Mutagenesis, Vol. 18, No. 3, 311-317,
May 2003
© 2003 UK Environmental Mutagen Society/Oxford University Press
MEETING REPORT |
Report of a symposium on the use of genomics and proteomics in toxicology*
Chemical Safety and Toxicology Division, Food Standards Agency, Aviation House, 125 Kingsway, London WC2B 6NH, UK and 1 Department of Health, Skipton House, 80 London Road, Elephant and Castle, London SE1 6LH, UK
Over the past few years, and especially since publication of the sequence of the human genome (International Human Genome Sequencing Consortium, 2001
), there has been increasing pressure to incorporate novel technologies such as proteomics and genomics into toxicological risk assessment.
The Committee on Toxicity of Chemicals in Food, Consumer Products and the Environment (COT) and its sister Committees on Carcinogenicity and Mutagenicity (COC and COM) held a symposium to consider this issue on 8 October 2001 at Skipton House, Elephant and Castle, London, UK. The meeting was attended by members of COT, COC and COM and also included members of the Advisory Committee on Hazardous Substances, the Advisory Committee on Pesticides, invited speakers from academia and industry and delegates from various Government Departments. It was also open to anyone with an interest in this area.
COT, COC and COM are independent expert advisory committees appointed by the Chief Medical Officer and the Chairman of the Board of the Food Standards Agency. The Committees advise on all aspects related to the toxicity, carcinogenicity and mutagenicity of chemicals. They also have a general remit to advise on important general principles or new scientific discoveries in connection with safety assessment of chemicals. This symposium brought together experts in the fields of genomics and proteomics to assist the committees in providing advice to Government Departments and Regulatory Agencies on the use of these technologies in risk assessment. The symposium started with a general overview of these subjects and introduced the areas of discussion for individual Working Groups. The concluding session included a report from each Working Group and overall recommendations.
Introductory session
Overview of use of genomics and proteomics in toxicology
The symposium was introduced by Professor Frank Woods (University of Sheffield, COT Chairman). Dr George Orphanides (Syngenta CTL, Macclesfield) presented brief details of the development of genomic and proteomic techniques used in toxicology. The term genomics was used to refer to transcript profiling, i.e. changes in levels of mRNA transcripts due to alterations in mRNA stability or gene regulation (Pennie et al., 2001). The techniques of genomics, transcriptomics (defined as the analysis of the entire transcribed cellular genes), proteomics and metabonomics provide information about different components of the pathway from action on the genome to functional effect(s). Thus genomics (or more correctly transcript profiling) and proteomics are complementary techniques that measure different aspects of the cellular regulatory process.
Key aspects for risk assessment were evaluation of the differences between gene expression associated with normal cell function and that resulting from exposure to exogenous test chemicals. Identification of changes in expression would potentially provide information on the toxic or adaptive responses to test chemicals.
The basic experimental approach evolved from the Northern blot methods developed in the 1970s, involving separation of mRNA by gel electrophoresis followed by immobilization on a membrane. Specific genes are then identified and quantified by hybridization to a labelled DNA probe of a complementary sequence. This was followed by the introduction of the more rapid dot blot assays using immobilized DNA sequences probed with a labelled RNA pool, ultimately leading to the microarrays used now. These can simultaneously screen thousands of gene sequences immobilized on nylon or glass. A pool of either fluorescently or radioactively labelled first-strand cDNAs is reverse transcribed from mRNA transcripts extracted from test cells and control cells for comparison. These are then bound to complementary sequences fixed at specific locations on the microarrays. The location of fluorescence/radioactivity identifies the gene expressed and the intensity of the signal is proportional to the expression level of the gene. A typical ‘toxicogenomic’ study involves analysing the differential expression of genes in response to the test chemical under consideration. In vivo studies use cells or tissues of treated and control animals. In vitro studies use treated and control cells, cell cultures or organ cultures. The basic approaches used in proteomics were also founded in the 1970s, with the use of two-dimensional gels to separate proteins firstly on the basis of pI (isoelectric point) and then by molecular weight. Proteomics extended this technique by identifying proteins separated in such gels by mass spectrometric fingerprinting using specific proteases. A ‘toxicoproteomic’ study involves separation of proteins from treated and control cells or tissues and analysis by mass spectrometry. Proteins are identified by comparison with predicted fingerprints derived from databases of protein sequences. However, proteins are often modified post-translationally, e.g. by phosphorylation. It is also possible to identify these types of modification using specific variations on the basic proteomic experiment as well as by more conventional techniques such as immunoblotting. In practical terms cDNA microarrays offer the prospect of high throughput screening of samples for large numbers of genes (including low abundance genes). At present proteomics has a lower sample throughput but can provide information that is relevant to the toxicological mechanisms of chemicals and can be used in samples, such as body fluids, which contain proteins but not mRNA. Both genomics and proteomics generate large quantities of data that are difficult to evaluate and require sophisticated and informed biomathematical and bioinformatic support.
Most toxicological effects, either directly or indirectly, involve changes in expression levels of genes and proteins. These changes can suggest a mechanism of action for the compound under investigation. Such changes may also have the potential to be used in a predictive capacity, under the premise that compounds with similar mechanisms of action will produce similar changes in gene and protein expression. Examples of the latter were the identification of biomarkers and expression fingerprints to help elucidate target organ effects or identify toxicant groups/classes of chemicals. A number of useful Internet sites are available to help identify relevant genes (e.g. http://www.genome.ad.jp/kegg/ and http://inn.weizmann.ac.il/look_2000/g.html). The criteria for using these techniques for the development of biomarkers may include selectivity, sensitivity, reliability and reproducibility. In addition, any putative biomarkers should be faster, more sensitive, cheaper and use fewer animals than existing biomarkers.
Dr Orphanides considered that a number of key challenges need to be met before genomic and proteomic data can be used in chemical risk assessment. Firstly, for each toxic process there is a need to establish a link between changes in gene/protein levels and cell phenotype. Secondly, it is important to be able to distinguish genes/proteins involved in toxicity from those showing regulatory changes of no consequence to the toxicological mechanism (‘bystander genes’). The potential for data to be over-interpreted may generate undue concern. Dr Orphanides considered that the key issues were experimental design and data analysis, interpretation and use in risk assessment. There might be a role for use of genetically engineered animals in the interpretation of critical genes in toxic responses. A key question in data interpretation is judging the extent of normal biological variation of key genes involved in mechanisms of toxicity. Topics for further research relate to development of appropriate statistical methods to detect trends in expression (e.g. detection of regulatory changes to maintain homeostasis in relevant gene clusters) and of databases for archiving data on gene expression together with classical toxicology data. Dr Orphanides concluded by expressing the view that, in the absence of classical toxicology data, changes in gene/protein expression cannot (currently) be used as indicators of adverse effect. There is a need for careful examination of whether changes in the critical genes/clusters can be related to toxicity. However, the clear rewards in using genomics and proteomics include the development of biomarkers, faster investigation of mechanisms of toxicity and screening of novel compounds.
Discussion. Distinguishing ‘critical’ from ‘bystander’ gene effects was considered problematic. It was noted that even within a class of chemicals such as peroxisome proliferators, individual chemicals had differing target organs and mechanisms of toxicity. Thus identification of relevant gene clusters for particular mechanisms of toxicity is a critical step in the evaluation process. Improved databases require an extensive level of data sharing between laboratories but there was concern that this could be hampered by commercial sensitivity. Using genomics/proteomics in screening might direct researchers to novel gene targets thus improving understanding of the biology of chemical-induced pathology. It was also noted that the use of genetically engineered animal models to confirm a specific mechanism will probably only apply to that specific mechanism. In addition, confidence in the outcome of a study of the overall toxicity of a compound presupposes a knowledge of the toxicity of the compound, currently imperfect or poor for many compounds. The use of in vitro models based on permanent cell lines may be of limited relevance due to loss of many normal cellular functions and should be treated with caution.
Genomics
Dr Valerie Baker’s (Unilever, Bedfordshire; now at CuDoS Ltd, Nottingham) overview of the material to be presented to Working Group 1 is outlined below.
There has been an exponential growth in the number of published scientific reports referring to microarray technology. The principles of toxicogenomics are based on the premise that most toxicologically relevant outcomes are preceded by changes in gene expression. The immediate goal of toxicogenomics is to identify gene expression patterns that accurately reflect and predict specific toxicological end points. The issues are whether molecular (mRNA) fingerprints identified in microarray screening experiments can be used to predict toxicity and/or help define mechanisms of toxicity and improve interspecies extrapolations. Dr Baker outlined the scope of the International Life Sciences Institute/Health and Environmental Sciences Institute (ILSI/HESI) collaboration on the ‘Application of genomics and proteomics in mechanism-based risk assessment’ (outlined at http://www.ilsi.org/index.cfm?pubentityid=51; factsheet at http://www.ilsi.org/file/genomics.pdf). The goal of this work is to develop a broad-based cross-sector (industry, government and academia) collaboration to advance the scientific application of diverse genomic and proteomic technologies to mechanism-based risk assessment. The approach relates studies conducted in vivo and in vitro analysed by microarrays to the results of studies using more conventional approaches (e.g. histopathology, clinical chemistry and toxicokinetics) for three areas of toxicity (hepatotoxicity, nephrotoxicity and genotoxicity) and aims to produce a publicly available database. The ILSI/HESI programme is considering a wide range of experimental design issues (e.g. chemical selection, temporal effects on gene expression, dose–response relationships and statistical aspects of good study design). Examples of data presented included information on inter-laboratory variation for effects on specific genes in response to chemical exposures. The programme has highlighted the many challenges to the development of screening methods (e.g. inter-laboratory reproducibility, optimum experimental design, gene selection, mRNA sampling and comparison of data with results of animal toxicology studies) and a need to achieve a consensus on data interpretation, building reference data sets and approaches to extrapolation of patterns of gene changes due to known and unknown chemicals. It was recognized that building partnerships and consensus across regulatory, industrial and academic forums was very important to the successful application of genomics in toxicity screening.
Discussion. Defining differences between adaptive and toxicologically relevant changes was considered a key issue, as was the question of whether data from toxicogenomics studies could be used to define thresholds for use in toxicological evaluations. Important steps were needed before toxicogenomic data could be used. These included a consistent response in microarray studies and the interpretation of these data with regard to the mechanism of toxicity. The need for caution when considering in vitro data was stressed as data correlating in vitro and in vivo studies are limited. In addition, it was noted that the initiator events in some toxicological effects might not require changes in gene expression.
Proteomics
Dr Cliff Elcombe (University of Dundee) gave an overview of the issues to be discussed in Working Group 2.
He gave a general overview of proteomics, including a summary of techniques starting with two-dimensional gel separation. A variety of pH gradients can be applied during the first dimension separation by isoelectric focusing, depending on the extent of separation needed. After running the second dimension on a SDS–polyacrylamide gel, proteins can be visualized by traditional approaches using Coomassie blue staining or radiolabelled compounds. More recently, fluorescent markers and immunostaining have been used. The most recent techniques involve affinity capture of proteins and a specialized platform (e.g. Ciphergen’s SELDI ProteinChip®) and separation using mass analysis by matrix-assisted laser desorption and ionization time-of-flight (MALDI-TOF) mass spectroscopy. Identification of individual proteins requires establishment of databases and there is currently considerable activity in this area. Theoretically proteomics can be used quantitatively to analyse the expression of all proteins in a cell. Hence proteomics may be a better predictor of functional changes during toxicological processes than genomics, since mRNA levels do not necessarily reflect changes in concentrations of functional proteins. However, the limited throughput and time-consuming analysis of proteomic experiments limits the number of experiments. Potential applications include screening for novel toxicants, identification of novel mechanisms of toxicity and application to epidemiological investigations. A key issue for discussion was whether toxicoproteomic data could assist in the identification of No Observed Adverse Effect Levels (NOAELs) in toxicological studies. Dr Elcombe noted that there are several problems which need to be evaluated further before toxicoproteomic studies can be used in chemical risk assessment. Levels of many proteins vary widely with the age of experimental animals and inter-animal variation may be significant. Lastly, there are few published studies of the function and relevance of protein changes in relation to the mechanism of toxicity in animals with extrapolation of the data to humans.
Discussion. It was suggested that in cases where proteomic changes were found to be causally related to the toxicity of a compound it would be prudent to presume that these effects would also be relevant to humans until shown otherwise. At present protein databases are too limited to ensure that the identity of proteins measured in animal experiments could be confirmed. It was also noted that a single spot on a gel does not necessarily represent a single protein and further separation is needed before mass spectrometry is carried out. A single spot on a gel could be a mixture of proteins and/or isoforms. It was possible that this approach would mask relevant changes in protein levels from toxicoproteomic experiments. Though it is possible to detect dose–response effects, these need to be validated by a temporal effect, especially as isoforms of the proteins may alter in different ways.
Risk assessment
Dr Tim Gant (MRC, Leicester) gave an overview of the issues to be discussed by Working Group 3.
Consideration of what can be achieved now and possibly in the future using genomic matching and pathway profiling is important. Suggested key outcomes would be: (i) prediction of adverse effects; (ii) better understanding of the relevance of toxic effects in animals to humans; (iii) potential identification of subgroups in the population with greater/lower sensitivity; (iv) eventually faster toxicological assessment of chemicals by reducing the time to undertake toxicological testing in animals. These points were expanded on as follows.
- It is now possible to screen for potential effects on gene expression of thousands of genes relatively quickly, using cluster analysis to investigate mechanisms and in some instances predict effects of chemicals. In the future information from collaborative projects such as ILSI/HESI should improve the reliability of predictive analysis.
- There is good reason to anticipate that within a relatively short time frame proteomic data will enhance extrapolation from experimental animals to humans. However, there is a need to generate comprehensive background genomic and proteomic profiling data in experimental animals and humans to provide information on normal variation.
- It is possible to sequence DNA samples rapidly and to identify single nucleotide polymorphisms (SNPs) in individuals. The challenge is to identify the role of SNPs in disease causation.
- If reliable pattern mapping of toxicological effects could be attained then the prospect of avoiding some toxicological tests is feasible. However, a great deal of collaborative work is required. A further challenge relates to development of appropriate statistical bioinformatic tools. Thus, for example, in studies using a microarray of 10 000 genes, up to 500 of these genes might be wrongly identified using a 95% confidence limit.
Discussion. The importance of establishing sound databases for predictive toxicity was deemed a considerable challenge for the future. The use of multiple comparisons was considered a problem and the need to have input from qualified statisticians to avoid inappropriate use of statistical packages and software was highlighted, together with the need for agreement amongst collaborative groups on this issue. However, it was also considered important to apply biological logic to the evaluation of data and not be driven by statistical evaluations.
Working Group sessions
Use of genomics (transcript profiling) in screening
Speaker: Dr Valerie Baker (Unilever, Bedfordshire)
Facilitator: Dr Philip Carthew (Unilever, Bedfordshire)
The Working Group’s discussion covered the following points.
- The relationship between toxicological end points and gene expression patterns. Some information is available but projects such as are currently being coordinated by ILSI/HESI are necessary to test the reproducibility of the techniques used in genomics and correlation of the findings with known pathologies. It is necessary to correlate the changes in gene expression (including down-regulation of genes) with time to the corresponding histopathology.
- Prediction of toxic response. This is unlikely at present but some participants considered it possible if sufficient reference databases could be established linking specific gene expression patterns to specific toxicological effects in a range of organs. There is evidence that a number of hepatotoxins can be grouped into mechanistic classes on the basis of gene expression patterns (Waring et al., 2001). However, it should be recognized that there may always be novel toxicants with novel mechanisms of action or patterns of gene expression which do not match any reference database.
- Identifying mechanisms of toxicity. Genomics may be useful in mechanistic studies to understand species and strain differences and idiosyncratic reactions in man. However, it may be less useful than proteomics for studying differences between man and test species because comparisons cannot be made on components of body fluids. Instead, comparisons would entail use of human and animal cells in vitro.
- Application to specific toxicological effects. Genomics may be useful in predicting mutations in error-prone DNA synthesis caused by genotoxic carcinogens, but use in predicting non-genotoxic carcinogenesis would depend on sampling the right tissue at the right time in the carcinogenic process. Genomics may be useful in establishing the potential for carcinogenesis related to hormonal effects arising from chronic exposure to chemicals and for adverse reproductive effects not predicted by histopathology. Due to differing concentrations of hormones such as estrogen the extrapolation of rodent data to humans has been a perennial problem in these areas and it is possible that genomic analysis may provide better biomarkers. Genomics is unlikely to be of use in teratology studies because of the rapid rate of change in the developing organism. There was no consensus about its potential use in immunotoxicity, although in vitro techniques might be useful if the appropriate cells could be identified for study. Use in neurotoxicity is also problematic because of a lack of knowledge of the areas of the brain anatomically associated with functional signs of neurotoxicity and the difficulty in validating the data when traditional methods for detecting neurotoxicity are limited. The use of laser capture microdissection, where focal areas can be examined for alterations in gene expression, would be one method of correlating gene expression to functional neurological signs.
- Application to deriving NOAELs. At present, a NOAEL cannot be based on gene expression data because of the difficulty of distinguishing expression levels associated with adaptive or pharmacological responses and those associated with clearly adverse responses. Overall, the group considered that it is important that genomics be used as part of the weight-of-evidence approach to hazard assessment but that considerably more validation is required before it could provide the primary basis for risk assessment.
Applications of proteomics in toxicology
Speaker: Dr Cliff Elcombe (University of Dundee)
Facilitator: Dr Sandy Kennedy [Oxford Glycosciences (UK) Ltd]
Dr Elcombe presented results from experiments to aid discussion. The group agreed that two-dimensional gel separation and mapping techniques provided a reliable approach. Greater separation of low abundance proteins would become readily available using high performance liquid chromatography/mass spectrometry and protein chip methods. The group discussed several potential applications.
- Screening/predictive toxicology. The use of proteomics in screening and predictive toxicology has two principal applications: establishing relationships between toxic effects and protein molecular markers, i.e. identifying toxicological biomarkers; recognition of patterns, e.g. class effects and structure–activity relationships. In addition, proteomics offers several potential practical benefits. It should be possible to screen for toxic effects more rapidly with the advent of the newer proteomic methodologies [e.g. isotope-coded affinity tags (ICAT) and antibody chips] than with conventional methods. The highly sensitive analytical techniques used in proteomics have the potential to detect toxic effects at lower doses than methods such as histology and clinical chemistry. However, currently the application of proteomics does not extend to primary hazard identification. A considerable amount of work is required to elucidate optimum methods for screening (e.g. duration of dosing, separation of cells or fractions and protein separation/identification methods). The ILSI/HESI research initiatives on genomics could be usefully extended to proteomics. There is a need to establish reliable proteomic databases and to avoid over-interpretation of results in the absence of appropriate histopathology or to endorse correlation of proteomic changes with mechanisms of toxicology. The group discussed the potential use of proteomic experiments in setting/refining NOAELs and uncertainty factors and agreed that, at present, proteomic studies could only be used in highly defined circumstances and only for ‘data-rich’ chemicals.
- Mechanistic toxicology. Proteomics, especially when combined with conventional methods, offers the prospect of new insights into toxic mechanisms. Such insights allow recognition of effects that may be species-specific, giving a more accurate assessment of likely human toxicity. An example is cyclosporin A nephrotoxicity in the rat, mediated by decreased levels of the 28 kDa protein calbindin (Aicher et al., 1998). Furthermore, understanding the mechanisms of toxicity of compounds may enable selection of derivatives with lower toxicity.
- Non-invasive biomarker identification. A particular advantage of proteomics is that not only tissues but also body fluids can be assayed to investigate the molecular correlates of disease and toxicity. This is possible because, unlike mRNA, many proteins are secreted in patterns that vary predictably with physiological state. As a result, proteomic analysis can be carried out on large numbers of blood or urine samples.
The evaluation of body fluids using proteomics can be of particular value in the search for non-invasive biomarkers, as they are representative of the final secreted protein. However, many samples from earlier studies may not have been stored appropriately, but the techniques are suitable for prospective studies. The capability to separate proteins is enhanced by the ability to remove high abundance proteins such as albumin, IgG and haptoglobulin and by the option of transferring proteins from two-dimensional gels. Using an immunoaffinity-based enrichment technique hundreds of proteins that would previously have been masked from detection on a gel can be revealed. This is therefore a rich source of data on biomarker identification for toxicity, efficacy of a drug or exposure to a xenobiotic in humans or animals. Once a biomarker protein or group of proteins is identified, standard methods such as immunoassays can be used for screening. The group considered that it would be essential to establish the reproducibility of any proposed proteomic biomarker technique and to compare sensitivity and specificity with existing methods before concluding on suitability. Background variation in the general population would need to be described.
Proteomic techniques are likely to make a considerable contribution not only to research but also to regulatory toxicology. However, in the short term proteomic methods are likely to complement rather than replace conventional approaches for regulatory purposes. Protein biomarkers offer great potential to improve the predictivity of animal studies and, in particular, provide the bridge between effects in animals and in man. Until a greater body of toxicoproteomic data has been acquired, it is unwise to use such evaluations to do a primary identification of target organ toxicity. However, identification of more sensitive biomarkers can be envisaged as enhanced clinical pathology tools in toxicity studies.
Use of genomics/proteomics in risk assessment
Speaker: Dr Tim Gant (MRC Leicester)
Facilitator: Dr Andrew Smith (MRC Leicester)
Dr Gant presented a synopsis of two genomics studies highlighting how this technology may be used to define toxic mechanisms and identify biomarkers of exposure and effect.
Examination of data from one study demonstrated that the differential expression of clusters of genes, grouped together by a common function, could be used as markers of biological mechanisms or effects. Information from animal pathology/clinical chemistry studies is needed to be able to relate patterns of gene expression to adverse effects in the first analysis. However, subsequently it could be anticipated that the gene expression pattern would itself be a reliable marker of pathological change and preferably indicate the likely onset of pathological change before it is observable by histopathological analysis. An example of recognition of inflammatory change in the liver identified by a specific gene expression profile was presented. The inflammation was produced by the administration of the ferrochelatase inhibitor griseofulvin. Over time, development of inflammation could be observed histopathologically mirrored by the development of a specific gene expression profile, which included many genes associated with white cell infiltration. Additionally, the onset of the inflammatory change was preceded by another gene expression profile that may be indicative of the onset of this pathology and thus may be useful in toxicogenomic analysis to indicate the propensity of such change in the absence of its actual observation. At the later stages of the time course a gene expression profile was observed which seemed to indicate early fibrosis, and careful histopathological analysis confirmed this. In this instance the gene expression profile was able to indicate pathological change that was difficult to observe due to the very early stage of the change and thus clearly identified the potential for gene expression profiling in toxicological assessment.
Data were also presented on gene expression profiling in the liver after the administration of a novel antiproliferative compound (Donald et al., 2002). A surprise finding was induction of the cell cycle switch gene (cdc2). This led directly to further analysis at the histopathological level using Ki67 to measure cell proliferation. A substantial degree of hyperplasia was indicated that had not been seen by histopathological analysis and would not have been considered for a compound with antiproliferative effects in tumours and other organs. Thus the gene expression profiling led to the recognition of a novel and unexpected toxicity.
- Interpretation of genomic and proteomic data. The difficulties associated with interpreting genomic and proteomic data were discussed. Knowledge of gene function was considered crucial to interpretation of the results and whether the specific patterns of gene expression could be related to adverse effects or other effects (e.g. an adaptive response). It was acknowledged that, as the function(s) of many genes is unknown, the biological significance of precise quantitative data could not be determined. Qualitative interpretation of differential gene expression may be more beneficial. It was recommended that gene expression following treatment with known toxins be investigated to establish whether dose–response relationships could be derived. In order for pattern recognition to be used predictively for adverse effects, large amounts of data on different chemicals would need to be generated and put in the public domain. This may well be inhibited by concerns regarding data confidentiality.
Although there is little consistency in design and conduct of genomic and proteomic experiments, it was considered that defined protocols are unnecessary if laboratories are confident in the techniques and statistics used and some degree of reproducibility can be obtained in the results. In view of the huge amount of data generated in studies using microarrays, appropriate expertise regarding statistical analysis is essential. Although commercial systems provide their own software for this, there is often controversy regarding whether these systems are appropriate. For public domain data, or that used in collaborative exercises, the raw data should be supplied in the form of the images, annotation file and the measure of the image fluorescence as determined by the generating laboratory. This would allow different statistical tools to be applied. The standards currently being published by the microarray gene expression database group (http://www.mged.org) should be taken into account and adhered to if at all possible.
The techniques utilized for the determination of gene expression are only poorly quantitative, resulting in a tendency towards variability in the analysis between replicate experiments. This combined with the very dynamic nature of gene expression and the problems inherent at sampling at exactly the same moment in time can result in a high variance in the determination of gene expression. This makes it imperative that appropriate statistical analysis is applied to the data sets. Arbitrary cut-off points should have no place in the analysis of such data. The recognition of a differentially expressed gene should be based on the use of replicate experiments and quantitation to which statistical analysis is applied. In addition, experiments need to be carefully designed to ensure that the data generated are the most appropriate for statistical analysis.
- Use of genomics and proteomics in epidemiology. It was noted that removing the high background variability of gene expression in long-term studies of whole populations could cause problems. In addition, as exposures are measured very late in such studies, and biomarkers are generally short-lived, only chronic exposures or exposures with long-term effects may be measurable.
It was considered that both technologies had strengths and weaknesses. In some areas, proteomics may be more useful as it is capable of analysing samples that can be easily obtained from humans. However, it was recognized that the study of proteomics was less advanced than genomics and could only offer a more limited level of sophistication. Genomics may be useful if transcript levels in blood lymphocytes can be used as surrogate markers for toxicity. However, this approach has not been investigated to date and gene regulation in lymphocytes may be related to induction of responses in this cell type to the properties of the chemical rather than toxicity elsewhere in the body.
It was suggested that epidemiological studies could be aided by genomic and proteomic analysis of the plasma samples to be collected as part of the UK Population Biomedical Collection (http://www.wellcome.ac.uk/en/1/biovenpop.html), now known as the UK Biobank (http://www.ukbiobank.ac.uk/).
- Use in risk assessment. These techniques may be used to identify possible mechanisms of action as well as biomarkers for exposure and predictors of effect. Their usefulness in these areas is dependent on knowledge of gene function as well as the relationships between gene expression and biological effects. It is important to distinguish adverse from adaptive effects (e.g. a response to changes in nutritional status, for example post-prandial). The limited information on natural variability and on baseline values restricts the applicability of gene expression patterns in predicting biological effects. These techniques could not currently be used for risk assessment in isolation and must be accompanied by additional supporting evidence from conventional toxicological studies.
Genomics and proteomics were regarded as inappropriate for routine biomonitoring due to the high background variability in gene expression but could potentially be useful for hazard identification. It was suggested that these techniques could be used in areas, such as endocrine toxicology, where histopathology data were absent or unclear but there would again be the need to define adverse effects as opposed to adaptive/pharmacological effects. They may also be applied to the analysis of mixtures of chemicals, although, at present, analysis of even single compound exposures was extremely difficult.
- Communication of results. The importance of communicating the results of genomic and proteomic studies to the layperson was considered. It was recommended that the limitations of these technologies should be highlighted and the problems and uncertainties in the data should be effectively communicated, especially the point that changes in gene or protein expression need not be of toxicological significance. In addition, data from such studies cannot at the present time be used to make statements on the health risks of chemicals in the absence of other supporting toxicological studies or where effects are seen below the NOAEL established by conventional toxicological studies.
- Future research. It was recommended that further work be undertaken on the background variability in gene expression so that expression changes in response to chemical exposures can be put into context. It was suggested that validation of genomic and proteomic techniques with known toxins, such as that being conducted by the ILSI project, should be encouraged.
It was acknowledged that such technologies would not reduce the use of animals in toxicology in the short term but they may provide sufficient information to focus and shorten the duration of toxicological studies, reducing the cost and number of animals required in the future.
Discussion session
Professor Peter Farmer (University of Leicester) chaired the discussion session. Members agreed a number of overall conclusions from each of the Working Groups.
Genomics
- Data from toxicogenomics studies are potentially suitable for use in hazard assessment but not currently in risk assessment. Changes in gene expression could not be used, at present, to identify NOAELs.
- There is a need to correlate changes in gene expression, including significant decreases with time, to corresponding histopathology.
- It is particularly important, but potentially difficult, to identify whether changes in gene expression represent adaptive rather than adverse effects.
- A number of toxicological end points might be suited to screening by genomics, including hormonally mediated carcinogenesis, mutation in error-prone DNA test systems and adverse effects on reproduction (but not teratology).
- The potential application to neurotoxicology needs further research to correlate gene expression with the location of functional/structural effects in the central and peripheral nervous system.
- Toxicogenomic changes should not be considered in isolation, but rather in a traditional weight of evidence approach to hazard assessment.
Proteomics
- Toxicoproteomic studies cannot be used for primary identification of hazards and should not be used for regulatory toxicology at present.
- The most probable uses are in screening out of lead candidates where there is some knowledge of the mechanism of toxicity.
- Proteomics may also be of use for identifying novel mechanisms.
- Currently there are very few studies of toxicological effects such as immunotoxicity, neurotoxicity and genotoxicity and it is therefore not possible to make preliminary comments on the suitability of proteomics for screening for these effects.
- There are difficulties with the reproducibility of studies. A collaborative initiative on proteomics similar to the ILSI/HESI project on toxicogenomics would provide much-needed validation data.
- Two-dimensional gel electrophoresis will eventually be replaced by other separation techniques such as ICAT and antibody chip identification.
- There is an urgent need to develop databases for protein identification.
- Use of data from proteomic studies to refine NOAELs is possible only in defined circumstances where the protein under investigation can be causally related to the toxic mechanism and observed pathology and the study includes dose–response data.
- Proteomics shows potential for applications in human biomarker studies but cannot be used retrospectively in situations where specimens have not been subject to storage at -70°C. It would be important to establish the reproducibility and dose–response relationship of any potential ‘proteomic biomarker’ and to compare these data with existing methods with regard to specificity and sensitivity.
- The potential benefits of this methodology, such as the development of non-invasive techniques, were also noted.
Risk assessment
- Genomic and proteomic technologies show considerable potential but at present the information produced is difficult to interpret and conclusions drawn from such data should be treated with caution.
- Risk assessments made purely on the basis of genomic and proteomic studies should not be regarded as reliable in the absence of supporting data from more conventional toxicological studies.
- The complex data sets generated from genomic and proteomic studies must be analysed using appropriate statistics and input from statisticians with relevant expertise/experience is essential.
- Further research into the natural background variations in gene expression is recommended to aid interpretation of the effects of toxins on gene expression.
- Communication of the results of genomic and proteomic studies requires care to avoid mis- or over-interpretation of the results.
- It is not anticipated that genomics and proteomics will lead to a reduction in the use of animals in toxicological studies in the short term. However, genomics should lead to the more appropriate use of both in vivo and in vitro model systems where extrapolation to man can be made on the foundation of a fundamental understanding of the nature of variation in gene expression between the two systems. In the future it may provide sufficient data to enable toxicological studies to be more focused, thus reducing the use of animals.
The meeting concluded by recognizing the future potential of genomics and proteomics in toxicological risk assessment. Routine application of these technologies in hazard identification and mechanistic research is a realistic probability in the near future. It was also agreed that the techniques might sometimes serve as adjuncts to conventional toxicological studies. However, there is a need for more basic research on genomic/proteomic databases, methods of statistical analysis of data and pattern recognition and, in particular, information on the normal range of gene expression before the prospect of their use in regulatory toxicological risk assessments can be envisaged.
Acknowledgments
The authors would like to thank members of COT, COC and COM and their secretariats and all speakers and participants who helped in the preparation of this paper.
Notes
2 To whom correspondence should be addressed. Tel: +44 20 7276 8520; Fax: +44 20 7276 8513; Email: caroline.tahourdin{at}foodstandards.gsi.gov.uk 
* The opinions expressed in this paper represent the opinions of the speakers and do not necessarily represent the views or policies of the Department of Health or the Food Standards Agency.
References
-
Aicher,L., Wahl,D., Arce,A., Grenet,O. and Steiner,S. (1998) New insights into cyclosporine A nephrotoxicity by proteome analysis. Electrophoresis, 19, 1998–2003.[CrossRef][Web of Science][Medline]
Donald,S., Verschoyle,R.D., Edwards,R. et al. (2002) Hepatobiliary damage and changes in hepatic gene expression caused by the antitumor drug ecteinascidin-743 (ET-743) in the female rat. Cancer Res., 62, 4256–4262.
International Human Genome Sequencing Consortium (2001) Initial sequencing and analysis of the human genome. Nature, 409, 860–921[CrossRef][Medline]
Pennie,W.D., Woodyatt,N.J., Aldridge,T.C. and Orphanides,G. (2001) Application of genomics to the definition of the molecular basis for toxicity. Toxicol. Lett., 120, 353.[CrossRef][Web of Science][Medline]
Waring,J.F., Ciurlionis,R., Jolly,R.A., Heindel,M. and Ulrich,R.G. (2001) Microarray analysis of hepatotoxins in vitro reveals a correlation between gene expression profiles and mechanisms of toxicity. Toxicol. Lett., 120, 359–368.[CrossRef][Web of Science][Medline]
Received on September 26, 2002; accepted on February 17, 2003.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. M Battershill Toxicogenomics: regulatory perspective on current position Human and Experimental Toxicology, January 1, 2005; 24(1): 35 - 40. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||