Uterine NK cells regulate endometrial bleeding in women and are suppressed by the progesterone receptor modulator asoprisnil.

Uterine Natural Killer cells (uNK) are the major leukocyte population in non-pregnant secretory endometrium and decidua during the first trimester of pregnancy ( 1 ). These cells develop and mature in the non-pregnant endometrium to acquire phenotypic and functional properties distinguishing them from peripheral blood NK cells (pNK). Evidence is accumulating from genetic and functional studies that the function of uNK in pregnancy is to regulate placentation ( 2 , 3 ). Recent data suggests that uNK contribute to the initial stages of remodelling of uterine spiral arterioles in the early stages of pregnancy before the more dramatic destruction of the media by invading trophoblast cells ( 4 ). This important observation is pertinent to pregnancy pathologies such as pre-eclampsia and fetal growth restriction in which the failure of spiral artery remodelling is implicated in disease pathogenesis. In contrast, little is known about the function of uNK in the normal cycling endometrium. They are sparse in the estrogen-dominated follicular phase, but then vigorously proliferate in the luteal phase after ovulation when circulating progesterone levels increase, leading to a dense accumulation of uNK in the stroma in the late secretory phase ( 5 , 6 ). Menstrual breakdown occurs a few days later unless the corpus luteum is maintained by a pregnancy, in which case, progesterone levels continue to rise and the endometrium is transformed to decidua. We have proposed that uNK have a homeostatic role in this crucial switch between menstruation and decidualisation because they die a few days pre-menstrually, coincident with involution of the corpus luteum and falling progesterone levels ( 7 ). Indeed, many studies indicate that uNK may initiate arterial remodelling and maintain vascular stability ( 2 , 4 , 8 - 10 ). Menstrual bleeding complaints affect quality of life and comprise a substantial societal burden, including major effects on healthcare use and costs ( 11 , 12 ). Thus, how endometrial bleeding is controlled in normal and pathological states and how uNK contribute to this are important questions. Progesterone receptor (PR) modulators, including PR antagonists and compounds with partial and mixed progesterone agonist and antagonist activity, have been developed over the last decade as primary therapeutics for a number of indications ( 13 ). Asoprisnil, a progesterone receptor modulator (PRM) ( 14 ), has been evaluated as a treatment for heavy menstrual bleeding, uterine fibroids and endometriosis ( 15 ). Marked, dose-dependent suppression of uterine bleeding has been the most profound and consistently reported clinical effect of asoprisnil, both in healthy pre-menopausal women ( 16 ) and in patients with heavy menstrual bleeding associated with uterine fibroids ( 14 ). The exact mechanism of action of asoprisnil on the endometrium remains to be established, but the most striking histological effect is on the spiral arteries ( 17 ), which appear unusually prominent due to abnormally thick muscular walls. This effect seems to be specific to asoprisnil, because it has not been observed with other PRMs, including mifepristone. Expression of PR is normally abundant in endometrial stroma, especially in perivascular cells ( 18 , 19 ) so that asoprisnil might initiate these striking changes by affecting stromal PR expression; an effect via alteration of uNK function is also possible. Uterine NK cells do not, however, express the PR ( 20 ) therefore, any PR-associated regulation of function will be indirect. IL-15 is crucial in the regulation of uNK development, expansion, and maturation and is highly expressed in perivascular stromal cells in the non-pregnant endometrium where uNK cluster ( 21 , 22 ). To investigate the mechanisms by which asoprisnil exerts its profound suppression of uterine bleeding, we used a whole genome expression array to study gene expression profiles in asoprisnil-treated versus placebo-treated patients with bleeding owing to symptomatic uterine fibroids. Robust statistical and stringent threshold analysis identified 245 genes, predominantly associated with immune-inflammatory genes. Based on the of statistical pathway analysis for functional enrichment of these genes, the IL-15 pathway was identified as the principal network altered upon asoprisinil treatment. Many of the down-regulated genes fed into the IL-15 pathway, known to have a key role in uNK development and function ( 23 ). We therefore examined uNK distribution in the endometrium of asoprisnil-treated women and strikingly found that uNK were substantially reduced or absent. It is notable that these women also reported a dramatic reduction or absence of menstrual bleeding ( 14 ). These data provide convincing evidence to support a pivotal role for an IL-15-uNK axis in the regulation of endometrial bleeding.

Samples of endometrium exposed to asoprisnil were derived from a phase II, multi-centre, randomised, double-blind, placebo-controlled study of asoprisnil administration in premenopausal patients with symptomatic uterine fibroids scheduled for hysterectomy, as described previously ( 14 , 17 ). Subjects received oral doses of asoprisnil 10 mg, 25 mg or placebo once daily for 12 wk until hysterectomy. The study protocol was approved by the Multicentre Research Ethics Committee (LREC/2002/6/17). Endometrial tissue was collected at hysterectomy and preferably obtained from areas not overlying any fibroids. Samples were processed for RNA extraction and immunohistochemistry. IL-15 mRNA expression was assessed (use of SYBR green-based quantitative real time PCR [QRT-PCR]) in endometrium from women not exposed to asoprisnil. Samples of normal endometrium (menstrual n=7; proliferative phase n=11; secretory phase n=16) were obtained from women with normal menstrual cycles and no endometrial pathology attending Addenbrooke’s Hospital, Cambridge for sterilisation. Patients with endometriosis, fibroids or a history of exogenous hormone administration were excluded. Human endometrial tissue specimens were obtained from women undergoing surgery for non-malignant gynaecological conditions; written informed consent was obtained from all subjects prior to surgery, and ethical approval was granted by the Lothian research ethics committee (LREC/05/51104/12; LREC/10/51402/59.) Samples of first trimester decidua were obtained from women undergoing elective surgical termination of pregnancy in the first trimester at Addenbrooke’s Hospital. All patients provided informed consent. Peripheral blood was collected by venepuncture of healthy volunteers, with informed consent. Blood was layered directly onto Lymphoprep (Nycomed, Oslo, Norway) and centrifuged (700 × g , 20 min) to enrich leukocytes. The interface was collected and washed in PBS. The study was approved by Cambridge Local Research Ethics Committee, study number 04/Q0101/23. RNA was extracted from endometrial samples and collected from women exposed to asoprisnil, as previously described ( 20 ). Quality and integrity of each RNA sample was checked with the Agilent 2100 Bioanalyser (Agilent Inc, Santa Clara, CA, USA) according to manufacturer’s instructions and the RNA Integrity Number (RIN) determined. Only samples with a RIN > 8.5 were processed for further analysis. Prior to full array analysis, RNA was subjected to the Affymetrix Test3 Array (Affymetrix Inc, Santa Clara, CA, USA) to exclude degraded samples with insufficient target. Human endometrial stromal cells (ESCs) were isolated from tissue specimens following collagenase and DNAse digestion as described previously. ESCs were subsequently maintained in vitro at 37°C under 5% CO 2 in air for a maximum of 4 passages and cultured in RPMI 1640 media (Sigma) supplemented with 10% FCS (Cat. No. 10082-147; Invitrogen), 10 mL/L Penicillin/Streptomycin (10,000 units penicillin and 10mg Streptomycin per 1 ml solution; Cat No. P-4333; Sigma), 2 mM L-glutamine (Cat. No. G-7513; Sigma), and 2.5 μg/ml Fungizone (Cat. No. 15290-018; Invitrogen). Prior to experimentation ESCs were transferred to medium that contained phenol-red free RPMI 1640 and charcoal stripped FCS but otherwise supplemented as previously described for 48 h. FCS was charcoal-stripped to remove endogenous steroids. Prior to decidualisation ESCs were transferred to serum depleted media (as above but 2% FCS) for 24 h. Decidualisation was induced by addition of decidualisation media (RPMI 1640, 2% FCS, 0.1 mg/ml 8-Br-cAMP and 1 μM progesterone) for up to 8 days. Control cells were incubated with vehicle (DMSO). Concentrations of IL-15 mRNA were determined using Taqman qRT-PCR. Cells were isolated from the decidua as previously described ( 24 ). Cells were then plated down on plastic for two hours in RPMI 1640 medium plus 10% fetal calf serum (FCS). The non-adherent cells were stained for CD3 and CD56 and flow sorted on CD3 + CD56 - cells (T cells) and CD3 - CD56 + cells (NK cells). Adherent cells were harvested by trypsin digestion (5 min, 37°C), stained for HLA-DR and CD10 and sorted on HLA-DR + CD10 - cells (myeloid antigen presenting cells, APC) and HLA-DR - CD10 + cells (stroma) ( 23 ). Trophoblast cells were cultured overnight as described ( 24 ), harvested by trypsin digestion (5 min, 37°C), stained for HLA-G and CD14 (macrophage marker) and sorted on HLA-G + CD14 - extravillous trophoblast (EVT) cells. Endometrial samples from placebo (n=6), 10 mg asoprisnil (n=11) and 25 mg asoprisnil (n=10) treatments were hybridised to 27 microarrays. The array platform employed was the Affymetrix Human Genome U133 plus 2.0 ( www.affymetrix.com/support/technical/datasheets/human ), whole human genome expression array. The microarray data were processed according to the following approaches: Between-array normalisation followed a standard Robust-Multiarray-Average model, providing background-corrected, quantile-normalised, gene-level agglomerated and log 2 transformed expression values. A non-specific filtering step was applied, removing probes that either fell below a threshold of detection or did not vary across samples. The corresponding filtering parameters were Affymetrix “Present” or “Marginal” calls in at least 3 arrays and variation greater than or equal to median interquartile range estimate across all samples. After filtering, 26111 of 54675 probes on the chip were used as the analysis set. For statistical analysis of endometrial samples, a basic cell means model was fitted for each gene probe, with contrasts extracted for 10 mg asoprisnil vs placebo and 25 mg asoprisnil vs placebo. The model was enhanced by empirical Bayes shrinkage of standard errors of genes towards a combined value. In this analysis, significance estimates for gene probes were corrected for multiple testing by the method described by Benjamini & Yekutieli ( 25 ). The data discussed in this publication have been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus ( 26 ) and are accessible through GEO Series accession number GSE47577 ( http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE47577 ). All genes showing a significance p5 fold change between placebo and asoprisnil were also subjected to Gene Ontology Enrichment analysis. Pathway analysis and network clustering used DAVID ( www.david.abcc.ncifcrf.gov ), KEGG ( www.genome.jp/kegg/pathway ) and IPA (Ingenuity Inc, USA) knowledge databases and software tools, applying hypergeometric test of significance for pathway membership. In addition, Pearson correlation, bi-clustering and K-means clustering and network analysis of co-regulated genes we used to explore how differentially regulated genes correlated and how their concerted alteration contributes to a particular biological process. TaqMan real-time quantitative PCR was used to determine the concentrations of target mRNAs, identified in the gene microarray derived from endometrial samples from women exposed to asoprisnil, according to standard protocols ( 20 ). The PCR reaction was run on the ABI Prism 7900 (Applied Biosystems, Life Technologies, Foster City, CA, USA) using standard conditions. Messenger RNAs encoding IL-15 and the uNK cell marker CD56 (NCAM1) were evaluated ( Table 1 ). For the SYBR green-based QRT-PCR, the reaction consisted of 1 × Blue QPCR SYBR low ROX mix (Thermo Fisher Scientific,Waltham, MA, USA) 1 μL template cDNA, 70 nM forward and reverse primers made up to 20 μL in ultra-pure water. All reactions were run in triplicate and a melt curve was run to check product purity. Relative concentrations of DNA present during the exponential phase of the reaction were determined using the ΔCt method, comparing unknowns to a standard curve. Results for each sample were normalised to 18S RNA. Total decidual or peripheral leukocytes were incubated with CD56 Microbeads (1ml of beads per 10 6 cells, 15 min, 4°C; Miltenyi Biotec, Bergisch Gladbach, Germany) and NK cells enriched magnetically. NK cells were then cultured for five days at 50,000 - 100,000 cells/well of a 24 well plate in 2 mL RPMI 1640 medium plus 10% FCS with 0.5 ng/mL recombinant human IL-15 (Peprotech) or 10 ng/mL IL-15Rα-Fc fusion protein (R&D Systems) or both. When both IL-15 and IL-15Rα-Fc were added, they were pre-incubated together for two hours to allow complexes to form. At the end of the culture period, the cells were counted, excluding dead cells using trypan blue, and replated in 200 μL RPMI 1640 medium plus 10% FCS at 100,000 cells/well of a 96 well plate, with the addition of 0.037 MBq tritiated thymidine (Amersham Biosciences, Amersham, UK) and cultured for an additional 16 h. Thymidine incorporation was assessed by measuring counts per minute. Immunohistochemistry was performed according to standard protocols with Ag retrieval in 0.01M sodium citrate ( 27 ). The Bond-X Staining System (Leica, Milton Keynes, Bucks, UK) was used for further processing ( 27 ). Monoclonal mouse antibodies (mAb) to CD56 (1:200; Zymed Laboratories, San Francisco, CA, USA) and to PR (1:400; Novocastra-Leica, Milton Keynes, Bucks, UK) together with negative control, isotope-matched mouse serum (MIgG 1 ; Sigma-Aldrich, St Louis, MO, USA,) at matched IgG concentrations. The automated staining process was completed according to the Bond Polymer Define Detection Kit (Leica Microsystems) 3h protocol per manufacturer’s instructions. The intensity and frequency of positively immunostained uNK (CD56 + ) in the endometrial stroma was quantified. CD56 immunostained tissue sections were observed under a Leitz DMRB research microscope under bright field illumination using a Q-imaging Qicam FAST-1394 digital camera and a Prior motorized stage. Media Cybernetics Image-Pro Plus software with a stereology module was used to tile areas of endometrium using a 4x objective. From the tiled overview image a representative area of interest was delineated by the operator and 10 random fields were then acquired via software at original magnification 40x. CD56 + cells were counted over the 10 frames. Results were unblinded at the end of the count. To identify components of the arterial walls, a monocloncal antibody to ∝ smooth muscle actin (αSMA) was used (1:5000; no. A-2547, Sigma-Aldrich, St Louis, MO, USA) ( 28 ). Masson Trichrome was used to identify collagen ( 29 ). Each group of asoprisnil treated subjects (10 and 25 mg) was compared to the subgroup of placebo-treated subjects who had undergone hysterectomy in the secretory phase of their menstrual cycle. These were determined by histological examination ( 17 ) and correlated with patients’ menstrual diaries ( 14 ). Samples from the progesterone-dominated cycle phase were deemed most appropriate for comparison to samples following treatment with asoprisnil. A comparison of treatment groups as well as analysis of immunohistochemical protein expression was done applying the Wilcoxon’s rank sum test with statistical significance determined at 0.05 level using Hochberg’s multiple comparison procedure. For QRT-PCR of IL-15 and IL-15RA across phases of the menstrual cycle and cell types, significant differences were detected using the Kruskal-Wallis test. When differences were detected, pairwise comparisons were then made using the Mann-Whitney U test. For IL-15 trans -presentation assays, paired samples, cultured either with or without IL-15Rα, were compared using the Wilcoxon signed ranks test.

Endometrial samples from 33 subjects were examined for immunolocalisation of PR (10, 12 and 11 patients had received placebo, 10 mg asoprisnil and 25 mg asoprisnil, respectively, as reported previously) ( 14 ). Control secretory phase endometrium (from women who received placebo) exhibited the characteristic strong expression of PR in the endometrial stroma with negligible expression in glandular epithelium ( Figure 1A ). In contrast, exposure of endometrium to asoprisnil (25 mg) resulted in virtual complete loss of stromal PR expression. PR expression was also suppressed in perivascular cells by both doses of asoprisnil compared with placebo ( Figure 1B ). Conversely, PR expression was significantly up-regulated in both surface and glandular epithelium in asoprisnil-exposed endometrium ( Figure 1B ). To our knowledge, this dramatic switch in the distribution of PR expression from stroma to glandular epithelium in response to a PRM has not been reported previously. To gain further insight into asoprisnil-mediated PR effects, genome-wide molecular profiling was used to provide an informative unbiased approach to identify the response of endometrial tissue to asoprisnil treatment. Microarray analysis of endometrial samples from the same 33 patients was performed ( 14 ). 27 of the 33 endometrial RNA samples (n=6 placebo [secretory]; n=11 asoprisnil 10 mg; n=10 asoprisnil 25 mg) were of suitably high quality (RIN > 8.5; Mean RIN 9.35) for analysis. 245 genes were significantly (p≤0.05 after multiple testing correction) up-regulated or down-regulated 5-fold or greater in either or both treatment groups (10 and 25mg asoprisnil) compared to placebo. Figure 2 provides an overview of this gene set in the form of a heat map representing gene expression levels and their similarity across all samples and treatment groups. The overall trend was a marked suppression of gene expression (166 genes) and gene ontology enrichment analysis of this set showed a significant over-representation of genes characteristic of NK cells, notably IL-15, IL-2RB, GZMB, NKG7 and GNLY. The down-regulation of many genes was dose dependent ( Table II and Supplementary Table 1 ), which is indicative of an association with drug action. Subsequently, 245 genes with significant differential change in expression were analyzed for association with biological functions using pathways and network analysis. Discrete biological networks of interacting genes can be integrated to test for connectivity and overlap. In this regard, integration of the networks associated with 166 down regulated genes uncovered IL-15 in the centre of an immune mediator axis in the down-regulated genes ( Figure 3 ). To validate the microarray experiments, QRT-PCR was performed on endometrial tissue samples to evaluate the effects of asoprisnil on IL-15. Sequences for QRT-PCR are described in Table I . IL-15 mRNA expression was significantly down-regulated by 25 mg asoprisnil compared to placebo (secretory) samples ( Figure 4A ). Because of the clear influence of IL-15 on NK cell development and function, we investigated how uNK numbers were affected by asoprisnil using immunohistochemistry to identify CD56 + cells. Figure 1C shows the typical distribution of uNK in the normal secretory phase scattered throughout the stroma. In contrast, CD56 + uNK cells were absent or dramatically reduced following treatment with asoprisnil (25 or 10 mg), compared to placebo ( Figure 1D , and Figure 5 ). The three samples in the placebo group that had only small numbers of CD56 + uNK, were histologically dated as early secretory phase that is before the major expansion of uNK ( Figure 5 ). We examined the histological components of the endometrial spiral arteries ( 17 ); following asoprisnil administration, they exhibit an unusual appearance with abnormally thick muscular walls. ∝SMA is normally expressed only in the tunica media of arteries and arterioles, whereas collagen is localized in the tunica adventitia. Morphological analysis of asoprisnil-treated endometrium showed that ∝SMA staining was increased within the tunica media and detected in the tunica adventitia ( Figure 1E ). Within the clusters of thick-walled arterioles, increased collagen was found in both the tunica adventitia and tunica media following administration of asoprisnil ( Figure 1F ). Neither effect on ∝SMA or collagen was dose-dependent. Figures 1G and 1H illustrate the histological appearance of normal endometrial spiral arteries in the secretory phase. To investigate whether IL-15 is induced by progesterone and to determine when IL-15 signalling to uNK cells mainly occurs, we measured IL-15 transcript in endometrial samples taken throughout the normal menstrual cycle and the first trimester of pregnancy. IL-15 transcripts were low in the menstrual and proliferative phases, increased in the secretory phase, and higher still in first trimester decidua ( Figure 4B ). In addition the non-pregnant endometrial secretory phase was modelled in vitro , and the response of primary human endometrial stromal cells decidualisation media (progesterone plus cAMP) was examined for IL-15 production. Significant increases in IL-15 mRNA expression were demonstrated ( Figure 4E ). As a result, we can confirm that levels of IL-15 mRNA increase in response to progesterone, consistent with previous reports ( 30 - 32 ). To investigate whether uNK respond to trans -presented IL-15, we used an assay first developed in mice and recently adapted for ex vivo use on human pNK ( 33 , 34 ). In this assay, CD56 + NK cells ( Figure 6A, B ) were cultured with recombinant human IL-15, which has been pre-incubated to form complexes with IL-15Rα-Fc fusion protein mimicking the effect of IL-15 trans -presentation. As a positive control, we first replicated experiments on the effect of the complexes on pNK from non-pregnant donors. After five days, there were significantly more cells and slightly more thymidine incorporation in the cultures with complexes than in those with IL-15 only ( Figure 6C, D ). The 1.7-fold median expansion of cells observed in the trans -presentation condition, compared with soluble IL-15, is consistent with the 2-fold mean expansion previously reported ( 34 ). The results of the assay assessing cell culture growth in the presence of soluble IL-15 versus complexes with IL-15Rα-Fc fusion protein were similar for uNK and pNK. Significantly higher cell numbers and thymidine incorporation were present after culture with complexes than after culture with IL-15 alone ( Figure 6E, F ). Therefore, we conclude that uNK respond to IL-15 trans -presentation. To identify which cells are responsible for IL-15 trans -presentation, we examined transcription of IL-15 and IL-15RA by different uterine cell subsets. Because we were unable to extract sufficient cell numbers from small endometrial biopsies, we isolated various decidual cell subsets. IL-15 and IL-15RA transcripts were significantly higher in decidual APCs than in NK or T cells ( Figure 4C, D ). IL-15 and IL-15RA were low in EVT, although the number of samples was too small to make a statistical comparison. In stromal cells, IL-15 was also expressed at significantly higher levels than in NK and T cells but IL-15RA expression varied considerably, such that it was neither significantly higher than in NK and T cells, nor significantly lower than that in APCs. Therefore, both IL-15 and IL-15RA transcripts are expressed in APCs and stromal cells at similar, high, levels.

Asoprisnil has been shown, along with other PRMs, to be useful clinically because of its ability to reduce dramatically or stop endometrial bleeding ( 13 , 14 , 16 ). Although we have reported previously that thick-walled clusters of spiral arterioles within the endometrium are characteristic of asoprisnil treatment ( 17 ), the mechanism by which it mediates this effect is not understood. We now report the striking observations from asoprisnil-treated endometrium that expression of the progesterone receptor (PR) in stromal cells is negligible, there is gain of PR expression in uterine epithelium and that uNK are absent. Stromal cell expression of PR and presence of uNK are both defining features of normal secretory endometrium. Our findings lead us to hypothesize that asoprisnil prevents endometrial bleeding by interfering with the complex interplay between endometrial stromal cells, uNK and the spiral arteries. These novel data also complement the recent report that uNK play a role in the remodelling of spiral arterioles in the early stages of pregnancy ( 4 ). Using whole genome expression arrays, we showed that asoprisnil down-regulates a number of innate immune genes centred on the IL-15 signalling pathway and this finding was confirmed by QRT-PCR. IL-15 mediates the differentiation of immature to mature uNK in humans ( 23 , 35 ) and is essential for their development in mice ( 21 ). In this study we have shown that IL-15 can also promote the proliferation of mature human uNK in the context of trans -presentation by IL-15Rα. Thus, the marked down-regulation of IL-15 in asoprisnil-treated endometrium is likely to account for the remarkable absence of uNK in these samples. In agreement with others, we found that IL-15 variation over the course of the menstrual cycle and early pregnancy correlates with circulating progesterone levels ( 30 - 32 , 36 ). In vitro modelling of secretory endometrium confirmed IL-15 is up-regulated during decidualisation of endometrial stromal cells, a process which occurs in response to peak progesterone levels in vivo . Moreover, IL-15 expression decreases in response to treatment with the progesterone antagonist RU486 (unpublished observations), lending further support that progesterone signalling causes increased IL-15 signalling, rather than the two merely being correlated. IL-15 expression in the endometrium is already well known to correlate with fluctuations in the number of uNK in mice ( 37 ). Thus, although uNK cells themselves do not express the PR, they are clearly influenced by progesterone levels ( 38 ). This suggests that progesterone-responsive cells in the endometrium, notably stromal cells, relay the signal to uNK via IL-15. Our finding that endometrial stromal cells and APCs express both IL-15 and IL-15RA, which are required to effectively signal to uNK via trans -presentation ( 39 ), suggests that either or both of these cells are capable of supporting uNK differentiation and expansion. However, APCs and stromal cells are not equally responsive to progesterone, and by extension to asoprisnil. A major impact of asoprisnil administration was the observed down-regulation of expression of PR in endometrial stromal cells. In contrast, APCs do not express PR ( 40 , 41 ), a finding we confirmed in both asoprisnil-treated and control patient s . Thus, because normally only stromal cells express PR, they are the likely candidates to promote uNK expansion in response to increased progesterone ( 20 , 40 - 42 ), and to mediate the dramatic effects of asoprisnil on endometrial uNK numbers and in turn the structural changes in spiral arteries. There are now several lines of evidence that uNK directly affect spiral arteries by modifying the structure of the tunica media. In humans, uNK are always preferentially located around the glands and arteries ( 43 ), and are thought to influence the loosening of the smooth muscle of the media by production of a range of angiogenic factors ( 2 , 4 , 8 - 10 ). By staining for αSMA and collagen, we observe in this study that spiral arterioles in asoprisnil-treated endometrium have a thicker tunica media in the absence of uNK. There is a clear parallel between the endometrium of asoprisnil-treated women and the implantation site of uNK-deficient strains of mice. In addition to lacking uNK, both have arteries with elevated wall-to-lumen ratios and thickened walls ( 21 ). Thus, asoprisnil treatment results in a human endometrial phenotype resembling a uNK-null mouse ( 21 ). Furthermore, a recent study of decidual spiral arteries after antibody depletion of uNK cells in rats also showed altered arterial development ( 44 ). Data on endometrial IL-15 expression and uNK numbers are not available for PRMs other than asoprisnil, and none of the morphological effects observed with asoprisnil occur in women treated with mifepristone. However, infrequent alterations in vascular morphology have been described with other PRMs, such as ulipristal acetate (CDB2914). Detailed histology from two large Phase 3 clinical trials of more than 500 subjects revealed thick walled vessels similar to those seen in asoprisnil-treated endometrium but only in ~10% of subjects treated with ulipristil acetate ( 45 ). Experience with other PRMs is more limited. In a smaller clinical trial of telapristone acetate (Proellex; CDB4124), unusual patterns of endometrial vasculature (dilatation but no thickening of the walls) were found in only 4% of subjects ( 46 ). To our knowledge none of the morphological effects observed with asoprisnil are found in women treated with progesterone antagonists such as mifepristone. Although uNK in normal endometrium are clearly regulated by progesterone, the effects of asoprisnil on stroma, uNK, and the stability of spiral arteries are likely to be due to the partial progesterone agonist activity of asoprisnil. A direct action of the PR ligand, asoprisnil itself, cannot be ruled out because PRs are expressed in the vessel wall of human and non-human primate endometrium ( 19 ). Menstrual bleeding disorders are a major health problem and studies with asoprisnil have revealed a potentially useful class of drug and have highlighted the complex interplay between progesterone, the stroma, uNK and spiral arteries in the endometrium, a pathway that is crucial for endometrial health and disease. In summary, we have shown that administration of asoprisnil inhibits endometrial expression of IL-15 and causes a profound reduction in the number of uNK. Furthermore, our findings that endometrial stromal cells express PR, that stromal cell PR expression is inhibited by asoprisnil, and that stromal cells are capable of trans-presenting IL-15 to uNK lead us to propose a model to account for the effects of asoprisnil on endometrial bleeding ( Figure 7 ). In normal non-pregnant endometrium uNK proliferate and mature under the influence of progesterone that induces IL-15 trans -presentation from stromal cells. Along with the other pre-decidual changes that are a feature of mid-late secretory endometrium, uNK modify the smooth muscle media of the arteries and maintain arterial stability. If the corpus luteum is maintained by a pregnancy, these changes persist into early decidua until the more dramatic arterial transformation mediated by trophoblast occurs in the decidua basalis ( 47 ). In a non-pregnant cycle, the corpus luteum involutes and progesterone and IL-15 levels fall, resulting in the demise of uNK, vascular collapse and the onset of menstruation. In miscarriage and failure of implantation following IVF, endometrial breakdown also occurs probably because of similar mechanisms. uNK have also been implicated in situations in which there are disturbances of hormonal regulation of the endometrium such as peri-menopausal bleeding and breakthrough bleeding on exogenous hormone therapy ( 48 ). We propose therefore that uNK not only function in the regulation of placentation but also in the homeostasis of the cycling endometrium and in maintenance of the decidua. Compared to other mucosal surfaces, the uterine mucosa is uniquely dynamic and can change dramatically from total breakdown to decidualisation in only a few weeks with uNK playing an integral part in this process.