In SilicoAnalysis Identifies a Novel Role for Androgens in the Regulation of Human Endometrial Apoptosis

Rigorous in silico comparisons of a gene set extracted from molecular phenotyping of human endometrium from normoovulatory women ( 15 ) with data sets of androgen-regulated genes identified in rat uterus, decidualized human cells, and androgen-responsive prostatic cell lines (Supplemental Table 1 and Supplemental Fig. 2) resulted in the identification of only 15 genes, hereafter referred to as the endometrial androgen target gene set ( Table 1 ). Consistent with the inclusion of genes identified by ChIP array ( 11 , 12 ) in our screening strategy, consensus androgen response elements were identified in the promoter regions of all candidate genes (Supplemental Fig. 3). Sequences with homology to consensus binding sites for other proteins implicated in steroid-dependent changes in gene transcription, including estrogen and progesterone receptors, activator protein-1, and SP-1 were also represented in the same promoter regions (Supplemental Fig. 3). mRNA encoded by all 15 of the androgen target gene set were detected in total tissue homogenates of endometrial tissue obtained during the proliferative phase ( Fig. 1 and data not shown) when immunoexpression of AR is restricted to the stromal compartment (Supplemental Fig. 1). Fourteen were also detected in extracts of first-trimester decidua (TRH, not detectable, Fig. 1 , and data not shown). Expression of phorbol-12-myristate-13-acetate-induced protein 1 ( PMAIP1 ) mRNA was higher in proliferative compared with midsecretory phase endometrium ( P < 0.01), whereas the opposite result was obtained for monoamine oxidase type A (MAOA) ( Fig. 1 ). Using commercially available antibodies, we were able to confirm expression of proteins encoded by PMAIP1 , MAOA , cAMP response element-binding protein (CBP)/p300-interacting transactivator 2 ( CITED2 ), prune homolog 2 ( PRUNE2/BNIPXL ), and CD44 using sections of human endometrium and decidua; all four proteins were expressed in the stromal compartment during the proliferative phase (Supplemental Fig. 4). Subcellular patterns of expression in primary hESC mimicked those in vivo with CITED2 detected in the nucleus, PRUNE2 localized to the cytoplasm, and PMAIP1 in both compartments (Supplemental Fig. 5). Expression of candidate androgen-regulated gene mRNA in total tissue extracts from human endometrium and decidua as determined by quantitative RT-PCR. Samples were homogenized from functional endometrium recovered during the proliferative (P) and midsecretory phases (MS) as well as from first-trimester decidua (Dec). Concentrations displayed relative to those in proliferative phase in each case (n = 4–6 per group). During in vitro incubation the primary hESC retained the phenotype of the stromal cells within the functional layer of the proliferative phase endometrium being immunopositive both CD10 [a stromal cell marker ( 22 )] and AR; the latter was also detected as a full-length protein on Western blots (Supplemental Fig. 5). We confirmed that expression of AR mRNA was maintained up to and including passage 5 but thereafter declined (not shown), and therefore, all subsequent evaluations were undertaken on primary hESC at passages 3–5. Expression of 12 of 15 of the endometrial target gene set was detected in primary hESC ( Fig. 2 ); concentrations of mRNA encoded by three putative AR target genes [ EVA1 , gastrin ( GAST ), TRH ] were too low to be investigated further. For nine of 15 of the genes [ PMAIP1 , SHC SH2 binding protein 1 ( SHC BP1 ), platelet-derived growth factor receptor-like ( PDGFRL ), acyl-CoA synthetase short-chain family member 2 ( ACSS2 ), matrix metallopeptidase 10 (stromelysin 2/ MMP10 , CITED2 , MAOA ] incubation of cells with DHT resulted in a significant reduction in their mRNA at one or more time points. The exception was CD44 , which was increased at 2 h, returning to control levels at 8/24 h ( Fig. 2 ). Time-dependent changes in androgen-regulated gene expression in primary hESC. Cells were incubated with vehicle ( gray bars ) or 10 −8 m DHT ( black bars ) for 2, 8, and 24 h (n = 6 each time point). Concentration of mRNA was quantified by quantitative RT-PCR and expressed as fold change compared with time-matched vehicle-treated hESC. Note that with the exception of CD44, incubation of cells with DHT resulted in either a significant reduction in concentration of mRNA or no significant change but with a trend to a reduction. For most genes, changes in mRNA expression were time dependent with the most striking change in total concentrations at 2 h for eight of 12 of the genes. *, P < 0.05; **, P < 0.01; ***, P < 0.001. Pathway analyses revealed 12 of 15 of the androgen target gene set interacted, directly or indirectly, on one network associated with androgen-regulated cell function ( Fig. 3 ) and an association with cell death ( Table 2 ). Additional bioinformatic analysis of subnetwork interactions formed between the androgen target gene set highlighted a relationship between AR-CD44-endothelin receptor B and gastrin associated with response to cell stress. A further association was identified between AR-CITED2-CD44 and PMAIP1 in control of apoptosis ( Table 3 ). Pathway analysis of endometrial androgen target gene set. Note 12 of 15 androgen-regulated genes identified by in silico screening (indicated with multicolored circles ) were predicted to interact in a single Metacore pathway centered on AR; intermediate molecules are indicated to show putative intermediate signaling molecules. Key shows the functional classification of the target genes and the arrows indicate predicted regulation ( red , negative; green , positive). Gene ontology processes associated with the androgen candidate gene set identified by androgen candidate gene ontology identified using Metacore Note that the term target indicates the number of candidate genes/proteins/compounds (objects) in a data set that are associated with a given network/process. The P value (the probability of a random intersection) indicates a measure of relevance of the intersection between a gene/protein and an entity in a particular ontology. The lower the P value, the higher is the nonrandomness of finding such intersection. The Z-score ranks the networks with regard to the number of objects present in the networks. The higher the Z-score, the higher the number of objects from the data set. Subnetworks within the candidate gene set and associated processes identified by androgen candidate gene ontology identified using Metacore Note that the term target indicates the number of candidate genes/proteins/compounds (objects) in a data set that are associated with a given network/process. The P value (the probability of a random intersection) indicates a measure of relevance of the intersection between a gene/protein and an entity in a particular ontology. The lower the P value, the higher is the nonrandomness of finding such intersection. The Z-score ranks the networks with regard to the number of objects present in the networks. The higher the Z-score, the higher the number of objects from the data set. GO, Gene ontology. The impact of DHT on apoptosis was investigated using a caspase-3/7 fluorescent assay. Caspase activity was not detected in control hESC cultures or after treatment of cells with DHT alone (not shown). When apoptosis was induced by treatment of cells with staurosporine (30–500 n m ), addition of DHT significantly reduced caspase activation. Addition of E2 had no effect and simultaneous addition of E2 and DHT partially abrogated the antiapoptotic impact of the DHT ( Fig. 4 A). Treatment of hESC with DHT alters apoptosis and proliferation. A, Apoptosis as measured by caspase-3/7 assay in cells treated with staurosporine. Note addition of DHT had a significant impact (*, P < 0.05) on the rate of apoptosis when compared with controls but that the impact of DHT was blunted in the presence of E2 (n = 4). B, Wound-healing assay. Addition of DHT had a significant impact (*, P < 0.05) compared with its vehicle control (ethanol), whereas flutamide alone had no impact compared with vehicle (methanol) and pretreatment of cells with flutamide blocked the DHT-dependent reduction in wound closure. DHT, 10 −8 m ; flutamide, 10 −5 m ; n = 6. To determine whether exposure to androgens could have a direct impact on migration of hESC, a wound-healing assay was performed using cells pretreated with vehicle or flutamide that were subsequently incubated with or without DHT for 24 h ( Fig. 4 B). A significant reduction in the rate of wound healing was detected when cells were incubated with DHT alone ( P < 0.05), whereas addition of flutamide alone had no effect pretreatment of cells with flutamide abrogated the impact of DHT ( Fig. 4 B).

Expression of AR and enzymes capable of local biosynthesis of androgenic ligands has been well documented in human endometrium, but little is known about the direct impacts of androgens on gene expression in this dynamic tissue. In this study, rigorous in silico comparisons of genomic data sets derived from normal human endometrium ( 15 ) with those identified as androgen regulated by array analysis of uteri from DHT-exposed rats ( 14 ), androgen-treated human decidualized cells ( 16 ), and genes identified as containing regions within their promoters capable of binding ligand activated androgen receptors ( 11 , 12 ) and has enabled us to identify, for the first time, a group of androgen-regulated genes expressed in primary hESC. Our strategy included data sets based on ChIP-on-chip ( 11 , 12 ) conducted using in prostate cancer-derived epithelial cells treated for short duration with androgen. We included the ChIP data sets because they provided a stringent, unbiased way of focusing on genes that were direct targets of androgen action; unfortunately, ChIP data sets for androgen-treated endometrial cells were not available. Only a small number of genes were identified as common to all of these data sets reflecting differences in tissue/cell context. One unexpected finding was that treatment of primary hESC with DHT resulted in a transient but significant reduction in expression of most of the mRNA encoded by our androgen target gene set. Consistent with the likely differences in availability of modulators of androgen action (coactivators/corepressors) between the AR-positive cells from endometrium (primary stroma) and those used for the ChIP arrays (transformed AR positive prostatic epithelial cells), four genes ( ACSS2 , CITED2 , MAOA , PPFIBP2 ) showed DHT-dependent changes in expression that were opposite in the two cell types [ Table 1 ( 11 , 17 )]. Notably, the regulation of three of our androgen target gene set, namely PRUNE2 , PMAIP1 , and ACSS2 , has never been investigated in the context of endometrial function. Promoter analysis identified multiple binding sites for both AR and Sp1 and pathway analysis (see Fig. 3 ) supported a role for Sp1 in mediating AR-dependent changes in expression of EDNRB and ACSS2. In this study we focused on the impact of androgens on stromal cells recovered from the functional layer of the endometrium during the proliferative phase at a time when AR is not expressed in epithelial cells (Supplemental Fig. 1). However, we did detect expression of MAOA, CD44, CITED2, PRUNE2, and PMAIP1 in endometrial epithelial cells by immunohistochemistry, and therefore, further studies are required to determine whether androgens also have an impact on their expression in this cell type. Analysis of the androgen target gene set with Metacore software (GeneGo) revealed a strong association with processes regulating programmed cell death ( Table 2 ). Apoptosis is an important regulator of endometrial function, with reports that it occurs predominantly in the late secretory and menstrual phases associated with alterations in the expression of Bcl-2 and Bax ( 23 ). Two members of our target gene set, PRUNE2 (BMCC1 ) and PMAIP1 , encode proapoptotic proteins not previously investigated in the context of endometrial function. PRUNE2 ( BMCC1 ) is highly expressed in the nervous system; although some studies have reported androgen regulation in prostate cancer cells ( 24 ), this has been disputed ( 25 ). PMAIP1 , also known as NOXA (Latin for damage), encodes a Bcl-2 homology 3-only member of the Bcl-2 family of proteins. The protein has been implicated in controlling cell death pathways in several cell types by virtue of competitive binding to proteins including the antiapoptotic proteins Bcl-2 protein A1 Bcl-xL and MCL1 ( 26 ). In neuroblastoma cells, gene knockdown of PMAIP1 significantly reduces apoptosis ( 26 ). Expression of PMAIP1 can be repressed by glucocorticoids in lymphoblasts of children with acute lymphoblastic leukemia ( 27 ), consistent with detection of a putative glucocorticoid response element in our TRANSFAC analysis. Conversely, introduction of AR into AR-negative prostate cancer cells is reported to up-regulate expression of PMAIP1 ( 28 ), highlighting the importance of cell context in the regulation of this gene. Although initial studies reported that PMAIP1 is a primary p53 response gene, it is now apparent that multiple signals can induce its expression (reviewed in Ref. 29 ). Reduced expression of PRUNE2 and PMAIP1 in the endometrium in response to androgens is consistent with reports that the amount of apoptosis is very low in the endometrium of patients with PCOS ( 30 ), a condition associated with elevated circulating androgens, elevated expression of endometrial AR, and infertility ( 6 ). CD44 , the only gene identified as up-regulated by androgen action in our primary stromal cells, has also been implicated in regulatory pathways involving the tumor suppressor p53. Its overexpression antagonizes the tumor-suppressive apoptotic and growth-inhibitory actions of p53 in mammary epithelial cells ( 31 ). Stromal cell decidualization during the secretory phase is associated with reduced expression of AR in the functional layer, but expression in the basal compartment is maintained (Supplemental Fig. 1). These temporal and spatial differences in the pattern of expression of AR could result in variations in expression of androgen-regulated genes in the endometrium and may protect basal stromal cells from apoptosis. In addition to those genes implicated in control of apoptosis and/or associated with the activities of the tumor suppressor p53 genes, several of the genes we have identified as androgen regulated in endometrial stromal cells have also been implicated in development of cancers. For example, SHCBP1 (SHC SH2 binding protein 1) has been cited as one of a panel of six markers for the diagnosis of early cervical cancer ( 32 ). PDGFRL , a gene that encodes a secreted protein with significant sequence homology to the ligand-binding domain of platelet-derived growth factor receptor-β, has been proposed as a tumor suppressor ( 33 ). CITED2, a CBP/p300-dependent transcriptional coactivator, was associated with increased expression of matrix metalloproteinase (MMP) 13 and enhanced invasiveness of colon cancer cells ( 34 ). All three genes have been detected in uterine tissue. For example, Kim et al. ( 35 ) reported that expression of SHCBP1 was down-regulated in endometrium recovered during the proliferative phase from women with PCOS compared with controls, a result consistent with our novel data demonstrating a striking and significant reduction in hESC treated with DHT. Talbi et al. ( 15 ) have reported that expression of PDGFL and CITED2 were cycle dependent, and studies in mice identified Cited2 as one of a group of genes up-regulated in the uterus 4 h after progesterone treatment ( 36 ). However, we believe the current study is the first to report that these genes might be targets for androgen action within the stromal compartment of the human endometrium. Data herein provide novel insight into the mechanisms that might contribute to the increased risk of endometrial cancer observed in women with PCOS ( 7 ). Endometriosis is a chronic condition characterized by the presence of endometrial tissue outside the uterine cavity and, like PCOS and endometrial hyperplasia, has been associated with a progesterone-resistant phenotype ( 37 ). The synthetic androgen, Danazol (17α-ethinyl testosterone) is an effective treatment for pain associated with endometriosis but is no longer used due to unacceptable side effects ( 38 ). Although disturbances in androgen-dependent gene expression have not been investigated in the context of this disease, it is notable that both AR and 5α-reductase enzymes have been detected in pelvic endometriosis ( 39 ). Altered expression of three of the genes in our androgen target set has also been recorded in women suffering from endometriosis. Cells recovered from menstrual effluent of women with endometriosis are reported to have increased expression of CD44 splice variants. Cells from the endometrium of women with endometriosis are reported to exhibit increased adherence to peritoneal mesothelial cells ( 40 ) and reduced expression of MAOA and CITED2 ( 37 ). It has been suggested that alterations in expression of MMP in endometriotic lesions would favor disturbances in tissue invasion/remodeling within the endometriotic tissue ( 41 ); expression of MMP10 has not been investigated to date and would merit further investigation in these patients. In conclusion, using rigorous bioinformatics to compare data sets between species and between tissues, we have identified a small cohort of androgen-responsive genes expressed in the human endometrium. Pathway analysis and functional assays suggest androgen-dependent changes in gene expression may have a significant impact on stromal cell migration and survival in this tissue. These novel data provide the platform for further investigations on the role of circulatory or local androgens in the regulation of endometrial function that will require studies on intact tissue in which the dynamics of stromal-epithelial interactions are maintained. These data identify androgens as candidates in the pathogenesis of common endometrial disorders including PCOS, endometrial cancer, and endometriosis.

Microarray, ChIP, and published gene data sets (Supplemental Table 1) were recovered from the Gene Expression Omnibus [ GSE4888 ( 15 ), GSE7868 ( 12 )] and published data sets ( 14 , 16 ). The strategy for comparison of data sets is outlined in Supplemental Fig. 2. The GSE4888 data set included raw signal intensities; global normalization to the 75th percentile was undertaken and all subsequent calculations were done using log2-transformed values. Fold change and false discovery rate was calculated using the SAM (Significance Analysis Microarray software developed at Stanford University and available free to academic users by registering at http://www-stat-class.stanford.edu/∼tibs/clickwrap/sam.html .). For the data set generated from rat uterus ( 14 ), gene identifiers were mapped on to the human ortholog. Data generated from array analyses of endometrium of normoovulatory women ( 15 ) were processed to identify those genes differentially expressed between the proliferative [AR immunoexpression intense in stromal cells ( 1 , 4 )] and midsecretory phases; this comparison was not made in the original study ( 15 ). The final endometrial androgen target gene set was generated by filtering for entries overlapping between all the data sets derived from the studies in both human and rat endometrium ( 14 – 16 ) and human prostatic cells ( 11 , 17 , 18 ) (Supplemental Fig. 2). Promoter regions of genes in the endometrial androgen target gene set ( Table 1 ) were screened for transcription factor binding sites using MAPPER ( http://bio.chip.org/mapper ( 19 ); using sequences 10 kb upstream and 4 kb downstream from the transcription start site of each gene as retrieved from Ensemble 54 (build NCBI36). Gene ontology and biological pathway analyses were performed using MetaCore version 5.4 build 19940 (GeneGo, Inc., St. Joseph, MI). Endometrium androgen target gene set identified by stringent in silico screening of publicly available data sets ID, Identification. The predicted change in expression in response to androgens is based on studies in prostatic cells ( 22 , 23 ). Endometrial tissues were collected by suction curette (n = 14, Pipelle; Laboratoire CCD, Paris, France) or after hysterectomy (n = 10) from women with regular menstrual cycles who had not received hormonal treatments during the 3 months before tissue collection. Samples were staged based on standard histological criteria, the patient's reported last menstrual period, and circulating estradiol (E2) and progesterone levels at the time of collection. First-trimester decidua (n = 4; 8–12 wk gestation) were collected from women undergoing the surgical termination of pregnancy. Written informed consent was obtained from all subjects and ethical approval granted by the Lothian Research Ethics Committee. Tissues were fixed in 4% neutral buffered formalin, immersed in RNAlater (Ambion, Austin, TX), or used for isolation of primary cells. Primary human endometrial stromal cells (hESC) were purified from proliferative stage samples as previously described ( 20 ). Cells were maintained at 37 C in DM-RPMI 1640 [containing 10% heat inactivated fetal calf serum, penicillin (100 U), streptomycin (0.1 mg/ml), Fungizone (1 μg/ml) (Life Technologies, Inc., Paisley, UK), l -glutamine (2 m m ), and gentamicin (0.02 mg/ml)]. For gene expression studies, they were seeded at 2 × 10 5 /well in six-well culture plates and allowed to reach 80% confluence, incubated for 24 h in phenol red-free DM-RPMI 1640 (10% charcoal stripped fetal calf serum and supplements as above) followed by 24 h in phenol red-free, serum-free medium. Treatments were performed using serum-free medium containing vehicle (ethanol) or DHT (10 −8 m ) for 2, 8, or 24 h (n = 6) for measurement of mRNA or 8, 24, or 48 h (n = 6) for measurement of protein. hESC derived from proliferative phase endometrium express both ERα and ERβ [( 21 ); Collins, F., unpublished observations]; therefore, DHT was chosen for these studies because, unlike T, it is not metabolized to E2. To ensure that concentrations of DHT were maintained throughout the culture period, media were refreshed after 8 h. Total RNA was purified using TRIzol (Invitrogen, Paisley, UK) according to the manufacturer's instructions; cDNA was prepared using the SuperScript VILO cDNA synthesis kit (Invitrogen) using 400 ng of template RNA. PCR was performed in 10-μl reactions that contained 5.0 μl of TaqMan master mix (Applied Biosystems, Warrington, UK), 200 n m of each primer, 5 n m of probe, and 4.5 μl of template (1:20 dilution of cDNA). Details of primer/probes are given in Supplemental Table 2; PCR was 95 C for 10 min plus 40 cycles of 95 C for 15 sec and 60 C for 60 sec. Target genes were quantified using standard template dilution curves and normalization to endogenous controls (18S, actin). Primary antibodies are listed in Supplemental Table 3; negative controls had nonimmune serum. Tissues sections were immunostained using standard protocols with antigen retrieval at pH 6. Primary hESC were fixed in cold methanol and permeabilized in 0.2% IGEPAL (octylphenoxypolyethoxyethanol; Sigma, St. Louis, MO) with 1% BSA and 10% blocking serum. Endogenous peroxidase was blocked using 0.15% hydrogen peroxide in methanol. For visualization of bound primary antibodies, cells were incubated with biotinylated secondary antibodies (goat antimouse Ig, goat antirabbit Ig, or rabbit antigoat Ig; Dako UK Ltd., Ely, UK) diluted to 1:500 for 30 min at room temperature followed by reagents from the ImmPACT diaminobenzidine kit (Vector Laboratories, Burlingame, CA). Apoptosis was measured using the Apo-ONE homogeneous caspase-3/7 assay kit (Promega Ltd., Hampshire, UK): the amount of fluorescent product generated was proportional to the amount of active caspase-3/7 in the sample. Cells were incubated with staurosporine (30–500 n m ) alone (positive control) or in the presence of DHT, E2, or DHT plus E2 all at 10 −8 m for 0.5–3 h. hESC (passage <5) were seeded at 2 × 10 5 well in six-well plates in DM-RPMI 1640, supplemented as above, and grown until confluent. Forty-eight hours before scratch, medium was replaced with serum/phenol-free RPMI 1640. One hour before the scratch, cells were pretreated for 1 h with 10 −5 m flutamide or medium supplemented with vehicle. Each well of cells was scratched with a sterile 200-μl pipette tip and then incubated in the presence of DHT (10 −8 m ) or vehicle in serum-free media (n = 6 experiments, duplicate wells). For each well, six images were captured along the length of each wound at 0 and 24 h using an Axiovert 200M inverted microscope (Carl Zeiss, Jena, Germany). Images were analyzed using AxioVision release 4.72, and calculations of average distance closed for each sample were based on three measurements at identical positions along the wound at 0 and 24 h. Values presented are averages ± sem . Differences in mean values were evaluated by an unpaired t test and two-way ANOVA analysis using Bonferroni for post hoc comparisons. P < 0.05 was considered to be statistically significant.