DNA Arrays to Monitor Gene Expression in Rat Blood and Uterus following 17ß-Estradiol Exposure: Biomonitoring Environmental Effects Using Surrogate Tissues

John C. Rockett1, Robert J. Kavlock, Christy R. Lambright, Louise G. Parks, Judith E. Schmid, Vickie S. Wilson, Carmen Wood and David J. Dix

Reproductive Toxicology Division, National Health and Environmental Effects Research Laboratory, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711

Received October 25, 2001; accepted April 30, 2002


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
We propose that gene expression changes in accessible tissues such as blood often reflect those in inaccessible tissues, thus offering a convenient biomonitoring method to provide insight into the effects of environmental toxicants on such tissues. In this pilot study, gene expression changes in peripheral blood leukocytes (PBL) were compared to those in the uteri of adult rats to identify genes that were altered in both tissues following estradiol treatment. Ovariectomized rats were treated with either 17ß-estradiol or vehicle control (corn oil) for 3 days. PBL and uterine RNAs were hybridized to arrays containing 1185 genes. One hundred and ninety three genes were expressed in common between the PBL and uterus. Eighteen were changed significantly in both tissues, 9 of which were treatment- but not tissue-specific (e.g., jun-D, phospholipase A2, thymidine kinase). These results demonstrate that many genes are coexpressed between PBL and uterus, and that some are coregulated by estradiol. Given the limited number of genes examined in this study and the estimated size of other mammalian genomes, we conclude that many more genes will also be coregulated and suggest that accessible tissues such as PBL can serve as surrogate tissues for observing gene expression changes in inaccessible target tissues.

Key Words: uterus; blood; rat; microarray; gene expression; estradiol; endocrine disrupting chemical; biomonitoring; surrogate tissue.


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The last 10 years or so have produced a body of evidence indicating that many environmental xenobiotics are having hormone-like effects on wildlife. Such chemicals have been termed Endocrine Disrupting Chemicals (EDCs). Although the relationship between human disease and exposure to such chemicals is unknown, it is believed that EDCs present a potential threat to public health. Compared to adults, children are seen as being at increased risk from such environmental toxicants because of vulnerable developing systems and enhanced exposure. However, a major obstacle to monitoring the impact of EDCs on juvenile reproductive development is that the developing organs cannot be easily assessed due to their inaccessible and incomplete nature.

Although epidemiological and nutritional data can provide some measure of exposure to certain EDCs, a more accurate representation is provided by biological samples. This idea has been used in developing the concept of biomonitoring. In its National Report on Human Exposure to Environmental Chemicals, the National Center for Environmental Health defines biomonitoring as "assessment of human exposure to chemicals by measuring the chemicals or their metabolites in human specimens, such as blood or urine" (www.cdc.gov/nceh/dls/report). This definition clearly limits the biomonitoring approach to measuring levels of exposure. Although such measurements can be used to assess the general risk of a toxic outcome, cultural and genetic factors make prognostic evaluations of possible toxic effects in specific individuals or populations highly speculative.

The challenge therefore exists to use the biomonitoring approach to create a stronger and more defined linkage between exposure and effects. This will require the development of assays that (1) can be applied to easily accessible biological samples, (2) can offer an accurate indication of whether growth and/or development is proceeding in a normal manner, and (3) can provide a detailed record for future reference. One way to do this might be to generate quantitative data on gene expression. This approach offers the opportunity not only to indicate or confirm exposure to specific chemical classes (Bartosiewicz et al., 2001Go; Waring et al., 2001aGo,bGo), but also to identify biomarkers (single genes or patterns of gene expression) that are indicative of the initiation/presence of toxicant mechanisms or mode of action (Burczynski et al., 2000Go; Waring et al., 2001aGo), or can be linked to specific pathology (Waring et al., 2001bGo).

Given that only a limited number of biological specimens are reasonably available from human studies, often in very small quantities, one possible solution is to use array technology to monitor the expression of thousands of genes in a single sample from the test subjects. For this approach to be of value, gene expression in readily available, so called "surrogate" tissues must reflect or correlate with that known to indicate or be predictive for adverse effects in target organs such as the uterus.

In this article we show that, in a limited pilot study of 1185 genes, there is a significant overlap of gene expression profiles in the uterus and blood, and that gene expression in several cases is changed in a similar manner by estradiol treatment, thus providing initial evidence that gene expression profiling might be a plausible method of biomonitoring, and that PBL may be a useful surrogate tissue to "see" what is happening in inaccessible target tissues.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Animals.
Animals were treated and housed in accordance with approved guidelines (DHEW Guidelines for the Care and Use of Laboratory Animals), and approved project reviews (Animal Care and Use Committees of the U.S. EPA and NIEHS). Animals were housed singly in polycarbonate cages with pine-shavings bedding and free access to food (Purina 5001, Durham, NC) and water (5 µm filtered). The room was temperature- and humidity-controlled and subject to a 14/10 h light/dark cycle (on 0800, off 2200).

Ovariectomies.
Twelve 68-day-old female Long-Evans rats (Charles River Laboratories, Raleigh, NC) were anesthetized with a mixture of ketamine (87 mg/kg; Sigma-Aldrich, St. Louis, MO) and xylazine (13 mg/kg; Sigma-Aldrich) administered ip, and bilaterally ovariectomized on Day 0. The surgeries were performed between 1000 and 1100 h.

Dosing.
On the sixth day after ovariectomy, animals were randomly assigned to control (C1–6) or treatment (T1–6) groups in a manner that provided similar body weight means and distribution. Animals were dosed subcutaneously (sc) in the morning (0800 h) for 3 consecutive days with either vehicle control (corn oil, Sigma-Aldrich, 0.1 ml/animal/day) or 17ß-estradiol (Sigma-Aldrich, 1 µg/animal/day in 0.1 ml vehicle). The animals were sacrificed during the afternoon following the third dose: Animals were anaesthetized using Halothane (Halocarbon Labs, Riveredge, NJ) delivered mixed with room air by a vapormatic vaporizor (A.M. Bickford Inc., Wales Center, NY). Blood was collected via cardiac puncture (18 gauge/5 cc syringe) and transferred to a 6 ml EDTA whole blood collection tube (7.5% EDTA, Becton Dickinson, Franklin lakes, NJ). Tubes were centrifuged at 2000 x g for 10 min. Five hundred microliters of the plasma layer was removed to a 1.5 ml microcentrifuge tube and frozen at –800C. The buffy coat layer (PBL) was then removed using a glass Pasteur pipette and transferred to a 1.5 ml microcentrifuge tube. TRI reagent BD (0.75 ml/0.2 ml of buffy coat collected; Molecular Research Center Inc., Cincinnati, OH) was added to each tube and samples stored at –800C overnight. Uteri were also removed and, following blotting and measurement of wet weight, placed in 5 ml tubes. Trizol Reagent (Molecular Research Center Inc.) was added (1 ml/75 mg), and samples homogenized and stored at –800C overnight.

Estradiol radioimmunoassay.
The concentration of estradiol in the plasma of each animal was measured using the coat-a-count Estradiol kit (DBP, Los Angeles, CA), according to manufacturer’s instructions. Briefly, duplicate calibrator and thawed plasma samples (100 µl) of each animal were aliquoted into the coat-a-count tubes. One milliliter of 125I-Estradiol was added to each tube, and the samples incubated for 3 h at room temperature. The liquid was decanted, and the tubes counted for 1 min on a gamma counter. The value of replicate samples was averaged, and the amount of estradiol in each plasma sample calculated from the standard plot derived from the calibrator samples.

Statistical analysis.
Changes in the average uterine wet weights and plasma estradiol concentration were compared using unpaired t-tests (GraphPad Instat v3.01). For comparison of uterine wet weights, a normal Gaussian distribution was assumed and a two-tailed p-value was calculated. Since the standard deviations of the plasma estradiol level in the two populations were significantly different, the data was tested with both a nonparametric unpaired t-test and a Welch-corrected unpaired t-test.

RNA extraction.
RNA was isolated from uteri using commercially available protocols (Invitrogen Life Technologies, Carlsbad, CA). Briefly, samples were removed from the –800C and brought to room temperature. Chloroform (Fisher, Pittsburgh, PA) was added at 0.2 ml per 1 ml Trizol, shaken for 15 s, and incubated at room temperature 3 min. Samples were centrifuged at 2600 x g for 35 min at 40C. The aqueous phase was placed into a sterile 15 ml conical tube, and isopropyl alcohol added at 0.5 ml per 1 ml of Trizol. Samples were incubated at room temperature for 10 min then centrifuged at 2600 x g for 35 min. The supernatant was discarded and 75% ethanol added at 1 ml/1 ml Trizol and centrifuged at 7500 x g for 5 min. The ethanol was discarded and pellet was air dried for 30 min. Pellet was resuspended in 100 µl sterile diethylpyrocarbonate (DEPC)-treated water and stored at –800C.

RNA was isolated from the PBL using the protocol from the manufacturer (Molecular Research Center Inc.). Briefly, samples were removed from the –800C and brought to room temperature. Chloroform (Fisher) was added at 0.2 ml per Trizol volume, and the samples shaken for 15 s before incubating 3 min at room temperature. Samples were centrifuged at 8500 x g for 15 min at 40C. Aqueous phase was place into a sterile 1.5 ml microcentrifuge tube, and isopropyl alcohol added at 0.5 ml per 1 ml of Trizol. Samples were incubated at room temperature for 10 min then centrifuged at 8500 x g for 15 min. The supernatant was discarded and 75% ethanol added at 1 ml/1 ml Trizol and centrifuged at 7500 x g for 5 min. The ethanol was discarded and the pellet dried for 30 min before resuspending in 50 µl sterile DEPC-treated water.

Contaminating DNA was removed from uterus and PBL samples using DNA Free (Ambion, Austin, TX), according to manufacturer’s instructions. The purity of the RNA was assessed using O.D.260/280 ratio, and the integrity checked on a 0.7% agarose gel. All RNA was stored at –800C.

RT-PCR of estrogen receptor-{alpha}.
Estrogen receptor-{alpha} (ER{alpha})-specific primers designed by Mohamed and Abdel-Rahman (2000) were used. Primers used were F1: 5‘- GCGGCTGCCACTGACCATG and R1: 5‘- CCTCGGGGTAGTTGAACACGG, allowing the amplification of a 185 base pair fragment corresponding to the 5‘ noncoding region and to the first part (N-terminal) of the coding region of the rat ER{alpha} gene. Approximately 1 µg of RNA from each PBL fraction of 3 treated animals was pooled, and reverse transcription (RT) reactions were performed with 200, 100, and 50 ng of RNA using the reaction conditions established with the uterine RNA. As a positive control, reactions with 6.25 ng of uterine RNA were performed at the same time. An RT reaction mix containing 2 µl of 10x RT buffer (Promega, Madison, Wisconsin), 4 µl 25 mM MgCl2, 2 µl 10 mM (each) dNTP mix, 0.5 µl Rnasin (Promega), 0.3 µl of 10 pmol/ml ER{alpha}-specific primer (F1), 0.4 µl of AMV reverse transcriptase (Promega, 10 U/ml), and 9.8 µl DEPC-treated water was prepared. Nineteen microliters of reaction mix was aliquoted to each tube containing 1 µl of the appropriate RNA. Tubes were incubated at 420C for 15 min, then at 950C for 5 min.

After heat inactivation, PCR was initiated by adding 30 µl of PCR master mix directly to each RT tube. PCR master mix contained 3 µl of 10X PCR buffer, 0.3 µl (10 pmol/µl) of each F1 and R1 primers, 0.25 µl Taq polymerase (1.25 U per reaction, Promega), and sterile water to a final volume of 30 µl per reaction (final volume in PCR reaction was 50 µl). The amplification was performed in a PTC-100 thermocycler (MJ Research, Inc., Waltham, MA) and consisted of 35 cycles (15 s denaturation at 940C, 1 min annealing at 550C, and 30 s elongation at 720C), with a final elongation at 720C for 5 min. A 2.8% MetaPhor agarose gel (FMC BioProducts, Rockland, ME) was used to analyze the PCR products. Bands were visualized with ethidium bromide.

Microarray hybridization.
32P-labelled cDNAs were produced from 2 µg of each individual RNA, according to the manufacturer’s instructions, using the reagents supplied with the Atlas rat 1.2i array kit (catalog #7854-1, Clontech Laboratories Inc., Palo Alto, CA). Hybridization to these nylon membrane arrays was carried out according to the manufacturer’s instructions. Briefly, each sample of total RNA was reverse-transcribed with 32P-dATP, and the labeled cDNA probe fractions collected using ChromaSpin-200 columns. Fresh filter arrays were used in most cases. However, where filters were reused (see Filter variations, below), they were first allowed to decay at –200C for 3–4 months. Just prior to reuse, they were stripped in boiling 0.5% SDS solution for 10 min. Filters were placed in 35 x 150 mm roller bottles and prehybridized with 5 ml ExpressHyb for 30 min at 68°C with continuous agitation. The labeled cDNA probe fraction (2–10 x 106 counts) was denatured and added to prehyb solution together with 5 µg of Cot-1 DNA. Hybridizations were carried out overnight at 68°C. After washing, the arrays were exposed to phosphorimaging plates for 1 to 5 days. Images were developed using a PhosphorImager (Molecular Dynamics Inc., Sunnyvale, CA) and quantitated using Atlas Image v1.2 or 2.0 (Clontech Laboratories Inc.).

Filter variations.
Each RNA sample was hybridized to 1 filter array, except 6 RNAs (3 from each of the control and treated uterine samples), which were hybridized a further two times each to additional filters in order to examine the extent of experimentally derived variation.

Two lots of filter arrays were used during the study to examine the degree of inter-lot variation. For the PBL RNA, 3 control and 3 treated samples were hybridized to filters from lot 1. Two control and two treated samples were hybridized to filters from lot 2. For the uterine RNA, 3 control and 3 treated samples were all done on filters from lot 1. The two repeats (rep1, rep2, described above) were all done on filters from lot 2.

Six filters were stripped and reused once in order to see if this provided satisfactory hybridizations. Thus, the remaining individual uterine RNA samples (3 control and 3 treated) were hybridized once each to reused membranes of lot 1 (which were originally used for uterine samples of the other three control and treated uterine samples).

Background correction and normalization of raw array data.
The filters included 1185 genes. The datasets were analyzed after subtraction of local background. Genes were considered "expressed" and retained for analysis only if they had intensity levels at least twice background in both PBL and uterus, in at least 3 of the 5 samples where both PBL and uterus were available in either the treated or control animals. Since local background could be quite high for spots where there was bleed-over from highly expressed adjacent spots, twice background was defined to be local background + median background, which left 193 genes in the analysis. After background subtraction, the data were log2-transformed. Normalization of the data and subsequent analysis followed a procedure similar to that proposed by Wolfinger et al.(2001). The data were normalized using a mixed effects model: log2 (yij) = intercept + {alpha}i x arrayi, where array was considered a random effect. The residuals from this model were the values that were further analyzed. Essentially, this centers each array around 0, so that values from arrays of different mean intensities can be compared.

Analysis of DNA array data.
Sufficient blood RNA was obtained from 5 of the 6 control and 5 of the 6 treated animals to carry out array hybridizations, and each RNA was hybridized once. Therefore, for each gene there were 5 treated and 5 control PBL values. Uterine RNA was obtained from all 6 control and all 6 treated animals. Three control and 3 treated RNAs were hybridized once each. The other 3 control and 3 treated RNAs were each hybridized in triplicate, and the average normalized values from these was used in the final analysis of gene expression changes between the 6 control and 6 treated uterine RNAs. The residuals from the normalization model described above were used in the analysis. For each gene, a mixed effects linear model was used, which included as predictors of intensity the fixed effects: treatment, tissue, treatment-by-tissue interaction, and the random effects: animal and lot-by-experiment. Two contrasts (the test of treatment difference for each tissue) were tested in the model. The genes with most potential as biomarkers were those where there was a significant treatment effect in both tissues.

The uterus and PBL data were also analyzed independently using the same normalization and analysis procedures, except for the following: in normalization, genes were kept in the PBL analysis if they were expressed in 3/5 samples, and in the uterus analysis if they were expressed in 4/6 samples. In the analysis, treatment was the only fixed effect in the model.

Quantitative real time RT-PCR.
1 µg each of DNase-treated uterine RNA samples was reverse transcribed with random hexamer primers using Applied Biosystems’ Reverse Transcription kit according to manufacturer’s instructions. cDNA from this reaction was diluted 1/10,000 for use with the 18S primers (eukaryotic 18S rRNA, Applied Biosystems). Five microliters (50 ng) of each random hexamer reaction was used in each PCR reaction with each primer set. PCR was performed with a premade PCR Master Mix (Applied Biosystems) according to manufacturer’s instructions. All primers were Pre-Developed TaqMan Assay Reagents (PDARS, rat c-jun, rat PCNA, rat TGFB, rat c-fos, rat Hsp-70, rat IGF-1, Applied Biosystems). Quantitative real time PCR was performed in a BioRad iCycler using the appropriate filter set. The 18S primer required a VIC filter. All the other primers used a FAM filter. In each plate all samples were amplified with 18S primers, and in separate tubes these same samples were also amplified with one of the other primer pairs. All samples were run in duplicate. The threshold crossing level for the duplicate sets was averaged, and the results from the 6 control samples compared against the 6 treated samples using unpaired t-tests (GraphPad Instat v3.01).


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The plasma estradiol levels were significantly greater in the estradiol-treated animals (130.8 pg/ml) than in the control group (23.7 pg/ml; Mann-Whitney test, p = 0.0022; Welch-corrected unpaired t-test, p = 0.0001). The average wet-weight for the uteri of estradiol-treated animals (0.290 g) was significantly greater than the control group (0.141 g; unpaired t-test, two–tailed p-value < 0.0001). These data show that, following the ovariectomies, the circulating levels of endogenous estradiol were significantly reduced in the control group, that the dose of estradiol administered to the treated group resulted in levels of circulating estradiol which were slightly higher than the peak concentrations in unaltered animals (approximately 100 pg/ml), and that the sc-administered estradiol was acting in the anticipated manner in stimulating uterine growth.

Typical hybridization patterns observed with RNA from PBL and uterus samples are shown in Figures 1A and 1BGo. Control and treated samples showed broadly similar patterns, with the greatest differences being in the change of intensity of specific genes. It is clear that there is a large difference in the expression patterns between the two tissues. Most striking perhaps is the relatively small number of genes expressed in the PBL RNA (330, out of 1185 present on the array), and the presence of three very highly expressed genes: microglobulin, neuropeptide Y, and mitochondrial cytochrome c oxidase, subunit IV. The uterus also expressed only a relatively small fraction (39%, 459 genes) of the total genes present on the array. There were no very abundantly expressed genes in the manner of the PBL RNA, perhaps because the uterus is a much more heterogeneous mix of cell types.



View larger version (103K):
[in this window]
[in a new window]
 
FIG. 1. Representative images of the PBL (A) and uterus (B) array hybridizations from control animals. DNased total RNA from PBL and uterine samples of control or estradiol-treated ovariectomized rats was used to produce 32P-labelled cDNA probes to Clontech Atlas Rat 1.2i Arrays. Probes were hybridized to the arrays overnight, washed, and the hybridization image captured using a phosphorimaging screen 1–5 days later. The images presented (derived from control animals) have, for illustrative purposes, been slightly overexposed to facilitate pattern observation. The 3 highly expressed genes, microglobulin (M), neuropeptide Y (N), and mitochondrial cytochrome c oxidase, subunit IV (C), are noted.

 
RT-PCR analysis confirmed that ER{alpha} mRNA is present in the uterus and in the pooled PBL fraction of 3 treated animals (Fig. 2Go).



View larger version (66K):
[in this window]
[in a new window]
 
FIG. 2. RT-PCR of RNA from uterine and PBL samples demonstrating presence of ER{alpha} receptor message. Total RNA from the PBL fraction of 3 estradiol-treated ovariectomized rats was pooled, and RT-PCR reactions were performed using the reaction conditions established with uterine RNA from the same animals. Lane 1, 1 kb DNA ladder; Lanes 2 and 3, PBL RNA, 200 ng; Lanes 4 and 5, PBL RNA, 100 ng; Lanes 6 and 7, PBL RNA, 50 ng; Lanes 8 and 9, uterus RNA, 6.25 ng; Lane 10, no RNA; Lane 11, no RT enzyme.

 
Statistical Analysis
The question of background.
There has been debate among users of Clontech membrane arrays about the value of subtracting background and what to use as background. Subtracting background is a standard scientific procedure in experiments where it is measured. In this technique however, subtracting background can greatly increase the variability around the lower intensity genes. The Atlas Image software provides several options for subtracting background. Thus, in this work we compared normalized data which had been either adjusted for local background (where the background for each gene is calculated based on the median intensity of the space directly surrounding the gene spot), adjusted for the median local background (where the background is set at the median intensity of a user-defined area of the array membrane), or unadjusted for background. An increased variability in the low end of the plot was observed where background is subtracted, and in both cases the resulting plots (data not shown) looked very similar. We used local background subtraction because some of the very highly expressed genes had bleed-over onto adjacent spots. If we had chosen not to subtract local background under such conditions, genes adjacent to such spots would have had inflated expression values.

The question of replicates, lots, and experiments.
Three treated and 3 control uterus samples were each run 3 times to look at the relative importance of running repeats of samples versus different samples. The first hybridization of each sample was run at the same time, using the first lot of arrays. The second and third hybridizations of these samples were run later, using the second lot of arrays. The two replicates appeared fainter than the first hybridization, and even after normalization appeared to be more similar to each other than to the first replicate. In order to investigate this further, a fixed effects linear model was run for each gene, which included as predictors of intensity: treatment, animal (nested within treatment), and lot (or experiment, these cannot be separated). The mean squared error of each predictor was compared with the error of the model. In this case the error of the model can be considered the within-sample error, and can be compared to the among-sample error, that is due to animal variance. The among-animal variance was on average 2.1 times the within-animal variance. Although this ranged widely across genes, (0.02 to 22.1, median 1.4), where the within-animal variance was higher than the among-animal variance, both variances were usually very small. In contrast, the variance due to treatment averaged 15.3 times the within-animal variance, and the variance due to lot/experiment 37 times the within-animal variance. These distributions were highly skewed, the medians of the ratios being 4.1 and 7.0 respectively. This effectively means that the very large ratios for relatively few genes greatly influenced the means. In the full dataset, there were 4 different "lot-by-experiment" categories. The variability of lot-by-experiment had on average a variance 11.2 times that of the error variance (median = 3.4). The variable, "lot," was thus included as a predictor of intensity in the analysis.

Gene Expression Changes
Eighty-one of the 330 genes expressed in the PBL RNA were significantly altered in the treated samples compared to the controls at p < 0.1, 55/330 at p < 0.05, 17/330 at p < 0.01, and 2/330 at p < 0.001. For these significantly altered genes, changes ranged from a 5.6-fold decrease to a 5.8-fold increase. For the uterus, 279 of the 459 expressed genes were significantly altered in the treated samples compared to the controls at p < 0.1, 240/459 at p < 0.05, 173/459 at p < 0.01, and 108/459 at p < 0.001. For these significantly altered genes, changes ranged from a 20-fold decrease to an 11.4-fold increase.

Gene expression changes in the PBL RNA were compared to those in the uterine RNA to look for similar effects of the estradiol treatment. A total of 193 genes (out of 1185 on the array) were found to be detectable in both the PBL and uterus RNA. Of these 193 genes, 18 were found to have significant changes in expression in both tissues following 17ß-estradiol treatment. Figures 3AGo (PBL) and 3B (uterus) show graphically which of the 193 genes were significantly changed in each tissue. On the whole, genes whose expression is not significantly changed by estradiol treatment are distributed along the line of identity (gray circles), while those whose expression is significantly changed lie further away from the line. It is apparent that certain genes are not significantly altered even when their expression appears to be changed 2-fold or more between treatments. In contrast, there are several cases where genes have significant changes that are much less than 2-fold.



View larger version (22K):
[in this window]
[in a new window]
 
FIG. 3. Scatterplot of 193 genes expressed in both PBL and uterus samples showing those that were significantly changed in one or both tissues. Comparison of the mean treated intensities versus mean control intensities of the (A) PBL and (B) uterine samples. In both cases, the blue stars represent the genes in the appropriate tissue that were significantly altered in the treated samples compared to the controls. The red stars represent those genes that were significantly altered in both the PBL and the uterus, i.e., they represent the same genes on each graph, although their location may be different. The central line is the line of identity (x = y), along which all genes having zero change (1:1 ratio) fall. The hatched lines represent the point where there is a 2-fold difference between control and treated genes. Genes under the line of identity are downregulated in the treated samples, and those above it are upregulated.

 
Table 1Go lists 29 genes that altered in a significant (p < 0.05), or nearly significant (0.05 < = p < 0.1) manner in both tissues. Eighteen genes showed significant changes in both tissues. Nine genes were included where the change was significant in one tissue and nearly significant in the other, and two were included where the gene changes in both tissues were nearly significantly. Though the inclusion of "nearly significant" changes is not necessarily standard practice, these genes are of possible interest since the use of larger numbers of arrays could potentially identify them as being genuinely significant. The genes in Table 1Go can be divided into two main groups:


View this table:
[in this window]
[in a new window]
 
TABLE 1 Genes Significantly Changed in Both PBL and Uterus following Estradiol Treatment of Ovariectomized Rats
 

As can be seen from Table 1Go, it is possible to break these two groups down still further, to discern those genes with different degrees of linkage significance, extent of change, and direction of change. Nevertheless, this group of 29 genes as a whole can be seen as forming the basis of a set of genes whose expression profile may change correspondingly (positively or negatively) in the PBL and uterus in response to challenge from estradiol or other estrogenic compound.

Quantitative Real Time RT-PCR (qRT-PCR)
Expression of 6 target genes was measured in the 6 control and 6 treated uterine samples. There was insufficient PBL RNA to perform qRT-PCR analysis. 18S RNA was measured on each occasion for all samples as an endogenous control. There was no significant difference in threshold crossing (i.e., expression level) for 18S RNA between treated and control samples on any single day (paired t-tests, p = 0.7809 to p = 0.2291) or between any samples on different days (1-way ANOVA, p = 0.3978). This indicates that the amount of starting RNA in the samples from the control and treated groups was never significantly different. The results of the qRT-PCR analyses, in comparison to the array data, are shown in Table 2Go.


View this table:
[in this window]
[in a new window]
 
TABLE 2 Comparison of Selected Gene Changes According to Arrays and qRT-PCR
 

    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
This pilot study, carried out with commercially available DNA membrane arrays, suggests that gene expression analysis of surrogate tissues could be developed into a valid approach for biomonitoring the effects of EDCs and other toxicant exposures on reproductive and other tissues with minimally or noninvasive procedures. This approach is ideally possible where 3 conditions are met:

The validity of gene expression analysis using DNA array technology is still of course the subject of debate, and there are some aspects of the methodology that require further validation. For example, there has been much debate over the relative merit of applying RNA from a small number of individuals on several arrays each, versus doing a larger number of individuals, each on a single array. Our analysis of the between- and within-sample error, and the error associated with array lot suggests two things: first, that since the within-animal variance is small, it would be advantageous to use more animals rather than more replicates to get a better estimate of the true mean across animals. Secondly, the variation due to lot (or experiment) tended to be large, and for some genes huge, which suggests that whenever possible experiments should be done close together in time, and with the same lot of filters. When that is not possible, differences due to filter lot or experiment should be examined and adjusted for statistically if necessary. The question of whether or not to subtract background, and if so, which one (median or local), is also difficult. The subtraction of either local or median background increases variability among the low-end expression values. For this reason, as long as there are no gross differences in background across the membrane and no significant bleed-over from the spots of highly expressed genes, it might be better not to subtract background, and rely instead on the normalization to adjust for differences in intensity among arrays. Since we had significant bleed-over in several spots, we decided on this occasion to subtract local background.

Membrane arrays, rather than chip-based arrays, are the current format of choice where limited quantities of RNA are available, such as from blood samples or biopsies. In this experiment we used the minimum amount (2 µg) of RNA recommended by the array manufacturer. Most chip-based arrays require 15 µg or more of total RNA, and are therefore inappropriate for small samples. Although some chip-based labeling systems (e.g., NEN’s "TSA" system and Genisphere’s "DNA dendromer" system) claim to work with as little as 0.5–1 µg total RNA, there is still some difference of opinion as to the reproducibility of these systems.

Array results are often validated to confirm changes in expression of certain genes with methods such as Northern blots, RT-PCR, or Real Time PCR. However, such confirmatory analyses are by no means universal, and there appears to be no consensus as to how many genes, if any, must be examined in order for such studies to be acceptable. The ability to carry out such supporting experiments can be confounded by lack of RNA, particularly where biopsies are used. In our work for example, the amount of RNA extracted from the blood samples was barely more than required for an array. Indeed, in two cases (one control and one treated animal), we were unable to extract sufficient PBL RNA to apply to the arrays, while in the others there was insufficient RNA to carry out confirmatory studies.

Nevertheless, we did perform a small number of confirmatory studies on the uterine RNA samples using qRT-PCR. On balance, the array data and qRT-PCR data were well matched: increases in expression were recorded for all 6 genes on both platforms, with the degree of increase being very similar in both platforms for 5 out of 6 genes. The one exception was C-fos, where the increase detected by qRT-PCR was a 6-fold higher than the increase detected on the arrays. This is by no means an unprecedented finding in such comparative studies. It is widely acknowledged that arrays tend to give "flatter" results than qRT-PCR, a phenomenon that is thought to be due largely to the specificity of the probe cDNAs on the arrays. In two cases, the observed increase is not significant on one of the two platforms (TGFß1 on the arrays; C-jun with qRT-PCR). According to the arrays, the change in TGFß1 was "nearly significant" (p < 0.1, see Table 1Go). The qRT-PCR data appears to support its inclusion in the table by confirming a significant (p < 0.05) increase. In light of this result, it would have been useful to carry out similar confirmatory studies on all such "nearly significant" genes.

In both these disparate cases, the p-values are very close to significance ("nearly significant"), and in previous studies both C-jun and TGFß1 have been shown to be significantly upregulated in the uterus following estradiol exposure (Rageh et al., 2001Go; Webb et al., 1990Go). Thus, the observed discrepancies are probably due largely to the fact that the measured increases in expression are not large, and it is possible that a slightly larger cohort of control and treated samples could have pushed the results into significance.

The issue of confirmation needs to be addressed more rigorously in future studies, perhaps by reducing the amount of RNA for the arrays to less than 1 µg. Although it is below the recommended amount for Atlas arrays, we have produced clear, reproducible results in trial experiments using 1 µg of RNA. Furthermore, Clontech supplies the Atlas SMART probe amplification kit, which can be used to preamplfy as little as 50 ng starting RNA for hybridization to their arrays. It has been reported (Dr. G. Hellmann, RJ Reynolds, personal communication), that as little as 10 ng total RNA is sufficient to generate clean, reproducible array data using this kit. Reducing the amount of RNA used for array hybridization in such a manner should free up enough RNA to carry out a limited number of confirmatory studies on some of the candidate surrogate markers listed in Table 1Go.

Perhaps the main disadvantage of the limited number of genes included on these off-the-shelf arrays is that they are generic in nature (to appeal to as broad a range of researchers as possible), and the genes are chosen arbitrarily according to available sequence and ontology information. Thus, the majority of genes on the arrays were uninformative for the blood or uterus (i.e., not expressed), and there is a relatively small overlap between RNA species that are expressed in both the blood and uterus. This clearly restricts the number of genes that can be considered as potential biomarkers of toxicant exposure. However, this could be rectified in one of two ways. One is to produce a custom array designed specifically to include genes that are expressed in both uterus and blood cells. In this fashion, we have previously produced a testis-expression array for analyzing testicular responses to environmental toxicants (Rockett et al., 2001Go). Furthermore, the array we used in the current work did not include many genes that are known to be involved in estrogen-mediated responses, including ER{alpha} itself. Thus, the utility of a custom array could be further enhanced by including a high number of genes that are involved in the model being measured, in this case the uterine response to the presence of estrogenic compounds. Secondly, broad coverage arrays containing many thousands of genes could be employed. Indeed, Clontech, among other manufacturers, is moving gradually toward this target, having recently introduced a human "plastic" array containing more than 8000 genes. It should not be too long before similar numbers of genes are identified in the rat (and other species) and become available for commercial and in-house array production. Strategically speaking, it may be possible at a later juncture to reduce the number of genes (and thus the cost of producing the arrays) to those that are known to provide an informative response in the tissues being analyzed.

Our results demonstrate that there are genes expressed in both the uterine cells and the PBL population that respond to 17ß-estradiol treatment with significant changes in their level of expression. Many of the changes observed in this study are supported by previous findings. For example, of the genes listed in Table 1Go, it has been documented that estradiol treatment induces Jun-D (Cicatiello et al., 1992Go; Nephew et al., 1993Go), phospholipase A2 (Pakrasi et al., 1983Go), thymidine kinase (Sakamoto et al., 1980Go), and insulin-like growth factor binding protein (IGFBP-) one (Yallampalli et al., 1993Go) expression in adult ovariectomized rats. In addition, cytoplasmic ß- actin mRNA has been shown to transiently increase in rat uterus after treatment with estradiol (Hsu and Frankel, 1987Go). In addition, interleukin-4 receptor, lipoprotein lipase, N-myc, protein kinase C, vascular endothelial growth factor (VEGF), c-Jun, c-Fos, and proliferating cell nuclear antigen have all previously been shown to be upregulated by estradiol in the uteri of ovariectomized rats, while IGFBP-3, IGFBP-5, and neuropilin 2 (VEGF receptor 2) have been confirmed as being suppressed. Indeed, of the 17 genes that we found in PubMed (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=PubMed) with previously documented expression in an estradiol-treated ovariectomized rat uterus model, the only contrasting data we discovered was that our results indicate that serotonin receptor 5B is downregulated in such animals, while previous studies suggest that it is upregulated (Ichida et al., 1983Go, 1984Go). However, this may well be a function of the subtype of receptor, since discrimination of the receptor subtypes was not performed in these older studies and there are currently 15 rat serotonin receptors listed under UniGene. It must be borne in mind, however, that in most of these cases the exact model used in the present was not necessarily applied. Thus, although the published experiments were carried out in ovariectomized adult rats treated with estradiol, the strain, age, dose, and time of harvest was not necessarily the same as we used. Nevertheless, the literature supports many of the changes we detected using the DNA array approach, suggesting that they were probably real. Moreover, many of the observed changes in this exposure model are intuitively understandable from a biological context. For example, it is no surprise to see an increase in the expression of Jun-D, C-jun, C-fos, and N-myc, all of which are important genes involved in transcription regulation. PCNA, ß-actin, and VEGF could also be expected to increase expression in a proliferating tissue, although the decrease in VEGF receptor mRNA is more difficult to explain.

Gene expression in lymphocytes in response to estradiol has not been characterized to any significant degree, and therefore a search of published abstracts did not reveal a similar support for the genes listed in Table 1Go. The fact that ER{alpha} and ERß are expressed in lymphocytes, and that estradiol appears to plays a role in lymphocyte development, indicates that further studies of the role of estradiol in immune-cell function could reveal some interesting similarities with nonimmune cells. This is supported by the work of Echeverria et al.(1994), the results of which led them to suggest that estradiol exerts its effects through a common nuclear mechanism in cells of male and female reproductive and nonreproductive organs. In agreement with previous immunohistochemical (Hiroi et al., 1999Go), in situ hybridization (Shughrue et al., 1998Go), RNase protection assays (Nephew et al., 2000Go) and RT-PCR studies (Khurana et al., 2000Go), the RT-PCR analysis confirmed that ER{alpha} mRNA is present in the uterus. Moreover, ER{alpha} mRNA was also found to be present in the pooled PBL fraction of 3 treated animals (Fig. 2Go). Although this has never been confirmed before in rat lymphocyte preparations, Tornwall et al.(1999) demonstrated that ER{alpha} RNA is present in several different populations of murine lymphocytes. That ER{alpha} message is indeed present in both the PBL and uterus fractions of estradiol-treated ovariectomized rats provides a possible common mechanism for mediating estradiol-induced effects. Whether this is the case could be addressed by the use of an ER{alpha}-blocker such as lasofoxifene (Ke et al., 1998Go) in combination with the estradiol treatment.

As can be seen in Table 1Go, some changes were in opposing directions in the two tissues, and this in itself is not necessarily surprising given the very different functions of the two tissues. However, such changes may still be useful if they are shown to be consistent. Whatever their direction of regulation, all the genes we have presented are possible candidates for inclusion in the fingerprint of a blood-uterus paradigm of biomonitoring. However, the genuine utility of these genes cannot be ascertained without further study. For example, it is important to know if the observed responses were specific or nonspecific, direct or indirect, and if they would continue to be regulated in a like manner across a range of estradiol doses. An ideal genomic biomarker would be an RNA species that increased or decreased in expression in both "target" and "surrogate" tissues proportionally to the dose level and have a wide dynamic range. If this condition were true, then it is probable that the gene was being regulated through a similar mechanism in both tissues. The best candidates for biomarkers of estrogenic action would have a known involvement in the uterotrophic response and be activated by interaction with one of the estrogen receptors. The identification of such markers would perhaps be facilitated by improving our exposure model, since the residual quantities of estradiol that were observed in the control animals were larger than expected or desired in such studies. They were probably produced by the adrenal glands, as these have been shown to produce estradiol in rats (Steger and Peluso, 1982Go; Tinnikov et al., 1988Go) and sheep (Adams et al., 1990Go). Furthermore, aromatases in peripheral tissues are capable of converting adrenal-synthesized androgen precursors into estrogen (Brodie, 1991Go; Simpson et al., 1999Go). Davidge et al.(2001) recently suggested that only ovariectomized rats treated with either aromatase inhibitor or calorie-controlled diet constitute an effective control condition for estrogen studies. In future studies, the exposure regimen we used could be extended to include model estrogenic and antiestrogenic compounds, of various potencies and acting through different mechanisms or modes of action. An analysis of gene expression data from these kinds of studies should help reveal biomarkers and patterns of gene expression that are indicative of the degree and type of exposure. Diel et al.(2000) used such an approach to compare the effects of different estrogenic compounds on the expression of a small set of genes on the uterus, and from their findings concluded that the fingerprint of uterine gene expression is a very sensitive tool to investigate estrogenicity of natural and synthetic compounds, and offers the possibility of obtaining information regarding the molecular mechanisms involved in the action of the respective compounds.

Although the number of coregulated genes discovered here (18) might appear rather limited, it should be remembered that the arrays used in this pilot study contained only 1200 genes. Given that the latest estimates are that the mouse and human genome contain around 34K and 42K genes respectively and the fair assumption that the rat will have a similar number, it is possible that there may be in excess of 500 genes in total that are coregulated in this PBL-uterus estrogen exposure model. Furthermore, gene expression differences were assessed at only one timepoint. Since gene expression is a dynamic process, it is probable that additional genes will be also change at different times after or during an exposure. Thus, there is likely to be an abundant pool of target genes expressed in both tissues from which to derive candidates for biomonitoring exposure and/or effect.

If similar findings to those described here can be reproducibly demonstrated with different classes of chemicals in response to chronic and acute exposures of different degree, then the use of surrogate tissues may prove useful as a noninvasive method for biomonitoring toxicant exposure. Any combination of surrogate and target tissues can potentially be used, although it may be advantageous to pair more compatible tissues where possible. For example, blood cells may be more informative of the status of the thyroid or spleen than reproductive organs. The use of a surrogate tissue to monitor gene expression changes in tissues or organs elsewhere in the body could ultimately pave the way for early diagnosis of environmental EDC (and other) exposures, offer prognostic indicators of toxicant exposure and health outcome, and permit the development of lifelong health monitoring programs. This pilot study suggests that this could well be the case and justifies further investigation into this new biomonitoring concept.


    ACKNOWLEDGMENTS
 
We gratefully acknowledge the following people for their help and assistance in planning and carrying out this work: with the U.S. EPA, Dr. Tammy Stoker and Mrs. Dorothy Guidici for performing the ovariectomies, Drs. Sue Fenton and Gerri Youngblood (now with Stratagene) for performing the serum estradiol assays, Dr. Robert Luebke for technical advice; we thank Dr. Barbara Abbott (U.S. EPA) and Dr. Valerie Baker (Unilever) for scientific review of the manuscript.


    NOTES
 
The information in this document has been funded wholly by the U.S. Environmental Protection Agency. It has been subjected to review by the National Health and Environmental Effects Research Laboratory and approved for publication. Approval does not signify that the contents reflect the views of the Agency, nor does mention of trade names or commercial products constitute endorsement or recommendation for use.

1 To whom correspondence should be addressed at U.S. Environmental Protection Agency, Reproductive Toxicology Division (MD-72), Research Triangle Park, NC 27711. Fax: (919) 541-4017. E-mail: rockett.john{at}epa.gov. Back


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Adams, N. R., Atkinson, S., Hoskinson, R. M., Abordi, J. A., Briegel, J. R., Jones, M. and Sanders, M. R. (1990). Immunization of ovariectomized ewes against progesterone, oestrogen or cortisol to detect effects of adrenal steroids on reproduction. J. Reprod. Fertil. 89, 477–483.[Abstract]

Bartosiewicz, M. J., Jenkins, D., Penn, S., Emery, J., and Buckpitt, A. (2001). Unique gene expression patterns in liver and kidney associated with exposure to chemical toxicants. J. Pharmacol. Exp. Ther. 297, 895–905.[Abstract/Free Full Text]

Brodie, A. (1991). Aromatase and its inhibitors—an overview. J. Steroid Biochem. Mol. Biol. 40, 255–261.[ISI][Medline]

Burczynski, M. E., McMillian, M., Ciervo, J., Li, L., Parker, J. B., Dunn, R. T., 2nd, Hicken, S., Farr, S., and Johnson, M. D. (2000). Toxicogenomics-based discrimination of toxic mechanism in HepG2 human hepatoma cells. Toxicol. Sci. 58, 399–415.[Abstract/Free Full Text]

Cicatiello, L., Ambrosino, C., Coletta, B., Scalona, M., Sica, V., Bresciani, F., and Weisz, A. (1992). Transcriptional activation of jun and actin genes by estrogen during mitogenic stimulation of rat uterine cells. J. Steroid Biochem. Mol. Biol. 41, 523–528.[ISI][Medline]

Davidge, S. T., Zhang, Y., and Stewart, K. G. (2001). A comparison of ovariectomy models for estrogen studies. Am. J. Physiol. Regul. Integr. Comp. Physiol. 280, R904–907.[Abstract/Free Full Text]

Diel, P., Schulz, T., Smolnikar, K., Strunck, E., Vollmer, G., and Michna, H. (2000). Ability of xeno- and phytoestrogens to modulate expression of estrogen-sensitive genes in rat uterus: Estrogenicity profiles and uterotropic activity. J. Steroid Biochem. Mol. Biol. 73, 1–10.[ISI][Medline]

Echeverria, O. M., Gonzalez M. A., Traish, A. M., Wotiz, H. H., Ubaldo, E., and Vazquez-Nin, G. H. (1994). Immuno-electron microscopic localization of estradiol receptor in cells of male and female reproductive and non-reproductive organs. Biol. Cell. 81, 257–265.[ISI][Medline]

Hiroi, H., Inoue, S., Watanabe, T., Goto, W., Orimo, A., Momoeda, M., Tsutsumi, O., Taketani, Y., and Muramatsu, M. (1999). Differential immunolocalization of estrogen receptor alpha and beta in rat ovary and uterus. J. Mol. Endocrinol. 22, 37–44.[Abstract/Free Full Text]

Hsu, C. Y., and Frankel, F. R. (1987). Effect of estrogen on the expression of mRNAs of different actin isoforms in immature rat uterus. Cloning of alpha-smooth muscle actin message. J. Biol. Chem. 262, 9594–9600.[Abstract/Free Full Text]

Ichida, S., Oda, Y., Tokunaga, H., Hayashi, T., Murakami, T., and Kita, T. (1984). Mechanisms of specific change by estradiol in sensitivity of rat uterus to serotonin. J. Pharmacol. Exp. Ther. 229, 244–449.[Abstract]

Ichida, S., Tokunaga, H., Oda, Y., Fujita, N., Hirata, A., and Hata, T. (1983). Increase of serotonin receptors in rat uterus induced by estradiol. J. Biol. Chem. 258, 13438–13443.[Abstract/Free Full Text]

Ke, H. Z., Paralkar, V. M., Grasser, W. A., Crawford, D. T., Qi, H., Simmons, H. A., Pirie, C. M., Chidsey-Frink, K. L., Owen, T. A., Smock, S. L., Chen, H. K., Jee, W. S., Cameron, K. O., Rosati, R. L., Brown, T. A., Dasilva-Jardine, P., and Thompson, D. D. (1998). Effects of CP-336,156, a new, nonsteroidal estrogen agonist/antagonist, on bone, serum cholesterol, uterus and body composition in rat models. Endocrinology 139, 2068–2076.[Abstract/Free Full Text]

Khurana, S., Ranmal, S. and Ben-Jonathan, N. (2000). Exposure of newborn male and female rats to environmental estrogens: Delayed and sustained hyperprolactinemia and alterations in estrogen receptor expression. Endocrinology 141, 4512–4517.[Abstract/Free Full Text]

Luft, J. C., and Dix, D. J. (1999). Hsp70 expression and function during embryogenesis. Cell Stress Chaperones 4, 162–170.[ISI][Medline]

Mohamed, M. K., and Abdel-Rahman, A. A. (2000). Effect of long-term ovariectomy and estrogen replacement on the expression of estrogen receptor gene in female rats. Eur. J. Endocrinol. 142, 307–314.[ISI][Medline]

Nephew, K. P., Long, X., Osborne, E., Burke, K. A., Ahluwalia, A., and Bigsby, R. M. (2000). Effect of estradiol on estrogen receptor expression in rat uterine cell types. Biol. Reprod. 62, 168–177.[Abstract/Free Full Text]

Nephew, K. P., Webb, D. K., Akcali, K. C., Moulton, B. C., and Khan, S. A. (1993). Hormonal regulation and expression of the jun-D protooncogene in specific cell types of the rat uterus. J. Steroid Biochem. Mol. Biol. 46, 281–287.[ISI][Medline]

Pakrasi, P. L., Cheng, H. C., and Dey, S. K. (1983). Prostaglandins in the uterus: Modulation by steroid hormones. Prostaglandins 26, 991–1009.[Medline]

Rageh, M. A., Moussad, E. E., Wilson, A. K., and Brigstock, D. R. (2001). Steroidal regulation of connective tissue growth factor (CCN2; CTGF) synthesis in the mouse uterus. Mol. Pathol. 54, 338–346.[Abstract/Free Full Text]

Rockett, J. C., Luft, J. C., Garges, J. B., Krawetz, S. A., Hughes, M. R., Kim, K. H., Oudes, A., and Dix, D. J. (2001). Development of a 950-gene DNA array for examining gene expression patterns in mouse testis. GenomeBiology 2, research 0014.1–0014.9.

Sakamoto, S., Yamada, N., and Okamoto, R. (1980). [The effects of estradiol on uterine estrogen receptor and thymidine kinase in immature and spayed adult rats (author’s transl.)]. Nippon Naibunpi Gakkai Zasshi 56, 926–932.[Medline]

Shughrue, P. J., Lane, M. V., Scrimo, P. J., and Merchenthaler, I. (1998). Comparative distribution of estrogen receptor-{alpha} (ER-{alpha}) and ß (ER-ß) mRNA in the rat pituitary, gonad, and reproductive tract. Steroids 63, 498–504.[ISI][Medline]

Simpson, E., Rubin, G., Clyne, C., Robertson, K., O’Donnell, L., Davis, S., and Jones, M. (1999). Local estrogen biosynthesis in males and females. Endocr. Relat. Cancer 6, 131–137.[Abstract/Free Full Text]

Steger, R. W., and Peluso, J. J. (1982). Effects of age on hormone levels and in vitro steroidogenesis by rat ovary and adrenal. Exp. Aging Res. 8, 203–208.[ISI][Medline]

Tinnikov, A. A., Bazhan, N. M., and Ivanova, L. N. (1988). [Secretion of steroid hormones by the adrenals and ovaries and estradiol receptors in the uterine cytosol in postnatal ontogenesis]. Probl. Endokrinol. (Mosk.) 34, 67–70.

Tornwall, J., Carey, A. B., Fox, R. I., and Fox, H. S. (1999). Estrogen in autoimmunity: Expression of estrogen receptors in thymic and autoimmune T cells. J. Gend. Specif. Med. 2, 33–40.[Medline]

Waring, J. F., Ciurlionis, R., Jolly, R. A., Heindel, M., and Ulrich, R. G. (2001a). Microarray analysis of hepatotoxins in vitro reveals a correlation between gene expression profiles and mechanisms of toxicity. Toxicol. Lett. 120, 359–368.[ISI][Medline]

Waring, J. F., Jolly, R. A., Ciurlionis, R., Lum, P. Y., Praestgaard, J. T., Morfitt, D. C., Buratto, B., Roberts, C., Schadt, E., and Ulrich, R. G. (2001b). Clustering of hepatotoxins based on mechanism of toxicity using gene expression profiles. Toxicol. Appl. Pharmacol. 175, 28–42.[ISI][Medline]

Webb, D. K., Moulton, B. C., and Khan S. A. (1990). Estrogen induced expression of the C-jun proto-oncogene in the immature and mature rat uterus. Biochem. Biophys. Res. Commun. 168, 721–726.[ISI][Medline]

Wolfinger, R. D., Gibson, G., Wolfinger, E. D., Bennett, L., Hamadeh, H., Bushel, P., Afshari, C., and Paules, R. S. (2001). Assessing Gene Significance from cDNA Microarray Expression Data via Mixed Models. http://brooks.statgen.ncsu.edu/ggibson/Pubs.htm. Accessed October 24, 2001.

Yallampalli, C., Rajaraman, S., and Nagamani, M. (1993). Insulin-like growth factor binding proteins in the rat uterus and their regulation by oestradiol and growth hormone. J. Reprod. Fertil. 97, 501–505.[Abstract]