fax: (973) 940-4159. E-mail: james.madonald{at}spcorp.com.


    REFERENCES
 TOP
 REFERENCES
 
Boorman, G. A., Maronpot, R. R., and Eustis, S. L. (1994). Rodent carcinogenicity bioassay: Past, present, and future. Toxicol. Pathol. 22(2), 105–111.[ISI][Medline]

Cohen, S. M., Cano, M., Earl, R. A., Carson, S. D., and Garland, E. M. (1991). A proposed role for silicates and protein in the proliferative effects of saccharin on the male rat urothelium. Carcinogenesis 12, 1551–1555.[Abstract]

Golden, R. J., Holm, S. E., Robinson, D. E., Julkunen, P. H., and Reese, E. A. (1997). Chloroform mode of action: Implications for cancer risk assessment. Regul. Toxicol. Pharmacol. 26, 142–155.[CrossRef][ISI][Medline]

Hard, G. C., Boorman, G. A., and Wolf, D. C. (2000). Re-evaluation of the 2-year chloroform drinking water carcinogenicity bioassay in Osborne-Mendel rats supports chronic renal tubule injury as the mode of action underlying the renal tumor response. Toxicol. Sci. 53(2), 237–244.[Abstract/Free Full Text]

Jack, D., Poynter, D., and Spurling N. W. (1983). Beta-adrenoceptor stimulants and mesovarian leiomyomas in the rat. Toxicology 27, 315–320.[CrossRef][ISI][Medline]

Klaunig, J. E., Babich, M. A., Baetcke, K. P. Cook, J. C., Corton, J. C., David, R. M., DeLuca, J. G., Lai, D. Y., McKee, R. H., Peters, J. M., Roberts, R. A., Fenner-Crisp, P. A. (2003). PPARalpha agonist-induced rodent tumors: Modes of action and human relevance. Crit. Rev. Toxicol. 33(6), 655–780.[ISI][Medline]

MacDonald, J. S., Lankas, G. R., and Morrissey, R. (1994). Toxicokinetics and mechanistic consideration in the interpretation of the rodent bioassay. Toxicol. Pathol. 22, 124–140.[ISI][Medline]

MacDonald, J. S., Robinson, D., and Pettit, S., Eds. (2001). An evaluation of alternative methods for carcinogenicity testing. Toxicol. Pathol. 29 (Supplement).

MacDonald, J. S., and Scribner, H. E. (1999). The maximum tolerated dose and secondary mechanisms of carcinogenesis, in Carcinogenicity: Testing, Predicting, and Interpreting Chemical Effects (K. T. Kitchen, Ed.), Marcel Dekker, New York, 124–144.

Marselos, M., and Vainio, H. (1991). Carcinogenic properties of pharmaceutical agents evaluated in the IARC Monographs Programme. Carcinogenesis 12, 1751–1756.[Abstract]

McClain, R. M. (1994). Mechanistic consideration in the regulation and classification of chemical carcinogens. In Nutritional Toxicology (F. N. Kotsonis, M. Mackey, and J. Hjelle, Eds.), pp. 273–303. Raven Press, New York.

McLean, A. E. M., Driver, H. E., and Sutherland, I. A. (1992). Liver Tumour Promotion by Phenobarbital: Comparison of Rat and Human Studies. In Relevance of Animal Studies to the Evaluation of Human Cancer Risk, (R. D'Amato, T.J., Salaga, W. Farland, C. Henry, Eds.) pp. 251–259. Wiley-Liss, Inc.

Meek, M. E., Bucher, J. R., Cohen, S. M., Dellarco, V. Hill, R. N., Lehman-McKeeman, L. D., Longfellow, D. G., Pastoor, T., Seed, J., and Patton, D. E. (2003). A framework for human relevance analysis of information on carcinogenic modes of action. Crit. Rev. Toxicol. 33, 591–653.[ISI][Medline]

Miller, E. C. (1978). Some current perspectives on chemical carcinogenesis in humans and experimental animals: Presidential Address. Cancer Res. 38, 1479–1496.[ISI][Medline]

Monro, A., and MacDonald, J. S. (1998). Evaluation of the carcinogenic potential of pharmaceuticals. Drug Saf. 18, 309–319.[ISI][Medline]

Pritchard, J. B., French, J. E., Davis, B. J., and Haseman, K. (2003). The role of transgenic mouse models in carcinogenic identification. Environ. Health Perspect. 111(4), 444–454.[ISI][Medline]

Rodent Bladder Carcinogenesis Working Group. (1995). Urinary bladder carcinogenesis: Implications for risk assessment. Food Chem. Toxicol. 33, 797–802.[CrossRef][ISI][Medline]

Russo, I. H., and Russo, J. (1998). Role of hormones in mammary cancer initiation and progression. J. Mammary Gland Biol. Neoplasia 3(1), 49–61.[CrossRef][ISI][Medline]

Ryffell, B. (1992). The carcinogenicity of ciclosporin. Toxicology 73(1), 1–22.[CrossRef][ISI][Medline]

Schyve, P. M., Smithline, F., and Meltzer, H. Y. (1978). Neuroleptic-induced prolactin level elevation and breast cancer: An emerging clinical issue. Arch. Gen. Psychiatry 35, 1291–1301.[Abstract]

Sonich-Mullin, C., Fielder, R., Wiltse, J., Baetcke, K., Dempsey, J., Fenner-Crisp, P., Grant, D., Hartley, M., Knaap, A., Kroese, D., et al. (2001). International Programme on Chemical Safety. IPCS conceptual framework for evaluating a mode of action for chemical carcinogenesis. Regul. Toxicol. Pharmacol. 34, 146–152.[CrossRef][ISI][Medline]

Sontag, J. M., Page, N. P., and Saffiotti, U. (1976). Guidelines for carcinogen bioassay in small rodents. National Cancer Institute, NCI-CG-TR-1. DHEW Publ. No. (NIH) 76–801. U.S. Government Printing Office, Washington, D.C.

Swenberg, J. A., and Lehman-McKeeman, L. D. (1999). {alpha}2Urinary-globulin associated nephropathy as a mechanism of renal tubule cell carcinogenesis in male rats. In Species Differences in Thyroid, Kidney and Urinary Bladder Carcinogenesis (C. C. Capen, E. Dybing, J. M. Rice, and J. D. Wilbourn, Eds.), IARC Scientific Publications No. 147, Lyon, France, 95–118

Weisburger, E. K. (1983). History of the bioassay program of the National Cancer Institute. Prog. Exp. Tumor Res. 26, 187–201.[ISI][Medline]

Welsch, C. W., and Nagasawa, H. (1977). Prolactin and murine mammary tumorigenesis: A review. Cancer Res. 37, 951–963.[Abstract]

Whysner, J, Ross, P. M., and Williams, G. M. (1996). Phenobarbital mechanistic data and risk assessment: Enzyme induction, enhanced cell proliferation, and tumor promotion. Pharmacol. Ther. 71, 153–191.[CrossRef][ISI][Medline]