【关键词】 endometrium
departments of obstetrics and gynecology (s.ta., a.e.h., k.c.v., s.tu., m.t.o., c.d., n.l.s., c.n.n., n.r.n., l.c.g.) and pathology (r.k.), stanford university, stanford, california 94305
department of obstetrics and gynecology (b.a.l.), greenville hospital system, greenville, south carolina 29605
abstract
histological evaluation of endometrium has been the gold standard for clinical diagnosis and management of women with endometrial disorders. however, several recent studies have questioned the accuracy and utility of such evaluation, mainly because of significant intra- and interobserver variations in histological interpretation. to examine the possibility that biochemical or molecular signatures of endometrium may prove to be more useful, we have investigated whole-genome molecular phenotyping (54,600 genes and expressed sequence tags) of this tissue sampled across the cycle in 28 normo-ovulatory women, using high-density oligonucleotide microarrays. unsupervised principal component analysis of all samples revealed that samples self-cluster into four groups consistent with histological phenotypes of proliferative (pe), early-secretory (ese), mid-secretory (mse), and late-secretory (lse) endometrium. independent hierarchical clustering analysis revealed equivalent results, with two major dendrogram branches corresponding to pe/ese and mse/lse and sub-branching into the four respective phases with heterogeneity among samples within each sub-branch. k-means clustering of genes revealed four major patterns of gene expression (high in pe, high in ese, high in mse, and high in lse), and gene ontology analysis of these clusters demonstrated cycle-phase-specific biological processes and molecular functions. six samples with ambiguous histology were identically assignable to a cycle phase by both principal component analysis and hierarchical clustering. additionally, pairwise comparisons of relative gene expression across the cycle revealed genes/families that clearly distinguish the transitions of peese, esemse, and mselse, including receptomes and signaling pathways. select genes were validated by quantitative rt-pcr. overall, the results demonstrate that endometrial samples obtained by two different sampling techniques (biopsy and curetting hysterectomy specimens) from subjects who are as normal as possible in a human study and including those with unknown histology, can be classified by their molecular signatures and correspond to known phases of the menstrual cycle with identical results using two independent analytical methods. also, the results enable global identification of biological processes and molecular mechanisms that occur dynamically in the endometrium in the changing steroid hormone milieu across the menstrual cycle in normo-ovulatory women. the results underscore the potential of gene expression profiling for developing molecular diagnostics of endometrial normalcy and abnormalities and identifying molecular targets for therapeutic purposes in endometrial disorders.
introduction
human endometrium, the anatomic prerequisite for establishing and sustaining pregnancy, undergoes remarkable histological and structural changes throughout the menstrual cycle, in preparation for embryonic implantation and subsequent shedding and regeneration in nonconception cycles (1). different histological appearances of this tissue were first described in 1908 by hitschmann and adler (2), and since the classic manuscript by noyes et al. (3) in 1950, it has been appreciated that endometrial histology directly correlates with changes in the circulating ovarian-derived hormones estradiol (e2) and progesterone (p) (4, 5). based on an ideal 28-d cycle, with cycle d 1 being the first day of menstrual flow and cycle d 14 the day of ovulation, noyes and colleagues (3, 6) identified, in nearly 8000 endometrial biopsies, different histological appearances (phases), which now are appreciated as menstrual; postmenstrual repair; early-, mid-, and late-proliferative; interval; and early-, mid-, and late-secretory. histological examination has been the gold standard for clinical evaluation of the endometrium (3, 6), including from which phase the tissue derives, whether ovulation has occurred, how many days postovulation the tissue is or is expected to be (dating the endometrium), the effects of exogenous estrogens and progestins, and detection of polyps and hyperplastic and/or neoplastic cells. in the proliferative phase, under the influence of e2, there are numerous cellular mitoses and tissue growth (approximately 10 mm over a 2-wk period) (1, 7, 8). the secretory phase is heralded, under the influence of p, by epithelial (glandular) secretion, stromal edema, and stromal cell differentiation (decidualization). in addition, there is extensive angiogenesis in the proliferative phase, coiling of the spiral arteries in the secretory phase, and an influx of leukocytes in the latter part of the mid-secretory phase and during the late-secretory phase (1, 7). in the menstrual phase, there is cellular apoptosis and extensive tissue breakdown. although many of the biological processes and molecular participants in these events have been elucidated, a complete understanding has yet to be achieved on a global scale and for transitioning from phase to phase.
recently, the usefulness of histological dating of the endometrium for couples with infertility has been questioned because histological delay in endometrial maturation fails to discriminate between fertile and infertile couples (9). in another recent study (10), histological features fail to distinguish, reliably, specific menstrual cycle days or narrow intervals of days, leading to the conclusion that histological dating has neither the accuracy nor the precision to be useful in clinical management. because histology has equivocal value in patient evaluation and management, the question arises as to whether molecular profiles of endometrium may distinguish among the phases of the cycle, define uterine receptivity to implantation, and identify a variety of endometrial disorders not apparent from histological evaluation. in the current study, we have investigated global gene expression profiling across the menstrual cycle in normo-ovulatory women with the goals of determining whether molecular profiles of human endometrium can distinguish among the phases of the menstrual cycle laying the foundation to identify endometrial disorders that may not be detectable with classical histological assessment. the data demonstrate specific patterns of gene expression that are characteristic of the different phases of the cycle, and they identify biological processes, molecular functions, and structural cellular participants in the dynamic histological events occurring across the cycle. importantly, the gene expression patterns and molecular signatures that are pathognomonic of a particular cycle phase have enabled classification of samples of unknown, equivocal, or ambiguous histology, thus heralding a new era of molecular diagnostics and targeted therapeutics for endometrial disorders.
materials and methods
sample collection and processing
endometrial samples (n = 45) were obtained from normally cycling women undergoing hysterectomy or endometrial biopsy after written informed consent, under an approved protocol by the stanford university committee on the use of human subjects in medical research. samples were also obtained under an approved protocol, after written informed consent, at the university of north carolina, chapel hill, and the center for women’s medicine at greenville hospital. of these 45 samples, 22 eventually were of adequate amount, yielded high-quality rna, and had histological diagnoses agreed upon by two or more pathologists. these samples comprise the well-characterized samples used for microarray analysis and are listed in table 1. all subjects of the well-characterized samples were normo-ovulatory with regular cycles (2435 d), had not been on steroid hormone medications within 3 months of endometrial sampling, and were between the ages of 23 and 50 yr old (median age, 42 yr; mean ± sem = 40 ± 6.6 yr old). indications for the operative procedures (n = 14 hysterectomies; n = 8 endometrial biopsies) in which samples were obtained included uterine prolapse (n = 4), uterine leiomyomata (fibroids, diameters ranging from 0.513 cm, median of 3 cm, not submucosal) (n = 9), normal volunteers (n = 7), pelvic pain (n = 1), and ovarian cyst (n = 1). none of the subjects had endometriosis (laparoscopy proven), and pathology reports revealed no inflammation within the endometrium in any of the specimens. samples were collected using a pipelle catheter or curetting the endometrium from hysterectomy specimens. they were kept at room temperature and transported to the laboratory in 1x pbs. portions of tissues were saved in 10% formalin for hematoxylin and eosin staining and histological evaluation, and portions were also snap frozen in liquid nitrogen for rna isolation. tissue specimens were examined blindly by up to four independent pathologists, and phases were assigned according to the criteria of noyes et al. (3). dating in the secretory phase was within a 2-d cycle window. of the well-characterized samples, five were defined as mid-late proliferative phase, three as early secretory, eight as mid secretory, and six as late secretory, for a total of 22 sufficiently dated samples (agreement between at least two pathologists) (table 1). of the remaining 23 samples, one was appropriately assessed histologically and was used as part of the validation studies but not for microarray processing. six samples had ambiguous interpretations (defined herein when two or more pathologists gave different histological interpretations of the same sample or when a sample was evaluated histologically by only one pathologist) (table 2) and were processed for microarray analysis. sixteen were excluded from further study on the basis of patient medical history or poor sample quality.
total rna isolation and microarray preparation
total rna was isolated from individual tissue samples using trizol reagent (invitrogen, carlsbad, ca) following the manufacturer’s protocol. the rna preparations were then dnase treated and purified using the rneasy mini kit (qiagen, valencia, ca.) some samples with a lower rna yield were concentrated by adding 10% sodium acetate, 1% glycogen, and 2.5 vol 100% cold ethanol. they were then incubated overnight at 20 c, centrifuged at 4 c for 30 min at 20,000 x g, and washed once with cold 80% ethanol, with air drying of the rna pellet. samples were stored in rnase-free h2o, and the purity was analyzed by both the 260/280 absorbance ratio as well as gel electrophoresis. first- and second-strand cdnas were prepared according to the affymetrix microarray preparation protocol (affymetrix, santa clara, ca). briefly, a dt24 primer (proligo, boulder, co) was incubated with 5 μg total rna at 70 c for 10 min. samples were then incubated with superscript ii reverse transcriptase (invitrogen) at 42 c for 1 h to generate first-strand cdna. samples were placed on ice and incubated with escherichia coli dna polymerase, ligase, and an rnaseh inhibitor (invitrogen) at 16 c for 2 h to generate second-strand cdna. t4 polymerase (invitrogen) was then added, and samples were incubated for another 5 min at 16 c. the reaction was terminated and the samples purified using the genechip sample clean-up module (affymetrix). biotinylated crna was then prepared using the enzo bioarray high-yield t7 transcript labeling kits (enzo, farmingdale, ny.) crna was subsequently purified using the clean-up module and fragmented using 5x fragmentation buffer [200 mm tris (ph 8.1), 500 mm koac, 150 mm mgoac). individual samples were hybridized overnight to high-density human genome (hg) u133 plus 2.0 arrays (affymetrix), containing 42,203 genes and 12,397 expressed sequence tags (ests), at the stanford university school of medicine protein and nucleic acid facility. subsequently, the chips were scanned using an hr3000 scanner, and the data were extracted using the affymetrix genechip operating software version 1.1.
data analysis
microarray gene expression data analysis.
the intensity values of different probe sets (genes) generated by affymetrix genechip operating software were imported into genespring version 7.2 software (silicon genetics, redwood city, ca) for data analysis. the data files (cel files) containing the probe level intensities were processed using the robust multiarray analysis algorithm (genespring) for background adjustment, normalization, and log2-transformation of perfect match values (11). subsequently, the data were subjected to per-chip and per-gene normalization using genespring normalization algorithms. the normalized data were then subjected to a series of pairwise comparisons that included all 22 well-characterized specimens. comparisons were made between successive cycle phases: early-secretory endometrium (ese) vs. proliferative endometrium (pe); mid-secretory endometrium (mse) vs. ese; and late-secretory endometrium (lse) vs. mse. the resulting gene lists generated from each pairwise comparison included only the genes that had a fold change value of 1.5 or higher and a p value of less than 0.05 by a one-way anova parametric test and a benjamini-hochberg multiple testing correction for false discovery rate. individual pairwise comparison gene lists were combined to produce a list of all of the genes that are differentially expressed by microarray analysis. this combined gene list includes 7231 genes and ests (4135 genes and 3096 ests).
to verify the genespring analysis, significant changes in differentially expressed genes were evaluated using a second, independent method, namely, statistical analysis of microarrays (sam) (12). this was performed for a select pairwise comparison group (lse vs. mse). normalized and log-transformed gene expression data were analyzed using the genespring implementation of the sam algorithm. the sam parameters were set as follows to analyze two classes of unpaired data (lse vs. mse) using cycle phases as the parameter; 1000 permutations were performed with a false discovery rate less than 0.05. the resulting gene list was filtered for a fold change of 1.5 or higher.
principal component analysis (pca).
pca is an unsupervised pattern recognition and visualization tool used to analyze large amounts of data derived from array gene expression analysis (13). it displays a multidimensional data set in a reduced dimensionality (three dimensions) to capture as much of the variation in the data as possible, allowing its summarization and further analysis. each dimension represents a component to which a certain percentage of variance in the data is attributed. we applied the unbiased pca algorithm in genespring to all 28 samples across the menstrual cycle, using all 54,600 genes and ests on the hg u133 plus 2.0 chip to look for similar expression patterns and underlying cluster structures.
hierarchical clustering.
hierarchical clustering is an unsupervised way of grouping samples based only on their gene expression similarities (14). herein, we conducted hierarchical cluster analysis of differentially expressed genes from all pairwise comparisons (the combined gene list) using the smooth correlation for distance measure algorithm (genespring) to identify samples with similar patterns of gene expression. (this approach differs from that which we used for pca in that for the latter, all of the genes on the array were used, whereas in the hierarchical clustering analysis, the pairwise comparisons-derived gene list was used.) in addition, the six ambiguous samples were included in the clustering analysis to interrogate in which cluster these samples would fall. the output data are displayed graphically as a heatmap, based on the measured intensity values of the genes in the gene list (above) and are represented as a hierarchical tree with branches to indicate the relationships among different groups.
k-means analysis.
the differentially expressed genes from all pairwise comparisons (combined gene list) were also subjected to k-means analysis (15) using genespring software. this analysis was used to detect trends of gene expression during different stages of the menstrual cycle. kinetic patterns of gene expression were analyzed across the cycle in pe, ese, mse, and lse. k-means was applied to the data using the smooth correlation of distance measure algorithm (genespring), because this algorithm was developed specifically for time-dependant samples and it allows clear separation of gene expression profiles. four cluster groups (a, b, c, and d) were the optimal number derived from the analysis, in that all genes were distributed among the four clusters, reflecting a mapping of gene expression profiles to the four endometrial cycle phases, pe, ese, mse, and lse. this optimal clustering allowed all genes to be classified into these clusters, and no genes were unclassified (i.e. did not fit into any of the clusters). gene expression values of members of each cluster group were averaged to show one profile for graphic representation of each cluster group (see results).
gene ontologies (go) classification.
a web-based tool, go tree machine (gotm) (16), was used to interpret biological, molecular, and cellular functions of genes identified by both pairwise comparisons and k-means analyses in different phases of the menstrual cycle. gotm uses go hierarchies to discover significant biological processes, molecular functions, and cellular components in a gene list. gotm also implements a statistical analysis of the go categories for the input gene list and suggests biological areas that warrant further study (17). first, the differentially expressed genes are classified by their corresponding go categories, and the observed number of genes in each of these go categories is recorded. genes represented on the hg u133 plus 2.0 affymetrix chip comprise the reference gene list. the expected number of genes in each go category corresponds to the number of genes falling into that go category in the reference gene list. a given go category is considered enriched when the observed number of genes in that category is greater than the expected number.
microarray validation by real-time pcr
using rna isolated as described above, cdna was generated from 1 μg total rna from each sample using the omniscript reverse transcription kit (qiagen) with a 1:1 ratio of oligo-(dt)1618 (invitrogen) and random hexamers (invitrogen). real-time pcr was performed in triplicate in 25-μl reactions using the quantitect sybr green pcr kit (qiagen), according to the manufacturer’s instructions. the template cdna was diluted from the reverse-transcribed product by 4-fold for use in the pcr. the housekeeping gene ribosomal protein l19 was used as normalizer gene, because it displayed the lowest variation in expression levels as judged by microarray analysis and the smallest sd in threshold cycle (ct) values when compared with other known normalizer genes (-actin, glyceraldehyde-3-phosphate dehydrogenase, 2-microglobulin, and ubiquitin) (hamilton, a., s. talbi, m. nyegaard, m. overgaard, and l. giudice, unpublished data). most pcr primers were designed to be intron spanning to serve as a control for contamination of genomic dna and to have an optimal amplicon range of 150260 bp (table 3). for the primer pairs that were not intron spanning, control templates with no reverse transcriptase enzyme were run for each sample. pcr were run using the mx4000 q-pcr system (stratagene, la jolla, ca). the thermal cycling conditions included an initial activation step at 95 c for 15 min, followed by 40 cycles of denaturation, annealing, and elongation (94 c for 30 sec, 5760 c for 30 sec, and 72 c for 30 sec, respectively). fluorescence data collection was performed during the annealing and elongation steps. pcr products were analyzed by thermal dissociation (5595 c) with a fluorescence measurement at every 1 c increment to ensure the production of a single pcr product. for the normalizer and each gene of interest, a standard curve was generated using serial dilutions of template dna. the template for each gene of interest standard curve was chosen based on the highest levels of overall expression during the menstrual cycle. the dilutions of template used were 1:4, 1:16, 1:64, and 1:256. the efficiencies of amplification (eff) for each gene were calculated by the following equation: eff = 10[1/slope] x 1. genes with standard curves that showed approximately 100% amplification efficiency (± 5%) were used for subsequent sample analyses. the relative expression ratio (r) was calculated based on the corresponding efficiencies of amplification for each treatment compared with the control and the differences in ct values (ct) using the following mathematical model:
(1)
where r represents the ratio at which a given gene of interest (goi) is expressed in the treatment group, relative to the control group, when normalized by ribosomal protein l19. ct values were calculated by the mx4000 software based on fluorescence intensity values after normalization with an internal reference dye and baseline correction. pairwise comparisons were performed between treatment vs. control: ese vs. pe, mse vs. ese, and lse vs. mse. all validation experiments were performed using n = 3 subject samples in each group. statistical analysis of the pcr data were performed using the relative expression software tool (rest) that employs a pairwise fixed reallocation and randomization test to determine significance (18).
results
endometrial genes: clustering, cluster trees, and cluster patterns
pca.
pca of the expression profiles (fig. 1), using all of the genes and ests on the affymetrix gene chip, demonstrates a clear segregation of the 22 well-characterized samples into four clusters, corresponding to pe, ese, mse, and lse. pca distributes samples into a three-dimensional space based on the variance in gene expressions, and samples that have similar trends in their gene expression profiles cluster close together in the pca plot. clustering (fig. 1) was not dependent on subject diagnosis or how the samples were obtained (curetting endometrium from hysterectomy specimens or by endometrial biopsy) (table 1). with regard to the six ambiguous samples, nos. 421 and 642 segregate with the pe group, and no. 389 segregates with the ese group (consistent with the single pathologist’s histological assessment of this sample (table 2). sample 407 segregates with the lse group, as do nos. 494 and 648. note that the cluster analysis assigned the latter to lse, which was one of the two different histological assessments by the pathologists (table 2).
hierarchical clustering analysis.
the microarray gene expression profiles of the 22 well-characterized endometrial samples were subjected to unsupervised hierarchical clustering analysis, based on the input of the combined list of genes differentially expressed throughout different phases of the cycle. the six samples with ambiguous histology were also included in this analysis to investigate where they might cluster. hierarchical clustering analysis draws relationships between samples and generates a dendrogram that illustrates these relationships, in contrast to pca, which does not draw relationships between samples. overall, hierarchical clustering analysis resulted in a tree-like dendrogram of sample clustering and a heatmap of gene expression (fig. 2a). striking segregation of samples into two major clustering branches (nos. 1 and 2) was observed. pe and ese samples self-cluster into branch 1, which contains no samples with histological dating in the mid- or late-secretory phase. all mse and lse samples self-segregate into the second major branch (no. 2), with no pe or ese samples in this cluster. further analysis of branch 1 reveals two major sub-branches, 1a and 1b (fig. 2a). sub-branch 1a contains all pe samples, whereas sub-branch 1b contains only ese samples. histologically ambiguous samples 421 and 642 cluster in the pe group, and sample 389 clusters with the ese group, as they did with pca (fig. 1). analysis of branch 2 reveals two major sub-branches, 2a and 2b, which each have further subdivisions. sub-branch 2a contains only mse, and sub-branch 2b contains only lse samples. ambiguous samples 407, 494, and 648 all cluster with lse samples, as they did in the independent analysis of pca (fig. 1).
with regard to the heatmap of gene expression (fig. 2a), within the major sub-branches there are common patterns of intensities of gene expression, although some dissimilarities are also apparent, especially in the two major lse sub-branches. as with the dendrograms above and the pca analysis, similarities and differences among samples were not dependent on subject diagnosis, method of specimen collection, or underlying pathology for which a surgical procedure was indicated. with regard to the ambiguous samples, their gene expression profiles are similar to other members of their cluster group (e.g. nos. 421, 389, 407, 494, and 648), except for no. 642, which is in its own sub-branch and has some obvious differences from the other samples in branch 1a. it is interesting that this sample was histologically read as atrophic by one pathologist and questionable hormone effect by another.
k-means analysis, cluster groups, and gene expression patterns.
k-means clustering is a method that identifies a set of genes whose expression is regulated in a similar way across different experimental conditions. this is particularly useful in identifying genes with similar regulation across the menstrual cycle. k-means clustering applied to differentially expressed genes across the endometrial cycle phases revealed four major groups (a, b, c, and d) (fig. 2b, left). these groups distinguish themselves by the following characteristics: cluster a, high expression of genes in samples that cluster as pe and low expression in the rest of the cycle; cluster b, high expression of genes in samples identified as ese and low expression in pe, mse, and lse; cluster c, high expression of genes in mse; and cluster d, high expression in lse compared with the rest of the cycle. figure 2b (right) illustrates the k-means analysis results (average gene expression profile for each cluster) that led to these clusters. these are described in more detail in the next section with regard to go, biological processes, molecular functions, and cellular localization of events.
cluster group go
cluster a.
go analysis (table 4) (complete go tables are published as supplemental data on the endocrine society’s journals online web site at http://endo.endojournals.org) revealed high expression of genes in the proliferative phase that are involved in cell adhesion, cell-cell signaling, cell cycle regulation, and cell division (e.g. dna replication, strand elongation, dna metabolism, chromatin cycle and segregation, mitosis, g1/s and g2m transitions, and response to dna damage). other genes/families included in cluster a include collagen metabolism and extracellular matrix regulation, signal transduction, development, and regulation of enzyme and ion channels. molecular functions highly represented in this cluster group are those related to cell cycle regulation and dna synthesis, steroid binding, receptor-mediated events, and extracellular matrix structural constituents. the cellular components involved in the proliferative phase of the menstrual cycle include primarily the chromosomes, the replication fork, dna polymerase/primer complexes, microtubule skeleton, the extracellular matrix, non-membrane-bound organelles, and collagen. these processes are highly consistent with the cell replication and growth of the tissue in the proliferative phase, extracellular matrix remodeling that occurs with this growth, and steroid hormone action and subsequent signaling to effect these processes.
cluster b.
cluster b has high gene expression in early secretory phase and low gene expression in the remainder of the cycle. this cluster is characterized by remarkably different biological processes, compared with the proliferative phase (table 4), which include genes that govern metabolism and/or biosynthesis of cholesterol, amino acids, organic acids, acetyl coenzyme a and lipids, fatty acids, steroids, and prostaglandins and also genes for transport of ions, peptides, and carboxylic acids. to affect these processes, a variety of coenzymes, hydrolases, isomerases, oxidoreductases, prostaglandin d synthase, coenzyme a carboxylase, and transporters are involved. the cellular components involved are cell membranes, microsomes, peroxisomes, and mitochondria. it is known that the early secretory phase is characterized by robust glycogen synthesis along with high energy demands and a concomitant increase in mitochondrial size and number of cristae (1, 7), which is consistent with the molecular events derived from the clustering analysis.
cluster c.
genes in cluster c have high expression in the mid-secretory phase. in mse (table 4), these include genes for cell communication, intracellular signaling cascades, negative regulation of cell proliferation, cell ion homeostasis, metabolism and synthesis of amino acids, organic acids, and polysaccharides, and the first appearance of genes involved in the immune response and in response to chemicals, ions, wounding, and stress. consistent with these biological processes are molecular functions of glutathione and metallothionein activities, a variety of protein and enzyme binding, glycoprotein biosynthesis (acetyl-glucosaminyl transferase activity), inhibition of protease activities, and a variety of transporters. these processes occur in the plasma membrane and other cellular membranes, in the cytoplasm, cell-cell adherens junctions, the basement membrane, and the extracellular region.
cluster d.
in cluster d, the following biological processes are represented in the late-secretory phase (table 4): cell motility and communication, cell adhesion and interactions with the matrix, signal transduction and intracellular signaling cascades including the mapk cascade, regulation of the cell cycle with the primary process of cell cycle arrest and negative regulation of cell proliferation, cellular physiological processes, and cell activation as well as apoptosis, cleavage of cytoskeletal and extracellular matrix proteins, and icosanoid biosynthesis. in addition, genes are highly expressed that regulate endocytosis, phagocytosis, blood coagulation/hemostasis, and the immune response [including regulation of lymphocyte and t-cell differentiation and activation, cellular and humoral immune responses, antimicrobial humoral response, and natural killer (nk) cell activity] as well as responses to inflammation, pathogens, chemicals and xenobiotics, and wounding. in addition, chemotaxis and wound healing are represented in this cluster. molecular functions include carbohydrate, protein, cytokine, and ig binding, il-1 receptor and tgf receptor binding, protease and protease inhibitor activity regulation, cyclin-dependent protein kinase inhibitory activity, signal transduction activity, icosanoid and prostanoid receptor activity, thromboxane receptor activity, and hematopoietic/interferon-class cytokine receptor activity (for details, see the supplemental data published on the endocrine society’s journals online web site at http://endo.endojournals.org). these processes occur primarily at the plasma membrane, vacuoles, lysosomes, and the extracellular matrix, and are consistent with this phase having known influx of leukocytes and, if no conception occurs, preparation for tissue desquamation with concomitant vasoconstriction.
relative gene expression across the menstrual cycle
in this section we present the results of pairwise comparisons of genes expressed in distinct phases of the menstrual cycle relative to each other, specifically, ese vs. pe, mse vs. ese, and lse vs. mse. this is in contrast to the k-means clustering in the previous section that is based on profiles throughout all four phases of the cycle. only genes that are statistically significantly different by one-way anova and have at least a 1.5-fold -change were considered. their gene ontology categories, fold change, and gene families are presented below and in accompanying tables.
early-secretory vs. proliferative endometrium
in comparing ese vs. pe, 1589 genes and ests were significantly up-regulated and 1470 were significantly down-regulated. the go classifications for up-regulated genes in ese vs. pe are shown in table 5. during the transition to the early-secretory phase there is an up-regulation of metabolism of alcohols, amino acids, lipids, fatty acids, and icosanoids, a large representation of transporters for biological molecules participating in these metabolic processes, and negative regulation of cell proliferation, among others. the most highly up-regulated genes in the early-secretory phase, transitioning from the proliferative phase, are shown in table 6. (select gene lists for select go categories are in supplement a on the endocrine society’s journals online web site at http://endo.endojournals.org). interestingly, these genes govern specifically estradiol availability within the endometrium [17-hydroxysteroid dehydrogenase (17-hsd) type 2, which converts e2 to estrone (e1)] and secretory proteins [including members of the secretoglobin family and osteopontin (secreted phosphoprotein)]. in addition, transporters for amino acids, peptides, and ions are highly up-regulated, as are enzymes involved in collagen metabolism and other processes [matrix metalloproteinase 26 (mmp-26)], a variety of cytochrome p450 proteins (important in electron transport and energy pathways), and regulators of blood coagulation (il-20r1 and tissue factor pathway inhibitor 2). inhibitors of wnt signaling are tightly regulated in this part of the cycle, with dkk-1 being up-regulated 6-fold, and secreted frizzled-related protein 1 (sfrp-1) down-regulated (see below). aquaporin 3, important in water transport, is up-regulated, and mucin 1 (muc-1), important as an antibacterial agent and surface lubricant (19), is first noted to be up-regulated in this part of the menstrual cycle. lipid metabolism, phospholipase activity, and icosanoid/prostaglandin metabolism are highly up-regulated. the data further demonstrate that metallothioneins are first noted to increase in the cycle in the early-secretory phase, as are arginase 2, phospholipase c, keratin 8, il-1r type i, the gaba-r -subunit, il-15, and monoamine oxidase, which are commonly attributed to being mid-secretory endometrial products (20). evidence of epidermal growth factor signaling is apparent, as is an antiapoptotic marker (foxoa1a).
with regard to down-regulated genes, go biological processes that are down-regulated in ese vs. pe (table 7) include cell motility, communication, and adhesion, wnt receptor signaling, and the i-b kinase/nuclear factor (nf)-b cascade. strikingly, and not unexpectedly, genes involved in cell cycle regulation and cellular mitosis are highly represented among the down-regulated processes in this phase of the cycle (table 7). also, tissue regeneration, proteolysis, blood coagulation, and wound healing are down-regulated. with regard to specific genes, the most highly down-regulated genes in the early-secretory phase are mmp-11 (stromelysin), tissue plasminogen activator (tpa, plat), and adam-12. there is also evidence for suppression of retinoic acid, as well as igf and tgf family members (tgf1, inhibin a, and bmp-1 and -7) and angiogenic factors (fgf1), hgf and fgf receptor (fgfr1, -2, and -3), il-4 signal transduction, and specific activators of protein kinase activities (table 8 and supplement b). the extracellular matrix reveals down-regulation of several members of the collagen family, laminin, dermatopontin, and tenascin c.
mse vs. ese
comparison of mse vs. ese revealed that 1415 genes and ests were significantly up-regulated and 1463 were significantly down-regulated. the early- to mid-secretory phase transition is characterized by biological processes that are strikingly distinct from the proliferative to early-secretory phase transition. these include (table 9) cell motility, adhesion, and communication; intracellular signaling cascades (including up-regulation of the i-b kinase/nf-b cascade); inhibition of apoptosis; cytoskeletal organization and anchoring; and metabolism and transport of a variety of amino acids and organic anions. also, the first appearance is made in the cycle of mechanisms put into place for hemostasis; nitric oxide biosynthesis; immune response and responses to chemicals, reactive oxygen species, metal ions, stress, and wounding; regulation of chemotaxis; and wound healing. the molecular functions accompanying these processes (table 9) include antioxidant activity; polysaccharide, calcium ion, and heparin binding; a variety of cytokines, growth factors, and their receptor binding and signal transducer activities; protease and inhibitor activities; transporter activities; and biosynthesis of secreted proteins.
the most highly up-regulated genes (table 10) are the small inducible cytokine subfamily member 14 (cxcl14), paep (glycodelin), glutathione peroxidase-3 (gpx3), a variety of solute carrier family members, cartilage oligomeric matrix protein (comp), decay-accelerating factor (daf), superoxide dismutase (sod) 2, defensin, monoamine oxidase a (mao a), complement component 4, and tanscobalamin and ceruloplasmin (transporters for vitamin b12 and copper, respectively). in addition, gastrin and laminin family members are highly up-regulated, as are genes involved the immune response, dkk-1, igfbp-1, claudin 4, il-15, apo d, lif, g0s2, and star (table 10).
several gene lists associated with the go classifications (supplement c) warrant further presentation because of the insight that they convey with regard to the molecular participants in these processes. in particular, angiogenic factors that are up-regulated in mse vs. ese include endothelial cell growth factor 1, vascular endothelial growth factor, and fibroblast growth factor 18. other growth factors include hepatocyte growth factor, inhibin b, leukemia inhibitory factor (lif), cxcl12 (stromal derived factor 1/), and tgf2 (supplement c). the gene list underscores the participation of cytokine signaling through ilr, il4r, il6r, and ccr1. with regard to the immune response, members of the complement family are up-regulated in mse, as are receptors for nk cells and enzymes associated with macrophages (granulysin, granzymes, and perforin) (table 11 and supplement c). in fact, protease activities are primarily associated with lymphocytes and also include multiple cathepsins and members of the adam family, and there is a spectrum of protease inhibitor activities concomitantly up-regulated. calcium homeostasis and binding are a hallmark of the mid-secretory phase (supplement c). the response to wounding and chemicals involves genes associated with angiogenesis and detoxification (e.g. multiple members of the metallothionein family). in addition, there is unique up-regulation of antioxidant activities in mse, including glutathione peroxidase 3 (gpx3), apoe, and prostaglandin synthase 1. also of interest is the receptome (21) in mse (supplement c) that demonstrates all of the receptors in this phase of the cycle. these include receptors for calcitonin, cytokines, and growth factors (see above), integrins, nk cells, cellular defense (tlr4), and prostanoids. the receptome also gives insight into signaling pathways, including il-6 signal transduction (lif, il-2, and oncostatin m receptor) and signaling through g protein-coupled receptors as well. finally, in the extracellular matrix, there is an up-regulation of laminin subunits, including 4, 5, 3, and c1, characteristic of the developing extracellular matrix as stromal cells begin to decidualize.
with regard to down-regulated processes in mse vs. ese (table 12), there is striking down-regulation of genes involved in cell division, metabolism, and biosynthesis of components required for cell division, biosynthesis of fatty acids, lipids, steroid hormones, and steroid hormone receptors. included in the latter are the progesterone receptor (pr) membrane component 1, the pr, estrogen receptor (er) (er did not show regulation), and indian hedgehog (ihh). among the most highly down-regulated genes in the transition from ese to mse (table 13) are matrix-degrading enzymes, alkaline phosphatase, genes involved in prostaglandin metabolism, endothelin 3, igf-i, and wnt members (sfrp4, frizzled homolog 1, dkk-3). homeobox genes (msx1, msx2, and hoxa10) were also down-regulated, as were the folate receptor, msh4, ephrin e3, cell cycle regulators, and transport activities (supplement d).
lse vs. mse
comparison of lse vs. mse revealed that 1190 genes and ests were significantly up-regulated and 1027 were significantly down-regulated. up-regulated genes derived from the genome-wide analysis of lse vs. mse reveal numerous biological processes (table 14) that include cell motility, cell communication, cell adhesion to the matrix, cell-cell signaling, apoptosis, proteolysis of cytoskeletal and extracellular matrix proteins, hemostasis, vasoconstriction, immune response, response to wounding and chemical substances, chemotaxis, and wound healing. the molecular functions include antigen, carbohydrate, protein, ig, calmodulin, cytokine, growth factor, and actin binding. of note is the up-regulation of proteases (especially metalloproteases) and their inhibitors, microfilament motor activity, transmembrane tyrosine kinase receptor signaling components, g protein-coupled receptor and integrin-mediated signaling, and involvement of the ras protein kinase cascade.
among the most highly up-regulated genes in lse vs. mse (table 15) are mmp-10, -7, and -11; ebaf; igfbp-1, -3, and -6; tgf family members (inhibin a, ltbp-2, and tgf3); thrombospondin 1; nidogen; asporin; potassium channels; timp-3; il-1; microfibrillar-associated protein 2; and granzymes a and b. several members of the integrin family (2, m, x, 2, and 3) are also up-regulated (supplement e). interestingly, the cytokines that are most prevalent in lse are il-1, inhibin a, bone morphogenetic protein-2 and are signaling through the il-2 receptor. with regard to extracellular matrix proteolysis (supplement e), there is up-regulation of metalloendopeptidase activities (mmps 2, 7, 10, 11, 19, and 27 and members of the adam family, adam 8, 12, 19, 28, ts2, and ts5). in addition, the plasminogen activators (tpa/plat and upa/plau) are also up-regulated. timp-2 and -3 are up-regulated, as are lxn (a carboxypeptidase inhibitor) and csta (cystatin a), as well as thrombospondin 1. the immune response is more robust and further up-regulated in lse (table 16) vs. mse (table 11), with the appearance of class ii molecule up-regulation (hla-dpa1, dpb1, dqb1, and dra) as well as a variety of nk cell receptors and, uniquely, compared with the mse vs. ese phase, up-regulation of il-1 and -, the il2r and , and the csf1r, as well as fc receptor subtypes (table 16). genes encoding members of the humoral immune response are also up-regulated in lse vs. mse. consistent with preparing for tissue desquamation (and accompanying cell debris), there is an up-regulation of genes involved in phagocytosis, wounding, and wound healing as well as cellular apoptosis and cytoskeletal protein degradation (supplement e). interestingly, there is up-regulation of components of g protein signaling, kal1, serotonin receptor, and deiodinase 2.
the receptome for lse vs. mse (table 16 and supplement f) primarily reflects immune cell receptors, including hematopoietin/interferon class 2 activities; major histocompatibility complex (mhc) class i, mhc class ii, and hla-c-specific inhibitory to the mhc class i receptor activities; nk inhibitory and stimulatory receptors; the t cell receptor -locus; and scavenger receptor activities. furthermore, there is receptor activity for il-2, il-8, tgf1, tgf2, inhibin a, and il-10 and membrane protein tyrosine kinase activity for csf1r, igf1r, kdr, and prlr.
among the down-regulated processes in lse vs. mse (table 17) are cell motility, migration, and adhesion; negative regulation of monocyte, osteoclast, and myeloid blood cell differentiation; ion homeostasis; cellular proliferation; and metabolism of sugars, amino acids, steroid hormones, lipids, fatty acids, and molecules needed for dna synthesis. genes that are most highly down-regulated (table 18 and supplement f) are mcp-2 (scya8), cxcl13 (scyb13), secretoglobins, s100p, metallothioneins, mmp-26 (collagen metabolism), dkk-1, igf-i, ptgds, mme, mao-a, adipsin, 17hsd2, lif, gastrin, and members of the complement family. most of these down-regulated genes in lse vs. mse are up-regulated in mse vs. ese and thus peak in mse.
a second, independent analysis of statistically significantly regulated genes in lse vs. mse was conducted using the sam algorithm. more than 85% of the genes in the analysis by one-way anova were also found to be regulated by the sam analysis, and genes up- and down-regulated that are common to both analyses are marked with an asterisk in tables 15, 16, and 18.
validation of microarray data by real-time pcr
for validation by real-time pcr, we tested 29 comparisons from the microarray data analysis. some gene changes have been previously identified as being regulated during various phases of the cycle, whereas others have not. this approach resulted in 29 comparisons regulated as in the microarray analysis for a concordance rate of 100%. of these 29 validated comparisons, 24 were statistically significant, yielding a concordance of 83% (fig. 3). also, 21 of the 29 comparisons had a fold change of 2.0 or higher by the microarray analysis, of which 18 were statistically significant by real-time pcr and rest analysis. the remaining eight comparisons fell into the 1.5- to 1.99-fold change range by microarray analysis, of which six were statistically significantly regulated in the real-time pcr analysis. in the comparison of ese vs. pe, there is consistent up-regulation of metallothionein 1h (mt1h), dkk-1, and timp-3. we also observed down-regulation of thrombospondin 1 (thbs1), mmp-11, and sex-determining region y-box 4 (sox4, a transcription factor that may be involved in apoptotic pathways) (22) (fig. 3a). in the comparison of mse vs. ese, numerous genes are consistent with up-regulation observed with the microarray data. igfbp-1 shows a higher increase by real-time pcr in the mse vs. ese comparison than observed with the microarray analysis. among other significantly up-regulated genes are cxcl14 (the most highly regulated gene by microarray analysis), decay-accelerated factor (daf), thbs1 (which continues to rise in the mid-secretory phase), il-6 signal transducer (il6st), dkk-1, leptin receptor (lepr), solute carrier family 16, member 3 (slc16a3), and star. timp-3 was up-regulated but not significantly. hoxa10 and igf-i are down-regulated, consistent with the microarray data, although the igf-i data were not statistically significant (fig. 3b). in lse vs. mse, significant up-regulation of endometrial-associated bleeding factor (ebaf), igfbp-1, mmp-11, sox4, thbs1, igf1r, and timp-3 is observed, consistent with the microarray data. among the down-regulated genes, significant down-regulation of mt1h, hoxa10, and hoxa11 was observed, and (nonsignificant) down-regulation of igf-i was also observed (fig. 3c).
discussion
general comments
this study provides the first, whole genome-wide gene, go, and gene clustering analyses across the entire menstrual cycle in normo-ovulatory women. several observations are worthy of comment and are presented in more detail below, including the comparison of the pca and hierarchical clustering analyses and assignment of endometrial samples with ambiguous histology; how consistent the cluster groups, specific genes, and biological processes are with histological observations and tissue-based analyses; and what the relationship is between the current observations, previous analyses, and the emerging literature on microarray analyses in different phases of the menstrual cycle. we also discuss whether the data provide further insight into e2 and p regulation of specific genes in endometrium and its cellular components, what information is new in the current study (specifically, comparison of ese vs. pe and lse vs. mse), and how the data compare with those in previous reports (pe vs. menstrual, mse vs. ese, and mse vs. pe) and a recent cdna array assessment (10,500 cdna sequences) across the cycle (23).
clustering and identification of histologically ambiguous samples
in the current study, we used two independent clustering algorithms to analyze data generated from the microarray experiments and to analyze how samples cluster together based on similarities in their gene expression profiles. all 54,600 probe sets on the affymetrix chip were used in the pca, whereas the hierarchical clustering analysis used a more limited gene set (7231) derived from the pairwise comparison combined gene list. the finding of equivalent clustering patterns generated by these two methods using different gene lists and different analytical approaches supports segregation of phases of the menstrual cycle based on their unique gene expression profiles (molecular signatures). furthermore, assignment of menstrual cycle stage of ambiguous samples, based on their gene expression profiles and their cluster grouping, is a powerful adjunct to the historical histological gold standard of endometrial assessment. that ambiguous samples were classified in the same cycle phase using both pca and hierarchical clustering, despite the fact that all genes and ests were used in the former and only the differentially expressed genes from the pairwise comparisons were used in the latter analysis, further underscores the potential for these approaches in endometrial diagnostics. it is important to note that although obtaining specimens in the secretory phase would be ideally timed to the lh surge for appropriate histological assignment and having circulating e2 and p levels would give insight into steroid hormone effects on tissues obtained, this study demonstrates that samples obtained at any time in the cycle have unique molecular signatures that preclude the need to assign a histological phase to the sample a priori. although unique gene signatures lead to clustering in groups (phases), it is also important to recognize that the molecular signatures are not identical and may reflect subject-to-subject variability, complement of cell types in a specimen, and other factors, described in more detail at the end of the discussion. these results set the stage for larger-scale studies to investigate the utility of identifying biologically similar samples of endometrial tissue through their gene expression profiling, used adjunctively with histological assessment. it remains to be seen whether abnormal endometria will have gene expression profiles that cluster according to cycle phase or whether new cluster patterns will evolve.
of interest is a recent cdna microarray study of gene expression across the menstrual cycle (23). initial hierarchical clustering analysis of 43 endometrial samples revealed two main branches: one containing menstrual, early pe (epe)/mid-pe (mpe), late pe (lpe), and ese/mse, and the second containing menstrual, menstrual/epe, mse, lse, and lse/menstrual. after removal of six outliers (for disagreement of histological stage between two pathologists or for poor hybridization), the two main branches for 37 samples contained two branches: one branch with menstrual, epe/mpe, lpe/ese, and ese/mse and the other having menstrual, mse/lse, and lse/menstrual. in our study, before we performed pairwise comparisons, we removed samples with poor hybridization and ambiguous histology, and our data fell into branches that correspond with phases of the cycle with no overlap from branch to branch. in the study by ponnampalam et al. (23), some cycle phases had too few samples for the analysis and thus were merged (e.g. mse and lse), in contrast to the current study where numerous samples were in these phases of the cycle, and this may account for some of the differences observed (discussed below).
k-means analysis
k-means analysis of gene expression profiles provides an unbiased approach to investigate patterns of gene expression in tissues and cells and to investigate biological similarities among specimens. in the current study, four cluster groups were identified that reveal biological processes in pe, ese, mse, and lse, some of which have already been identified and some of which are new. it is well known, e.g. that in lpe, under the influence of e2, cellular constituents of the endometrium undergo proliferation, as evidenced by high mitotic indices, extensive dna synthesis (24), and increasing height of the tissue (20). the genes, biological processes, molecular functions, and cellular components in cluster a are consistent with mitotic activity and tissue remodeling necessary for growth in pe. however, of particular interest are ion channels in pe, which, to date, have received little attention in terms of the physiology of the endometrium, their dependence on e2, and as potential therapeutic targets (e.g. clinically thin endometrium with concomitant fertility compromise or thickened endometrium and hyperplasia/cancer). in contrast, the large number of genes related to cellular metabolism in cluster b underscore the unique metabolic capacity of the ese phase. in this phase, the cluster analysis is consistent with energy generation for glycogen synthesis, which occurs in the absence of excessive glycogen intake (7). major transporters and some secretions as well as genes for germ cell migration may facilitate sperm transport and assure an aseptic environment. clusters c and d further reveal processes related to the immune response with regard to the innate immune system and the cellular and humoral immune responses that are even more clearly revealed in the pairwise analyses (see below). we are unable to compare this go analysis with the data of ponnampalam et al. (23), because the latter were not analyzed by go.
in the current study, we observed four major kinetic patterns of gene expression across the menstrual cycle (pe, ese, mse, and lse). in the cdna microarray analysis of gene expression in samples obtained from across the cycle (23), more cycle phases were evaluated (epe/mpe, mpe, lpe/ese, ese/mse, mse/lse, lse/menstrual, and menstrual), and thus more k-means patterns were obtained. because the k-means categories were not equivalent in the two studies, it is difficult to compare genes that are regulated in similar ways across the cycle. nonetheless, some genes do show similar regulation, including high expression in ese (keratin-8), high expression in mse [glutathione peroxidase 3 (gpx3), annexin 4, scyb14 (cxcl14), s100p, and aquaporin-3], high expression in mse and lse [apo-e and stanniocalcin 1 (stc1)], and high expression in lse (sox-4, adamts5, gng-4, integrin 2, and the prolactin receptor).
pairwise comparisons
ese vs. pe.
ese in situ is stimulated by high levels of circulating e2 during the proliferative phase and then by low, but rising, levels of p (and e2). pairwise comparison of genes differentially expressed in ese vs. pe has not heretofore been reported. of interest in the gene list for the comparison of ese vs. pe is a mixture of potentially e2- and p-regulated genes, although for most genes regulated in this transition, it is not known with certainty which steroid hormone (e2 or p) regulates their expression. a recent study on human endometrium, using the same affymetrix chip as in the current study, may shed some light into e2-regulated genes in this tissue (25). in this study, gene expression in lpe (high e2) vs. menstrual endometrium (very low e2 levels) was investigated, as were genes regulated in explant cultures of tissues from both phases and treated with e2. genes up-regulated in lpe vs. menstrual endometrium included, e.g. oviductal glycoprotein-1, connexin-37, olfactomedin-1, and sfrp4, and down-regulated genes included several mmps (-1, -3, and -10), il-1, il-8, il-11, inhibin a, sox4, and cc-ligand 20, among others (25). in our data, olfactomedin-1 was down-regulated in ese, suggesting that p inhibits its expression in this phase of the cycle. sox4 was down-regulated in lpe vs. menstrual endometrium (25), and it is also down-regulated in ese vs. pe (table 8), suggesting that e2 down-regulates this gene. in a recent study on global gene expression (12,000 genes/ests) in response to e2 treatment of cultured human endometrial cells, n-cadherin was up-regulated by e2 (26). however, in the current study, n-cadherin was down-regulated in ese vs. pe, suggesting that n-cadherin expression is inhibited by p. of interest, also, is the up-regulation of foxo1a (2.1-fold) in ese vs. pe, especially in view of recent data in breast cancer cells that demonstrate the importance of foxo1a in e2 action (27). whether there is dysregulation of this transcription factor in endometrial abnormalities awaits further investigation.
in a recent study by tan et al. (28), global gene profiling of mouse uterus during the estrous cycle was investigated. of interest among regulated genes in estrus vs. diestrus is the up-regulation of 17hsd-2; cathepsins l, h, and s; 1 protease inhibitors 1 and 5; n-myc downstream regulator (ndr1); regulation of g protein signaling 2 (rgs2); and complement c3 and h. the up-regulation of 17hsd-2 in ese vs. pe in the current study would suggest, based on analogy of the data from estrus vs. diestrus study, that 17hsd-2 is e2 regulated in ese vs. pe (table 6). however, the data are convincing that 17hsd-2 in human endometrium is regulated vi
