People are exposed daily to diverse chemical and physical agents in the environment, some of which may adversely affect human health. These agents are present across an enormous range of concentrations, which results in a wide variety of human exposures.
In this context, it is useful to consider the distinction between exposure and dose. Exposure is the amount of a given agent presented by the immediate environment, whereas dose is the amount of an agent absorbed by the individual. Thus, dose has an implicit time component, which depends on the duration of exposure. Individuals in the same environment for the same amount of time may experience different doses of a given compound because of differences in absorption or metabolism. Human behavior adds to the variables. For example, individuals consuming arsenic-contaminated drinking water from the same source may absorb different doses simply because they drink different amounts of the water. Individual variations in physiology or metabolism can further modify the dose by affecting the rate of clearance or enzymatic processing. Finally, the dose to specific organs often depends on metabolic pathways, as in the concentration of radioactive iodine in the thyroid.
Despite these challenges, environmental toxicologists have long sought to use the modifications or expression changes of specific biomolecules to determine actual exposures in individuals and in populations. This chapter describes the current status of exposure assessment with biologic molecules as indicators, focusing on the application of toxicogenomic technologies to assess human exposure to environmental agents. The hope and expectation is that these new toxicogenomic approaches will allow more refined and sensitive assessment of exposure through the measurement of more subtle and broader changes in gene and protein expression and through identification and quantification of greater varieties of metabolites. The greater detail provided by toxicogenomic information might also enable the distinction of individual components of exposure in complex mixtures or the detection of exposures after time has elapsed, providing “fingerprints” of biologic responses to agents that are not retained in tissues, such as ionizing radiation.
This chapter focuses on applications of toxicogenomic technologies to measure environmental exposures in populations exposed to occupational and environmental agents, an important task in risk assessment and toxicology research. The related problem of exposure to pharmaceutical agents is not considered here. The relevance of toxicogenomics to risk assessment modeling is discussed after a description of the state of the art.
CONVENTIONAL BIOMARKERS OF EXPOSURE AND RESPONSE
The term biomarker has been widely applied to describe quantifiable molecular species that reflect biologic states, exposures, and disease. Under this broad heading, two biomarker subcategories can be distinguished: biomarkers of exposure, reflecting the occurrence of an exposure; and biomarkers of response, which indicate the response of an organism to an exposure.
Biomarkers of exposure may be specific modifications of specific molecules, such as the adducts formed on hemoglobin due to exposure to benzene or pyrolysis products (Skipper et al. 1994; Medeiros et al. 1997; Alexander et al. 2002), the DNA adducts produced by exposure to vinyl chloride or urethane (Skipper et al. 1994), or the polycyclic aromatic hydrocarbons from cigarette smoke (Shugart et al. 1983; Perera et al. 1986). Although xenobiotics and the metabolites that persist in tissues (for example, polychlorinated biphenyls, dioxins) are used as exposure biomarkers, adducts are the most commonly used biomarkers of exposure. Adducts can persist detectably in the organism, constituting biomarkers of a past exposure. Exposure biomarkers such as hemoglobin adducts might not be involved in toxic effects, but others (for example, mutagenic DNA damage) may be.
Single-molecule species have been most commonly used as biomarkers, and traditionally they have been measured by conventional approaches such as gas chromatography and high-performance liquid chromatography. However, a collection of biomarkers may also be used as a fingerprint of exposure. Monitoring multiple biomarker molecules could boost the sensitivity of detection as well as its specificity in reporting a particular exposure type. One reason is that different agents can produce overlapping profiles of adducts, so that measuring multiple types of adducts could distinguish related but different exposures. Such distinctions can help define exposures associated with different disease risks. For example, weakly carcinogenic methylating agents such as methyl methanesulfonate produce N7-methylguanine as the predominant DNA adduct, which is also the main DNA lesion formed by potent carcinogens such as N-methyl-N'-nitro-N-nitrosourea. The key mutational and carcinogenic effects of these compounds, however, are mainly due to differences in the levels of relatively minor lesions: the strongly mutational adduct O 6-methylguanine accounts for only about 0.1% of DNA damage caused by the weak mutagen methyl methanesulfonate, but it accounts for about 7% of the damage caused by the potent mutagen N-methyl-N'-nitro-N-nitrosourea (Montesano et al. 1980; Pegg 2000).
Adducts and persistent metabolites are just one measure of exposure. Cells and tissues also alter their metabolism or gene expression in response to exposure. Well-known examples include the induction of specific groups of genes in response to heat shock, hypoxia, or osmotic stress (Finkel and Holbrook 2000; Chellappan 2001; Berra et al. 2006). These responses can produce patterns of specific changes in gene expression, proteins, or metabolic profiles that report the exposure to a particular agent or class of agent. Complex sets of gene expression changes and corresponding changes in protein networks and metabolite profiles have been beyond the reach of older technologies. Toxicogenomic technologies offer new opportunities to detect and quantify these changes and apply them as new classes of biomarkers. Some of the initial work along these lines is described in the next section.
STATE OF THE ART: TOXICOGENOMIC APPROACHES TO THE DEVELOPMENT OF EXPOSURE BIOMARKERS
As discussed in Chapter 2, transcriptional profiling seeks to catalog, at the level of messenger RNA (mRNA), the changes in gene expression that are provoked by exposure to chemical and physical agents or to compare differences among tissue types, developmental stages, genetic variants, and so forth. The approaches most commonly rely on microarray technologies.
The use of mRNA molecules to report cellular exposure to toxicants is not new. For example, Northern blot analysis (see Chapter 2) demonstrated that genotoxic exposure (for example, to ultraviolet (UV) light) activates the expression of GADD (growth arrest and DNA damage) genes (Fornace et al. 1988). More recently, conventional analysis of the transcriptional function of the tumor suppressor protein p53 revealed a number of genes whose expression is activated by p53 in response to DNA damage (Lakin and Jackson 1999). Thus, changes in individual p53-regulated transcripts, such as the GADD genes or the gene encoding the cell cycle arrest protein p21, are now often used to report cellular exposure to DNA-damaging agents.
The transcriptomic approach extends information of this type to most or all of the expressed genome. In principle, transcriptomics could exploit the overall pattern of gene expression moving beyond changes in just a few genes to generate more specific signatures of exposure. Greater specificity in tying patterns of change in gene expression to a particular exposure in this way is an important potential advantage of transcriptional profiling and of toxicogenomic technologies in general. For example, whereas many DNA-damaging agents activate p21 expression, other genes respond to UV light, and still others re spond to X-rays (Lu and Lane 1993). To be used as biomarkers, altered levels of a given transcript need not be connected to a specific biologic end point, nor do the specific functions of all the mRNA molecules have to be known, although such information would be valuable.
Application to Human Exposure
Although applying such toxicogenomic technologies to determining human exposure lies in the future, experiments with model organisms or human cells in culture support the transcriptomic approach in its broad outline. In baker’s yeast Saccharomyces cerevisiae, exposures to an oxidant (t-butyl hydroperoxide), UV light, or a DNA-alkylating agent led to transcriptional profiles with both common and agent-specific components (Begley and Samson 2004). Similar approaches in mammalian cell lines (Dickinson et al. 2004; Newton et al. 2004; Kim et al. 2005) also gave signatures for different agents that could be distinguished by their transcriptional profiles (van Delft et al. 2005). Of course, such carefully controlled exposures, using genetically uniform cell populations, are a far cry from the highly variable and complex exposures that occur in the human population.
Unlike microorganisms or cell lines in culture, additional complexity in studying human responses results from the existence of multiple organs and tissue types. Not all tissues or cell types will respond equally to a given agent, nor will all tissues be equally exposed as a function of the organism’s exposure. This may constitute a significant challenge because of the practical limitations on which human tissues can be readily sampled. However, at least some modifications that are informative about exposure are likely to be available through analysis of DNA damage in circulating lymphocytes. For example, single-transcript studies showed that changes in CYP1A1 expression could be detected in the peripheral lymphocytes of railroad workers exposed to creosote, although no strong correlation with exposure could be established (Cosma et al. 1992). Additional complexity will come from genetic variation and perhaps individual life history that enhance the idiotypic nature of transcriptional responses in a human population. Nevertheless, the high level of information inherent to the transcriptomic approach may confer the ability to distinguish various types of exposure.
Applying transcriptional profiling to determine the dose of a given agent presents another level of challenge. All the processes affecting mRNA levels—transcription, processing, and turnover—are subject to thresholds for a response and maximum changes. Researchers often refer to the linear range for transcriptional (or other) responses, and the correlation of exposure and transcriptional induction is usually most accurate in this range (Baggerly et al. 2001; Larkin et al. 2005). Thus, for any given gene, the extent of the linear response in an individual basically determines the range of assignable exposures. This limitation may be partly overcome by toxicogenomic approaches because the overall pattern of gene expression can be more informative than the behavior of an individual gene. On the other hand, significant changes in biologic mechanism may occur with increasing dose, so that the toxicogenomic readout—in metabolomics, for example—is altered qualitatively as well as quantitatively (Keun et al. 2004).
Further issues concern the persistence of a detectable change after an exposure and whether changes in gene expression or other indicators can be detected after very brief exposures or after long-term exposures to very low levels of agents. Even large changes in gene or protein expression, or in metabolites, will eventually return to baseline levels with increasing time after an exposure. The high sensitivity of toxicogenomic approaches may extend the interval during which a brief or low-level exposure can be detected (exposures that are currently out of reach with conventional approaches), but the degree to which this is true will have to be determined by fundamental studies.
Whereas transcriptional profiling is a powerful method, proteins are generally the functional “business end” of genomic expression affecting cellular metabolism and regulation. Chemical exposures are manifested by two types of proteome changes. In the first, altered gene expression, mRNA stability, protein stability, or some combination of the three alters protein expression levels. In the second type of response, there is a change in the distribution of a protein between two or more modified forms. Such redistribution may result from physical modification of proteins by xenobiotic metabolites, directly or indirectly, or from effects of chemical exposure on the cellular systems that modify proteins (for example, protein kinases and phosphatases, protein cleaving activities, ubiquitylation).
New technologies enable changes in protein levels and protein modification to be assessed on a proteome-wide scale, with both types of changes referred to as proteomics. Analysis by two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) is one approach for detecting such changes. As of this writing, approximately a dozen studies describe the application of 2D-PAGE-based proteome analysis to identify protein biomarkers of chemical exposure. Xiao et al. (2003) examined the responses of a macrophage cell line exposed to diesel exhaust particles and identified expression changes for 32 proteins, which were linked functionally by their known roles in oxidative stress and antioxidant response networks (but see comment at the end of this paragraph). 2D-PAGE analysis of liver tissue has been used to assess proteome changes in animal models of chemical exposure. An earlier study noted dozens of proteome changes that distinguish a toxic dose of acetaminophen from an equivalent dose of a nontoxic compound of a closely related structure (Myers et al. 1995). Changes in protein expression levels were detected in rat liver after treatment with bromobenzene, but few of the protein species involved were identifiable (Heijne et al. 2003). A more recent study detected 45 2D-PAGE features that were differentially altered in rat liver treated with the carcinogen N-nitrosomorpholine (Fella et al. 2005). Of the proteins identified (for example, antioxidant enzymes, heat shock and chaperone proteins), several were already known as characteristic of stress responses to multiple chemicals, some of which were found in earlier studies. The utility of these approaches hinges on whether the observed changes represent generalized responses to stress versus truly chemical-exposure-specific changes. This issue will be resolved only with a much larger body of studies with diverse chemicals in relevant biologic models.
Although the examples described below are interesting, they represent only a few published studies describing proteomic approaches to identify biomarkers of exposure. A key question to be answered about proteomic approaches is whether agent-specific changes can be identified. Satisfactory resolution of this issue will be possible only after a much larger body of proteomic data on exposed cells, tissues, and biofluids is collected and analyzed.
The foregoing discussion shows the detail that can accrue in studying a model organism, but it is not expected that tissues such as liver will be routinely available for analysis in exposure studies of human populations. Monitoring changes in blood-borne cells such as macrophages seems more realistic. Nevertheless, it is also expected that some changes in gene expression and metabolism in “inaccessible” organs will indirectly produce altered profiles in “accessible” tissues and fluids, and these in turn may make valuable contributions to the development of exposure indicators.
A potentially useful development is the proteomic analysis of biofluids that can be obtained in a noninvasive manner to identify biomarkers of exposure. An advantage of this approach is that certain biofluids (for example, bile, urine, nasal or bronchoalveolar lavage) are in direct contact with target tissues of interest, and changes in the proteomes of these fluids may closely reflect the tissue changes that result from exposure. Moreover, protein profiles of these biofluids may be less complex than the proteomic expression of tissues, which could make it easier to detect biomarkers. Lindahl and colleagues used 2D-PAGE analyses of bronchoalveolar and nasal lavage fluids to identify biomarkers of exposure to cigarette smoke or the lung irritant dimethylbenzylamine (Lindahl et al. 1995 , 1999). These analyses identified changes in the levels of several proteins as well as changes in the distribution of apparently modified protein forms. Most of these protein changes could be explained on the basis of known functions in airway inflammation and protection. However, these studies also identified a novel protein, palate lung nasal epithelium clone, which is modulated by airway irritants and specifically binds lipopolysaccharides, thus suggesting a role in innate immunity in airways (Weston et al. 1999).
“Shotgun” Proteome Analysis
Shotgun proteome analyses (described in Chapter 2) present an attractive alternative to 2D-PAGE because this approach allows more proteins to be analyzed. For example, Welch et al. (2005) used this approach to analyze proteome changes in mouse liver. They identified and quantified more than 100 proteins that changed more than 2-fold upon treatment with acetaminophen, almost 10 times the number of protein changes observed with 2D gel analyses. In another example, shotgun proteome analysis revealed proteome changes in the bile of rats given two hepatotoxicants that differ in the severity of the injury they produce (Jones et al. 2003). Although the observed changes clearly represented toxic injury, follow-up studies can help determine whether such changes can lead to specific biomarkers of exposure.
Some chemical exposures directly modify proteins (for example, through the formation of adducts), while others perturb endogenous posttranslational modifications. Although much previous work on exposure markers focused on specific adducts in the accessible proteins in blood (Skipper et al. 1994; Ehrenberg et al. 1996), protein adducts have not been measured in proteomes more complex than blood. The main problem, even in relatively simple proteomes, is limited sensitivity for detection of adducted proteins in the presence of excess unmodified protein. Analyses of protein posttranslational modifications often use methods that enrich the modified protein or peptide forms in the samples—for example, with specific antibodies or by chromatography methods. Unfortunately, analogous affinity tools for enrichment of xenobiotic-modified proteins are not generally available. The placement of immunoreactive or other tags on proteins in cell model systems provides an alternative means of enriching samples for analysis. For example, the addition of such small tags to ubiquitin and small ubiquitin-related modifier (SUMO) proteins enabled the capture of cellular proteins to which ubiquitin or SUMO had become attached as a result of environmental and chemical stresses (Manza et al. 2004; Zhou et al. 2004; Kirkpatrick et al. 2005). Although this approach is not directly applicable to human populations, cell culture models can indicate which exposure-responsive stress-responsive proteome changes might be analyzed by other means.
Because diverse exposures sometimes produce similar changes in gene or protein expression, additional approaches may be needed to allow more precise identification of the exposure agent. As noted above, changes in gene and protein expression can alter metabolism in particular ways that can provide distinct signatures. Metabolomics involves measuring collections of small compounds in cells or biologic fluids, providing considerable biochemical detail in that the measured molecules include both metabolized products of environmental chemicals and endogenous metabolites. Endogenous metabolites may also be persistently changed by exposure, enabling chemical exposures to be detected for longer periods of time than the environmental compounds or their metabolites persist. The nuclear magnetic resonance- (NMR) and mass-spectrometry-based methods used for metabolomics in principle can specifically identify metabolites, but such identification may not be necessary if one is interested only in the use of metabolomic patterns as biomarkers.
Whereas combining information from metabolomics with other toxicogenomic data may turn out to be especially powerful for analyzing subtle or complex exposures (Griffin 2004), metabolomics is likely to offer several advantages over other toxicogenomic technologies (for example, transcriptomics and proteomics). Metabolomic analysis can be conducted on biofluids collected non-invasively (for example, urine, saliva), which would greatly facilitate sampling large populations of humans. Moreover, detection of continuing changes by metabolomic technologies may yield a timeline that allows extrapolation back to an estimate of the time of the exposure. Metabolomics may also have some advantages over proteomics because metabolism is often conserved across species—for example, the similarities and differences in the species-dependent pathophysiology and metabolomic profiles noted in mice and rats treated with hydrazine (Keun et al. 2004; Bollard et al. 2005).
Global metabolomic profiling might also be a simpler task than transcriptomic or proteomic profiling and thus more amenable to high-throughput screening. This was suggested by studies showing that the main yeast metabolome consists of fewer than 600 low-molecular-weight compounds (Oliver et al. 1998), but significant metabolomic analysis of exposure effects will be needed to address the possibility of high-throughput screening. Global metabolomic analysis in exposure studies in animals is just beginning and has not yet addressed the low-level exposures of interest in large populations. As noted above, metabolomic studies can readily be performed with noninvasive samples such as biofluids and breath condensate as well as on tissues in vivo. Profiles within a tissue or cell can be compared with profiles in biologic fluids or with cell secretion products to understand the metabolic consequences of xenobiotic-induced toxicity. Excellent examples of this approach are the studies by Waters and colleagues that used NMR and pattern recognition analysis to investigate time-related metabolic effects of α-naphthylisothiocyanate on liver, urine, and plasma in the rat (Waters et al. 2001 , 2002). One way this has been investigated is with high-resolution “magic angle spinning” NMR spectroscopy of intact tissue, which made it possible to link hepatic and renal histopathology to urinary and plasma metabolites (Garrod et al. 2001; Waters et al. 2001 , 2002 , 2005). Waters and colleagues found that hepatic lipidosis was associated with the increased urinary excretion of taurine and creatine. In addition, there was reduced urinary excretion of intermediates in the tricarboxylic acid cycle and increased excretion of plasma ketone bodies. These studies enabled a clearer understanding of key metabolic effects during development of and recovery from a toxic lesion. As with other approaches, however, development of this technology to generate biomarkers will require performing studies at subtoxic exposures—the low-level exposure of interest for assessing human exposures.
Larger scale work in metabolomics comes from the Consortium for Metabonomic Toxicology, which involves six pharmaceutical companies and the Imperial College of Science, Technology and Medicine, London, U.K. They are applying NMR-based metabolomic analysis of urine and blood serum for preclinical toxicologic screening and have completed studies of more than 80 candidate hepatic and renal toxicants (Lindon et al. 2003). This effort has yielded initial recommendations for standardization and reporting of metabolic analyses (Lindon et al. 2005a).
EVALUATION OF EXPOSURE IN RISK ASSESSMENT
The text to this point describes how toxicogenomic technologies may help improve exposure assessment, which is a critical element in risk assessment. However, it is possible that toxicogenomic technologies will be useful in risk assessment beyond directly improving exposure assessment. Because direct measurement of exposure to environmental contaminants (such as through biomonitoring of body fluids or personal or area monitors) is extremely rare, even in occupational settings, the U.S. Environmental Protection Agency (EPA) relies on exposure models that consider factors such as the quantities of water typically consumed in a day, respiratory rates, and activity patterns that would affect exposures. In the past, exposure analysis relied primarily on point estimates developed for illustrative populations (such as “reasonable worst-case exposures”). EPA now generally relies on probabilistic models (such as Monte Carlo methods) for exposure calculations that generate a distribution of population exposures. From this distribution, the risk assessor can choose a meaningful percentile value—such as the 90th, 95th, or 99th percentile—to assess risk to the population from a compound under evaluation.
Exposure assessment can also be modified by the availability of data on the mode of action (MOA) of toxicants, and toxicogenomic data may be able to inform this process. Typically, MOA approaches consider pharmacokinetic and pharmacodynamic variables that affect dose- and species-dependent responses. The emergence of physiologically based pharmacokinetic (PBPK) models has sometimes improved the understanding of dose as affected by kinetics (Meek et al. 2003a). For those chemicals whose toxicity is mediated by rate-controlled activation or detoxification processes, information provided by PBPK models has enabled exploration of the relevance of high-dose and route-of-administration specific animal toxicity responses to potential human risks. This type of MOA data refinement is well illustrated with the example of chloroform, a water disinfection by-product, which produced liver tumors in mice when administered as boluses in high-dose oral gavage cancer bioassays but not when administered in comparable doses in the animals’ drinking water. PBPK dosimetry assessments examining critical rates of chloroform metabolism to its toxic intermediate and subsequent detoxification demonstrated that the results of bolus gavage bioassays did not appropriately predict risks of chloroform carcinogenicity under conditions of long-term, low-level drinking water exposure. Information from these assessments resulted in significant modification of estimated human risks (Meek et al. 2003a).
PBPK models are not without their limitations. There can be considerable uncertainty using these models because their predictions depend on selection from the multiple modeling and parameter assumptions that are consistent with the data. Toxicogenomics may be able to improve such modeling by informing selection of assumptions and parameters. For example, the kinetic behavior and toxicity of chemicals often depend on metabolism, and toxicogenomic approaches can be used to rapidly identify potential key and rate-limiting enzyme targets whose activities, with further characterization, can be incorporated into PBPK models that better refine dose-dependent toxicity observations. Thus, toxicogenomic data provide mechanistic insights that can help refine the PBPK models used to predict target organ doses and hence provide a more accurate assessment of dose response.
Establishing Toxicogenomic Profiles
The toxicogenomic approaches are distinguished by their ability to reveal patterns of change involving many individual molecules. The resolving power of such patterns, when they can be recognized, will likely be much greater than that provided by individual molecules. The key scientific challenge is to identify reliable patterns (signatures) that report specific exposures and their intensities. There will be significant statistical challenges in establishing criteria for recognizing transcriptomic, proteomic, and metabolomic signatures of exposure. One challenge is that human exposure to most toxicants occurs at rather low levels and the resulting signals might be obscured by greater noise than occurs in experimental models. Another important challenge will be to develop the means to detect exposures at relatively long times after they occur so that long-term health effects might be assessed. Considerably more work is needed in transcriptomics, proteomics, and metabolomics to establish toxicogenomic signatures of exposure.
Experimental studies typically have focused on assessing exposure to a single agent (Yang 1994), but, in real life, humans are exposed to combinations of substances. For example, chlorinated drinking water contains hundreds of chlorinated organic compounds. The air we breathe, particularly in urban areas, contains a suite of toxic ingredients, including combustion products, oxides of nitrogen and sulfur, ozone, particulates, and more, in addition to living organisms (bacteria, fungi, spores). The latter will surely produce their own transcriptional, proteomic, and metabolomic signatures, which must be distinguished from those generated by chemical and physical agents. Beyond this, dietary and pharmaceutical components are likely to add to the signals and variation among individual humans.
The committee was able to identify only two studies that had generated microarray data to evaluate multicomponent exposures. Although these studies focused on the biologic effects of the agents, it can readily be seen that the changes in both gene expression and DNA modification might contribute to exposure assessment.
Working with MCF-7 (human breast carcinoma) cells in culture, Mahadevan et al. (2005) compared the effects of a standard reference particulate material (SRM 1649a) alone or in combination with two well-studied carcinogenic compounds, benzo[a]pyrene (BP) and dibenzo[a,l] pyrene (DBP), on gene expression, metabolic activation, and the formation of DNA adducts. Such polycyclic aromatic hydrocarbons (PAHs) are already known to be involved in activation, detoxification, and DNA repair, and thus they can alter genomic integrity. Researchers focused on the impact on two PAH metabolism genes (CYP1A1 and CYP1B1, encoding two cytochrome P450s), because the PAH compounds in the mixture are known to affect and be affected by these enzymes. Global analyses of the gene expression data revealed a clear additive induction of CYP1A1 and CYP1B1 upon cotreatment with SRM 1694a and BP as well as other effects. DBP had no effect alone or in combination with SRM 1649a. SRM 1649a decreased the total level of BP DNA adducts.
Bae et al. (2002) compared gene expression changes in human keratinocytes exposed to N-methyl-N'-nitro-N-nitrosoguanidine (a DNA-alkylating carcinogen), arsenic, or a metal mixture containing arsenic, cadmium, chromium, and lead. Arsenic alone induced DNA-protective genes in the exposed cells (consistent with an anticarcinogenic effect). However, of the DNA repair genes activated in cells treated with arsenic alone, only hNTH1 was induced in the cells treated with the arsenic-containing mixture. In fact, the metals mixture actually suppressed expression of four DNA repair transcripts. Two metallothionein genes showed increased expression in the mixture compared with arsenic alone, perhaps because the mixture contained cadmium (Bae et al. 2002).
Human Genetic Variation
Genetic diversity across human populations constitutes its own challenge in the application of toxicogenomic approaches to exposure assessment. A hint of this problem can be seen even in the study of individual gene expression. One study sought to test whether DNA sequence differences in the promoter regions of the metallothionein IIA gene could be associated with differences in the inducibility of this defense protein in response to metals (Wu et al. 2000). Instead, the individual variation in zinc inducibility was so great that it precluded definitive identification of effects due to genetic variation in the promoter. Presumably, the overshadowing differences in response lay elsewhere—in the regulatory proteins governing the response or the cellular uptake of zinc. Determining whether agent-specific profiles can be recognized with toxicogenomic approaches will require a considerable amount of human study, perhaps beginning with workplace exposures, which may be high enough to produce substantial toxicogenomic signals.
There is also the issue of epigenetic variation—changes in gene expression that are due not to DNA sequence differences but rather to alterations such as DNA cytosine methylation, changes in chromatin structure, or effects arising from differences in the genes of mitochondria or their expression. It is already clear that such effects can contribute to disease development in humans (notably cancers), but little information is available to relate such effects to environmental exposure. One intriguing study in mice (Waterland and Jirtle 2004) showed that early nutritional differences could alter imprinting and hence expression of a specific gene. It is not hard to imagine that epigenetic variation, like genetic sources of variation, could overshadow differences in response attributable to a chemical exposure.
One study evaluated exposure to metallic fumes among welders by assessing transcriptomic profiles in whole blood (total RNA extracted) before and after acute exposures to metal fumes and in nonexposed controls (Z. Wang et al. 2005). A self-controlled study design, which involved taking measurements before and after individual exposure, was used to overcome the problems of large interindividual variations compared with the small changes caused by environmental exposure. The group was stratified according to smoking status (which profoundly affected the whole blood expression profiles), and nonsmokers exhibited altered gene expression in 35 genes from eight functional pathways, in cluding processes related to oxidative stress, proinflammatory responses, phosphate metabolism, cell proliferation, and apoptosis.
Another study (Forrest et al. 2005) used microarrays to analyze transcriptomic profiles of blood lymphocytes of Chinese shoe factory workers chronically exposed to relatively high levels of benzene (mean = 47 parts per million) and matched nonexposed controls. This analysis identified 29 genes with highly altered expression in the benzene-exposed group. Real-time polymerase chain reaction confirmed changes in four of six selected genes. These transcripts may provide reliable molecular indicators of benzene exposure.
The ultimate goal of using toxicogenomic technologies as exposure indicators is to develop more sensitive, specific, and practically implementable exposure assessment methods. The application of toxicogenomic technologies to exposure assessment is new, and its use for human exposure assessment is in its infancy. A large gap exists between the current state of the art, with only a few studies to date having used gene expression analysis successfully to determine occupational exposure. Although high-sensitivity and data-rich toxicogenomic approaches may already be feasible for analyzing human exposure in some settings, applying these technologies to human populations exposed to low-level environmental contaminants will require considerably more development. Major challenges include the following:
- A lack of suitable information to define what constitutes a transcriptomic, proteomic, or metabolomic profile as a signature of specific toxicant exposure. Especially important with regard to “real-world” exposures is the question of sensitivity: what are the minimum exposures that such signatures can detect?
- Determining which accessible fluids and tissues display changes related to low-level exposures that will typify the human condition.
- A lack of information on how exposure to multiple agents affects toxicogenomic signatures and minimum thresholds that can be detected.
- A lack of information on how the time interval after an exposure affects the ability to recognize a toxicogenomic signature. Some long-term changes in gene expression may occur, but this has not been well studied.
These challenges should be addressed in research on the application of transcriptomic, proteomic, and metabolomic profiling to exposure assessment. In this context, the integration of information from multiple toxicogenomic approaches may provide considerably more analytical power than any one approach alone. On the question of sensitivity, a useful beginning would be a focus on occupational exposures, in which the toxicants are often known and the levels are higher than exposures to the population in general.
Use transcriptomic, proteomic, and metabolomic technologies to identify signatures of environmental exposures in target and surrogate tissues and fluids, primarily with animal models.
Begin testing complex mixtures for possible identification of distinct exposure signatures.
Examine the time course of chemical versus toxicogenomic signature persistence after initial chemical exposures.
To enable the further development of toxicogenomic measures of exposure, include transcriptomic, metabolomic, and/or proteomic analysis of samples in large human population studies and studies designed to assess exposures at toxicant levels commonly encountered in the workplace and certain communities1. This would be especially useful for chemicals whose toxicity across a range of exposures is well established. Use these studies to begin addressing issues of interindividual variability, background noise, confounding effects of combined exposures, the ability of toxicogenomic approaches to report exposures quantitatively, and determining ranges of quantifiable responses.
To enable the further development of toxicogenomic measures of exposure, include toxicogenomic analysis of samples in relevant case-control, cohort and panel studies that involve repeated measurements over time, as well as in clinical trials when possible and appropriate.
Use the information collected from studies to help develop and populate a database that will support further development of toxicogenomic exposure assessment.
Adapt and apply toxicogenomic measurements to assess exposure by developing signatures of exposure to single compounds and complex mixtures that can be used in animal and human population studies.
National Academies Press (US), Washington (DC)
National Research Council (US) Committee on Applications of Toxicogenomic Technologies to Predictive Toxicology. Applications of Toxicogenomic Technologies to Predictive Toxicology and Risk Assessment. Washington (DC): National Academies Press (US); 2007. 4, Application to Exposure Assessment.