Microbial Status and Genetic Evaluation of Mice and Rats: Proceedings of the 1999 US/Japan Conference (2000)

Joseph J. DeGeorge

Center for Drug Evaluation and Research

Food and Drug Administration

Rockville, MD

The US Food and Drug Administration (FDA) comprises a very diverse group in Foods and Biologics and Veterinary Medicine and Drugs. We all have very different views on how animals should be used, in what kinds of studies they should be used, and how we use information from studies in assessing risk. My remarks are from the perspective of the Center for Drug Evaluation and Research and not from the perspective of any of the other centers within FDA.

Genetic marker information is not required in study protocols. It is not submitted as part of data analyses, even for explanations of deviations in study results. As an example, when I surveyed many of the reviewers within the Center for Drug Evaluation and Research about how often they received genetic profiles of the test animals, none had ever seen such information.

Industry simply does not provide it.

The information we typically receive is required information: strain identification, the source of the animals, and sometimes site-specific control data sets, which we may request in an effort to understand the significance of some finding because we do not have the control data of all sponsors or contractors. We do have complete control data, including line listings, from the animals on the study.

For example, if we have a particular question about vasculitis either in your breeding colony or from your source, we may request the historical response rate so that we can calculate whether there is a significant effect of the pharmaceutical in relation to variable background rates. Another very recent example involves

our carcinogenicity assessment committee, which evaluates all carcinogenicity studies that come into the Center for study conduct and results review, including a report received last week from an international company that had an unlicensed product. For their dose range finding study, the available general toxicology studies had been done in Japan, and the carcinogenicity study had been done in the United States. For their animals, they had gone to the same global supplier, which had two different colonies of animals—one in Japan and one in the United States. In the Japanese studies that were submitted, there was a phenotypic response in the animals to the drug that was unmistakable. Females lost about 100 g/kg or up to 20% of their body weight. Males were not affected. Thus, the effect that occurred in the dose range finding studies at 3 months (and later, in a separate study, at 6 months) were persistent and obviously drug related. However, in the carcinogenicity study, with the same dose levels as in the other studies and with animals from the US source, there was no effect on body weight at any time during the study.

The question is whether this difference is one of genetics, the source of animals, or the source of feed. It could be caused by many things. We are still wondering why this happened, and the issue may require 2 or 3 years and $1 to $2 million to resolve.

Another area from the regulatory view (although we are not really regulating in this area) is that of microarray technology. In the pharmaceutical arena of this technology, we are beginning to look for better ways to interpret study results—possibly to understand mechanisms for responses and to eliminate the conduct of some studies. Using this technology will be a learning experience.

Currently, I participate in an International Life Sciences Institute group, which collects and attempts to compare across market-ready platforms in an effort to characterize platform responses. The purpose of this effort is to gain a similar experience base and even, in fact, build a standard response library to enable an understanding of what kind of toxic insult might reveal a predictable pattern. It will be years before this database is built.

One of my questions is what animals will be used when specialists validate or characterize this microarray platform response data set. During our last meeting, we discussed whether we should use SD or Wistar rats when trying to compare the response with a single chemical at multiple test sites, to compare platforms. That comparison can be carried farther, and there may be data to address this issue. Differences in the source of animals should be considered, even if you get what can be called a particular strain.

Finally, it is important to determine a course of action for studies in a different strain or source of animal that result in a different signal with an unknown chemical. We need to know with certainty that a different response is due to a

genotypic difference in the animals. We must learn to understand the genetics and what expression changes are associated with what strains or whether to control for this difference in test systems.

We obtain genotypic and phenotypic expression information—response data for response elements, specifically in the p53 knockout and Tg.AC models. For example, we test to learn whether the transgene has been inserted in the Tg.AC model. In this process, we have discovered that even with a genetic marker, it is possible to be fooled. In the case of the Tg.AC model, there was a long delay in the response characterization because the genotype marker being assessed was not responsible for the phenotype. Because people were not tracking the genotype that was associated with the phenotype of interest, confusion resulted and the model was almost lost. The animals expressing the transgene suddenly were no longer responding to carcinogen treatment. That example clearly reflects a need to have very good characterization of the genotype and phenotype and identification of a genotype that is responsible for the phenotype of interest.

We also now have the p53 knockout model expressed on three different background strains, and we have no information as to whether those accessory genes influence the expression rate in the different p53 animals and whether they all respond with a particular signal. Within FDA, we are looking at the Hras2 mouse model to learn whether it has the same potential liability as the Tg.AC, because it was made with a technique similar to the Tg.AC model. To the best of our knowledge today, it does not appear to carry this potential problem.

In a regulatory setting, there is very minimal information on the genetics of animals. However, that information can be very important from a regulatory perspective. We understand the importance of genetically engineered animals, but we do not necessarily appreciate that this importance can also apply to our standard toxicology models. Perhaps in the examples cited above, if we had the genetic information from the Japanese source of animals in the dose ranging study and the source for that carcinogenicity study, resulting linkage information may have enabled us to resurrect the validity of that study. Of course, it is necessary to know what genetic markers are important to follow.

I believe there are emerging issues regarding genotype and interpretations of results, particularly in the area of microarray technology that we are all rushing into headlong without resolution. We must be careful about what kind of data we collect. We must consider the meaning of an observed effect that no longer appears as a signature signal —consider the source of animals and any other factors that could have influenced that animal's response.

DR. FESTING: What is the implication of a case in which genetic markers showed that both the Japanese rats and the US study rats were, in fact, genetically different?

DR. DE GEORGE: The implied result is basically the same as the current situation, which is that the study must be repeated. According to that 2-year assay, the doses were not selected properly for that group of animals, which means that we did not learn about the carcinogenic potential of the drug. If the rodent to human dose margin had been huge or the dose had been close to the appropriate dose, we probably would have been able to accept those data. However, in this case, the doses were clearly too divergent. The result is $1 million wasted and 2 years lost, the latter of which is probably more important.