Abstract
Dr. Cohen’s commentary provides an excellent review of human risk evaluation for rodent liver carcinogens, based on the mode of action (MOA) framework. This approach identifies key steps in the pathogenesis of neoplasia and then determines whether these key steps are likely to occur in humans exposed to the test article. While this provides an excellent basis for assessment of human risk for rodent carcinogens, it does not provide a compelling argument for application in the opposite direction, that is, that all relevant modes of action for all rodent carcinogens have been identified or could be identified in shorter term (e.g., thirteen-week) studies.
The two-year rodent bioassay (carcinogenicity study) has recently been criticized as being imprecise and too costly. The costs involve time and animal use as well as monetary costs for laboratory activities and professional evaluation. The bioassay is also criticized as being overly sensitive, lacking specificity, and not necessarily defining human risk. One has to agree that the two-year rodent bioassay is not a perfect model, if one defines a perfect model as having 100% sensitivity and 100% specificity for the target outcome. By that standard, there are no perfect models. Even well-controlled tests in humans are not perfect for outcomes in all humans because of variability in genetic makeup and environment that may not be evident as being relevant a priori. This is more likely to occur when dealing with new molecular targets in pharmaceutical development. The rodent bioassay is a hazard identification tool and does not necessarily provide all information needed for risk assessment.
Dr. Cohen reviews the evolution of the MOA framework as applied to rodent carcinogenicity studies and applies this to rodent liver carcinogens. The MOA framework has been applied previously in determining the relative human risk for rodent liver tumors (Holsapple et al. 2006). The application of the MOA framework depends on the identification of key events necessary for development of neoplasms in the rodent model. Note that this is not necessarily the mechanism of action, the actual molecular events associated with primary or secondary alterations in DNA, but key events in the progression of tumor development. Dr. Cohen points out that carcinogenicity is due to two basic processes, direct genotoxicity and nongenotoxic (epigenetic) effects, which result in increased cell replications that lead to an increased chance of spontaneous genetic defects. Potential relevant epigenetic effects also include changes in cell cycle control, genetic repair mechanisms, or signals for cellular differentiation that allow cells with genetic alterations to survive. These effects may not be associated directly with increased cell replications.
A basic premise of Dr. Cohen’s argument is that all key events associated with development of liver neoplasms in rodents can be detected in shorter term (thirteen-week) studies (Allen et al. 2004). This is likely correct, at least for all key events identified to date. Since the liver is a common site of carcinogenicity in the rodent, carcinogenic effects in the liver have been extensively studied, and the key events associated with most rodent carcinogens have been identified. It should be noted that the reference Dr. Cohen cites (Allen et al. 2004) involves chemicals in tests in the National Toxicology Program. Most effects of these chemicals are likely associated with chemical toxicity or nonspecific epigenetic events (e.g., enzyme induction) and do not represent novel pharmacologic target mediated events. Dr. Cohen also points out that detection of these changes has high sensitivity but low specificity, which would then be considered to be hazard identification rather than risk assessment. His approach then applies the MOA process to assess potential human cancer risk of these identified key events and may involve additional mechanistic studies.
Dr. Cohen’s proposal is in line with the current direction of proposals for improvements in toxicology testing recently proposed by the National Research Council (2007; reviewed by Krewski et al. 2009). This vision of testing is to eventually move to toxicology testing in vitro that tests critical pathways of human toxicity. The envisioned outcome of this exercise would predict lower expense, an increased relevance to man, and an eventual elimination of animal testing. Dr. Cohen’s proposal is in line with the first level of advancement (Option II in the National Research Council scheme) that still involves animal testing/biology but reduces use of animals and requires less time. This vision is admirable and should be pursued, especially from the scientific perspective of increased specificity. This vision, however, faces multiple challenges (Krewski et al. 2009). The primary challenges involve characterization and understanding of toxicity pathways in humans and means to identify them. This will require considerable effort over a period of a decade or more. As toxicity pathways in humans are identified, distinctions will have to be made between critical pathways and alterations that are without significant health consequences.
Dr. Cohen’s argument for replacement of the two-year bioassay with shorter term studies can be outlined as follows: Development of liver neoplasms in rodents is always preceded by lesions detectable in shorter term studies (e.g., thirteen weeks; hazard identification for rodent carcinogens). This premise appears to be valid and supported with data with minimal weak points. As pointed out earlier, the liver is a common site of rodent carcinogenicity and as such has been extensively studied. Since the premise is based on an evaluation of known rodent liver carcinogens, the possibility of a newly discovered rodent liver carcinogen acting through a previously undescribed MOA is possible but has a low probability. Note that Dr. Cohen points out that the test is hazard identification and has low specificity. Many compounds identified as potential rodent liver carcinogens on the basis of the ninety-day data will not cause liver neoplasia in rodents, let alone be relevant to humans. This approach does allow for less cost and animal use than the two-year bioassay but provides even less specificity for hazard identification. In addition, some identified modes of action involved in liver carcinogenesis may be transient (e.g., increased mitosis may be detectable for only a short period of time during treatment), and extensive testing may be required to detect all possible relevant changes. These lesions can be put into context for human carcinogenic risk but may require additional mechanistic studies (risk assessment for human carcinogenicity). This premise is valid and supported by data. However, the additional mechanistic studies that may be required to define MOA may be similar to those required to explain carcinogenicity responses in two-year studies. Identification of precursor lesions and definition of human carcinogenic risk can be applied to other organs/tissues (hazard identification and risk assessment for all other tissues). This premise has serious deficiencies that prevent application to deleting the two-year bioassay. The deficiencies with the premise can be outlined as follows: All potential precursor lesions to neoplasia in rodents have not been identified and characterized, especially in tissues other than liver. Even if all precursor lesions were characterized, identification of all potential precursor lesions in all tissues in a rodent thirteen-week study is not practical and may not be possible. Identification of all potential precursor lesions and additional work to put them into context for potential human relevance could result in more cost, time, and animal use than conducting a two-year bioassay.
Considerations for Genotoxic Agents
As pointed out by Dr. Cohen and others, most genotoxic agents can be assumed to possess some carcinogenic risk to humans. Conduct of a two-year bioassay may provide some additional information, basically whether the genotoxic event can lead to carcinogenicity in an in vivo system. Risk assessment would still involve studies focusing on metabolism in rodents and humans to characterize formation or elimination of genotoxic metabolites. For agents with evidence of metabolism to strong genotoxic moieties in humans, assumption of human risk may be great enough that further testing to define risk is not warranted. However, even when considering genotoxic agents, there is an assumption that all possible relevant metabolic processes have been characterized and can be identified in in vitro and/or short-term in vivo tests. Dr. Cohen points out that improved assays for detecting DNA binding of compounds are available, and structure-activity relationship evaluation is increasingly effective in predicting DNA reactivity. However, he also mentions that tamoxifen was associated with metabolic processes in rats that led to the production of DNA reactive moieties that were not predicted prior to conduct of the rodent bioassay (Randerath et al. 1994; Tryndyak et al. 2006).
Considerations for Nongenotoxic Carcinogens: Recent Examples
The primary issue (i.e., that all potential precursor lesions to neoplasia in rodents have been and/or could be identified and characterized) can be illustrated with newly discovered rodent carcinogens with nongenotoxic modes of action that had not been previously characterized as having carcinogenic potential. Examples of nongenotoxic rodent carcinogenic responses that were not anticipated continue to occur. Many of these have been subsequently characterized and determined to have little human carcinogenicity relevance. However, the primary issue in the current argument is whether all potential modes of action can be recognized a priori (before the conduct of the rodent bioassay) and not subsequent to human risk assessment.
Beta-2 agonists were shown many years ago to be associated with the development of leiomyomas in rodents. These leiomyomas were associated with the female reproductive tract, and the hazard was not recognized before conduct of the rodent carcinogenicity studies. Subsequent mechanistic work identified the precursor lesions and key events and defined the lack of relevance to humans (Gibson et al. 1987; Gopinath and Gibson, 1987; Jack, Poynter, and Spurling 1983). Later evaluations indicated that precursor lesions were present and would be detectable in a thirteen-week rodent study, but this recognition was in hindsight.
Peroxisome proliferator-activated receptor (PPAR) γ agonists and α/γ dual agonists have been associated more recently with a number of types of neoplasms in rodent bioassays. One of the PPAR agonists associated neoplasm types in rats is sarcomas, described morphologically as fibrosarcoma and liposarcoma (El Hage 2005; Hardisty et al. 2007). The histogenesis of the neoplasms is incompletely defined (Hardisty et al. 2007), but the overall incidence of sarcomas of various histologic types associated with a PPAR α/γ dual agonist suggested a common histogenic origin (Long et al. 2009).
Rats treated with PPAR γ agonists and α/γ dual agonists have increased deposition of adipose tissue, including grossly overt increases in the amount of subcutaneous fat, and morphologic changes in the adipose tissue (de Souza et al. 2001; Hellmold et al. 2007; Tannehill-Gregg et al. 2007). Morphologic changes in adipose tissue include proliferation of mesenchymal cells and fibrosis in adipose tissue (Long et al. 2009; Hellmold et al. 2007; Herman et al. 2002; Tannehill-Greg et al. 2007). These proliferative changes may be associated with later development of sarcomas in rats, but association is not consistent by compound and/or doses. These changes may provide evidence of a precursor lesion for PPAR agonist–associated sarcomas in rats, but this was not recognized before the neoplasms were identified. Potential early recognition of the carcinogenic effects of PPAR γ agonists was even more problematic because many early in vitro and in vivo studies indicated that PPAR γ agonists had antiproliferative effects: increased cell differentiation and apoptosis and reduced angiogenesis (Grommes et al. 2004; Na and Surh 2003; Panigrahy et al. 2003; Peraza et al. 2006). In addition, not all PPAR agonists are associated with development of sarcomas in rats, and the actual relationship of the extent of the precursor lesion in rodents and eventual tumor development has not been defined.
Thyroid C-cell tumors in rodents have been associated with the glucagon-like peptide 1 (GLP-1) agonist liraglutide, recently submitted for approval. In reviewing the public data made available for liraglutide submission (Food and Drug Administration [FDA] 2009; Novo Nordisk 2009), it is apparent that while GLP-1 receptors are expressed on many cell types, only thyroid C cells appear to respond to GLP-1 agonists with the development of focal hyperplasia and neoplasia in rodents. Although it was considered during the review that rodents may be relatively sensitive to the development of C-cell tumors, the conclusion of the FDA and the Advisory Panel was that the MOA was not sufficiently defined to determine if C-cell proliferation is specific to the rodent and what the relative risk to humans is. Of importance to the topic of this commentary, the data presented by the sponsor indicated that rats did not develop clearly defined precursor lesions in short-term studies, nor did they develop diffuse C-cell hyperplasia when treated with liragultide, and they developed an increased incidence of focal hyperplasia and adenomas only after longer treatment (i.e., more than seven months). Mice reportedly developed focal C-cell hyperplastic lesions after four or nine weeks of treatment. Increased serum calcitonin levels were suggested as a key step in the MAA for liraglutide. This may be problematic since functional C-cell effects such as calcitonin release are dependent on multiple factors (i.e., serum calcium, various secretagogues) and are therefore difficult to monitor and interpret in routine studies. Whether or not the MOA for liraglutide is characterized with additional research, the fact remains that the sponsor reported no precursor lesions indicative of tumor development in rats before the identification of the tumorigenic response in the bioassay.
In summary, Dr. Cohen presents additional justification for the use of the MOA framework for establishing human risk related to rodent carcinogenicity findings. However, he does not present a compelling argument to reverse and expand the use of the MOA framework to replace the two-year rodent carcinogenicity study. Dr. Cohen focused on carcinogenic responses in the rodent liver that are associated with well-characterized MOAs. The role of safety assessment is to test for effects in any tissue, by any MOA. Examples of recently identified rodent carcinogens indicate that potential MOAs for rodent carcinogenicity are still being identified and characterized. Even if one assumed that all MOAs for rodent carcinogenicity are now characterized, it would be extremely difficult, if not impossible, to adequately test for the absence of all possible relevant changes in short-term rodent studies in a manner that would involve less overall effort and expense than a two-year rodent bioassay.