| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Deparments of Pediatrics (Y.W., P.J., X.W., S.G., S.-W.G.), Obstetrics and Gynecology (E.S., G.H.), and Pathology (Z.B.), Medical College of Wisconsin, Milwaukee, Wisconsin 53226; Department of Pathology (A.K.-B.), University of Illinois, Chicago, Illinois 60612; and Department of Statistics and Applied Probability (Y.W.), University of California, Santa Barbara, California 93106
Address all correspondence and requests for reprints to: Sun-Wei Guo, Ph.D., Department of Pediatrics, Medical College of Wisconsin, 8701 Watertown Plank Road, MS 756, Milwaukee, Wisconsin 53226-0509. E-mail: swguo{at}mcw.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
As with the eutopic endometrial tissue, its ectopic counterpart responds to cyclic changes in steroid hormones by proliferation differentiation and by the production of autocrine and paracrine factors. As reported by numerous studies, however, ectopic endometrium appears to behave quite differently from its eutopic counterpart in many other ways (5, 6). Therefore, the characterization of the differences and similarities between the eutopic and ectopic endometrium is arguably a first important step toward the understanding of the pathogenesis of endometriosis. It also would serve the dual purposes of better defining the disorder through the comparison and contrast between eutopic and ectopic endometrium, and of finding better cell markers for the disorder. Despite numerous documentations of the differences between eutopic and ectopic endometrium, most, if not all, studies have focused on a single or few proteins/genes/molecules, and the characterization appears to be fragmentary. In fact, although overwhelming evidence points to various differences between endometriotic lesions and endometrium, the molecular definition of these changes has been difficult to characterize.
The cDNA microarray technology (7) provides a powerful tool for quantifying expression levels of thousands of genes simultaneously. With this approach, we can compare gene expression patterns between eutopic and ectopic endometrium from the same patient or identify genes differentially expressed (DE) in endometrium between samples with and without endometriosis. In the last 3 yr, several gene expression studies on endometriosis have been published. In a pilot study, Eyster et al. (8) compared gene expression levels between the eutopic and ectopic endometrium from three patients with endometriosis using cDNA microarrays and identified eight genes to be up-regulated in endometriotic implants. Lebovic et al. (9) profiled eight patients and four controls and reported Tob-1, a cell-cycle inhibitor gene, to be differentially responsive to IL-1ß stimulation in endometriotic stromal cells compared with their eutopic counterparts. More recently, Arimoto et al. (10) used spotted cDNA microarrays and analyzed expression profiles of ovarian endometrial cysts from 23 patients. Using fixed fold-changes of 0.5 or 2.0, they identified 15 genes that were commonly up-regulated in the ovarian cysts during both proliferative and secretory phases, and 42 and 40 that were up-regulated only in the proliferative and secretory phases, respectively. In addition, they identified genes that were down-regulated in the proliferative, secretory, and both phases. Kao et al. (11), using an oligonucleotide chip, identified 91 (115) genes that were up-regulated (down-regulated) in the endometrium from seven women with endometriosis compared with eight normal controls. Similarly, Konno et al. (12) identified genes involved in immunoreactions in endometriotic lesions. Using cDNA chips printed on a nylon membrane, Matsuzaki et al. (13) identified several possible pathways that are involved in the pathogenesis of deep endometriosis using cDNA microarrays.
All these studies have provided much needed insight into the transcriptional changes related to endometriosis, and the field of endometriosis research is now poised to further characterize and delineate some specific pathways involved in endometriosis as identified in these studies. However, there is still ample room for improvement in terms of methodological refinement. Table 1
lists the characteristics of six previously published gene expression studies of endometriosis and this study.
|
We conducted the present study to further profile transcriptional differences between the ectopic and eutopic endometrium, overcoming some deficiencies in previous studies. Once we identified genes that are DE, we attempted to identify biological themes and pathways based on these genes. Furthermore, we examined their utility in classifying endometriosis.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
Tissue processing and staining
Five-micrometer frozen tissue sections were mounted on uncharged, noncoated slides, immediately fixed in 75% ethanol for 30 sec, and stained immediately using Histogene LCM Frozen Section Staining Kit (Arcturus Engineering). Briefly, the slides were rinsed in distilled water for 30 sec, then stained as follows: stained in 100 µl Histogene Staining solution for 20 sec, rinsed with distilled water for 30 sec, then fixed in 75% and 95% ethanol for 30 sec each, and finally dehydrated in 100% ethanol for 30 sec followed by incubation in xylene for at least 5–10 min. The slides were then air-dried in the hood for 5 min and stored in a dissector for no more than 2 h before laser capture microdissection.
Laser capture microdissection (LCM)
LCM was performed with a Pixcell II laser capture microscope (Arcturus Engineering). Epithelial cells were captured on thermoplastic caps (Arcturus Engineering) by using a 7.5-µm diameter laser spot and 50 mW laser power. The average number of cells captured on each cap from an individual endometriotic lesion and from normal endometrium was approximately 250 and 1000, respectively. After microdissection, the caps were placed on 0.5-µl microfuge tubes containing 200 µl denaturing buffer and 1.6 µl ß-mercaptoethanol for RNA extraction.
Array printing
The Research Genetics (Huntsville, AL) sequence-verified human library consisting of 41,472 clones was used as a source of probe DNA. Amplification of clone insertions and array fabrications were described in Ref.18 .
RNA extraction
Total RNA was extracted from LCM-harvested cells with the Micro RNA Isolation Kit (Stratagene, San Diego, CA). Samples were incubated with 200 µl denaturing buffer plus 1.6 µl ß-mercaptoethanol at room temperature for 25 min; then the cell lysates were mixed with 20 µl of 2 M sodium acetate (pH 4.0), 220 µl of water saturated phenol, and 60 µl of 24:1 chloroform:isoamyl alcohol and incubated on ice for 15 min. After centrifugation at 12,000 x g at 4 C for 30 min, the aqueous phase was collected and mixed with 1 µl of 10 mg/ml glycogen and 200 µl cold isopropanol. After overnight incubation at –20 C, the RNA pellet was precipitated and washed twice with 75% DEPC water-treated ethanol. After air-drying, the RNA pellet was resuspended in 10 µl nuclease-free water and ready to use for RT PCR and RNA amplification.
Antisense RNA (aRNA) synthesis (RNA amplification)
aRNA was synthesized using RiboAmp RNA amplification kit (Arcturus Engineering). Two rounds of linear amplifications were performed according to the manufacturer’s protocol. The aRNA was quantified by DU-64 UV/Vis Spectrophotometer (Beckman Coulter, Inc., Fullerton, CA), and the quality of the aRNA was routinely checked on 1% agarose gels.
Array hybridization
Modified labeling and hybridization protocols, as described previously (19), were used. Briefly, 3.0 µg of aRNA was labeled by RT in a total volume of 20.0 µl, including 4.0 µl first-strand buffer 1.0 µl of 8.0 µg/µl random hexamer, 2.0 µl of 10x lowT-dNTP, 2.0 µl of 0.1 M DTT, 1.0 µl RNAsin, and 2.0 µl Cy3-dUTP or Cy5-dUTP (Amersham Pharmacia, Piscataway, NJ). The reaction mixtures were preheated at 65 C for 5 min, then 2.0 µl of 200 U/µl Superscript II (Invitrogen, Carlsbad, CA) were added to each mixture and incubated at 42 C for 1 h. The reactions were terminated by adding 2.5 µl of 0.5 mM EDTA at 65 C for 1 min, 5.0 µl of 1 M NaOH at 65 C for 15 min, followed by neutralizing with 12.5 µl of 1 M Tris-HCl (pH 7.4). Cy3- and Cy5-labeled cDNA targets were purified by Bio-6 Chromatograph column (Bio-Rad, Cambridge, MA). After purification, Cy3- and Cy5-labeled cDNA was combined and savant dried to 8.0 µl. The hybridization solution was adjusted to approximately 16.0 µl for 22 x 22 cover slips by adding 1 µl of 50x Denhardt’s blocking solution (Sigma, St. Louis, MO), 1.0 µl of 8.0 mg/ml poly dA (Amersham Pharmacia), 1.0 µl of 4 mg/ml yeast tRNA (Invitrogen Life Technologies, Carlsbad, CA), 1.0 µl of 10.0 mg/ml human cot I DNA (Invitrogen Life Technologies), and 2.6 µl 20x SSC to the 8.0 µl of labeled cDNA mixture. The hybridization solution was heated at 99 C for 2 min and cooled to room temperature. Then 0.6 µl of 10% SDS was added to the hybridization solution before applying it onto the array chip. Hybridizations were performed at 65 C for 18 h in a humidity chamber. After hybridization, the slides were washed at room temperature in 2x SSC, 0.1% SDS for 1 min, then 1x SSC for 1.5 min, 0.2x SSC for 1.5 min, and 0.05x SSC for 30 sec. The slides were dried immediately at 500 x g for 5 min.
Real-time RT-PCR for microarray validation
Numerous gene expression profiling studies using cDNA microarray with validations using either real-time RT-PCR or Northern blotting have shown that results from microarrays are reliable (20, 21, 22). Therefore, we randomly selected eight genes from 904 (identified to be DE) for validation using samples from patients 4 and 9 with real-time RT-PCR. These eight genes were: TNFAIP1 (TNF), KIAA0095, DOC-1R (tumor suppressor deleted in oral cancer-related 1), INDO (indoleamine-pyrrole 2, 3 dioxygenase), GGTLA1 (
-glutamyltransferase-like activity 1), HSPA1A (heat shock 70 kDa protein 1), CHL1 (cell adhesion molecule), and APPBP2 (amyloid ß precursor protein-binding protein 2). Total RNA (10 ng) was treated with DNase I to remove potential DNA contamination and then reverse-transcribed using Superscript II Reverse Transcriptase (Invitrogen). Real-time polymerase chain reactions were carried out on a Smart Cycler System (Cepheid, Sunnyvale, CA), and monitored by SYBR Green I (Qiagen, Valencia, CA). The PCR products of the expected size were also visualized on a 0.8% agarose gel. The relative mRNA level of each gene was calculated using Relative Quantitation of Gene Expression (Applied Biosystems, Foster City, CA) with 18S mRNA as an endogenous control. The primer sequences for RT-PCR and their product sizes are listed in Table 2. All primers were designed to span two exon boundaries, thus restricting PCR amplifications to cDNA templates only. The universal 18s internal standard was purchased from Ambion (Austin, TX).
|
For all 4 x 12 = 48 slides, a total of 4916 spots with quality scores less than 0.2 were removed from the analysis. All data were normalized using a procedure as described by Yang et al. (24).
Statistical analysis
Because the phase of the menstrual cycle and the location of the endometriotic lesion may influence gene expression levels, the two variables were recorded and incorporated in the data analysis. To do this, linear mixed effects models were constructed to identify genes that are DE in lesions as compared with endometrium. In choosing the model, we considered six factors: location of lesion (ovarian or nonovarian), menstrual phase (proliferative or secretory), subject, array, dye, and tissue (ectopic or eutopic endometrium). The subject factor is nested within the combination of the phase and location factors, and the array factor is nested within the subject factor. Because we used the Latin-square design for each patient, there was some confounding, and we considered lower-order, effects only (25). We treated both subject and array as random factors. Therefore, for each gene, we considered the following linear mixed-effects model
 |
ijklmn are random errors. We assumed that random effects and random errors follow normal distributions and are mutually independent. This model accounts for the intricate nested treatment structure and Latin-square design structure of the experiment. In addition, the model detects the interaction between the tissue, phase, and lesion location, and once such interaction is detected, the contrasts between the two tissues are evaluated at different levels of phase and location for statistical significance. Specifically, we looked at the three-way interaction LPT first. If it is significant at the P = 0.01 level, we looked at the tissue contrasts for each combination of the location and phase factor to identify genes that were significant at the P = 0.01 level. If the three-way interaction was not significant, we examined the two-way interactions LT and PT. If one or both of them were significant at the P = 0.01 level, we examined the tissue contrasts for each level of the location, or phase, or both, depending on which one or both are significant. Again, we identified genes that are significant at P = 0.01 level. If none of the interactions LPT, LT, and PT is significant, then we examined the tissue effect directly and identified genes that are significant at the P = 0.01 level. The inclusion of menstrual phase and the location of the lesion effectively controlled the effects of both variables in identifying genes DE in ectopic and eutopic endometrium. Once DE genes were identified, a multidimensional scaling (MDS) analysis was performed.
We used SAS procedure PROC MIXED to fit linear mixed effects models and construct tissue contrasts (26). Other computations were carried out in R (version 2.0.1, http://www.r-project.org).
Hierarchical cluster analysis was carried out using Cluster 3.0. Pearson’s correlation coefficient (uncentered) was used as the similarity metric and the average linkage as the clustering method. The resulting dendrogram was viewed using MapleTree.
Identification of biological themes and pathways from the list of DE genes
Once DE genes are identified, we used Expression Analysis Systematic Explorer (EASE) (27) to annotate gene functions and identify biological themes. From a given list of genes, EASE identifies sets of genes (typically with known functions) that are overrepresented and converts it into an ordered table of robust biological themes that summarize the biological result of the experiment. Because our inclusion of genes/ESTs into our cDNA microarray did not have any present rules and thus was unbiased, a score can be attached to any biological pathway that designates the overrepresentation of the genes identified to be DE.
With the list of DE genes with known symbols and functions, we also searched Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://www.genome.jp/kegg/) to identify biological pathways in which these genes are involved. KEGG is a knowledge base for systematic analysis of gene functions, linking genomic information with higher order functional information (28). The identified pathways provide us with a better perspective on the roles of those genes identified to be DE, instead of an isolated and fragmented view of individual genes.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
. Almost all of these patients have revised American Fertility Society (29) stage III/IV endometriosis.
|
The total epithelial cells captured using LCM and the aRNA yields from linear amplification for select patients are shown in Table 4. The result shows that similar input of LCM captured cells for linear amplification produced little variability of aRNA yields, suggesting that the quality and quantity of input total RNA were fairly uniform and adequate for our purposes.
|
The statistical analysis of the filtered and normalized microarray data involved close to a half-million data points (12 patients x 4 slides x 2 colors x 4,684 spots = 449,664 data points). Using the linear mixed model mentioned above, we found that the dye effect is significant and we identified 388 genes/ESTs that were DE (30) between the eutopic and ectopic endometrium, regardless of the phase of the menstrual cycle or the location of the ectopic endometrium (P 
0.01). In addition, 21 and 25 genes/ESTs were DE, respectively, depending on the phase of the menstrual cycle (proliferative vs. follicular) but not on the location of the lesion; 265 and 169 genes/ESTs were DE, respectively, depending on the location of the lesion analyzed (ovarian vs. nonovarian) but not on the menstrual phase. Moreover, five, seven, six, and eight genes/ESTs were DE, respectively, depending on the joint status of menstrual phase and the location of the lesion without the three-way interaction among menstrual phase, location of the lesion, and the tissue (eutopic vs. ectopic endometrium). Finally, 16, 26, 23, and 27 genes/ESTs were DE, respectively, depending on the joint status of menstrual phase and the location of the lesion in the presence of the three-way interaction among menstrual phase, location of the lesion, and the tissue. Removing some overlap among the above categories, 904 genes/ESTs were identified to be DE between the eutopic an ectopic endometrium, representing approximately one in every five genes/ESTs interrogated.
MDS analysis based on expression levels of the identified 905 DE genes/ESTs, using Euclidean distance, revealed that the 48 slides can be distinctly divided into roughly two groups: ovarian and nonovarian endometriosis (Fig. 1). This suggests that the location of the lesion from which the ectopic endometrial samples were taken is an important factor that affects the magnitude of differential gene expression levels between the eutopic and ectopic endometrium. From Fig. 1, it is also evident that there are considerable slide-to-slide variations within the same subject, but this variation is much smaller than the person-to-person variations.
|
). As shown in the figure, the 12 patients were clearly clustered by the locations of endometriotic lesions, ovarian and nonovarian. Accordingly, the left six leaves of the array tree are all ovarian (Pt. 1, 3, 5, 4, 8, and 9, in that order), whereas the right six leaves are all nonovarian (Pt. 7, 6, 2, 10, 11, and 12). Visual inspection of the genes/ESTs cluster tree revealed five distinct clusters (Fig. 2), depending on the different magnitudes of the expression ratios of ectopic vs. eutopic endometrium between the ovarian and nonovarian endometriosis groups. The first cluster of genes/ESTs are those that are down-regulated in the ovarian group but moderately up-regulated in the nonovarian group (hereafter, we will use up- and down-regulation to mean that the gene expression is increased or reduced in the ectopic endometrium as compared with the eutopic endometrium. Strictly speaking, the up- or down-regulation does not necessarily mean the absolute gene expression in the ectopic endometrium has been increased or reduced, because we have only measured the ratios of the expression levels in the ectopic vs. the eutopic endometrium. It is possible that both endometria have increased (or reduced) expression yet the ratio still could be less or greater than 1 because the magnitudes of increase or reduction may be different in the ectopic and eutopic endometria). The second cluster is the reversal of the first cluster. The third cluster contains genes that are lower expression in both ovarian and nonovarian groups. The fourth cluster contains genes that are mostly lower expression in the ovarian group but are substantially higher expression in the nonovarian group. The last cluster contains genes that are mostly higher expression in both groups. Some important genes with known functions in each cluster are listed in Table 5.
|
|
.
|
KEGG pathway search
Among 904 identified DE genes/ESTs, 390 of them are genes with known symbols and functions. With these 390 genes, we carried out a pathway search using the KEGG database and identified 79 pathways from 113 of 390 genes. Among the 79 identified pathways, 39 of them were each identified with one gene from the list of 113 genes, and 15 of them were identified based on at least four genes from the list (Table 7). Note that some genes were involved in more than one pathway. One of the identified pathways is shown in Fig. 3.
|
|
Validation of gene expression
We performed real-time RT-PCR experiments to confirm the differential expressions identified by microarray analysis. In the labeling experiment, cDNA from endometriotic lesion was labeled with Cy5 and cDNA from corresponding endometrium was labeled with Cy3 (dye-switched for the cross-labeling experiment). A red spot or green spot indicates the gene has higher or lower expression level in the lesion than that in the endometrium. APPBP2, GGTLA1, and CHL1 had higher expression in ectopic endometrium as compared with its eutopic counterpart, whereas the expression of INDO, TNFAIP1, DOC-1R, HSPA1A, and KIAA0095 were lower.
The results of real-time RT-PCR were highly correlated with that of microarray results (r = 0.87, P = 0.005) and are shown in Fig. 4. Each fold change represents the relative expression ratio of lesion to endometrium. Note that the magnitude of fold changes obtained from microarrays appeared to be "compressed" as compared with those obtained from real-time RT-PCR. This is consistent with what other investigators have observed (20, 22).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
Six studies on gene expression profiling using microarrays have been published in endometriosis (see Table 1
). Our study differs from these studies in several important ways. First, we have used LCM to harvest the desired cell type. LCM has been proven to be a tissue microdissection procedure that allows accurate, even single-cell, tissue sampling from small target tissues such as endometriotic lesions (31, 32). The purity of epithelial cell purification was ascertained by histological review of laser capture microdissected cells as previously described (33). The use of LCM ensures us an accurate and reliable acquisition of cells of the desired type from specific microscopic regions of tissue sections under direct visualization, which in turn, permits molecular genetic analysis of pure populations of epithelial cells taken from lesion samples. It greatly minimizes or even eliminates any possibility of contamination. Because epithelial cells and stromal cells often have different expression patterns (34, 35), the use of LCM adds more specificity to our findings. In all six published studies, only one recent study (13) also uses LCM.
For this study, we chose to focus on epithelial cells only, using LCM to avoid stromal and host tissue cells in our preparations. Stromal cells are much more difficult to separate from the underlying host tissues and inflammatory cells even with current technology. However, there is strong evidence in the literature to support the importance of stromal cells and host tissue microenvironment (35). It is likely that some, if not many, of the differences we found between ovarian and nonovarian endometriosis could be ascribed to hormonal microenvironment differences—the "soil" vs."seed" analysis of this complex problem.
Second, we carried out a rigorous statistical analysis of the array data, taking into account the location of the lesion and menstrual phases. It has been known that using the arbitrarily set, fixed fold change thresholds for declaring whether a gene is DE or not has little statistical validity (36).
Third, we used dye swaps to protect against the possibility of a well-documented phenomenon called labeling effect, referring to the difference in hybridization efficiency between the two dyes (37). Dye-swapping is a highly recommended procedure for cDNA microarrays (38).
Fourth, we used replication arrays to increase the reliability and precision of the estimated fold changes, because it is well documented that any single microarray output is subject to substantial variability (21, 39), and is likely to confound individual-to-individual variation with array-to-array variation. As we can see from Fig. 1, both individual-to-individual variation and array-to-array variation clearly exist.
Fifth, we have used stringent criteria to sift out imperfect spots in microarrays and gone through careful data normalization procedures, despite the fact that each of our printed cDNA array passed quality assurance using the third dye (40). Because all hybridized microarrays, including commercial oligonucleotide chips, contain imperfect spots (41, 42), exclusion of low-quality spots ensures data quality.
Lastly, after identifying DE genes, we carried out extensive database queries to identify biological themes and pathways that are involved in endometriosis pathogenesis. This step is important because this provides an objective way to annotate these genes and to identify major themes and pathways that may be embedded in hundreds of DE genes. Although a manual annotation could be performed, there is a risk of relying too heavily on often subjective knowledge and of missing important themes or pathways. The identification of these pathways not only provides new insight onto the molecular mechanisms underlying the disease but also offers the opportunity to identify up- and down-stream genes of the genes identified to be DE, that are also involved in the disease process. This process is important for generating new hypotheses.
Agreement with previously published studies
Our study has confirmed some previously reported findings. For example, Chegini et al. (43) reported that ectopic endometrium of women with endometriosis express IL-15 mRNA and protein with elevated levels compared with eutopic and control endometrium, irrespective of the phases of the menstrual cycle. Our study found that IL-15 expression is higher in ectopic endometrium of women with nonovarian endometriosis, but the reverse is true in the same tissue of women with ovarian endometriosis. We also found that the expression of IL-15 receptor 
(IL15RA) is reduced in the lesions. Kao et al. (11) reported down-regulation of IL-15 in the eutopic endometrium of women with endometriosis.
Several studies have reported an increase of pro-inflammatory chemoattractant cytokines such as IL-8 in the peritoneal fluid from patients with endometriosis (44, 45). IL-8 is known to facilitate expression of surface adhesion molecules on neutrophils, angiogenesis, and mitogenesis of epidermal and vascular smooth muscle cells and also induce the proliferation of endometrial stromal cells acting as autocrine growth factor to the endometrium (46, 47, 48). It has been shown recently that the expression of IL-8 in ectopic endometrium is higher than in the eutopic endometrium in women with endometriosis (12, 47). Our study confirmed this finding.
Pathways likely to be involved in the pathogenesis of endometriosis
We found higher expression levels of platelet-derived growth factor receptor (PDGFRA) and platelet-derived growth factor ß in lesions. In addition, a closely related gene, KIT, also has increased expression only in ovarian endometriosis patients, but decreased expression in nonovarian endometriosis patients. Platelet-derived growth factor receptors (PDGFRs) and their ligands, platelet-derived growth factors (PDGFs), play critical roles in mesenchymal cell migration and proliferation. In adults, PDGFR/PDGF is important in wound healing, inflammation, and angiogenesis (49). Autocrine signaling as a consequence of platelet-derived growth factor ß overexpression has been implicated in the pathogenesis of dermatofibrosarcoma protruberans (50, 51). Overexpression of PDGFRs and/or their ligands has been described in many solid tumors. PDGFs have been shown to promote proliferation in endometrial epithelial cells (52). Matsuzaki et al. (13) found that PDGFRA is up-regulated in endometriosis stromal but not in epithelial cells relative to eutopic endometrium. This discrepancy could be attributed to our higher statistical power in our study due to a larger sample size (12 patients in our study vs. six in their study).
Besides PDGFRA, Matsuzaki et al. (13) also found that protein kinase C ß1 and Janus kinase were up-regulated, whereas Sprouty2 and MAPK kinase 7 were down-regulated, suggesting that the RAS/RAF/MAPK signaling pathway through PDGFRA is involved in endometriosis pathogenesis. Unfortunately, our chip did not contain protein kinase C ß1, Janus kinase 1, Sprouty2, and MAPK kinase 7, and thus we were unable to validate their finding. However, our study lends further support for the involvement of the MAPK signaling pathway in endometriosis pathogenesis (Fig. 3). In fact, our study identified 13 genes in all three distinct, classical subfamilies of the MAPK signaling pathways (53, 54): ERK, cJun N-terminal kinase/stress-activated protein kinase (JNK/SAPK), and p38 MAPK. In addition, we also identified one gene in the recently characterized MAPK pathway the Big MAPK-1/ERK5 (55). In the RAS/RAF/MAPK/ERK pathway, not only PDGFRA, but also its upstream gene PDGF, was identified. In addition, six more genes downstream of PDGFR and in the pathway were also identified: RAF1, MAPK6 (a member of ERK family), DUSP5 (a member of MAPK phosphatase family), PLA2G5 (a member of cytosolic phospholipase A2 family), MKNK1, and RPS6KA3 (a member of ribosomal S6 kinase 2). In the JNK/SAPK and p38 pathways, TGFB3 (a member of TGFB family), RCA1 (a member of Cdo42/Rac family), AKT1 (a member of AKT family), and DUSP5 have been identified. In addition, HSPB2, which codes for heat shock 27-kDa protein 2, has been identified, suggesting that certain stress response has been evoked in endometriosis. In the MAPK1/ERK5 route, MAPK7 has been identified.
The ERK1/ERK2 route is activated by growth factors and has been linked to the stimulation of cell proliferation in several cellular systems (56, 57). The two other MAPK routes, the JNK/SAPK and the p38 pathways, are triggered largely by cytokine and stress stimuli, and their activation has been shown to regulate apoptosis responses (58, 59, 60).
The MAPK1/ERK5 route has been implicated recently in the control of proliferation (61, 62, 63, 64). It participates in cellular responses to oxidative and mechanical stresses (65, 66), regulates apoptotic responses (55), and plays important roles in angiogenesis and vasculogenesis (67, 68).
Although the absence of other genes involved in the MAPK pathways in our microarray and the lack of knowledge of absolute gene expression levels in either eutopic or ectopic endometrium of women with endometriosis preclude us from knowing exactly how the MAPK pathways are involved in the pathogenesis of endometriosis, the 13 genes identified to be DE clearly indicate that the pathways are involved somehow. In addition, they point out the need to further elucidate the roles of other genes linking the 13 genes identified here in the MAPK pathways in future studies. For example, genes that are downstream of PDGFR and upstream of RAF1, such as GRB2, SOS, and RAS.
Another interesting pathway is oxidative stress, which is particularly interesting given the reports that antioxidant agents suppress cell proliferation of endometrial cells in vitro (69, 70). Our study identified GSTM1, GGTLA1, GSTP1, GSS, and GPX4 to be DE, that are involved in glutathione metabolism. The glutathione S-transferase (GST) gene family encodes genes that are critical for certain life processes, as well as for detoxication and toxification mechanisms. The identification of GPX4 is consistent with the report that GPx is aberrantly expressed in eutopic and ectopic endometrium of women with endometriosis (71).
The identified HSPB2, coding for heat shock 27-kDa protein 2, in our study also is consistent with report that the protein, along with other heat shock proteins, is aberrantly expressed in ectopic and eutopic endometrium of women with endometriosis (72). An essential function of these proteins is to "chaperone" protein synthesis, in that it prevents abnormal interactions and participates in protein synthesis while remaining separate from the final structure (5). As stress response proteins, they can be activated by numerous stimuli, including oxidative stress (73). Because the GST genes are activated in response to oxidative stress (74), it lends further support for involvement of oxidative stress in the pathogenesis of endometriosis (73, 75, 76, 77, 78).
Some other lines of evidence further support the involvement of oxidative stress. We also identified superoxide dismutase 1 (SOD1) and cytochrome C to be DE. SODs are a critical antioxidant enzyme that protects the cells against oxidative stress by scavenging superoxide anions. It has been reported that both Zn- and Mn-SOD expressions are higher in the endometrium of women with endometriosis, suggesting that oxidative stress may play a key role in endometriosis (79). Cytochrome C is involved in oxidative-phosphorylation (80).
Other genes and pathways
In the KEGG pathway search, 314 genes with known names were not identified to be involved in known pathways. This by no means suggests their lack of importance in the endometriosis pathogenesis because 1) the KEGG itself is evolving; 2) the functions of many genes are yet to be defined under various contexts; and 3) our arrays did not contain all known genes. Here we mention three genes, MMP16, TIMP-2, and ICAM5, that were identified to be DE but not included in any of the pathways identified.
The matrix metalloproteinase (MMP) system consists of the enzymatic component, the MMPs, and the enzyme inhibitory component, the tissue inhibitors of metalloproteinases (TIMPs) (81). The MMP family has 22 members identified so far in humans (82), which are structurally related endopeptidases and are collectively capable of degrading all components of the ECM. Its proteolytic activity happens in routine physiological processes such as tissue remodeling, wound healing, angiogenesis, and reproduction (83, 84). The TIMPs, four homologs identified so far, are a family of 20- to 29-kDa secreted proteins that bind to and inhibit the active MMPs and the major regulators of MMPs at the tissue or cellular level (84). The aberrant or elevated levels of MMPs in endometriosis have been reported at MMP-1 (85, 86), MMP-2 (87), MMP-3 (88, 89), MMP-7 (90), and MMP-9 (91). The reduced levels of MMP inhibitors TIMP-1 and TIMP-2 also have been reported (85, 89). This study suggests that MMP-16, too, may be involved in endometriosis pathogenesis.
Intercellular adhesion molecules or ICAMs are an immunoglobulin superfamily and play important roles as adhesion molecules in the hematopoietic system (92, 93). In endometriosis research, the involvement of ICAM-1 in endometriosis has long been recognized. Somigliana et al. (94) found that soluble ICAM-1 was constitutively shed from the surface of endometrial stromal cells harvested from women with endometriosis into the culture medium. Soluble ICAM-1 levels in serum and peritoneal fluid of women with endometriosis also have been reported to be elevated (95, 96, 97). A significantly reduced expression of ICAM-1 in the secretory endometrial cells of women with endometriosis also has been reported (96, 98). However, the involvement of ICAM-5 in endometriosis has not been reported so far.
Pathways and disease pathogenesis
Although 79 pathways identified appear to be diverse, it should be noted that many pathways are interconnected. This can be seen first by the fact that many genes fall into multiple pathways. In addition, TGF-ß, along with GnRH, has been shown to activate MAPK in a dose-, time-, and cell-dependent manner in endometrial cells (99). The MAPK pathways, as signaling pathways, also regulate through TAK1 and NLK canonical Wnt signaling pathway (100). One obvious question regarding the involvement of MAPK pathways is why they are involved. The identification of the glutathione metabolism pathway in which five genes (GGTLA1, GPX4, GSS, GSTM1, and GSTP1) have been identified to be DE in our study, coupled with other pathways such as cytokine-cytokine receptors, suggest that the cellular response to oxidative stress and inflammatory cytokines occurs by signaling through MAPK pathways (101). MAPK pathways themselves are linked with apoptosis (Fig. 3).
With over 900 genes DE, 79 pathways identified, and numerous biological themes, it is easy to see that there are vast differences, at least at the transcriptional level, between the ectopic and eutopic endometrium, despite the fact that endometriosis is defined as the ectopic presence of endometrial glands and stroma. Because these vast differences are merely a snapshot in the long process of endometriosis pathogenesis, many genes, especially those involved in the initiation of the disease, may not be captured. This is especially true because our microarrays do not contain all genes in our genome. In addition, our transcriptional analysis cannot capture any epigenetic changes (102) or constitutive changes in the genome (103). Some, but not all, transcriptional changes that we saw are surely linked with the pathogenesis of endometriosis. Other changes may also have occurred as a result of ectopic relocation of endometrial cells. Further studies are warranted to distinguish these primary and secondary changes in gene expressions.
Like many other chronic diseases, endometriosis is a complex and progressive disease that may well be etiologically heterogeneous, involving many genes and/or gene products (104), just as this study has shown. In addition, the lengthy process from disease initiation to onset of overt disease with observable clinical presentation adds to the difficulty of investigating its possible causes. This is further complicated by the lack of noninvasive diagnostic procedures for endometriosis, precluding observation of the natural history of the disease. Indeed, the traditional approach of studying one gene/protein at a time provides a desirable, yet piecemeal glimpse at the inner secret of endometriosis pathogenesis. As is becoming increasingly evident, endometriosis appears to be a system-wide disease affecting many aspects of reproductive health and well-being (5, 105). As in complex systems, "while in many cases properties of individual components can be well characterized in a laboratory, these isolated measurements are typically of relatively little use in predicting the behavior of large scale interconnected systems or mitigating the cascading spread of damage due to the seemingly innocuous breakdown of individual parts" (106). Indeed, the heterogeneity observed in endometriosis is to be expected, because, like tumor, heterogeneity among endometriotic lesions may be a result of "a high level of redundancy, and hence increased chances of survival and growth" (107). The heterogeneity is very likely caused by structural and functional changes in the genome and by equally complex epigenetic changes. Because the female reproductive system itself is very complex, involving several organs, a more fundamental system or near system-wide approach is needed because it is unlikely that endometriosis is caused by a single gene or gene product.
The high throughput molecular genetic technologies such as cDNA microarrays have provided us with powerful tools that we can use to approach the endometriosis pathogenesis from new angles. As a 10-yr-old technology, microarrays have swept all fields of biomedical research and have proven to be a valuable, efficient, and reliable tool for discovery and for classification of disease subtype (14). The fact that many genes identified to be DE in this study confirmed previously reported results, such as TIMP-2, IL-8, and IL-15, also attests the efficiency and reliability of this technology.
Classification of endometriosis
The most widely adopted classification system for endometriosis, the revised American Fertility Society scheme (29), has so far not done well in terms of predicting treatment responses for either infertility or chronic pelvic pain (108, 109, 110, 111, 112, 113, 114, 115). Considering the enormity of variables in classification/staging of endometriosis—location, number, size, depth, and morphology of lesions, and, above all, the difference in treatment goals (achievement of pregnancy, alleviation of pelvic pain, and reduction in recurrence), establishing a useful classification system for endometriosis based on clinical observations can be daunting.
The successful classification of endometriosis by the location of the lesion based on these DE genes/ESTs suggests the possibility of a molecular genetic classification of the disease. This seems to echo the view that three types of endometriotic lesions—peritoneal, ovarian, and rectovaginal—should be considered separate entities, each with a different pathogenesis (116). In addition, our cluster analysis indicates that both types of endometriosis had lower expression levels in genes involved in cell adhesion, Wnt signaling, and induction of apoptosis (cluster 3), and higher expression levels in genes responsible for acute-phase response, cell proliferation, cell cycle, and regulation of transport (cluster 5). They differ in expression levels in genes responsible for glycoprotein (cluster 1), response to oxidative stress (cluster 2), and G protein-coupled receptor (cluster 4). Although a more precise definition of the five clusters and the reason why the two types of endometriosis differ in this way still warrant further research, it is clear that the two types are transcriptionally different.
Because the gene expression profiles often are predictive of survival or prognosis in patients with cancer (15, 117), an interesting question would be whether the expression profiles are of any predictive value in predicting the time to recurrence since recurrence risk after surgery is quite high in endometriosis (16).
|
Conclusions |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
|
Acknowledgments |
|---|
|
Footnotes |
|---|
First Published Online September 29, 2005
Abbreviations: aRNA, Antisense RNA; DE, differentially expressed; EASE, Expression Analysis Systematic Explorer; EST, expressed sequence tag; GO, gene ontology; GST, glutathione S-transferase; JNK/SAPK, cJun N-terminal kinase/stress-activated protein kinase; KEGG, Kyoto Encyclopedia of Genes and Genomes; LCM, laser capture microdissection; MDS, multidimensional scaling; MMP, matrix metalloproteinase; PDGF, platelet-derived growth factor; PDGFR, PDGF receptor; PDGFRA, PDGFR ; SOD, superoxide dismutase; TIMP, tissue inhibitor of metalloproteinase.
Received April 12, 2005.
Accepted for publication September 17, 2005.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsReferences |
|---|
