Abstract
Much attention and angst have been focused on the decline in new drug approvals, the increased failure rate of drug candidates in late stage development, and the soaring costs of developing new drugs. The causes for this crisis have been much discussed and debated. Oft cited are the following: consolidation in the pharmaceutical industry; more complicated multigenic diseases such as cancer, diabetes, and obesity; competition with generics; and perhaps a lower tolerance for adverse side effects. It is abundantly clear however that advances in applied science have not kept pace with the frenetic speed of developments in basic sciences (http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.pdf). 1 This discrepancy is especially true for the discipline of regulatory toxicology. In spite of the tremendous advances in our ability to assess xenobiotic-induced changes in gene expression, protein and metabolite levels, the tools that are routinely employed in preclinical safety testing have changed little in the last quarter of a century. Many of the endpoints we routinely evaluate are subjective and only semiquantitative such as clinical signs, functional observations, gross pathology, and histopathology. This article is designed to identify specific advances in applied regulatory toxicology that could help facilitate the movement of new drugs from discovery to the clinic.
It is instructive to consider which advances in toxicology would provide the greatest benefit in advancing drug development. While nonclinical studies still comprise a relatively small percentage of overall drug development costs, the rate of increase for this aspect of drug development is rising dramatically along with other development costs (http://csdd.tufts.edu/InfoServices/OutlookPDFs/Outlook2008.pdf). 11 This is especially true for biologics, which frequently rely on costly nonhuman primate studies. For small molecules, the costliest studies, in terms of time and resources are chronic studies which have in-lives of 6 to 24 months. Clearly, a major contribution to the field of regulatory toxicology would be the ability to predict chronic outcomes from relatively short-term studies. Arguably, this would be the most important contribution of the new “omics” endpoints.
One of the costliest, and certainly the most time-consuming requirement in nonclinical safety testing is the 2-species, 2-year chronic bioassay for carcinogenicity. In addition to the time and resource requirements, cancer bioassays are often criticized as being overly sensitive (subject to false positives), especially for drugs that induce tumors at only 1 site, in 1 sex or in 1 species. Perhaps the most frustrating aspect of this assay is its lack of mechanistic information. Often when drugs are found to induce tumors, the basis for the result is not known, and therefore little useful information is provided for designing the next generation drug that does not induce tumors. Shorter drug development timelines in the pharmaceutical industry have resulted in the need to start these studies earlier in the development cycle. Since many drugs will fail in phase 1 and phase 2 clinical trials, early initiation of cancer bioassays can result in wasted resources and needless use of animals.
Predicting carcinogenicity from short-term assays is a challenge. It was both a surprise and disappointment when in the late 1980s it was learned that while most mutagens are carcinogens, the opposite is often not true, especially for pharmaceuticals. 10 Pharmaceutical companies screen drug candidates early in development and generally discard genotoxic molecules except for life-threatening indications such as oncology. Nevertheless, approximately one third of the drugs in the Physicians' Desk Reference (PDR) for which carcinogenicity data are available are either positive or equivocal. 9 Most of these drugs are either negative for genotoxicity or positive only in tests with low specificity (in vitro cytogenetics or mouse lymphoma gene mutation assay).5 The low specificity of these 2 assays suggests that these positive genetox results may be unrelated to the mechanisms of carcinogenesis. The prevalence of nongenotoxic carcinogens presents a serious challenge in creating a short-term carcinogenicity assay because the mechanisms of carcinogenesis can be highly varied. For pharmaceuticals, the most common mechanisms include exaggerated pharmacology, hormonal imbalance, or immune suppression. For example, proton pump inhibitors increase stomach pH and gastrin secretion, which can result in enterochromaffin-like (ECL) tumors, the Leydig cell adenomas seen in mice given fenasteride are most likely related to increased luteinizing hormone (LH) levels, and Elidel-associated lymphomas are likely the result of immune suppression. Is it realistic to expect that any short-term assay will be able to predict carcinogenicity from such a diverse array of mechanisms?
A study by Nie et al. 6 examined expression arrays from livers of rats treated with over 100 “paradigm compounds.” Animals were administered a single high dose (30 to 50% of published LD50) of each chemical, and the livers were removed 24 hours after dosing. A training set consisting of 24 nongenotoxic carcinogens and 28 noncarcinogens was used to identify a 6-gene signature, which identified nongenotoxic carcinogens with an accuracy of 88%. This was the case whether or not the liver was the ultimate target organ. Such a high success rate for compounds having many modes of action is encouraging. A similar study by Fielden et al. 2 examined 100 structurally and mechanistically diverse nongenotoxic hepatocarcinogens and nonhepatocarcinogens. Rats were treated for 5 days, and a novel multigene biomarker (i.e., signature) was derived to predict the likelihood of nongenotoxic chemicals to induce liver tumors in longer term studies. Independent validation of the signature on 47 test chemicals indicated an assay sensitivity and specificity of 86% and 81%, respectively.
In a collaboration between several pharmaceutical companies and the Food and Drug Administration (FDA) organized by the Critical Path Institute, a meta-analysis of microarray data from the 2 studies cited above was performed. 3 The analysis examined data from short-term rat studies on over 150 compounds. The merged data sets indicated that the accuracy of the Fielden et al. 3 and Nie et al. 6 signatures was approximately 65% and 60%, respectively. Because of the differences in study design and different microarray platforms, it is not surprising that predictivity was reduced relative to internal validation estimates reported in the individual studies. The authors concluded that “while the signatures were not suitable for regulatory decision making, they were deemed worthwhile in the early assessment of drugs to aid in decision making in drug development.”2
Another approach for predicting carcinogenicity from shorter term studies relies on the fact that many of the mutations in oncogenes and tumor suppressor genes associated with animal and human tumors are known. For example, Parsons et al., 8 using “Allele-specific Competitive Blocker-PCR,” have investigated specific mutations in the mouse H-ras codon 61. The method uses polymerase chain reaction (PCR) primers designed to specifically block amplification of the wild-type codon, while selectively amplifying the mutant. These authors have demonstrated that this method can detect a base pair substitution in the presence of a 105-fold excess of wild-type DNA. 8 Using this method, increased frequencies of ras codon 61 mutations could be detected in the livers of mice treated the carcinogen 4-aminobiphenyl. Neonatal animals were injected twice with the chemical, and livers were harvested after 8 months, a time when this tissue is still morphologically normal. 7
A survey of the Center for Drug Evaluation and Research (CDER) database was used to determine if the pathology seen in shorter term animal studies could predict the outcome of 2-year chronic studies. 4 Final reports on a total of 60 rat oral carcinogenicity studies submitted to CDER between January of 2002 and December 2005 were included. This survey looked at a number of target organ sites including liver, kidney, mammary, thyroid, adrenal, urinary bladder, lymph node/spleen, and lung. The correlations were based primarily on 13-week range finding studies and some 6-month studies. The data indicated that preneoplastic lesions seen at 3 and 6 months did not always convert to neoplasms at 24 months, and reliance on shorter term studies would lead to many false positives.
A recent survey begun at Merck and soon to be extended to other Pharmaceutical Research and Manufacturers of America (PhRMA) member companies asks the question whether histologic biomarkers seen at the end of a 6- or 12-month toxicology study in rats can predict the outcome of a 2-year carcinogenicity study (F. Sistare, unpublished data). These markers include hyperplasia, hypertrophy, and eosinophilic and basophilic foci. A survey of 78 drugs from a combined set of studies obtained from FDA Freedom of Information files and Merck's own internal database for which 6- and/or 12-month and 2-year studies were available suggested that histologic endpoints had a negative predictivity of 90%. That is, if there were no preneoplastic lesions seen at the end of the 6 or 12 months, there was a 90% chance that the carcinogenicity study would be negative. Four drugs, all approved, were missed with this paradigm. An examination of the mechanisms by which these 4 drugs induced cancer suggests that the mechanisms may not be relevant to humans at clinical exposure levels. It should be noted that the positive predictivity of these biomarkers was poor; preneoplastic lesions seen in 6- or 12-month studies often do not progress to tumors. This is in agreement with conclusions reached from the CDER database using shorter term study results. It would therefore be necessary to extend studies with positive signals to determine the final outcome of the carcinogenicity study. However, it was estimated that if negative 6- or 12-month studies obviated the need for a 2-year study, approximately half of currently performed rat carcinogenicity studies could be eliminated. These preliminary data suggest that the combination of a negative 6-month transgenic mouse assay and a negative 6-month rat chronic assay may be sufficient to conclude that a drug lacks oncogenic potential. This initial examination of the publicly available data and the Merck database will be extended to include other PhRMA member companies and expand the database to more than 200 compounds. The results of this survey are expected to be available in the latter part of this year.
It is tempting to speculate that a 6-month rodent study combined with multigene signatures and/or quantification of oncogene/tumor suppressor gene mutation may be sufficient to predict the ultimate outcome of 2 carcinogenicity studies. Not only would such an approach save time and resources, it would reduce animal usage and provide information on mechanisms of carcinogenesis. While there is reason to be optimistic that these new approaches could supplant the need for traditional cancer bioassays, for the time being, CDER continues to follow International Conference on Harmonization (ICH) guidance for this endpoint.