Research Resource: Whole-Genome Estrogen Receptor α Binding in Mouse Uterine Tissue Revealed by ChIP-Seq

Abstract

To advance understanding of mechanisms leading to biological and transcriptional endpoints related to estrogen action in the mouse uterus, we have mapped ERα and RNA polymerase II (PolII) binding sites using chromatin immunoprecipitation followed by sequencing of enriched chromatin fragments. In the absence of hormone, 5184 ERα-binding sites were apparent in the vehicle-treated ovariectomized uterine chromatin, whereas 17,240 were seen 1 h after estradiol (E₂) treatment, indicating that some sites are occupied by unliganded ERα, and that ERα binding is increased by E₂. Approximately 15% of the uterine ERα-binding sites were adjacent to (<10 kb) annotated transcription start sites, and many sites are found within genes or are found more than 100 kb distal from mapped genes; however, the density (sites per base pair) of ERα-binding sites is significantly greater adjacent to promoters. An increase in quantity of sites but no significant positional differences were seen between vehicle and E₂-treated samples in the overall locations of ERα-binding sites either distal from, adjacent to, or within genes. Analysis of the PolII data revealed the presence of poised promoter-proximal PolII on some highly up-regulated genes. Additionally, corecruitment of PolII and ERα to some distal enhancer regions was observed. A de novo motif analysis of sequences in the ERα-bound chromatin confirmed that estrogen response elements were significantly enriched. Interestingly, in areas of ERα binding without predicted estrogen response element motifs, homeodomain transcription factor-binding motifs were significantly enriched. The integration of the ERα- and PolII-binding sites from our uterine sequencing of enriched chromatin fragments data with transcriptional responses revealed in our uterine microarrays has the potential to greatly enhance our understanding of mechanisms governing estrogen response in uterine and other estrogen target tissues.

Estrogens, including the endogenous ovarian hormone estradiol (E₂), interact with the nuclear estrogen receptor, estrogen receptor (ER)α, to modulate transcription rates of genes, leading to biological effects in target tissues (1). The mouse uterus is one such highly sensitive tissue with well-characterized responses to acute estrogen and as such has proven to be an effective model in which to study the biological and biochemical mechanisms of ERα-mediated response (2). Using the ovariectomized mouse uterus we have previously demonstrated profiles of transcriptional responses that underlie the biphasic biochemical and biological events that follow an acute dose of E₂ (3). Through these studies we have confirmed that responding genes are primarily seen initially (2 h) or later (24 h) after dosing, with some transcripts peaking at intermediate times (6–12 h). These patterns suggest that unique regulation of transcripts direct the observed early (hypertrophy, fluid uptake) or later (hyperplasia, epithelial cell proliferation) biological responses (3). These analyses have also indicated that E₂ is driving both up- and down-regulation of transcription rates. Additionally, studies using ERα or β-null mice indicate the uterine responses are mediated solely by ERα (3).

One aspect of responsiveness that has never been comprehensively evaluated in the uterine tissue, however, is the interaction between the ERα protein and target genes. Previous studies have used chromatin immunoprecipitation (ChIP) on a chip or sequencing of ChIP enriched chromatin fragments (ChIP-seq) approaches in cancer cells or engineered cell models to address the relationship between ERα chromatin interaction and response to E₂ (Refs. 4 and 5); reviewed in Ref. 6), or have evaluated interaction with candidate gene-regulatory sequences using ChIP-PCR (7, 8). This Research Resource reports our ChIP-seq analysis of ERα- and RNA polymerase II (PolII)-binding sites in ovariectomized mouse uterine tissue that is treated for 1 h with saline vehicle (V) or E₂. We have included our mapped sequences as well as our peak analysis and initial evaluation of the data for enrichment of ERα DNA-binding protein motif sequences. Although in general our findings mirror those from MCF7 cell ChIP-chip and ChIP-seq studies, our results are derived from an in vivo mouse model using comprehensive whole-genome sequencing rather that tiled promoter arrays, yielding a more relevant investigation of endogenous gene profiles.

Materials and Methods

Animals

Female C57bl/6J/129 mice were bred in our off-site colony housed at Charles River Laboratories (Wilmington, MA). For ChIP-PCR, stock C57bl6/J ovariectomized mice were purchased from Charles River Laboratories (Raleigh, NC). All procedures were carried out in accordance with National Institutes of Health guidelines for humane care and use of laboratory animals under an National Institute of Environmental Health Sciences (NIEHS) ACUC-approved protocol. For ChIP-seq, mice were ovariectomized at NIEHS at 16–23 wk of age, rested for 10–14 d, and then injected ip with 0.1 ml of saline (six mice) or 2.5 μg/ml estradiol (Research Plus Inc. Barnegat, NJ; in saline, six mice). Uteri were harvested 1 h after injections, snap frozen in liquid nitrogen, and then shipped to Genepathway, Inc. (San Diego, CA) for Factorpath analysis.

ChIP-Seq

ERα or PolII chromatin immunoprecipitation

Uteri were submerged in PBS + 1% formaldehyde, cut into small (∼1 mm³) pieces with a razor blade, and incubated at room temperature for 15 min. Fixation was stopped by the addition of 0.125 m glycine (final). The tissue pieces were then treated with a TissueTearor (Biospec Products, Bartlesville, OK) and finally spun down and washed twice in PBS. Chromatin was isolated from the sample by adding 5–10 ml lysis buffer containing piperazine-N,N′bis(ethanesulfonic acid); 1,4-piperazinediethanesulfonic acid, Igepal, phenylmethylsulfonyl fluoride, and Protease Inhibitor Cocktail, followed by disruption with a Dounce homogenizer. Samples were pelleted by centrifugation and resuspended in buffer containing Na deoxycholate, sodium dodecyl sulfate, and Triton X-100. Lysates were sonicated using a Misonix Sonicator 3000 (Misonix, Farmingdale, NY) equipped with a microtip to shear the DNA to an average length of 300–500 bp. Lysates were cleared by centrifugation, and the chromatin suspensions were transferred to new tubes and stored at −80 C. To prepare Input DNA (genomic DNA), two aliquots of 10–25 μl each (∼1/50 of each chromatin preparation) were removed and treated with ribonuclease for 1–2 h at 37 C, proteinase K for 3 h at 37 C, and 65 C heat for at least 6 h to overnight for de-cross-linking. DNA were purified by phenol-chloroform extraction and ethanol was precipitated. Pellets were resuspended in 1/5 10 mM Tris, 1 mM EDTA. Resulting DNA were quantified on a Nanodrop spectrophotometer (Thermo Scientific, Wilmington, DE). Extrapolation to the original chromatin volume allowed determination of the yield for each chromatin preparation, as measured by the DNA content. Before use in ChIP, protein A agarose beads (Invitrogen Technologies, Carlsbad, CA) were preblocked using blocking proteins and nucleic acids for 3 h. For each ChIP reaction, an aliquot of chromatin (20–30 μg) was precleared with 30 μl preblocked protein A agarose beads for 1–2 h. ChIP reactions were set up using precleared chromatin and antibody to ERα (sc-542; Santa Cruz Biotechnlogy, Inc., Santa Cruz, CA) or PolII phospho Ser5 (ab5095; Abcam, Cambridge, MA) in a buffer containing Na deoxycholate and incubated overnight at 4 C. Preblocked protein A agarose beads were added and incubation at 4 C was continued for another 3 h. Agarose beads containing the immune complexes were washed two times each with a series of buffers consisting of the deoxycholate sonication buffer, high-salt buffer, LiCl buffer, and Tris-EDTA buffer. An sodium dodecyl sulfate-containing buffer was added to elute the immune complexes from the beads, and the eluates were subjected to ribonuclease treatment at 37 C for 20 min and proteinase K treatment at 37 C for 3 h. Cross-links were reversed by incubation overnight at 65 C, and ChIP DNA were purified by phenol/chloroform extraction and ethanol precipitation. Quality of ERα ChIP enrichment was assayed by quantitative PCR (qPCR) using primers against candidate estrogen response element (ERE) in the Igf1 and Stat5a promoters (9). Quality of PolII ChIP enrichment was assayed by qPCR using primers against region in Intron 1 of the housekeeping genes Actin B and Gapdh. Input DNA was queried at the same sites in parallel.

ChIP sequencing

ChIP DNA was amplified by following the Illumina ChIP-Seq DNA Sample Prep Kit (Illumina, Inc., San Diego, CA) protocol. In brief, DNA ends were polished and 5′-phosphorylated using T4 DNA polymerase, Klenow polymerase, and T4 polynucleotide kinase. After addition of 3′-A to the ends using Klenow fragment (3′-5′ exo minus), Illumina genomic adapters were ligated and the sample was size fractionated (200–250 bp) on a 2% agarose gel. After a final PCR amplification step (18 cycles, Phusion polymerase; New England Bioloabs, Ipswich, MA), the resulting DNA libraries were quantified and tested by qPCR at the same specific genomic regions as the original ChIP DNA to assess quality of the amplification reactions. DNA libraries were sequenced on a Genome Analyzer II (Illumina). Sequences (36 nucleotide reads) were aligned to the mouse genome (National Center for Biotechnology Information Build 37.1/mm⁹) using Eland (Illumina pipeline) software. Alignments were extended in silico at their 3′-ends to a length of 130 bp, which is the average genomic fragment length in the size-selected library, and assigned to 32-nt bins along the genome. The resulting histograms were stored in Binary Analysis Results files. Data are deposited to Gene Expression Omnibus (GEO) accession no. GSE36455.

Quality control and ERα peak selection

Reads were first selected using a cross-correlation analysis as described in Ref. 10 and those that passed this qc, had no more than two mismatches, and mapped uniquely [mouse reference sequence (National Center for Biotechnology Information Build 37)] were used in further analysis. Estimated average fragment length is shown in Table 1. The Partek Genomics Suite (Partek, Inc, St Louis, MO) was used for detection of ERα peaks. Peaks were then filtered using intra- and intersample statistics, with a binomial test between sample and control, a Mann-Whitney U test for the separation of forward and reserve reads, and application of false discovery rate cut off of 0.05. Peak height was determined based on extending all mapped reads to the estimated average fragment length. The sum of the mapped (extended) bases under a given peak was normalized to ensure that the chipped count totals were scaled to the input count total. Average height for a given peak was then calculated as chipped count minus input count over peak width.

Table 1.

Characteristics of ERα ChIP-seq tags and results of mapping and peak calling

	V	1-h E2	Input
Estimated average fragment length	132 nt	140 nt	120 nt
Tags mapped	33.2 × 106	24.6 × 106	33 × 106
Peaks	5184	17,240
Genes within 100 kb of an ERα peak	7907	13,417
Mean ERα peak height	45	137
Genes within 100 kb of an ERα peak with PolII signal at TSS	4672	6519

nt, Nucleotides.

Analysis of PolII ChIP-seq

As a simple heuristic of whether a transcript is potentially active, ΔRPKM (reads per kb per million mapped reads; PolII chipped minus input) values were calculated for 500-nt bins across the genome. A transcript for which the annotated transcription start site (TSS) was located in a bin with ΔRPKM greater than or equal to 3 was considered to have PolII signal. Transcript entries were collapsed by locus to give a count of genes with PolII signal. To identify transcripts that are transcriptionally poised and/or up-regulated upon E₂ treatment, RPKM values were calculated for the promoter proximal (defined as −250 to +250 relative to the annotated TSS) and gene body [defined as +251 relative to TSS to transcription end site] regions of each transcript. The thresholds for differences and ratios in these calculated RPKM values (between gene regions and between treatments) are specified in Table 4.

Table 4.

Promoter-proximal PolII transcripts

	Transcripts (n = 29,208)	Genes represented (n = 22,611)
Too small	1247 (4.3%)	986 (4.4%)
Not (highly) up-regulated in E2	27,508 (94.2%)	21,299 (94.2%)
Up-regulated in E2, not poised in vehicle, active in vehicle	31 (0.1%)	25 (0.1%)
Up-regulated in E2, not poised in vehicle, not active in vehicle	62 (0.2%)	53 (0.2%)
Up-regulated in E2, poised in vehicle, active in vehicle	132 (0.5%)	111 (0.5%)
Up-regulated in E2, poised in vehicle, not active in vehicle	228 (0.8%)	178 (0.8%)

Promoter proximal (PP) = −250 to +250 relative to TSS; gene body (GB) = +251 (relative to TSS) to transcription end site (TES).

1. Minimum gene locus size (TSS to TES): 750 nucleotides.
2. Highly up-regulated in E₂: E₂ GB ΔRPKM − vehicle GB ΔRPKM ≥1 and E₂ GB ΔRPKM /vehicle GB ΔRPKM ≥3.
3. Poised in vehicle: vehicle PP RPKM/vehicle GB RPKM ≥2 and vehicle PP ΔRPKM ≥1.
4. Active in vehicle: vehicle GB RPKM ≥0.5.

ΔRPKM = ChIP sample − input sample.

Locations relative to genes

Locations of ERα peaks relative to annotated RefSeq TSS were calculated. Gene models for this analysis are based on RefSeq entries (annotations for the mm9 assembly) downloaded from the University of California Santa Clara Genome Bioinformatics site (http://genome.ucsc.edu/) on August 14, 2011. Peak-to-gene distances were determined according to the TSS of a given RefSeq entry. For calculating the total number of genes within a potential interacting window, all RefSeq entries with annotated TSS within 100 kb of a called ERα peak were considered (Table 1). When assigning a peak to a (unique) closest gene, multiple RefSeq transcripts from the same gene locus were collapsed, and the shortest distance was used (Supplemental Fig. 1 published on The Endocrine Society's Journals Online web site at http://mend.endojournals.org).

ERE motif and peak annotation

The sequences corresponding to the 17,000+ peaks identified as above from the 1-h E₂ sample were input into GADEM (a genetic algorithm guided formation of spaced dyads coupled with an EM algorithm for motif discovery) (11) for a seeded analysis with the ERE position weight matrix (PWM) constructed from 48 experimentally identified ERE (15 bp in length) (12) as the starting PWM. Further analysis was carried out with both V and E₂-treated peaks to evaluate the relative quantities of ERE motifs in the peaks and associated genes. Next, each peak was annotated with RefSeq transcripts, using a threshold distance of 100 kb between the given peak and the RefSeq TSS. The set of RefSeq transcripts for a given peak was then collapsed to gene identifications, and the average number of peaks per gene was calculated based on those assignments

Motif enrichment

The sequences from the 1-h E₂ peaks were divided into those containing ERE motifs (10,157) and those with no ERE motifs (7083). To identify enriched motifs in each category, both data sets were scanned with all PWM in the TRANSFAC database (Release 2010.1). For each PWM and each data set, numbers of binding sequences containing at least one binding site and the number of sequences without a binding site were counted, and the relative enrichment P value was calculated using the Fisher exact test.

Homeodomain transcription factor (Hox) motif scans

All 1-h E₂ peaks were scanned for Hox motifs using GADEM with TRANSFAC Hox9b PWM (accession: M014262) with the default parameters except –em 0 (without EM optimization). The sequences with Hox motifs were then compared with 9000 randomly selected sequences that have the same length (300 bp) and CG content as the putative Hoxb9-containing sequences. Both sets of sequences were then scanned with PWM from the TRANSFAC database for enriched motifs in the Hoxb9 set compared with the random set. Annotations for genes mapped to the regions with the Hox sites were used to select probes that were then clustered using Rosetta Resolver software to evaluate RNA expression data.

Comparison of ChIP-Seq and microarray datasets

To compare ERα recruitment to transcriptional responses, RefSeq probes with fold changes more than or equal to 2 vs. V at 2, 6, or 24 h (GEO no. GSE23072 and unpublished 6-h dataset) were compared with RefSeq probes within 100 kb of ChIP-seq peaks (combined V and 1-h E₂). Analyses were also carried out on individual time points with V and E₂ ChIP-seq peaks separated.

Functional analysis

Ingenuity Pathway Analysis (Ingenuity Systems, Inc., Redwood CA) software package was used to analyze enriched functions in gene lists.

ERα ChIP-PCR

Chromatin was isolated from frozen uterine tissue, and ERα was enriched by immunoprecipitation as previously described (9) with the following modifications. Sonication was increased to a total of 18 min (six repetitions of 3-min cycles 50% on/off), Dynabeads protein A (Invitrogen) was used to assemble the anti-ERα immunoprecipitation resin, beads were prewashed in PBS+5 mg/ml BSA, preblocked for 30 min in PBS/BSA+salmon sperm DNA. Chromatin was diluted 1:10 with homogenization immunoprecipitation (IP) buffer and precleared with blocked resin for 30 min. Anti-ERα antibody (Santa Cruz SC7202; 100 μl per 100 μl beads in 1 ml PBS/BSA) or normal rabbit IgG (50 μl per 100 μl beads in 1 ml PBS/BSA) was prebound to the resin for 30 min, washed in PBS/BSA, and then cleared chromatin was added to anti-ERα or IgG resins overnight. After washing three times in homogenization IP buffer, resin was eluted and DNA purified as previously detailed (9). Samples included three each of saline and 1-h E₂ treated. Isolated DNA was evaluated using real-time PCR as before (primers used: Supplemental Table 1). Values for each probe set were calculated using a standard curve of input DNA to determine the amount in each condition. A ratio of ERα IP to IgG IP represents the fold enrichment.

Results

ERα and PolII ChIP-Seq of uterus

Analyses were conducted on data derived from whole uterine tissues collected from ovariectomized mice 1 h after injection of saline V or E₂. These conditions were selected because they represent the baseline and the potential initial ERα-binding events that precede the transcriptional regulation observed in microarray studies beginning 1–2 h after administering E₂ (3). Surprisingly, in the V-treated condition, in the absence of hormone, 5184 ERα peaks were detected (Table 1), indicating the presence of numerous ERα-binding sites in the absence of estrogenic ligand. Similarly, previous studies have indicated the ability of ERα to interact with DNA sites in the absence of E₂ ligand (13, 14). When each ERα-binding peak was mapped to RefSeq-annotated transcripts within 100 kb (either direction), the 5184 peaks were associated with 7907 genes (Table 1). As might be anticipated, E₂ treatment increased the number of binding peaks to 17,240 peaks that were mapped to 13,714 genes. Most of the genes (7280) associated with the V peaks overlapped with those in the E₂-treated peaks. This analysis was derived by comparison with mouse transcript annotations, without regard to indications of transcriptional activity in the tissue. Therefore, to add relevance to our assessment, we evaluated PolII interactions in the same chromatin samples. This analysis showed that 4672 and 6518 genes in the V and E₂, respectively, are within 100 kb of an ERα peak and have PolII signal at their TSS (Table 1). The normalized mean ERα peak height increased approximately 3-fold (Table 1), indicating that both the numbers and the magnitudes of the ERα-binding sites are increased by E₂.

Next, locations of the ERα-binding sites were divided into categories (>100 kb distal (5′ or 3′), 100 to 10 kb 5′, 10 to 0 kb 5′, and 0 to 100 kb 3′) based on their locations relative to RefSeq genes. For this analysis two different methods were employed. In the first method, sites were assigned to the closest RefSeq TSS (Supplemental Fig. 1A, Analysis 1). The second method used the PolII ChIP-seq data to exclude Refseq genes considered inactive (without strong PolII signal at TSS; Supplemental Fig. 1A, Analysis 2, and Supplemental Fig. 1B). Interestingly, although E₂ more than tripled the number of ERα-binding sites in comparison with the V sample, the percentage of peaks in each of the categories was not significantly changed between V and E₂ using either relative mapping strategy (RefSeq-annotated TSS or limiting the RefSeq transcripts to those with observed PolII signal), indicating E₂ acted to recruit ERα to more sites, but overall did not alter the binding locations relative to genes. Comparison of the mapping methods (relative to all RefSeq TSS or only those RefSeq TSS with observed PolII signal) indicated similar relative locations of ERα; however, a greater percentage of the sites were observed to be more than 100 kb distal from RefSeq TSS with PolII signal. The ERα peaks are distributed such that 10–20% of the sites are adjacent to promoters (within 10 kb 5′), whereas many of sites are located distal (>100 kb) from genes (20–25% relative to closest RefSeq TSS; 40–43% relative to RefSeq TSS with PolII peak). Yet although most sites are not adjacent to promoters, the density (sites per bp) of peaks in the 10-kb region 5′ of TSS is greater than in other regions (e.g. ∼15% per 10 kb adjacent vs. approx. 25–40% per 100 kb 3′).

Motif analysis

To evaluate the consensus ER-binding motif in our in vivo dataset, the 17,240 sequences associated with the 1-h E₂ ERα ChIP-seq peaks were analyzed for ERE motifs using the motif discovery tool GADEM [A Genetic Algorithm Guided Formation of Spaced Dyads Coupled with an EM Algorithm for Motif Discovery (11)]. The computed consensus sequence of ERα binding in our dataset (Fig. 1A) matched an ERE motif computed from 48 experimentally identified ERE using GADEM (Fig. 1B), which agrees with the consensus ERE sequence (Fig. 1C) (1), indicating that ERα has preference for this reported consensus ERE in mouse uterine tissue. To evaluate the potential function of ERE sequences in our datasets, the number of peaks and RefSeq genes associated with ERE motifs were compared in the V- and E₂-treated samples. Interestingly, 35% of peaks and 44% of genes (mapped within 100 kb of peaks) had ERE motifs (Table 2), even when no ligand was present. Administering E₂ resulted in a significant increase in the percentage of peaks (59%) or genes (84%) with ERE motifs, as well as an increase of the average number of ERα-bound sequences from 1.7 sites to 3.3 sites per gene.

Fig. 1.

ERE motifs in WT 1-h E₂ ChIP-seq peaks. Motif discovery: ERE motifs were identified by A Genetic Algorithm Guided Formation of Spaced Dyads Coupled with an EM Algorithm for Motif Discovery (GADEM). A, ERE motif identified from 17,240 1-h E₂ peak sequences. B, ERE motif constructed from 48 experimentally identified ERE. C, Consensus ERE (1).

Table 2.

Genes with ERE motifs

	Peaks	Peaks with ERE motif	Genes (TSS <100 kb)	Genes with ERE motif	Average peaks/gene
V	5184	1804 (35%)	7907	3491 (44%)	1.7
1-h E2	17,240	10,157 (59%)	13,417	11,248 (84%)	3.3

Our data were further evaluated using the Igf1 gene (Fig. 2), which has been extensively characterized as an ERα target gene in the uterus (3, 9, 15–17). Igf1 expression is increased by E₂ in the uterus, and we have previously shown E₂-dependent recruitment of ERα to several ERE sites (9). Interestingly, the ERα ChIP-seq analysis indicates that the sites we previously evaluated (adjacent to the promoter, shown by arrows) were minor in comparison with a set of binding regions more than 50 kb 5′ of the promoters (Fig. 2A). Remarkably, two of these sites were bound by ERα in the absence of E₂. Using primers to these regions, we verified enrichment of ERα using ChIP-PCR (Fig. 2B). Motif analysis suggests the presence of ERE motifs in each of these sites in addition to the ERE we had previously reported. Interestingly, two of these 50-kb 5′-sites (peaks 2 and 4) have motifs that match a consensus ERE.

Fig. 2.

ERα-binding sites near Igf1. A, University of California Santa Cruz Genome Browser representation of analysis of ERα ChIP-seq binding sites of Igf1 gene. Several ERα ChIP-seq peaks from the V or 1-h E₂ sequences were mapped near the Igf1 gene. Two were adjacent to the two Igf1 TSS and were previously studied and validated by ChIP-PCR (9) (lower panel; shown by arrows). Five peaks 50 kb 5′ of the TSS were apparent in our ChIP-seq data set and were further analyzed for ERE motifs. The lower panel has the maximum scale on the ERα ChIP-seq tracks set to emphasize the smaller peaks. Peaks 2 and 4 contain consensus ERE: Peak 2, GGTCAtgaTGACC; Peak 4, GGTCAcctTGACC. B, Validation of ERα binding at Igf1 50 kb 5′-sites by ChIP-PCR. Primers flanking regions in five ChIP-seq peaks in the region 50 kb 5′ from the Igf1 TSS or closer to the TSS (−6215/−8270; −29/−2084) were used to analyze independent ChIP samples. Fold enrichments are calculated as the ratio of ERα IP to IgG IP. WT, Wild type.

To evaluate other motifs associated with the ERα-binding regions, the 17,240 1-h E₂ ChIP-seq peak sequences were divided into those with a predicted ERE motif (10,157 sequences) and those with no apparent ERE motif (7083 sequences), and queried using TRANSFAC PWM (12). The sets (ERE or non-ERE) were tested against each other to reveal significantly enriched motifs (Table 3 and Supplemental Table 2). Interestingly, in the 10,157 ERE-containing sequences, 50 motifs were enriched (P <10⁻⁵). The ERα motif was the most significant; other enriched motifs included other members of the nuclear receptor family, including peroxisome proliferator associated receptor-γ, TR, and vitamin D receptor, which bind to motifs containing ERE half-sites. Additionally, motifs for additional nuclear receptor family motifs were represented, including androgen receptor (AR), glucocorticoid receptor (GR), Couptf2, DAX1, retinoid-related orphan receptor-α, and estrogen-related receptor (ERR). In the 7083 non-ERE-containing sequences, 240 motifs were enriched (P <10⁻⁵) and included motifs for numerous transcription factors that are not in the nuclear receptor superfamily. Notably, there were more than 50 occurrences of homeobox (Hox) motifs. As such, the entire set of 17,240 sequences with ERα-binding sites from the 1-h E₂-treated sample was evaluated for the presence of Hox motifs. Approximately 3700 sites were found on 3200 sequences (Fig. 3A). The motif logo constructed from predicted Hox binding sites (18) using a Hox PWM is shown in Fig. 3A. Hox motifs lack complexity and are thus unable to provide a high degree of DNA sequence specificity (19). Specificity is dictated by binding of Hox cofactors such as Pbx and Meis (19); thus the sequences with Hox motifs were evaluated for other motifs that are indicative of Hox cofactors. As might be expected, TF motifs that include ERE full- or half-sites as well as Hox motifs were the predominantly enriched motifs. However, significant enrichment of Meis1 and Pbx1 was also seen (Supplemental Table 3).

Table 3.

Motif enrichment of 1-h E₂ ERα−binding sites: peaks divided into ERE- or non-ERE-containing sequences

Group	ERE-containing sequences (10,157)	non-ERE-containing sequences (7083)
No. of enriched motifs (P < 10−5)	50	240
Motifs identified	ERα most significant	Enriched motifs include HOX, CEBP, E2f, SP1, ELK, CDX, TAL1/E47, FOX
Distinctive characteristics	Other enriched motifs include nuclear receptors: PPARγ, TR, VDR, AR, GR, ERR, RORA, COUPTF2	More than 50 occurrences of HOX motifs

COUPTF2, Chicken ovalbumin upstream promoter transcription factor 2; PPAR, peroxisomal proliferator-associated receptor; RORA, retinoid-related orphan receptor; VDR, vitamin D receptor.

Fig. 3.

Hox motifs. A, 17,240 1-h E₂ ERα peaks (consisting of both predicted ERE and non-ERE) were scanned for Hox motifs. Approximately 3700 sites were found in about 3200 sequences (18.6% of peaks). B, Levels of different Hox factor or cofactor transcripts in uterine sample derived from microarray analysis. Log signal intensities for Hox family and cofactor transcripts in V- or E₂-treated (2, 6, or 24 h) are displayed. C, Estrogen regulation of genes with ERα binding and Hox motif. Genes associated with the 3700 Hox sites in the ERα ChIP-seq peaks were examined in our microarray data (V vs. 2-, 6-, or 24-h E₂) and clustered to show those with P < 0.001 and fold change greater than 2. These uterine genes are E₂ regulated at various time points, suggesting a role for Hox in ERα-mediated uterine responses.

The presence of Hox motifs on ERα-binding sequences was of potential significance because Hox proteins are known to be important in reproductive tract patterning and in reproductive functions of the uterus (18). A screen of Hox RNA in the ovariectomized uterus by microarray indicates expression of a number of Hox family members and cofactors (Fig. 3B). To investigate a potential functional role for Hox motifs in ERα-mediated uterine gene responses, our microarray data were examined. Genes with ERα-binding sites (within 100 kb) as well as Hox motifs, as indicated by our ChIP-seq and subsequent motif analysis, were selected for comparison (Fig. 3C). The patterns of up- or down-regulation exhibited by many of the selected genes are consistent with association between ERα binding, Hox motifs, and estrogen regulation. Supplemental Fig. 2 shows the positions of predicted Hox motifs identified in ChIP-seq peaks relative to ERα recruitment sites and PolII ChIP-seq data near several transcripts that are regulated by E₂ at indicated times.

Integration with microarray data

To comprehensively evaluate potential roles of the ERα binding in transcriptional regulation of genes, a comparison was made with known differentially regulated transcripts from our microarray datasets (GEO no. GSE23072 and unpublished 6-h dataset) and our ChIP-seq analysis. ERα-ChIP-seq peaks (6019) could be mapped to within 100 kb of genes on the microarray chip (Supplemental Fig. 3A). Of these probes, 46% (2739 sequences) showed at least 2-fold response to E₂ at 2, 6, or 24 h. Similarly, of the 6746 probes on the array showing at least 2-fold regulation after 2-, 6-, or 24-h E₂ treatment, 40% (2739) had an associated (within 100 kb) ERα binding peak in the ChIP-seq dataset. To evaluate whether the ERα-binding sites in our study either in vehicle-treated or 1 h of E₂ were more likely to be involved in regulation of the earlier transcript responses, the comparisons were also separated by each time (2 h, 6 h, and 24 h; Supplemental Fig. 3, B–E), however, the amount of overlap was not notably changed, indicating the interactions detected in our ChIP-seq analysis at 1 h are not limited to those associated with earlier transcript responses. A recent report has shown that in MCF7 cells, ERα-binding sites are significantly closer to up-regulated transcripts than to down-regulated transcripts (20). To examine whether this relationship is also seen in our uterine samples, we mapped ERα-binding sites from our 1-h E₂ ERα ChIP-seq dataset relative to transcripts that were increased or decreased at least 2-fold relative to V after 2, 6, or 24 h. Consistent with the observation in MCF7 cells (20), at 2 or 6 h, significantly more ERα-binding sites mapped close to (≤10 kb 5′ of TSS) up-regulated transcripts than down-regulated transcripts (Supplemental Table 4). The opposite relationship occurs for distal (>100 kb) sites. For transcripts regulated at 24 h, however, this difference was not apparent.

Promoter-proximal poised PolII

Although we have extensively studied rapid (1–2 h) transcriptional responses in the ovariectomized uterus (3), the mechanism allowing for rapid activation has not been described in detail. One mechanism described in in vitro systems is the presence of poised promoter-proximal RNA PolII, which is preloaded, enabling very rapid increase in transcript production (21, 22). To establish whether there was evidence of poised promoter-proximal PolII in uterine tissue, the PolII ChIP-seq data was first evaluated to select transcripts that have increased PolII over the gene body after E₂ treatment (highly up-regulated by E₂). Then the TSS/gene body ratios of these transcripts were evaluated (Table 4). Examples of transcripts that show evidence of poised promoter-proximal PolII were evident (Supplemental Table 5 and Supplemental Fig. 4). The potential functions of transcripts that appeared to exhibit promoter-proximal poised PolII were evaluated using Ingenuity Pathway Analysis (IPA; Supplemental Table 6). Analysis of the 111 poised and active in the V sample genes indicated enrichment of cell growth, cell cycle progression, and gene trans-activation functions. Evaluation of the 178 poised and inactive in the V sample genes suggested roles in tissue development in addition to cell cycle arrest and gene trans-activation. Interestingly, IPA analysis reveals that both sets of poised genes include transcription factor networks that are targets of p53 (Supplemental Table 6).

Overlap with MCF7 analysis

Next, we compared our results with those observed previously with ERα ChIP-seq of MCF7 breast cancer cells (5). Unlike our observation in the vehicle-treated uterine tissue, there are few ERα peaks in the V control MCF7 sample (Supplemental Table 7). Interestingly, despite a much lower number of sequence reads in the MCF7 dataset, there is significant overlap between the genes within 10 kb of the ERα peaks in the E₂-treated samples (Table 5 and Supplemental Tables 7 and 8). Analysis of the functions of the 548 genes present in our ERα ChIP-seq and the MCF7 ERα ChIP-seq datasets using IPA indicate that the most enriched functions were cancer, cell death, tissue development, cellular movement, and cellular growth and proliferation. (Supplemental Table 8).

Table 5.

comparison of MCF7/uterus ERα ChIP-seq

No. of genes with an ERα peak within 10 kb	Vehicle	E2
Mouse genes	1376	3261
Human genes	68	1812
Corresponding mouse and human genes	3	548
Probability	0.938	5.60 × 10−27

Note: comparison limited to 16,079 Homologene loci.

Validation by ChIP-PCR

Several-estrogen regulated genes selected from previous studies were evaluated for ERα enrichment in the ChIP-seq dataset and tested by ChIP-PCR of independent samples (Supplemental Fig. 5). Errfi1, Cyr61, Wnt4, and Cdkn1a transcripts are rapidly increased by E₂ (3, 23, 24). ERα interaction sites are apparent in the vicinity of the Errfi1 gene (Supplemental Fig. 5A). E₂-dependent enrichment of ERα at two of the sites was verified by ChIP-PCR, one adjacent to the first exon and one 3′of the gene. Interestingly, this transcript has two different potential first exons; RT-PCR using exon-selective primers indicate both are present in uterine RNA; however exon 2 is more robustly increased by E₂ (Supplemental Fig. 5A). ChIP-seq analysis indicated a peak of ERα binding 5′ of the Cyr61 transcript that was verified by ChIP-PCR (Supplemental Fig. 5B). An ERα peak in the first intron of the Wnt4 transcript was confirmed (Supplemental Fig. 5C) but not a smaller peak adjacent to the first exon. A peak in the first intron of the Cdkn1a transcript was validated by ChIP-PCR (Supplemental Fig. 5D). As was seen with the Errfi1 transcript above, Cdkn1a has two potential first exons. RT-PCR using selective primers indicated that only exon 2 is present in the uterine RNA (Supplemental Fig. 5D).

Discussion

In this Research Resource we report, for the first time, a comprehensive analysis of specific sites of interaction between ERα and RNA polymerase II and uterine chromatin in a physiological tissue model using a ChIP-seq approach. Our study includes interactions both in conditions without hormone as well as 1 h after E₂ injection. Most interestingly, even in the biologically quiescent situation of ovariectomy without hormone treatment, more than 5000 sites of ERα interaction are apparent, indicating significant levels of binding. This finding lends biological significance to early reports that ERα binding to ERE could be independent of E₂ (13, 14). Several previous analyses of ERα-binding sites used breast cancer cell lines but did not evaluate hormone-free conditions (6). Some studies included vehicle-treated cells but did not report detailed findings from the unliganded condition (4). Others have dismissed the minimal observed interactions as compromised peak calling resulting from copy number variations (5) or have attributed hormone-independent interactions to background (25). Comparison with our dataset demonstrated that few sites were present in a vehicle-treated MCF7 ERα ChIP-seq dataset (Ref. 5, Supplemental Table 6, and Table 6); further investigation will be required to assess whether estrogen-independent interaction of ERα with chromatin is a property of tissue models that cancer or cell models lack. Recruitment of other nuclear receptors, including glucocorticoid receptor (GR), vitamin D receptor, and progesterone receptor (PR) have been observed independent of their respective ligands in similar comprehensive ChIP-seq studies (Refs. 26 and 27 and Rubel, C.A., R. B. Lanz, R. Kommagani, H. A. Franco, J. P. Lydon, F. J. Demayo, submitted for publication).

Addition of E₂ did lead to a significant increase in the magnitudes and numbers of ERα interaction sites, as well as the numbers of sites per gene, and the percentages of binding peaks and associated genes that include ERE motifs. Together, these patterns suggest that in an unstimulated state, ERα is arranged in a manner that would allow acute response, by prerecruitment, and a potential manner of rapid signal amplification, upon hormone binding, via increased binding to basal sites as well as recruitment to additional sites. Site multiplicity would assure the specificity and the magnitude of required gene response. Analysis of PolII ChIP-seq sites revealed evidence of poised polymerase, which is also situated in a manner that could potentiate rapid response.

As has been reported now in numerous global ERα-binding studies (6), the sites of interaction are often quite distant from regions adjacent to promoters. Currently, most nuclear receptor whole-genome ChIP studies are derived from in vitro cell culture systems, whereas ours uses living tissue. ERα-binding sites in ovariectomized 2-h E₂-treated mouse liver have been evaluated by ChIP-chip (28, 29), also revealing the prevalence of binding distal (>10 kb) from ResSeq TSS annotations. Similarly, mouse caput epididymis tissue was used in a recent androgen receptor (AR) ChIP-seq study. Interestingly, this AR study indicated that 75% of the nearly 20,000 sites of interaction were more proximal (within 20 kb) to TSS (30) than observed with ERα in our study and in other comprehensive ERα ChIP reports. One important biochemical difference between our uterine model and the epididymis analysis is that our samples represent the initial acute phase of ERα interaction occurring within 1 h of ligand administration, whereas the epididymis sample represents a homeostatic state of continuous exposure to endogenous circulating AR ligands. The initial ERα interactions that our analysis captures may, with time, progress to ERα accumulation at sites more proximal to TSS. Our future studies will evaluate this possibility.

Our observations emphasize that ERα interaction occurs at distal sites under a normal physiological situation in a biologically relevant tissue, and although similar initial reports from in vitro cell models were counter to expectations built from early gene-by-gene studies (31), future hypotheses and studies of ERα-mediated gene regulation will need to account for these distal arrangements. To account for the distal sites, mechanisms involving looping of DNA to allow interaction with promoter-proximal regions have been proposed (32). The presence of unrevealed transcriptionally active regions more proximal to these ERα sites of interaction remain a possibility, because global analysis of noncoding and micro-RNA, for example, are not yet well described. Future datasets that include comprehensive RNA-seq as well as histone modification ChIP-seq datasets aligned to our ERα and PolII ChIP-seq analysis will aid in addressing such questions.

Determining how ERα interactions correlate with mechanisms of response remains as an enormous challenge. Overlap with transcriptional regulations that have been demonstrated using PolII ChIP-seq, microarray, and RT-PCR are suggestive, but are naively based on annotations of the sites of ERα-ChIP interaction according to binding locations. Traditionally such relationships are evaluated using cis-reporter assays; however, the complexity of ERα-binding site arrangements observed in this and previous studies, at distances far greater than are feasible to study with such approaches, added to the complexities associated with tissue or cell specificity of response mechanisms that are difficult to reproduce in in vitro models, emphasizes the potential limits of such approaches. Using the example of the Igf1 transcript (Fig. 2), our previous analysis in uterine tissue demonstrated E₂-dependent recruitment of ERα to ERE sites adjacent to the promoters of this gene and in an intron (9); however, our study did not appreciate that, in addition to these sites, there is strong recruitment of ERα to sites 50 kb 5′ of the promoters. Interestingly, there are peaks of PolII binding at these distal sites as well (Supplemental Fig. 2D). Discrete PolII peaks distal from genes have also been described in enhancer regions (33, 34). We have observed similar regions of ERα and PolII peaks distal from some other estrogen-regulated transcripts (e.g. see Fos, Inhbb, Pgf, Dcn, Igfbp3, Cdk1, Dhcr24; Supplemental Fig. 2), suggesting that these enhancer regions may be important in uterine tissue responses. Future studies will focus on elaborating this interesting arrangement. Neither our previous study nor the knowledge that ERα is also interacting with these distal sites assures mechanistic understanding of the mode of regulation of Igf1 transcript in the uterus. We know that ERα interaction is enhanced after E₂ injection, and that mutation of the first zinc finger of ERα disrupts the interaction as well as Igf1 induction (Ref. 9 and data not presented). Interestingly, ERα protein and Igf1 transcript are both detected in the liver, yet injection of E₂ does not change the liver Igf1 transcript levels, despite the estrogen-dependent enrichment of ERα on one of the −50-kb distal sites (peak 4 in Fig. 2; and Grontved, L., unpublished data). Thus, follow-up studies to delineate mechanisms of response involving ERα interaction sites identified using the ChIP-seq approach present technical challenges, but the strength of the observations lays in their derivation from an authentic biological system that reflects such complexity.

One unique finding from our study was the coenrichment of motifs for other nuclear receptor family members in ERα-binding regions with ERE motifs (Table 4 and Supplemental Table 2), which in some instances is likely due to similarity in motifs (ERR and TR) but also suggests potential colocalization of ERα and other nuclear receptors (AR or GR). A second related finding was the enrichment of nonnuclear receptor motifs in ERα-binding regions that lack ERE motifs (Table 3 and Supplemental Table 2), consistent with the transcription factors that interact with these motifs acting as tethering factors, to mediate indirect ERα interaction with DNA. Some, such as SP factors and AP1, have previously been demonstrated to fulfill such roles (35); others have not been investigated as ERα tethering-factors. Interestingly, binding motifs for Hox protein were especially prevalent. Hox transcription factors are encoded by a series of genes in four separate clusters (A, B, C, and D), each cluster encoding nine to 13 Hox genes (18). Previous studies have noted the presence and importance of Hox family members, in particular, Hoxa10 and Hoxa11, for normal uterine patterning and function, as demonstrated by the uterine dysfunction of Hoxa10- or Hoxa11-null mice (18). Much of the work on roles of Hox factors in uterine function has focused on response to progesterone, because the phenotypes in the Hoxa10- and Hoxa11-null mice were associated with the progesterone-dominated processes of implantation and decidualization (36). Other studies have indicated disruption of female reproductive tract development as a result of diethylstilbestrol or environmental endocrine disruptor exposures that involves inappropriate ERα-mediated imprinting of Hoxa10 expression in Mullerian tissue (37–40). Our findings, including indications that uterine genes with ERα binding and potential Hox motifs are also regulated by E₂, suggest that Hox may be involved in ERα-mediated uterine responses. A recent study reported that the Hox cofactor Pbx acts as a pioneering factor to facilitate ERα binding in breast cancer cells, and enhances E-dependent proliferation (41), consistent with our suggestion of a role for Hox factors in estrogen responses. Interestingly, several of the genes that are regulated by E₂, bind ERα, and have Hox motifs have been demonstrated to have roles in reproduction, including progesterone receptor (PR) (42), ERα (43), Igf1 (44), and Itgb3 (45). Some, such as Itgb3 (46), PR (47), and Cdkn1a (48) have been demonstrated to be regulated by Hox factors. Our future work will further evaluate potential roles for Hox in estrogen responses in uterine tissue.

It has become evident that emerging ChIP-seq datasets will cause reevaluation of models of nuclear receptor response. The prevalence of distal binding sites, considered together with the incomplete overlap between binding and expression data, point toward more complex roles for ERα in mechanisms of transcriptional regulation, or potential novel roles unrelated to transcription. Our uterine ERα and PolII ChIP-seq datasets provide maps of basal and initial E₂-dependent sites of interaction in a biologically and physiologically relevant model. The added context provided by comparisons and correlation to extensive gene expression studies as well as similar findings reported using Progesterone Receptor ChIP-seq from uterine tissue (Rubel et al., submitted) promise to provide useful tools to investigators that will advance understanding of molecular mechanisms and biological processes involved in uterine responses to ovarian hormones and comparison with other responsive target tissues.

NURSA Molecule Pages^†:

• Nuclear Receptors: ER-α;

• Ligands:17β-estradiol.