Toxicogenomics is a powerful new tool to study the adverse effects of drugs and can provide valuable insight into the mechanism of ADRs. However, analysis of toxicogenomic data can be challenging. Gene expression is a sensitive measure of the overall status of a cell or tissue as reflected by the genome. A number of experimental factors, in addition to drug treatment, may contribute to the overall gene-expression changes in the microarray data. A flow chart of a toxicogenomic study is displayed in . Although not all of the factors in the flow chart will be involved in a single toxicogenomic study, many of them may greatly improve the efficiency of the study in some cases. For instance, one of the difficulties in the interpretation of toxicogenomic data generated from drug exposure is the differentiation of the intended drug pharmacology from off-target adverse effects. Although phenotypic anchoring of transcriptional data is a key in linking cause to effect in order to generate biologically meaningful hypotheses of toxic mechanism, gene regulation caused by therapeutic effects may also change in the same fashion as that caused by off-target toxic effects. In human patients it is even more complicated considering the differences in the health status of the subjects and possible drug interactions. However, in animal models, it has been demonstrated that the use of a reference drug from the same pharmacological class that is devoid of toxicity is very helpful in terms of differentiating gene-expression changes caused by pharmacological effects from those caused by toxic effects [60
]. In addition, timing is a critical factor in the experimental design. It is important to capture the transcriptional changes in early exposure preceding the occurrence of toxicity, but after any nonspecific general acute stress responses, because these changes carry important mechanistic information regarding the mode of action of the drug, while transcriptional changes at the occurrence of toxicity can be more related to an adaptive response to the toxicity or to secondary effects of the toxicity.
Currently, there are numerous techniques developed for microarray data analysis, the description of which is beyond the scope of this review. However, statistical modeling and pattern recognition (such as principle component analysis, clustering and support vector machines) are among the most basic approaches. The application of these analyses provides a high-level view of the overall transcriptomic data by reducing the dimensionality and complexity of the data, and may also generate biologically relevant expression signatures. In addition, although pathway analysis, assisted by a variety of bioinformatics software tools and databases, has been widely adopted in gene-expression data analysis, literature searching is nevertheless important in connecting different bodies of knowledge to discover new genes and pathways involved in a particular adverse response and gain a better interpretation of the toxicogenomic data. Last but not least, transcriptomic data has its own limitations and it can not reflect post-transcriptional changes. Therefore, hypotheses generated based on toxicogenomic studies need to be validated by additional scientific approaches and the extrapolation of a mechanistic hypothesis to human can be assisted by comparative genomic approaches, as well as comparative systems biology approach.
One of the most important issues in the management of ADRs is to develop noninvasive biomarkers that can be used in clinic for the early detection of drug-induced adverse effects. Blood, and specifically the blood transcriptome, is an excellent resource for such noninvasive early-detection makers. The use of human peripheral blood transcriptional profiling to assess the status of diseases, including cancer and cardiovascular diseases, and environmental exposure risks has been described by several reviews [50
]. It has also been reported recently that a gene-expression signature in the peripheral blood mononuclear cells, which consists of 29 genes, can distinguish individuals with early-stage non-small-cell lung cancer from individuals with nonmalignant lung disease with approximately 80% accuracy [63
]. Although there have not been many successful clinical applications of peripheral blood transcriptome analysis in the detection of drug-induced toxicity, the study of APAP toxicity in both rats and humans has demonstrated that the peripheral blood transcriptome certainly holds the potential to be used as a biomarker for the early detection of ADRs [57
In the last decade, there have been collaborative efforts among academia, government and industry in generating toxicogenomic data on a variety of industry chemicals and pharmaceutical compounds. The purpose was to better understand the mode of action of these compounds, generate detection biomarkers, and also improve the data analysis strategy. Several databases have been generated as an outcome of these efforts and provide a wonderful resource for queries about genome response to a variety of compounds, including some pharmaceuticals. For example, the Chemical Effects in Biological Systems database at NIEHS hosts rodent toxicogenomic data, which was generated by in-house studies at the NIEHS, the National Toxicology Program and a number of consortia, including the HESI toxicogenomics projects and the Toxicogenomics Research Consortium, as well as research laboratories from various universities and companies [64
]. Hepatotoxicity and nephrotoxicity data have also been generated by several additional consortia, including the FDA and GB Medicine Liver Toxicity Biomarker Study, the C-Path: Predictive Safety Testing Consortium, and projects carried out in Europe and Japan. Information regarding these consortia and associated databases have been reviewed in two papers [65
]. In addition to these specific databases, Gene Expression Omnibus is another important public database for repository of array-and sequence-based data, including toxicogenomic data. The usefulness of these databases is still underappreciated, owing to limitations, such as variation in array platforms, and lack of integrated high-content data and analysis tools, although some of the databases accept other experimental data in addition to transcriptomic data. For example, Chemical Effects in Biological Systems contains traditional toxicology data and a small number of proteomic studies. Currently, many novel genes and pathways that were discovered in transcriptomic studies, which may play important roles in the mechanism of specific toxicities, need further study in order to draw solid conclusions. Full appreciation of the toxicogenomic data needs collaborative efforts from the field of toxicology. Hopefully, in the foreseeable future, integrated `omics data analyses will be available from these datasets within these databases, which will significantly improve data interpretation and serve as a better resource for the toxicology community.
Toxicogenomics must be placed in the context of traditional scientific approaches to investigate the mechanisms of adverse effects from drugs. No technology is able to stand alone and great progress can be made as information is integrated from disparate data sources. One of the clear strengths of toxicogenomics is to reveal novel roles for biological pathways and processes that had not previously been considered to be involved in a drug response. Thus, mechanistic hypotheses can be generated that can then be experimentally tested using the full arsenal of pharmacology, toxicology, molecular biology and genetics. Extrapolation of mechanistic information from model systems into human cell systems and ultimately to human populations can then be fulfilled. As more information is obtained regarding the functional significance of genetic polymorphisms through genome-wide association studies, this information will need to be interpreted with toxicogenomic information regarding critical pathways and processes in ADRs. Then one may begin to be able to achieve the elusive goal of predictive toxicology associated with drugs in development and perhaps even be able to anticipate an `idiosyncratic' adverse response in a specific subpopulation. The challenges are great but the prospects have never been brighter.
Beginnings of toxicogenomics in adverse drug reaction studies
■ Drug-induced gene-expression changes can be causally linked to or downstream of the toxicity that is elicited.
■ Toxicogenomic approaches are very useful, not only in identifying potential predictors for drug-induced toxicity, but also in understanding the underlying mechanisms.
Moving forward with toxicogenomic studies in adverse drug reaction research
■ Toxicogenomic studies of nephrotoxic drugs revealed genes and pathways that are involved in the toxic mechanism and provided novel hypotheses for further studies.
■ Toxicogenomic studies of acetaminophen-induced hepatotoxicity revealed genes and pathways that are involved in the toxic mechanism and provided novel hypotheses for further studies.
■ Although using blood as a surrogate tissue to study the mechanisms of acetaminophen-related hepatotoxicity is challenging, it may provide valuable insight in the future as more is understood regarding organ-to-organ communication.
■ Toxicogenomic studies can help identify environmental risk factors and critical genetic components in idiosyncratic drug reactions.
Conclusion & future perspective
■ Analysis of toxicogenomic data can be challenging and a well-designed experiment is the base of an informative toxicogenomic study.
■ The peripheral blood transcriptome may be used for early detection of drug-induced adverse effects.
■ Toxicogenomic databases with integrated analysis tools and collaborative efforts from the field of toxicology are needed for better use of available toxicogenomic data.
■ The strength of a toxicogenomics approach in investigating the mechanisms of adverse effects will be further enhanced when the toxicogenomic study is integrated with traditional scientific approaches and systems biology.