Using a customized DNA microarray for expression profiling of the estrogen-responsive genes to evaluate estrogen activity among natural estrogens and industrial chemicals.

We developed a DNA microarray to evaluate the estrogen activity of natural estrogens and industrial chemicals. Using MCF-7 cells, we conducted a comprehensive analysis of estrogen-responsive genes among approximately 20,000 human genes. On the basis of reproducible and reliable responses of the genes to estrogen, we selected 172 genes to be used for developing a customized DNA microarray. Using this DNA microarray, we examined estrogen activity among natural estrogens (17beta-estradiol, estriol, estrone, genistein), industrial chemicals (diethylstilbestrol, bisphenol A, nonylphenol, methoxychlor), and dioxin. We obtained results identical to those for other bioassays that are used for detecting estrogen activity. On the basis of statistical correlations analysis, these bioassays have shown more sensitivity for dioxin and methoxychlor.

other interactions is high, estrogen activity can be masked or disguised by the second pathway. Furthermore, the major estrogen activity is not conducted by a unique pathway. First, there are at least two types of ERs, ER-α (Green et al. 1986) and ER-β (Kuiper et al. 1996), which differ in their affinity for ligands and the way in which they transduce signals (Katzenellenbogen and Katzenellenbogen 2000). Differences in affinity between ER-α and ER-β were reported for methoxychlor and its analog DDT (Jacobs et al. 2003). As for the ER-α, tamoxifen is an antagonist against natural estrogen but has agonist activity in the uterus, whereas ICI 182,780, a well-known pure antagonist, does not show such activity (Branham et al. 1996). This difference can be explained by the difference in the liganddependent or -independent activation functions assisted by coactivators and has been observed in other chemicals [reviewed by ; McKenna and O'Malley (2002)]. This indicates that, even for the first pathway, using any one of these signaling pathways as an indicator of estrogen activity would be biased and specific signals could be enhanced, resulting in differences between the expected and real biological outcomes.
The second pathway is more complex. It may include various metabolic and modification pathways for chemicals, and estrogen activity could be higher or lower than the original, depending on the products (Beresford et al. 2000). Methoxychlor, for example, is metabolized to mono-and bisphenolic forms by oxygenase (Bulger et al. 1978) or by cytochrome P450 isoforms (Hu and Kupfer 2002). These metabolites have more estrogen activity than methoxychlor. Such a metabolic activation of estrogenic chemicals was also reported for bisphenol A and bisphenol B (Yoshihara et al. 2001), 2-nitrofluorene , and styrenes . Metabolic inactivation or inactivation by modification could also occur in many chemicals. As estrogen activity results in growth and proliferation of the cell through the activity of transducing signals by means of hormones, growth factors, cytokines, and others, monitoring estrogen activity at the steps close to such cellular responses rather than at the beginning (receptor binding, for example) is crucial for reliable evaluation of estrogenicity.
Previously we found that a significant number of genes responded to estrogen in a DNA microarray analysis and we characterized some of them, including solute carrier family 7, member 5 (SLC7A5), retinoblastoma-binding protein 8 (RBBP8), and c-myc promoter-binding protein 1 (IRLB) (Inoue et al. 2002b). We also found that many of these genes responded to estrogen in a manner similar to that in cancer cells from the breast, ovary, stomach, kidney, and other sites. Here, using a customized DNA microarray with newly selected estrogen-responsive genes, we outline an experimental system with more sensitivity for evaluation of estrogen activity in natural and industrial chemicals on the basis of statistical analysis of gene response. Our goal is to establish an experimental system with more sensitivity for the evaluation of estrogen activity in these chemicals, which can be applied even to those having low activity.

cDNA Microarray Analysis
GeneChip analysis was conducted using human U95A oligonucleotide probe arrays (Affymetrix, Santa Clara, CA, USA) according to the supplier's protocols, as follows. Total RNA (1 µg) was used to generate a cRNA probe by T7-transcription. The fragmented cRNA (10 µg) was hybridized to the microarrays in 200 µL of a hybridization cocktail by incubation at 45°C for 16 hr in a rotisserie oven set at 60 rpm. The microarrays were then washed with a nonstringent wash buffer [6× NaCl/NaH 2 PO 4 /EDTA (SSPE)] at 25°C, followed by a stringent wash buffer [100 mM MES (pH 6.7), 0.1 M NaCl, and 0.01% Tween 20] at 50°C. The microarrays were stained with streptavidin phycoerythrin (Molecular Probes, Eugene, OR, USA), washed again with 6× SSPE, stained with biotinylated antistreptavidin IgG followed by streptavidin phycoerythrin, and washed a third time with 6× SSPE. The arrays were scanned using a GeneArray scanner (Affymetrix) at a resolution of 3 µm, and the scanned image was quantitatively analyzed with Microarray Suite 4.0 (Affymetrix). For normalizing the data to compare mRNA expression levels among samples, we unified the values to 1,000 as an average of average difference scores corresponding to the signal intensities of all probe sets in each sample.
A custom cDNA microarray (EstrArray) was manufactured by InfoGenes Co., Ltd. (Tsukuba, Japan) by mechanical spotting of cDNA (∼500 bp to ∼1.5 kb) of the genes selected from the above DNA microarray assays [see Inoue et al. (2002b) for details]. The analysis using EstrArrays was performed as follows: After the cells were cultured for 72 hr in the presence of chemicals at indicated concentrations, mRNA was purified using the PolyATract System 1000 (Promega, Madison, WI, USA) according to manufacturer instructions. The quality of mRNA was confirmed by examining the optical density and also by reverse transcription-polymerase chain reaction (RT-PCR) assay for several marker genes (β-actin for all, and pS2 and ER-α for the chemicals with high estrogen activity). Each mRNA was labeled with fluorescent Cyanine 3 (Cy3)-dUTP (for the treatment of chemicals) or Cy5-dUTP (for the control) at 37°C for 1.5 hr using SuperScript II (Invitrogen, Carlsbad, CA, USA) and random primers (a mixture of 6 mers and 9 mers). Both Cy3-and Cy5-labeled probes were mixed and denatured under alkaline conditions for 1 hr. After free fluorescent nucleotides were removed using Microcon-30 columns (Millipore, Bedford, MA, USA), probes were hybridized to EstrArrays for 16 hr in 5× NaCl/Na citrate (SSC) and 0.5% sodium dodecyl sulfate at 65°C. After hybridization, slides were washed twice with 0.05× SSC for 5 min at room temperature. The fluorescent intensities were scanned with a ChipReader (Virtek, Waterloo, Ontario, Canada), and scanned images were analyzed using IPLab (Scanalytics, Fairfax, VA, USA) according to manufacturer instructions. The ratio (Cy3/Cy5) was calculated for each spot, and after transforming the ratio into a logarithmic value (log 2 ), the value was normalized using internal control genes. Clustering analysis was performed using the Cluster program and the results were displayed with the TreeView program [for both programs see Eisen et al. (1998)]. The genes spotted on EstrArray or GenBank accession numbers (http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?db=nucleotide)] were as follows: SuperScript II (Invitrogen). Quantitative PCR was carried out using a LightCycler-FastStart DNA Master SYBR Green I kit (Roche Molecular Biochemicals, Mannheim, Germany). The PCR conditions were as follows: denaturation at 95°C for 1 min, followed by 40 cycles of denaturation at 94°C for 10 sec, annealing at 57°C for 5 sec, and extension at 72°C for 20 sec. After PCR a melting curve was constructed by increasing the temperature from 72 to 95°C. The product was resolved in agarose gels to ensure that the correct product was amplified in the reaction. PCR was repeated 3 times for each gene, and the average and standard deviations were calculated. The PCR primers were as follows: SLC7A11, 5´-ACAGTG CCAGAGTGAAGAAACTC-3´ and

Results
We first screened the estrogen-responsive genes in a human mammary tumor cell line, MCF-7, using two different comprehensive DNA microarray systems, UniGem, version 2 (IncyteGenomics) containing 9,182 genes and GeneChip U95A (Affymetrix) containing 12,625 genes ( Figure 1). Approximately 300 genes in UniGem, and 850 genes in GeneChip U95A showed a response higher than 2-fold and 3-fold, respectively. To examine the response to estrogen by monitoring transcription of the genes, we selected 172 genes after the reproducibility of their upregulation or downregulation on estrogen treatment (10 nM E 2 for 3 days) was confirmed by repeated DNA microarray and/or RT-PCR analyses (Inoue et al. 2002b; also, data not shown). To confirm that the data obtained were reliable for the genes with various expression levels, we arbitrarily divided the genes into high-and low-expression types. The genes categorized as the high expression type characterized by abundant transcript are summarized in Table 1. These genes had transcripts with expression levels higher than those of the solute carrier family gene 2, member 1 (SLC2A1) and the keratin 6B gene in each DNA microarray analysis (both appeared in both DNA microarrays and showed identical expression levels), and included the genes for amino acid transporters, and structural, ion-related, translation-, transcription-, and cell cycle-associated proteins. The expression of most of the tRNA synthetase genes and genes for the TATAbox binding protein-associated factor and histone deacetylase was probably upregulated for enhancing protein synthesis or transcription, respectively. Meanwhile, the genes associated with specific tissues, such as those for the nervous system, showed downregulation. A similar analysis was performed for the genes categorized as the low expression type ( Table 2). Upregulation of the genes related to various synthetases, transcription-related, and cell cycle or growth-associated proteins as well as receptors and ion or amino acid transporters was also prominent in this type. Among the tumor-associated genes, oncogenic genes such as for c-fos, AML (acute myeloid leukemia) 1b, FOS-like antigen 2, and a v-jun homolog were upregulated, whereas tumor suppressor-related genes (absent in melanoma 1) (Ray et al. 1996) were slightly downregulated. Expression of the ER-α gene was downregulated as observed for progressive breast tumors (Lapidus et al. 1998;Yoshida et al. 2000).
On the basis of the information obtained from the estrogen-responsive genes shown above, we constructed a customized DNA microarray, EstrArray, that contains 203 genes, including genes showing either upregulation (108 genes) or downregulation (64 genes) in their expression. EstrArray also contains calibration markers for adjusting the fluorescent levels between Cy3-and Cy5-labeled cDNAs (28 genes) and expression markers such as the genes for trefoil factor, the ER-α and ER-β, steroid sulfatase, and other estrogenrelated proteins (14 genes, 11 showing estrogen responsiveness, resulting in a total of 203 genes).

Customized DNA Microarray
DNA microarray technology is one of the most potentially powerful tools in modern toxicogenomics because it can shorten the time for elucidating toxicological phenotypes and widen the way for drug discovery (Inoue 2003). However, determining the relationship between specific gene expression profiles and toxicological phenotypes will be accelerated by the development of customized DNA microarrays, the accumulation of profiles specific to chemicals, and an increase in the knowledge of gene functions (Adachi et al. 2002;Inoue et al. 2002a;Watanabe et al. 2002;Wong et al. 2003).
Here we developed a customized DNA microarray, EstrArray, for expression profiling of estrogen-responsive genes. EstrArray contains 172 estrogen-responsive genes  Figure 2. Clustering of gene expression after the treatment of various estrogens and industrial chemicals examined using EstrArrays. E 2 -2, E 2 twice. Gene expression profiles were obtained after treatment with 10 nM of E 2 , estrone, estriol, and DES, 10 µM nonylphenol, bisphenol A, genistein, and methoxychlor, or 50 µg/mL dioxin. The results of EstrArray analysis are shown as values of log 2 (fluorescent intensity for chemical plus/fluorescent intensity for chemical minus), which were colored according to the color scale.
selected from approximately 20,000 genes, almost half the estimated number in the whole human genome. As approximately 95% of the genes examined did not respond to estrogen or were not expressed in MCF-7 cells, the genes used for EstrArray were considered to represent the genes most suitable for monitoring estrogen responsiveness. As we reported previously, some of these genes were characterized extensively to show reproducible estrogen responsiveness by Northern blot analysis (Inoue et al. 2002b) and to examine their potential functions (data not shown). EstrArray also contains marker genes for the calibration of fluorescent levels that cover a wide range of expression levels for normalizing signals between the presence and absence of chemicals. The genes, which show estrogen responsiveness, can be classified into several types according to their function (Tables 1 and 2; summarized in Figure 5). Among the genes related to tumor-associated genes, oncogenes and tumor-promoting genes are generally upregulated, whereas the genes related to tumor suppression and the ER-α gene are downregulated. This is consistent with the effects of estrogen, namely, the promotion of tumorigenesis. For growth-and ionassociated genes and other genes, the expression of various transporters, synthetases, transcription factors, growth response genes, and structural genes was upregulated, indicating enhancement of growth and proliferation of the cell. Meanwhile, the genes related to specific differentiation of the cell, such as those for neuronal proteins, were downregulated.

Genes Responding to Estrogenic Chemicals
Among the estrogen-responsive genes used for EstrArray, the AR and AGTR1 were examined in detail ( Figure 4). Both showed a relatively high response to E 2 (5.4-fold increase for AR and 2.8-fold decrease for the AGTR1) and a similar tendency of response to the other chemicals examined here. Estrogen responsiveness was low for estriol, estrone, and DES compared with E 2 when they were examined at the concentration of 10 nM, except for the AGTR1 with estriol. These data and the result of the statistical correlation study (Figure 3) indicate that the genes responded to most chemicals analyzed here in similar ways and suggest that these genes commonly respond to estrogen activity. The difference in the degree of response for each gene, however, might be due to the difference in biological effects originating from structural differences. This difference is particularly important for the evaluation of estrogen activity,  Figure 3. Estrogenicity of chemicals examined using EstrArrays. E 2 -2, E 2 twice. Gene expression profiles of estrogen-responsive genes were compared between the independent E 2 treatments and shown in a scatterplot graph (A). The same comparison was performed for the genes of the (B) low-expression or (C) highexpression types. Gene expression profiles were compared between (D) E 2 and estriol, (E) estrone, (F) DES, (G) genistein, (H) nonylphenol, (I) bisphenol A, (J) methoxychlor, and (K) dioxin. The axes are shown in log 2 (fluorescent intensity for chemical plus/fluorescent intensity for chemical minus) calculated for each chemical. The correlation coefficient (R) between two profiles was calculated for each graph on the basis of linear regression between the two profiles. CYP1A1, a dioxin marker, is indicated in K.
especially when the activity is low, giving an advantage to this assay (discussed below). The functional relationship of these genes to estrogen signaling is mostly unknown. AR is an epidermal growth factor and is expressed in invasive mammary tumors together with its receptor, forming a potential autocrine loop for tumor progression (Ma et al. 2001). AR is also a target gene for vitamin D3 (Akutsu et al. 2001) and progesterone (Das et al. 1995) and may go through the ErbB pathway for oncogenic activity by inhibiting apoptosis (Hurbin et al. 2002). Therefore, activation of the AR gene may well explain the progression of estrogen-independent breast cancer. The AGTR1 is a type 1-angiotensin II receptor whose expression is downregulated by estrogen in several tissues. This explains the estrogen deficiency in hypertension and other diseases (Krishnamurthi et al. 1999;Nickenig et al. 1998), although the explanation at the molecular signaling level is not so clear. The pathways common to the epidermal growth factor receptor or the insulin-like growth factor could be potential signaling mechanisms (Touyz and Berry 2002).

Evaluating Estrogenicity with EstrArray
The chemicals used here have estrogen activity in reporter gene assays (Demirpence et al. 1993;Gaido et al. 1999;Inoue et al. 2002a;Pons et al. 1990) and cell proliferation/uterotrophic assays [reviewed by Kanno et al. (2003)] and upregulate estrogen target genes in responsive cells (Nagel et al. 2001;Vivacqua et al. 2003). Dioxin does not have estrogen agonist activity (Astroff and Safe 1988;Spink et al. 1990). Cluster analysis shown in Figure 2 clearly demonstrated similar expression profiles among estrogenic chemicals, E 2 , estriol, estrone, genistein, nonylphenol, and DES. Note that the data were obtained for 10 µM in the case of genistein, nonylphenol, and bisphenol A, whereas a concentration of 10 nM was used for the others. Bisphenol A at 10 µM showed less of a tendency to enhance the gene response, although it may show a higher tendency when examined at a higher concentration. Methoxychlor at 10 µM showed an even lower response but showed a meaningful correlation with the profile for E 2 . Dioxin, as expected, was classified as the most distant chemical in the clustering here.
The evaluation of the estrogenicity of chemicals used here is unique. First, the estrogenicity of chemicals was compared as expression profiles of estrogen-responsive genes, giving multiple scales provided by the expression of each gene used here Toxicogenomics | Evaluating estrogen activity using a customized DNA microarray Environmental Health Perspectives • VOLUME 112 |   Figure 4. Expression profiles of the genes showing upregulation or downregulation in response to estrogen and estrogenic chemicals. (A) Responses to various chemicals analyzed using EstrArrays. The vertical axis marked as log 2 (C+/C-) indicates log 2 (fluorescent intensity for chemical plus/fluorescent intensity for chemical minus) calculated for each chemical. (B) The response to E 2 examined by real-time quantitative RT-PCR. The assays were repeated 3 times and the average and the SD (bracketed) in the log 2 values are shown. The genes examined are SLC7A11 (solute carrier family 7, member 11), EGR3 (early growth response 3), PDZK1 (PDZ domain-containing protein), S100P (S100 calcium-binding protein P), AR (amphiregulin), WARS (tryptophanyl-tRNA synthetase), SELENBP1 (selenium binding protein 1), ENO2 (enolase 2), ARHGDIA (Rho GDP dissociation inhibitor alpha), AGTR1 (angiotensin II receptor type 1), IGFBP5 (insulin-like growth factor binding protein 5), and SLC12A2 (solute carrier family 12, member 2). compared with the ligand-binding method and reporter gene assays. This is even advantageous when the estrogenicity of chemicals is low, as multiple scales can give statistically significant evaluations. The estrogenicity of methoxychlor was not detected clearly by some assays (Shelby et al. 1996), but here it showed a distinct tendency. Second, the estrogenicity shown here is based on biological effects because not only the target genes of estrogen/estrogen receptor complex but also the genes that are presumably located downstream of the estrogen signaling pathway were included (Inoue et al., in press; Rho et al. in press). Third, with more information, the data can be classified according to the tendency of response among chemicals, specific to steroids, phenol, and phthalate, for example, which are expected to have different effects on the genes. To apply DNA microarray data for the evaluation of estrogen activity among various compounds, we are now constructing a database consisting of DNA microarray data of genes, chemicals, and cells. VOLUME 112 | NUMBER 7 | May 2004 • Environmental Health Perspectives

Neuronal proteins
Integrins Enolase, synaptogyrin Catenin, protocadherin, cadherin Figure 5. The genes responding to estrogen. The genes were categorized into tumor-, growth-and ionassociated genes and other genes including those for structural and neuronal proteins. Upregulation and downregulation are indicated by arrows on the left side.