Microarray Analysis of Endometrial Carcinomas and Mixed Mullerian Tumors Reveals Distinct Gene Expression Profiles Associated with Different Histologic Types of Uterine Cancer Free

The American Cancer Society estimated that ∼40,100 new cases of cancer of the uterine corpus will be diagnosed during 2004 and ∼6,800 women are expected to die of their disease during that time (1). The majority of endometrial cancers are carcinomas, which may be characterized as type I or type II on the basis of both clinical presentation as well as histopathologic variables (2). Type I endometrial carcinomas are usually endometrioid in histology, well-differentiated, present with early-stage disease, and are often associated with a hyperestrogenic milieu (3). These tumors display a high incidence of alterations in the PTEN tumor suppressor gene (4–6) as well as defects in mismatch repair that results in microsatellite instability (7, 8). In contrast, type II endometrial cancers are more often poorly differentiated, at advanced stage at the time of diagnosis, and are nonendometrioid in histology (3). These tumors rarely, if ever, contain PTEN mutations or microsatellite instability (9) but are more likely to be characterized by p53 mutation and widespread aneuploidy (10–12).

Although the majority of uterine cancers are carcinomas that arise from the endometrial lining, ∼2% to 4% of uterine cancers are sarcomas that arise in the smooth muscle of the uterine wall (1). The majority of uterine sarcomas are classified as mixed mullerian tumors, which contain both carcinomatous and sarcomatous elements. Chemotherapeutics for mixed mullerian tumors have traditionally been similar to those effective in the treatment of other types of soft tissue sarcomas. There is, however, molecular evidence [i.e., X-chromosome activation experiments (13, 14), allelotyping studies (15), and mutation analysis (16)] to suggest that the carcinomatous component of mixed mullerian tumors is the cell type of origin and that the sarcomatous component is derived from the carcinoma through metaplastic transformation or from a stem cell that undergoes divergent differentiation (17, 18). The association of mixed mullerian tumors with obesity, exogenous estrogen use, and tamoxifen suggests clinical similarities with endometrioid endometrial carcinomas (19, 20). However, unlike most endometrioid carcinomas, mixed mullerian tumors are aggressive with a prognosis similar to papillary serous adenocarcinomas that are associated with poor outcome. Despite the clinical features that mixed mullerian tumors can share with endometrial adenocarcinomas, little is known regarding the molecular features that distinguish uterine carcinomas from sarcomas.

Our group has previously used cDNA microarray to examine the gene expression profiles of different histologic types of endometrial adenocarcinoma (21). The results of our initial analysis suggested that the gene expression profile for endometrioid, clear cell, and papillary serous endometrial cancers are distinct, and we identified several additional pathways important in the development of endometrial cancer. We have hypothesized that the gene expression profiles of mixed mullerian tumors are also distinct from both common types of uterine adenocarcinoma. The aim of this study is to present a more comprehensive genomic analysis of uterine cancer to better characterize the molecular expression profiles of different histologic types of uterine cancer. Elucidation of these molecular expression signatures may be useful in predicting the clinical behavior of uterine cancers as well as identifying candidate cellular pathways that can be targets for future therapeutics.

Tissue specimens. Flash-frozen cancer specimens were obtained from 119 patients undergoing surgery for uterine cancer at Duke University Medical Center. These included 66 endometrioid, 24 papillary serous, and 29 mixed mullerian tumors of low and high grades. All tissues were collected under an Institutional Review Board–approved protocol at Duke University Medical Center. Specimens were harvested by pathologists using gross specimens within 30 minutes of specimen removal at the time of surgery. Each uterine tumor was then frozen until the time of the analysis. Tissue specimens were evaluated by H&E to confirm that the specimen to be analyzed contained at least 50% or greater cancer cells. During preparation of the specimens for analysis, care was taken to macroscopically dissect the cancer away from any adjacent myometrium. Tissue samples were subjected to RNA isolation using TRIzol followed by an additional level of purification with the RNeasy kit (Qiagen, Valencia, CA). RNA was successfully extracted from each of the cancer specimens and 10 of the 15 normal endometrium samples. The integrity of each of the RNA samples was confirmed using denaturing gel electrophoresis (22).

Gene expression analysis. The gene expressions were assessed using the Affymetrix human genome U133A and B Genechips (45,000 gene transcripts covering 28,473 UniGene clusters). Approximately 5 μg total RNA from each sample were labeled using high yield transcript labeling kit (Enzo Life Sciences Inc., Farmingdale, NY) and labeled RNAs were hybridized, washed, and scanned according to manufacturer's specifications (Affymetrix, Inc., Santa Clara, CA). Affymetrix Microarray Suite 5.0 software (MAS5) was used to estimate transcript signal levels from scanned images (Affymetrix) by one-step Tukey's biweight algorithm. The probe annotations of HG-U133 chips and MAS5 statistical algorithms are available at Affymetrix website (http://www.affymetrix.com). The signals on each array were normalized to a trimmed mean value of 500, excluding lowest 2% and highest 2% of the signals. An Affymetrix probe set representing a unique Genbank sequence is referred as a probe or gene hereafter for convenience. To verify any errors in the expressions caused by image defects, the correlation coefficient of each array to an idealized distribution was determined where the idealized distribution is mean of all arrays. Visual inspection of scatter plots revealed that 4 of 136 arrays have abnormally high scatter that have correlation coefficients smaller than 0.85. All the arrays having correlation coefficients <0.85 were excluded from further study. The genes were filtered from the remaining arrays using detection P value reported by MAS5. The genes having P > 0.065 in 95% of the arrays were eliminated and all other signals were included for statistical comparisons of classes.

For multidimensional scaling (computed by Partek Pro Discover software build 5, Partek, Inc., St. Charles, MO), the genes included were at P < 0.065 in at least 50% of the arrays. Statistical calculations were done using logarithmic values of normalized signals.

Binary class comparison was done on individual comparisons of different histologic groups using BRB Array tools software (BRB Array tools ver. 3.0c, Richard Simon, Amy Peng, Biometric research branch, National Cancer Institute, NIH, http://linus.nci.nih.gov/BRB-ArrayTools.html). Differentially expressed genes were identified by parametric Student's t tests on genes having at least 50% or more present calls. In each of the comparisons, genes differentially expressed above 2-fold were clustered by the similarity of their expression profiles. Hierarchical clustering was done on logarithmic values of expressions using 1 − ρ as distance metric (16). The heat map was color-coded, using red for up-regulation from normal endometria and green for down-regulation. All the statistical calculations were done on the logarithmic values of signals to the base 2.

Validation of gene expression using quantitative PCR. The expressions of genes chosen for validation were determined by multiplex PCR using TaqMan Gene Expression Assays purchased from Applied Biosystems (Foster City, CA) with β-actin as reference. Samples were run on the ABI Prism 7700 Sequence Detection System according to manufacturer's suggested protocols. The relative quantitation, using the comparative C_T method, was calculated for each sample. The weighted average of the mean ratios of each histologic group was presented with the SE of mean values as error bars.

Unsupervised analysis including all three histologies suggested different global expression patterns associated with each of these groups. We subsequently chose to perform three separate comparisons (endometrioid versus papillary serous, mixed mullerian tumor versus papillary serous, and mixed mullerian tumor versus endometrioid) to better discriminate differences in gene expression patterns between histologic types of uterine cancer.

Endometrioid versus papillary serous carcinoma. Multidimensional scaling on all the genes having 50% present calls suggested that the gene expression of endometrioid and papillary serous carcinomas were different, further supporting the paradigm that these two types of endometrial cancer develop in part via different pathways (Fig. 1A). In a supervised comparison of 66 endometrioid and 24 papillary serous carcinomas, 1,055 genes were found to be differentially expressed at F test P < 0.001, of which 151 of genes had at least at 2-fold change. The tumor to normal expression ratios of 25 most up-regulated and 25 most down-regulated genes are shown as heat map in Fig. 1B. Examples of genes that were notably associated with a >2-fold papillary serous/endometrioid expression ratio included IGF2, PTGS1 (COX1), and p16, whereas genes with a >2-fold endometrioid/papillary serous carcinoma expression ratio included TFF3, FOXA2, and MSX2.

Mixed mullerian tumor versus endometrioid carcinoma. Grossly separated clusters of global expression profiles were noted in the unsupervised comparison of mixed mullerian tumor and endometrioid adenocarcinoma, (Fig. 2A). To identify transcripts that are responsible for this delineation, we did supervised analysis of 29 mixed mullerian tumors and 66 endometrioid carcinomas, which revealed 5,212 genes at F test P < 0.001, including 1,132 genes that were differentially expressed by at least 2-fold and 122 genes that were differentially expressed by at least 5-fold. The tumor to normal expression ratios of 25 most up-regulated and 25 most down-regulated genes are shown as heat map in Fig. 2B. Greater expression of IGF2 and lower expression of MUC1, SCGB2A1, HOXB6, and TFF3 was observed in mixed mullerian tumor specimens when compared with endometrioid carcinomas (Fig. 2B). To further examine the differences between mixed mullerian tumors and endometrioid carcinomas, we examined the global expressions using several class prediction modeling programs.

Mixed mullerian tumor versus and papillary serous carcinoma. Unsupervised analysis using multidimensional scaling showed that the genomic expression profiles of mixed mullerian tumors and papillary serous carcinomas also clustered according to histologic type (Fig. 3A). Supervised analysis of 29 mixed mullerian tumors and 24 papillary serous tumors revealed 1,208 genes at F test P < 0.001, of which 509 genes were differentially expressed by at least 2-fold. The heat map of tumor to normal expression ratios of 25 most up-regulated and 25 most down-regulated genes is shown in Fig. 3B.

Quantitative PCR analysis. We further evaluated the expression of six of these genes (MUC1, SCGB2A1, HOXB6, PTGS1, TFF3, and IGF2) in each of the histologic groups using real-time quantitative PCR to validate the results obtained from the array analysis. Each of the six genes revealed patterns of statistically significant differences in gene expression that were consistent with the microarray analysis. Several of these genes were differentially expressed in two of three comparisons and, therefore, have expression profiles reflective of all three groups in these instances ().

Cross-referencing of gene lists. In a final effort to find genes that are associated with a specific histologic variant of uterine cancer, we identified genes that were differentially expressed at least 2-fold (P < 0.001) on at least two of three gene lists for the following comparisons: endometrioid versus papillary serous; endometrioid versus mixed mullerian tumor, and mixed mullerian tumor versus papillary serous. Using this approach, we identified 66 genes associated with endometrioid cancers present on both the endometrioid versus papillary serous carcinoma and endometrioid versus mixed mullerian tumor list and consistent with a endometrioid carcinoma profile (Table 1); 21 genes associated with papillary serous cancers that were present on both the papillary serous versus endometrioid carcinoma list and the papillary serous carcinoma versus mixed mullerian tumor list and associated with a papillary serous carcinoma profile (Table 2) and 361 genes associated with mixed mullerian tumors that were present on both the mixed mullerian tumor versus endometrioid carcinoma and the mixed mullerian tumor versus papillary serous carcinoma lists and reflective of a mixed mullerian tumor profile (data not shown). The complete cross-referenced lists of differentially expressed genes (P < 0.001) for each of the three comparisons are posted on http://home.ccr.cancer.gov/lbc/risinger/pubs/hist.asp and may provide gene candidates that seem to be associated with a specific type of uterine cancer. Expression Analysis Systematic Explorer software was used to provide an analysis of gene ontology among the cross-referenced gene list for each of the three profiles: endometrioid, papillary serous, and mixed mullerian tumor. The complete data reflective of this analysis are also provided electronically (http://home.ccr.cancer.gov/lbc/risinger/pubs/hist.asp) and further support the paradigm that gene expression associated with different types of uterine cancer seems to be unique.

Previous data from our group (21) have suggested that the gene expression patterns associated with different epithelial types of uterine cancer are distinct. Only one other group has previously reported the details of expression profiles among different histologic types of endometrial cancer (23). In this cDNA microarray analysis, Moreno-Bueno et al. (23) identified only 66 genes that were differentially expressed by at least 2-fold (P ≤ 0.05) between endometrioid and nonendometrioid cancers. In our current microarray analysis, we detected 160 genes that were differentially expressed among endometrioid versus papillary serous cancers despite more stringent statistical criteria (P < 0.001). Our more inclusive list is most likely reflective of an increased sample number and the Affymetrix platform that enabled us to evaluate ∼45,000 gene transcripts covering 28,473 UniGene clusters in contrast to 9,726 clones corresponding to 6,386 different genes used by the other investigators. Although we did obtain similar results for several of the transcripts previously reported (i.e., BUB1, CCNB2, MYC; ref. 23), we did not find that STK15 was significantly overexpressed (at least 2-fold increased expression at P < 0.001) among the papillary serous carcinomas when compared with the endometrioid tumors (23). Our different results may be reflective of the fact that we chose to not include clear cell cases with papillary serous carcinomas in a comparative analysis of endometrioid and nonendometrioid carcinomas. Both our group (21) and other investigators (24) have independently determined that the expression profile of clear cell carcinomas is distinct from that of papillary serous and endometrioid histologic types among cases of endometrial cancer. Although Moreno-Beuno et al. (23) determined that the expression profiles of clear cell cases and papillary serous cases were similar, the number of cases and the array platform used may have prohibited detection. Nevertheless, the investigators determined that STK15 was amplified in five of the nine cases of nonendometrioid cancer available for fluorescence in situ hybridization analysis; the histologic type of positive cases was not reported (23) and it is possible that the major proportion were clear cell. When we did supervised analysis on a predominantly advanced group of endometrioid and papillary serous carcinomas that were matched for stage and grade, we obtained a slightly shorter list of differentially expressed genes (data not shown). These findings suggest that although endometrioid and papillary serous carcinomas are distinct, they may also share genetic alterations that are common to both types of endometrial carcinoma, especially when matched for other clinical prognostic factors.

In our validation of genes differentially expressed between endometrioid and papillary serous carcinomas, we noted that the expression of cyclooxygenase I (PTGS1) was increased among the papillary serous adenocarcinomas when compared with endometrioid cancer or mixed mullerian tumors (Fig. 3). Cyclooxygenase I (PTGS1) is constitutively expressed in most tissues in the body, whereas cyclooxygenase II (PTGS2) is induced in response to certain stimuli. Both isoforms result in production of prostaglandins, some of which have been implicated in carcinogenesis (i.e., PGE-3 and 6-keto PGF_1α) and angiogenesis (25). COX-2 overexpression has been observed in endometrial adenocarcinomas (26) and its expression may be associated with parameters of aggressiveness. The only report that evaluated COX-1 in endometrial cancer did not report histologic type in association with their results but found COX-1 expression to be negligible (27). In vitro studies have indicated that AKT induces COX-2 expression in mutated PTEN endometrioid endometrial cancer cells. Although these studies suggest an association between COX-2 and endometrioid endometrial cancer, there have not been any reports evaluating COX-1 or COX-2 expression in papillary serous cancers. Our findings would suggest that papillary serous adenocarcinomas of the endometrium overexpress COX-1, indicating that further investigations comparing these types of tumors to normal endometrium is warranted to determine whether COX-1 inhibitors might have a role in the prevention of these types of endometrial cancer.

IGF2 was noted to be overexpressed in the analysis of mixed mullerian tumors (P < 0.001) when compared with both endometrioid and papillary serous endometrial cancers (Fig. 4). Although the data are limited, several studies have suggested that IGF2 is associated with sarcomas of the uterus. In vivo analysis of the SK-UT-1 cell line, derived from a uterine mixed mesodermal tumor, has revealed increased binding of IGF2 compared with insulin and IGF-I (28). In addition, IGF2 was found to have a stimulatory effect on the growth of these cells, whereas IGF-I had no effect (29). Finally, loss of imprinting associated with overexpression has been reported in association with both leiomyosarcoma and mixed mullerian tumor of the uterus (30). Together with our findings, the evidence suggests that IGF2 may be overexpressed among mixed mullerian tumors of the uterus.

Two additional genes previously described in association with soft tissue sarcomas were also noted to be differentially expressed between the mixed mullerian tumors and uterine carcinomas. In the supervised analysis of both the mixed mullerian tumor versus endometrioid and the mixed mullerian tumor versus papillary serous carcinoma, we observed up-regulation of SNAIL2, which induces epithelial-mesenchymal transition, cell spreading, and cell separation in vitro (31). SNAIL2 is also a direct repressor of the tumor suppressor gene E-cadherin, which also encodes a cell-to-cell adhesion molecule. Increases in SNAIL2 can result in loss of adhesiveness associated with reduction in E-cadherin leading to increased invasiveness (32). Reduction of E-cadherin has been observed in analysis of soft tissue sarcomas (33), but there is limited evidence regarding E-cadherin expression in mixed mullerian tumor (34). Although our group has previously noted cadherin mutation in association with endometrial carcinomas (35), there have been no prior reports by our group or others regarding E-cadherin expression or SNAIL2 in uterine sarcomas.

In the analysis of mixed mullerian tumor versus either endometrioid or papillary serous carcinoma, there was a limited number of mixed mullerian tumors that seemed to have a gene expression intermediate between the mixed mullerian tumors and either type of carcinoma (Figs. 2B and 3B). These tumors did not seem to differ in terms of stage from the other mixed mullerian tumors that were more distinct in gene expression. It is possible that these cases may have had less carcinomatous component comprising the mixed mullerian tumor. There are no prior reports that quantify the proportion of carcinomatous elements that typically comprise most uterine mixed mullerian tumors. In the absence of this type of data, we did not choose to dissect the tumors to guarantee a set proportion of carcinomatous and sarcomatous components. Similarly, a small subset of endometrioid carcinomas seemed to have a gene expression profile that was somewhat similar to that of the mixed mullerian tumors (Fig. 2B). These tumors also did not seem to be more advanced in stage or grade compared with the other endometrioid carcinomas that were more distinct in gene expression profile.

Investigators have previously suggested that mixed mullerian tumors are characterized by molecular features that are more consistent with a carcinoma than a sarcoma. Many have subsequently advocated the use of chemotherapeutics for mixed mullerian tumors that have been traditionally used in the treatment of uterine papillary serous carcinomas, instead of regimens commonly used in the treatment of soft tissue sarcomas (18). Although mixed mullerian tumors are associated with a poor outcome that is characteristic of uterine papillary serous carcinomas, our findings suggest that the majority of mixed mullerian tumors have gene expression profiles that are distinct from both common histologic types of common endometrial carcinomas and may optimally benefit from therapies that target the unique molecular profile characteristic of these tumors.

The purpose of the current study was to identify genes that were differentially expressed between types of endometrial cancer, not those that distinguish normal endometrium from endometrial cancer subtypes. Comparison of endometrial cancer to normal endometrium is a complex undertaking and would require careful selection of normal samples with consideration given to age, menopausal status, and stage of the menstrual cycle.

In conclusion, data from our group has previously suggested that the gene expression patterns associated with different histologic types of uterine cancer are distinct. Using a robust microarray platform that queried over 28,000 UniGene clusters in combination with a large set of uterine cancer specimens, we have determined that the gene expression of endometrioid and papillary serous carcinomas as well as mixed mullerian tumors seem to be distinct, further supporting the paradigm that different histologic types of uterine cancer may develop in part via alternate pathways.