The Principles and Practice of Toxicogenomics: Applications and Opportunities

William D. Pennie*,1, Jonathan D. Tugwood{dagger}, Gerry J. A. Oliver{dagger} and Ian Kimber*

* AstraZeneca, Central Toxicology Laboratory and {dagger} Pharmaceuticals: Safety Assessment, Alderley Park, Macclesfield, Cheshire SK10 4TJ, United Kingdom

Received September 8, 1999; accepted December 1, 1999


    INTRODUCTION
 TOP
 INTRODUCTION
 REFERENCES
 
The last decade has seen major developments in large-scale genome sequencing and in the development of technical platforms to support this. Several genomes, such as yeast (DeRisi et al., 1997Go), have now been completely sequenced and the sequencing of the entire human genome is anticipated to be completed early in this decade, as a result of major effort in both the academic and industrial sectors. The availability of both the sequence information for many thousands of genes and, in many cases, physical clones of the coding regions of these genes, has allowed the construction of gene microarrays that enable quantitative measurement of the transcriptional activity of potentially tens of thousands of genes in biological samples (Schena et al., 1995Go). Microarray technology promises to revolutionize investigative biology (Khan et al., 1999Go), including drug discovery (Marton et al., 1998Go). Application of genomics to toxicology, toxicogenomics, may also yield a number of substantial dividends, including assisting predevelopment toxicology by facilitating more rapid screens for compound toxicity; allowing compound selection decisions to be based on safety as well as efficacy; the provision of new research leads; a more detailed appreciation of molecular mechanisms of toxicity; and an enhanced ability to extrapolate accurately between experimental animals and humans in the context of risk assessment. In this article, we provide a brief overview of microarray technology and its applications to mechanistic and predictive toxicology research, outlining both the strengths and limitations of this approach.

Microarray Technology Platforms
DNA chips, or microarrays, allow quantitative comparisons of the expression levels of potentially thousands of individual genes between different biological samples. This facilitates, for instance, comparisons of normal tissue with diseased, and of control with toxicant-treated cell lines. Construction of arrays involves the immobilization of DNA sequences (either cDNA sequences or oligonucleotides), corresponding to the coding sequence of genes of interest, on a solid support such as a glass slide or nylon membrane (Bowtell, 1999Go). The mRNA prepared from cells or tissues can be labeled and hybridized, usually in the form of a reverse transcribed cDNA copy, to a microarray and visualized using phosphorimager scanning or other appropriate methodologies. Subsequent analysis using appropriate software allows determination of the extent of hybridization of the labeled probes to the corresponding arrayed cDNA spots, and a comparison of control with test samples permits quantitative assessment of changes in gene expression associated with treatment (Brown and Botstein, 1999Go).

Many commercial microarrays are available, varying from those comprising several hundred genes (usually immobilized on a nylon membrane and probed with radiolabeled cDNA) to those harboring tens of thousands of oligonucleotide sequences (immobilized on glass slides and analyzed using dual fluorescent-probe technology). These arrays can be either "broad spectrum" or custom designed to profile particular tissues, biological pathways, or even disease states. In this regard, microarrays designed to profile genes involved in response to toxic insult have been developed by commercial vendors, pharmaceutical companies, and academic institutions (see, for example, Nuwaysir et al., 1999).

Possibilities and Caveats
A major part of the developmental cost of every successful new pharmaceutical or agrochemical product is the recovery costs of compounds that have failed in development, due to potential or observed toxicity. The application of toxicogenomics to mechanistic and predictive toxicology will enable the identification of more effective markers with the potential for adverse health effects, thus highlighting possible "show-stopping" toxicity earlier in a compound's development, and potentially enhancing the predictive power of in vitro model systems (Davila et al., 1998Go). This may prove particularly valuable if toxicogenomics can improve our understanding of processes such as nongenotoxic carcinogenesis or reproductive toxicity, which normally require both long-term animal assays and significant amounts of chemical to be detected accurately.

Clearly, the possibility of toxicogenomics giving an early alert to potential adverse effects at low levels of exposure, while powerful, raises many issues in the context of interpreting gene expression changes with respect to hazard and risk assessment. It is reasonable to predict that in the first instance, accurate interpretation of expression changes will only be possible when toxicogenomics analyses are conducted as part of a larger experimental design to understand observed toxicity at the physiological, histological, and/or biochemical levels. Also, there is a vast amount of information already available on in vivo toxic responses to compounds, particularly in rodents. Strategically designed toxicogenomics experiments will allow the exploitation and annotation of some of these data.

As microarrays only profile gene expression at the mRNA level, this technology alone cannot identify corresponding changes in the level of functional protein. Additionally, protein modifications such as phosphorylation, which may be critical for the function of many proteins (and indeed entire biochemical pathways such as signal transduction), cannot be detected directly by microarray analysis, although indirect detection is likely where the protein modifications result in downstream changes in gene expression. The complementary technology of proteomics may help resolve issues around such differential protein modification, leading to changes in the activity of gene products. Briefly, proteomics allows the separation of the total protein complement of cell extracts, consisting usually of thousands of different proteins, by 2D polyacrylamide gel electrophoresis. Stained gels are scanned and analyzed with bioinformatics software, in order to compare protein extracts of cells in different physiological states. Compounds with particular toxicological effects will induce changes in the proteome, producing a signature proteomic expression profile (for example, see Anderson et al., 1996). The throughput of proteomics is currently much lower than that of genomic microarrays, largely due to the requirement for mass spectroscopy analysis (or similar technologies) of differentially expressed proteins to determine identity. Recent reports have suggested, however, that microarray-based approaches to protein detection may overcome this limitation (Lueking et al., 1999Go). Protein function can also be affected by mutation or polymorphism of the encoding gene (without influencing transcript levels), and such events cannot be resolved directly by transcript profiling experiments. Methodologies such as pharmacogenetics are being developed to define polymorphic responses. Nevertheless, it is reasonable to assume that mechanistic events will give rise to diagnostic changes in the transcription of other genes that can be detected by microarrays. However, careful and informed interpretation of microarray results will obviously be required to "backtrack" to the important initiating events, under such circumstances.

Toxicogenomics Applications
In general terms, the applications of toxicogenomics can be characterized into two broad and overlapping classes: mechanistic or investigative research and predictive toxicology.

The biological relevance of the experimental system for transcript profiling is clearly of major importance where a mechanistic understanding of a toxic process or a mode of action is required. In all likelihood, the toxic endpoint is known in advance (at a physiological, histological, and/or biochemical level) and an appropriate test system (in vitro or in vivo) can be designed to model the endpoint as closely as possible. An example of such an endpoint, non-genotoxic carcinogenesis, is usually evaluated in the context of long-term cancer bioassays in rodents (Chhabra et al., 1990Go). Toxicogenomic applications may help to identify surrogate markers for the development of this phenotype, and indeed the exposure of rodent hepatocytes to the nongenotoxic carcinogen phenobarbital has been studied, using both microarray and gel-based expression technologies. In excess of 300 genes have been identified where expression is modulated by this compound (Rodi et al., 1999Go). Many other toxic endpoints could be profiled using these methods, with combinatorial approaches such as transgenic or knockout models (Ryffel, 1997Go) potentially providing insights into the role of specific genes.

The possibility that a specific group or class of compounds (grouped by toxic endpoint, mechanism, structure, target organ etc.) may induce signature patterns of gene expression changes is the basis for the application of toxicogenomics to predictive toxicology. The use of these technologies to analyze genome-wide changes in mRNA expression following treatment of in vitro systems with known reference toxicants may permit the identification of diagnostic gene expression patterns. Pattern recognition may, in turn, allow the design and construction of miniarrays, customized to detect specific toxicity endpoints or pathways. The throughput requirements of this approach will almost certainly necessitate employing in vitro culture systems. While in vitro systems have practical advantages, there are major drawbacks to consider. Even where appropriate cells in vitro, such as primary hepatocytes, are available, compound-induced changes in transcription may not necessarily reflect accurately the response of the corresponding organ in vivo. In addition, availability of appropriate cell lines may be limited (although where mechanistic information is not sought, generic cell lines may still be of value), and in some cases, metabolism may be required to produce the active chemical reactive species, although often this can be accomplished by pre-incubation with metabolically active cell extracts. Importantly, where the toxicity is specific to species, strain, sex, or route of administration, in vitro modeling is unlikely to be fully diagnostic. Nevertheless, even if these systems were able to detect potential adverse health effects of only a small subset of development compounds, their application in predevelopment toxicology screening would be of substantial benefit in providing an early view of compound safety in advance of traditional studies.

Development of reference data sets to allow a "pattern recognition" approach to toxicology is likely to require the application of complex computer algorithms and statistical approaches. For example, statistical clustering techniques have been applied to microarray data to analyze the temporal patterns of gene expression that characterize serum-responsiveness and wound repair (Iyer et al., 1999Go), and to distinguish cancerous tissue from normal tissues and cell lines (Alon et al., 1999Go). The building of reference data sets, possibly by comparison of microarray output across different laboratories, will require consistency in data analysis and format. A number of resources exist in both the academic and commercial sectors for such purposes (Bassett et al., 1999Go). One example, the software platform ArrayDB, has been developed at The National Human Genome Research Institute. The system facilitates the storage, retrieval, and analysis of microarray data along with information linking some 15,000 genes to public domain sequence and pathway databases (Ermolaeva et al., 1998Go).

Custom Toxicology Microarray Construction: ToxBlot
In addition to the use of high-density "discovery" arrays from commercial vendors, our laboratories have a program of in-house custom toxicology array design and manufacture: ToxBlot. A custom approach allows the selection of the genes to be represented on the array, permitting arrays to be focused on areas of particular interest to mechanistic or investigative toxicology research programs. (See Table 1Go). While the major costs of custom array manufacture are capital investment in appropriate technologies, once cDNA sets have been amplified, purified, and verified, the unit array cost is relatively modest, being in the order of $20 per array. For the construction of the ToxBlot arrays, approximately 2000 sequences of human or murine origin were identified as being of potential relevance to various forms of toxicity or normal cell regulatory or signaling pathways. In-house cDNA clone sets, of both public domain and proprietary origin, were used as the source material for array construction. Following selection, PCR amplification, purification, and (where appropriate) sequence verification, the representative cDNA sequences were immobilized on nylon membranes. Both a human and a mouse ToxBlot array have been constructed, in each case comprising approximately 2400 cDNA sequences, spanning about 600 genes of the relevant species. Reproducibility is ensured by each gene being represented by 4 individual spots on each array and the inclusion of two non-overlapping cDNAs for each gene (in duplicate) wherever possible. To aid in the interpretation of differential gene-expression results, we have assembled a database on gene function, distribution, and known allelic variations for each represented gene. In this regard, public-domain databases such as the GeneCards system developed at the Weizmann Institute (Rebhan et al., 1997Go), or the Kyoto Encyclopedia of Genes and Genomes (KEGG) databases (http://www.genome.ad.jp/kegg/) can provide valuable supplementary information.


View this table:
[in this window]
[in a new window]
 
TABLE 1 Broad Gene Classes Included on the ToxBlot Microarrays
 
The ToxBlot series of arrays are now in wide use in our laboratories and some examples of current applications are outlined below:

Endocrine Disruption
Concern that some environmental compounds (of natural and man-made origin) may affect estrogen receptor function, with consequences for normal endocrine function, has been a topic of debate among both the general public and the scientific community. Ligand-mediated estrogen receptor (ER) activation, followed by activation (or repression) of estrogen receptor-responsive genes has a complex mode of action which is not captured by many of the current in vitro screens for estrogen action. Recent research from our laboratory and others has indicated that the precise DNA sequence of the estrogen-response elements within the promoter of estrogen responsive genes may have a great influence on ER action (Pennie et al., 1998Go, Vanacker et al., 1999Go). Hence assays which measure the activation of a single gene product as a marker for estrogenicity may be misleading. Toxicogenomics analysis of ER action should help resolve pathways of gene regulation involved in estrogen action and lead to the development of more sophisticated in vitro assays for ER activation. In this regard, the combination of suppression subtractive hybridization (Diatchenko et al., 1999Go) and cDNA construction has been employed to characterize the differences between ER-positive and -negative cell lines (Yang et al., 1999Go). Our own laboratory is employing ToxBlot microarrays to characterize gene expression changes that take place in cultured cells exposed to natural estrogens (e.g., estradiol), synthetic (e.g., diethyl stilbestrol), and phytoestrogens (e.g., genestein). By treatment of appropriate cell lines (e.g., the ER + ve breast carcinoma cell lines T47D and MCF7) with estrogens at multiple doses and at a variety of time points, a pattern of consistent gene expression changes has begun to emerge that may permit evaluation of endocrine disruption at the molecular level. In addition, the use of cell lines of uterine origin, such as the human line SKUT (Bamberger et al., 1997Go), may aid the understanding of gene expression changes in particular target cells and possibly assist prediction of endocrine disruption, thus reducing the number of animals used to predict this endpoint by traditional methods such as the uterotrophic assay (Odum et al., 1997Go). Transcript profiling in these systems clearly demonstrates the need to profile gene expression changes at multiple time and dose points, as individual responsive genes can vary considerably in the kinetics of their regulation.

Hepatocyte Toxicity
Many liver cell models retain tissue-specific characteristics in culture and have proved useful in, for instance, cytotoxicity and genotoxicity assays, drug-drug interactions, and mode of action studies (Guillouzo, 1998Go). The characterization of cytotoxicity in cultured hepatocytes provides an ideal model for profiling induced changes in gene expression. A number of hepatoxicants has been characterized for cytotoxicity in cultured human hepatocyte cell lines (such as HepG2), including ethanol (Neuman et al., 1993Go), paracetamol (Nicod et al., 1997Go), hydrogen peroxide (Yang, C.-F. et al., 1999Go) and carbon tetrachloride (Dai and Cederbaum, 1995Go). At appropriate doses, these compounds all cause liver necrosis in vivo (and cytotoxicity in HepG2 cells), induce periportal hepatocyte proliferation, cause elevation of enzyme levels (e.g., cypP4502E1) and modulate oxidative stress and lipid peroxidation. ToxBlot microarrays have been employed to characterize the gene expression changes induced in HepG2 cells when they are exposed to varying concentrations of these 4 cytotoxicants. By calculating cytotoxicity over a range of concentrations (based on release of cellular LDH), we were able to select appropriate high and low doses for each compound (set at 40% and 25% LDH release, respectively). Transcript profiling at these selected doses for all 4 compounds (at a fixed time point) revealed genes that were up- or down-regulated consistently by different compounds. These investigations permit the identification of patterns of gene changes, which may prove to be diagnostic for hepatotoxicity.

Bone Marrow Toxicity of Insulin-Sensitizing Compounds
As discussed above, one of the most important applications of toxicogenomics is to provide information on safety early in the development process, to assist in compound selection. We have used microarrays for transcript profiling in parallel with a program to develop novel insulin-sensitizing agents as therapies for non-insulin-dependent diabetes mellitus (NIDDM). Such a program provides a useful "proof of concept" opportunity to use transcript profiling to obtain information for safety evaluation, for several reasons. First, the existing therapeutic agents for NIDDM, the thiazolidinediones (TZDs), have undesirable toxic properties that limit their clinical application. In particular, TZD administration to both rats and dogs results in bone marrow fat deposition (adipogenesis), impaired hematopoiesis, and anemia (Deldar et al., 1993Go; Williams et al., 1993Go). Thus, there is an incentive to develop alternative therapeutic agents with a more favorable toxic profile. Second, the TZDs are known to activate a nuclear receptor, the peroxisome proliferator-activated receptor gamma (PPAR{gamma}; Lambe and Tugwood, 1996), which represents a potential mechanism for both the toxic and therapeutic effects of these compounds, and assists the design and interpretation of array experiments. Third, a reliable in vitro system is available for evaluating adipogenic potential of compounds, using human bone-marrow stromal-cell culture (Gimble et al., 1996Go).

This in vitro model has been used to investigate adipogenesis caused by a number of insulin sensitizing compounds, including TZDs. Differentiation of the stromal cells along an adipogenic pathway develops over 7–10 days and fat deposition can be detected at this stage. The goal was to characterize transcriptional changes that precede this, and whether such changes are specific for TZDs. To this end, a custom ToxBlot array was designed that comprised genes associated with hematopoiesis, osteogenesis and adipogenesis in the bone marrow, together with genes known to be regulated by PPAR{gamma}. The final array consisted of approximately 2400 individual "spots", representing 802 discrete genes, and this is being employed currently to examine changes in gene transcription of TZD and non-TZD compounds at apidogenic concentrations.

Future Issues and Opportunities: A Perspective
The examples cited above serve to illustrate the ways in which the application of toxicogenomics can contribute significantly to our understanding of toxic processes. The potential utility of this new technology has necessarily stimulated a lively debate about how best genomics can be applied to the toxicological sciences and what, in practice, may be achieved. The debate is vigorous and essential and one that is continued here in considering several of the issues posed by the current and future application of toxicogenomics.

The first of these is that of "certainty": a question posed commonly as to what extent we can be certain that the changes in gene expression detected by microarray technology are real and, if real, whether the altered gene expression necessarily translates into protein production. There is of course no certainty, as such, on either count. In the context of identifying new research leads, or changes that may provide clues to a particular toxic mechanism, it is probably best to regard the results of microarray analyses as the springboard to more detailed and more focused investigations (using other experimental approaches) that would confirm (or otherwise) the robustness of the changes observed. Furthermore, it is certainly not the case that induced changes in gene expression, associated with no altered protein production, are of no interest or are without relevance. This is particularly not so if these gene expression changes are proven to represent reliable, sensitive, and selective markers of a toxic process or the actions of a particular class of toxicant. This introduces a broader issue, of course, relating to the certainty of biological relevance. There is a comfort in understanding the molecular basis for any normal or adverse biological process, but if we are to make full use of toxicogenomics, and the results of microarray technologies, it is likely that we will have to accept greater intellectual discomfort and learn to embrace data that display a much more holistic view of biological changes, which cannot necessarily be reconciled into known and discrete mechanistic pathways.

Although to make full use of toxicogenomics, at least for some applications, an acceptance of a frameshift in data analysis will be necessary, one thing that remains unchanged is the absolute requirement for appropriate experimental design. The principles of good experimental design and the need for appropriate controls and consideration of single variables remain critical. The results of toxicogenomic analyses will be only as valuable as the rigor of the experiment from which they derive. The old adage—the best research comes from asking Nature simple questions, one at a time—is no less true for microarray analyses. The answers that toxicogenomics can provide may be considerably more complex, but the questions we pose should be no less simple and specific than those asked using other experimental approaches. In this context, it is important that genomic and proteomic technologies are not blighted by precipitate or inappropriate interpretation of poor quality data from badly designed investigations.

Another important aspect of experimental biology is the need to understand and acknowledge the limitations of any approach. There has been, in the toxicological sciences, and particularly in predictive toxicology, a desire for validation, which in this context can be defined as a level of confidence that a particular method or test has the sensitivity and selectivity to permit conclusions to be drawn about the likely properties of a chemical. While these are important attributes of any test method, of at least equivalent and probably of greater importance is an appreciation of the constraints and limitations of the approach. Ultimately, test methods are the tools we employ as toxicologists to aid us in making decisions about the likely properties of chemicals in biological matrices and the potential for adverse health effects. It is a great arrogance to believe that we are able to design any experimental approach that will reflect accurately all the myriad effects xenobiotics may induce. No single test method will ever be absolute or comprehensive for all toxic endpoints, anymore than one single drug is ever likely to treat all disease. It is for this reason that the experience and expertise of toxicologists and their ability to accurately interpret the results of tests remain the final arbiter. The same is true of toxicogenomics. It is a tool that provides data that require interpretation—not an end in itself.

It is important that microarrays are tailored to the needs of the experimental system. At one extreme, where the intention is to identify novel genes of relevance to a toxic process, the use of an extensive array incorporating as many sequences, including those of unknown function, will be the favored approach. In contrast, for more discrete objectives such as, for instance, interrogating the potential of chemicals to induce a particular toxic response, it may be more appropriate to use an array comprising a smaller number of carefully selected genes of known function. There is a middle ground of course. As described above, the ToxBlot incorporates 600 genes of diverse function, and it is being used to explore different biological and toxicological responses. This serves to combine the benefits of functional genomics, while retaining the opportunity, to some extent, to explore unexpected changes in gene expression.

In this regard, it may be argued that potentially the most valuable aspect of genomics is that it will encourage, and indeed require, toxicologists to consider biological responses in a more holistic fashion. It is inevitable that when there exists opportunity only to evaluate changes in the expression of very few genes, those genes will be selected carefully, based upon our knowledge, expectations, and/or prejudice. What toxicogenomics offers and demands of us, is that consideration be given also to changes in the expression of genes that were believed to be of no apparent relevance. This will, and has already, yielded some surprises. Perhaps the most exciting advances in defining toxicological mechanisms and identifying new research leads will derive from the inter-disciplinary and more integrated toxicology that genomics will undoubtedly encourage.

Toxicogenomics is not a promise for the future, it is a tool that is available to us now, and which, if used correctly and within the guiding principles of good experimental biology, will bring huge dividends. Concern has been voiced already that a potential problem is the misinterpretation, or over-interpretation, of genomic analyses, particularly in the context of determining product safety. It must be recognized that the interaction of xenobiotics with biological systems will in many instances result in some changes in gene expression, even under circumstances where such interactions are benign with respect to adverse effects. The challenge again is to ensure that sound judgment and the appropriate toxicological skills and experience are brought to bear on the data generated, so that toxicologically relevant changes in gene expression are distinguished from those that are of no concern.


    NOTES
 
1 To whom correspondence should be addressed. Fax: 01625 586396. E-mail: bill.pennie{at}ctl.astrazeneca.com. Back


    REFERENCES
 TOP
 INTRODUCTION
 REFERENCES
 
Alon, U., Barkai, N., Notterman, D. A., Gish, K., Ybarra, S., Mack, D., and Levine, A. J. (1999). Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc. Natl. Acad. Sci. U S A 96, 6745–6750.[Abstract/Free Full Text]

Anderson, N. L., Taylor, J., Hofmann, J. P., Esquer-Blasco, R., Swift, S., and Anderson, N. G. (1996). Simultaneous measurement of hundreds of liver proteins: Application in assessment of liver function. Toxicol. Pathol. 24, 72–76.[ISI][Medline]

Bamberger, A. M., Jenatschke, S., Erdmann, I., and Schulte, H. M. (1997). Progestin-dependent stimulation of the human leukemia inhibitory factor promoter in SKUT-1B uterine tumor cells. J. Reprod. Immunol. 33, 189–201.[ISI][Medline]

Bassett, D. E., Jr., Eisen, M. B., and Boguski, M. S. (1999). Gene expression informatics—it's all in your mine. Nat. Genet. 21(Suppl. 1), 51–55.

Bowtell, D. D. (1999). Options available—from start to finish—for obtaining expression data by microarray. Nat. Genet. 21(Suppl. 1), 25–32.

Brown, P. O., and Botstein, D. (1999). Exploring the new world of the genome with DNA microarrays. Nat. Genet. 21(Suppl. 1), 33–37.

Chhabra, R. S., Huff, J. E., Schwetz, B. S., and Selkirk, J. (1990). An overview of prechronic and chronic toxicity/carcinogenicity experimental study designs and criteria used by the National Toxicology Program. Environ. Health Perspect. 86, 313–321.[ISI][Medline]

Dai, Y., and Cederbaum, A. I. (1995). Inactivation and degradation of human cytochrome P4502E1 by CCl4 in a transfected HepG2 cell line. J. Pharmacol. Exp. Ther. 275, 1614–1622.[Abstract]

Davila, J. C., Rodriguez, R. J., Melchert, R. B., and Acosta, D., Jr. (1998). Predictive value of in vitro model systems in toxicology. Annu. Rev. Pharmacol. Toxicol. 38, 63–96.[ISI][Medline]

Deldar, A., Williams, G., and Stevens, C. (1993). Pathogenesis of thiazolidinedione-induced hematotoxicity in the dog. Diabetes 42, 179.

DeRisi, J. L., Iyer, V. R., and Brown, P. O. (1997). Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 278, 680–686.[Abstract/Free Full Text]

Diatchenko, L., Lukyanov, S., Lau, Y. F., and Siebert, P. D. (1999). Suppression subtractive hybridization: A versatile method for identifying differentially expressed genes. Methods Enzymol. 303, 349–380.[ISI][Medline]

Ermolaeva, O., Rastogi, M., Pruitt, K. D., Schuler, G. D., Bittner, M..L., Chen, Y., Simon, R., Meltzer, P., Trent, J. M., and Boguski, M. S. (1998). Data management and analysis for gene expression arrays. Nat. Genet. 20, 19–23.[ISI][Medline]

Gimble, J. M., Robinson, C. E., Wu, X., Kelly, K. A., Rodriguez, B. R., Kliewer, S. A., Lehmann, J. M., and Morris, D. C. (1996). Peroxisome proliferator-activated receptor-{gamma} activation by thiazolidinediones induces adipogenesis in bone marrow stromal cells. Mol. Pharmacol. 50, 1087–1094.[Abstract]

Guillouzo, A. (1998). Liver-cell models in in vitro toxicology. Environ. Health Perspect. 106(Suppl. 2), 511–532.

Iyer, V. R., Eisen, M. B., Ross, D.T., Schuler, G., Moore, T., Lee, J. C. F., Trent, J. M., Staudt, L. M., Hudson, J., Jr., Boguski, M. S., Lashkari, D., Shalon, D., Botstein, D., and Brown, P. O. (1999). The transcriptional program in the response of human fibroblasts to serum. Science 283, 83–87.[Abstract/Free Full Text]

Khan, J, Bittner, M. L., Chen, Y., Meltzer, P. S., and Trent, J. M. (1999). DNA microarray technology: The anticipated impact on the study of human disease. Biochim. Biophys. Acta. 1423, M17–28.[ISI][Medline]

Lambe, K. G., and Tugwood, J. D. (1996). A human-peroxisome proliferator-activated receptor-{gamma} is activated by inducers of adipogenesis, including thiazolidinedione drugs. Eur. J. Biochem. 239, 1–7.[Abstract]

Lueking, A., Horn, M., Eickhoff, H., Bussow, K., Lehrach, H., and Walter, G. (1999). Protein microarrays for gene expression and antibody screening. Anal. Biochem. 270, 103–111.[ISI][Medline]

Marton, M. J., DeRisi, J. L., Bennett, H. A., Iyer, V. R., Meyer, M. R., Roberts, C. J., Stoughton, R., Burchard, J., Slade, D., Dai, H., Bassett, D. E., Jr., Hartwell, L. H., Brown, P. O., and Friend, S. H. (1998). Drug target validation and identification of secondary drug target effects using DNA microarrays. Nat. Med. 4, 1293–1301.[ISI][Medline]

Neuman, M. G., Koren, G., and Tiribelli, C. (1993). In vitro assessment of the ethanol-induced hepatotoxicity on HepG2 cell line. Biochem. Biophys. Res. Commun. 197, 932–941.[ISI][Medline]

Nicod, L., Viollon, C., Regnier, A., Jacqueson, A., and Richert, L. (1997). Rifampicin and isoniazid increase acetaminophen and isoniazid cytotoxicity in human HepG2 hepatoma cells. Hum. Exp. Toxicol. 16, 28–34.[ISI][Medline]

Nuwaysir, E. F., Bittner, M., Trent, J., Barrett, J. C., and Afshari, C. A. (1999). Microarrays and toxicology: The advent of toxicogenomics. Mol. Carcinog. 24, 153–159.[ISI][Medline]

Odum, J. Lefevre, P. A., Tittensor, S., Paton, D., Routledge, E. J., Beresford, N. A., Sumpter, J. P., and Ashby, J. (1997). The rodent uterotrophic assay: Critical protocol features, studies with nonyl phenols, and comparison with a yeast estrogenicity assay. Regul. Toxicol. Pharmacol. 25, 176–188.[ISI][Medline]

Pennie, W. D., Aldridge, T. C., and Brooks, A. N. (1998). Differential activation by xenoestrogens of ER alpha and ER beta when linked to different response elements. J. Endocrinol. 158, R11–14.[Abstract]

Rebhan, M., Chalifa-Caspi, V., Priluski, J., and Lancet, D. (1997). GeneCards: Encyclopedia for Genes, Proteins, and Diseases. Weizmann Institute of Science, Bioinformatics Unit and Genome Center (Rehovot, Israel), World Wide Web URL: http://bioinfo.weizmann.ac.il/cards.

Rodi, C. P., Bunch, R. T., Curtiss, S. W., Kier, L. D., Cabonce, M. A., Davila, J. C., Mitchell, M. D., Alden, C. L., and Morris, D. L. (1999). Revolution through genomics in investigative and discovery toxicology. Toxicol. Pathol. 27, 107–110.[ISI][Medline]

Ryffel, B. (1997). Impact of knockout mice in toxicology. Crit. Rev. Toxicol. 27, 135–154.[ISI][Medline]

Schena, M., Shalon, D., Davis, R. W., and Brown, P. O. (1995). Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470.[Abstract]

Vanacker, J. M., Pettersson, K., Gustafsson, J. A., and Laudet, V. (1999). Transcriptional targets shared by estrogen receptor-related receptors (ERRs) and estrogen receptor (ER) alpha, but not by ERbeta. EMBO J. 18, 4270–4279.[Abstract/Free Full Text]

Williams, G. D., Deldar, A., Jordan, W. H., Gries, C., Long, G. G., and Dimarchi, R. D. (1993). Subchronic toxicity of the thiazolidinedione Tanabe-174 (LY282449) in the rat and dog. Diabetes 42, 186.

Yang, C. F., Shen, H. M., and Ong, C. N. (1999). Protective effect of ebselen against hydrogen peroxide-induced cytotoxicity and DNA damage in HepG2 cells. Biochem. Pharmacol. 57, 273–279.[ISI][Medline]

Yang, G. P., Ross, D. T., Kuang, W. W., Brown, P. O., and Weigel, R. J. (1999). Combining SSH and cDNA microarrays for rapid identification of differentially expressed genes. Nucleic Acids Res. 27, 1517–1523.[Abstract/Free Full Text]