1 Department of Biochemistry and Molecular Biology, and National Food Safety and Toxicology Center, Michigan State University, East Lansing, Michigan 48824
2 Department of Biostatistics, Virginia Commonwealth University, Richmond, Virginia 23298
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
uterus; microarray; GeneChip; temporal expression
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Despite these diverse functions, the uterus continues to be a preferred model for elucidating the effects of estrogenic compounds. Temporal changes in circulating estrogen levels are critical to the normal progression of the uterine cycle and stimulate events associated with implantation in the event of ovum fertilization including the stimulation of cell cycle progression, transcription and translation of specific proteins, cellular water intake (imbibition), vascularization, recruitment of immune cells, and cell and tissue architecture modifications. ER is the predominant isoform in the uterus (74). Studies in mice with targeted disruptions have confirmed that ER
is required for the early hyperemia and water imbibition responses induced by estrogen, as well as later stimulation of DNA synthesis, induction of specific estrogen-inducible transcripts, and uterine growth and morphogenesis (22).
The profound complexity of estrogen signaling underscores the limitations of current assays aimed at the detection and characterization of exogenous estrogens. Current in vitro assays do not adequately address the consequences of ER binding and subsequent gene expression (84). By contrast, in vivo assays are based on the ability of the estrogenic substance to induce uterine weight in an immature and/or ovariectomized animal (57). Although this protocol has been used for decades (42), a lack of specificity has been observed (38), despite the incorporation of complementary uterine endpoints (i.e., epithelial cell height and number, gland number) (57, 83). Although in vitro tests have some clear advantages, they underrepresent the sophisticated processes intrinsic to an intact animal, and conversely in vivo studies offer enhanced ability to extrapolate responses to humans, but frequently focus only on a limited number of uterine-based endpoints.
To address these issues, recent studies have examined female reproductive tract responses to estrogen using microarrays (48, 60, 75, 76). However, the studies vary substantially in design and conditions used, as well as in the specific transcripts under consideration. Furthermore, since the dosing protocol for measuring increased uterine weight (i.e., several consecutive daily doses) may not be optimal for capturing transcriptional responses, a time course experiment examining estrogen response is desired. In the present study, a GeneChip oligonucleotide array time course was performed using the standardized uterotrophic protocol to determine the temporal effects on gene expression responses elicited by 17-ethynylestradiol (EE). EE is a pharmaceutical with enhanced oral bioavailability (7) that induces an estrogenic transcriptional profile similar to that of the endogenous estrogen E2 (34). Rigorous statistical approaches were used to identify treatment-induced changes in gene expression responses while accounting for variability between replicates. Significant responses were clustered and then associated with functional annotations extracted from public databases to identify pathways involved in the uterotrophic response.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
RNA Extraction, Sample Labeling, and GeneChip Hybridization
Uteri were homogenized individually as previously described (23), and total RNA was extracted with TRIzol (Life Technologies, Rockville, MD) according to the manufacturers instructions, with yields of 3.0 µg RNA/mg uterine tissue. RNA was further purified using RNeasy spin columns (Qiagen, Valencia, CA) according to the manufacturers instructions and eluted in 20 µl water. RNA yield and purity were assessed by spectrophotometer and denaturing gel electrophoresis. Ten micrograms total RNA from each uterus was then separately reverse transcribed, made into cRNA, fragmented, and hybridized to either Test2 or Test3 GeneChip arrays (Affymetrix, Santa Clara, CA) to verify target quality, and then to Mu11KSubA GeneChip arrays, which contain 6,523 probe sets directed at murine targets, according to the manufacturers protocols.
GeneChip Data Acquisition and Clustering
Signal values and detection confidence levels were obtained using MicroArray Suite 5.0 (MAS5; Affymetrix) using the default settings. In accordance with MAS5 absolute call default parameters, a transcript was considered to be "present" (P; expressed) within a sample if the detection P value for the corresponding probe set was below 0.04 and to be "marginal" (M) if the P value fell between 0.04 and 0.06, with all other observations considered to be "absent" (A). Hierarchical and K-means clustering were performed using GeneSpring 4.2 (Silicon Genetics, Redwood City, CA).
Obtaining Updated Gene Annotation Information for Probe Set Accession Identifiers
A gene name for each probe set was obtained by using the GenBank accession identifier (GenBank ID) associated with the Affymetrix probe set identifier (probe set ID) to query a local copy of the National Center for Biotechnology Information (NCBI; http://www.ncbi.nlm.nih.gov) mouse UniGene database information (build 118), using a Perl script that returned and saved the corresponding cluster name (UniGene name), cluster identifier (UniGene ID), and gene abbreviation. mRNA RefSeq accession IDs were also saved where available. For GenBank IDs not found within UniGene, the probe set exemplar sequence was obtained from NetAffx (http://www.affymetrix.com), a BLAST search was performed against all other mouse sequences in GenBank, and a probable gene identity was assigned in cases where the five closest sequence matches all had an expect value of e 10-20 and were representative of a single UniGene cluster. Probe sets are referred to in the following text by their official gene abbreviation as obtained from NCBI.
RefSeq IDs were then used by another Perl script to query a local copy of the NCBI LocusLink database information, returning information of interest including LocusLink ID, RefSeq category, chromosomal location, Gene Ontology IDs and descriptions, and relevant NCBI PubMed references. The LocusLink ID of the NCBI-MGD human homolog was also noted from the mouse LocusLink record, which in turn provided a link to additional public functional information available in the Online Mendelian Inheritance in Man (OMIM; NCBI) and GeneCards (Weizmann Institute) databases. For probe sets with a UniGene name but lacking a RefSeq ID, a modified version of the Perl script used the gene name to query the LocusLink database information, and records were inspected manually to verify that the desired record had been returned. As well, additional sequence and annotation information about each probe set was obtained from NetAffx. All Perl scripts are available upon request, in accordance with software availability guidelines proposed by the International Society for Computational Biology (http://www.iscb.org/pr.shtml#software). All tables are available in Excel format at http://www.bch.msu.edu/zacharet.
In cases in which two different probe sets interrogated the same GenBank ID (referred to here as a probe set double), Microsoft Excel was used to identify probe set doubles and to determine the concordance of absolute calls on each array between probe set doubles and to examine links between these results and the probe set suffix flags (defined in Supplemental Table A, available at the Physiological Genomics web site,1
and at http://www.bch.msu.edu/zacharet) of the probe set double members.
Raw GeneChip data is available in the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo) as time course experiment GSE280 and sample files GSM3870 to GSM3890. Probe set IDs, gene names and abbreviations, K-means cluster assignments, fold change calculations, and other supplemental information is available in Supplemental Table B.
Quantitative RT-PCR
RNA was isolated from uteri obtained from two additional replicates of the time course experiment described above. Verification by SYBR Green quantitative real-time PCR (QRT-PCR) of each replicate used samples consisting of equal amounts of uterine RNA pooled from five identically treated animals, so that for each treatment condition, data were collected from two separate pools each representing five different animals. Data for each time point were therefore collected separately from each of the two pools, which represent ten animals in total, and then averaged.
For each pool, 1 µg total RNA was reverse transcribed using SuperScript II (Invitrogen, Carlsbad, CA) and an anchored oligo-dT primer in a 20-µl reaction as described by the manufacturer. Then, 0.6 µl of cDNA was added to a 30-µl PCR reaction containing 0.1 µM each of forward and reverse gene-specific primer, 3 mM Mg2+, 0.33 mM dNTPs, 0.5 IU AmpliTaq Gold, and 1 x SYBR Green PCR buffer (Applied Biosystems, Foster City, CA). Primers (Table 1) were selected using Primer3 (62) to obtain an amplicon of 125 bp length with Tm of 60°C. Amplicons were selected from within the Affymetrix exemplar sequence region and were compared by BLAST to all other sequences in GenBank to ensure specificity. Default cycling parameters were employed, using an ABI7000 Prism Sequence Detection System (Applied Biosystems).
|
For the GeneChip data, an empirical Bayes model (21), modified to account for an experimental design with multiple comparative variables, was used as a screening tool to identify genes that have the highest probability of being regulated due to treatment and/or time. That is, the empirical Bayes model in Ref. 21 addressed a two-sample comparison (treatment vs. control), whereas with the current experiment it is of interest to compare treatment vs. control across multiple time points. The modified approach reduces the dimensionality of the data down to a single summary statistic per gene, comparing all responses to the average response at the 2-h time point per gene, while simultaneously accounting for multiple testing issues. The empirical Bayes model estimates the a posteriori probability of being active (called p1z) with respect to treatment and/or time for each gene. In other words, the p1z value is a single summary statistic for each probe set, in which a high p1z value is indicative of a high level of response (either induction or repression) of a probe set relative to the variability in the response. The false-positive rate for a given rejection region was estimated using a Bayesian false discovery rate (FDR) (21), and was found to be 0.52% using p1z > 0.99 as the selected threshold.
An ANOVA model was then fit to the genes with p1z 0.99, and each response was classified as being active due to either a treatment effect, a time effect, or a treatment-by-time interaction. Responses with a significant (P < 0.01) treatment-by-time interaction were considered to be of interest, as were gene responses with only a significant treatment effect. Conversely, gene responses with only a significant time effect (indicative of strong vehicle-dependent or circadian responses), or responses in which the treatment and the time effects were separately significant but their interaction was not, were rejected.
For the QRT-PCR data, ß-actin-normalized responses were standardized to the average response at the vehicle 2-h time point for each gene to adjust for gene-specific differences in expression level. An ANOVA model was fit to the data using the logarithm-transformed standardized responses from the two independent pooled samples, and gene responses were also classified as having a treatment effect, a time effect, or a treatment-by-time interaction. In addition, the Pearson correlation between the temporal responses observed by QRT-PCR and microarray prior to time-matched normalization for each gene was calculated using Microsoft Excel. A technical paper describing the details of these statistical methods is available at http://www.bch.msu.edu/zacharet.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
GeneChip Analysis
Summary statistics.
A modified empirical Bayes screening approach (21) was used to identify active responses, followed by an ANOVA to remove vehicle-induced and circadian effects. This approach reduces the data to a single summary statistic (p1z value) for each probe set that is indicative of the overall level of the response relative to its variability. The p1z value reflects the responsiveness and reproducibility of a change in gene expression over time. Responsive genes were ranked according to their p1z value and a threshold value of p1z > 0.99 was initially chosen to identify 15% of the probe set results for further investigation. The FDR, or false-positive rate, which addressed multiple testing issues, indicated that 0.5% of probe sets passed the threshold, or
4 of the nearly 900 responses, may falsely be classified. The ANOVA did not consider multiple testing, and therefore a stringent threshold of P < 0.01 was used.
Probe sets with an absolute call by MAS5 of either present (P) or marginal (M) were relatively uniform among the arrays (45.4 ± 7.9% and 2.8 ± 0.4%, respectively). In addition, 17.3% of probe sets were called P on all 21 arrays, while 28.9% were called absent (A) on every array. Hierarchical clustering in GeneSpring using default settings indicated that the untreated sample, the 24 h EE-treated samples, and the 3 x 24 h EE-treated samples formed three distinct groups, whereas the 8 h and 12 h EE-treated samples together formed an additional group, and the 2 h EE-treated samples could not be distinguished from the vehicle-treated samples (not shown).
A gene name and associated publicly available annotation were linked to each probe set where possible by submitting the GenBank ID for the probe set to the NCBI mouse UniGene database (build 118, December 2002; Fig. 1). We found that 57.2% of the probe sets had an associated gene name, whereas 36.7% were denoted as ESTs. The remaining 6.1% were no longer represented in UniGene, but for those that were retained by subsequent screening steps, a tentative gene identity was assigned based on BLAST searching of the GenBank database. In some cases several probe sets were associated with the same UniGene identifier, and as a result, 6,261 unique UniGene IDs were found to be represented on the GeneChip (i.e., 4.0% redundancy).
|
GeneChips often have GenBank IDs represented by multiple probe sets, with each probe set associated with a different suffix flag. On the Mu11KSubA array, 433 GenBank IDs are represented by two probe sets (referred to here as doubles), which facilitates a comparison of results obtained from members of a double. All "g" probes have a corresponding "none" probe, and this comprises the majority of the probe set doubles ("g-none" 87.8%). The remaining pairs were "f-i" (5.8%), "i-s" (2.5%), "r-s" (2.3%), "f-r" (1.4%), and "i-r" (0.2%).
Absolute "present" (P) and "marginal" (M) calls were compared across the 21 arrays to evaluate the performance of probe pair doubles. At the extremes, 29.1% of doubles had at least 20 identical calls, while 5.1% of doubles had only one or zero identical calls. Upon further examination, no association between the proportion of identical calls within the double and the specific pair of suffix flags within the double was evident (not shown). Interestingly in 22 cases, one probe set within the double was called P/M on all 21 arrays while the other probe set was called A on every array (i.e., all-P/M-all-A probe set doubles). In 16 of the 17 cases (94.1%) of a "none-g" double, the "none" array was all-A while the "g" array was all-P/M. Similarly, for the 34 doubles in which there were between 16 and 19 differences in absolute call across the arrays, there were 20 doubles (58.9%) with a highly A "none" probe and a highly P/M "g" probe, but only two doubles (5.9%) with the reverse pattern. Overall, this suggests that the "none" probe set, designed using the ideal probe set rules, is not able to detect the presence of the corresponding transcript, whereas the permissive "g" probe set interrogates a transcript which may be similar but not identical to that indicated by the probe set ID.
Cluster analysis: complete dataset.
Temporal patterns of gene expression within the data were initially analyzed by K-means clustering using the correlation distance metric on the entire dataset for the five time points. The average of the duplicate responses at each time point divided by the average of the corresponding duplicate vehicle responses was examined (i.e., time-matched normalization). Eight K-means clusters best described the main temporal patterns based on visual inspection (Table 2). Cluster H, the largest cluster, contained the least responsive probe sets and had the lowest proportion of all-A probe sets (17.8%). All clusters had a majority of probe sets corresponding to named genes in UniGene (53.659.5%), with the exception of cluster D (46.6%), which had a majority of probe sets being represented by ESTs.
|
Screening to identify most active gene responses.
To eliminate negligible or confounding responses, an empirical Bayes model was used to generate p1z values ranging from 0 to 1 for each probe set, a summary statistic that indicated the degree of the response relative to its reproducibility as a measure of probe set variability. At a threshold of 0.99, 881 probe sets were retained, of which nearly 500 were associated with a p1z greater than 0.999, indicative of a robust response with low variability between replicates. Of the 881 significant probe sets, 537 (62.2%) had a gene name in UniGene, a percentage slightly higher than the proportion of named probe sets represented on the array as a whole (57.2%), suggesting that probe sets found in the present experiment to be significantly regulated were not much more likely to be annotated than randomly selected probe sets on the array.
The second ANOVA screening step was designed to identify significant (P < 0.01) treatment effects or treatment-by-time interactions, thereby eliminating vehicle or circadian effects for which the only significant effect would be time. Of the 392 that were considered to have a significant treatment effect or treatment-by-time interaction, 247 (63.0%) had an associated UniGene gene name (Fig. 1), again indicating that the proportion of uncharacterized ESTs among the highly responsive final probe set list was no less than the proportion in the entire probe set population on the Mu11KSubA arrays. The complete list of 392 highly responsive probe sets is available in Supplemental Table B and at http://www.bch.msu.edu/zacharet.
Quantitative RT-PCR.
QRT-PCR was used to verify selected response patterns in uteri collected from experiments using the same protocol. A subset of 26 probe sets was randomly selected for verification that spanned the range of p1z values. Overall, there was good agreement between the GeneChip and QRT-PCR results for responsive probe sets and lower correlation in cases where no response was evident, since values then tended to fluctuate without pattern by both methods. Furthermore, in a few cases no correlation value was calculated since no transcript was detectable by QRT-PCR. Therefore, in addition to correlation analysis, a qualitative assessment based on visual inspection was also provided (Table 3).
|
At the other extreme, Saa4, Fcgr1, and Krtap8-2 were all undetectable in every sample when measured by either technique. Nxf1 was also absent or nearly absent in all samples, consistent with the microarray data. By contrast, only QRT-PCR was able to detect Tbx6, C2ta, and Crygs with a treatment effect at or near P = 0.05, whereas these responses were muted or not evident in the GeneChip data. Mapk1 was also absent or nearly absent in all samples by microarray, whereas slight induction at 12 h was measured by QRT-PCR.
Repression of Tgfb3, Lepr, Capn10, and Ltbr was more pronounced in QRT-PCR measurements, and statistical correlation between the two measures was moderate, although qualitatively there was good agreement. Pdi2 responses for the two measures were highly correlated but more pronounced in the GeneChip data. Comparable results were also observed with Sftpd, C3, Farsl, Idb3, Vegfb, and Ccnb1. Csf1, S100a6, Idh2, and Fkbp10 were detectable by both methods but not greatly affected by treatment. Overall, responses were qualitatively similar, while statistical differences appeared to be due to variation between the duplicate measurements in isolated cases or to the higher sensitivity of QRT-PCR in detecting some transcripts.
Cluster analysis: reduced dataset.
K-means clustering requires that the number of clusters be specified a priori. Visual inspection of 215 K-means clusters indicated that the time-normalized data for the reduced list of probe sets was best described using seven clusters (Table 4 and Fig. 2). Viewing the corresponding non-time-normalized data revealed, as expected, that overall these probe sets were very responsive to estrogen treatment but showed little response to vehicle alone. Attempts to further stratify the data into subclusters confirmed the clustering of only a single temporal gene expression pattern. While clusters J and L both contained some responses that were induced at both 8 and 12 h, seven clusters categorized the observed temporal responses more completely than did six. Clusters corresponding to late responses (i.e., clusters M, N, and O) had the highest proportion of probe sets with UniGene annotation information (72.078.6%; Table 4).
|
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
Cell cycle.
Estrogen treatment of sexually immature or ovariectomized rodents with low circulating estrogen levels initiates synchronized uterine cell proliferation (10). Estrogen causes quiescent (G0) cells to enter the G1 growth phase, followed by DNA synthesis (S phase), a second phase of growth (G2), and finally mitosis (M), after which the daughter cells start a new G1 phase. Although many cell cycle-related changes are nontranscriptional events, others are wholly or partially regulated at the mRNA level.
Early induced transcripts (212 h) serve essential roles in the G1/S phase transition, whereas those directly involved in DNA synthesis or in the G2/M or M/ G1 transitions were upregulated at 24 h and possibly sustained to 3 x 24 h (Supplemental Table C). This is consistent with reports that only 3% of cells are in S phase at the time of dosing, whereas 75% have progressed to S phase by 15 h after E2 treatment (10), and that the rate of DNA synthesis is maximal by 24 h (28, 41). Few responses involved in cell cycle control or DNA synthesis were regulated as rapidly as 2 h (i.e., as were members of cluster I) or after maximal uterine proliferation was achieved at 3 x 24 h (i.e., as were members of cluster O).
Gadd45a (cluster I), an early response, provides a crucial link between the p53-dependent cell cycle checkpoint and DNA repair. Its levels are highest in G1, with a sharp drop during S phase, and it is upregulated following estrogen treatment (48, 60, 75). Cyclin D2 (Ccnd2; cluster J), which was also upregulated in the G1/S transition, is induced at mRNA and protein levels in the early proliferative stage of the human endometrium (11). Studies have also shown that cyclin D2 is highly induced at 28 h after estrogen injection and returns to baseline levels by 12 h (3). As well, the cdc2 activator cyclin B1 (Ccnb1; p1z = 0.964) was upregulated late, consistent with its reported upregulation during G2/M (85). Cdc2a (cluster N) was induced at later time points, although this has been reported not to be reflected at the protein level (82). Moreover, cyclin G2 (Ccng2; p1z = 0.999), which is upregulated in the late S phase (33), was repressed 10-fold at all treatments between 83 x 24 h. To our knowledge, other cell cycle-related transcripts (Supplemental Table C) have not been previously reported to be affected by estrogen.
Chromosome replication is an integral part of the cell cycle, and several EE-regulated genes are involved in DNA synthesis including Nme1 (cluster L), a nucleoside diphosphate kinase important in dNTP synthesis (75); the two subunits of ribonucleotide reductase, Rrm1 (cluster K), which is typically undetectable in quiescent cells (51); a thioredoxin-like transcript, Txnl2 (cluster J) (45); and thymidine kinase Tk1 (cluster M). Tk1 upregulation is associated with proliferating cells and serves as a marker for the onset of S phase (60, 64).
Several protooncogenes, including Fos, Jun, and Myc (78), show rapid, well-characterized estrogen inducibility in the uterus but were not represented on the array. Myb (cluster N), however, is upregulated at the G1/S phase transition during cell proliferation (71) and was upregulated at 3 x 24 h in the present study. In addition, the Myb binding protein Mybbp1a (cluster J) was also upregulated, possibly to assist Myb, which may play a critical role in cell proliferation and development.
RNA and protein synthesis and modification.
Uterine transcription increases 68 h after estrogen exposure (5). Several transcripts with roles in RNA synthesis were significantly altered following estrogen treatment (Supplemental Table D). Novel responses identified in cluster J (812 h) include: arginine methyltransferase (Hrmt1l2), which methylates H4 histones and may be essential to pre-mRNA processing (47); small nuclear ribonucleoprotein Snrnpa, which functions as a spliceosome component to excise introns from pre-mRNA; EBNA1 binding protein Ebp2, which plays a role in pre-rRNA processing, and poly(A) binding protein Pabpn1, which is required for efficient poly(A) tail formation and control of tail length; and Paf53 (p1z = 0.973), an activating subunit of RNA polymerase I, which was induced as previously reported (75). Importantly, the use of global GeneChip signal normalization suggests that in samples where overall transcript levels are increased, the intensity of the signals on these arrays may be underestimated. When numerous responses are identified that are not estrogen regulated, however, these may be used to normalize overall signal intensity levels across arrays.
EE also regulated several amino acid synthesis and transfer genes including cysteinyl-tRNA synthetase (Cars), alanyl-tRNA synthetase (Aars), and asparagine synthetase (Asns) in cluster J and asparaginyl-tRNA synthetase (Nars) in cluster L. Phenylalanine-tRNA synthetase (Farsl, p1z = 0.678), seryl-tRNA synthetase 1 (Sars1; p1z = 0.248), and arginyl-tRNA synthetase (Rars; p1z = 0.846) transcripts displayed nearly identical patterns of induction but with greater variability. No response was observed for other tRNA synthetase probe sets [aspartyl-tRNA synthetase (Dars; p1z = 0.938), seryl-tRNA synthetase 2 (Sars2; p1z = 0.504), tyrosyl-tRNA synthetase (Yars; p1z = 0.900), or lysyl-tRNA synthetase (Kars; p1z = 0.131)]. EE induction of histidyl-, glutamyl-, lysyl-, phenylalanine-like-, and seryl-tRNA synthetases as well as asparagine synthetase were also previously reported (75). In addition, the dehydrogenase Mthfd2 (cluster M), which may provide the formyltetrahydrofolate for formylmethionyl-tRNA synthesis, was also induced (75). Several chaperonin transcripts [i.e., Hspa8, Hspa4, and Hspa5 (also known as BiP); clusters J and L] were also induced consistent with previous reports (17, 58), as well as numerous translation initiation factors [i.e., Eif2a, Eif2s2, Eif4a1, Itgb4bp (also known as Eif6), Eif1a, and Eif2b1].
Protein modification or degradation genes such as peptidyl arginine deiminase Pdi2 (cluster M), which converts arginine to citrulline residues, and transglutaminase Tgm2 (cluster N), which links proteins through peptide bond formation and is also implicated in apoptotic body packaging, were upregulated. Cystatin B (Cstb; cluster J), an inhibitor of apoptosis (76) and of the cathepsin family of thiol proteases, was also induced, as was one of its target cathepsins (Ctsb; cluster L). Meanwhile Spi1-3 and Spi1-5 (cluster O), members of the serine protease inhibitor family with similarity to 1-antitrypsin, and proteasome subunits Psmb3, Psmd2, Psma3, Psmb6, and Psmb7 (clusters J and L) exhibited upregulation.
Matrix metalloproteinase (MMP) family members are involved in structural changes associated with the cycling human endometrium (65). MMPs degrade extracellular matrix components such as proteoglycans, fibronectin, elastin, casein, and gelatins and are involved in the breakdown of extracellular matrix during reproduction and tissue remodeling. Mmp7 (cluster O) was induced, whereas other members of the MMP family including 2, 12, 14, 15, and 24 were unaffected by EE treatment in this study. None of the tissue inhibitor of metalloproteases (TIMP) isoforms was represented on the array.
Immune and complement activation.
Early uterine responses to estrogen include characteristic inflammatory responses such as edema, enhanced vascular permeability, eosinophil infiltration, and subsequent production of inflammatory mediators (14). Eosinophil infiltration is also stimulated at estrus, with eosinophils undergoing degranulation during ovulation and characteristically releasing eosinophil peroxidase following estrogen exposure (52). Recruited eosinophils are not required for induction of uterine wet weight, protein synthesis, or epithelial complement C3 synthesis, but may play a role in endometrial stromal remodeling (54).
Decreases in progesterone levels during the uterine cycle, rather than direct action of estrogen, may stimulate cytokine release, attraction of eosinophils, release of the degradative MMPs, vasoconstriction, and hypoxia. Hypoxia in turn induces mediators such as vascular endothelial growth factor (Vegf), which is angiogenic and may promote edema by enhancing vascular permeability (14). However, the increase in uterine neutrophil and macrophage content is independent of progesterone or its receptor, confirming the critical role of estrogen (72). Moreover, the chemokine responsible for estrogen-induced eosinophil infiltration was recently identified as eotaxin (26).
Immunologically relevant genes induced by EE treatment (Supplemental Table E) include chemokine orphan receptor Cmkor1 (cluster I) (75), two members of the suppressor of cytokine signaling (SOCS) family Socs3 (cluster I) and Socs1 (cluster J), interleukins Il25 (cluster L), and Il17 (cluster M), interleukin-4 receptor Il4ra (p1z = 0.999) (58), the protooncogene tyrosine kinase Lyn (cluster N), which is required in immunoglobulin-mediated signaling, the well-characterized estrogen-inducible transcript for polymeric immunoglobulin receptor (Pigr; p1z = 0.952), formerly referred to as secretory component (67), and the monocyte chemotactic protein Ccl2 (also known as MCP1 or JE) (cluster L) (12). A probe set highly similar to macrophage migration inhibitory factor (Mif; cluster L), a lymphokine that stimulates immune cell activation and cytokine production, was also induced in the present study at 83 x 24 h (60).
Complement component factor precursor H2-Bf (cluster N) (also known as Factor B), which stimulates the proliferation of activated B-cells and is expressed at high levels in the luteal phase of the human and rodent uterine cycle (29), was also induced. The active H2-Bf cleavage product associates with the complement C3 degradation product C3b to form C3 convertase of the alternative complement pathway. However, complement component factor Cfi (cluster O), which inactivates complement components including C3b, was also significantly upregulated. C3 itself was upregulated at 3 x 24 h also, although associated with a low p1z value (p1z = 0.131) due to variability. C3 is a well-characterized ERE-regulated response that is involved in phagocytic and immunoregulatory processes (58), but the significance of the concurrent upregulation of C3 as well as both the H2-Bf activator and Cfi inactivator is confounding. Furthermore, the antigen Cd59a, or protectin (cluster I), which inhibits the terminal step of the complement activation cascade, and the activation of complement component C1r (cluster I), which is involved in activation of the first step of the classic complement system, were significantly repressed by EE. This complex transcriptional regulation of the complement pathways is poorly understood and requires further examination at the protein level and in situ studies to identify the cell types involved.
Other responses.
Several genes representing a multitude of pathways and functions consistent with uterine physiology were found to be affected by EE treatment (Supplemental Table F). For example, upregulation of proapoptotic genes [i.e., Egr1, Klf4, Cdkn1a (also known as p21), lymphotoxin B receptor (Ltbr), Pea15, Tia1] may contribute to the attenuation of cell proliferation as maximal growth is achieved, similar to the apoptotic phase that follows endometrial proliferation in the cycling uterus (32, 66). Secreted frizzled-related protein Sfrp2 (cluster I), which appears to function as a Wnt receptor and may have a role in apoptosis (17), was downregulated. The repression of other apoptotic inducers (i.e., Bcl2-like 11, caspase 2 and 6, Bok) suggests complex interactions between cell types within the proliferated uterus.
Ion transporters [i.e., cystic fibrosis transmembrane regulator Cftr (59), amiloride binding protein Abp1, or diamine oxidase (cluster O)] critical to uterine fluid balance were induced by EE. Cftr and the epithelial sodium channel (ENaC, or Scnn1; not represented on the Mu11KSubA array) isoforms regulate chloride and sodium balance, respectively (9). Furthermore, the ion channel Fxyd5 (cluster J), was induced, and although its ion transport capacity is poorly characterized, it is the homolog of human dysadherin, which inhibits E-cadherin (Cdh1; not represented on the array) function, an important component of epithelial cell adhesion (36). Mucin Muc1 (cluster N), an anti-adhesion molecule induced by estrogen through an ERE-independent mechanism (86), and another inhibitor of E-cadherin, increased steadily from 8 h to 3 x 24 h. Muc1 has been shown to be important in the structure and function of the uterine luminal epithelial layer (8).
The hypoxia-inducible factor Hif1a (cluster K), which is implicated in angiogenesis, apoptosis, and energy metabolism, was upregulated in agreement with a previous report (16), while the poorly characterized hypoxia-induced gene Hig1 (cluster K) was also induced. Regulation of Hif1a in the uterus is not well understood, but a direct role of hypoxia in inducing Hif1a in the estrogen-stimulated uterus has been questioned (16). By contrast, the rapid induction of the ERE-containing vascular endothelial growth factor, Vegf, in the uterus is well established (16, 35), although in the present study Vegfa was highly variable (p1z = 0.052). Vegf induces vascular permeability, and its inhibition abrogates the estrogen-induced uterine edema response (61).
Protein S (Pros1; cluster I), which inhibits blood clotting, and the tissue plasminogen activator Plat (cluster J), a serine protease that activates the fibrinolytic enzyme plasmin, were both highly repressed. No change was observed in the plasminogen activator inhibitor known as Pai1 or Serpine1 (p1z = 0.846), while two other probe sets with important roles in the cycling uterine endometrium, the urokinase plasminogen activator Plau and the potent vasoconstricting endothelin Edn1 (65), were not represented on the array.
The dramatic changes in the proliferating uterus have extensive energy requirements. Responses induced to meet this need included the late upregulation of the carbonic anhydrase Car2 (cluster M), which is involved in many biological processes due to its role in the reversible hydration of carbon dioxide. Creatine kinase Ckb (cluster J), which has well-characterized rapid estrogen inducibility (40) and is important in generating ATP in tissues with high and fluctuating energy demands, was also upregulated (34, 48). In addition, probe sets for the ATP transporter Slc25a5 or ANT2 (cluster L) were found to be induced, further facilitating energy needs (1, 60).
A Proposed Model for Arginine and Ornithine Utilization
Integration of these results with previous reports provides compelling evidence of the convergence of multiple pathways regulated by estrogens that support the uterotrophic response. Although not fully elucidated, Fig. 4 depicts estrogen regulation of arginine and ornithine utilization and a model for EE-induced promotion of uterine proliferation and its subsequent mitigation.
|
As maximal uterine growth is achieved, marked induction of the polyamine acetyltransferase Sat (p1z = 0.952) and the arginase Arg1 (cluster O) at 3 x 24 h reduce ornithine and arginine availability to slow proliferation. Sat is the rate-limiting enzyme in the polyamine degradation pathway, and its involvement in uterotrophy has been demonstrated by the uterine hypoplasia in mice overexpressing Sat (55). Induction of uterine arginase activity is well established as a classic biomarker for estrogen action (24); however, it has also been reported to be induced by Tgfb1 (not represented on the array), which itself is known to be rapidly induced by estrogen (69). Although the biological significance of Arg1 mRNA (230-fold at 3 x 24 h) and enzyme induction has not been previously characterized, results from this study indicate an important role in the depletion of ornithine as maximum induction of uterine weight is realized. In addition, Arg1 induction, mediated by decreased ornithine levels or by the transcriptional activator c/EBPß or the growth factor Tgfb1 (not measured in the present study), may also curtail macrophage nitric oxide synthesis (20), thus diminishing its support of uterine proliferation.
General Conclusions
Estrogen elicits a myriad of responses from multiple pathways that culminate in the uterotrophic response. This study provides statistically rigorous gene expression data that support decades of molecular and physiological research into uterine function and its response to estrogens. Despite the comprehensiveness of this study, many important genes were not represented on the GeneChip, and therefore it was not possible to place all responses into mechanistic or biological context. Moreover, limited information regarding RNA stability, timing of translation, posttranslational modification, and protein-protein interactions, as well as the sparse amount of gene annotation information, further confounded the interpretation of the temporal pattern of groups of genes that may share a pathway relationship. Nevertheless, several gene expression changes have been placed into cellular response context, thus providing more predictive bioindicators of estrogen action in the uterus. Overall, the uterus provides an ideal model for systematic assessment of the effects of estrogens at multiple levels, while also being a key target of estrogen action that displays numerous physiological outcomes.
![]() |
ACKNOWLEDGMENTS |
---|
GRANTS
Funding for this study was provided by National Institute of Environmental Health Sciences Grant ES-011271 (to T. R. Zacharewski and C. Gennings) and by fellowships to K. C. Fertuck from the Society of Toxicology and the Michigan State University College of Natural Science.
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: T.R. Zacharewski, Michigan State Univ., Dept. of Biochemistry and Molecular Biology, Biochemistry Bldg., Wilson Road, East Lansing, MI 48824-1319 (E-mail: tzachare{at}msu.edu; http://www.bch.msu.edu/zacharet/).
10.1152/physiolgenomics.00058.2003.
1 The Supplementary Material for this article (Tables AF) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00058.2003/DC1.
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|