Mol. Hum. Reprod. Advance Access originally published online on October 22, 2004
Molecular Human Reproduction 2004 10(12):879-893; doi:10.1093/molehr/gah121
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Molecular classification of human endometrial cycle stages by transcriptional profiling
1Centre for Women's Health Research, Monash University Department of Obstetrics & Gynaecology, 246 Clayton Road, Victoria 3168, 2Victorian Bioinformatics Consortium, Monash University, Clayton, 3CSIRO, Mathematical and Information Sciences, Clayton South, 4Anatomical Pathology, Monash Medical Centre, Clayton, Victoria, Australia
5 To whom correspondence should be addressed. Email: anna.ponnampalam{at}med.monash.edu.au
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Endometrium is a dynamic tissue that undergoes cyclic changes each month, under the overall control of estrogen and progesterone. The aims of this study were to investigate the changing global gene expression profile of human endometrium during the menstrual cycle using microarray technology and to determine the correlation between histopathological evaluation and molecular profile of the samples. Standard two-colour cDNA microarrays were performed on the 43 samples against a common reference, using a 10.5 K cDNA glass slide microarray. The results were validated using real-time PCR. Analysis of expression data was carried out using parametric analysis of variance with Benjamini–Hochberg correction. Hierarchical clustering reveals a strong relationship between histopathology and transcriptional profile of the samples. The study identified 1452 genes that showed significant changes in expression (P
0.05) across the menstrual cycle, with 425 genes having changes that are at least 2-fold. The data were also independently analysed by a CSIRO algorithm called GeneRaVETM that identified a small subset of genes whose expression profiles could be used to classify nearly all the biopsies into their correct cycle stage. We also identified and validated three genes [(natural cytotoxicity triggering receptor (NCR)3, fucosyl transferase (FUT)4 and Fyn-binding protein (FYB)] that had not been shown to have significant cyclic changes in the human endometrium, previously. We have shown for the first time that endometrial cycle stage prediction is possible based on global gene expression profile. Key words: endometrium/FUT4/FYB/menstrual cycle/NCR3
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The endometrium undergoes cyclic changes each month, under the overall control of fluctuating levels of estrogen and progesterone. This hormone dependent endometrial remodelling is a continuum of structural and functional changes that make up the menstrual cycle. The endometrium regenerates and grows under the influence of estrogen during the proliferative phase of the cycle, while during the secretory phase it undergoes secretory transformation under the combined influence of estrogen and progesterone. In the absence of pregnancy, the abrupt fall in the levels of circulating estrogen and progesterone leads to the functional layer of the endometrium being shed during menstruation. The typical menstrual cycle lasts for about 28 days, although its normal length can vary from 21 to 35 days (Glasser et al., 2002, Ch.2
).
Over five decades have passed since the histological criteria for dating of the endometrial biopsy were established by Noyes et al. (1950). This approach has remained the gold standard among methods in clinical practice for the evaluation of endometrial maturation and infertility. Recent advances in reproductive physiology and molecular biology have raised questions about the usefulness of morphological criteria alone, in determining clinical or biological function. According to Noyes et al.'s (1950) criteria, a biopsy specimen is considered abnormal when there is out of phase histological maturation of more than 2 days, based on the assessment of a relatively small number of structural features. A recent study found that these structural features may be much less temporally distinct and discriminating than originally thought (Murray et al., 2004). Numerous studies have shown that significant variability in the length of the secretory phase is normal, adding further uncertainty to temporal classification solely based on structural appearance (Treloar et al., 1967; Johannisson et al., 1982; Lenton et al., 1984; Munster et al., 1992). Hence, the dating criteria proposed by Noyes et al. (1950) not only has the potential to miss pathologies, which are not reflected by obvious histological change, but can also generate inter-observer variability that may, in extreme cases, lead to different clinical management strategies. Attempts to refine histopathological assessment by combining it with measurement of the LH surge, doppler ultrasound and serum progesterone have produced contradictory results which on broad terms have failed to improve the outcome of endometrial dating (Kim-Bjorklund et al., 1991; Corsan et al., 1992; Martinez et al., 1992; Arthur and Khan, 1996; Santoro et al., 2000 Sterzik et al., 2000).
The significant histological, biological and physiological changes that occur during the proliferative, secretory and menstrual phases of the cycle are ultimately the result of changes that occur at the level of gene transcription. Although the use of chronological and histological dating has improved the understanding of anatomical changes within the endometrium during the menstrual cycle, most of the molecular pathways that are responsible for these changes remain unknown. Viewing the broader transcriptional map of the endometrium throughout the menstrual cycle is likely to enhance overall understanding of endometrial function as well as giving an insight into the molecular processes that initiate histological changes throughout the menstrual cycle. Microarray gene expression profiling allows large numbers of genes to be investigated simultaneously, and the resulting gene expression patterns can then be correlated with different structural, functional or clinical parameters (Liotta and Petricoin, 2000). We propose that classifying the stages of the menstrual cycle based on global gene expression of the endometrium will improve understanding of the functional changes that occur in the endometrium, and help to identify abnormalities that are not apparent histologically. Hence, the aims of our study were to investigate the global gene expression profile of the human endometrium during the menstrual cycle, and to compare histopathological dating of the endometrium with the molecular profile. Our hypotheses were that gene expression profiles of the different phases of the menstrual cycle would correlate with the histopathological dating of the endometrium, and that we would identify some endometria with apparently normal histology but abnormal gene expression profiles.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Tissue collection and cycle staging
Ethical approval for the study was obtained from Southern Health Human Research and Ethics Committee B. Endometrial curettings were obtained from 43 normal cycling women after informed consent. All endometrial samples used in the study were classified as normal by routine histopathology. Subjects ranged in age between 18 and 47 years and had not used hormonal contraception in the 3 months prior to tissue collection. A small portion of the tissue was sent to histopathology to evaluate the cycle stage. An experienced independent pathologist later reconfirmed the cycle stage of each tissue sample. The menstrual cycle was divided into seven stages by histopathology, based on well-established criteria (Noyes et al., 1950
): early-proliferative (EP), mid-proliferative (MP), late-proliferative (LP), early-secretory (ES), mid-secretory (MS), late-secretory (LS) and menstrual (M). Endometrial tissues were snap frozen on dry ice immediately after collection and stored at –80°C until RNA extraction.
RNA extraction
Total RNA was extracted using Trizol reagent (Invitrogen Life Technologies, Australia), as previously described (Weston et al., 2002). Curettings were homogenized (1 ml Trizol/100 mg tissue) and incubated at room temperature for 5 min. After the addition of chloroform (0.2x volume of Trizol), samples were incubated for another 3 min at room temperature, centrifuged for 15 min at 12 000 g (4°C) and the aqueous phase RNA was precipitated with an equal volume of 100% ethanol. Total RNA was further purified by running it through a Qiagen RNeasy column (Qiagen, Germany) according to manufacturer's protocol. RNA was resuspended in RNase-free water and ethanol precipitated by storing overnight at –20°C in two volumes of 100% ethanol and 0.1 volume of 3 M sodium acetate. RNA was centrifuged at 12 000 g for 40 min at 4°C, the pellet was washed twice with 70% ethanol and the RNA was resuspended in RNAse-free water. The final concentration of total RNA was 
5 µg/µl. Quantification and the estimation of purity were derived by measuring the absorbance of each RNA sample at 260 and 280 nm. A reference population of RNA for the gene array experiments was created by mixing RNA from various endometrial samples. All RNA samples were stored at –80°C until use.
Microarray hybridization
Glass microarray slides spotted with approximately 10 500 cDNA sequences were purchased from the Peter MacCallum Cancer Institute (PMCI) microarray facility, Melbourne. Total experimental RNA of 50–70 µg and an equal amount of reference RNA were used for each microarray. The protocol followed for labelling and hybridization of slides was already published by our group (Weston et al., 2002, 2003). Fluorescently labelled cDNAs were created by a direct-labelling procedure.
RNA samples were centrifuged at 15 000 g for 3 min and the volume was made up to 20.8 µl with RNAse-free water, mixed with 2 µl oligo dT (2 µg/µl) and incubated at 65°C for 10 min. Next, 8 µl of 5x first strand buffer, 4 µl 0.1 M dithiothreitol (DTT), 0.8µl 50x low CdNTP and 2 µl of either Cy3-dUTP (cat. no. PA53201; Amersham Bioscience, UK) (reference RNA) or Cy5-dUTP (cat. no. PA55201; Amersham) (experimental RNA) were added and the mixture was incubated at 42°C for 5 min before adding 2 ml of superscript 
.
The reverse transcription step was performed at 42°C for 2.5 h in a PCR machine and RNA was degraded by the addition of 5 µl 0.5 M EDTA and 10 µl 0.25 M NaOH and incubated for 20 min at 65°C. The mixture was neutralized with 25 µl 0.2 M acetic acid. The cDNAs labelled with Cy3 or Cy5 were combined and purified using Qiagen PCR purification columns (Qiagen, Germany), according to the manufacturer's instructions. A combination of blockers [3 µl yeast tRNA 4 mg/ml, 3 µl human cot-1 DNA (cat. no. 15279-011; Gibco) 10 mg/ml, 0.75 µl poly-dA 8 mg/ml, 0.75 µl 50x Denharts herring sperm DNA (cat. no. 14430-029; Gibco), and 2 µl Cy3- and Cy5-labelled luciferase mRNA control] was added to the combined probe to reduce cross-hybridization. The solution was dried in a heated vacuum and the pellet was resuspended in 8 µl of 6.25 x standard saline citrate (SSC) and 8 µl of 100% deionized formamide was added to the solution. The mixture was placed at 100°C for 3 min to denature the cDNA and cooled immediately on ice. Sodium dodecyl sulphate (SDS) (0.2 µl of 10%) was added to the labelled cDNA mixture before it was hybridized to the microarray slide under a cover slip. Hybridization occurred in a humidified chamber at 42°C for 16 h. Hybridized slides were washed at room temperature with 0.5 x SSC and 0.01% SDS for 1 min, 0.5 x SSC for 3 min, 0.06 x SSC for 3 min and then rinsed once with sterilized water. The slides were dried by centrifuging at 1500xg for 5 min.
Scanning and quantification
The microarray slides were scanned using a ScanArray 5000 UV laser scanner (Perkin–Elmer, USA) and the data were extracted using Quantarray software (Perkin–Elmer).
Verification of the clones
cDNA clones of 24 genes (selected because they exhibited significant changes in expression across the cycle, and/or because they were of biological interest to the authors) were purchased from PMCI for sequence verification. Individual colonies were picked and grown in LB medium. Plasmid DNA was isolated using QIAprep® spin kit (cat. No. 27104, Qiagen) and sequenced in order to verify the sequence identity.
Real-time quantitative RT–PCR (RT-QPCR)
Sufficient RNA remained from 32 out of 37 endometrial samples following gene array for subsequent RT-QPCR. RNA from each sample used for PCR was DNase treated using the DNA-free kit (Ambion, USA) according to the manufacturer's protocol to remove any genomic DNA contamination. One microgram of RNA from each sample was used for reverse transcription. One microgram of RNA was mixed with 1.7 µl Random Primers (Invitrogen), 2 µl 10 mM dNTPs (Roche, Australia), 4 µl 5x RT buffer (Roche), 0.5 µl RNAsin (Promega, USA), 2 µl DTT (Promega), 0.2 µl AMV reverse transcriptase (Roche) and sterile water to make up a total volume of 20 µl. The mixture was placed in a PCR machine with a heated lid at 42°C for 1 h.
A Roche Light Cycler and LC fast start DNA master SYBR green kit were used to perform the real-time PCR, according to manufacturer's instructions. The primer sequences used are shown in Table I. Primer concentrations were 0.5 µmol/l. Each set of primers was optimized for annealing temperature and extension times as shown in Table II. Relative mRNA expressions for each of the genes tested were determined by measurement against a specific cDNA standard. 18S rRNA was used as a housekeeping gene to normalize all results.
|
|
Statistical analysis
Data were imported into Gene Spring (Silicon Genetics, USA) version 5.0 and unless otherwise indicated statistical analysis was performed using default settings in this software. Per chip-intensity-dependent normalization (LOWESS) was carried out for each array and a group of 9993 genes was identified that satisfied the criteria of a mean signal intensity of at least 200 in at least 50% of the hybridizations. Parametric analysis of variance (Welsh ANOVA) was performed on log-transformed ratios of the normalized data for these 9993 genes with Benjamini–Hochberg false discovery rate correction at P
0.05, to identify differentially expressed genes across seven different histopathologically defined stages of the menstrual cycle. Hierarchical clustering was performed on samples to group them based on gene expression profile. Hierarchical clustering is an unsupervised clustering method that groups samples according to similarities in their expression patterns (Gene Spring Version 5.0 User manual).
The expression patterns of selected significantly regulated genes were analysed using k-means clustering within the Gene Spring software package. Briefly, k-means clustering was used to divide the genes into a user-defined number of groups (seven in the present study) based on similarities of their expression profiles (Gene Spring Version 5.0 User manual).
The array data were also examined by an algorithm called GeneRaVETM, developed by CSIRO, Mathematical and Information Sciences, Australia. Details of this algorithm have been described elsewhere (Kiiveri, 2003). Briefly, the algorithm uses Bayesian stochastic variable selection to identify differentially expressed genes. The validation strategy of GeneRaVETM uses permutation distributions and cross-validation. This algorithm has been validated on public domain microarray data sets such as the B-cell lymphoma data of Alizadeh et al. (2000) and the prostate cancer data of Luo et al. (2001). In one of its formulations, GeneRaVETM is able to obtain parsimonious sets of solution genes as sample classifiers. For the data of the current paper, GeneRaVETM was first run using every microarray with the full complement of probe genes. The solution genes identified by GeneRaVETM in this step were placed in the initial gene solution set. Next, in a leave-out-one-sample-at-a-time procedure, GeneRaVETM was run using all arrays except for the first array, then using all arrays except the second and so on, finishing with running GeneRaVETM on all arrays except for the last array. At each run, a solution set of genes was obtained. This procedure provided a test of solution stability and had the potential to identify other solution genes. Although repeated applications of GeneRaVETM in this fashion have the potential to generate many predictor genes for group classification, we were only interested in selecting a small set of genes that could define each cycle stage.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Data analysis
One-way parametric ANOVA with Benjamini–Hochberg false discovery rate correction performed on the 9993 genes that passed quality filtering, identified 571 genes as significantly differentially expressed in at least one histopathological stage of the cycle at P
0.05.
Hierarchical clustering of samples and cycle stages
Unsupervised hierarchical clustering of all 43 samples was performed, based on the expression profile of these 571 differentially expressed genes. Results are displayed as a dendrogram in Figure 1A. The length and the subdivision of the dendrogram branches show the relatedness of the samples, with samples joined by short branches being most similar in expression profiles. The 43 samples were sorted into nine clusters based on the similarity of the gene expression profiles of the samples: M–EP (n=2); EP–MP (n=5); MP (n=5); LP–ES (n=5); ES–MS (n=6); MS–LS (n=7); LS (n=3); LS–M (n=2); M (n=4). Samples in these nine groupings agreed with their histopathology by either being in the same group, or an adjacent group, apart from four samples [E355 (LP), EA30 (MS), EA85 (LP), and EA2 (EP)]. These four outliers were removed from further analysis (Figure 1A, B). Three out of the four outliers (355, A30 and A2) found by hierarchical clustering showed significant inter-observer variability between the two histopathology evaluations, while there was agreement between the two pathologists' reports for all other samples. Visual inspection of the fourth outlier's (A85) microarray image showed that it had very poor hybridization.
|
Following the removal of outliers, group M–EP was excluded from further analysis due to lack of replicates, and MS and LS groups were merged to produce the final molecularly defined classification of the menstrual cycle. The final groupings, shown in Figure 1B, were: EP–MP (n=5); MP (n=5); LP–ES (n=5); ES–MS (n=6); MS–LS (n=7); LS–M (n=5); M (n=4).
Parametric ANOVA with Benjamini–Hochberg correction was repeated on 9993 genes for the 37 remaining samples in Figure 1B, and 1452 genes were identified as differentially expressed in at least one stage of the cycle at P0.05. Of these 1452 genes, 425 were chosen for further analysis, as they showed a minimum 2-fold increase in at least one stage of the cycle.
Inconsistent sequence fidelity of spotted cDNA fragments on microarray
Plasmid DNA was isolated from the selected 24 clones (including the clones for the eight genes chosen for validation) and the sequences were verified. Thirteen out of 24 cDNA fragments contained incorrect sequences (data not shown).
Validation
Following sequence verification of clones obtained from the PMCI microarray facility, eight differentially expressed genes [Fyn-binding protein (FYB), P=0.003; transforming growth factor (TGF)
, P=0.0003; endometrial bleeding-associated factor (EBAF), P=0.009; outer dense fibre protein (ODF)2, P=0.0001; natural cytotoxicity triggering receptor (NCR)3, P=0.0002; fucosyl transferase (FUT)4, P=0.0011; annexin (ANX)4, P=0.007; prostaglandin E2 receptor type 4 (EP4), P=0.0007] identified by our analysis and selected for further study were chosen to be validated by RT-QPCR. One-way ANOVA (Kruskal–Wallis) was performed on all RT-QPCR data. For three out of eight genes (TGF, NCR3 and FUT4), RT-QPCR results matched the microarray expression profile, as determined by a significant correlation between microarray data and RT-QPCR data for each sample (TGF, r=0.712, P0.01; NCR3, r=0.631, P0.01; FUT4, r=0.501, P0.05). The data from both microarray and RT-QPCR for these genes are shown in Figure 2. RT-QPCR results for FYB, ANX4 and EBAF mRNA expression did not correlate with the microarray data, but did show significant differential expression during the cycle (FYB, P=0.01; ANX4, P=0.005; EBAF, P=0.001). ODF2 and EP4 mRNAs did not show any differential expression by RT-QPCR (ODF2, P=0.26; EP4, P=0.17). Results are shown in Figure 3.
|
|
K-means clustering
To identify biologically significant patterns of gene expression, the 425 genes with expression that varied a minimum of 2-fold in at least one cycle stage were clustered using k-means clustering. The genes were divided into seven user-defined clusters, with each cluster containing groups of genes with similar temporal expression patterns throughout the menstrual cycle (Figure 4).
|
GeneRaVE
As an alternate bioinformatics approach, the GeneRaVETM algorithm (Weston et al., 2003
) was applied to the 37 samples based on the 425 genes' expression profile. The algorithm identified a set of six genes, the expression profiles of which could be used to correctly define 31 out of 37 samples into their exact molecular cycle stage. Of the remaining six samples, five were misclassified by only one cycle stage. The details of the six genes are given in Table III.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The present study reports the biggest set of microarray data on endometrium to date, and is the first to look at endometrial gene expression across the whole menstrual cycle. We have found a strong relationship between the grouping of endometrial samples based on gene expression profiles, and their histopathological cycle stages. We have also shown that discordant patterns of gene expression may help to identify endometrial samples with subtle abnormalities not readily dissembled by routine histopathology. The study has validated gene array expression profiles with RT-QPCR for three genes: TGF
, FUT4 and NCR3, and the RT-QPCR data have further identified ANX4, FYB and EBAF as having significant changes in endometrial gene expression during the menstrual cycle. Three of the genes investigated (FUT4, NCR3 and FYB) have not been previously shown to have significant cyclic changes in human endometrium. Hierarchical clustering algorithm was applied to 1452 differentially expressed genes to group the endometrial samples based on their transcriptional profiles. The dendrogram generated from hierarchical clustering shows that 37 out of 43 endometrial samples with normal endometrial histology can be separated into seven histopathologically relevant groups based on distinct gene expression patterns (Figure 1). Hence, the study has demonstrated a strong agreement between the grouping of the endometrial samples by gene expression profile and their histopathological classification into menstrual cycle stages.
While many of the samples are grouped together in the same stage of the cycle as classified by histopathology, some are grouped with the adjacent stage of the cycle (Figure 1). For example, sample E353 from the menstrual stage of the cycle has been clustered together with the LS group, while the rest of the menstrual samples have been clustered as one group. Histopathological findings suggested that E353 was from day 28 to 1 of the menstrual cycle, whereas the rest of the menstrual samples were from day 2 to 4. Hence the array data have identified that the gene expression profile of a sample that is from day 1 (the first day of menstrual shedding) is more closely related to day 27–28 (LS) than day 2–4 (M) samples. Thus, the clustering emphasizes the fact that the menstrual cycle is a continuum, and that any classification system that creates discrete groups or stages will inevitably separate some very closely related samples into adjacent, rather than similar groups.
Even though the histopathological reports of two independent pathologists showed significant variability in the cycle stages of the three outliers (E355, EA30 and EA20), they did not detect any apparent abnormalities. However, the gene expression profiles of these samples were significantly different from their histopathological grouping. This result suggests that molecular profiles may be used to identify abnormalities that may not be apparent histologically.
A k-means clustering algorithm was also applied to re-cluster 425 genes based on their expression patterns during the cycle. The genes were grouped into seven different profiles. Each cluster represents a group of genes that peak at a particular cycle stage and have a similar expression profile during the menstrual cycle (Figure 4). The list of genes in the seven clusters is shown in Table IV. Given that endometrial function is primarily driven by circulating estrogen and progesterone, we had hypothesized that major changes in groups of genes might follow peripheral levels of these two hormones. Somewhat surprisingly, there is little evidence of a major change in gene expression that correlates with the rise in estrogen during the proliferative phase of the cycle. By contrast, the rise in cluster 6 mimics the rise in secretory phase circulating progesterone.
|
Our data demonstrate that major changes in gene expression occur around the time of implantation (clusters 1, 4 and 6). Implantation is a complex event involving biological processes such as cell-adhesion, cell growth and differentiation and signal transduction. The above three clusters include previously known, and novel genes associated with the biological processes involved in implantation (Table IV). Adhesion molecules include Trophinin, which has been shown to mediate homophilic cell adhesion between cells of a human trophoblast cell line and an endometrial adenocarcinoma line, and is also expressed by human endometrial epithelium during the early secretory phase (Fukuda et al., 1995
; Suzuki et al., 1998). The implantation associated gene clusters also include several growth factors such as TGF, platelet-derived growth factor alpha polypeptide (PDGFA), fibroblast growth factor 2 (FGF2) and placental growth factor (PlGF). PlGF has been reported as being expressed at higher levels in the secretory than proliferative stage of the cycle (Li et al., 2001; Suzumori et al., 2003). Another major biological event during the cycle, for which it is likely that specific patterns of gene expression exist, is menstruation. Our data show that clusters 5 and 7, which peak leading up to and at the time of menstruation, include apoptosis inducers and regulators, and inflammatory mediators, both of which processes can be associated with menstruation. These clusters also include many genes involved in DNA repair, transcription and signal transduction.
In addition to the genes in each cluster that can be linked to putative functions associated with the relevant cycle stage (e.g. apoptosis at menstruation), it is interesting to note that the majority of genes in each cluster do not readily identify with known functions relevant to each cycle stage. It should also be emphasized that most of the genes have not been validated; hence some will be incorrectly identified.
There have been previous studies that have tried to improve the value of histopathological dating in mid-luteal biopsies by combining it with molecular markers such as 
ß3 integrin expression, to predict pregnancy outcome of infertility patients (Lessey et al., 2000; Creus et al., 2001; Lindhard et al., 2002; Ordi et al., 2002). Given that the anatomical changes which occur in the endometrium throughout the cycle are the results of the interaction of thousands of genes simultaneously, it may be overly simplistic to hope that a single gene or marker will accurately represent endometrial maturation. In the present study, we have identified (using GeneRaVETM algorithm) a small set of predictor genes rather than one or two, whose expression patterns were able to classify 36 out of 37 endometrial samples into the exact and/or adjacent menstrual cycle stage (Table III). Although this concept needs further refining, it may serve as a useful yet easy technique to complement histopathological dating in the future.
Eight genes were selected for validation by RT-QPCR. Real-time PCR results for three of the genes (TGF, NCR3 and FUT4) completely agreed with the microarray results (Figure 2). The RT-QPCR results for FYB, ANX4 and EBAF did not correlate with the microarray data, but did show significant up-regulation during the secretory phases of the cycle (Figure 3). ANX4 mRNA was also shown to be significantly up-regulated during MS phase compared to LP phase and to ES phase by Kao et al. (2002) and Riesewijk et al. (2003), respectively. Byrjalsen et al. (1995) found that ANX4 precursor protein was maximally expressed during the secretory phase compared to proliferative phase of the menstrual cycle. The EBAF mRNA profile obtained by real-time PCR was also in agreement with the existing literature (Tabibzadeh et al., 1998). EP4 and ODF2 mRNA expression profiles by RT-QPCR did not show any cyclic changes and did not agree with the microarray data.
Problems encountered during validation may be due to several different reasons. It has been observed in the past that consistent validation may not be achieved for genes showing less than a 4-fold difference in expression by microarray (Rajeevan et al., 2001; Taniguchi et al., 2001). This might explain the lack of correlation between real-time PCR and array data for the expression of EBAF and EP4. We also found from the clone sequence verification that 13 out of 24 cDNA fragments spotted on the microarray contained incorrect sequences. Other reports in the past have highlighted similar inconsistencies with array data. Halgren et al. (2001) reported that 38% of IMAGE clones purchased from a commercial vendor were contaminated. Even the use of sequence verified clone libraries may not be enough to prevent the introduction of significant errors during the production of a microarray. Taylor et al. (2001) found that by the time the PCR products from a sequence-verified library were ready for printing, only 79% of the clones matched the original vendor database. While some of the errors arose from mistakes in the vendor database itself, the rest were introduced during the preparation of PCR products for printing. Hence even though the eight genes chosen for validation had correct cDNA fragments in the clones, there is still a possibility that cDNAs on the array slide may contain incorrect sequences. Kothapalli et al. (2002) evaluated microarray data obtained from two different commercially available systems. With spotted cDNA microarrays, they purchased 17 probes for sequence verification from genes that were found to be differentially expressed. Four of the plasmids (23.5%) contained incorrect sequences. They tried to verify the results of the 17 genes by Northern blotting and found only eight genes (47%) showed similar results to the microarray data. These results emphasize the importance of validation of any microarray data before proceeding to further investigations.
Aberrant results are inevitable in a large set of microarray data, and can be influenced by cross-hybridization, alternative splicing (Chuaqui et al., 2002) and the fact that hybridization conditions are ‘one-size-fits-all’ for all genes across the array. However, the consistent patterns of agreement in the current study for transcripts across many samples in the array data are statistically unlikely to have arisen by accident. Hence, even though there is the possibility of some of the transcripts being incorrectly labelled, it seems likely that these are still real genes and the concept that the menstrual cycle can be classified based on the transcriptional profile still holds true.
Of the eight genes that were chosen to be validated, we focus on five (NCR3, TGF, FUT4, ANX4 and FYB) that showed significant changes throughout the cycle by real-time PCR.
NCR3/NKp30 is one of the three (NKp30, NKp44, NKp46) existing NCRs involved in natural killer (NK) cell triggering upon recognition of non-HLA ligands. NKp30 and NKp46 are selectively expressed by resting and activated NK cells, while NKp44 is only expressed by activated NK cells (Pende et al., 1999). Uterine natural killer (uNK) cells are present in the endometrium throughout the menstrual cycle and they are particularly abundant during the secretory phase (Pace et al., 1989; King and Loke, 1991). Our data shows that NCR3 mRNA expression mimics the profile of uNK cells in the endometrium (Figure 2).
TGF is a well-studied growth factor and a cytokine. It belongs to the epidermal growth factor (EGF) family and binds to the EGF receptor. EGF family members appear to play an important role in implantation (Glasser et al., 2002, Ch. 14) and may also modulate the effects of estrogen or progesterone or each other by altering receptor expression (Giudice, 1994; Smith, 1998). TGF mediates a variety of cellular processes including proliferation, migration, adhesion and differentiation (Yarden, 2001). TGF mRNA expression was significantly up-regulated from MS–LS stage onwards compared to the early stages of the cycle (Figure 2). These results differ slightly from a previously published study which showed that TGF mRNA and protein expressions were significantly increased during the late proliferative and secretory stages compared to EP stages (Imai et al., 1995).
The FUT4 gene family encode enzymes that transfer fucose in 1, 3/4 linkage on a variety of glycans. Fucosylated glycans linked to proteins are involved in many biological processes such as cell adhesion during development, inflammatory response, leukocyte trafficking and fertilization (Javaud et al., 2003). FUT4 also catalyses the synthesis of the non-sialylated antigens (Lewis x and Lewis y). It has been reported that oligosaccharides Le(x) and Le(y) may play an essential role at the initial stage of implantation (Glasser et al., 2002, Ch.15). These antigens may act as mediator molecules for recognition and adhesion between the surface of the blastocyst and the epithelial cells. Our array data show that the expression of FUT4 mRNA is significantly up-regulated at the time of implantation (Figure 2), suggesting that FUT4 may play an important role during blastocyst implantation.
ANX4 is a member of the lipocortin family of calcium-dependent phospholipid binding proteins thought to be involved in membrane trafficking and membrane organization within the cells (Kaetzel et al., 1989; Gerke and Moss, 2002). FYB also known as adhesion and degranulation-promoting adaptor protein, regulates leukocyte signal transduction and is a modulator of integrin-mediated adhesion of T-cells and mast cells (Geng et al., 2001; Peterson, 2003). ANX4 and FYB mRNA expression are significantly up-regulated during the secretory phase of the menstrual cycle (Figure 3).
There have been four microarray publications in the past 2 years on human endometrium (Kao et al., 2002; Borthwick et al., 2003; Carson et al., 2002; Riesewijk et al., 2003). All four used the affymetrix HG-U95A chip (12 000 transcripts); Borthwick et al. (2003) also used HG-U95 B-E chips (60 000 transcripts). Kao et al. (2002) looked into differential expression between the average values of four late proliferative phase samples (cycle day 8–10) and seven other individual MS phase (LH+8–LH+10) samples. They found 114 up-regulated and 218 down-regulated genes in the MS stage. Borthwick et al. (2003) compared global gene expression profiles of pooled samples of five women in the proliferative phase (day 9–11) with pooled samples of five women in the secretory phase (LH+6–8) and found 89 up-regulated and 57 down-regulated genes in the secretory phase. Both experiments were identical except that Borthwick et al. (2003) pooled before hybridizing onto the microarray. There is about 25–30% agreement between the studies for genes that were up-regulated and 10% agreement for genes that were down-regulated.
Carson et al. (2002) compared three pooled samples from early secretory phase (LH day 2–4) with three pooled samples from MS phase (LH day 7–9) and found 98 up-regulated and 119 down-regulated genes in the MS phase. Riesewijk et al. (2003) compared gene expression between five pairs of ES (LH+2) and MS (LH+7) samples from five individual women. They found 153 up-regulated and 58 down-regulated genes in the MS phase. These two groups have only around 10% agreement between their up-regulated gene set and even less agreement between their down-regulated genes. Finally, even though there were significant similarities in the experimental set-up between the four studies, and they were performed on the same microarray platform, the results only share four genes in common. These are granulysin, secreted phosphoprotein 1, apolipoprotein D and dickkopf 1.
A major difference between the above studies and our own is that they only give information on relative gene expression differences between two fixed time points (ES versus MS or LP versus MS), not across the whole cycle. In a continuous repetitive cycle, relative expression between two discrete time points gives very little information on the functional profile of genes. For example, Riesewijk et al. (2003) identified and validated three genes: glutathione peroxidase 3 (GPx3), claudin 4 and solute carrier family 1 member 1 (SLC1A1) as being significantly up-regulated at LH+7 compared to LH+2 by their microarray data. They also analysed the mRNA expression of these genes by quantitative PCR (Q-PCR) throughout the menstrual cycle. Although Q-PCR data confirmed that GPx3, claudin 4 and SLC1A1 mRNA were significantly up-regulated during MS compared to ES, it also showed that the expression of GPx3 and SLC1A1 mRNA sharply increased further in the LS phase. Hence, GPx3 and SLC1A1 may be more involved in the biological processes leading to menstruation rather than implantation. By dividing the menstrual cycle into several stages to study the transcriptional profile of the whole cycle, far more information becomes available on the putative role that individual genes may have with respect to the different events that occur during the cycle.
In conclusion, our study shows for the first time that it is possible to classify the menstrual cycle according to the transcriptional profile of the endometrial samples, and that this classification agrees with the existing histopathological system of classification.
Further studies need to be undertaken to establish the functions of validated genes as well as to determine whether the predictor genes are functionally relevant to endometrial biology. Our study also highlights some of the problems associated with microarray experiments and especially emphasizes the need for validation studies to be performed on genes of interest before conducting further functional studies.
|
Acknowledgements |
|---|
This study was supported by the Arthur Wilson Medical Scholarship awarded to Dr Gareth Weston by the Royal Australian and New Zealand College of Obstetrics and Gynaecology. We thank staff at PMCI Microarray facility for technical support with the microarray experiments and Dr Yoko Murata (Prince Henry's Institute for Medical Research) for assistance with the real-time PCR. The endometrial tissue samples were collected by research nurses, Nancy Taylor and Nikki Sam. We are grateful to Dr Mark Lawrence, Professor David Healy, Dr Beverley Vollenhoven, Dr Elizabeth Farrell and other gynaecologists at Southern Health for their help in obtaining the tissue samples. PAWR is a National Health and Medical Research Council of Australia (NHMRC) Principal Research Fellow (grant no. 143805). GCW was in receipt of a NHMRC medical postgraduate research scholarship during this study.
|
References |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X et al. (2000) Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511.[CrossRef][Medline]
Arthur ID and Khan SM (1996) The application of standard and morphometric dating criteria for luteal phase endometrial specimens collected by outpatient sampling. Br J Obstet Gynaecol 103, 473–474.[ISI][Medline]
Borthwick JM, Charnock-Jones DS, Tom BD, Hull ML, Teirney R, Phillips SC and Smith SK (2003) Determination of the transcript profile of human endometrium. Mol Hum Reprod 9, 19–33.
Byrjalsen I, Mose Larson P, Fey SJ and Christiansen C (1995) Human endometrial proteins with cyclic changes in the expression during the normal menstrual cycle: characterization by protein sequence analysis. Hum Reprod 10, 2760–2766.
Carson DD, Lagow E, Thathiah A, Al-Shami R, Farach-Carson MC, Vernon M, Yuan L, Fritz MA and Lessey B (2002) Changes in gene expression during the early to mid-luteal (receptive phase) transition in human endometrium detected by high-density microarray screening. Mol Hum Reprod 8, 871–879.
Chuaqui RF, Bonner RF, Best CJ, Gillespie JW, Flaig MJ, Hewitt SM, Phillips JL, Krizman DB, Tangrea MA, Ahram M et al. (2002) Post-analysis follow-up and validation of microarray experiments. Nat Genet 32 (Suppl), 509–514.
Cornet PB, Picquet C, Lemoine P, Osteen KG, Bruner-Tran KL, Tabibzadeh S, Courtoy PJ, Eeckhout Y, Marbaix E and Henriet P (2002) Regulation and function of LEFTY-A/EBAF in the human endometrium. mRNA expression during the menstrual cycle, control by progesterone, and effect on matrix metalloprotineases. J Biol Chem 277, 42496–42504.
Corsan GH, Kemmann E, Bohrer M, Blotner M, Smilow P and Shelden R (1992) Use of urinary luteinizing hormone immunoassays in the assessment of luteal function in infertile women. Am J Obstet Gynecol 166, 41–46.[ISI][Medline]
Creus M, Ordi J, Fabregues F, Casamitjana R, Vanrell JA and Balasch J (2001) Mid-luteal serum inhibin-A concentration as a marker of endometrial differentiation. Hum Reprod 16, 1347–1352.
Fukuda MN, Sato T, Nakayama J, Klier G, Mikami M, Aoki D and Nozawa S (1995) Trophinin and tastin, a novel cell adhesion molecule complex with potential involvement in embryo implantation. Genes Dev 9, 1199–1210.
Geng L, Pfister S, Kraeft SK and Rudd CE (2001) Adaptor FYB (Fyn-binding protein) regulates integrin-mediated adhesion and mediator release: differential involvement of the FYB SH3 domain. Proc Natl Acad Sci USA 98, 11527–11532.
Gerke V and Moss SE (2002) Annexins: from structure to function. Physiol Rev 82, 331–371.
Giudice LC (1994) Growth factors and growth modulators in human uterine endometrium: their potential relevance to reproductive medicine. Fertil Steril 61, 1–17.[ISI][Medline]
Glasser SR, Aplin JD, Giudice LC and Tabibzadeh S (2002) The endometrium. St Edmundsbury Press, Suffolk, UK.
Halgren RG, Fielden MR, Fong CJ and Zacharewski TR (2001) Assessment of clone identity and sequence fidelity for 1189 IMAGE cDNA clones. Nucleic Acids Res 29, 582–588.
Imai T, Kurachi H, Adachi K, Adachi H, Yoshimoto Y, Homma H, Tadokoro C, Takeda S, Yamaguchi M, Sakata M et al. (1995) Changes in epidermal growth factor receptor and the levels of its ligands during menstrual cycle in human endometrium. Biol Reprod 52, 928–938.[Abstract]
Javaud C, Dupuy F, Maftah A, Julien R and Petit JM (2003) The fucosyltransferase gene family: an amazing summary of the underlying mechanisms of gene evolution. Genetica 118, 157–170.[CrossRef][ISI][Medline]
Johannisson E, Parker RA, Landgren BM and Diczfalusy E (1982) Morphometric analysis of the human endometrium in relation to peripheral hormone levels. Fertil Steril 38, 564–571.[ISI][Medline]
Kaetzel MA, Hazarika P and Dedman JR (1989) Differential tissue expression of three 35-kDa annexin calcium-dependent phospholipid-binding proteins. J Biol Chem 264, 14463–14470.
Kao LC, Tulac S, Lobo S, Imani B, Yang JP, Germeyer A, Osteen K, Taylor RN, Lessey BA and Giudice LC (2002) Global gene profiling in human endometrium during the window of implantation. Endocrinology 143, 2119–2138.
Kiiveri H (2003) A Bayesian approach to variable selection when the number of variables is very large. In: Science and Statistics: A Festschrift for Terry Speed, Goldstein TR (ed.), IMS Lecture Notes—Monograph Series, Volume 40, Institute of Mathematical Statistics, Beachwood, OH, pp. 127–143.
Kim-Bjorklund T, Landgren BM, Hamberger L and Johannisson E (1991) Comparative morphometric study of the endometrium, the fallopian tube, and the corpus luteum during the postovulatory phase in normally menstruating women. Fertil Steril 56, 842–850.[ISI][Medline]
King A and Loke YW (1991) On the nature and function of human uterine granular lymphocytes. Immunol Today 12, 432–435.[CrossRef][ISI][Medline]
Kothapalli R, Yoder SJ, Mane S and Loughran TP, Jr (2002) Microarray results: how accurate are they? BMC Bioinformatics 3, 22.[CrossRef][Medline]
Lenton EA, Landgren BM and Sexton L (1984) Normal variation in the length of the luteal phase of the menstrual cycle: identification of the short luteal phase. Br J Obstet Gynaecol 91, 685–689.[ISI][Medline]
Lessey BA, Castelbaum AJ, Wolf L, Greene W, Paulson M, Meyer WR and Fritz MA (2000) Use of integrins to date the endometrium. Fertil Steril 73, 779–787.[CrossRef][ISI][Medline]
Li XF, Charnock-Jones DS, Zhang E, Hiby S, Malik S, Day K, Licence D, Bowen JM, Gardner L, King A et al. (2001) Angiogenic growth factor messenger ribonucleic acids in uterine natural killer cells. J Clin Endocrinol Metab 86, 1823–1834.
Lindhard A, Bentin-Ley U, Ravn V, Islin H, Hviid T, Rex S, Bangsboll S and Sorensen S (2002) Biochemical evaluation of endometrial function at the time of implantation. Fertil Steril 78, 221–233.[CrossRef][ISI][Medline]
Liotta L and Petricoin E (2000) Molecular profiling of human cancer. Nat Rev Genet 1, 48–56.[CrossRef][ISI][Medline]
Luo J, Duggan DJ, Chen Y, Sauvageot J, Ewing CM, Bittner M, Trent JM and Isaacs WB (2001) Human prostate cancer and benign prostatic hyperplasia: molecular dissection by gene expression profiling. Cancer Res 61, 4683–4688.
Martinez AR, Bernardus RE, Vermeiden JP and Schoemaker J (1992) Reliability of home urinary LH tests for timing of insemination: a consumer's study. Hum Reprod 7, 751–753.
Munster K, Schmidt L and Helm P (1992) Length and variation in the menstrual cycle—a cross-sectional study from a Danish county. Br J Obstet Gynaecol 99, 422–429.[ISI][Medline]
Murray MJ, Meyer WR, Zaino RJ, Lessey BA, Novotny DB, Ireland K, Zeng D and Fritz MA (2004) A critical analysis of the accuracy, reproducibility, and clinical utility of histologic endometrial dating in fertile women. Fertil Steril 81, 1333–1343.[CrossRef][ISI][Medline]
Noyes RW, Hertig AW and Rock J (1950) Dating the endometrial biopsy. Fertil Steril 1, 3–25.
Ordi J, Creus M, Ferrer B, Fabregues F, Carmona F, Casamitjana R, Vanrell JA and Balasch J (2002) Midluteal endometrial biopsy and alphavbeta3 integrin expression in the evaluation of the endometrium in infertility: implications for fecundity. Int J Gynecol Pathol 21, 231–238.[ISI][Medline]
Pace D, Morrison L and Bulmer JN (1989) Proliferative activity in endometrial stromal granulocytes throughout menstrual cycle and early pregnancy. J Clin Pathol 42, 35–39.
Pende D, Parolini S, Pessino A, Sivori S, Augugliaro R, Morelli L, Marcenaro E, Accame L, Malaspina A, Biassoni R et al. (1999) Identification and molecular characterization of NKp30, a novel triggering receptor involved in natural cytotoxicity mediated by human natural killer cells. J Exp Med 190, 1505–1516.
Peterson EJ (2003) The TCR ADAPts to integrin-mediated cell adhesion. Immunol Rev 192, 113–121.[CrossRef][ISI][Medline]
Rajeevan MS, Vernon SD, Taysavang N and Unger ER (2001) Validation of array-based gene expression profiles by real-time (kinetic) RT-PCR. J Mol Diagn 3, 26–31.
Riesewijk A, Martin J, van Os R, Horcajadas JA, Polman J, Pellicer A, Mosselman S and Simon C (2003) Gene expression profiling of human endometrial receptivity on days LH+2 versus LH+7 by microarray technology. Mol Hum Reprod 9, 253–264.
Santoro N, Goldsmith LT, Heller D, Illsley N, McGovern P, Molina C, Peters S, Skurnick JH, Forst C and Weiss G (2000) Luteal progesterone relates to histological endometrial maturation in fertile women. J Clin Endocrinol Metab 85, 4207–4211.
Smith CL (1998) Cross-talk between peptide growth factor and oestrogen receptor signalling pathways. Biol Reprod 58, 627–632.
Sterzik K, Abt M, Grab D, Schneider V and Strehler E (2000) Predicting the histologic dating of an endometrial biopsy specimen with the use of Doppler ultrasonography and hormone measurements in patients undergoing spontaneous ovulatory cycles. Fertil Steril 73, 94–98.[CrossRef][ISI][Medline]
Suzuki N, Zara J, Sato T, Ong E, Bakhiet N, Oshima RG, Watson KL and Fukuda MN (1998) A cytoplasmic protein, bystin, interacts with trophinin, tastin, and cytokeratin and may be involved in trophinin-mediated cell adhesion between trophoblast and endometrial epithelial cells. Proc Natl Acad Sci USA 95, 5027–5032.
Suzumori N, Sugiura-Ogasawara M, Katano K and Suzumori K (2003) Women with endometriosis have increased levels of placental growth factor in the peritoneal fluid compared with women with cystadenomas. Hum Reprod 18, 2595–2598.
Tabibzadeh S, Lessey B and Satyaswaroop PG (1998) Temporal and site-specific expression of transforming growth factor-beta4 in human endometrium. Mol Hum Reprod 4, 595–602.
Taniguchi A, Suga R and Matsumoto K (2000) Expression and transcriptional regulation of the human alpha 1, 3-fucosyl-transferase (FUT4) gene in myeloid and colon adenocarcinoma cell lines. Biochem Biophys Res Commun 273, 370–376.[CrossRef][ISI][Medline]
Taylor E, Cogdell D, Coombes K, Hu L, Ramdas L, Tabor A, Hamilton S and Zhang W (2001) Sequence verification as quality-control step for production of cDNA microarrays. Biotechniques 31, 62–65.[ISI][Medline]
Treloar AE, Boynton RE, Behn BG and Brown BW (1967) Variation of the human menstrual cycle through reproductive life. Int J Fertil 12, 77–126.[ISI][Medline]
Weston GC, Haviv I and Rogers PA (2002) Microarray analysis of VEGF-responsive genes in myometrial endothelial cells. Mol Hum Reprod 8, 855–863.
Weston G, Trajstman AC, Gargett CE, Manuelpillai U, Vollenhoven BJ and Rogers PA (2003) Fibroids display an anti-angiogenic gene expression profile when compared with adjacent myometrium. Mol Hum Reprod 9, 541–549.
Yarden Y (2001) The EGFR family and its ligands in human cancer. Signalling mechanisms and therapeutic opportunities. Eur J Cancer 37 (Suppl 4), S3–S8.
Submitted on August 19, 2004; resubmitted on September 23, 2004; accepted on October 3, 2004.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  A. P Ponnampalam and P. A W Rogers Expression and regulation of fucosyltransferase 4 in human endometrium Reproduction, July 1, 2008; 136(1): 117 - 123. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Dassen, R. Kamps, C. Punyadeera, F. Dijcks, A. de Goeij, A. Ederveen, G. Dunselman, and P. Groothuis Haemoglobin expression in human endometrium Hum. Reprod., March 1, 2008; 23(3): 635 - 641. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Mitko, S. E Ulbrich, H. Wenigerkind, F. Sinowatz, H. Blum, E. Wolf, and S. Bauersachs Dynamic changes in messenger RNA profiles of bovine endometrium during the oestrous cycle Reproduction, February 1, 2008; 135(2): 225 - 240. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. Hanifi-Moghaddam, B. Boers-Sijmons, A. H.A. Klaassens, F. H. van Wijk, W. F. Van Ijcken, P. Van der Spek, H. A.M. Verheul, H. J. Kloosterboer, C. W. Burger, and L. J. Blok Difference in signalling between various hormone therapies in endometrium, myometrium and upper part of the vagina Hum. Reprod., February 1, 2008; 23(2): 298 - 305. [Abstract] [Full Text] [PDF]
|
|