preclinical development handbook || toxicogenomics in preclinical development

44
867 24 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT Eric A. G. Blomme, Dimitri Semizarov, and Jeffrey F. Waring Global Pharmaceutical Research and Development, Abbott Laboratories, Abbott Park, Illinois Preclinical Development Handbook: Toxicology, edited by Shayne Cox Gad Copyright © 2008 John Wiley & Sons, Inc. Contents 24.1 Introduction 24.1.1 Toxicogenomics and Other Emerging Technologies in Perspective 24.1.2 Definitions and Basics of Toxicogenomics 24.1.3 History of Toxicogenomics 24.1.4 Applications of Toxicogenomics 24.2 Practical and Logistic Aspects of Toxicogenomics 24.2.1 Technical Considerations 24.2.2 Species Considerations 24.2.3 Toxicogenomics in In Vitro and In Vivo Studies 24.3 Toxicogenomic Reference Databases 24.3.1 The Need for Toxicogenomic Reference Databases 24.3.2 Design and Development of Toxicogenomic Reference Databases 24.3.3 Existing Toxicogenomic Databases 24.4 Toxicogenomics in Drug Discovery 24.4.1 Predictive Toxicology 24.4.2 Development of Predictive Gene Expression Signatures 24.4.3 Case Examples 24.4.4 Predicting Species-Specific Toxicity 24.5 In Vitro Toxicogenomics 24.5.1 Objectives of In Vitro Toxicogenomics 24.5.2 Proof-of-Concept Using Primary Rat Hepatocytes 24.5.3 Use of Gene Expression Profiling to Assess Genotoxicity 24.5.4 Current and Future Use of In Vitro Toxicogenomics 24.6 Toxicogenomics in Mechanistic Toxicology 24.6.1 Objectives of Mechanistic Toxicology 24.6.2 Case Examples of Mechanistic Toxicology

Upload: shayne-cox

Post on 02-Dec-2016

214 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

867

24 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

Eric A. G. Blomme , Dimitri Semizarov , and Jeffrey F. Waring Global Pharmaceutical Research and Development, Abbott Laboratories, Abbott Park, Illinois

Preclinical Development Handbook: Toxicology, edited by Shayne Cox GadCopyright © 2008 John Wiley & Sons, Inc.

Contents

24.1 Introduction 24.1.1 Toxicogenomics and Other Emerging Technologies in Perspective 24.1.2 Defi nitions and Basics of Toxicogenomics 24.1.3 History of Toxicogenomics 24.1.4 Applications of Toxicogenomics

24.2 Practical and Logistic Aspects of Toxicogenomics 24.2.1 Technical Considerations 24.2.2 Species Considerations 24.2.3 Toxicogenomics in In Vitro and In Vivo Studies

24.3 Toxicogenomic Reference Databases 24.3.1 The Need for Toxicogenomic Reference Databases 24.3.2 Design and Development of Toxicogenomic Reference Databases 24.3.3 Existing Toxicogenomic Databases

24.4 Toxicogenomics in Drug Discovery 24.4.1 Predictive Toxicology 24.4.2 Development of Predictive Gene Expression Signatures 24.4.3 Case Examples 24.4.4 Predicting Species - Specifi c Toxicity

24.5 In Vitro Toxicogenomics 24.5.1 Objectives of In Vitro Toxicogenomics 24.5.2 Proof - of - Concept Using Primary Rat Hepatocytes 24.5.3 Use of Gene Expression Profi ling to Assess Genotoxicity 24.5.4 Current and Future Use of In Vitro Toxicogenomics

24.6 Toxicogenomics in Mechanistic Toxicology 24.6.1 Objectives of Mechanistic Toxicology 24.6.2 Case Examples of Mechanistic Toxicology

Page 2: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

868 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

24.7 Toxicogenomics and Target - Related Toxicity 24.8 Toxicogenomics and Idiosyncratic Toxicity

24.8.1 Defi nition of Idiosyncratic Toxicity 24.8.2 Preclinical Models of Idiosyncratic Toxicity 24.8.3 Case Example: Idiosyncratic Hepatotoxicity

24.9 Toxicogenomics in Regulatory Submissions 24.9.1 Overview of the FDA Pharmacogenomics Guidance 24.9.2 Future Impact of Toxicogenomic Data in Regulatory Decision Making

24.10 Conclusion References

24.1 INTRODUCTION

24.1.1 Toxicogenomics and Other Emerging Technologies in Perspective

The cost of drug discovery and development has risen exponentially in the last decades. According to the Pharmaceutical Research and Manufacturers of America (PhRMA), the industry ’ s trade group, pharmaceutical companies spent $ 33 billion on R & D in 2003, a threefold rise since 1990 and nearly 30 - fold since 1977 [1] . This increase in investment has so far failed to deliver a surge of new medicines, and this is refl ected by a concerning low productivity of pharmaceutical R & D [1, 2] . The cost of developing a new chemical entity (NCE) ranges from $ 800 million (U.S.) to $ 1.1 billion [1, 3] . These rising R & D costs are not sustainable and the lack of productivity of pharmaceutical R & D units has to be addressed. There are several factors under-lying this change in the drug development economics. First, pharmaceutical compa-nies are now tackling diseases of greater complexity than in the past and the industry ’ s interest in the development of blockbusters requires running longer and more expensive clinical trials [1] . Second, the requirements for approval are notably higher than in the past because of the enhanced standard of care for most diseases and more stringent regulations intended to improve drug quality and safety [4] .

One striking aspect of the drug discovery and development process is the high failure rate of compounds with an estimated 99.9% of compounds eliminated from the discovery and development pipeline [1] . Obviously, the vast majority of these compounds are eliminated very early in the process because of suboptimal pharma-cological, physicochemical, pharmacokinetic, or toxicologic properties. Nevertheless, failure rates in the subsequent, more costly stages of development are substantially high with the vast majority of clinical attrition occurring in Phases IIb and III [4] . Eliminating unsuccessful drugs earlier than in full development is defi nitely a pre-requisite for a decrease in the overall R & D costs. Indeed, the recently implemented laboratory technologies (such as combinatorial chemistry, genomics tools, or high throughput screening) driving discovery efforts are resulting in a constantly increas-ing number of novel compounds being synthesized and a similarly rising number of therapeutically interesting targets. Therefore, approaches that allow for an earlier, multidirectional characterization of compounds are needed to face these major challenges.

Page 3: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

INTRODUCTION 869

The major causes of attrition in the clinic in 2000 were lack of effi cacy and safety, both accounting for approximately 30% of failures, respectively [4] . This is in con-trast to what was occurring in the late 1980s when poor pharmacokinetic properties were the main reason for termination (around 40%), while lack of clinical safety already accounted for 30% of failures [5] . This remarkable improvement in the pharmacokinetic properties of advanced compounds was mostly the result of a sig-nifi cant effort by the pharmaceutical industry to develop preclinical tools to better predict the pharmacokinetic properties of experimental compounds. While the industry has successfully addressed the failures related to pharmacokinetics, it has not signifi cantly improved its ability to better characterize early the toxicologic properties of compounds. In fact, one may argue that with an increased number of compounds to be evaluated, less toxicologic characterization has been possible, and often compounds are selected for animal testing without suffi cient data regarding their toxicologic potential. Furthermore, traditional toxicologic evaluation through in vivo studies typically creates a bottleneck in the R & D process because of its length and cost, and due to the requirement for signifi cant amounts of compound. Approaches designed to characterize the toxicologic profi le of compounds earlier would allow discovery scientists to select the molecules with an optimal, or at least adequate, toxicological profi le for these costly studies. To be cost effective and applicable in the drug discovery setting, these approaches must use small amounts of compounds (typically an amount that would not require scale - up chemistry), have acceptable accuracy (the level of acceptability being dependent on the stage of testing) and reproducibility, and have an appropriate throughput [6] . Various tech-nologies, including the “ omics ” technologies, potentially meet these criteria and are addressed in this textbook in various chapters.

24.1.2 Defi nition and Basics of Toxicogenomics

In this chapter, we use the term toxicogenomics to refer to the use of gene expres-sion analysis in the fi eld of toxicology. The sequencing of several whole genomes has led to the development of methodologies that make it feasible to monitor in several animal species (humans, mice, rats) the expression levels of large numbers of genes expressed in a specifi c tissue at a certain time, an activity referred to as gene expression profi ling or simply expression profi ling. Among these laboratory tools reviewed in Chapter 22 , molecular toxicologists have mostly used DNA micro-arrays to generate transcription profi les from tissues collected from in vivo studies or cells derived from in vitro experiments. Consequently, this chapter heavily empha-sizes the use of microarrays in toxicology studies. It is noteworthy, however, that other technologies are available or may become available in the future for gene expression profi ling. In particular, for specifi c applications, it is generally agreed that more cost - effective platforms with a higher throughput will be needed for toxicoge-nomics to realize its full potential in drug discovery and development [7] .

Toxicogenomics is based on the relatively simple assumption that toxicants acting through a similar mechanism of action will generate similar gene expression profi les or at least affect similar pathways, leading to common gene expression changes (Fig. 24.1 ). In other words, these gene expression changes (either upregulation or down-regulation) induced in common by toxicants with similar toxicologic properties could represent an easy and sensitive endpoint to identify and classify toxicants.

Page 4: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

870 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

These common gene expression changes are typically referred to as fi ngerprints or signatures of toxicity. Proof - of - principle for the signature concept has already been nicely demonstrated in various types of cancer, as mentioned in Chapter 22 . Using gene expression analysis, investigators have identifi ed new classes of hematological malignancies or predicted prognosis in lung cancer and breast cancer [8] . Further-more, because microarrays allow for the global evaluation of the cell transcriptome, the assumption is also that the identifi ed gene expression changes will allow toxi-cologists to understand better the molecular mechanisms whereby toxicants injure cells. Again, this is clearly supported by the numerous mechanistic observations made in various diseases using microarrays [8] . These observations are typically a stepping stone for the articulation of new hypotheses on the mechanism of action, which can then be evaluated and confi rmed by subsequent, appropriately designed experiments.

The use of most of the early microarray platforms was associated with reproduc-ibility and accuracy issues. Since then, the technology has rapidly improved and several recent studies have demonstrated the reproducibility and accuracy of gene expression data [9 – 11] . This rapid improvement has led to a change of practices. For instance, it was initially recommended to confi rm specifi c gene expression changes observed with microarrays with techniques that were considered more accurate, such as real - time reverse transcription - PCR (RT - PCR) [12] . In our laboratory, we do use RT - PCR to validate microarray data, but this confi rmation step is now only occasional. An example demonstrating the good correlation between a microarray platform and RT - PCR is illustrated in Fig. 24.2 . For gene expression profi ling to reliably identify and characterize toxicity, gene expression data must be suffi ciently reproducible following exposure to chemicals. The consistency and reproducibility were evaluated by different industrial, governmental, and academic laboratories

FIGURE 24.1 Principle of toxicogenomics. Toxic compounds induce changes in the cellular transcriptome, including changes in the expression of particular gene sets that correlate with the mechanism of toxicity. These gene sets are typically referred to as signatures. These sig-natures can be used for the identifi cation and characterization at the molecular level of toxic changes.

Page 5: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

INTRODUCTION 871

using various commercial and customized microarray platforms and shown to be suffi cient for assessing toxic reactions in tissues such as liver, kidney, and heart [9, 13, 14] .

Any experimental in vivo study is associated with some degree of interindividual variability. This is the reason behind the use of treatment groups of appropriate size in toxicology studies. Developing an understanding of the interindividual variability in gene expression data was critical to fully understanding the optimal number of animals per group to be used in toxicogenomic studies. Clearly, gene expression data are also subject to interindividual variability inherent to any in vivo model. In our experience, gene expression data generated following short in vivo exposures (typi-cally less than 24 hours) are typically quite variable. However, gene expression profi les from tissues exposed for longer periods of time (1 – 5 days) are usually less variable than other endpoints typically used in toxicological assessment, such as serum chemistry, hematology, or histopathology. Figure 24.3 illustrates this concept using liver as an example. This low variability has clear implications and advantages, as it allows the toxicologist to use as few as 2 – 3 rats/group to generate interpretable and reliable data.

The question of species extrapolation has been evaluated and studied by toxi-cologists for decades. In most instances, especially in the case of pharmaceuticals, the objective of the toxicologist is to identify hazards and assess their risks for humans. A particular interest with toxicogenomics is to understand if this technol-ogy can lead to an improved prediction of toxicologic reactions in humans, and in particular whether gene expression profi ling allows toxicologists to reliably deter-

FIGURE 24.2 Correlation between expression levels of cytochrome P450, 1b1 (CYP1B1) measured with Affymetrix microarrays and RT - PCR (TaqMan). The log10 values of the fold changes determined by Affymetrix microarrays ( Y axis) are plotted against those determined by RT - PCR ( X axis). Included are the gene expression levels from 32 hepatocyte treatments measured by both platforms. The values represented by squares indicate genes that exhibit the same directionality in terms of gene expression changes using both technologies. Circles indicate genes that exhibit different directionality between the two platforms.

Page 6: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

872 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

mine which of the changes occurring in traditional toxicologic preclinical species would not be relevant to humans. Several studies have shown that, in general, toxi-cogenomics improves the robustness of cross - species extrapolation [15] . Indeed, despite the differences in genomes, many responses to toxicants are evolutionary conserved. Therefore, since analysis of gene expression changes following the expo-sure to a toxicant improves the understanding of the mechanism of toxicity and of the major cellular subsystems affected, it becomes easier to assess the relevance of a specifi c toxic change or to predict how a specifi c species would react to a com-pound. However, it is important to point out that most of these studies have focused on a specifi c class of toxicants or a particular mechanism of toxicity. Furthermore, in vitro studies have mostly been used to evaluate how humans may react to a spe-cifi c compound and these studies make it diffi cult to fully comprehend a cause - and - effect relationship. Consequently, it is prudent to state that while a toxicogenomic analysis in the context of a particular toxicologic study can and will ultimately improve the overall risk assessment to humans, it is likely that exceptions will remain.

24.1.3 History of Toxicogenomics

Soon after the microarray technology was fi rst invented in 1995, toxicologists and molecular biologists rapidly realized its potential to generate large amounts of valu-

FIGURE 24.3 Heatmap with hierarchical clustering illustrating gene expression changes in the liver of rats treated for 3 days with various hepatotoxicants at toxic doses. Genes shown include genes that were up - or downregulated by at least two fold with a p < 0.01. Despite signifi cant variability in response observed with clinical pathology and histopathology, there is limited interindividual variability in gene expression profi les.

Page 7: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

INTRODUCTION 873

able molecular data, which could help in understanding the pathogenesis of various toxic changes [16, 17] . This technology has rapidly been embraced by the pharma-ceutical industry as a potentially useful tool to identify safer drugs in a faster, more cost - effective manner [18] . Almost all major pharmaceutical companies now have dedicated groups applying gene expression analysis in toxicology. Academic and governmental institutions have also aggressively adopted this technology, and the regulatory community has clearly identifi ed toxicogenomics as an important part for the success of its critical path initiative [19] . Growing interest in this fi eld is best exemplifi ed by the vast numbers of workshops, committees, and consortia created to address various technical issues, to improve the science base of the community, to allow contributors to keep pace with these new emerging technologies, and to establish guidelines for determining how genomic data should be submitted to regu-latory agencies. An exponentially growing number of studies are being published on this topic. Some studies have used microarrays to identify the mechanism of toxicity of pharmaceutical agents or standard toxicants [20 – 22] . Other studies have evaluated the predictive power of toxicogenomics by identifying potential toxic liabilities before the development of other manifestations of toxicity, such as clinical chemistry or histopathology [23 – 25] . Finally, several investigators have used this technology in an attempt to identify new markers of toxicity [26] . The majority of published studies have evaluated reference and tool compounds that induce well - characterized, general, or specifi c toxicities. These compounds either have been on the market for many years or have never been developed as pharmaceutical agents. However, in the last few years, several pharmaceutical companies have clearly moved beyond this proof - of - concept stage and have applied this technology to drug development programs to address critical, development - limiting toxicologic issues, as indicated by recent submissions of gene expression profi ling datasets to regula-tory agencies as well as by various published studies.

24.1.4 Applications of Toxicogenomics

It would clearly be beyond the scope of this chapter to provide an in - depth review of all published studies using toxicogenomics. Rather, in this chapter, we focus on potential applications of toxicogenomics using selected examples as illustrations. Regrettably, numerous very valuable published studies have not been referenced here because of space limitations, and therefore, we encourage the reader interested in a specifi c application to proceed to a more elaborate literature search.

In the drug discovery and development process, toxicogenomics has been or can be applied at different stages to address different issues (Fig. 24.4 ). Its applications include the following:

• Prediction and characterization of the toxic properties of experimental com-pounds from short - term in vivo studies or in vitro systems. This activity typically takes place in the discovery setting at the lead identifi cation, lead optimization, and candidate selection stages.

• Elucidation of the mechanism of toxic changes. These types of studies are more often conducted on candidate compounds that do induce unexpected changes of unknown mechanism in repeat - dose toxicology studies. The objectives are to develop an understanding of the relevance of these changes for humans or

Page 8: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

874 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

to generate mechanistic data for the establishment of appropriately designed counterscreens for selecting backup compounds.

• Characterization of new therapeutic targets and the proactive identifi cation of potential toxicity issues associated with their modulation (also referred to as toxicity related to primary pharmacology or on - target toxicity). The purpose is to investigate further the biology of the target as early as the target selection stage. Given the rising number of targets that pharmaceutical companies are working with, learning what makes a good or bad target at the gene expression level with respect to toxicology can lead to better target prioritization in the discovery pipeline.

• Identifi cation of selective gene expression signatures that could be used as sensitive biomarkers. There is a defi nite need for additional biomarkers that could help in the identifi cation and monitoring of specifi c toxicologic changes, in both preclinical and clinical studies. In the last few years, very few additional sensitive and specifi c correlates of particular toxic changes have been made available to the toxicology community. Whole - genome analysis represents an ideal approach for the identifi cation of genes or signaling pathways whose deregulation leads to specifi c toxic effects.

• Molecular characterization of compounds associated with idiosyncratic reac-tions in the clinic. Idiosyncratic toxic events are low - incidence toxic events of unknown mechanism that occur in humans, usually in a non - dose - dependent manner and that were not observed in preclinical species. This type of toxicity

FIGURE 24.4 Use of toxicogenomics in drug discovery and development. In drug discovery and development, toxicogenomics can be applied at different stages. An understanding of potential target - related toxicity with gene expression profi ling can help to prioritize for the most drugable targets. Toxicogenomic studies using short - term in vivo studies or in vitro assays (predictive toxicology) are useful for the early characterization of the toxic properties of compounds at the lead identifi cation/lead optimization stages. Toxicogenomics can also be used to elucidate the mechanism of toxicity associated with compounds. These mechanistic studies typically take place after candidate selection but are also useful to establish appropri-ate counterscreens for backup compound selection. Finally, gene expression profi ling studies can help identify novel biomarkers for the identifi cation and monitoring of toxicologic changes, in both preclinical and clinical studies.

Page 9: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

leads to costly failures of drugs in clinical trials or market withdrawal. Approaches that would allow for understanding the mechanism of idiosyncratic adverse events could have an enormous impact on the productivity of the pharmaceutical industry. They could rescue useful drugs through identifi cation of populations sensitive to the toxic effects or through early prediction of idio-syncratic adverse events.

24.2 PRACTICAL AND LOGISTIC ASPECTS OF TOXICOGENOMICS

24.2.1 Technical Considerations

At the time when this chapter is written, gene expression profi ling is still a rapidly evolving discipline for which efforts have so far essentially focused on addressing key technical issues and on validating approaches through proof - of - concept studies. Because of the immaturity of the microarray technology, there are many different technical issues, including external or internal control selection, probe design, plat-form variation and comparison, scanner performance characteristics, data normal-ization, or analysis software packages. These technical issues are being aggressively addressed by various academic, governmental, and industry groups, often in the form of consortia or collaborations [9, 10, 14, 27, 28] . While technical issues are criti-cal to the reliability and correct interpretation of microarray data, most of them are closely related to the instrumentation being used and are consequently rapidly evolving and becoming obsolete. For this reason, in this chapter, we only address technical issues that will have some relevance for a signifi cant period of time and that need to be understood by molecular toxicologists. For other issues, the reader is referred to Chapter 22 or other existing publications [10,12, 27 – 30] .

24.2.2 Species Considerations

The vast majority of toxicogenomic studies are conducted using rat tissues or cell lines of rat or human origin. The major reason is that the rat is the most commonly used small laboratory animal species for toxicology testing in the pharmaceutical industry. Another reason is the incomplete genome annotation for other species (dog or monkey) and the lack of historical gene expression data for these species. Nevertheless, gene expression studies have been conducted in large animal species (dog and monkey) and several investigators have experimental microarray plat-forms available for these large species [31, 32] .

24.2.3 Toxicogenomics in In Vitro and In Vivo Studies

Tissue Considerations Toxicogenomic studies have been performed using cell cultures and tissues from in vivo studies. Cell cultures are very homogeneous, and gene expression changes induced by toxicants can be reproduced reliably in a labo-ratory under similar experimental conditions. The use of tissues can be more chal-lenging. Several tissues, such as liver or heart, are relatively homogeneous in their phenotypes and transcriptomes, and robust, consistent changes in gene expression can reliably be detected, as long as a consistent tissue collection protocol is followed.

PRACTICAL AND LOGISTIC ASPECTS OF TOXICOGENOMICS 875

Page 10: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

876 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

Other tissues, such as brain or testis, are more heterogeneous and complex, and their gene expression profi les are therefore less consistent. For instance, the brain is composed of various cell types (neurons, glia) with immense phenotypic and tran-scriptional diversity [33, 34] . In addition, depending on the region of the brain, marked differences in functions and transcriptomes are present between cells of the same origin. This complexity explains the limited number of attempts to apply toxi-cogenomics to these tissues. Technologies, such as laser capture microdissection (LCM) and RNA amplifi cation protocols, have greatly enhanced the ability to perform expression analysis on single cell populations [35] . However, these tech-nologies also require a more extensive investment of time and labor with an overall reduced capacity. In addition, the analysis of gene expression changes in single cell populations may limit one ’ s ability to fully understand major interactions between cells that may play a signifi cant role in the pathogenesis of a toxicologic change. These limitations will be illustrated in our section covering testicular toxicity (Section 24.6.2 ). Finally, it should be reemphasized that an accurate interpretation of gene expression changes can only succeed if tissue collection protocols are suffi ciently consistent. Even in the case of rather homogeneous tissues like heart or kidney, it is critical to collect samples for gene expression analysis in a consistent manner [36] . Likewise, relatively homogeneous tissues like kidney contain several compartments that are clearly different in structure, function, and transcriptome. Toxicants may induce changes in only specifi c compartments or cellular subpopulation. Failure to cover all compartments of a tissue would limit one ’ s ability to detect toxicant - induced gene expression changes. In our laboratory, we have established collection protocols for all tissues that we routinely evaluate. For example, we collect kidneys in a manner such that appropriate and consistent proportions of cortex and medulla are included for RNA extraction. Inappropriate collection procedures would lead to the identifi cation of gene expression changes related to the collection procedure rather than to the toxicity being evaluated.

Samples for gene expression analysis should be collected immediately after sac-rifi ce and fl ash frozen in liquid nitrogen or preserved in an appropriate RNA stabi-lization solution. They can then be stored for a prolonged period of time at − 80 ° C without signifi cant RNA degradation. Inappropriate collection procedures or storage conditions will result in RNA degradation, as typically revealed by an overall poor RNA quality after RNA extraction procedures (Fig. 24.5 ). It is therefore recom-mended to always evaluate RNA quality after extraction to ensure that the quality is suffi cient to justify the costly step of hybridization to microarrays, but also to ensure that interpretation of the microarray data is feasible.

Hybridization Design in Toxicogenomic Studies Three common hybridization designs are used in experiments using two - color or two - channel microarrays. These designs do not pertain to experiments using one - color microarrays. These designs are referred to as direct, reference , and loop [37, 38] . In the direct design , the test article - treated samples are hybridized against their appropriate control samples. This allows for the identifi cation of differentially expressed genes at a specifi c time. In the reference design , which is the most commonly used design within the biologi-cal community, all study samples are hybridized against a common reference sample, and this can be useful in the case of a study containing one control group for several treatment groups. This design is also well suited for the characterization of the

Page 11: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

magnitude of gene expression changes and temporal relationships. The loop design is seldom used, although it has been shown to attain a higher precision [30, 38] . It entails sequential hybridization of all study samples against one another; it may offer advantages for time course experiments where one needs to understand expression changes over time.

Experimental Design in Toxicogenomic Studies The design of in vivo and in vitro toxicogenomic studies is determined by the questions to be answered or the issues to be addressed. Specifi c design will be discussed as different applications are reviewed. However, several considerations are relevant to the current discussion.

Duration of Dosing In the case of a study to address the mode of action of a spe-cifi c toxicologic change, a time - course experiment may be very appropriate to iden-tify gene expression changes linked to the development of the toxicity. Time points are selected based on the particular toxicologic change to be investigated. An impor-tant aspect to remember is that gene expression changes are transient and therefore timing is of critical importance. Changes that are more relevant to generate a mecha-nistic understanding of toxicologic effects are typically those occurring before the effects are fully developed. Therefore, it is usually more insightful to evaluate gene expression during the development of a tissue change, rather than when the change

FIGURE 24.5 Evaluation of RNA quality after extraction using an Agilent Bioanalyzer. Rat spleens were collected after sacrifi ce and fl ash frozen. For samples 1 – 3, the collection procedure was inappropriate, resulting in RNA degradation as evidenced by the multiple bands and smear observed on the Bioanalyzer - generated electrophoretic image. These samples should not be used for hybridization to microarrays. Note the sharp contrast with samples 4 and 5, where RNA quality is optimal for microarray experiments.

PRACTICAL AND LOGISTIC ASPECTS OF TOXICOGENOMICS 877

Page 12: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

878 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

has already occurred and is fully established. In the case of a study used to rank - order several potential lead molecules for candidate selection, the study needs to be designed with a consideration for compound availability, reference database, performance characteristics of the predictive signatures of interest, and so on. For instance, in our laboratory, an in vivo toxicogenomic hepatic reference database has been generated based on 3 - day repeat - dose studies in rats. Consequently, we typi-cally assess our compounds for hepatotoxicity in 3 - day repeat - dose studies. In several companies, toxicogenomic analysis is now integrated in most 2 – 4 week rat repeat - dose toxicology studies. The objective is usually to be proactive in case unex-pected toxicologic changes occur in the studies or to enhance knowledge and exper-tise in gene expression analysis. In addition, these types of practice are useful to promote a better acceptance of these new technologies in a company.

Dose In a mechanistic in vivo study, the dose will be selected based on the best chance to consistently reproduce the toxicologic change of interest. In contrast, the method used to develop the predictive signatures will dictate the dose to be used in a predictive study. Validation of predictive signatures should address to some extent their predictive power and accuracy for various dose levels. For instance, in our laboratory, our in vivo databases and signatures have been developed based on a low and a high dose. The low dose corresponds to an estimated pharmacological dose (a dose resulting in an exposure similar to that achieved in the ED 50 in the most appropriate animal model), while the high dose corresponds to a maximal tolerated dose (MTD) for a 3 - day study (defi ned by the highest dose before rats exhibit clinical signs of toxicity, lack of body weight gain, or a signifi cant decrease in food consumption). It is clear that in the vast majority of cases, because these studies are typically conducted early in a project, only limited information is avail-able for selecting the doses. Consequently, dose setting can be a challenging task and requires extensive communication among all project stakeholders. An adequate dose selection is, however, crucial for proper decision making and, in our experience, is central to a successful prediction of toxicologic changes.

Samples When conducting an in vivo study, we strongly recommend that all major tissues, including blood, be collected for concurrent histopathologic and clinical pathologic examinations. The latter are relatively inexpensive compared to the current cost of microarrays, can be performed quickly in a discovery setting, and provide critical information that helps interpret gene expression data in the context of changes in the homeostasis of the tissue being analyzed and in the context of the overall status of the animal. For instance, if clinical observations and analysis of the clinical pathology and histopathology changes suggest that an animal is moribund, gene expression changes may be more refl ective of the overall poor condition of the animal than of a specifi c mechanism of toxicity. In our laboratory, for predictive studies, we typically select or prioritize the tissues to be evaluated with microarrays based on a prior histopathologic and clinical pathologic evaluation. For instance, if a test article demonstrates dose - limiting hepatotoxicity as evidenced by hepatocel-lular necrosis and elevations of serum transaminases, it would be moot and redun-dant to evaluate the liver by gene expression analysis. However, if the project team considers it important to understand the pathogenesis of this hepatoxicity, microar-ray analysis of the liver may be warranted.

Page 13: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

24.3 TOXICOGENOMIC REFERENCE DATABASES

24.3.1 The Need for Toxicogenomic Reference Databases

Gene expression changes, when viewed in isolation, can often lead to more ques-tions than answers [39] . Furthermore, although changes in expression of individual genes may be of importance, it is usually more appropriate to examine gene expres-sion changes by looking at pathways being regulated. The analysis of pathways increases the confi dence that a change in expression of specifi c genes has biological implications, but increases the accuracy of the overall interpretation. Abundant historical data are not yet available and general experience is still insuffi cient for the meaningful interpretation of thousands of simultaneous gene expression changes, which may often appear disconnected. Only after a specifi c toxicologic change has been observed consistently can its real toxicologic signifi cance be under-stood. The same holds true for gene expression changes. A recent study nicely illustrated this aspect. In this study, over 300 liver microarray experiments covering three different classes of compounds (genotoxic carcinogens, nongenotoxic carcino-gens, and noncarcinogens) were clustered across 72 putative oncogenes [40] . The three classes of compounds were strikingly interdispersed within the cluster, indicating that upregulation of oncogene expression was not a surrogate marker for carcinogenesis, as both noncarcinogens and carcinogens were upregulating the expression of these selected genes. This study also reiterates the danger of focusing on the change in expression of single or a limited set of genes in a microarray experiment.

Gene expression changes induced by toxicants typically refl ect a large number of complex pharmacological, physiological, and biochemical processes [18, 41] . To generate a plausible mechanistic hypothesis for the pathogenesis of a toxicologic change, the gene expression changes related to the toxicity need to be identifi ed and separated from those that are adaptative, benefi cial, or unrelated to the develop-ment of the toxicologic change. This requires an appropriate study design, including the evaluation of multiple time points, but mostly the access to reference data. The use of reference compounds may clarify or confi rm which gene expression changes are related to a specifi c lesion or how the lesion develops. However, contextual information from large, established reference databases is optimal to properly inter-pret gene expression data by correlating unique gene changes to those associated with treatment with a large repository of compounds or with specifi c toxicologic mechanisms [39] . The large number of compounds, tissues, corroborative toxicologic and pathologic changes, and gene expression data in these reference databases allow one to strengthen statistical inferences [42] . The concept of databases in toxicology is not novel, and as an illustration, one can think of databases for serum chemistry, hematology, pathology, or carcinogenesis.

24.3.2 Design and Development of Toxicogenomic Reference Databases

The ideal toxicogenomic database contains gene expression profi les induced in various tissues following the treatment of the reference species (most often rats) with a variety of reference toxicants (known pharmaceutical agents, prototypical toxicants) and control compounds, at multiple doses and time points [23, 43 – 47] .

TOXICOGENOMIC REFERENCE DATABASES 879

Page 14: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

880 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

The reference compounds profi led in the database should refl ect a variety of toxic mechanisms and represent different structure – activity relationships [42] . The use of multiple doses of the reference compounds (e.g., an effi cacious dose and a maximum tolerated dose) is extremely useful to distinguish a pharmacological effect from a toxicological effect. In some situations, time - course data can be very useful to iden-tify gene expression changes linked to a time - dependent toxic response and can increase the chances of observing a true toxic response. The number of animals required for each time point or dose is also an important consideration. Biological replicates are useful to establish both the biological and technical variability. As mentioned earlier, when rats are exposed for a suffi cient period of time to toxicants, the interindividual variability of gene expression changes is relatively small, so that 3 animals per group and per time point are usually considered suffi cient for the generation of meaningful gene expression profi les.

In addition to gene expression profi les, a useful database also contains suffi cient technical and biological information. The realization by the scientifi c community that gene expression data can only be correctly understood and put in a perspective if the critical amount of associated information is available had led to the publica-tion in 2001 by the Microarray Gene Expression Data Society of the minimum information about a microarray experiment (MIAME) guidelines for the reporting and publication of microarray experiments [48] . These guidelines have become a reference for the submission of gene expression data to scientifi c journals and are used as standards by some public and commercial databases.

Experimental inconsistencies can lead to some confusion, as not all datasets relate to one another. In particular, the platform used to generate the reference profi les can limit the value of some experimental datasets. While some marked improvements have been achieved in the ability to extrapolate data from one plat-form to another and in establishing more consistent gene nomenclature, different platforms still generate slightly different datasets or may be evaluating different genes [49] . Furthermore, even when the same platform is used, gene expression data have been shown to vary across laboratories because of differences in protocols or instrumentations [37] . These issues are the focus of several major initiatives involv-ing academic, industry, and government laboratories with the objectives of identify-ing, validating, and implementing standards that could be used for gene expression analysis [9, 10, 14] . Overall, however, several confounding factors still exist that cur-rently limit the comparison of gene expression data from one laboratory to another. These experimental aspects should be carefully evaluated when selecting the data-base to work with.

The nature of the species or strain used can also be a limitation. Ideally, one should compare gene expression profi les generated in a particular species with those generated in the same species or even strain. Practically, this may not always be feasible because of a lack of reference data in some species or strains. Therefore, certain circumstances require extrapolating across species. There have been signifi -cant improvements in the annotations of the genome of the major preclinical species currently used in toxicology [15] . However, the mapping of orthologous genes should still be viewed as approximate and needs considerable additional effort to be optimized. Furthermore, not all species react similarly to a specifi c toxicant and this difference in response also limits the ability to extrapolate from one species to another. Different rat strains may also have different responses to some toxicity and

Page 15: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

the difference in response at the transcriptome level has also been addressed. Overall, the transcriptome response following exposure to toxicants has been shown to be usually very similar across different rat strains [50] .

24.3.3 Existing Toxicogenomic Databases

Several publicly available, commercial or proprietary databases exist for the analysis of gene expression datasets [28] . Although not all of these repositories are specifi c to toxicology, they can still represent a useful source of data for the analysis of gene expression. Several recently developed databases are mostly focused on toxicology. For instance, the National Institute of Environmental Health Sciences has recently established the National Center for Toxicogenomics to create a reference knowl-edge database (Chemical Effects in Biological Systems or CEBS) that would ulti-mately allow scientists to understand mechanisms of toxicity through the use of gene expression analysis as well as proteomics and metabolite profi ling [51, 52] . Compa-nies such as GeneLogic (Gaithersburg, MD) or Iconix (Mountain View, CA) have created large toxicogenomic reference databases containing gene expression pro-fi les induced by prototypical reference compounds with corroborating toxicologic and pathologic endpoints [42, 53] . A list of selected public databases with a brief description of their general attributes is presented in Table 24.1 .

24.4 TOXICOGENOMICS IN DRUG DISCOVERY

24.4.1 Predictive Toxicology

In this chapter, the term predictive toxicology is used to refer to the use of short - term assays that allow the toxicologist to predict with suffi cient accuracy toxic changes that would occur after longer exposure and to extrapolate toxic reactions from preclinical species to humans [54, 55] . Predictive toxicology assays can be in the form of in vitro assays or short - term in vivo studies. Because most attempts to use toxicogenomics in predictive toxicology have so far focused on the use of short - term in vivo studies, we limit our discussion to this approach. The use of cell cultures is covered later in this chapter.

Large - scale expression analysis is an extremely sensitive approach to detect deregulated genes and signaling pathways that contribute to toxic changes. The main assumption of toxicogenomics is that following exposure to toxicants at relevant doses, transcriptional changes occur before the development of a toxic phenotype as assessed by traditional endpoints, such as clinical observations, hispathologic examination, or clinical pathology measurements. In our experience, this assumption is accurate for the vast majority of toxicities with few exceptions. For this reason, toxicogenomics offers the unique opportunity to reliably identify compounds with toxic liabilities early in the drug discovery process, and to thus signifi cantly improve the productivity of drug discovery [18, 54, 55] . As mentioned earlier, the develop-ment of more sensitive and predictive technologies that would allow for the char-acterization of toxicology early is critical for the selection of molecules with an optimal toxicologic profi le. Pharmaceutical R & D units and private toxicogenomics companies have consequently invested signifi cant resources in the development of

TOXICOGENOMICS IN DRUG DISCOVERY 881

Page 16: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

882 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

TABLE 24.1 Some Public Databases

Database Attributes URL

Gene Expression Omnibus (GEO)

World ’ s largest public repository Adherence to MIAME guidelines Toxicology data available Exploration, analysis, and visualization

tools

http://www.ncbi.nlm.nih.gov/projects/geo/

ArrayExpress Large public repository Adherence to MIAME guidelines Toxicology data available in

Tox - MIAMExpress Expression data from normal human

and mouse tissues

http://www.ebi.ac.uk/arrayexpress/

Chemical Effects in Biological Systems (CEBS)

Evolving public toxicogenomics repository from the National Institute of Environmental Sciences (NIEHS) National Center for Toxicogenomics (NCT)

Designed to house data from complex studies having multiple data steams (genetic, proteomic, metabonomic data)

Exploration and analysis tools

http://cebs.niehs.nih.gov/

Environment, Drugs, and Gene Expression (EDGE)

Public toxicogenomics repository Standardized experimental conditions

including standardized microarray platform

Mostly focused on mouse liver microarray data

Useful bioinformatics tools (clustering, BLAST searching, rank analysis, classifi cation tools)

http://edge.oncology.wisc.edu/

Symatlas Product of the Genomics Institute of the Novartis Research Foundation

Expression data from a large panel of normal human and mouse tissues or cell culture models

http://symatlas.gnf.org/SymAtlas/

DbZach System Toxicogenomic database allowing data mining and full knowledge - based understanding of toxicologic mechanisms

Contains correlating clinical chemistry parameters and histopatholigc data

http://dbzach.fst.msu.edu/

predictive gene expression - based assays. However, relatively few of these efforts have been formerly published, such that the majority of the information used here is based on personal experience.

Predictive toxicology should ideally be applied at the lead selection and lead optimization stages, concurrently with the other assays used to assess drug - like

Page 17: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

properties of molecules (pharmacologic and physicochemical properties, ADME pharmacokinetic characterization). Predictive toxicogenomics is not yet amenable to such an early stage because of its relative immaturity, low throughput, and sig-nifi cant cost. In addition, while most available data suggest that expression analysis will allow for the use of gene expression - based assays early in the discovery process, not enough data related to the accuracy of this approach are available to fully assess the value added by this technology. Nevertheless, several companies are using gene expression analysis to rank - order or prioritize compounds using short - term rat studies based on their toxic potential in specifi c tissues, in particular, liver.

24.4.2 Development of Predictive Gene Expression Signatures

As pointed out earlier, changes in expression of single genes are typically not suffi -cient to predict or identify existing toxic changes. Rather, changes in the expression of a gene set are more likely to correlate with toxicity. Consequently, the fi rst step in predicting toxic changes entails the development of gene expression signatures that strongly correlate with toxic changes. Albeit simple in concept, this task has proved to be substantially more diffi cult and more resource intensive than initially anticipated. Not all recently developed genomics approaches have demonstrated superiority compared to traditional methods or have been able to adequately vali-date signatures based on external samples (forward validation process). Indeed, sophisticated statistical tools and biostatistical expertise are needed to address the complexity of the various changes in the transcriptome and to develop predictive toxicogenomic signatures [18, 53, 56] . Some of the most commonly used statistical methods are briefl y reviewed next.

Toxicants typically induce hundreds to thousands of gene expression changes, a large number diffi cult to manage for most statistical methods. Therefore, the fi rst step in developing predictive signatures is to reduce the number of parameters, focusing on the ones relevant to the classifi cation model. Two major approaches are traditionally used for dimensionality reduction. The most commonly used approach is to rank genes with respect to differences in expression between experimental groups using parametric or nonparametric statistical tests, such as standard or permutation t - or F - test, Wilcoxon statistics, or signifi cance analysis of microarrays (SAMs) [43, 44, 57, 58] . The genes that are differentially expressed at a specifi ed signifi cance level or a fi xed number of top ranking genes are then selected for inclusion in the prediction model. The second approach for dimension reduction is to use noise reduction methods, such as principal component analysis (PCA) [59] . The PCA method reduces a large set of genes into several components, where each new component is a weighted linear combination of all genes. These components are rank - ordered according to the amount of variance, and the fi rst component represents the greatest variability among the samples. By selecting the fi rst n components as most informative, the dimension of gene expression data is vastly reduced.

Once data have been reduced and informative genes have been selected, predic-tive models can be developed with the objective of classifying compounds as toxic or nontoxic in the relevant tissue. This step requires a training set consisting of a repository of gene expression profi les encompassing both toxic and nontoxic com-pounds, and the use of computational algorithms that will allow for the accurate

TOXICOGENOMICS IN DRUG DISCOVERY 883

Page 18: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

884 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

classifi cation of unknown samples. These computational methods require signifi cant numbers of gene expression profi les to generate useful predictive models. Several computational algorithms have been used for the analysis of microarray - generated gene expression data. They include logistic regression, linear discriminant analysis (LDA), naive Bayesian classifi ers, artifi cial neural networks (ANNs), and support vector machines (SVMs). Both logistic regression and LDA use statistical inference to weigh the contributions of each signature gene expression value in sample pre-diction and typically require tens to hundreds of datasets, depending on the vari-ability of the gene expression data. In situations where there is a clear difference between groups, these methods can be quite robust [43, 53] . Naive Bayesian classi-fi cation is a popular approach for classifi cation. In one study, gene expression pro-fi les were generated from the liver of mice treated with 12 compounds, representing 5 well - characterized classes of hepatotoxicants (peroxisome proliferators, aryl hydrocarbon receptor agonists, noncoplanar polychlorinated biphenyls, infl amma-tory agents, and hypoxia - inducing agents) [45] . Using a naive Bayesian classifi cation, a predictive signature consisting of 12 genes accurately classifi ed all samples into their chemical groups. Artifi cial neural networks (ANNs) are analogous in concept to a biological nervous system. They are composed of a number of highly interconnected processing elements called neurons or nodes tied together with weighted connections. An iterative learning process begins by feeding input data into the network, which calculates the predicted output based on predeter-mined weights. Comparison of the predicted output and the targeted output leads to adjustment of the weights of the connections, and a new output is calculated. This process is reiterated until the network output closely matches the targeted output. ANNs have the advantage of learning complex patterns and of learning from new information; they have gained increasing popularity for the classifi cation of gene expression profi les in many disciplines. In our laboratory, as illustrated later, the ANN approach has proved to be extremely powerful for the generation of pre-dictive signatures of hepatotoxicity in rats. Support vector machines (SVMs) are a relatively new type of learning algorithm with robust performance with respect to sparse and noisy data [60] . SVMs operate by fi nding an optimal demarcation that most distantly separates positive and negative samples. Unlike other classifi cation algorithms, SVMs perform very well with a large number of data and this feature makes SVMs especially attractive for the classifi cation of gene expression data, which usually have a large number of gene expression endpoints and a limited number of samples [61, 62] . One can easily be confused with the choice and com-plexity of these various classifi cation algorithms. The ultimate criterion to select an optimal prediction model is the prediction accuracy, which is estimated using a testing set that is different from the training set during a validation step. Although most studies evaluate the robustness of predictive models using various validation approaches, no exhaustive survey of the various prediction methods is available in the literature. For predictive toxicogenomics, the choice of methods is likely depen-dent on the experimental design. For instance, an LDA approach is an easy alterna-tive and performs very well if there is a clear difference between the classes of toxicants and/or the sample size is relatively large. However, in the case of a more complex system, such as general hepatotoxicity, with a limited number of gene expression profi les, a machine learning approach, like SVMs or ANNs, could be more appropriate [18, 62] .

Page 19: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

24.4.3 Case Examples

Prediction of Hepatotoxicity Hepatotoxicity has been the toxicity of choice for most toxicogenomic studies. Indeed, the liver is a common target organ for many toxicants and has been extensively studied. This has promoted the generation of comprehensive gene expression databases that facilitate interpretation of liver - derived gene expression profi les. Additionally, the liver is a rather homogeneous tissue, as opposed to tissues such as intestines or brain, for instance. The liver is composed mostly of hepatocytes sharing similar biochemical functions that translate into relatively uniform gene expression profi les. This homogeneity makes identifi ca-tion and interpretation of gene expression changes easier. For these reasons, a wealth of published and proprietary gene expression information is available for the liver, and hepatotoxicity can now be relatively well predicted and understood with gene expression profi ling.

As a proof - of - concept, our laboratory has developed, a few years ago, a quantita-tive approach to predict hepatotoxicity based on gene expression profi les [18] . We fi rst constructed an internal database containing microarray - generated liver gene expression profi les from 3 - day rat toxicology studies using over 50 hepatototoxi-cants and nonhepatotoxicants. All compounds were administered to 3 male rats/group at a high dose (a dose expected to induce hepatotoxicity after 1 week of treatment) and at a lower, nonhepatotoxic dose. A set of marker genes was identi-fi ed that distinguished the hepatotoxicants from the nonhepatotoxicants using ANOVA analysis. Using an artifi cial neural network algorithm coupled with prin-cipal component analysis for dimensionality reduction, a quantitative model was established to classify compounds according to a composite toxicity score (Fig. 24.6 ). A forward validation step was conducted using additional compounds that were not part of the original database. The neural network algorithm could successfully clas-sify these compounds based on their potential to cause hepatotoxicity with a high degree of sensitivity and specifi city. This predictive hepatotoxicity assay is now routinely used to prioritize compounds using exploratory 3 - day repeat - dose rat studies for various projects.

Prediction of Nephrotoxicity A recent publication illustrates how microarray - generated gene expression profi les from the kidneys of rats treated for short periods of time with various nephrotoxicants and nonnephrotoxicants can be used to predict toxic changes in longer term studies [25] . In this study, using a large commercial reference database, a predictive gene expression signature of renal tubular toxicity was developed and shown to predict with good accuracy renal tubular changes that would typically occur after longer exposure to the toxicants. These gene expression profi les were derived from the kidneys of rats treated for 5 days, a time point where no obvious toxic changes were evident, as evidenced by the lack of histopathologic observations and the lack of changes in serum chemistry parameters (most notably serum creatinine and blood urea nitrogen). In a step to confi rm that tubular injury would ultimately occur in longer term studies, the authors dosed rats up to 28 days with the 15 nephrotoxicants used in their positive class, as a phenotypic anchor to the predictive signature. In addition, the signature was validated using compounds naive and not structurally related to the training set. These test compounds included

TOXICOGENOMICS IN DRUG DISCOVERY 885

Page 20: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

886 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

9 nephrotoxicants and 12 nonnephrotoxicants. The signature correctly predicted the future presence or absence of renal tubular injury in 76% of the compound treatments.

Another study used a similar, yet slightly different, approach to study time - and dose - dependent gene expression changes associated with proximal tubular injury in the rat [63] . In this study, Sprague – Dawley rats were treated with a wide variety of classical nephrotoxicants and renal gene expression profi les were evaluated 1, 3, and 7 days after initiation of dosing. The gene expression profi les nicely clustered based on the severity and nature of the pathologic changes and were consistent with tubular degeneration, regeneration, and necrosis. In addition, using an SVM - based approach and a training set of 120 gene expression profi les, a predictive classifi er was developed that was able to predict the type of pathology of a testing set com-posed of 28 gene expression profi les with 100% selectivity and 82% sensitivity.

These two studies are good illustrations of the validity of the toxicogenomic approach for predictive toxicology but are also a good indication of the scale involved in the generation of predictive signatures of suffi cient accuracy to be useful.

FIGURE 24.6 Schematic representation of the ANN - based approach used for the develop-ment of a predictive hepatotoxicity toxicogenomic assay. An internal rat liver gene expression database was fi rst constructed. Reference gene expression profi les were then reduced in dimensionality using PCA. The fi ltered gene expression profi les were fed into an ANN, which processes the information from one layer to the next using multiple weighting factors and transfer functions. The output of the ANN is compared with the ideal toxicity classifi cation and the model is readjusted. This learning process is repeated until the model is able to make an accurate prediction. Using microarray - generated gene expression profi les from male Sprague – Dawley rats treated for 3 days with experimental compounds, the neural network algorithm classifi es the compounds based on their potential to cause hepatotoxicity in rats on a four - category scale ranging from no potential to high potential. (From Ref. 18 , with permission from Elsevier.)

Page 21: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

Both studies required the use of signifi cant numbers of prototypical toxicants for both the training and testing sets, multiple time points, 3 or 4 rats/treatment group/time point, and sophisticated computational algorithms. In other words, such studies are associated with a signifi cant investment of resources, and it is likely that contin-ued validation of these early signatures by a combined effort of the broad scientifi c community will be required to improve accuracy, but also to better estimate their predictivity and suitability in early toxicology testing.

Prediction of Carcinogenicity Assessment of the carcinogenic potential of com-pounds in the rodent bioassay is expensive and lengthy and cannot be performed until late in a program. In addition, the endpoint of the rodent bioassay consists essentially of macroscopically and microscopically detectable tumors, which may spontaneously occur in naive, aging animals and whose incidence varies tremen-dously among animals, thereby requiring a signifi cant number of animals to obtain suffi cient statistical power [64] . Toxicogenomics applied in a predictive mode would allow for a profound reduction in the duration of dosing, thereby reducing the amount of compound required and permitting an earlier assessment of the com-pound. Furthermore, a response at the transcriptome level more homogeneous than the development of tumors could be expected, thereby allowing for a signifi cant reduction in the number of animals to be dosed. It is obvious that any predictive signature of a carcinogenic effect would not dispense with the requirement to conduct the regulatory carcinogenicity studies. However, such predictive signatures would clearly be valuable when applied early to make more informed decisions on compounds.

Using a 5 - day repeat - dose toxicity study in rats, Kramer et al. [65] evaluated whether gene expression profi ling could help identify candidate molecular markers that may predict hepatic carcinogenicity induced by either nongenotoxic or geno-toxic compounds. They hypothesized that there might exist a small number of criti-cal genes whose expression may be predictive of the early events of carcinogenesis initiated by multiple mechanisms. Their study included three dose levels of a number of well - characterized compounds, including fi ve nongenotoxic carcinogens, one genotoxic carcinogen, one carcinogen that may act via genotoxicity, a mitogen, and a noncarcinogenic hepatotoxicant. Analysis of the hepatic gene expression profi les resulted in the identifi cation of several genes (including CYP - R and TSC - 22, which were upregulated and downregulated, respectively) whose expression correlated well with the estimated carcinogenic potential. CYP - R catalyzes the transfer of electrons from NADPH to heme oxygenase, cytochrome - b 5 , and a variety of cyto-chromes P450. As correctly stated by the authors, the upregulation of this gene may simply represent a surrogate marker for cytochrome P450 induction, since many nongenotoxic carcinogens are also cytochrome inducers. Alternatively, CYP - R induction may refl ect a role for oxidative stress in rodent hepatocarcinogenesis. TSC - 22 belongs to a subfamily of leucine zipper transcription factors and had been previously shown to be regulated by various treatments, such as TGF - β , phorbol ester, progesterone, or vesnarionone. Although the exact role of TSC - 22 is uncertain, its repression may refl ect regulation mediated directly by the test articles or an adaptive response to these compounds. In either case, its regulation supports an altered balance between proliferation and apoptosis, consistent with what would be expected with carcinogens.

TOXICOGENOMICS IN DRUG DISCOVERY 887

Page 22: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

888 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

A second study used gene expression profi ling to evaluate whether known geno-toxic carcinogens would induce a common set of genes belonging to defi ned biologi-cal pathways and whether these genes could be used as a predictive signature for hepatic genotoxic carcinogens [44] . This study used potent carcinogens (such as dimethylnitrosamine or 2 - nitrofl uorene) dosed daily for up to 14 days in rats. Gene expression analysis of livers indicated that the following biological pathways were mostly deregulated: DNA damage response, specifi c detoxifi cation response, prolif-eration and survival, and structural changes. This common pattern of deregulation is consistent with what would be expected in the early events of tumorigenesis and could be predictive of later tumor development. The same investigators followed up on these encouraging preliminary results with a study incorporating nongeno-toxic carcinogens [66] . In contrast to what was seen with the genotoxic compounds, nongenotoxic carcinogens impacted distinct cellular pathways/response (including oxidative DNA or protein damage, cell cycle progression, tissue regeneration) that were consistent with compound - specifi c mechanisms and the two - stage model of carcinogenesis.

Several important lessons can be learned from these pioneering studies. First, neither a single gene nor a single pathway will be suffi cient to predict and discrimi-nate the two classes of carcinogens. Second, a predictive gene expression signature of relatively good accuracy can likely be generated once a suffi cient repository of gene expression profi les from a larger variety of carcinogens at different doses and different time points become available. Indeed, for a project of that scale, it is likely that a collaborative effort from the scientifi c community will be necessary for the refi nement and validation of any predictive signature of carcinogenicity.

24.4.4 Predicting Species - Specifi c Toxicity

Toxicologic changes occurring in preclinical species are not necessarily relevant to humans because of species differences in cell biology, physiology, or responses to changes induced by compounds [67] . A classic example of a toxicity with no rele-vance to humans involves the peroxisome proliferators, such as the fi brate class of cholesterol - lowering drugs, that activate the peroxisome proliferator - activated receptor - α (PPAR - α ). Upon chronic administration, these compounds cause hepa-tomegaly and eventually hepatic neoplasms in rats [68, 69] . There are marked species differences in the response to peroxisome proliferators, with mice and rats being highly responsive in contrast to humans. This differential species response correlates directly with the number of hepatic PPAR - α ; PPAR - α is expressed in human liver at only 5 – 10% of rodent liver levels. Consequently, humans are at minimal or no risk to develop hepatic tumors following chronic exposure to peroxi-some proliferators. Toxicogenomics has furthered the understanding of the molecu-lar mechanisms associated with the various effects of several peroxisome proliferators [23, 70, 71] . This allows one to more specifi cally demonstrate the mechanisms of action by which certain compounds lead to rodent hepatomegaly and hepatic car-cinogenesis, thereby improving overall risk assessment. The value of toxicogenomics to understand species - specifi c responses is also illustrated with the case of cyclospo-rine - induced nephrotoxicity [72, 73] . In the kidneys of cyclosporin A - treated rats, a marked downregulation of calbindin - D28kDa, a calcium binding protein, correlates with and causes the accumulation of calcium in tubules and ultimately renal tubular

Page 23: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

calcifi cation in this species. In contrast, cyclosporine does not regulate calbindin - D28kDa expression in the kidneys of dogs and monkeys, two species resistant to cyclosporine - mediated renal toxicity.

24.5 IN VITRO TOXICOGENOMICS

24.5.1 Objectives of In Vitro Toxicogenomics

Compounds are ultimately assessed in animal toxicology studies. Not surprisingly, the vast majority of published toxicogenomic studies to date have been using tissue from animals dosed in vivo . In vivo studies require large amounts of compound and consequently do not allow for an early characterization of the toxicologic profi les of compounds. Moreover, the number of compounds that can be analyzed in animal studies is limited, in part because of the cost and practicalities of these studies. Thus, in vitro systems may signifi cantly improve the throughput and increase the value of toxicogenomics in drug discovery by allowing for an early toxicological character-ization of compounds. In addition, gene expression profi ling using in vitro systems may identify biomarkers of toxicity in the form of gene sets that could be transferred and investigated in preclinical or clinical studies to monitor possible toxic reactions. Finally, gene expression studies in human cells, such as primary human hepatocytes, may, in some cases, be more relevant to the clinical situation or allow for a better understanding of the relevance of toxic changes and for a better assessment of safety risks for humans.

The selection of an appropriate cell system should be guided by the questions to be addressed. If one desires to identify the mechanism of toxicity of a compound, using a cell system that most closely mimics the target organ, such as a primary cell system, would be preferable. However, if one wishes to identify markers of general toxicity, then the cell type may not be as important. In fact, it is likely that identifying general markers of toxicity (such as DNA damage, apoptosis, or oxida-tive stress) would be feasible using cell lines of various origins, such as HeLa or Jurkat cells.

The major limitation of in vitro systems is their inability to recapitulate the overall complexity of the living organism, which limits their potential in detecting lesions associated with multicellular interactions. Most in vitro systems are also short term and therefore inadequate to detect chronic effects [74] . Understanding the predictive value of in vitro systems has remained a major challenge for toxicologists for decades and in vitro toxicogenomics falls into the same predicament. Several studies have addressed the relationship between in vitro and in vivo toxicogenomic results. These studies have demonstrated that different mechanisms of toxicity can be identifi ed using gene expression profi les generated from in vitro systems and that consequently the concept of predictive signatures was also relevant to in vitro systems [75, 76] .

Although gene expression profi ling using in vitro systems can distinguish com-pounds with different mechanisms of toxicity, signatures of satisfactory accuracy and cost - effective gene expression platforms with adequate throughput are necessary for its implementation in a discovery setting. For an in vitro toxicogenomic assay to have practical applications in drug discovery, gene expression signatures need to be

IN VITRO TOXICOGENOMICS 889

Page 24: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

890 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

generated and validated for several relevant toxicologic endpoints. In addition, in order to increase the throughput of gene expression profi ling (e.g., adapting it to a 96 - or 384 - well format), it would be advantageous to reduce the number of genes being monitored. Ideally, one would want to rapidly evaluate compounds for several toxicologic endpoints in a simultaneous fashion in a limited number of wells.

The selection of an appropriate dose represents a critical issue. At this point, there is no clear consensus of what represents an ideal dose for in vitro toxicoge-nomic assessment, and it is likely that the dose selection will depend on the cell type. In our experience, development of robust predictive signatures and character-ization of the toxicologic profi les of compounds require the use of relatively high doses, suffi cient to cause cytotoxicity. For instance, in our primary rat hepatocyte model, we typically characterize compounds at concentrations suffi cient to cause death of 20% of cells. Failure to reach these cytotoxic concentrations will result in an insensitive assay of limited value.

24.5.2 Proof - of - Concept Using Primary Rat Hepatocytes

Most published in vitro toxicogenomic studies have evaluated rat liver cells for several reasons. First, liver is a common target organ of toxicity. Second, rat liver cells are most commonly used for in vitro toxicologic studies. Third, the use of hepa-tocytes offers the opportunity to assess the toxicity associated with certain metabo-lites without prior metabolic activation. Finally, since most in vivo studies have focused on liver, liver - derived cells can be used to correlate in vitro and in vivo data. An in vitro system for hepatotoxicity could consist of isolated perfused livers, liver slices, isolated hepatocytes, or liver cell lines. Isolated perfused livers and liver slices maintain intact cellular interactions and spatial arrangements and allow for long - term studies [74, 77] . These models are also the most appropriate for studying toxic effects on the biliary system, because they contain phenotypically and functionally intact biliary epithelial cells. However, isolated livers and liver slices are resource intensive, low - throughput systems. Cell lines are readily available, cost effective, and generally yield reproducible results over time. However, liver cell lines are quite different from liver or primary hepatocytes in terms of function and phenotype. Gene expression profi les were compared for rat livers, rat liver slices, primary rat hepatocytes cultured on collagen monolayer or collagen sandwich, and two rat liver cell lines (BRL3A and NRL clone 9 cells) [20] . Liver slices were the most similar to intact rat livers, followed by primary hepatocytes in culture. In contrast, the two rat liver cell lines showed little correlation to intact rat livers. In particular, the cell lines expressed very low or undetectable levels of phase I metabolizing enzymes, both at the RNA and protein levels [20] . These results are consistent with data generated in our laboratory (Fig. 24.7 ).

Isolated hepatocytes are not identical but are suffi ciently close to intact livers in terms of gene expression analysis (Fig. 24.8 ) [20, 78] . In addition, they maintain the enzyme architecture and metabolizing capabilities of intact liver in short - term cul-tures [79, 80] . However, isolated hepatocytes, especially of human origin, can be very diffi cult and expensive to obtain. Furthermore, in the case of human hepatocytes, lifestyle differences of the donors, such as smoking or drinking habits, medications, or general health, lead to substantial interindividual variability in gene expression profi les. However, this variability is overall not a major concern, since preliminary

Page 25: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

FIGURE 24.7 Principal component analysis of Affymetrix HG - U133A microarray - generated gene expression profi les from primary human hepatocytes, HeLa cells, and HepG2 cells. This analysis illustrates the major differences at the transcriptome level between primary human hepatocytes and liver cell lines, such as HepG2 cells, which are almost as dissimilar to human hepatocytes as HeLa cells, which are derived from a cervical carcinoma.

FIGURE 24.8 Heatmap illustrating the differences in gene expression profi les between rat liver and primary rat hepatocytes. Gene expression profi les were generated using Affymetrix rat RAE230A microarrays. Genes that are at least showing a two fold difference in expres-sion levels with a p value less than 0.01 between the two systems are shown.

IN VITRO TOXICOGENOMICS 891

Page 26: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

892 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

studies indicate a surprisingly low interindividual variability in response to high exposure to toxicants for the most robustly regulated genes [81] .

In our laboratory, we routinely use primary rat hepatocytes cultured on collagen to characterize compounds at the gene expression level. In this model, we have profi led a large number of compounds, thereby generating an internal database that has allowed us to develop predictive signatures for several toxicologic endpoints (Fig. 24.9 ). For instance, our laboratory has reported results from a study profi ling at the gene expression levels 15 well - characterized hepatotoxicants in primary rat hepatocytes [75] . Compounds with similar mechanisms of toxicity, such as the aro-matic hydrocarbon (Ah) receptor ligands Aroclor 1254 and 3MC, resulted in similar expression profi les and, using unsupervised hierarchical clustering, could clearly be distinguished from other agents such as carbon tetrachloride and allyl alcohol. This study also demonstrated a signifi cant correlation between the genes regulated in vivo and in vitro for some toxicants, such as the Ah receptor ligands. Similar results have been confi rmed, reproduced, or expanded in studies conducted by others. For instance, in a study using primary rat hepatocytes exposed to 11 different hepato-toxicants and a low - density array platform containing only 59 genes, compounds could correctly be classifi ed into different mechanistic hepatotoxic classes [82] . Other studies have also used liver - derived cell lines with some success. Three sepa-rate studies used a human hepatoma cell line (HepG2 cells) to demonstrate that transcriptional analysis differentiates compounds and that in vitro toxicogenomics can be used to further the understanding of toxic mechanisms [21, 83, 84] .

24.5.3 Use of Gene Expression Profi ling to Assess Genotoxicity

Toxicogenomics has also been applied in other cell systems. In particular, gene expression analysis has been evaluated as a potential tool to gain a better under-

FIGURE 24.9 Heatmap of gene expression changes following treatment of primary rat hepatocytes with 16 aryl hydrocarbon receptor (AhR) agonists, 18 negative control com-pounds, and 18 peroxisome proliferator - activated receptor - α (PPAR - α ) agonists for 48 hours. The genes shown were selected by linear discriminant analysis. Using a small internal data-base of in vitro gene expression profi les in primary rat hepatocytes, small gene sets can be identifi ed using various statistical algorithms (in this case linear discriminant analysis) to classify compounds according to their toxic properties for specifi c toxicologic endpoints.

Page 27: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

standing of genotoxic mechanisms. The current in vitro genotoxicity assays using mammalian cells (mammalian mutation and/or chromosomal damage assays) provide a limited insight into genotoxic mechanisms [22, 85] . Not surprisingly, the majority of positive genotoxicity fi ndings for marketed drugs with negative carci-nogenicity data have been observed in these in vitro mammalian assays, demonstrat-ing their low specifi city and the need to develop approaches enabling mechanism - based risk assessment [86] . More specifi cally, the differentiation of DNA - reactive versus DNA - nonreactive mechanisms of genotoxicity would facilitate risk assessment of positive fi ndings in the in vitro mammalian cell - based assays. Several studies have addressed the question of whether gene expression profi ling of in vitro systems would allow a better risk assessment of genotoxicants [85, 87] . Data from these studies demonstrated differences in gene expression profi les between DNA damag-ing and non - DNA - damaging compounds. In various cell types (p53 - defi cient mouse lymphoma cells L5178Y/TK +/ − , TK6 cells), DNA - damaging compounds regulated genes involved in cell cycle regulation, DNA repair, apoptosis, and cellular signaling that were distinct from those regulated by non - DNA - damaging agents [22] . This suggests that, although toxicogenomics will not replace the current standard geno-toxicity assays for hazard identifi cation, it can serve as a useful follow - up experi-mental approach to evaluate compounds with positive fi ndings in these standard assays.

24.5.4 Current and Future Use of In Vitro Toxicogenomics

A tremendous amount of work remains to be done for in vitro toxicogenomics to become a routine tool for toxicologic characterization of compounds in discovery. However, this area is moving at an extremely rapid pace, and several companies are already using gene expression - based in vitro assays for compound characterization and prioritization. For instance, our laboratory has identifi ed predictive signatures for specifi c toxicologic endpoints in the rat hepatocyte model. Some of these signa-tures, as well as the methods used to generate them, have already been published [18, 75, 88] . One can envision that efforts in this fi eld will soon clarify the strengths and limitations of the in vitro toxicogenomic approach and thus establish the role of in vitro toxicogenomic assays in drug discovery.

24.6 TOXICOGENOMICS IN MECHANISTIC TOXICOLOGY

24.6.1 Objectives of Mechanistic Toxicology

The impact of gene expression profi ling in drug discovery and development has so far been mostly evident when used to elucidate the mechanism of a specifi c toxicity. Toxic changes are commonly identifi ed in preclinical studies and obviously not all toxic changes are worth investigating. The decision regarding whether a specifi c toxicologic change needs to be mechanistically understood is based on multiple factors. These factors include, for instance, the nature of the toxicologic change, the exposures at which the change occurs, the species affected, the availability of good backup compounds, or the stage of the program [18] . For instance, there is clear value in trying to understand the mechanism of tumorigenesis for test article - related

TOXICOGENOMICS IN MECHANISTIC TOXICOLOGY 893

Page 28: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

894 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

tumors detected in a lifetime bioassay. At this stage of a program, clinical trials are well underway and an enormous amount of resources have already been invested. In contrast, in an early program, if several backup compounds with a different chemistry but similar physicochemical and pharmacological properties are available, it may not be worth investigating the mechanism of a specifi c toxicologic change occurring with a single lead compound. However, if backup compounds are not yet available, understanding the molecular basis of toxicologic changes may be useful to properly select backup compounds without this toxicologic liability through early structure – toxicity relationship studies during lead optimization. Indeed, when the mechanism of toxicity is understood, appropriate counterscreens can rationally be developed that allow for the selection of backup compounds unlikely to induce the same toxic change. Finally, there is a lack of sensitive and specifi c biomarkers for some toxic changes. Because gene expression profi ling provides a global view of the transcriptional effects induced by a compound, it may identify biomarkers that can subsequently be used in preclinical and conceptually in clinical studies to monitor toxicity.

24.6.2 Case Examples of Mechanistic Toxicology

Hepatotoxicity Early studies have shown that changes in expression of small gene sets can reliably discriminate compounds with distinct mechanism of toxicity in the liver. These gene sets can be used to mechanistically classify compounds and assign a compound with an unknown toxicologic mechanism into predefi ned classes based on mechanism [23, 70] . This allows one to identify toxicity using predictive gene expression signatures as discussed before and also to establish mechanistic hypoth-eses by comparing expression profi les with those present in a database. For instance, our laboratory has investigated the hepatic effects of A - 277249, a thienopyridine inhibitor of NF - κ B - mediated expression of cellular adhesion molecules [89] . This compound induced hepatic changes in rats in a repeat - dose toxicity study, including increased liver weights, changes in serum chemistry (elevations of serum transami-nases, alkaline phosphatase, and gamma glutamyl transferase), and histopathologic changes (hypertrophy and hyperplasia of hepatocytes and biliary epithelial cells). To investigate the mechanism of this hepatotoxicity, a 3 - day repeat - dose rat toxicity study was conducted. Livers were collected and gene expression profi les were gener-ated using microarrays. The compound was observed to induce extensive changes in gene expression. Using a proprietary gene expression database of known hepa-totoxicants, agglomerative hierarchical cluster analysis demonstrated that A - 277249 had a gene expression profi le similar to Aroclor 1254 and 3 - methylcholanthrene (3MC), two well - characterized activators of the aryl hydrocarbon nuclear receptor (AhR), indicating that A - 277249 hepatic changes were, at least in part, mediated by the AhR either directly or through effects on NF - κ B [89] . The AhR is a nuclear receptor that mediates responses to various toxicants, such as the halogenated aro-matic toxicants [90] . In this particular case, the chemical class was abandoned. But, in a case where backup compounds would have been available, one could have used these data to set up an appropriate counterscreen to rapidly evaluate the backup compounds for this toxic mechanism. In particular, one could have evaluated whether these compounds induce an upregulation of CYP1A1 in primary rat hepa-tocytes, since ligands of the AhR are known to induce CYP1A1.

Page 29: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

Intestinal Toxicity and Notch Signaling Gene expression profi ling has rarely been applied to study toxicity of the gastrointestinal tract. The complexity and het-erogeneity of this tissue makes it very diffi cult to investigate toxic changes at the level of the transcriptome. However, two recent studies demonstrated that toxicoge-nomics can be used to elucidate the toxic mechanisms and identify markers of toxic-ity for this tissue [26] . In the last few years, functional γ - secretase inhibitors (FGSIs) have been developed as potential therapeutic agents for Alzheimer disease [26, 91] . FGSIs can block the cleavage of several transmembrane proteins, including the cell fate regulator Notch - 1, which plays an important role in the differentiation of the immune system and gastrointestinal tract. Rats treated with several FGSIs develop a unique gastrointestinal toxicity, characterized by an increase in gastrointestinal weight, distended stomach and small and large intestines, and a mucoid enteropathy related to goblet cell hyperplasia [26, 91] . Microarray analysis of the duodenum or ileum of FGSIs - treated rats identifi ed changes in the expression of several genes, and these changes confi rmed that perturbation in Notch signaling was the mecha-nism for this characteristic enteropathy. These gene expression studies went further and also identifi ed that the gene encoding the serine protease adipsin was signifi -cantly upregulated following treatment with FGSIs. The investigators followed up on this interesting fi nding and demonstrated elevated levels of the adipsin protein in gastrointestinal contents and feces of FGSIs - treated rats, as well as increased numbers of ileal enterocytes expressing adipsin by immunohistochemistry. Based on these data, both laboratories concluded that adipsin may be potentially exploited as a specifi c, sensitive, and noninvasive biomarker of FGSIs - induced gastrointestinal toxicity.

Testicular Toxicity Early toxicant - induced testicular changes are typically subtle in early stages without striking morphologic changes, which may easily be missed unless more sophisticated techniques, such as tubular staging, are used [92] . However, in longer term in vivo studies (4 weeks or longer), changes are typically more pro-nounced and advanced. Approaches that would allow for an earlier detection of testicular changes would be benefi cial in preclinical safety assessment. Current well - established correlating biomarkers (such as serum FSH or semen analysis) are not sensitive enough to allow for an early detection of toxic changes both in preclinical studies and in clinical trials. Recently, a project sponsored by the ILSI Health and Environmental Sciences Institute (HESI) evaluated the suitability and limitations of additional biomarkers, such as Inhibin B, to detect modest testicular dysfunction in rats [93] . While the complete results of this study have not been communicated at this point, preliminary data have indicated that plasma inhibin B is not a useful marker. Finally, little is known in general about the mechanism of toxicity for tes-ticular toxicants. This lack of understanding relates probably to a lower degree of interest by toxicologists in general in contrast to tissues like liver, but also to the complexity of the tissue. The testis is a tissue composed of several different cell types with striking differences in functions and morphology, which can all be targets for toxicants. Furthermore, these different cell types closely interact with each other, modulating their respective functions and status. Consequently, a toxicant targeting a specifi c cell type will ultimately affect, by a secondary mechanism, the status of another cell type. For instance, a toxicant that affects the function of Sertoli cells, the cell supporting the growth, differentiation, and release of germ cells, will

TOXICOGENOMICS IN MECHANISTIC TOXICOLOGY 895

Page 30: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

896 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

ultimately lead to effects on germ cells, such as failure of sperm release or germ cell depletion. This complex cellular interdependence has limited the use of in vitro studies to investigate the mechanism of testicular toxicity.

Several elegant studies have demonstrated that gene expression profi ling can elucidate the molecular basis of testicular toxicity [94 – 96] . For instance, gene expres-sion changes in the testis were evaluated following exposure of mice to bromoch-loroacetic acid, a known testicular toxicant. Using a custom nylon DNA array, numerous changes in gene expression were detected in genes with known functions in fertility, such as Hsp70 – 2 and SP22, as well as genes encoding proteins involved in cell communication, adhesion, and signaling, supporting the hypothesis that the toxicologic effect was the result of disruption of cellular interactions between Sertoli cells and spermatids [94, 97] . Our laboratory has used DNA microarrays to investi-gate the testicular toxicity of another halogenated acetic acid, dibromoacetic acid (DBAA). Oral treatment of rats with DBAA at high doses (250 mg/kg/day) induces specifi c early morphologic changes in the testis, characterized by failed spermiation or failure of release by Sertoli cells of mature step 19 spermatids (Fig. 24.10 ). While this morphologic change strongly suggests that the Sertoli cell is the target cell for DBAA, our results using whole testes indicated that DBAA induced a small but consistent downregulation of cytochrome P450c17 α (CYP17), an enzyme essential for the production of testosterone by gonads (Fig. 24.10 ). These results led us to hypothesize that DBAA may induce, at least in part, its toxicity through an effect on testicular testosterone production. In fact, we were able to show that following treatment of rats with DBAA for as little as 4 days, testicular testosterone contents were signifi cantly decreased, indicating that the decrease in CYP17 expression likely has biological implications. This specifi c study illustrates two important points. First, because gene expression analysis represents a new and

FIGURE 24.10 Histologic and gene expression changes induced in rat testis by dibromo-acetic acid (DBAA). Oral treatment of rats with DBAA at high doses (250 mg/kg/day for 4 days) leads to failed spermiation, characterized by the failure of release by Sertoli cells of mature step 19 spermatids (arrows). This histologic change correlates with a signifi cant down-regulation of cytochrome P450c17a (CYP17) mRNA in the 11 rats evaluated.

Page 31: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

global approach to study the molecular mechanism of various pathologies, one should be extremely cautious not to focus exclusively on earlier hypotheses and not to rely on old dogma. In fact, microarray analysis offers the unique opportunity to generate global molecular data that should be used to generate new, unbiased hypotheses. Once generated, these hypotheses should be addressed using appropri-ately designed studies. Second, although laser - capture microdissection represents a useful technique to study gene expression in complex tissues by focusing on single cell populations, it should also be used with the understanding that critical cell – cell interactions may govern changes in gene expression. In the present case, should we have focused on Sertoli cells only, we probably would have been unable to implicate the decrease in testosterone production as a signifi cant part of the DBAA - induced testicular toxicity.

24.7 TOXICOGENOMICS AND TARGET - RELATED TOXICITY

The recent advances in genomics have spurred a proliferation of novel potential therapeutic targets. The traditional target validation procedure is designed to dem-onstrate that modulation (inhibition or activation) of target activity in relevant disease models can lead to a therapeutic benefi t. The target validation step fre-quently does not include an assessment of the effects of target modulation on normal cellular, organ, or body function. Yet, developing a good understanding of the potential safety liabilities associated with a target is clearly a critical phase of target drugability assessment. Most of the genomics - derived novel targets play major roles in normal cellular function, and consequently modulation of their activ-ity can lead to toxic changes or what is often referred to as target - related toxicity, mechanism - based toxicity, or on - target toxicity.

Gene expression profi ling can provide invaluable information about the biology of these novel targets and help proactively identify target - related safety liabilities. Ideally, this assessment should be conducted at the earliest stages of the discovery process, namely, the target identifi cation/validation stage. With the proliferation of novel attractive targets and the limited resources of discovery units to focus on novel programs, this step becomes especially critical, as it allows a discovery organization to prioritize novel targets based not only on expected therapeutic benefi ts but also on potential safety liabilities. This ultimately leads to a greater focus on programs most likely to ultimately succeed.

Part of this step involves the evaluation of target expression in normal tissues from both preclinical species and human beings, with the development of quantita-tive expression tissue maps [54, 98 – 100] . These tissue maps allow one to proactively identify tissues more likely to be affected by toxic changes. The availability of com-plete tissue repositories or banks is necessary to generate these tissue maps. Target expression can be evaluated at the level of mRNA expression, protein expression, and enzymatic activity in the case of an enzyme or using assays such as receptor binding assays. Evaluation of mRNA levels is clearly the easiest way to generate expression data, but one should remember that, for a large proportion of proteins, mRNA levels do not necessarily correlate well with protein or activity levels. Therefore, it is usually recommended to confi rm or complement mRNA data with appropriate secondary assays. Furthermore, total tissue expression may not

TOXICOGENOMICS AND TARGET-RELATED TOXICITY 897

Page 32: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

898 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

always refl ect the critical role played by some targets, especially in cases where a target is expressed in a small compartment of a complex tissue. These situations can be more effectively addressed with either in situ hybridization or immunohistochemistry.

The generation of expression tissue maps, albeit useful, is usually insuffi cient to evaluate potential safety liabilities associated with a target. In particular, these novel targets are frequently identifi ed based on overexpression in diseased samples. Expression in normal tissues does not necessarily translate into nondrugability or on - target toxicity. The difference in expression levels may result in safety margins suffi cient enough for development. Determination of a safety window, however, is not a trivial task early in a program. Ideally, it should be determined using appropri-ate tool compounds (i.e., compounds with adequate pharmacologic activity against and specifi city for the target). These tool compounds do not need to have candidate properties as long as suffi cient systemic exposure can be achieved for a short repeat - dose study. To control for toxicity related to a chemical class, one should use several tool compounds from different chemical classes and, if available, inactive com-pounds with close structural similarities (such as inactive enantiomers). In our labo-ratory, we typically use a 3 - day repeat - dose study in male rats using carefully selected tool compounds for target assessment. Doses are selected to result in expo-sures suffi cient to achieve effi cacy (as determined by concurrent studies in relevant preclinical effi cacy models), but also in higher exposures, so target - related safety margins can be determined. These high exposures are obviously dependent on the characteristics of the tool compounds available (pharmaceutical properties, ADME/PK, and toxicity) and are often limited early in a program. Tissues for gene expres-sion profi ling are then selected and prioritized based on the biology of the target (partly generated from the expression tissue maps, but also from the literature), as well as prior evaluation of clinical pathology and histopathology changes. Gene expression changes are evaluated in the context of available reference databases and of our battery of gene expression signatures.

Other approaches have been proposed to proactively evaluate on - target toxicity. These approaches use the same tools as for target validation, such as antibodies, genetically engineered mice, or technologies to modulate mRNA expression levels, such as antisense oligonucleotides, ribozymes, or siRNA [101 – 103] . In particular, siRNA has gained popularity in the last few years. We have explored the use of siRNA to understand on - target and off - target effects of compounds in vitro with mixed results. Clearly, issues such as potency and selectivity of siRNA need to be better understood for this approach to be useful in toxicogenomics. Furthermore, gene silencing approaches induce varying degrees of mRNA and protein downregu-lation, which may not mimic the pharmacologic inhibition necessary to achieve effi cacy. Recent advances in viral and nonviral delivery methods have led to the use of RNA interference for in vivo functional genomic studies with successful gene downregulation [104, 105] . Because of the costs of siRNA, these in vivo studies are typically conducted in mice, a species for which robust toxicogenomic databases are not available. Furthermore, the required delivery techniques also induce mild adverse effects that complicate the interpretation of gene expression profi les. Our current experience with in vivo RNAi studies is too limited to provide a reliable assessment of the potential use of gene silencing to better predict target - related toxicity. However, conceptually, this approach appears promising.

Page 33: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

24.8 TOXICOGENOMICS AND IDIOSYNCRATIC TOXICITY

24.8.1 Defi nition of Idiosyncratic Toxicity

Idiosyncratic toxicity refers to a toxicity not related to the pharmacology of the drug that occurs unexpectedly in a small proportion of treated patients, often in a non - dose - dependent manner [106] . This toxic effect is typically not detected in animals and consequently cannot be predicted during preclinical testing or the early phases of clinical trials. This could be due to host specifi city of the toxic reaction or to the insuffi cient size of preclinical animal studies. Examples of drugs that result in idio-syncratic toxicity include troglitazone (withdrawn in 2000 due to liver toxicity), bromfenac (withdrawn in 1998 due to liver toxicity), fenfl uramine (withdrawn in 1997 because of heart valve disease), and cerivastatin (withdrawn in 2001 due to rhabdomyolysis).

Several mechanisms have been proposed to explain the development of idiosyn-cratic drug toxicity. These include the formation of reactive metabolites in certain individuals due to the presence of polymorphisms in drug - metabolizing enzymes, the development of immune - mediated responses to the drug or one of its metabo-lites, a synergistic effect of concurrent low - level infl ammatory reactions, and changes in mitochondrial function and integrity [107 – 111] . While substantial experimental evidence exists that these mechanisms occur for certain drugs, there is currently no reliable way to proactively identify compounds that may lead to idiosyncratic responses in a large patient population. Because gene expression profi ling allows for a global view of responses occurring simultaneously in cells or tissues, it repre-sents a new approach to study the mechanisms of idiosyncratic reactions and to potentially identify predictive signatures of idiosyncratic toxicity. We illustrate this concept using idiosyncratic hepatotoxicity.

24.8.2 Preclinical Models of Idiosyncratic Toxicity

Several models have been used to study the mechanism of idiosyncratic toxicity. Only a few models are covered here. The brown Norway rat has been commonly used as an animal model of idiosyncratic toxicity, especially for the study of sus-pected immune - mediated idiosyncratic reactions. For instance, following exposure to nevirapine or D - pencillamine, brown Norway rats develop toxic changes similar to those seen in humans [112, 113] . Another model is a two - hit model for idiosyn-cratic toxicity, where compounds with potential idiosyncratic liabilities require another underlying factor (such as alcohol intake or concurrent infection) to cause toxicity. This has led to the development of a lipopolysaccharide (LPS) - enhanced toxicity rat model, where rats are coadministered a single dose of compound and LPS. Using this model, several compounds (such as chlorpromazine, ranitidine, and trovafl oxacin) known to cause idiosyncratic toxicity in humans have been shown to induce similar changes in rats [114 – 116] . This LPS - enhanced toxicity rat model has also been investigated at the gene expression level to generate novel hypotheses regarding the exact molecular cascade associated with the toxicity [116, 117] .

Idiosyncratic toxicity has also been studied in in vitro models. In particular, an assumption is that these types of toxicity are host specifi c and that, consequently, only the use of human cells will lead to an accurate insight into the molecular and

TOXICOGENOMICS AND IDIOSYNCRATIC TOXICITY 899

Page 34: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

900 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

biochemical events occurring during toxicity. Using this approach, for instance, troglitazone was shown to lead to decreases in cellular ATP and mitochondrial membrane potential in HepG2 cells, and that, consequently, mitochondrial dysfunc-tion may be, at least in part, the cause of the idiosyncratic toxicity induced by troglitazone [118] . Since the liver is often the target organ of idiosyncratic drug reactions, the latter are commonly studied using cultured human hepatocytes or liver - derived cell lines. Although cell lines such as HepG2 can be useful, recent advances in culturing of human hepatocytes have facilitated their use in toxicoge-nomic studies.

24.8.3 Case Example: Idiosyncratic Hepatotoxicity

As mentioned earlier, the liver is a common target organ of idiosyncratic toxicity and the focus of most research. In particular, the quinolone trovafl oxacin has been a tool compound often used in our laboratory. Quinolones are antibacterial agents that act by inhibiting bacterial DNA gyrase and DNA topoisomerase IV [119] . As a class, they are generally well tolerated, except for trovafl oxacin [120] . Before its regulatory approval in 1997, trovafl oxacin had been tested in over 7000 patients and had not caused any hepatic failures or deaths. Over 2 million people have since received trovafl oxacin and 150 cases of liver toxicity have been reported, including 14 cases of acute liver failure. Four patients required liver transplants and an addi-tional fi ve patients died [121] . Because of this hepatotoxicity, severe restrictions were placed on the use of trovafl oxacin. The drug can now only be administered in life - threatening situations. The mechanism underlying this adverse effect has not yet been determined.

Our objective was, using human hepatocytes and gene expression profi ling, to determine the molecular mechanism of the idiosyncratic toxicity induced by trova-fl oxacin. Human hepatocytes from four different donors were treated with six qui-nolone agents (trovafl oxacin, levofl oxacin, grepafl oxacin, gatifl oxacin, ciprofl oxacin, and clinafl oxacin). Using gene expression profi ling, trovafl oxacin could clearly be distinguished from the other quinolones; the treatment with trovafl oxacin resulted in far more gene expression changes than the other compounds [81] . Many of these gene expression changes involved crucial biological pathways that may be involved in the mechanism underlying trovafl oxacin - induced hepatotoxicity. In particular, trovafl oxacin regulated a number of mitochondrial genes that were not regulated by the other quinolones [81] . In parallel, male Sprague – Dawley rats were treated with levofl oxacin (600 mg/kg/day) and trovafl oxacin (200 mg/kg/day) for 7 days. Consistent with other measures of toxicity (serum chemistry, histopathology), micro-array analysis of the rat livers failed to identify unique gene expression changes induced by trovafl oxacin.

Troglitazone is also commonly investigated as a reference compound for idiosyn-cratic hepatotoxicity. Troglitazone is a thiazolidinedione PPAR - γ agonist that was developed for the treatment of type II diabetes. Troglitazone induced idiosyncratic hepatotoxicity in a small percentage of patients and was removed from the market [106] . In a study similar to ours, human hepatocytes were treated with three thia-zolidinedione compounds (troglitazone, rosiglitazone, and pioglitazone). Trogli-tazone also resulted in a large number of gene expression changes that were not observed with the two other thiazolidinedione compounds [122] .

Page 35: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

Albeit preliminary, these studies demonstrate two important points. First, the combined use of human hepatocytes and gene expression profi ling allowed for the distinction of compounds associated with idiosyncratic hepatotoxicity (trovafl oxa-cin, troglitazone) from compounds of the same chemical classes and not associated hepatotoxicity, suggesting that this in vitro system may be appropriate for studying idiosyncratic hepatotoxicity, and potentially for the early identifi cation of safety liabilities. Second, rat repeat - dose studies did not differentiate trovafl oxacin from another quinolone known not to induce liver failure in humans, indicating that, at least in the case of the quinolones, the traditional rat model may not be suited to detect potential idiosyncratic toxic liabilities even when coupled with a toxicoge-nomic evaluation.

24.9 TOXICOGENOMICS IN REGULATORY SUBMISSIONS

24.9.1 Overview of the FDA Pharmacogenomics Guidance

Because of the potential of gene expression profi ling to improve the safety assess-ment of new chemical entities, the FDA issued a guidance in March 2005 for the regulatory submission of pharmacogenomic data ( http://www.fda.gov/cder/guid-ance ). This guidance refl ects the effort of the FDA to promote the use of genomic technologies in drug development and is also designed to enhance the agency ’ s knowledge of these emerging technologies. In fi nalizing these guidelines, the FDA has openly cooperated with the various stakeholders and has organized appropriate forums to focus on the major issues and principles that the document should cover. The pharmaceutical industry has welcomed this guidance, as it represents an impor-tant stepping stone toward the development of genomics - based drugs and the use of genomics - based safety data. In addition, this guidance provided reassurance to companies that early - stage toxicogenomic experiments would not bring negative regulatory consequences, an important aspect for the wider acceptance of this new technology in the relatively conservative environment of drug safety evaluation.

The guidance clarifi es the FDA ’ s policy on the use of pharmacogenomic data in the drug application review process and covers the application of genomics concepts and technologies to nonclinical, clinical pharmacology, and clinical studies. It pro-vides guidelines to sponsors on pharmacogenomic data submission requirements, the format and procedure for data submission, and how the data will be used in regulatory decision making. In general terms, gene expression data for which sub-mission is required include data used for decision making within a specifi c trial; data used to support scientifi c arguments about mechanism of action, dose selection, safety, or effectiveness; data that will support registration or labeling language; and data generated on previously validated biomarkers. This guidance demonstrates that the FDA is open to and expects the submission of gene expression profi ling data that were generated to support scientifi c contentions related to toxicity.

The guidance defi nes pharmacogenomic tests as follows: “ An assay intended to study interindividual variations in whole - genome or candidate gene, single - nucleotide polymorphism (SNP) maps, haplotype markers, or alterations in gene expression or inactivation that may be correlated with pharmacological and thera-peutic response. In some cases, the pattern or profi le of change is the relevant

TOXICOGENOMICS IN REGULATORY SUBMISSIONS 901

Page 36: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

902 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

biomarker, rather than changes in individual markers. ” This implies that gene expres-sion datasets could ultimately be recognized as validated biomarkers. The guidance also defi nes “ valid biomarkers ” and distinguishes between “ known valid biomark-ers ” and “ probable valid biomarkers. ” Valid biomarkers are measured in an analyti-cal test system with well - established performance characteristics, and an established scientifi c framework or body of evidence exists to understand the signifi cance of the test results. For a known valid biomarker, a widespread agreement exists in the medical or scientifi c community about the signifi cance of the results. In contrast, for a probable valid biomarker, there is no widespread agreement, but only a scientifi c framework or body of evidence suffi cient to elucidate the signifi cance of the test results. An example would be a biomarker developed by a sponsor and not available for public scientifi c scrutiny or for independent verifi cation. This distinction also reemphasizes the enormous amount of work and improvement that will be needed in the future to make gene expression data suitable for regulatory decision making. This includes obviously an improved scientifi c framework for data interpretation through the use of larger, more complete reference databases, but also improved quality control of laboratory procedures, a better understanding of the compara-bility of different platforms, and some better - defi ned processes to validate biomarkers.

The FDA recognizes that, currently, most gene expression profi ling data are exploratory and would therefore not be required for submission. However, to be prepared to appropriately evaluate future submissions, FDA scientists need to develop an understanding of a variety of relevant scientifi c issues. The Voluntary Genomic Data Submission (VGDS) provides the material necessary to develop this understanding and is reviewed by a cross - center Interdisciplinary Pharmacoge-nomic Review Group (IPRG). For more information, the reader is referred to the FDA web site, which reviews the frequently asked questions regarding VGDS ( http://www.fda.gov/cder/genomics/FAQ.htm ). All VGDS data are protected from disclosure either outside the FDA or to review divisions, are routed directly to the IPRG, and stored on a secured, separate server. These data are not distributed outside the IPRG without the prior agreement of the sponsor and are not to be used for regulatory decision making. The concept of VGDS has in general been well received by the pharmaceutical industry. Voluntary submissions allow sponsors to familiarize FDA scientists with genomic data and their interpretation and, at the same time, to learn about the regulatory decision - making process and expectations involving genomic data. Ultimately, this could prevent delays in future submissions containing required genomic data. So far, many formal submissions have occurred.

24.9.2 Future Impact of Toxicogenomic Data in Regulatory Decision Making

The development of a guidance demonstrates that the FDA is expecting genomic data to become an integral part of the risk assessment of pharmaceuticals. It is, however, diffi cult at this point to objectively predict the role that a rapidly evolving technology will have in regulatory decision making. When the concept of toxicoge-nomics fi rst emerged, expectations were high and to some extent unrealistic. While the technology and the development of appropriate analytical tools have consider-ably improved the ability to generate and interpret large sets of data, it still should

Page 37: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

REFERENCES 903

not be considered a mature approach to evaluate toxicology. Furthermore, it is unlikely that toxicogenomics will replace most traditional toxicology studies that are currently part of a regulatory package. Rather, toxicogenomics will more likely complement and increase the value of these studies by providing an improved understanding of the relevance of preclinical toxicologic changes to humans. Toxi-cogenomics also represents a largely needed novel approach to identify additional biomarkers of safety that could potentially be used to improve monitoring of adverse events in the clinics, resulting in safer clinical trials and potentially earlier identifi ca-tion of outliers with increased sensitivity to particular adverse events. However, it is still unclear how genomics - based biomarkers will be validated to become an integral part of regulatory decision making. The validation of these biomarkers is complex and clearly context specifi c.

24.10 CONCLUSION

To address the high failure rate due to toxicity, the preclinical toxicologist in the pharmaceutical industry needs to more accurately identify a hazard earlier and provide an improved risk assessment of compounds in discovery and development. In contrast to what was available in the past, toxicogenomics requires a broader and slightly different expertise, often only achieved by multidisciplinary teams com-posed of toxicologists, molecular biologists, bioinformaticians, and biostatisticians, among others. The formation of productive teams composed of people with strik-ingly different scientifi c backgrounds, diverse expertise, and sometimes confl icting interest can be a signifi cant challenge. Properly generated and curated reference databases are also needed to fully exploit the potential benefi ts of toxicogenomics. Furthermore, major improvements are needed for the microarray technology to become cost effective and meet performance characteristics amenable to its routine implementation in preclinical risk assessment. All these requirements can only be achieved through a substantial investment in human resources and hardware. Most major pharmaceutical companies have committed to signifi cant investments, but so far toxicogenomics still has not been fully integrated in many organizations. This may refl ect the current shortage of experts with a pragmatic vision of the future use of this technology, as well as the traditionally conservative nature of toxicology departments in the pharmaceutical industry. Nevertheless, if the trend seen in the last few years continues, it is realistic to predict that molecular toxicology and toxi-cogenomics will play a growing strategic role in the risk assessment of new chemical entities in the pharmaceutical industry.

REFERENCES

1. Service RF . Surviving the blockbuster syndrome . Science 2004 ; 303 : 1796 – 1799 . 2. Grabowski HG , Vernon JM . Returns to R & D on new drug introductions in the 1980s .

J Health Econ 1994 ; 13 : 383 – 406 . 3. Rawlings MD . Cutting the cost of drug development ? Nat Rev Drug Discov

2004 ; 3 : 360 – 364 .

Page 38: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

904 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

4. Kola I , Landis J . Can the pharmaceutical industry reduce attrition rates ? Nat Rev Drug Discov 2004 ; 3 : 711 – 715 .

5. Prentis RA , Lis Y , Walker SR . Pharmaceutical innovation by the seven UK - owned pharmaceutical companies (1964 – 1985) . Br J Clin Pharmacol 1988 ; 25 : 387 – 396 .

6. Ulrich RG , Friend SH . Toxicogenomics and drug discovery: will new technologies help us produce better drugs ? Nat Rev 2002 ; 1 : 84 – 88 .

7. MacNeil JS . Genomics goes downstream . Genome Technol 2005 ; 54 : 24 – 30 . 8. Segal E , Friedman N , Kaminski N , Regev A , Koller D . From signatures to models: under-

standing cancer using microarrays . Nat Genet 2005 ; 37 (Suppl): S38 – S45 . 9. Chu TM , Deng S , Wolfi nger R , Paules RS , Hamadeh HK . Cross - site comparison of gene

expression data reveals high similarity . Environ Health Perspect 2004 ; 112 : 449 – 455 . 10. Shi L , Tong W , Fang H , Scherf U , Han J , Puri RK , et al. Cross - platform comparability of

microarray technology: intra - platform consistency and appropriate data analysis proce-dures are essential . BMC Bioinformatics 2005 ; 6 ( Suppl 2 ): S12 .

11. Yauk CL , Berndt ML , Williams A , Douglas GR . Comprehensive comparison of six microarray technologies . Nucleic Acids Res 2004 ; 32 : e124 .

12. Chuaqui RF , Bonner RF , Best CJ , Gillespie JW , Flaig MJ , Hewitt SM , et al. Post - analysis follow - up and validation of microarray experiments . Nat Genet 2002 ; 32 (Suppl): 509 – 514 .

13. Baker VA , Harries HM , Waring JF , Duggan CM , Ni HA , Jolly RA , et al. Clofi brate - induced gene expression changes in rat liver: a cross - laboratory analysis using mem-brane cDNA arrays . Environ Health Perspect 2004 ; 112 : 428 – 438 .

14. Waring JF , Ulrich RG , Flint N , Morfi tt D , Kalkuhl A , Staedtler F , et al. Interlaboratory evaluation of rat hepatic gene expression changes induced by methapyrilene . Environ Health Perspect 2004 ; 112 : 439 – 448 .

15. Thomas RS , Rank DR , Penn SG , Zastrow GM , Hayes KR , Hu T , et al. Applications of genomics to toxicology research . Environ Health Perspect 2002 ; 110 : 919 – 923 .

16. Schena M , Shalon D , Davis RW , Brown PO . Quantitative monitoring of gene expression patterns with a complementary DNA microarray . Science 1995 ; 270 : 467 – 470 .

17. Nuwaysir EF , Bittner M , Trent J , Barrett JC , Afshari CA . Microarrays and toxicology: the advent of toxicogenomics . Mol Carcinog 1999 ; 24 : 153 – 159 .

18. Yang Y , Blomme EA , Waring JF . Toxicogenomics in drug discovery: from preclinical studies to clinical trials . Chem Biol Interact 2004 ; 150 : 71 – 85 .

19. Innovation or Stagnation? Challenge and Opportunity on the Critical Path to New Medical Products . Washington, DC: US Department of Health and Human Services, FDA; 2004 .

20. Boess F , Kamber M , Romer S , Gasser R , Muller D , Albertini S , et al. Gene expression in two hepatic cell lines, cultured primary hepatocytes, and liver slices compared to the in vivo liver gene expression in rats: possible implications for toxicogenomics use of in vitro systems . Toxicol Sci 2003 ; 73 : 386 – 402 .

21. Hong Y , Muller UR , Lai F . Discriminating two classes of toxicants through expression analysis of HepG2 cells with DNA arrays . Toxicol In Vitro 2003 ; 17 : 85 – 92 .

22. Newton RK , Aardema M , Aubrecht J . The utility of DNA microarrays for characterizing genotoxicity . Environ Health Perspect 2004 ; 112 : 420 – 422 .

23. Hamadeh HK , Bushel PR , Jayadev S , Martin K , DiSorbo O , Sieber S , et al. Gene expres-sion analysis reveals chemical - specifi c profi les . Toxicol Sci 2002 ; 67 : 219 – 231 .

24. Lee J , Richburg JH , Shipp EB , Meistrich ML , Boekelheide K . The Fas system, a regulator of testicular germ cell apoptosis, is differentially up - regulated in Sertoli cell versus germ cell injury of the testis . Endocrinology 1999 ; 140 : 852 – 858 .

Page 39: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

REFERENCES 905

25. Fielden MR , Eynon BP , Natsoulis G , Jarnagin K , Banas D , Kolaja KL . A gene expression signature that predicts the future onset of drug - induced renal tubular toxicity . Toxicol Pathol 2005 ; 33 : 675 – 683 .

26. Searfoss GH , Jordan WH , Calligaro DO , Galbreath EJ , Schirtzinger LM , Berridge BR , et al. Adipsin, a biomarker of gastrointestinal toxicity mediated by a functional gamma - secretase inhibitor . J Biol Chem 2003 ; 278 : 46107 – 46116 .

27. Shi L , Tong W , Goodsaid F , Frueh FW , Fang H , Han T , et al. QA/QC: challenges and pitfalls facing the microarray community and regulatory agencies . Expert Rev Mol Diagn 2004 ; 4 : 761 – 777 .

28. Mattes WB , Pettit SD , Sansone SA , Bushel PR , Waters MD . Database development in toxicogenomics: issues and efforts . Environ Health Perspect 2004 ; 112 : 495 – 505 .

29. Shi L , Tong W , Su Z , Han T , Han J , Puri RK , et al. Microarray scanner calibration curves: characteristics and implications . BMC Bioinformatics 2005 ; 6 ( Suppl 2 ): S11 .

30. Churchill GA . Fundamentals of experimental design for cDNA microarrays . Nat Genet 2002 ; 32 (Suppl): 490 – 495 .

31. Higgins MA , Berridge BR , Mills BJ , Schultze AE , Gao H , Searfoss GH , et al. Gene expression analysis of the acute phase response using a canine microarray . Toxicol Sci 2003 ; 74 : 470 – 484 .

32. Sugai T , Kawamura M , Iritani S , Araki K , Makifuchi T , Imai C , et al. Prefrontal abnor-mality of schizophrenia revealed by DNA microarray: impact on glial and neurotrophic gene expression . Ann NY Acad Sci 2004 ; 1025 : 84 – 91 .

33. Mirnics K , Pevsner J . Progress in the use of microarray technology to study the neuro-biology of disease . Nat Neurosci 2004 ; 7 : 434 – 439 .

34. Galvin JE , Ginsberg SD . Expression profi ling and pharmacotherapeutic development in the central nervous system . Alzheimer Dis Assoc Disord 2004 ; 18 : 264 – 269 .

35. Todd R , Lingen MW , Kuo WP . Gene expression profi ling using laser capture microdis-section . Expert Rev Mol Diagn 2002 ; 2 : 497 – 507 .

36. Irwin RD , Boorman GA , Cunningham ML , Heinloth AN , Malarkey DE , Paules RS . Application of toxicogenomics to toxicology: basic concepts in the analysis of microarray data . Toxicol Pathol 2004 ; 32 : 72 – 83 .

37. Hayes KR , Bradfi eld CA . Advances in toxicogenomics . Chem Res Toxicol 2005 ; 18 : 403 – 414 .

38. Vinciotti V , Khanin R , D ’ Alimonte D , Liu X , Cattini N , Hotchkiss G , et al. An experi-mental evaluation of a loop versus a reference design for two - channel microarrays . Bioinformatics 2005 ; 21 : 492 – 501 .

39. Kolaja K , Fielden M . The impact of toxicogenomics on preclinical development: from promises to realized value to regulatory implications . Preclinica 2004 ; 2 : 122 – 129 .

40. Fielden MR , Pearson C , Brennan R , Kolaja KL . Preclinical drug safety analysis by che-mogenomic profi ling in the liver . Am J Pharmacogenomics 2005 ; 5 : 161 – 171 .

41. Guerreiro N , Staedtler F , Grenet O , Kehren J , Chibout SD . Toxicogenomics in drug development . Toxicol Pathol 2003 ; 31 : 471 – 479 .

42. Ganter B , Tugendreich S , Pearson CI , Ayanoglu E , Baumhueter S , Bostian KA , et al. Development of a large - scale chemogenomics database to improve drug candidate selection and to understand mechanisms of chemical toxicity and action . J Biotechnol 2005 ; 119 : 219 – 244 .

43. Bushel PR , Hamadeh HK , Bennett L , Green J , Ableson A , Misener S , et al. Computa-tional selection of distinct class - and subclass - specifi c gene expression signatures . J Biomed Inform 2002 ; 35 : 160 – 170 .

Page 40: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

906 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

44. Ellinger - Ziegelbauer H , Stuart B , Wahle B , Bomann W , Ahr HJ . Characteristic expres-sion profi les induced by genotoxic carcinogens in rat liver . Toxicol Sci 2004 ; 77 : 19 – 34 .

45. Thomas RS , Rank DR , Penn SG , Zastrow GM , Hayes KR , Pande K , et al. Identifi cation of toxicologically predictive gene sets using cDNA microarrays . Mol Pharmacol 2001 ; 60 : 1189 – 1194 .

46. Waring JF , Jolly RA , Ciurlionis R , Lum PY , Praestgaard JT , Morfi tt DC , et al. Clustering of hepatotoxins based on mechanism of toxicity using gene expression profi les . Toxicol Appl Pharmacol 2001 ; 175 : 28 – 42 .

47. Waring JF , Cavet G , Jolly RA , McDowell J , Dai H , Ciurlionis R , et al. Development of a DNA microarray for toxicology based on hepatotoxin - regulated sequences . Environ Health Perspect 2003 ; 111 : 863 – 870 .

48. Brazma A , Hingamp P , Quackenbush J , Sherlock G , Spellman P , Stoeckert C , et al. Minimum information about a microarray experiment (MIAME) — toward standards for microarray data . Nat Genet 2001 ; 29 : 365 – 371 .

49. Mattes WB . Annotation and cross - indexing of array elements on multiple platforms . Environ Health Perspect 2004 ; 112 : 506 – 510 .

50. Luhe A , Suter L , Ruepp S , Singer T , Weiser T , Albertini S . Toxicogenomics in the phar-maceutical industry: hollow promises or real benefi t ? Mutat Res 2005 ; 575 : 102 – 115 .

51. Tennant RW . The national center for toxicogenomics: using new technologies to inform mechanistic toxicology . Environ Health Perspect 2002 ; 110 : 8 – 10 .

52. Waters MD , Fostel JM . Toxicogenomics and systems toxicology: aims and prospects . Nat Rev Genet 2004 ; 5 : 936 – 948 .

53. Castle AL , Carver MP , Mendrick DL . Toxicogenomics: a new revolution in drug safety . Drug Discov Today 2002 ; 7 : 728 – 736 .

54. Searfoss GH , Ryan TP , Jolly RA . The role of transcriptome analysis in pre - clinical toxi-cology . Curr Mol Med 2005 ; 5 : 53 – 64 .

55. Suter L , Babiss LE , Wheeldon EB . Toxicogenomics in predictive toxicology in drug development . Chem Biol 2004 ; 11 : 161 – 171 .

56. Natsoulis G , El Ghaoui L , Lanckriet GR , Tolley AM , Leroy F , Dunlea S , et al. Classifi ca-tion of a large microarray data set: algorithm comparison and analysis of drug signatures . Genome Res 2005 ; 15 : 724 – 736 .

57. Golub TR , Slonim DK , Gaasenbeek JR , Caligiuri MA , et al. Molecular classifi cation of cancer: class discovery and class prediction by gene expression monitoring . Science 1999 ; 286 : 531 – 537 .

58. Tusher VG , Tibshirani R , Chu G . Signifi cance analysis of microarrays applied to the ionizing radiation response . Proc Natl Acad Sci USA 2001 ; 98 : 5116 – 5121 .

59. Khan J , Wei JS , Ringner M , Saal LH , Westermann F , et al. Classifi cation and diagnostic prediction of cancers using gene expression profi ling and artifi cial neural networks . Nat Med 2001 ; 7 : 673 – 679 .

60. Cristianini N , Shawe - Taylor J . An Introduction to Support Vector Machines . Cambridge UK : Cambridge University Press ; 2000 .

61. Furey TS , Cristianini N , Duffy N , Bednarski DW , Schummer M , Haussler D . Support vector machine classifi cation and validation of cancer tissue samples using microarray expression data . Bioinformatics 2000 ; 16 : 906 – 914 .

62. Steiner G , Suter L , Boess F , Gasser R , de Vera MC , Albertini S , et al. Discriminating different classes of toxicants by transcript profi ling . Environ Health Perspect 2004 ; 112 : 1236 – 1248 .

Page 41: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

REFERENCES 907

63. Thukral SK , Nordone PJ , Hu R , Sullivan L , Galambos E , Fitzpatrick VD , et al. Prediction of nephrotoxicant action and identifi cation of candidate toxicity - related biomarkers . Toxicol Pathol 2005 ; 33 : 343 – 355 .

64. Greim H , Gelbke HP , Reuter U , Thielmann HW , Edler L . Evaluation of historical control data in carcinogenicity studies . Hum Exp Toxicol 2003 ; 22 : 541 – 549 .

65. Kramer JA , Curtiss SW , Kolaja KL , Alden CL , Blomme EA , Curtiss WC , et al. Acute molecular markers of rodent hepatic carcinogenesis identifi ed by transcription profi ling . Chem Res Toxicol 2004 ; 17 : 463 – 470 .

66. Ellinger - Ziegelbauer H , Stuart B , Wahle B , Bomann W , Ahr HJ . Comparison of the expression profi les induced by genotoxic and nongenotoxic carcinogens in rat liver . Mutat Res 2005 ; 575 : 61 – 84 .

67. Waring JF , Ulrich RG . The impact of genomics - based technologies on drug safety evalu-ation . Annu Rev Pharmacol Toxicol 2000 ; 40 : 335 – 352 .

68. Cattley RC . Peroxisome proliferators and receptor - mediated hepatic carcinogenesis . Toxicol Pathol 2004 ; 32 ( Suppl 2 ): 6 – 11 .

69. Holden PR , Tugwood JD . Peroxisome proliferator - activated receptor alpha: role in rodent liver cancer and species differences . J Mol Endocrinol 1999 ; 22 : 1 – 8 .

70. Hamadeh HK , Bushel PR , Jayadev S , DiSorbo O , Bennett L , Li L , et al. Prediction of compound signature using high density gene expression profi ling . Toxicol Sci 2002 ; 67 : 232 – 240 .

71. Kramer JA , Blomme EA , Bunch RT , Davila JC , Jackson CJ , Jones PF , et al. Transcription profi ling distinguishes dose - dependent effects in the livers of rats treated with clofi brate . Toxicol Pathol 2003 ; 31 : 417 – 431 .

72. Badr MZ , Belinsky SA , Kauffman FC , Thurman RG . Mechanism of hepatotoxicity to periportal regions of the liver lobule due to allyl alcohol: role of oxygen and lipid per-oxidation . J Pharmacol Exp Ther 1986 ; 238 : 1138 – 1142 .

73. Butterworth KR , Carpanini FM , Dunnington D , Grasso P , Pelling D . The production of periportal necrosis by allyl alcohol in the rat . Br J Pharmacol 1978 ; 63 : 353P – 354P .

74. Amin K , Ip C , Jimenez L , Tyson C , Behrsing H . In vitro detection of differential and cell - specifi c hepatobiliary toxicity induced by geldanamycin and 17 - allylaminogeldana-mycin using dog liver slices . Toxicol Sci 2005 ; 87 : 442 – 450 .

75. Waring JF , Ciurlionis R , Jolly RA , Heindel M , Ulrich RG . Microarray analysis of hepa-totoxins in vitro reveals a correlation between gene expression profi les and mechanisms of toxicity . Toxicol Lett 2001 ; 120 : 359 – 368 .

76. Burczynski ME , McMillian M , Cirvo J , Li L , Parker JB , Dunn RT , et al. Toxicogenom-ics - based discrimination of toxic mechanism in HepG2 human hepatoma cells . Toxicol Sci 2000 ; 58 : 399 – 415 .

77. Gomez - Lechon MJ , Ponsoda X , Bort R , Castell JV . The use of cultured hepatocytes to investigate the metabolism of drugs and mechanisms of drug hepatotoxicity . Altern Lab Anim 2001 ; 29 : 225 – 231 .

78. Waring JF , Ciurlionis R , Jolly RA , Heindel M , Gagne G , Fagerland JA , et al. Isolated human hepatocytes in culture display markedly different gene expression patterns depending on attachment status . Toxicol In Vitro 2003 ; 17 : 693 – 701 .

79. Li AP , Reith MK , Rasmussen A , Gorski JC , Hall SD , Xu L , et al. Primary human hepa-tocytes as a tool for the evaluation of structure – activity relationship in cytochrome P450 induction potential of xenobiotics: evaluation of rifampin, rifapentine and rifabutin . Chem Biol Interact 1997 ; 107 : 17 – 30 .

80. Ulrich RG , Bacon JA , Cramer CT , Peng GW , Petrella DK , Stryd RP , et al. Cultured hepatocytes as investigational models for hepatic toxicity: practical applications in drug discovery and development . Toxicol Lett 1995 ; 82 – 83 : 107 – 115 .

Page 42: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

908 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

81. Liguori MJ , Anderson LM , Bukofzer S , McKim J , Pregenzer JF , Retief J , et al. Microarray analysis in human hepatocytes suggests a mechanism for hepatotoxicity induced by trovafl oxacin . Hepatology 2005 ; 41 : 177 – 186 .

82. de Longueville F , Surry D , Meneses - Lorente G , Bertholet V , Talbot V , Evrard S , et al. Gene expression profi ling of drug metabolism and toxicology markers using a low - density DNA microarray . Biochem Pharmacol 2002 ; 64 : 137 – 149 .

83. Harries HM , Fletcher ST , Duggan CM , Baker VA . The use of genomics technology to investigate gene expression changes in cultured human cells . Toxicol In Vitro 2001 ; 15 : 399 – 405 .

84. Morgan KT , Ni H , Brown HR , Yoon L , Crosby LM , et al. Application of cDNA microar-ray technology to in vitro toxicology and the selection of genes for a real - time RT - PCR - based screen for oxidative stress in Hep - G2 cells . Toxicol Pathol 2002 ; 30 : 435 – 451 .

85. Aubrecht J , Caba E . Gene expression profi le analysis: an emerging approach to investi-gate mechanisms of genotoxicity . Pharmacogenomics 2005 ; 6 : 419 – 428 .

86. Snyder RD , Green JW . A review of the genotoxicity of marketed pharmaceuticals . Mutat Res 2001 ; 488 : 151 – 169 .

87. Dickinson DA , Warnes GR , Quievryn G , Messer J , Zhitkovich A , Rubitski E , et al. Dif-ferentiation of DNA reactive and non - reactive genotoxic mechanisms using gene expression profi le analysis . Mutat Res 2004 ; 549 : 29 – 41 .

88. Yang Y , Abel S , Ciurlionis R , Waring JF . Development of gene expression - based in vitro assays for the effi cient toxicity characterization of compounds . Pharmacogenomics 2006 ; 7 : 177 – 186 .

89. Waring JF , Gum R , Morfi tt D , Jolly RA , Ciurlionis R , Heindel M , et al. Identifying toxic mechanisms using DNA microarrays: evidence that an experimental inhibitor of cell adhesion molecule expression signals through the aryl hydrocarbon nuclear receptor . Toxicology 2002 ; 181 – 182 : 537 – 550 .

90. Denison MS , Nagy SR . Activation of the aryl hydrocarbon receptor by structurally diverse exogenous and endogenous chemicals . Annu Rev Pharmacol Toxicol 2003 ; 43 : 309 – 334 .

91. Milano J , McKay J , Dagenais C , Foster - Brown L , Pognan F , Gadient R , et al. Modulation of notch processing by gamma - secretase inhibitors causes intestinal goblet cell metapla-sia and induction of genes known to specify gut secretory lineage differentiation . Toxicol Sci 2004 ; 82 : 341 – 358 .

92. Creasy DM . Evaluation of testicular toxicity in safety evaluation studies: the appropriate use of spermatogenic staging . Toxicol Pathol 1997 ; 25 : 119 – 131 .

93. Stewart J . Inhibin B as a potential biomarker of testicular toxicity . Toxicologist 2005 ; 74 : 6 .

94. Richburg JH , Johnson KJ , Schoenfeld HA , Meistrich ML , Dix DJ . Defi ning the cellular and molecular mechanisms of toxicant action in the testis . Toxicol Lett 2002 ; 135 : 167 – 183 .

95. Adachi T , Ono Y , Koh KB , Takashima K , Tainaka H , Matsuno Y , et al. Long - term altera-tion of gene expression without morphological change in testis after neonatal exposure to genistein in mice: toxicogenomic analysis using cDNA microarray . Food Chem Toxicol 2004 ; 42 : 445 – 452 .

96. Adachi T , Koh KB , Tainaka H , Matsuno Y , Ono Y , Sakurai K , et al. Toxicogenomic dif-ference between diethylstilbestrol and 17beta - estradiol in mouse testicular gene expres-sion by neonatal exposure . Mol Reprod Dev 2004 ; 67 : 19 – 25 .

97. Rockett JC , Christopher LJ , Brian GJ , Krawetz SA , Hughes MR , Hee KK , et al. Devel-opment of a 950 - gene DNA array for examining gene expression patterns in mouse testis . Genome Biol 2001 ; 2 : 14.1 – 14.9 .

Page 43: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

REFERENCES 909

98. Zappa F , Ward T , Pedrinis E , Butler J , McGown A . NAD(P)H: quinone oxidoreductase 1 expression in kidney podocytes . J Histochem Cytochem 2003 ; 51 : 297 – 302 .

99. Su AI , Wiltshire T , Batalov S , Lapp H , Ching KA , Block D , et al. A gene atlas of the mouse and human protein - encoding transcriptomes . Proc Natl Acad Sci USA 2004 ; 101 : 6062 – 6067 .

100. Su AI , Cooke MP , Ching KA , Hakak Y , Walker JR , Wiltshire T , et al. Large - scale analysis of the human and mouse transcriptomes . Proc Natl Acad Sci USA 2002 ; 99 : 4465 – 4470 .

101. Honore P , Kage K , Mikusa J , Watt AT , Johnston JF , Wyatt JR , et al. Analgesic profi le of intrathecal P2X(3) antisense oligonucleotide treatment in chronic infl ammatory and neuropathic pain states in rats . Pain 2002 ; 99 : 11 – 19 .

102. Semizarov D , Frost L , Sarthy A , Kroeger P , Halbert DN , Fesik SW . Specifi city of short interfering RNA determined through gene expression signatures . Proc Natl Acad Sci USA 2003 ; 100 : 6347 – 6352 .

103. Zambrowicz BP , Turner CA , Sands AT . Predicting drug effi cacy: knockouts model pipe-line drugs of the pharmaceutical industry . Curr Opin Pharmacol 2003 ; 3 : 563 – 570 .

104. Lu PY , Xie F , Woodle MC . In vivo application of RNA interference: from functional genomics to therapeutics . Adv Genet 2005 ; 54 : 117 – 142 .

105. Li L , Lin X , Staver M , Shoemaker A , Semizarov D , Fesik SW , et al. Evaluating hypoxia - inducible factor - 1alpha as a cancer therapeutic target via inducible RNA interference in vivo . Cancer Res 2005 ; 65 : 7249 – 7258 .

106. Waring JF , Anderson MG . Idiosyncratic toxicity: mechanistic insights gained from analy-sis of prior compounds. Curr Opin Drug Discov Dev 2005 ; 8 : 59 – 65 .

107. Williams DP , Park BK . Idiosyncratic toxicity: the role of toxicophores and bioactivation . Drug Discov Today 2003 ; 8 : 1044 – 1050 .

108. Knowles SR , Uetrecht J , Shear NH . Idiosyncratic drug reactions: the reactive metabolite syndrome . Lancet 2000 ; 356 : 1587 – 1591 .

109. Uetrecht J . Prediction of a new drug ’ s potential to cause idiosyncratic reactions . Curr Opin Drug Discov Dev 2001 ; 4 : 55 – 59 .

110. Roth RA , Luyendyk JA , Maddox JF , Ganey PE . Infl ammation and drug idiosyncrasy — is there a connection ? J Exp Ther 2003 ; 307 : 1 – 8 .

111. Park BK , Kitteringham NR , Powell H , Pirmohamed M . Advances in molecular toxicol-ogy — towards understanding idiosyncratic drug toxicity . Toxicology 2000 ; 153 : 39 – 60 .

112. Shenton JM , Teranishi M , Abu - Asab MS , Yager JA , Uetrecht JP . Characterization of a potential animal model of an idiosyncratic drug reaction: nevirapine - induced skin rash in the rat . Chem Res Toxicol 2003 ; 16 : 1078 – 1089 .

113. Masson MJ , Uetrecht JP . Tolerance induced by low dose d - penicillamine in the brown Norway rat model of drug - induced autoimmunity is immune - mediated . Chem Res Toxicol 2004 ; 17 : 82 – 94 .

114. Luyendyk JP , Maddox JF , Cosma GN , Ganey PE , Cockerell GL , Roth RA . Ranitidine treatment during a modest infl ammatory response precipitates idiosyncrasy - like liver injury in rats . J Pharmacol Exp Ther 2003 ; 307 : 9 – 16 .

115. Buchweitz JP , Ganey PE , Bursian SJ , Roth RA . Underlying endotoxemia augments toxic responses to chlorpromazine: is there a relationship to drug idiosyncrasy ? J Pharmacol Exp Ther 2002 ; 300 : 460 – 467 .

116. Waring JF , Liguori MJ , Luyendyk JP , Maddox JF , Ganey PE , Stachlewitz RF , et al. Microarray analysis of LPS potentiation of trovafl oxacin - induced liver injury in rats suggests a role for proinfl ammatory chemokines and neutrophils . J Pharmacol Exp Ther 2006 ; 316 : 1080 – 1087 .

Page 44: Preclinical Development Handbook || Toxicogenomics in Preclinical Development

910 TOXICOGENOMICS IN PRECLINICAL DEVELOPMENT

117. Luyendyk JP , Mattes WB , Burgoon LD , Zacharewski TR , Maddox JF , Cosma GN , et al. Gene expression analysis points to hemostasis in livers of rats cotreated with lipopoly-saccharide and ranitidine . Toxicol Sci 2004 ; 80 : 203 – 213 .

118. Tirmenstein MA , Hu CX , Gales TL , Maleeff BE , Narayanan PK , Kurali E , et al. Effects of troglitazone on HepG2 viability and mitochondrial function . Toxicol Sci 2002 ; 69 : 131 – 138 .

119. Drlica K , Zhao X . DNA gyrase, topoisomerase IV and the ;4 - quinolones . Microbiol Mol Biol Rev 1997 ; 61 : 377 – 392 .

120. Ball P , Mandell L , Niki Y , Tillotson G . Comparative tolerability of the newer fl uoroqui-nolone antibacterials . Drug Safety 1999 ; 21 : 407 – 421 .

121. Bertino J , Fish D . The safety profi le of the fl uoroquinolones . Clin Ther 2000 ; 22 : 798 – 817 .

122. Kier LD , Neft R , Tang L , Suizu R , Cook T , Onsurez K , et al. Applications of microarrays with toxicologically - relevant genes (tox genes) for the evaluation of chemical toxicants in Sprague Dawley rats in vivo and human hepatocytes in vitro . Mutat Res 2004 ; 549 : 101 – 113 .