National Toxicology Program, National Institute of Environmental Health Sciences, National Institutes of Health, Research Triangle Park, North Carolina 27709
Received September 7, 2004; accepted September 22, 2004
![]() |
ABSTRACT |
---|
![]() ![]() ![]() |
---|
Key Words: carcinogen identification; approach; mechanism-based biological observations.
We thank the editors and appreciate the opportunity to contribute to this timely series of articles on approaches for generating data appropriate for assessing the carcinogenic risk of chemicals. As the last scheduled submission in this series, we acknowledge the thoughtful opinions expressed by the authors of earlier submissions, and urge readers to review those articles, as we have in preparing these remarks, for more complete development of some issues that are only briefly addressed herein.
Since 1981 the US National Toxicology Program (NTP) has directed a rodent carcinogenicity bioassay program begun by the National Cancer Institute (NCI). Together the NCI/NTP have issued approximately 530 draft or final technical reports on cancer studies of over 500 unique substances, mixtures, or physical agents. Most of these have followed a traditional design involving exposure of male and female rats and mice to three concentrations of the agent under study, plus controls, for two years (Chhabra et al., 1990). Although the basic study designs are similar and all aspects of these GLP studies are carefully controlled, each study is individually designed and tailored to provide information on attendant toxicity and in some cases, mechanistic information that can be used to better understand key events in the biological response(s) leading to cancer.
The rodent bioassay was originally envisioned as a cancer screen useful for identifying agents that would be examined in human epidemiology studies, assuming that relatively few compounds would induce tumors in animals. It has evolved to occupy a primary role in the identification of agents considered "possible" or "probable" human carcinogens by the International Agency for Research on Cancer (IARC), and those "reasonably anticipated to be human carcinogens" by the NTP Report on Carcinogens. When combined with evidence from human epidemiology studies and/or in vitro studies using human tissues, the cancer bioassay provides critical supporting evidence for agents designated as "known" human carcinogens by these agencies. Evaluations for both the IARC Monographs and the NTP Report on Carcinogens use bioassay findings in a weight-of-evidence context, and it is becoming less common for bioassay results alone to drive a listing decision by IARC or NTP. See for example the listings and background documents for steroidal estrogens and nickel compounds in the 10th NTP Report on Carcinogens and those supporting pending listings for diazoaminobenzene, lead and lead compounds and neutrons in the 11th edition (http://ntp-server.niehs.nih.gov/NewHomeRoc/AboutRoC.html). The bioassay also provides information on dose response that is used in quantitative risk assessments by regulatory agencies. Jacobs and Jacobson-Kram (2004) provide a nice example of how traditional rodent bioassay results are used in the FDA drug approval process.
The sensitivity of the bioassay for identifying agents that are thought to pose a carcinogenic risk to humans is essentially 100% (Huff, 1994), but the specificity is often criticized, and many tumor responses in animals have been questioned for their relevance for predicting human cancer hazard (Cohen, 2004
; MacDonald, 2004
). It is recognized that the strain of rat or mouse used in a given bioassay can affect the tumor outcome (Takahashi et al., 2002
), and the perception that strain specific responses inordinately influence bioassay results was a driving force behind the examination of transgenic or knockout mouse models as cancer screens (Tennant, 1997
). The models used as cancer screens harbor a disabled tumor suppressor gene or an activated oncogene in signaling pathways involved in many human cancers. The genetic predisposition results in a shortened latency period for the expression of cancer for agents presumably acting through those pathways. Several models have been evaluated over the last decade as general cancer screens and as replacements for the conventional mouse study in the two-year rodent bioassay (MacDonald, 2004
). Proposals for the use of genetically modified mouse models in cancer hazard identification have been published (Schwetz and Gaylor, 1997
; (International Conference on Harmonization S1B) in MacDonald et al., 2004
).
The NTP performed extensive evaluations of several of the more promising genetically modified mouse models (Pritchard et al., 2003, see also MacDonald, 2004
). Through workshops and peer review exercises the NTP has tried to identify an appropriate niche for the results of these assays that is accepted by the scientific/regulatory community in predicting human cancer risks.
The NTP experience in this endeavor has been instructive and has identified serious problems that hinder acceptance of any new screening method. Perhaps predictably, all of the evaluated models (Tg.AC, P53 +/, Hras2) failed to identify certain agents that are considered to be potential or in some cases, known human carcinogens. Hindsight would suggest that this might be expected given the construct of the models. Although about equal in their ability to identify carcinogens, the p 53 +/ and the Hras2 models enjoy greater acceptance than the Tg.AC model in large part because of the perception that there is a greater relevance of the presumed biological events involved in the tumor response. This persists despite accumulating evidence that the p 53 +/ model currently in use (C57Bl/6 background) is unresponsive to nongenotoxic agents, as expected, but also to many genotoxic drugs as well (MacDonald et al., 2004).
The carcinogen testing scheme put forth by Schwetz and Gaylor (1997) was based on a decision tree approach to filter agents through a tiered testing design that utilized genetically modified mice as aids to decide whether standard two-year cancer studies were needed. To be effective in reducing the costs, time, and numbers of animals associated with identifying carcinogens, it is essential that the scientific/regulatory communities come to agreement on the value of both positive and negative results in the genetically modified mouse or other alternative models. The evidence to date would argue that agents that are positive in these models should be considered carcinogens, and negative studies should prompt further evaluations. However, our experience has been that positive findings have been rejected (e.g., studies of genotoxic multifunctional acrylates in the Tg.AC) and negative findings accepted (e.g., studies of nongenotoxic nonnutritive sweeteners in the Tg.AC and P 53+/) as valid measures of carcinogenic hazard.
This situation illustrates the subjective nature of the current approaches to carcinogen identification. As has been pointed out previously, a large fraction of chemicals tested in the rodent bioassay result in a positive response for one or more tumor types in one or both species (MacDonald, 2004), although estimates of the proportion of carcinogens among the "world" of chemicals are lower (Fung et al., 1995
). A positive tumor response in the two-year bioassay is a strong stimulus to toxicologists and risk assessors to attempt to fully understand the reasons for the response. Consideration is given to many aspects of the response, such as the history of response of the tumor type in that organ to other substances; the relationship between any toxicity exhibited in the organ to the tumor response, or evidence of altered oncogene or tumor suppressor gene expression. Similarly, measures of cell proliferation, genotoxicity, or evidence of altered hormonal stimulation to the target organ are examined. Toxicokinetic models are developed to more fully explore the response in relation to the dose to the tissue. Ultimately, the relevance of the rodent tumors to human cancer hazard is assessed by applying an informal or formal "mode of action" analysis similar to that outlined by Meek et al. (2004)
. Despite the expenditure of time, effort, animals, and capital resources, we are still left with uncertainties. These uncertainties can hamper regulatory actions, preventing the regulation of a true hazard or regulating an agent needlessly. As public health professionals, we have a responsibility to utilize the best available science in making a decision and to prevent situations such as that with formaldehyde, where despite recognition as an animal carcinogen in 1983, human exposures have continued to the point where the epidemiological evidence for formaldehyde as a human carcinogen is now sufficient (Cogliano et al., 2004
).
All of the preceding articles in this series have emphasized the growing role of mechanistic understanding in improving the interpretation of the currently used cancer models, or in devising new schemes for carcinogen identification. Cohen (2004) specifically mentions the potential for advances in genomics to clarify modes of action, but he rightly points out potential difficulties in the interpretation of data from these data intensive methods.
Given this situation we suggest that it may be time to consider a new approach. The National Toxicology Program has recently established a Vision for the Future: "To support the evolution of toxicology from a predominantly observational science at the level of disease-specific models to a predominantly predictive science focused upon a broad inclusion of target-specific, mechanism-based, biological observations."
Currently and historically, NTP programs have been oriented towards a particular disease endpoint, be it cancer, impaired fertility, dysmorphogenesis, hypersensitivity, or some other affliction. Each substance is studied for its potential to produce disease in a model organism using a test protocol developed for that purpose. As science progresses and interests expand, new disease endpoints, e.g., developmental neurotoxicity, endocrine disruption, etc., require new testing paradigms, and often, additional studies of relatively well studied substances.
Scientific and societal pressures on this resource intensive, disease-model approach to toxicology are growing and are pulling the field in different directions. The EPA through its High Production Volume (HPV) chemical initiative, and the European Union through it Registration, Evaluation, and Authorisation of Chemicals (REACH) program are creating large and in part publicly available databases of chemical toxicity information that are populated with data on chemicals that were not studied based on a suspicion of hazard, but based on high production volumes. The NTP is making all of the data from its studies available in an interactive, searchable, electronic mode. This will allow these databases to be probed in ways not before possible, potentially revealing new associations between toxicity endpoints and thus new ways of interpreting these tests. In an inverse sense, the new technologies of toxicogenomics and proteomics offer a similar opportunity to provide data unbiased by the model selected to study the disease, through the collection of vast amounts of information on gene and protein expression representing the global response of biological systems exposed to chemicals. For the first time it may be possible to apply computational, or systems biology analysis approaches to toxicology databases of sufficient magnitude to discern meaningful patterns or profiles of toxic responses. Efficiency in the collection, dissemination, and use of this information will become more important in light of our constant obligation to use animal (and other) resources judiciously.
The NTP Vision calls for the Program to provide data and leadership to allow evolution of the field of toxicology from a science based on the observation of disease to a science focused on the prediction of disease through collection of mechanism-based, biological observations. This vision is based in part on the belief that many environmentally influenced disease processes have an underlying similarity in their basic causal mechanisms (such as mitochondrial dysfunction, altered signal transduction, receptor activation, DNA repair inhibition, etc.), and that cellular and organ system responses to maintain homeostasis in the face of chemical stressors are discrete and limited in number.
Predictive toxicology efforts are not new, and indeed a major NTP predictive effort in the 1980 s involved winnowing a large number of genetic toxicity assays down to only a few that were most predictive of cancer in rodents (Tennant et al., 1987). Two subsequent predictive exercises followed in which a prospective challenge was issued to the scientific community to use whatever tools were available to predict the outcome of ongoing NTP rodent cancer studies (Bristol et al., 1996
). These challenges illustrated how difficult it is to create a body of knowledge sufficient to allow prediction of an outcome as complicated as a disease. In general, predictive systems that relied solely on decision rules derived from toxicology and cancer data sets existing at that time performed more poorly than those that relied on decision rules modified by human judgment. This suggests that the predictive models were not yet of sufficient sophistication to capture the wide variety of information needed to succeed. It is conceivable that the extent of information obtainable through genomic and proteomic analyses, high throughput mechanistic toxicology screens, and better toxicokinetic and metabolism information may ultimately strengthen these predictive knowledge bases and allow them to be used to predict rodent carcinogens, dysmorphogens, etc. But predicting rodent disease outcomes is not our primary goal. As outlined above, a serious limitation of all experimental animal-based procedures for predicting human responses is the influence of species- and strain-specific factors that affect either the qualitative or quantitative nature of the disease outcome. Moving from a disease-based science to a mechanism-based science could conceivably bypass these limitations.
It is assumed that the most promising predictive toxicology screens and databases will focus on general mechanisms of disease etiology common to both laboratory animals and humans. It is likely that these tools might be most effectively developed with an eye toward ones that can assess a toxicological response in humans or in human tissues as well as in rodents or rodent tissues. It is probably unrealistic to expect that predictive systems, in a broad sense, will ever be developed to the point that they entirely replace in vivo toxicology and cancer studies. There will always be exposure situations as well as biological responses that cannot be replicated in vitro.
What would it take to move toxicology from a disease-based science to a mechanism-based science? Clearly there would have to be broad scientific support for a set of biological observations that are considered to represent adverse, and perhaps irreversible events. There will also have to be a regulatory framework that allows the use of mechanistic information alone for regulation. Recent uses of mechanistic information in listing agents in the NTP Report on Carcinogens have withstood legal challenge and provide support to the notion that this is feasible. Acceptance of mechanism-based observations in place of traditional disease endpoints will require that toxicologists and risk assessors reorient their thinking to place a higher value on early biochemical or gene or protein expression changes than on traditional disease endpoints. It is likely that some regulations will need to be rewritten if we are ever able to take full advantage of scientific advancements in predictive toxicology.
How might this work? In keeping with the thrust of this Forum series, consider key biological events in cancer. Although we as scientists may weigh the significance of positive signals for mutagenesis, inhibition of DNA repair, altered DNA methylation, and alterations of proliferation/apoptosis (for example) differently, we would probably agree that exposures to agents that cause these responses in human or animal cell systems would best be avoided. Those signals or sets of signals most closely associated with carcinogenesis could be determined by comparison with our accumulated bioassay results.
What would be lost by such a move? Clearly estimating and extrapolating quantitative risks for specific human diseases would prove more difficult for regulatory agencies. A testing framework would need to be designed to integrate mechanism-based biological observations with toxicokinetic and other absorption, distribution, metabolism, and elimination (ADME) data from short-term rodent studies, to establish usable dose metrics for adverse biological outcomes.
Another major challenge to the establishment of a public health paradigm based on predictive, mechanism-based, biological observations is the requirement that such a system weigh positive findings of potential adverse events more heavily than the lack of such events. This is an inherent weakness in any predictive or reductionist system that relies on the presumed scientific understanding of a mechanism of adverse action.
Acceptable performance characteristics of predictive systems is also an area on which agreement must be reached within the scientific community. All predictive systems will have inherent error rates, as do the current in vivo toxicology models used for predicting human responses. An important assumption in the NTP Vision is that predictive systems designed to detect mechanism-based adverse biological outcomes will have inherently more power to predict similar human responses than do current in vivo or rodent disease-based models to predict human diseases. This has important implications with regard to the approach one would take toward the design of any process that would evaluate and validate mechanism-based predictive systems for their public health value.
In conclusion, at some point toxicologists will have to decide when our collective understanding of adverse biological responses in short-term in vivo, or in in vitro assays, with all their recognized limitations, has advanced to the point that data from these assays would support decisions that are as protective of the public health as are current approaches relying on the results of the two-year rodent bioassay. The NTP Vision proposes enhanced efforts to develop comprehensive databases of mechanism-based biological observations to enable this decision to be reached. More information on the NTP Vision for the 21st Century including a Roadmap for its implementation and timelines for critical steps can be found at http://ntp-server.niehs.nih.gov/main_pages/NTPVisionPg.html.
![]() |
ACKNOWLEDGMENTS |
---|
![]() |
NOTES |
---|
1 To whom correspondence should be addressed at N.I.E.H.S. PO Box 12233, Research Triangle Park, NC 27709. Fax: (919) 541-4255. E-mail: bucher{at}niehs.nih.gov.
![]() |
REFERENCES |
---|
![]() ![]() ![]() |
---|
Chhabra, R. S., Huff, J. E., Schwetz, B. S., and Selkirk, J. (1990). An overview of prechronic and chronic toxicity/carcinogenicity experimental study designs and criteria used by the National Toxicology Program. Environ. Health Perspect. 86, 313321.[ISI][Medline]
Cohen, S. M. (2004). Human carcinogenic risk evaluation: An alternative approach to the two-year rodent bioassay. Toxicol. Sci. 80, 225229.
Cogliano, V., Grosse, Y., Baan, R., Straif, K., Secretan, B., Ghissassi, F. E., and WHO International Agency for Research on Cancer (2004). Advice on formaldehyde and glycol ethers. Lancet Oncol. 5, 528.[CrossRef][ISI][Medline]
Fung, V. A., Barrett, J. C., and Huff, J. (1995). The carcinogenesis bioassay in perspective: Application in identifying human carcinogens. Environ. Health Perspect. 103, 680683.[ISI][Medline]
Huff, J. E. (1994). Chemicals causally associated with cancers in humans and in laboratory animals: A perfect concordance. In Carcinogenesis (M. P. Waalkes and J. M. Ward, Eds.), pp. 2537. Raven Press, New York.
Jacobs, A., and Jacobson-Kram, D. (2004). Human carcinogenic risk evaluation, Part III: Assessing cancer hazard and risk in human drug development. Toxicol. Sci. 81, 260262.
MacDonald, J. S. (2004). Human carcinogenic risk evaluation, Part IV: Assessment of human risk of cancer from chemical exposure using a global weight-of-evidence approach. Toxicol. Sci. 82, 38.
MacDonald, J., French, J. E., Gerson, R. J., Goodman, J., Inoue, T., Jacobs, A., Kasper, P., Keller, D., Lavin, A., Long, G., McCullough, B., Sistare, F. D., Storer, R., and van der Laan, J. W. (2004). The utility of genetically modified mouse assays for identifying human carcinogens: A basic understanding and path forward. Toxicol. Sci. 77, 188194.
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. (2004). A framework for human relevance analysis of information on carcinogenic modes of action. Crit. Rev. Toxicol. 33(6), 591653.[ISI]
Pritchard, J. B., French, J. E., Davis, B. J., and Haseman, J. K. (2003). The role of transgenic mouse models in carcinogen identification. Environ. Health Perspect. 111, 444454.[ISI][Medline]
Schwetz, B., and Gaylor, D. (1997). New directions for predicting carcinogenesis. Mol. Carc. 20, 275279.[CrossRef][ISI]
Takahashi, M., Dinse, G. E., Foley, J. F., Hardisty, J. F., and Maronpot, R. R. (2002). Comparative prevalence, multiplicity, and progression of spontaneous and vinyl carbamate-induced liver lesions in five strains of male mice. Toxicol. Pathol. 30, 599605.[CrossRef][ISI][Medline]
Tennant, R. (1997). Identifying carcinogensa strategy for use of abbreviated models. Toxicol. Pathol. 25, 240241.[ISI][Medline]
Tennant, R. W., Margolin, B. H., Shelby, M. D., Zeiger, E., Haseman, J. K., Spalding, J., Caspary, W., Resnick, M., Stasiewicz, S., and Anderson, B. (1987). Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicity assays. Science 236, 933941.[ISI][Medline]
|