Novel and future applications of microarrays in toxicological research.

Microarray technologies have both fascinated and frustrated the toxicological community since their introduction around a decade ago. Fascination arose from the possibility offered by the technology to gain a profound insight into the cellular response to chemically mediated stress, and the potential that this genomic signature would be indicative of the biological mechanism by which that stress was induced. Frustrations have arisen primarily from technical factors such as data variance, the requirement for the application of advanced statistical and mathematical analysis, and difficulties associated with actually recognising signature gene expression patterns, and discerning mechanisms. Toxicogenomics was predicted to make toxicological assessment and extrapolation easier, faster and cheaper. The reality has been somewhat different; toxicogenomics is difficult. However, its potential when properly applied has been indicated by some well designed toxicogenomics studies, particularly in the differentiation of genotoxins from non-genotoxins. Technology waits though for no man. While the toxicological community has been working to apply transcriptomics (mRNA levels) in toxicology, the technology has moved beyond this application into new arenas. Some have application to toxicology and are reviewed here, except transcriptomics which has been extensively written about before. This review discusses the application of microarray technologies applied to the genome per se (amplifications, deletions, epigenetic change), mRNA translation and its control mechanisms through miRNA. Which of the new genomics technologies will find most application in toxicology? In the opinion of the author there are three potentially major applications: i) arrayCGH in assessment and recognition of genotoxicity; ii) epigenetic assessment in developmental and transgenerational toxicology; and iii) miRNA assessment in all toxicology types, but particularly developmental toxicology.

Novel and future applications of microarrays in toxicological research 1

. Introduction
Using microarrays to measure mRNA levels for genes across the whole or large parts of the genome has become a familiar technology in recent years. There are still many challenges associated with the interpretation of the data obtained but the actual technology for data generation has matured with oligo-based arrays in various formats becoming the de facto standard. Recent data from the FDA has confirmed that the platforms are becoming robust, and that quantitatively similar data can be obtained across platforms [1] .
The determination of mRNA levels for many genes simultaneously has been the major application of microarray technology [2]. However, to consider that this is their only application in regulatory and investigative toxicology is to take a myopic view of the potential application of this important methodology. The microarray format is as adaptable as its ancestor hybridisation technologies and so useful in all experimental applications involving hybridisation. This article explores some of these applications, their role in making toxicological assessments and position in the drug development pipeline.
As the use of microarrays for the determination of mRNA levels has been very well presented in a large number of articles it is not addressed here. Rather, this article focuses on some of the less explored applications of microarray technology and their application in toxicological assessment. The various applications are presented in overview under separate subheadings with the potential application to toxicological assessment dealt with at the end of each. The potential spectrum of microarray applications in toxicological assessment from the genome to assessment of mRNA translation is shown in Figure 1 . In addition, this article is restricted to nucleic acid-based arrays rather than those made with proteins or designed for high-throughput transfections [3] . This focus though does not imply that non nucleic acid microarrays, such as those employing proteins as targets, do not have application in the understanding of toxicology. Indeed, there are profound opportunities for the use of non-nucleic acid microarrays in toxicological research.
What should be apparent is that although many articles have been written on the application of the microarrays in toxicology we are really still only at the initial stages of realising their true potential [4,5] .

Array comparative genome hybridisation
Outside of the determination of mRNA levels, one of the first alternative applications of microarray technology was in the determination of gene changes (amplification and deletion) in the genome [6] . This method has recently been used to measure global variation in copy number through the human genome [7] . The process of determining chromosomal changes is essentially the same as that of measuring mRNA transcript levels, except that the probe is genomic DNA (gDNA) and not mRNA [8,9] . Hybridisation and data collection occur in a similar manner to that used for expression microarrays with (for a two-colour system) probes from a control and test system hybridised to the same microarray to produce the familiar red and green spot image. The ratio of fluorescent dye in the hybridised spot indicates either an amplification or deletion in the genome. These data can be plotted against the chromosomal location of the probe to produce a map of the chromosome. An example of this can be seen in Figure 2 where hybridisation of gDNA from a rat with a gene mutation giving rise to a Wilson's disease type was hybridised against that from a control Fischer rat. A deletion of one gene (cadherin 11) can be seen. The areas of the genome assayed in array comparative genome hybridisation (arrayCGH) are dependent on the targets on the microarray. Therefore, targets with concensus sequences to gene coding regions will detect chromosomal changes in open reading frames, whereas, for detection of changes in intragenic regions, intronic probes are required. The resolution of the method depends on the number of probes used with the best resolution being provided by tiled probes such as those used in the study of Ekong et al . in Axenfeld-Rieger syndrome [10] .
It is often informative to compare the data obtained from an arrayCGH assay with that of expression level assayed by measuring mRNA levels, which can indicate whether the genes have been amplified whole and if the increased copy number is reflected in the amount of mRNA transcript. An example of this application is shown in Figure 3 for doxorubicin resistance cells where the BAK1 and ABCB4 genes are shown amplified but not overexpressed, MMP1 is overexpressed but not amplified, but ABCB1 is both amplified and overexpressed. Furthermore, ABCB1 and ABCB4 are adjacent on human chromosome 7 and so, although both have probably been amplified as a cassette, only ABCB1 has been amplified in a manner which has led to its overexpression [11] . Many of the assays used for the assessment of genotoxicity are actually looking for the amplification or deletion of large sections of the genome. For example, micronucleus formation essentially represents the amplification of a piece of DNA and indicates there should be a role for the use of arrayCGH in the detection of genetic change after xenobiotic exposure. One example study, although not using microarrays, has shown the amplification of the SV40 sequence in cell lines after exposure to various clastogens and aneugens [12] . A much wider perspective could be gained from the use of arrayCGH.
There is a potential problem in the application of arrayCGH in the detection of genotoxicity and that is the number of cells affected. Established assays, such as fluorescent in situ hybridisation, micronucleus and reversion assays for example, work at a single cell level in that single affected cells are counted to determine the effect of the xenobiotic. Unless the effect of a genotoxic xenobiotic is very substantial, effects at the genome level in a minority of cells could be so diluted by unaffected material as to not be detected by arrayCGH. However, in tumours patterns of genomic alteration have been associated with disease outcomes, such as survival and genomic alteration in primary gastric adenocarcinomas [13] , despite these tumours consisting of a heterogeneous cell population. These data suggest that dilution effects on minority cell populations may not severely disable the potential of the assay. Perhaps most encouragingly though is the recent demonstration of the application of arrayCGH to single cells [14] . If verified then this technology would give arrayCGH application for detection of genotoxicity using single cells. The information gained would be more informative than that achieved from established assays as it would show whether there was a gene or region-specific effect.
Use of the arrayCGH technique has demonstrated the diversity that exists of copy number in the human genome [7] . Such genetic change affects resistance and susceptibility to toxicity, and understanding the effect of genome variation could have a fundamentally important application in the assessment of drug efficacy and safety. These data also indicate that there is an important need to use arrayCGH to characterise cells and animal strains used for testing purposes. This has not been carried out extensively to date, but could yield valuable data that would inform on the resistance and susceptibility of strains and species, and also act as a reference for confirming the genomic stability of cells used for testing purposes in culture.

Epigenetic modifi cation
The recent sequencing of the human epigenome for three chromosomes has brought this application of microarray technology into sharp focus [15] . For toxicology, there is potential for: i) understanding how drugs and chemicals alter DNA methylation patterns in cells and how these might affect toxicity; and ii) elucidating how epigenetic changes affect gene expression and, in turn, susceptibility and resistance to toxicity. Most work in genetic susceptibility has focussed on polymorphic mechanisms, but the small amount of epigenetic work in toxicology published indicates that this is substantial contributory mechanism to the genetically modulated response to xenobiotic exposure [16] . One means of epigenetic modulation is via the differential cytosine methylation. This can be assessed using microarray technology by first breaking up the DNA using either a mechanical or enzymatic means followed by immunoprecipitation of those fragments methylated using an antibody against 5-methylcytosine. A microarray-based comparison of the differences in DNA sequence content between the immunoprecipitated fractions indicates those

. Various combinations of gene amplifi cation and/or expression in MCF7 cells resistant to doxorubicin compared with the wild-type in each case .
Those with acquired resistance have been derived from exposure of a population of naïve cells to increasing concentrations of doxorubicin and have thus acquired resistance. It is hypothesised that those cells with an acquired resistance arise though selection of cells with an intrinsic resistance due to genome alteration. Those cells labelled as intrinsically resistant here are representative of this population of cells having been derived without drug exposure (an antibody selection technique was utilised) [11]. The similarity of their differential gene expression and genome alteration compared with wild-type progenitors supports the hypothesis that cells with an acquired resistance are indeed toxin selected intrinsically selected cells from a majority sensitive population. regions of the chromosome where differential methylation has occurred [17] .
However, there is another potentially important role for epigenetic assessment in toxicology; in the understanding the mechanisms of transgenerational toxicology. What is meant by transgenerational toxicology is genome alteration, which may give rise to a phenotype, and is present in progeny as a result of germline transmission from the exposed parents [18] . This is different from reproductive toxicity where the toxicity arises in the fetus as a result of direct xenobiotic exposure in the uterus. Transgenerational genome instability arising from chemical exposure has been indicated by minisatellite mutation in both individual sperm and somatic cells of first generation mice where the father has been exposed to radiation or genotoxins [19,20] .
Apart from transgenerational effects, modification of epigenetic status in the embryo through maternal exposure may have important long-term effects for the fetus and subsequent generations. One such example is diethylstilbestrol (DES) where effects are then seen in the subsequent male and female lineages after exposure in utero [21] . Effects observed in the male lineage mice were an increased incidence of proliferative lesions of estrogen target tissues and tumours of the reproductive tract. In females an increased incidence of uterine adenocarcinoma was observed. Transmission was only investigated through the female germ line. The mechanism of this effect was not investigated, but previous work from this group has shown an effect of DES on DNA methylation patterns. Thus, the hypothesis that altered methylation may be the mechanism is entirely plausible [22,23] . The demonstration of the inheritance to the F2 and F3 generations in the mouse of vinclozolin-induced DNA methylation changes lends weight to the hypothesis [24] .
Using microarrays containing probes for promoter regions combined with immunoprecipitation of methylated DNA offers profound opportunities for the study of epigenetic modification in terms of assessing contribution to xenobiotic susceptibility and resistance, and for recognition and understanding of mechanisms of transgenerational and reproductive toxicity. The modulation of epigenetic status is certainly a potentially important mechanism of toxicity, not least because of the inherited nature of the effects. The microarray method of analysis for epigenetic effects has a major role to play both in the future assessment of, and understanding of, such toxicities. The technique of single cell polymerase chain reaction amplification of minisatellites as a measure of transmissible genomic alteration [20] combined with a microarray-based assessment of epigenetic change has enormous potential for the assessment of xenobiotic-induced transmissible genomic change.

Chromatin immunoprecipitation analysis
The methodology of chromatin immunoprecipitation (ChIP) analysis is very similar to that used for epigenetic analysis.
A microarray containing gene promoter region target sequences is required, as it is for the assessment of epigenetic analysis. Similarly, an immunoprecipitation step is involved. The difference is that, although the assessment of epigenetic status requires the use of an antibody against 5-methylcytosine, ChIP analysis requires an antibody against the transcription factor of interest. As there are two proteins involved, the antibody and the transcription factor, as well as the DNA fragment, it is necessary to chemically crosslink the transcription factor to the DNA before immunoprecipitation [25] . Gene expression at the transcriptional level is controlled by the binding of transcription factors to the promoter regions of genes, which may then recruit RNA polymerase II to initiate the transcription of the gene, resulting in increased mRNA levels that may be measured using microarray expression technology. An example of a well known toxicant that alters gene expression by causing the translocation of an intracellular receptor and subsequent binding of a transcription factor is 2,3,7,8-tetrachlorodibeno-p -dioxin (TCDD). Other examples would include the orphan receptor agonists, such as phenobarbital and retinoic acid. Data from the ChIP assay is able to show the gene promoter regions binding the transcription factor in one condition relative to another. As an example this assay was used to recognise and assess the binding sites of HNF6 in the liver, and elucidate the functions of GATA factor complexes [26,27] . The limitation of this assay to date has been the availability of microarrays containing suitable promoter fragments. However, there are now several companies supplying such microarrays which are likely to lead to an increased application of this assay method.
An assay such as ChIP is not likely to find application in the recognition of hazard. Rather, this is an assay which may be used to understand mechanisms and by doing so make an informed assessment of risk. For example, if, following a gene expression assessment, a novel chemical entity (NCE) is suspected to have Ah receptor type agonist properties from its gene expression profile, a ChIP assay may be carried out with antibodies against the Ah receptor. The hypothesis would be that the NCE would cause increased immunoprecipitation of promoter regions containing Ah receptor binding sites and that the profile obtained in this assay would also be very similar to that obtained with TCDD. Such an assay would inform that the mechanism of action of the NCE was similar to that of TCDD and, therefore, was likely to constitute a high risk level.

Transcription rate analysis
The nuclear run-on assay has been employed to assess whether a gene is increased in its transcriptional rate in a certain circumstance (e.g., after chemical exposure). For example, the nuclear run-on assay has been used to show that transcription of the ABCB1 gene can be increased after chemical exposure in the liver [28] . It should be possible to undertake such an analysis on a global scale using a microarray. Indeed, a crude version of a microarray (the dot blot) has always been employed for this assay so the microarray would mean just an increase in the number of genes being probed.
The principle of the assay would be that nuclei would be isolated from untreated and exposed cells or tissues. These nuclei would then be allowed to finish making transcripts in an in vitro environment and incorporate a labelled nucleotide into the RNA made. The RNA would then be isolated from the nuclei in the same manner as for a nuclear run-on and hybridised to an expression type microarray consisting of gene coding regions. Where a gene is transcriptionally activated by exposure to the xenobiotic, this will cause an increased amount of transcript for that gene during the run-on in vitro . The net result will be that this differential transcription should be detected by increased hybridisation to the relevant target sequence on the microarray.
This assay would be a compliment to the ChIP type assay and would find use in toxicological assessment for an understanding of mechanisms of toxicity. For example, with TCDD, which leads to the transcriptional activation of genes through the Ah receptor, an assay such as this should show increased transcription from the same battery of genes which are both differentially expressed, and transcriptionally regulated. Application of the ChIP assay should show increased Ah receptor binding to the promoter regions of these genes.

mRNA translation
All of the above microarray based assays give information about the upstream events that can lead to differential mRNA levels. The subject of the determination of gene expression by measurement of mRNA levels is not covered in this article for the reasons explained in the introduction. One of the criticisms often levelled at the assessment of mRNA levels is that measurement has little application in mechanistic toxicological assessment because 'only proteins matter' and there is no evidence that the mRNA is translated into protein in a quantitative manner. Thus, it could be that a gene is increased in its transcriptional regulation resulting in an increased mRNA level but that this does not lead to an increase in its protein level per se . Addressing this question is difficult because of limitations in technologies. Proteomic techniques do not have the dimensionality of genomic techniques due the inherent problems of two-dimensional gel resolution. Even when a protein can be measured, the problems of quantification and of quantification methods, the necessity of measuring actual number of molecules (which is usually not done) and other technical factors make any quantitative comparison of data from genomics and proteomics techniques notoriously difficult, or impossible. Investigation of the role of microRNA (miRNA) in the cell (see Section 6.1) has shown that these species can target mRNA molecules for storage in processing bodies (P-bodies) within the cytoplasm [29] . These can potentially be retrieved and used for rapid translation and, thus, there is a possibility of an increase in protein level without new transcription. This is most likely a fast response mechanism and, in the longer term, new transcription becomes necessary. Data from our laboratory (A Paun pers. commun. in collaboration with R Currie of Syngenta CTL) has indicated that phenobarbital causes an increase in translation in rat liver, which is dose-and time-dependent. Using the microarray mRNA translation assay (see below) it has been shown that many of the genes that are transcriptionally increased in expression are also translated more actively, indicating that both transcription and translation control mechanisms are important in the cellular response to xenobiotics.
Therefore, there appears to be a fundamental requirement to understand the effect of chemicals on mRNA translation. In particular, if chemical exposure can lead to mRNA species being differentially translated as the cells increase their requirement for particular proteins to counter xenobiotic effects.
The mRNA microarray translation assay, which has been developed in the authors' laboratory and elsewhere [30] , seeks to resolve both the questions of ' is the mRNA translated? ' and ' is the mRNA differentially translated? '. The assay relies on separation of mRNA with ribosomes attached from the whole RNA by density, the polysomal fraction. The mRNA without attached ribosomes combined with the ribosomal RNA itself is the monosomal fraction ( Figure 4 ). An assumption is made mRNA found in the polysomal fraction is being actively translated, and there is a linear relationship between the number of ribosomes bound to the mRNA and the amount of protein formed. With this assumption, the basis of the assay is in fact very simple. A cytoplasmic cell extract is prepared in which the ribosomes are stabilised on the mRNA through the use of the translation inhibitor cycloheximide. This is then layered onto a continuous sucrose density gradient and centrifuged. The proteins fail to enter the gradient and remain in a layer at the top. The mRNA species band through the gradient according to their density, and their density is dependent on the number of ribosomes bound. Due to its higher density, the polysome fraction finds density equilibrium at higher sucrose concentrations, and the monosome fraction at the lower sucrose concentrations. These gradients can then be unloaded using a collection apparatus with continuous monitoring at 254 nm, allowing the profile of the separation to be determined and the fractions collected. The RNA can be isolated from the collected fractions and obtained is a series of RNA fractions from the heavy polysomal layers through to the lighter polysomal layers and the monosomes.
Either alone, or by pooling, the mRNA content of each of the fractions can then be assessed for mRNA content and diversity using a microarray containing consensus sequences for gene coding regions. Careful reference of the ratio of RNA in the monosome layers of two systems compared with that in the polysome layers allows the question ' is an mRNA species differentially translated in this system? ' to be addressed [30] . The question of ' is an mRNA translated? ' is answered more simply by comparing the proportion in the polysome fraction with that in the monosome fraction. As the RNA has been fractionated in this method, care needs to be taken both with the amount used for hybridisation and the method employed for normalisation of the data. We use the assumption of majority non-differential translation as the basis for normalisation. To achieve this on a two colour microarray, the monosome fractions of the two samples are hybridised together on one microarray and the polysomes from the same two samples on another.
It is not possible to normalise between polysomal fractions from control and treated tissues because this fails to correct for the amount of mRNA present in the sample for any particular species, which could be altered due to differential transcription. Comparison of the polysomal fractions only would lead to the massive likelihood of identifying a gene as differentially translated between two samples when it is in fact just differentially expressed.
The approach of density RNA fractionation with microarrays has been used in fundamental biology to understand translational changes during spermatogenesis and cancer progression, but remains underused in toxicology [31,32] . A few studies have, however, indicated its potential toxicological application, for example, the translational response to redox stress in yeast and in mammalian cells to ultraviolet (UV) light [33,34] . Shenton et al . showed that in yeast hydrogen peroxide (H 2 O 2 ) at concentrations of 2.0 mM causes a suppression of global translation. Using microarrays in the manner described, the effect of H 2 O 2 on specific genes was determined. At low concentrations (0.2 mM) of H 2 O 2 there was specific upregulation of the translation of transcripts associated with membrane transport and iron regulation, among others, and at 2.0 mM H 2 O 2 transcripts associated with ribosome biogenesis were upregulated, perhaps indicating the importance to the cell of maintaining its translational capability in the face of a substantial toxic insult [34] .
For toxicity profiling with an aim to identify types of potential toxicity in novel agents, it is not required to assay for translation. Whether a gene is translated or not is of no consequence for this analysis as it is only the pattern of gene expression that is required for database matching. However, for an understanding of the mechanisms of toxicity and appropriate risk assessment, first understanding whether there is a change in the protein complement of the treated system and whether this is regulated per se is of primary importance.

microRNA
One of the main control methods of gene translation occurs through the recently recognised miRNA species [35] . These miRNA species are mainly transcribed from the genome by RNA polymerase II, although some transcription from RNA polymerase III may occur [36,37] . The immature transcripts then undergo processing in the nucleus and cytoplasm to give rise to a 21 -23 nt miRNA molecule which, by interaction the RISC complex, can control the translation of mRNA species against which it has a degree of homology [38] .
For toxicology, there are some interesting aspects to these miRNA species. First, they are transcribed from polycistronic regions and the control of the transcription appears to be very similar to that used for protein coding genes [36] . This implies that these genes may well be differentially regulated under the influence of chemicals which could then substantially change the translational profile of the cell. Second, each miRNA species has an effect on the translation of many mRNA species and so a change in its level of expression could substantially affect the protein complement of the cell by altering the mRNA translation and lead to profound responses to chemical exposure. Finally, because of the potential control of the polycistronic regions at the transcriptional level, they form an expression profile in a similar manner to mRNA species. Therefore, the pattern of their expression could potentially be used to identify the type of toxicity that may be associated with the xenobiotic exposure. Such miRNA profiling has been shown to differentiate cancer types and the same could occur for different xenobiotic exposures [39] .
To assess the expression of the miRNA species using microarrays there are some technical challenges to overcome. First, if there is a desire to assess the mature miRNA forms only, it is necessary to use a RNA tailing method for the labelling because the miRNA species are too short for conventional labelling techniques. Second, as the miRNA species are short, it is often necessary to place targets on the microarray which contain modified nucleotides called locked nucleic acid nucleotides. These have the effect of making the hybridisation more like an RNA/RNA hybridisation, which has a higher melting temperature and thus gives rise to an increased stability and sensitivity [40] .
The recognition of toxicity by profiling the miRNA species may have little to offer over conventional expression profiling. However, their expression is fundamental to the control of the protein complement of the cells and this may be material both to the understanding of mechanisms of toxicity as well as differential susceptibility. There remains much work to be done to determine if miRNAs are differentially expressed between populations, the effect of this on mRNA translation and, ultimately, susceptibility or resistance to chemical toxicity. To date there has been no application of this technology in toxicology.

Expert opinion
The technologies described in this review give profound opportunities for the further understanding, recognition and assessment of toxicity. The last few years have seen a steady improvement in the quality of microarray technology as well as its breadth of application, and for many laboratories the use of microarray technology is now routine. In genotoxicity there is a recognised need to improve the predictive power of genotoxicity assays, particularly with respect to their relevance to actual carcinogenicity and non-genotoxic carcinogenic potential. The application of arrayCGH could inform, not only on the potential of a compound for aneugen or clastogen action, but also where the effect occurs on the genome. This is relevant because an increase in DNA instability in a known area of genomic area of instability could potentially be of less relevance than, for example, oncogene amplification. Epigenetic modification is achieving importance driven by understanding in cancer where changes in the epigenetic DNA profile are recognised as having a role the causing phenotypic changes in the tumour. The role of epigenetics in relation to chemical toxicity has not been explored extensively to date, but as relationships between epigenetic change and phenotype are defined then the assessment of epigenetic change in response to xenobiotics will become essential. This will be particularly so if a relationship between epigenetic change and phenotypic change in the progeny of exposed individuals is established. Of particular relevance for investigative toxicology will be the new application of microarray technology in the field of translation because it is becoming clear there is an important role for altered translation in toxic responses, particularly stress responses. This may be controlled by miRNA species which have promoter structures that may allow for transcriptional regulation by xenobiotics. Alteration of specific miRNA transcript levels can quickly result in the increase in the specific mRNA translation that may have been stored in an intracellular P-body. Recent data from the authors' laboratory has indicated that under the influence of certain chemicals some mRNA species are not only increased in transcription but also in translation. For genes in a rapid response network this seems a biochemically sensible approach. This then leads to the probable requirement of assessing differential miRNA expression in cells under the influence of xenobiotics.
The above techniques are an addendum to, and not a replacement for, the more established expression profiling applications which will remain the first choice microarray technique for both the assessment of toxicity and investigation of mechanisms. For further mechanistic understanding there is a need to understand the regulation of genes under the control of xenobiotics. ChIP-and microarray-based analysis of transcription rates will indicate genes that are similarly transcriptionally controlled. Combined with bioinformatics (analysis of the control regions involved) these assays will give knowledge on the network regulation of genes at the transcriptional level and those that are important for regulating cellular responses to xenobiotics. Analysed together with other data, in particular that of mRNA levels, it should be possible to further identify gene expression biomarkers specific for certain xenobiotic types from the whole cohort of genes differentially expressed in response to xenobiotic exposure. This in turn may ease the difficulties in analysing mRNA transcriptome data for the purposes of identifying hazards and quantifying risk.
For the drug development pipeline the following are highly desirable: • rapid identifi cation of potential toxicity in early stage development prior to Phase I • recognition of the potential involvement of genetic polymorphisms that may give rise to adverse drug reactions • some understanding of the mechanism of toxicity such that rational evaluation of risk can be made particularly with reference to species and strain responses.
To achieve the above, the use of mRNA profiling analysed with data from established end points such as histopathology and clinical chemistry will be the first microarray-based assay employed. These data should then be bioinformatically analysed by various methods such as phenotyping and pathway mapping and alongside other data sets from similar chemicals in order to discern possible mechanisms of toxicity. The use of transcriptomics has highlighted the variability in biological background, which is highly relevant to the assessment of toxicity. In vitro established and primary cells may be genetically unstable, easily contaminated and very responsive to changes in environment and so are particularly problematic in this regard. For this reason it is probably sensible to use inbred animal strains where there is control on the genetic background to reduce variability in the data. Reduction of variability is essential for high dimension data sets. It is conceivable that the time period required for assessment in vivo can be reduced such that, for example, the same data gained from a 4-or 5-day test as could be achieved with a 28-day (or longer) test. This would result in considerable cost saving with the added advantage of mechanistic understanding. ArrayCGH has been used to indicate the degree of variability in the human genome [7] and should be applied to inbred animal strains and cells used for toxicology testing to show genetic variability. These data can then be built into the risk assessment of xenobiotics which may indicate particular genome changes that give rise to differential susceptibility. ArrayCGH could also be used in vivo or in vitro in the same biological samples used for transcriptomics to indicate any genomic change associated with the xenobiotic exposure. Careful analysis of the regions affected could indicate whether these are likely to be important for toxicity. Similarly, the role of epigenetics in either conferring differential susceptibility or in mechanisms of toxicity should be explored. The data available indicating the involvement of epigenetic modification with the transgenerational toxicity of DES suggests this is an area of importance. Once again, the samples from a short exposure could be used with appropriate microarray techniques to assess epigenetic change. Single cell minisatellite analysis could be carried out to show any genome instability that may be associated with epigenetic change. Finally, the role of translation as a primary response to cellular stress is becoming apparent. Translation and miRNA analysis on the same samples from the short-term exposure studies could be carried out to discern if any xenobiotic related effect on translation has occurred, and if this is mediated through differential miRNA expression.
For mechanistic understanding, which may be necessary for an appropriate risk assessment, ChIP and transcriptional analysis could be carried out. Data from the transcriptome and translational assays would indicate if there was a requirement for these assays.
Therefore, microarrays have potential application both in recognition and understanding of toxicity from xenobiotics. The challenge has been, is now, and for the foreseeable future will be, not so much use of the technology but data interpretation. These technologies have exciting potential in toxicology but also bring with them the challenge of data analysis and integration for real success to be achieved.