Gene Expression Profiling in Ecotoxicology | SpringerLink

Ecotoxicology, 12, 475±483, 2003 # 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.

Gene Expression Profiling in Ecotoxicology TERRY W. SNELL,* SARA E. BROGDON AND MICHAEL B. MORGAN School of Biology, Georgia Institute of Technology, Atlanta, GA 30332-0230, USA Accepted 27 January 2003 Abstract. Gene expression profiling is a powerful new end point for ecotoxicology and a means for bringing the genomics revolution to this field. We review the usefulness of gene expression profiling as an end point in ecotoxicology and describe methods for applying this approach to non-model organisms. Since genomes contain thousands of genes representing hundreds of pathways, it is possible to identify toxicant-specific responses from this wide array of possibilities. Stressor-specific signatures in gene expression profiles can be used to diagnose which stressors are impacting populations in the field. Screening for stress-induced genes requires special techniques in organisms without sequenced genomes. These techniques include differential display polymerase chain reaction (DD PCR), suppressive subtractive hybridization PCR (SSH PCR), and representational difference analysis. Gene expression profiling in model organisms like yeast has identified hundreds of genes that are upregulated in response to various stressors, including several that are well characterized (e.g., hsp78, metallothionein, superoxide dismutase). Using consensus PCR primers from several animal sequences, it is possible to amplify some of these well characterized stress-induced genes from organisms of interest in ecotoxicology. We describe how several stress-induced genes can be grouped into cDNA arrays for rapidly screening samples. Keywords: gene expression; profiling; microarrays; toxicogenomics; biomarkers; corals Introduction Gene expression profiling in a wide variety of organisms has become technically feasible in the last decade (Gibson, 2002). The strategy of this approach in ecotoxicology is to compare gene expression patterns in two samplesÐone control and one exposed to toxicants. Toxicant exposure transiently alters gene expression profiles as cells regulate certain genes to protect cellular structures and repair damage (Causton et al., 2001). mRNA abundance increases from these toxicant-induced genes as transciption is modulated to protect against the stressor. Toxicant-induced genes are only a small portion of the genome and * To whom correspondence should be addressed. Tel.: (404) 894-8906; Fax: (404)894-0519; E-mail: [email protected]

define specific pathways activated by stress which are often unique signatures of the stressors. Enumeration of these gene-expression profiles could provide a key for interpreting stress responses in natural populations and for diagnosing specific causes of the stress response. Gene expression profiling, therefore, represents a powerful new end point for ecotoxicology and is a means for bringing the genomics revolution to this field. Progress in the molecular description of model organisms has produced knowledge that is generalizable to the non-model organisms of interest to ecotoxicology. Rapidly growing sequence databases are providing genetic tools for dissecting stress responses in many species, even those for which little sequence is available.

476 Snell et al. The current challenge for ecotoxicology is how to reap the benefits of this explosion of biological knowledge in a field where the technology evolves rapidly. As a first approach, it is possible to identify short-term goals to address questions likely to yield immediate benefits. These include: identification of conserved genes that are up-regulated in response to toxicant exposure, description of how these geneexpression profiles can be used to diagnose stressors, and determination of which genes are most informative to incorporate into stress gene arrays. The objective of this paper is to review the usefulness of gene expression profiling as an end point in ecotoxicology, to compare some of the techniques applicable to nonmodel organisms. Stress gene expression as end point in ecotoxicology Gene expression profile is indicative of stressor Organisms have evolved mechanisms to cope with natural stressors over million of years, so that stress responses are integral adaptations for the survival of most organisms (Hoffmann and Parsons, 1991). Wellcharacterized metabolic pathways exist for dealing with thermal, oxidative, and osmotic stress, natural toxins, and starvation (Sanders, 1993). These pathways are stressor-specific/selective and their activation represents a signature of the stressor impacting a population. For example, yeast grown at 25 C and shifted to 37 C respond by changes in expression of 854 heat-responsive genes, representing about 15% of the yeast genome (Causton et al., 2001). Many of these heat-induced genes like chaperonins are involved with protein folding and transport which are presumably needed to contend with the protein denaturation resulting from heat shock. The function of several of these genes has been described (Travers et al., 2000). However, despite the intensive study of heat-responsive genes in yeast, about 50% still have no defined function and yet some of these uncharacterized genes are highly expressed under stressed conditions (Wodicka et al., 1997). Many anthropogenic stressors also activate these pathways. Cadmium, for example, causes oxidative stress by changing intracellular glutathione levels. This causes a more than four-fold induction in the expression of about 50 genes in yeast, including the glutathione synthetase gene (GSH1), a rate limiting

enzyme of glutathione biosynthesis (Momose and Iwahashi, 2001). Several other enzymes involved in the sulfur amino acid metabolic pathway also are up-regulated, especially adenylylsulfate kinase MET14 and sulfhydralase MET17. These 50 upregulated genes represent less than 1% of the more than 6,000 genes in the yeast genome. Some of the up-regulated genes are generally responsive to stress like the heat shock genes HSP12 and HSP26. Others, like GSH1, seem to be toxicant specific. By profiling gene expression changes in response to cadmium exposure in controlled experiments, a library of expression patterns can be assembled that can be used to diagnose stressors when they are not known. Since genomes contain thousands of genes representing hundreds of pathways, it is possible to identify toxicant-specific responses from such a wide array of possibilities. It is therefore feasible to characterize patterns of stress-induced gene expression and correlate them to particular stressors. These expression profiles then could be used to diagnose the stressors impacting natural populations and prioritize their impact. This could be especially valuable in multiple stressor environments where toxicity may result from the cumulative effects of many stressors, each with many interactions. Diagnosing major stressors in natural populations A full scientific understanding of the ecological effects of toxicants on natural populations requires identification of the stressors having an impact, the metabolic pathways impaired, their modes of action, and their consequences for growth, survival, and reproduction. At a more practical level, stressor identification is essential for formulating successful management efforts, designing discharge and disposal permits to minimize stressor impacts, and to convince the public to support management actions. Recognizing the importance of stressor identification, EPA has produced a framework for collecting and analyzing data relevant for making stressor diagnoses (US EPA, 2000). Gene expression profile data can be particularly useful in stressor identification. Expression profiling of thousands of genes representing all metabolic pathways can pinpoint impairment of particular pathways. A causal relationship between pathway impairment and stressor often can be established because of decades of investigation into physiology and cell biology

Gene Expression Profiling 477

of model organisms. Since many stress responses are conserved across animal phyla, up-regulation of certain genes can implicate a stressor. For example, the CUP1 gene is up-regulated in yeast in response to copper and cadmium exposure (Tohyama et al., 1992). This gene codes for the metal-binding protein metallothionein which removes free metal ions from cells, preventing their binding to essential proteins and nucleic acids. Also activated are genes for exporting metals and sequestering them in cells. Detection of the expression of this suite of genes in a natural population would be strongly indicative of copper and/or cadmium exposure. A causal link could be established by spiking experiments at the environmental concentrations of copper and cadmium exposure and observation of induction of the same suite of genes. Gene expression profiling, therefore, allows for rigorous experimental designs amenable to hypothesis testing in the field and the confirmation of causation in manipulative experiments. Gene expression profiles also could be useful for defining thresholds for toxicant effects in natural populations. Small tissue samples of a few milligrams usually provide enough mRNA to hybridize to cDNA arrays. Ultrasensitive detection methods allow accurate quantitation of mRNA abundance. Samples could be taken repeatedly from the same individual over time so that a temporal profile of gene expression could be assembled. The response time of many gene expression changes is on the order of hours or days, so that spikes in gene expression should closely track stressor exposure. Measurements of gene expression changes coupled with chemical analyses of environmental exposures could quantify thresholds for stress-gene induction in natural populations. Because of the distinct signature of individual stressors on gene expression profiles, this approach has a good chance of separating out the effects of natural and anthropogenic stressors. Likewise, the cumulative effects of multiple stressors could be detected in the gene expression profile. Gene expression profiling promises to provide a large array of biomarkers for organisms that currently lack them. A broad suite of similar characters whose function is comparable across phylogenetically diverse species is useful for comparative studies in ecotoxicology. Such comparisons are likely to yield insight into which species are prone to succumb to stressor exposure and which are likely to persist.

Experimental approaches to gene expression profiling A major challenge for applying gene expression profiling to ecotoxicology is how to apply it to organisms without sequenced genomes and that are poorly defined genetically. Although these organisms are not as favorable experimental models as the classical model organisms, they have high ecotoxicological relevance and are worthy of investigation. The methods described below illustrate that it is technically feasible to develop gene expression profiles for nonmodel organisms and that the rewards for ecotoxicology are likely to be great. The ideal approach to gene expression profiling is to use full genome microarrays to identify genes up-regulated in response to toxicant stress. This requires fully sequenced genomes which currently exist for only a few model eukaryotic organisms like humans, Drosophila, Caenorhabditis elegans, and yeast. Applying this technology to the analysis of natural and anthropogenic stressors has progressed furthest using yeast where whole genome chips are widely available and relatively affordable (Gasch et al., 2000; Causton et al., 2001; Momose and Iwahashi, 2001). These studies provide insight into the whole genome stress responses of eukaryotes. Causton et al. (2001) used genome-wide expression analysis in yeast to explore how gene expression is remodeled in response to environmental stressors like temperature, oxidation, nutrients, pH, and osmolarity. They scored, 5,594 genes and found that 66% were significantly induced or repressed by at least threefold in response to at least one stressor. The response common to most environmental changes involved 499 genes (about 10% of genome), with 216 genes induced and 283 repressed. Of the 216 genes induced some classes were prominently represented. Many genes involved in glycolysis were up-regulated to increase energy production. Genes involved in trehalose synthesis were up-regulated to provide energy and protect cell membranes from denaturation. A variety of molecular chaperones and proteins involved in degradation were up-regulated to prevent accumulation of damaged proteins. Many induced genes were involved with defense against oxidizing agents, ion homeostasis, DNA repair, and metal transport and sequestration. However, more than half of the induced genes have no known cellular role. Changes in gene expression were transient for some

478 Snell et al. genes and long lasting for others, with elevated expression often being maintained as long as the stressor was present. The stress response was graded to the level of stress and additive for multiple stressors. Similar results for yeast were reported for different stressors by Gasch et al. (2000) and Momose and Iwahashi (2001). The generalization to emerge from these studies is that certain vital cellular processes are protected from stress-induced damage under any circumstances. These include energy generation and storage, protection against reactive O2 species, homeostasis of internal osmolytes, protein folding and turnover, and DNA repair. The stress response of yeast provides guidance in examining stress responses in non-model organisms lacking sequenced genomes. Screening for stressinduced genes in these organisms is possible, but it requires special techniques to identify stress-induced genes. Such techniques exist and are beginning to be applied in ecotoxicology. These include differential display polymerase chain reaction (DD PCR), suppressive subtractive hybridization PCR (SSH PCR), and representational difference analysis (RDA). The overall objective is to identify and isolate as many stress-induced genes as possible and to characterize their stressor specificity. Differential Display PCR The technique of DD PCR was first described by Liang and Pardee (1992) and Liang et al. (1994). These techniques are constantly being refined and modified as they are applied to the non-model organisms used in ecotoxicology (Rhodes and Van Beneden, 1996). Methods for DD PCR vary in detail, but they all begin with RNA extraction, purification to remove DNA, and reverse transcription to produce cDNAs. The cDNAs are then amplified by DD PCR using all combinations of several degenerate and anchored primers. A commercial kit (RNA Image) to perform DD PCR is available from GenHunter (Nashville, TN, USA), and is a good place to start, but a kit is not necessary for performing DD PCR. The PCR reactions are carried out using a thermocycler with the following conditions for amplification: 94 C for 30 s, followed by 40 C for 2 min, 72 C for 30 for 40 cycles, followed by 72 C for 5 min. Separation of cDNA fragments is accomplished by electrophoresis of samples on a 6% denaturing polyacrylamide gel. After electrophoresis, the gel is transferred to blotting

paper and dried under vacuum for 1 h at 80 C. Visualization of PCR products is possible by incorporating 33 P or 35S dATP during PCR and exposure of the radioactive cDNA fragments to autoradiographic film. The polyacrylamide gel permits separation of cDNA fragments in the size range of 100±600 base pair (bp) to be analyzed for each primer pair. Individual PCR reactions vary in the number of bands generated. To be considered a gene differentially up-regulated due to toxicant exposure, a band needs to be expressed in all replicates of the exposed group and absent in all control samples. Down-regulated genes need to be expressed in all control samples and absent in all exposed replicates. Typically, only cDNAs whose regulation changes by more than twofold are considered induced or repressed. Examples of DD PCR used in ecotoxicology include Rhodes and Van Beneden (1996) who identified genes associated with pollution-induced gonadal neoplasias in soft shell clams. Denslow et al. (1999, 2001a) isolated up-regulated vitellogenin and estrogen receptor genes from sheepshead minnows treated with 17-b-estradiol. In subsequent work (Denslow et al., 2001b), mRNA fingerprint analysis of approximately 18% of the total expressed mRNAs revealed 33 upregulated and 15 downregulated genes. Genes identified in this group included vitellogenin and two chorionic proteins ZP2 and ZP3. Sultan et al. (2000) used DD PCR to identify several genes of unknown function in the bivalve Donax trunculus that were up-regulated in response to coastal industrial pollution. Nine genes induced by insecticides, metals, and PAHs were identified in the reef-building coral Acropora cervicornis using DD PCR by Morgan et al. (2001) and Morgan and Snell (2002). Larkin et al. (2002) used DD PCR of sheepshead minnows exposed to 100 ng/l of 17-b-estradiol to isolate fifteen upregulated and seven down-regulated cDNA clones. Two up-regulated genes were identified, vitellogenin and the choriogenic protein ZP2, and one downregulated gene, transferrin, a protein involved in iron transport. Suppressive subtractive hybridization PCR A second approach to screening for stress-induced genes is SSH PCR. This technique was first described by Diatchenko et al. (1996) and Gurskaya et al. (1996) and has since been widely used in biomedical applications. In SSH PCR, differentially expressed


genes are identified by comparing two populations of mRNA, a control and one exposed to a stressor. First, RNA is extracted from tissues and both populations are reverse transcribed into cDNA. Tester (stressorexposed cDNA) and driver (control cDNA) are digested using the restriction enzyme RsaI to obtain shorter, blunt-ended fragments. The tester cDNA population is divided into two portions and each is ligated with a different adaptor to allow annealing of the PCR primer. Driver cDNA has no adaptors. The control driver cDNA is hybridized to tester cDNA from the stressor-exposed sample and the amplification of hybridizing sequences is suppressed in a PCR. This is because only molecules that have two different adaptors are amplified exponentially, thus selecting for differentially expressed sequences (Fig. 1). SSH PCR has the advantage of screening the entire genome for differentially expressed genes and identifying a small subset induced by specific stressors. One commercial kit is available from Clontech to perform SSH PCR starting with 1±2 mg of poly A mRNA, but others are now on the market. There are few examples of SSH PCR applied to environmental problems. It has been used to investigate diapause in Drosophila (Daibo et al., 2001), desiccation in fucoid algae (Pearson et al., 2001), and heat shock in the parasite Trichinella (Mak et al., 2001). At present, there appears to be only a single published application in ecotoxicology, a study of mercury exposure in the plant Arabidopsis (Heidenreich et al., 2001). Plant rosettes grown in tissue culture were exposed to 20±40 mM HgCl2. mRNAs were extracted and subjected to SSH PCR and 576 cDNA clones were isolated and screened for up-regulation by Northern hybridization. Thirty one of these cDNA clones were induced 1.5±10-fold by mercury, but for only seven clones was mercury the exclusive inducer. The induction of 14 clones was influenced by a diurnal rhythm and 10 were modulated by the day± night cycle. Functional analysis of the genes induced exclusively by mercury indicated some encoded proteins for the photosynthetic apparatus or the antioxidative system. Representational difference analysis A third approach to identifying stress-induced genes is RDA. This technique was originally developed to examine genomic differences (Lisitsyn and Wigler, 1993), modified and applied to mRNA expression by

Hubank and Schatz (1999), and optimized for gene expression studies by Pastorian et al. (2000). The basic procedure is similar to SSH PCR in that it requires reverse transcription followed by multiple rounds of digestion, ligation, amplification, hybridization, and visualization by electrophoresis. A detailed protocol can be found in Pastorian et al. (2000). The advantage of RDA is that cDNA fragments representing differentially expressed genes are sequentially enriched while gene fragments common to both populations are selectively eliminated. This approach has effectively isolated very rare differentially expressed transcripts using as little as 20 mg of total RNA as the starting material (Edman et al., 1997; Pastorian et al., 2000). As yet, there have been no applications in ecotoxicology, but this technique is beginning to be used in evolutionary studies. Zoldos et al. (2001) have used RDA to compare genomic differences among closely related oak species. PCR primers from consensus sequences of known stress-induced genes In addition to DD PCR, SSH PCR, and RDA, stressinduced genes can be identified using a bioinformatics approach. Gene expression profiling in model organisms like yeast has identified hundreds of genes that are up-regulated in response to stressors (Gasch et al., 2000; Causton et al., 2001), including several that are well known stress-induced genes (e.g. hsp78, metallothionein, superoxide dismutase). These studies have provided many candidate genes that could be explored for stress-induced expression in non-model organisms. These include genes for carbohydrate metabolism (hexokinase, fructose-2,6-bisphosphate, trehalose synthetase, glycerol-3-phosphate dehydrogenase), cellular redox reactions (cytochrome C peroxidase, superoxide dismutase, glutathione peroxidase, glutathione transferase), protein folding (chaperonins hsp104, hsp78, hsp26), protein degradation (polyubiquitin enzyme, ubitquitin conjugating enzyme, carboxypeptidase), DNA damage repair (methyladenine DNA glycosylase), and intracellular signaling (phosphodiesterase, protein kinase IKS1, phosphatidyl-inositol kinase). We have had some success designing PCR primers from consensus sequences of well characterized stressinduced genes from several animals. This approach has enabled us to amplify a metallothionein-like gene from the coral Acropora cervicornis. Conserved

480 Snell et al. Adaptor 1 Adaptor 2

T1

Ligation • After digestion with RsaI tester cDNA is

subdivided into two portions (T1 & T2), each ligated to a different adaptor. Driver cDNA has no adaptor. • A portion of tester cDNA with adaptor 1 is mixed with a portion of tester cDNA with adaptor 2. This is the unsubtracted tester control (UTC) - a negative control for subtraction.

UTC T2

T2

T1 a

First Hybridization

b

• Excess driver cDNA is added to each

c

• After heat denaturing and annealing,

sample of tester cDNA (T1 & T2). type a, b, c, and d molecules are formed. d

Second Hybridization • The two primary hybridization samples

a, b, c, d

+

(T1 & T2) are.

e

• New e type hybrids are formed (ds tester

molecules with different adaptor ends). • Fresh driver cDNA is added to enrich for

differentially expressed e sequences.

T1

T2

a b

Incubation

c

• Primer ends are filled in by a short incu

bation step. d e

b , 5 t 3

a, d

no amplification

b,

no amplification

PCR • Primers are added (

c

linear amplification

e

exponential amplification

) and all populations of molecules (T1, T2, & UTC) are subjected to PCR. • Differentially expressed molecules are amplified. • Only type e molecules, which have two different adaptors, can be amplified exponentially.

Figure 1. Outline of suppressive subtractive hybridization PCR.

regions of metallothionein genes were identified from GenBank sequences of several animals and PCR primers were designed to amplify fragments of 100±300 nucleotides. The forward primer was AGCCCTTGTAATTGCATTGA (aa 7±13, Table 1) and the reverse primer was CGAACAACTGGAGTCACATTTA (aa 67±73 Table 1). Total RNA was

extracted from coral fragments exposed to 50 mg/l copper. This RNA was reverse transcribed to cDNA which then was used as template in a PCR reaction with the consensus primers. We were able to amplify a A. cervicornis gene that was shown by sequence analysis to likely be a metallothionein (Table 2). Since metallothionein is a short protein (60 aa),


E-values from a BLASTx search (http://www.ncbi.nlm. nih.gov/BLAST/ ) are of limited value in identifying similarities to known proteins. However, three lines of evidence suggest that we have isolated a coral metallothionein-like gene. First, a position-specificinterative (PSI) BLAST search (http://www.ncbi.nlm. nih.gov/BLAST/ ) provides a protein profile based on common functional motifs and compares this to profiles of all known proteins. This search revealed that our coral sequence was related to 63 other metallothionein genes, mostly from aquatic animals. Second, an analysis using the program Pfam (http://pfam. wustl.edu, PF00131) identified metallothionein-like regions within our coral sequence consisting of conserved cysteines at positions 9, 11, 18, 20, 33, 52, 60, 72, and 82. Third, multiple sequence alignment with 15 well characterized metallothionein genes using the program Clustal (http://www.ebi.ac.uk/clustalw/ ) showed that the coral sequence possessed many Table 1. A. cervicornis metallothionein sequence for MT1. Gene expression was induced by exposure of coral to 50 mg Cu/l for 4 h Gene 50 ATCGGGAACATTAACGTAACTACTGGTGCGTCCCTTACAGCGTTTAAATTAGCTGGAGGAGCTTTATTCGTTGAAAAGCACGTTTCATGGCACGTCATTTACACTGAGGTCAACAAGCATTTACACTGAGGTCAACAAGC 30 Protein SPCNCIEIAPCNCIELHSVEKCPGRKNV P T F I L V R G I F A L C E P A T C K C K Y Q R S T KC DSSC

highly conserved residues in the same positions as metallothionein genes (Table 2). In particular, cysteines are abundant and conserved in metallothioneins. Four cysteines were conserved in all 16 of the sequences and four more were conserved in 90% of the sequences. Additional conservation at the 80% level was demonstrated with other residues using the program Consensus (http://www.bork.embl-heidelberg. de/Alignment/consensus.html). This approach is slow and tedious and is unlikely to work for all genes. However, when it does work, it permits incorporation of genes of known function into stress gene arrays. Incorporating stress-induced genes into cDNA arrays Once several stress-induced genes are identified from the organism of interest, they are grouped together for screening samples. The most efficient way to accomplish this is to spot these cDNAs onto glass slides or nylon membranes as an array. Microarrays were developed in mid-1990s (Schena et al., 1995) and have quickly become established as an essential tool for investigating gene expression (Gibson, 2002). These stress gene arrays as they are used in ecotoxicology are libraries of cDNAs representing expressed sequence tags (ESTs) of genes that are responsive to toxicant exposure. Arraying technology is available to produce high-density arrays with duplicate 100±200 mm spots spaced about 250 mm

Table 2. Multiple alignment of the A. cervicornis sequence with metallothionein sequences from 15 other animals (GenBank accession numbers). These species included: Crassostrea virginica (P23038), C. angulata (AAK15581), Mytilus edulis (P80251), Dreissena polymorpha (Q94550), Helix pomatia (P33187), Callinectes sapidus (P55949), Astacus astacus (P55951), Chaenocephalus aceratus (O93593), Parachaenichthys charcoti (O93450), Chionodraco rastrospinos (Q92145), Dicentrarchus labrax (Q9PTG9), Oreochromis mossabicus (P52756), Gadus morhua (P51902), Cyprinus carpio (O13269), Caenorhabditis elegans (NP 506482), A. cervicornis. Scale at bottom indicates amino acid number. Histogram shows percentage conservation at each amino acid site

482 Snell et al. apart. The EST cDNAs on the array are chemically denatured and hybridized with test cDNAs reverse transcribed from mRNAs from the organism of interest. These test cDNAs are labeled during the RT with either a radioiostope, fluorescent, or chemiluminescent tag. These tags allow the degree of hybridization to be quantified between the EST cDNAs on the array with the test cDNAs from the sample. mRNA abundance for each gene being expressed in the sample can be estimated by comparing the intensity of each spot relative to controls. Although technologies are changing rapidly, an illustration of arrays used to detect changes in gene expression in ecotoxicology has been provided by Larkin et al. (2002). Using DD PCR, they isolated cDNA clones from sheepshead minnow induced by 17-b-estradiol exposure. Several estrogen responsive genes including vitellogenin, vitelline envelope protein (ZP2) and the iron transport protein transferrin, along with 17 constitutive genes that served as controls were spotted as 100 nl spots onto a nylon membrane. Test cDNAs were labeled with [a-33P] dATP and hybridized to the array. Spot intensity was quantified using an image analysis system. Of the 54 cDNAs spotted onto the array, 15 were up-regulated by estradiol exposure, 7 were down-regulated, and 32 were unaffected. An analysis of variability in hybridization intensities among identical membranes showed that the measurements are highly reproducible. One of the challenges in analyzing microarray data is being able to compare data between experiments (Mutch et al., 2002). Another challenge is the assumption that signal intensities are linearly related to expression levels (Ramdas et al., 2001) Proper controls facilitate meeting these challenges by enabling researchers to normalize signals to standard samples, quantifying the sensitivity and specificity of hybridization reactions. Controls also permit quality control of array fabrication and validate labeling reactions and RNA quality. Controls need to have stable, unchanging expression and there are ready-to-spot oligos available from Stratagene (La Jolla, CA) and others that are convenient and reliable. Methods for analysis of microarray data are developing rapidly and it appears that standard statistical approaches will suffice. Mixed model ANOVA techniques have been described by (Kerr and Churchill, 2001). Sample sizes of 5±6 replicates are sufficient to demonstrate that a 1.5-fold change in transcript level is statistically significant (Gibson, 2002). Mapping

networks of genes that are coordinately regulated in response to toxicant exposure is possible using principal components analysis (Holter et al., 2000). Strengths and limitations of gene expression profiling Like any technique, gene expression profiling has strengths and limitations. Its strengths are that it is a fast and sensitive end point. More importantly, the up-regulation of specific genes provides information about which biochemical pathways are impacted by toxicant exposure and the likely mechanism of toxic action. Identifying the biochemical pathways impaired permits diagnosis of the likely stressors. This diagnosis of stressors is perhaps the most exciting result of the application of microarrays in ecotoxicology. The limitations of gene expression profiling are that it can be difficult to interpret array data. Continued improvements in statistical analysis are needed to improve the rigor of data handling. It is also difficult to relate changes in gene expression to whole organism responses that have ecotoxicological significance like survival, growth, and reproduction. Comparisons are needed so that changes in gene expression can be correlated to adverse effects. Acknowledgements We acknowledge with appreciation the support of the US Environmental Protection Agency, NCERQA grant R82-7105-010. We thank Melissa Hicks and Dr. Igor Zhulin for help with the metallothionein sequence analysis and Kristen Marhaver for technical assistance. References Berger, A., Mutch, D.M., German, J.B. and Roberts, M.A. (2002). Unraveling lipid metabolism with microarrays: effects of archidonate and docosahexaenoate acid on murine hepatic and hippocampal gene expression. Genome Biology 3: preprint 0004.1± 0004.53. Causton, H.C., Ren, B., Koh, S.S., Harbison, C.T., Kanin, E., Jennings, E.G., Lee, T.I., True, H.L., Lander, E.S. and Young, R.A., (2001). Remodeling of yeast genome expression in response to environmental changes. Mol. Biol. Cell. 12, 323±37. Daibo, S., Kimura, M.T. and Goto, S.G. (2001). Up-regulation of genes belonging to the drosomycin family in diapausing adults of Drosophila triauraria. Gene 278, 177±84. Denslow, N.D., Bowman, C.J., Robinson, G., Lee, H.S., Ferguson, R.J., Hemmer, M.J. and Folmer, L.C. (1999). Biomarkers of endocrine disruption at the mRNA level. In

Gene Expression Profiling 483 D.S. Henshel, M.C. Black and M.C. Harrass (eds) Environmental Toxicology and Risk Assessment: Standardization of Biomarkers of Endocrine Disruption and Environmental Assessment, Vol. 8, pp. 24±35. West Conshohocken, PA, USA: American Society for Testing and Materials. Denslow, N.D., Bowman, C.J., Ferguson, R.J., Lee, H.S., Hemmer, M.J. and Folmar, L.C. (2001a). Induction of gene expression in sheepshead minnows (Cyprinodon variegatus) treated with 17B-estradiol, diethylstilbestrol, or ethinylestradiol: the use of mRNA fingerprints as an indicator of gene regulation. Gen. Comp. Endocrinol. 121, 250±260. Denslow, N.D., Lee, H.S., Bowman, C.J., Hemmer, M.J. and Folmar, L.C. (2001b). Multiple responses in gene expression in fish treated with estrogen. Comp. Biochem. Physiol. B 129, 277±282. Diatchenko, L., Lau, Y., Campbell, F.C., Chenchil, A.P., Moqadam, A., Huang, F., Lukyanov, B., Lukyanov, S., Gurskaya, K., Sverdlov, N. and Siebert, E.D. (1996). Suppression subtractive hybridization: a method for generating differentially regulated or tissue-specific cDNA probes and libraries. Proc. Natl. Acad. Sci. USA 93, 6025±30. Edman, C.F., Prigent, S.A., Schipper, A. and Feramisco, J.R. (1997). Identification or ErbB3-stimulated genes using modified representational difference analysis. Biochem. J. 323, 113±18. Gasch, A.P., Kao, C.M., Huang, M., Carmel-Harel, O., Storz, G., Elledge, S.J., Botstein, D. and Brown, P.O. (2000). A common genomic expression program in the response of S. cerevisiae to diverse environmental stresses. Mol. Biol. Cell. 11, 4241±42. Gibson, G. (2002). Microarrays in ecology and evolution: a preview. Mol. Ecol. 11, 17±24. Gurskaya, N.G., Diatchenko, L., Chenchik, L., Siebert, P.D., Khaspekov, G.L., Lukyanov, G.L., Vagner, K.A., Ermolaeva, L.L., Lukyanov, O.D. and Sverdlov, E.D. (1996). Equalizing cDNA subtraction based on selective suppression of polymerase chain reaction: cloning of Jurkat cell transcripts induced by phytohemaglutinin and phorbol 12-myristate 13-acetate. Anal. Biochem. 240, 90±97. Heidenreich, B., Mayer, K., Sandermann, H. Jr and Ernst, D. (2001). Mercury-induced genes in Arabidopsis thaliana: identification of induced genes upon long-term mercuric ion exposure. Plant, Cell Environ. 24, 1227±34. Hoffmann, A.A. and Parsons, P.A. (1991). Evolutionary Genetics and Environmental Stress, 284 p. New York: Oxford University Press. Holter, N., Mitra, M., Maritan, A., Cieplak, M., Banavar, J. and Federoff, N. (2000). Fundamental patterns underlying gene expression profiles: simplicity from complexity. PNAS USA 97, 8409±14. Hubank, M. and Schatz, D.G. (1999). cDNA representational difference analysis: a sensitive and flexible method for identification of differentially expressed genes. Meth. Enzymol. 303, 325±49. Kerr, M.K. and Churchill, G. (2001). Statistical design and the analysis of gene expression microarray data. Genetical Res. 77, 123±28. Larkin, P., Folmar, L.C., Hemmer, M.J., Poston, A.J., Lee, H.S. and Denslow, N.D. (2002). Array technology as a tool to monitor exposure of fish to xenoestrogens. Mar. Environ. Res. 54, 395±9. Liang, P. and Pardee, A.B. (1992). Differential display of eukaryotic messenger RNA by means of the polymerase chain reaction. Science 257, 967±71.

Liang, P., Zhu, W., Zhang, X., Guo, Z., O'Connell, R.P., Averboukh, L., Wang, F. and Pardee, A.B. (1994). Differential display using one-base anchored oligo-dT primers. Nuc. Acids Res. 22, 5763±64. Lisitsyn, N. and Wigler, M. (1993). Cloning the differences between two complex genomes. Science 259, 946±51. Mak, C.H., Sun, K.W. and Ko, R.C. (2001). Identification of some heat-induced genes of Trichinella spiralis. Parisitology 123, 293±300. Momose, Y. and Iwahashi, H. (2001). Bioassay of cadmium using DNA microarray: genome-wide expression patterns of Saccharomyces cerevisiae response to cadmium. Environ. Toxicol. Chem. 20, 2353±60. Morgan, M., Vogelien, D.L. and Snell, T.W. (2001). Assessing coral stress responses using molecular biomarkers of gene transcription. Environ. Toxicol. Chem. 20, 537±43. Morgan, M.B. and Snell, T.W. (2002). Characterizing stress gene expression in reef-building corals exposed to the mosquitoside dibrom.. Mar Pollut. Bull. 44, 1206±18. Pastorian, K., Hawel, L. and Byus, C.V. (2000). Optimization of cDNA representational difference analysis for the identification of differentially expressed mRNAs. Anal. Biochem. 283, 89±98. Pearson, G., Serrao, E.A. and Cancela, M.L. (2001). Suppression subtractive hybridization for studying gene expression during aerial exposure and desiccation in fucoid algae. Europ. J. Phycol. 36, 359±66. Ramdas, L., Coombes, K.R., Baggerly, K., Abruzzo, L., Highsmith, W.E., Krogmann, T., Hamilton, S.R. and Zhang, W. (2001). Sources of nonlinearity in cDNA microarray expression measurements. Genome Biol. 2(11), 1±7. Rhodes, L.E. and Van Beneden, R.J. (1996). Gene expression analysis in aquatic animals using differential display polymerase chain reaction. In G.K. Ostrander (ed.) Techniques in Aquatic Toxicology, pp. 161±83. Boca Raton, FL: CRC Lewis Publishers. Sanders, B.M. (1993). Stress proteins in aquatic organisms: an environmental perspective. Crit. Rev. Toxicol. 23, 49±75. Schena, M., Salon, D., Davis, R.W. and Brown, P.O. (1995). Quantitative monitoring of gene expression patterns with a cDNA microarray. Science 270, 467±70. Sultan, A., Abelson, A., Bresler, V., Fishelson, L. and Mokady, O. (2000). Biomonitoring marine environmental quality at the level of gene expressionÐtesting the feasibility of a new approach. Water Sci. Tech. 42, 269±74. Tohyama, H., Tomoyasu, T., Inoue, M., Joho, M. and Murayama, T. (1992). The gene for cadmium metallothionein from cadmiumresistant yeast appears to be identical to CUP1 in copper-resistant strain. Curr. Gen. 21, 275±80. Travers, K.J., Patil, C.K., Wodicka, L., Lockhart, D.J., Weissman, J.S. and Walter, P. (2000). Functional and genomic analyses reveal an essential coordination between the unfolded protein response and ER-associated degradation. Cell 101, 249±58. US EPA. (2000). Stressor Identification Guidance Document, Washington, DC: Office fo Water. EPA-B-025. Wodicka, L., Dong, H., Mittmann, M., Ho, M. and Lockhart, D.J. (1997). Genome-wide expression monitoring in Saccharomyces cerevisiae. Nat. Biotech. 15, 1359±67. Zoldos, V., Siljak-Yakovlev, S., Papes, D., Sarr, A. and Panaud, O. (2001). Representational difference analysis reveals genomic differences between Quercus robur and Q. suber: implications for the study of genome evolution in the genus Quercus. Mol. Gen. Genomics 265, 234±41.