Purpose: Ovarian and uterine carcinomas manifest several differentiation patterns resembling those seen in nonneoplastic epithelia of the gynecologic tract. Specific oncogene and tumor suppressor gene defects have been associated with particular differentiation patterns in carcinomas arising in either the uterus or ovary. For instance, ovarian and uterine carcinomas with endometrioid differentiation frequently show β-catenin mutations. Whereas type of differentiation is considered in the treatment of uterine carcinomas, it does not presently contribute to decisions about treatment of ovarian carcinomas. A widely accepted view is that the accumulation of specific gene defects and gene expression changes underlies phenotypic traits of cancers, including their response to treatment.

Experimental Design: Using oligonucleotide microarrays to assess gene expression in 103 primary ovarian and uterine carcinomas, we sought to address whether organ of origin or type of differentiation (histotype; endometrioid versus serous) had a more substantial effect on gene expression patterns.

Results: We found that effects on gene expression due to organ of origin and histotype are similar in magnitude and are parallel in that organ effects are similar in the two histotypes and histotype effects are similar in the two organs. In addition, ovarian and uterine endometrioid adenocarcinomas with β-catenin defects show a common gene expression signature largely distinct from that seen in tumors lacking such defects.

Conclusions: Our results illustrate how organ of origin, type of differentiation, and specific molecular defects all contribute to gene expression in the most common types of ovarian and uterine cancers. The findings also imply gene expression data will be of value for stratifying ovarian cancer patients for new treatment approaches.

Gynecologic malignancies account for nearly 29,000 cancer-related deaths in the United States each year; of these, carcinomas of the uterine endometrium are most common (1). Uterine carcinomas have been classified into two primary types: estrogen dependent (type I) and nonestrogen dependent (type II; refs. 2–5). Type I uterine carcinomas are usually well-differentiated endometrioid adenocarcinomas that arise via a characteristic progression of glandular epithelial proliferation, with increasing degrees of atypia, to frank adenocarcinoma. They are most commonly found in premenopausal and perimenopausal women and account for the vast majority of uterine carcinomas. Type II uterine carcinomas usually exhibit serous or, less frequently, clear cell differentiation and are thought to originate from atrophic endometrium. These type II uterine carcinomas more commonly develop in older women, are clinically aggressive, and seem unrelated to estrogen exposure. Molecular analyses have shown characteristic genetic alterations and gene expression signatures in each type of uterine carcinoma (6). For example, microsatellite instability and alterations of CTNNB1, PTEN, and K-RAS are common in type I uterine carcinoma, whereas p53 mutations are more frequent in type II uterine carcinoma (7–11). The dissimilar pathogenesis and clinical behavior of the two major types of uterine carcinoma have provided the basis for therapeutic strategies tailored to each type. In addition, molecularly based strategies are now being investigated in phase I and II clinical trials to target each uterine carcinoma type more precisely, including tumors of patients with advanced or recurrent disease (12).

Ovarian carcinomas are less common than uterine carcinomas but account for the largest proportion of deaths attributable to gynecologic cancers (1). Compared with the uterine carcinomas, less is known about ovarian cancer pathogenesis, and classification of ovarian carcinomas has traditionally been based on morphologic criteria alone. Ovarian carcinomas usually exhibit one of four major types of differentiation: serous, endometrioid, clear cell, or mucinous. Ovarian serous carcinomas account for over 50% of all ovarian carcinomas and almost always present as stage III or IV disease (13). Ovarian endometrioid carcinomas account for ∼25% of all ovarian carcinomas and also often present with advanced stage disease. Mucinous and clear cell ovarian carcinomas are less common, and are more frequently confined to the ovary at the time of diagnosis. In contrast to type I (endometrioid) versus type II (serous and clear cell) uterine carcinomas, the different histologic types of ovarian carcinomas are currently treated with identical therapeutic strategies. Even recently described novel chemotherapeutic approaches do not differentiate between ovarian carcinoma subtypes (14), despite evidence that histologic type may influence response to platinum-based first-line chemotherapy (15).

In recent years, increasing support has been given to the hypothesis that each ovarian carcinoma subtype may represent a biologically and pathogenetically distinct entity (16–18). In support of this notion, and comparable to the situation in the uterine endometrium, particular genetic alterations have been found to characterize each ovarian carcinoma tumor subtype. For example, ovarian serous carcinomas show frequent mutations of p53, whereas p53 mutations are rare in all other subtypes. Similarly, although K-ras and β-catenin oncogenic mutations are uncommon in ovarian carcinoma as a group, 85% of mucinous carcinomas show K-RAS mutations and ovarian endometrioid carcinomas frequently (40%) exhibit β-catenin mutations (19–23). Additionally, comprehensive gene expression profiling has revealed distinctive, albeit partially overlapping, expression signatures for each ovarian carcinoma subtype (17).

Microscopically, endometrioid and serous ovarian carcinoma subtypes are histologically similar to their counterparts in the uterus. This is perhaps not surprising, in light of the fact that all female genital tract epithelia seem to arise from the coelomic mesothelium during embryogenesis. In addition, recent molecular analyses have identified comparable genetic alterations in tumors with similar differentiation, regardless of their organ of origin. For example, microsatellite instability and PTEN and CTNNB1 mutations are frequently observed in both uterine endometrioid carcinomas and ovarian endometrioid carcinomas, whereas mutations of p53 are frequently observed in uterine serous carcinomas and ovarian serous carcinomas (9, 24–27). Such findings suggest that type of differentiation (e.g., endometrioid versus serous) may be at least as important as organ of origin (i.e., uterus versus ovary) in determining a given tumor's gene expression signature. If verified, this would have obvious implications for the development and implementation of novel histotype-specific strategies for treatment of ovarian cancer. In previous studies, we have shown that deregulation of a specific molecular signaling pathway (i.e., the canonical Wnt signaling pathway) is a major determinant of the gene expression pattern seen in a subset of ovarian carcinomas (i.e., the endometrioid carcinomas). Whether such defects lead to a similar gene expression signature in endometrioid-type uterine carcinomas has not yet been determined. To address relationships among histologic type, organ of origin, and Wnt pathway status in ovarian carcinomas and uterine carcinomas, we employed high-density oligonucleotide microarrays to compare gene expression profiles among 103 primary uterine and ovarian adenocarcinomas with endometrioid or serous differentiation.

Tumor Samples. One hundred three snap-frozen primary tumors were analyzed: 75 from the Cooperative Human Tissue Network/Gynecologic Oncology Group tissue bank (Columbus, OH), 22 from the University of Michigan Health System, 4 from New York Presbyterian Hospital-Cornell Center, 1 from the Johns Hopkins Hospital, and 1 from Kumamoto University Hospital (Japan). The 103 tumors included 52 ovarian serous carcinomas, 5 uterine serous carcinomas, 33 ovarian endometrioid carcinomas, and 13 uterine endometrioid carcinomas (Table 1). Tumors were classified as well differentiated (grade 1), moderately differentiated (grade 2), or poorly differentiated (grade 3) using standard criteria (13). Tumors were staged according to the International Federation of Gynecology and Obstetrics (FIGO) criteria (28). Analysis of tissues from human subjects was approved by the University of Michigan's Institutional Review Board (IRB-MED 2001-0568 and 1999-0428).

Table 1

Distribution of tumors by location, histologic type, and tumor stage and grade

TypenStage
Grade
I/IIIII/IVUnknown1/23
Ovarian serous carcinoma 52 46 20 32 
Uterine serous carcinoma  
Ovarian endometrioid carcinoma 33 20 12 20 13 
Uterine endometrioid carcinoma 13 11  10 
TypenStage
Grade
I/IIIII/IVUnknown1/23
Ovarian serous carcinoma 52 46 20 32 
Uterine serous carcinoma  
Ovarian endometrioid carcinoma 33 20 12 20 13 
Uterine endometrioid carcinoma 13 11  10 

RNA Isolation, cDNA Synthesis, and Gene Expression Profiling. Primary tumor tissues were manually microdissected before RNA extraction to ensure that each tumor sample contained at least 70% neoplastic cells. Total RNA was extracted from frozen tissue biopsies using Trizol (Invitrogen, San Diego, CA), then further purified using RNeasy spin columns (Qiagen, Valencia, CA) according to manufacturers' protocols. Oligonucleotide microarrays [HuGeneFL arrays (7,129 probe sets), Affymetrix, Santa Clara, CA] were used in this study. The preparation of cDNA, hybridization, and scanning of the microarrays were done according to the manufacturers' protocols, as reported previously (17, 29). Gene expression data from the ovarian carcinomas have been previously described in detail (17) and the microarray data from these tumors and the uterine cancers are publicly available (http://dot.ped.med.umich.edu:2000/pub/Ovary/index.html). Gene expression data from the uterine carcinomas were processed using the same methods applied to the ovarian tumors. Briefly, to obtain an expression measure for a given probe set, the mismatch hybridization values were subtracted from the perfect match values, and the average of the middle 50% of these differences was used as the expression measure for that probe set. A quantile normalization procedure was done to adjust for differences in the probe intensity distribution across different chips.

Data Processing and Statistical Analysis. The data were transformed via log2[max(X,1)]. Four differences were then calculated for each gene: (i) serous histotype mean minus endometrioid histotype mean within ovary, (ii) serous histotype mean minus endometrioid histotype mean within uterus, (iii) uterus organ mean minus ovary organ mean within endometrioid histotype, and (iv) uterus organ mean minus ovary organ mean within serous histotype. These differences were compared in scatterplots to assess whether in the data set as a whole, organ specific differences were invariant between histotypes, and histotype specific differences were invariant between organs.

For identification of differentially expressed genes, we deemed a gene differentially expressed if it exhibited at least 2-fold mean difference and t test P < 0.05 in a given one of the four comparisons listed above. To assess significance, the number of such genes was counted for each of the four comparisons. Then the samples were randomized 1,000 times (random reassignment to histotype and organ groups), differential expression was assessed on the randomized data as described above, and the 95th and 99th percentiles for the number of differentially expressed genes in randomized data were calculated for each of the four comparisons. The observed number of differentially expressed genes was compared with the randomized percentiles for each of the four comparisons. The identical selection procedure and randomization approach was used to identify genes differentially expressed between Wnt pathway–intact and Wnt pathway–defective ovarian endometrioid carcinoma samples.

Analysis of Wnt Signaling Pathway Status. A training set of 33 well-characterized ovarian endometrioid carcinoma samples was used to form a gene expression profile distinguishing the Wnt pathway–intact samples (n = 21) from Wnt pathway–deregulated samples (n = 12). Molecular analysis of these tumors has been previously described in detail (23, 30, 31). A graphical discriminant analysis procedure (described below) was used to determine whether gene expression differences between Wnt pathway–defective and Wnt pathway–intact uterine endometrioid carcinoma samples conformed to the ovarian endometrioid carcinoma expression profile.

Genes associated with Wnt pathway status in the ovarian endometrioid carcinoma samples were selected if they had at least a 2-fold difference between the pathway-deregulated and pathway-intact within-group means, and a t test P value for the comparison of wild-type to mutant samples was <0.05. Using these criteria, 83 marker genes associated with Wnt pathway status were identified based on the ovarian samples. In a randomization analysis of the ovarian endometrioid carcinoma samples, the average number of genes meeting the selection criteria was 12, with 95th and 99th percentiles equal to 30 and 54, respectively. This indicates that most of the 83 Wnt pathway marker genes are likely to be true positives.

Next, a two-dimensional linear projection of the 83 marker gene expression levels separating pathway-intact from pathway-deregulated ovarian endometrioid carcinoma samples was constructed as follows. Principal component analysis was used to project the expression levels of the 83 marker genes down to 20 dimensional summary vectors (accounting for 57% of the overall variation), in the same manner as done in a previous analysis of the ovarian tumors (17). Then linear discriminant analysis was done on the 20 dimensional summaries to determine an optimal two-dimensional projection for distinguishing Wnt pathway–intact from Wnt pathway–deregulated ovarian endometrioid carcinoma samples. Because there were fewer ovarian training specimens than the number of marker genes (33 compared with 83), linear discriminant analysis was not applied directly. The uterine endometrioid carcinoma samples were then placed onto the same set of axes as the ovarian endometrioid carcinoma samples, and their locations were analyzed according to their Wnt pathway status, as determined by the presence or absence of nuclear β-catenin immunostaining. Combined analysis in which marker genes and multivariate projections are learned from both uterine endometrioid carcinoma and ovarian endometrioid carcinoma samples was not carried out due to the smaller number of uterine samples.

Immunohistochemical Analysis of β-Catenin Expression. Five-micrometer-thick sections of formalin-fixed, paraffin-embedded sections from all ovarian endometrioid carcinoma and uterine endometrioid carcinoma samples were immunohistochemically stained with mouse monoclonal anti-β-catenin antibody (Transduction Laboratories, Lexington, KY) diluted 1:500 as described previously (23). Antigen-antibody complexes were detected with the avidin-biotin peroxidase method using 3,3′-diaminobenzidine as a chromogenic substrate (Vectastain avidin-biotin complex kit, Vector Laboratories, Burlingame, CA). Immunostained sections were lightly counterstained with hematoxylin and then examined by light microscopy. Immunoreactivity was scored independently by three observers (M.P.K., D.R.S., and K.R.C.) for the presence of nuclear versus membranous staining. Samples were designated Wnt pathway deregulated if at least 5% of cells displayed unequivocal nuclear immunoreactivity. Nuclear staining was usually accompanied by reduction or loss of the membranous staining pattern noted in Wnt pathway–intact tumor samples.

Type of Differentiation and Organ of Origin Both Contribute Significantly to Gene Expression Profile. Several studies have shown that gene expression profiles can be used to distinguish among many different types of cancers (32–34) and to identify clinically relevant tumor subsets (35). However, previous studies have also shown that highly related tumors, such as those derived from the Müllerian system, can be difficult to classify based on gene expression signature alone, particularly if histologic type is not taken into account (32). To determine the relative contribution of organ of origin versus type of differentiation to gene expression in the most common histologic types of ovarian and uterine carcinoma, we profiled gene expression in 103 primary ovarian and uterine serous and endometrioid carcinomas, including 52 ovarian serous carcinomas, 5 uterine serous carcinomas, 33 ovarian endometrioid carcinomas, and 13 uterine endometrioid carcinomas. Details regarding the distribution of histologic type, tumor stage, and grade of the tumors studied are provided in Table 1. Four lists of genes derived from comparisons of four pairs of tumor subgroups were generated, using 2-fold differential expression and t test P < 0.05 as selection criteria. Specifically, genes differentially expressed in ovary versus uterus within either of the two histotypes, and genes differentially expressed in serous versus endometrioid carcinomas within either organ were selected. After a randomization procedure, the 95th and 99th percentiles for the chance number of differentially expressed genes were calculated for each of the four comparisons. The observed number of differentially expressed genes was then compared with the randomized percentiles for each comparison. To the extent that more genes were selected in the actual data than are expected by chance, we can state that at least some genes are likely to be differentially expressed even after accounting for the large number of tests carried out.

The observed numbers (95th percentiles) of differentially expressed genes for each comparison are shown in Table 2 and summarized as follows: serous versus endometrioid expression within uterus, 216 (146); serous versus endometrioid expression within ovary, 62 (5); uterus versus ovary expression within serous histotype, 121 (111); and uterus versus ovary expression within endometrioid histotype, 98 (28). These results clearly indicate that each of the four groups exhibits a characteristic pattern of gene expression. For certain comparisons (e.g., serous versus endometrioid histotype within ovarian carcinomas), the list of genes exhibiting characteristic expression can be stated with high certainty. For other comparisons (e.g., uterus versus ovary within carcinomas of serous histotype), the list may contain many false positives, but nevertheless, there is strong evidence that at least some genes are characteristically expressed. The top 15 genes, based on fold increase in gene expression for each comparison group, are listed in Table 3 (complete lists of differentially expressed genes will be provided upon request).

Table 2

The number of 2-fold differentially expressed genes for each comparison

Uterus
Ovary
ObservedExpectedObservedExpected
Up in serous 103 45 (87, 110)* 43 1 (3, 5) 
Up in endometrioid 113 38 (84, 111) 19 1 (3, 5) 
Total 216 83 (146, 180) 62 1 (5, 8) 
     
 Serous
 
 Endometrioid
 
 
 Observed
 
Expected
 
Observed
 
Expected
 
Up in uterus 69 32 (63, 89) 74 6 (18, 30) 
Up in ovary 52 24 (59, 86) 24 5 (14, 26) 
Total 121 54 (111, 147) 98 11 (28, 43) 
Uterus
Ovary
ObservedExpectedObservedExpected
Up in serous 103 45 (87, 110)* 43 1 (3, 5) 
Up in endometrioid 113 38 (84, 111) 19 1 (3, 5) 
Total 216 83 (146, 180) 62 1 (5, 8) 
     
 Serous
 
 Endometrioid
 
 
 Observed
 
Expected
 
Observed
 
Expected
 
Up in uterus 69 32 (63, 89) 74 6 (18, 30) 
Up in ovary 52 24 (59, 86) 24 5 (14, 26) 
Total 121 54 (111, 147) 98 11 (28, 43) 
*

Next to the expected numbers indicated, upper percentiles are given in parentheses (95th percentile, 99th percentile) for the number of genes exhibiting 2-fold differential expression by chance.

Table 3

Top 15 up-regulated genes (fold increase)

Up in uterus versus ovary within serous histotypeUp in uterus versus ovary within endometrioid histotype
HBBb 5.9 IGKC 9.1 
HBBc 5.1 MSX1 5.7 
MMP7 4.9 S71043 5.7 
PLS3 3.8 IGHG3 
HGD 3.7 MSX1b 4.4 
HBA2 3.4 TFF3 4.3 
IGHMc 3.3 MMP7 4.3 
S71043 3.2 HBBb 3.6 
LAMB1 MMP12 3.5 
GPNMB 2.9 IL8 3.5 
TCF2b 2.8 S100P 3.4 
ABP1b 2.8 ID1b 3.3 
PLAB 2.8 NT5 3.3 
RGS2 2.8 FABP4 3.1 
HOXB5 2.8 ID3 
    
Up in ovary versus uterus within serous histotype
 
 Up in ovary versus uterus within endometrioid histotype
 
 
MSLN DLK1 5.6 
GAS6 4.5 C7 4.5 
PTMS 4.3 ASS 4.3 
CKB 4.2 GAS6 4.2 
KLK7 PEG3 4.2 
FOLR1 3.8 LY6E 3.2 
UQCRH 3.8 SMARCD3 2.9 
STHM 3.6 WT1 2.6 
MT1G 3.5 SST 2.6 
MT2Ab 3.3 STAR 2.6 
PCP4 3.3 ATP6B1 2.5 
APOA1 3.3 TNNT1 2.4 
HE4 3.2 PFKM 2.4 
S100A1 PRKCI 2.3 
SMARCD3 2.9 DPYSL3 2.2 
    
Up in endometrioid versus serous histotype within uterus
 
 Up in endometrioid versus serous histotype within ovary
 
 
MSX1 15 MSX1 3.7 
TFF3 14.3 TFF3 3.6 
MSX1b 9.3 DLK1 
CKB 7.2 MSX1b 
X123 GAD1b 2.8 
ESR1 4.5 CST4 2.7 
PLA2G4A 4.4 ALDH1 2.4 
SFN 4.4 CEACAM1b 2.4 
LTF 4.3 SFN 2.2 
UQCRH 4.2 PIK3R1 2.2 
S100P 4.1 PLAB 2.2 
SERPINA3 MGB1 2.1 
KRT5 3.9 NMA 2.1 
CEACAM1b 3.9 STC1 2.1 
SERPINA1b 3.8 MSX2b 2.1 
    
Up in serous versus endometrioid histotype within uterus
 
 Up in serous versus endometrioid histotype within ovary
 
 
MAL 5.9 FOLR1 8.1 
LHX1 5.7 MSLN 5.5 
KLK6 4.2 PTGS1 4.8 
C7 S100A1 4.8 
PRSS1 3.9 WT1 4.1 
PLTP 3.9 KLK7 3.6 
WT1 3.9 GAS6 3.5 
ASS 3.8 APOA1 3.5 
PTGS1 3.7 IGF2 2.9 
SST 3.4 KLK6 2.8 
GAS6 3.3 IGFBP2 2.8 
FOLR1 3.2 LU 2.7 
MAGP2 3.1 KIAA0275 2.7 
APOE NPR1 2.6 
ABP1b CRABP2 2.6 
Up in uterus versus ovary within serous histotypeUp in uterus versus ovary within endometrioid histotype
HBBb 5.9 IGKC 9.1 
HBBc 5.1 MSX1 5.7 
MMP7 4.9 S71043 5.7 
PLS3 3.8 IGHG3 
HGD 3.7 MSX1b 4.4 
HBA2 3.4 TFF3 4.3 
IGHMc 3.3 MMP7 4.3 
S71043 3.2 HBBb 3.6 
LAMB1 MMP12 3.5 
GPNMB 2.9 IL8 3.5 
TCF2b 2.8 S100P 3.4 
ABP1b 2.8 ID1b 3.3 
PLAB 2.8 NT5 3.3 
RGS2 2.8 FABP4 3.1 
HOXB5 2.8 ID3 
    
Up in ovary versus uterus within serous histotype
 
 Up in ovary versus uterus within endometrioid histotype
 
 
MSLN DLK1 5.6 
GAS6 4.5 C7 4.5 
PTMS 4.3 ASS 4.3 
CKB 4.2 GAS6 4.2 
KLK7 PEG3 4.2 
FOLR1 3.8 LY6E 3.2 
UQCRH 3.8 SMARCD3 2.9 
STHM 3.6 WT1 2.6 
MT1G 3.5 SST 2.6 
MT2Ab 3.3 STAR 2.6 
PCP4 3.3 ATP6B1 2.5 
APOA1 3.3 TNNT1 2.4 
HE4 3.2 PFKM 2.4 
S100A1 PRKCI 2.3 
SMARCD3 2.9 DPYSL3 2.2 
    
Up in endometrioid versus serous histotype within uterus
 
 Up in endometrioid versus serous histotype within ovary
 
 
MSX1 15 MSX1 3.7 
TFF3 14.3 TFF3 3.6 
MSX1b 9.3 DLK1 
CKB 7.2 MSX1b 
X123 GAD1b 2.8 
ESR1 4.5 CST4 2.7 
PLA2G4A 4.4 ALDH1 2.4 
SFN 4.4 CEACAM1b 2.4 
LTF 4.3 SFN 2.2 
UQCRH 4.2 PIK3R1 2.2 
S100P 4.1 PLAB 2.2 
SERPINA3 MGB1 2.1 
KRT5 3.9 NMA 2.1 
CEACAM1b 3.9 STC1 2.1 
SERPINA1b 3.8 MSX2b 2.1 
    
Up in serous versus endometrioid histotype within uterus
 
 Up in serous versus endometrioid histotype within ovary
 
 
MAL 5.9 FOLR1 8.1 
LHX1 5.7 MSLN 5.5 
KLK6 4.2 PTGS1 4.8 
C7 S100A1 4.8 
PRSS1 3.9 WT1 4.1 
PLTP 3.9 KLK7 3.6 
WT1 3.9 GAS6 3.5 
ASS 3.8 APOA1 3.5 
PTGS1 3.7 IGF2 2.9 
SST 3.4 KLK6 2.8 
GAS6 3.3 IGFBP2 2.8 
FOLR1 3.2 LU 2.7 
MAGP2 3.1 KIAA0275 2.7 
APOE NPR1 2.6 
ABP1b CRABP2 2.6 

Next, we wished to determine whether type of differentiation and organ of origin affect gene expression in parallel; that is, that histotype differences are similar in both organs, and organ differences are similar in both histotypes. Two scatterplots were formed; one plots uterus/ovary gene expression differences within the endometrioid carcinomas against uterus/ovary gene expression differences within the serous carcinomas (Fig. 1). The other plots serous/endometrioid differences within the ovary against serous/endometrioid differences within uterus (Fig. 2). For clarity, only the 1,000 most variably expressed genes are shown in each scatterplot. Both scatterplots show statistically significant correlations (r2 = 0.18 for Fig. 1 and r2 = 0.29 for Fig. 2). This indicates that many of the same genes that are specifically expressed in either serous or endometrioid tumors in the ovary tend to follow the same pattern in the uterus. Similarly, genes preferentially expressed in either ovarian or uterine tumors within the serous histotype tend to follow the same pattern in the endometrioid histotype. Collectively, these findings suggest that both type of differentiation and organ of origin contribute to gene expression profiles in the tumors studied. Interestingly, effects due to organ and to histotype are similar in magnitude and are approximately parallel in the sense that organ effects are similar in the two histotypes and histotype effects are similar in the two organs.

Fig. 1

Graphical depiction of uterus/ovary gene expression differences within the endometrioid carcinomas against uterus/ovary gene expression differences within the serous carcinomas. Only 1,000 genes with the greatest overall variation. Positive correlation (r2 = 0.18) indicates that many of the same genes were differentially expressed between uterine and ovarian carcinomas within both histologic types of carcinoma.

Fig. 1

Graphical depiction of uterus/ovary gene expression differences within the endometrioid carcinomas against uterus/ovary gene expression differences within the serous carcinomas. Only 1,000 genes with the greatest overall variation. Positive correlation (r2 = 0.18) indicates that many of the same genes were differentially expressed between uterine and ovarian carcinomas within both histologic types of carcinoma.

Close modal
Fig. 2

Graphical depiction of serous/endometrioid differences within the ovary against serous/endometrioid differences within uterus. Only 1,000 genes with the greatest overall variation. Positive association (r2 = 0.29) indicates that many of the same genes were differentially expressed between the two histologic types within both ovary and uterus.

Fig. 2

Graphical depiction of serous/endometrioid differences within the ovary against serous/endometrioid differences within uterus. Only 1,000 genes with the greatest overall variation. Positive association (r2 = 0.29) indicates that many of the same genes were differentially expressed between the two histologic types within both ovary and uterus.

Close modal

Ovarian and Uterine Carcinomas with Deregulated Wnt Signaling Cluster Distinctly from Samples with Intact Wnt Signaling Pathways. The canonical Wnt signaling pathway modulates many developmental and adult tissue processes, including cell fate specification, proliferation, and differentiation (36). Wnt pathway deregulation has been shown in a number of human cancers (37). We and others have shown that ovarian endometrioid carcinomas with canonical Wnt signaling pathway defects show nuclear accumulation of stabilized β-catenin protein, usually attributable to mutations of CTNNB1, the gene encoding β-catenin, and occasionally to mutational inactivation of proteins involved in regulating levels of free cytosolic β-catenin such as APC or AXIN (23, 38, 39). CTNNB1 mutations and nuclear accumulation of β-catenin have also been identified in a subset of uterine endometrioid carcinomas (40, 41).

In previous work, we established that Wnt pathway status is a major determinant of global gene expression in ovarian endometrioid carcinomas (30). In this study, we wished to determine whether a similar gene expression profile exists for both ovarian endometrioid carcinomas and uterine endometrioid carcinomas carrying Wnt pathway defects. Immunohistochemistry was done on all endometrioid samples from the ovary (previously published data; ref. 23) and uterus to define the pattern of intracellular β-catenin localization. Unequivocal nuclear staining was found in 36% (12 of 33) of ovarian endometrioid carcinoma samples and 31% (4 of 13) of uterine endometrioid carcinoma samples (Fig. 3). These results are consistent with previously published data (38–40). Next, the training set of 33 ovarian endometrioid carcinomas (21 Wnt pathway intact, 12 Wnt pathway deregulated) was used to generate a list of 83 genes from our microarray data that were significantly associated with Wnt pathway status in the 33 ovarian endometrioid carcinomas. Of these 83 genes, only 12 are expected to be false positives as determined by our randomization analysis. We then used a multivariate statistical procedure (see Materials and Methods) to construct a graph, shown in Fig. 4, which displays the ovarian endometrioid carcinoma samples in a way that reflects the expression of these 83 Wnt pathway marker genes. Samples lying near each other are more similar in expression of these 83 genes than those lying further apart. Coordinates in Fig. 4 combine the 83 marker gene expression levels in a way that is optimized to separate the ovarian Wnt pathway–deregulated samples from the ovarian Wnt pathway–intact samples. Importantly, the construction of these coordinate functions did not in any way reference the uterine endometrioid carcinoma data. Next, we placed the uterine endometrioid carcinoma samples on the graph, using uterine endometrioid carcinoma gene expression values and the coordinate functions developed from the ovarian endometrioid carcinoma samples. The most striking feature of Fig. 4 is that three of the uterine endometrioid carcinoma samples fall at a great distance from the other 10 uterine endometrioid carcinoma samples, lying closer to the Wnt pathway–deregulated ovarian endometrioid carcinoma samples than the Wnt pathway–intact samples from either organ. The three “outlier” uterine endometrioid carcinoma samples, two of which represent different areas from the same tumor specimen, all showed nuclear accumulation of β-catenin, indicative of Wnt pathway deregulation in these cases. One uterine endometrioid carcinoma with nuclear β-catenin did not cluster with the others, which may reflect other tumor-specific genetic alteration(s) affecting gene expression. The nine uterine endometrioid carcinomas with intact Wnt signaling are generally more comparable in marker gene expression to ovarian endometrioid carcinomas with intact Wnt signaling than to either uterine endometrioid carcinomas or ovarian endometrioid carcinomas with deregulated Wnt signaling. The abscissa in Fig. 4 apparently contains the important information about Wnt pathway status, allowing separation of nearly all ovarian endometrioid carcinoma and uterine endometrioid carcinoma samples correctly according to whether Wnt signaling is intact versus deregulated. The ordinate seems to reflect gene expression changes unrelated to the status of the Wnt signaling pathway.

Fig. 3

Immunohistochemical staining for β-catenin protein in (A) representative uterine endometrioid carcinoma (UEC) with intact Wnt signaling pathway (note the membranous staining pattern) and in (B) representative UEC with dysregulated Wnt signaling (note the nuclear localization of β-catenin).

Fig. 3

Immunohistochemical staining for β-catenin protein in (A) representative uterine endometrioid carcinoma (UEC) with intact Wnt signaling pathway (note the membranous staining pattern) and in (B) representative UEC with dysregulated Wnt signaling (note the nuclear localization of β-catenin).

Close modal
Fig. 4

Parallelism of gene expression changes associated with Wnt pathway dysregulation in ovarian endometrioid carcinomas (OEC) and uterine endometriod carcinomas (UEC). Most uterine endometriod carcinomas with Wnt pathway dysregulation (mutant) show expression of Wnt pathway marker genes more similar to that in Wnt pathway–defective OECs than to Wnt pathway intact (wild type) uterine endometriod carcinomas. Similarly, UECs with intact Wnt signaling are generally more comparable in marker gene expression to OECs with intact Wnt signaling.

Fig. 4

Parallelism of gene expression changes associated with Wnt pathway dysregulation in ovarian endometrioid carcinomas (OEC) and uterine endometriod carcinomas (UEC). Most uterine endometriod carcinomas with Wnt pathway dysregulation (mutant) show expression of Wnt pathway marker genes more similar to that in Wnt pathway–defective OECs than to Wnt pathway intact (wild type) uterine endometriod carcinomas. Similarly, UECs with intact Wnt signaling are generally more comparable in marker gene expression to OECs with intact Wnt signaling.

Close modal

Although the Wnt pathway gene expression signature was initially determined exclusively using ovarian endometrioid carcinomas, it produced Wnt pathway status-related clustering of uterine endometrioid carcinomas. This indicates that gene expression alterations associated with Wnt pathway status are largely similar in endometrioid carcinomas regardless of origin in the ovary or uterus. The differential clustering of tumors with intact versus aberrant Wnt pathway status suggests that the signaling pathway acts to modulate gene expression in uterine endometrioid carcinomas, as previously shown in ovarian endometrioid carcinomas, and that gene expression profiles of Wnt pathway-deregulated tumors in the female genital tract share a considerable level of similarity to one another. Moreover, these results further support the proposal that gene expression variations in ovary and uterus are substantially parallel. Above, we noted a parallelism in histotype-specific gene expression between uterus and ovary, and here we have provided evidence of a parallelism in Wnt pathway–specific gene expression between uterus and ovary.

Advances in our understanding of the molecular heterogeneity underlying the different histologic types of uterine carcinoma has led to the recognition that each type may respond differently to specific therapeutic approaches (44, 45). Physicians and researchers are beginning to regard ovarian carcinomas in a similar light to their uterine counterparts (i.e., as a pathogenetically heterogeneous group of neoplasms that may eventually be better targeted with treatments specific to type of differentiation and genetic/biochemical features, rather than simply to organ of origin). As tumor differentiation is largely a reflection of a given tumor's genotype and gene expression signature, a major goal for the cancer field as a whole and the gynecologic cancer field in particular, is to determine which molecular alterations confer a selective advantage to tumor cells and to develop therapies targeting these specific molecular defects in the tumor cells.

This study was undertaken to determine the extent to which type of differentiation and organ of origin influence the gene expression signature of endometrioid and serous carcinomas of the female genital tract. To our knowledge, our study is the first to examine differential gene expression between the two most common histologic types of carcinoma in the uterus and ovary using high-density oligonucleotide microarrays. Most recent studies comparing genetic similarity between gynecologic carcinoma subsets have either evaluated only a limited number of molecular defects, or have been restricted to a single organ. We conclude that both type of differentiation and organ of origin and the status of specific molecular signaling pathways play significant roles in determining gene expression in the most common types of ovarian and uterine carcinomas.

Advances in our understanding of the clinical and molecular heterogeneity amongst ovarian carcinoma histotypes has recently prompted the creation of a dualistic model for ovarian tumorigenesis, similar to the two-type system in the uterus (18). In the dualistic model, ovarian carcinomas are proposed to belong to one of two major categories, each corresponding to a different mode of pathogenesis. Type I ovarian carcinomas are low-grade neoplasms that progress in a stepwise fashion from borderline tumors and pursue a more indolent course; type II ovarian carcinomas are high-grade carcinomas that arise de novo and have a more aggressive course. The dualistic model is less focused on type of differentiation and more on tumor grade, although most type II carcinomas are high-grade serous carcinomas and most type I tumors are endometrioid, mucinous, and low-grade serous carcinomas. Our previous study using oligonucleotide microarrays to compare gene expression profiles in a large number of ovarian carcinomas provides some support for the dualistic model (17). Specifically, we found that high-grade carcinomas (mostly serous) were largely separable from the low-grade (mostly endometrioid and mucinous) and clear cell neoplasms based on global gene expression.

As serous tumors account for the vast majority of ovarian cancers and are the second most common type of adenocarcinoma in the uterus, we wanted to compare gene expression profiles of tumors from both organ sites. In support of results from previous studies showing similar genetic anomalies among ovarian and uterine serous carcinomas (20, 46), our microarray data revealed that many of the genes with greater expression in uterine endometrioid carcinomas versus uterine serous carcinomas likewise show greater expression in ovarian endometrioid carcinomas versus ovarian serous carcinomas. For example, endometrioid carcinomas in either organ often highly express the MSX1, TFF3, SFN, and CEACAM1b genes. Similarly, serous carcinomas in either organ often highly express the FOLR1, PTGS1, WT1, and GAS6 genes. Recent microarray-based analysis of gene expression in uterine carcinomas also identified increased expression of FOLR1 and PTGS1 in uterine serous carcinomas compared with uterine endometrioid carcinomas (6). Within the endometrioid carcinomas, there were statistically significant differences in gene expression between uterine and ovarian tumors. This suggests that although endometrioid carcinomas in the uterus and ovary may share some molecular defects, there are likely additional “organ-specific” factors that ultimately contribute to a given endometrioid carcinoma's gene expression signature.

Wnt pathway deregulation has been shown in a number of human cancers, including colorectal carcinomas (47), melanomas (48), hepatoblastomas (49), prostatic adenocarcinomas (50, 51), carcinomas of the uterus (40, 52), and those of the ovary (23, 38, 53). In both the uterus and ovary, tumors with Wnt pathway defects are usually low grade and often show squamous differentiation (23, 38, 40, 42, 53). In our study, the uterine endometrioid carcinomas with Wnt pathway defects show a gene expression signature that is more similar to Wnt pathway–defective ovarian endometrioid carcinomas than to carcinomas with presumptively intact Wnt signaling, regardless of organ of origin. Notably, however, the ovarian and uterine endometrioid carcinomas did not show entirely overlapping patterns of Wnt pathway status marker gene expression, again indicating a role for organ-specific effects and effects of other genetic alterations affecting gene expression.

In closing, we wish to emphasize that this study illustrates how organ of origin, type of differentiation, and specific molecular defects all contribute to gene expression in the most common types of ovarian and uterine cancers. There is increasing recognition of the fact that cancers arising within a given organ may have marked differences in their pathogenesis, clinical behavior, and response to therapy. Whereas this point is widely accepted for uterine carcinomas, it is hoped that the results of this study will help in shaping views about the importance of additional efforts to define biologically and clinically distinct subgroups of ovarian cancers.

Grant support: National Cancer Institute grants U19 CA84953 and RO1 CA94172 and University of Michigan Comprehensive Cancer Center Tissue Core grant P30 CA46952.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: K. Shedden and M. Kshirsagar contributed equally to this study.

We thank Rork Kuick for his expert assistance with microarray data management and Barbara Lamb for outstanding technical assistance.

1
Ries LAG, Eisner MP, Cossary CL, et al. SEER Cancer Statistics Review 1975-2001, National Cancer Institute. Bethesda (MD), http://Seer.Cancer.Gov/Csr/1975_2001/, 2004.
2
Bokhman JV. Two pathogenetic types of endometrial carcinoma.
Gynecol Oncol
1983
;
15
:
10
–7.
3
Lax SF, Kurman RJ. A dualistic model for endometrial carcinogenesis based on immunohistochemical and molecular genetic analyses.
Verh Dtsch Ges Pathol
1997
;
81
:
228
–32.
4
Ellenson LH, Wu TC. Focus on endometrial and cervical cancer.
Cancer Cell
2004
;
5
:
533
–8.
5
Sherman ME. Theories of endometrial carcinogenesis: a multidisciplinary approach.
Mod Pathol
2000
;
13
:
295
–308.
6
Risinger JI, Maxwell GL, Chandramouli GV, et al. Microarray analysis reveals distinct gene expression profiles among different histologic types of endometrial cancer.
Cancer Res
2003
;
63
:
6
–11.
7
Machin P, Catasus L, Pons C, Munoz J, Matias-Guiu X, Prat J. CTNNB1 mutations and β-catenin expression in endometrial carcinomas.
Hum Pathol
2002
;
33
:
206
–12.
8
Tashiro H, Blazes MS, Wu R, et al. Mutations in PTEN are frequent in endometrial carcinoma but rare in other common gynecological malignancies.
Cancer Res
1997
;
57
:
3935
–40.
9
Lax SF, Kendall B, Tashiro H, Slebos RJ, Hedrick L. The frequency of p53, K-ras mutations, and microsatellite instability differs in uterine endometrioid and serous carcinoma: evidence of distinct molecular genetic pathways.
Cancer
2000
;
88
:
814
–24.
10
Tashiro H, Isacson C, Levine RU, Kurman RJ, Cho KR, Hedrick L. p53 gene mutations are common in uterine serous carcinoma and occur early in their pathogenesis.
Am J Pathol
1997
;
150
:
177
–85.
11
Tashiro H, Lax SF, Gaudin PB, Isacson C, Cho KR, Hedrick L. Microsatellite instability is uncommon in uterine serous carcinoma.
Am J Pathol
1997
;
150
:
75
–9.
12
Elit L, Hirte H. Current status and future innovations of hormonal agents, chemotherapy and investigational agents in endometrial cancer.
Curr Opin Obstet Gynecol
2002
;
14
:
67
–73.
13
Scully RE, Young RH, Clement PB. Tumors of the ovary, maldeveloped gonads, fallopian tube, and broad ligament. Atlas of tumor pathology, third series, Fascicle 23 1998.
14
Copeland LJ, Bookman M, Trimble E. Clinical trials of newer regimens for treating ovarian cancer: the rationale for Gynecologic Oncology Group Protocol GOG 182-ICON5.
Gynecol Oncol
2003
;
90
:
S1
–7.
15
Hess V, A'Hern R, Nasiri N, et al. Mucinous epithelial ovarian cancer: a separate entity requiring specific treatment.
J Clin Oncol
2004
;
22
:
1040
–4.
16
Feeley KM, Wells M. Precursor lesions of ovarian epithelial malignancy.
Histopathology
2001
;
38
:
87
–95.
17
Schwartz DR, Kardia SL, Shedden KA, et al. Gene expression in ovarian cancer reflects both morphology and biological behavior, distinguishing clear cell from other poor-prognosis ovarian carcinomas.
Cancer Res
2002
;
62
:
4722
–9.
18
Shih IM, Kurman RJ. Ovarian tumorigenesis: a proposed model based on morphological and molecular genetic analysis.
Am J Pathol
2004
;
164
:
1511
–8.
19
Pieretti M, Cavalieri C, Conway PS, Gallion HH, Powell DE, Turker MS. Genetic alterations distinguish different types of ovarian tumors.
Int J Cancer
1995
;
64
:
434
–40.
20
Fujita M, Enomoto T, Inoue M, et al. Alteration of the p53 tumor suppressor gene occurs independently of K-ras activation and more frequently in serous adenocarcinomas than in other common epithelial tumors of the human ovary.
Jpn J Cancer Res
1994
;
85
:
1247
–56.
21
Teneriello MG, Ebina M, Linnoila RI, et al. p53 and Ki-ras gene mutations in epithelial ovarian neoplasms.
Cancer Res
1993
;
53
:
3103
–8.
22
Cuatrecasas M, Villanueva A, Matias-Guiu X, Prat J. K-ras mutations in mucinous ovarian tumors: a clinicopathologic and molecular study of 95 cases.
Cancer
1997
;
79
:
1581
–6.
23
Wu R, Zhai Y, Fearon ER, Cho KR. Diverse mechanisms of β-catenin deregulation in ovarian endometrioid adenocarcinomas.
Cancer Res
2001
;
61
:
8247
–55.
24
Sato N, Tsunoda H, Nishida M, et al. Loss of heterozygosity on 10q23.3 and mutation of the tumor suppressor gene PTEN in benign endometrial cyst of the ovary: possible sequence progression from benign endometrial cyst to endometrioid carcinoma and clear cell carcinoma of the ovary.
Cancer Res
2000
;
60
:
7052
–6.
25
Obata K, Morland SJ, Watson RH, et al. Frequent PTEN/MMAC mutations in endometrioid but not serous or mucinous epithelial ovarian tumors.
Cancer Res
1998
;
58
:
2095
–7.
26
Fujita M, Enomoto T, Yoshino K, et al. Microsatellite instability and alterations in the hMSH2 gene in human ovarian cancer.
Int J Cancer
1995
;
64
:
361
–6.
27
Caduff RF, Svoboda-Newman SM, Ferguson AW, Johnston CM, Frank TS. Comparison of mutations of Ki-RAS and p53 immunoreactivity in borderline and malignant epithelial ovarian tumors.
Am J Surg Pathol
1999
;
23
:
323
–8.
28
Shepherd JH. Revised FIGO staging for gynaecological cancer.
Br J Obstet Gynaecol
1989
;
96
:
889
–92.
29
Giordano TJ, Shedden KA, Schwartz DR, et al. Organ-specific molecular classification of primary lung, colon, and ovarian adenocarcinomas using gene expression profiles.
Am J Pathol
2001
;
159
:
1231
–8.
30
Schwartz DR, Wu R, Kardia SL, et al. Novel candidate targets of β-catenin/T-cell factor signaling identified by gene expression profiling of ovarian endometrioid adenocarcinomas.
Cancer Res
2003
;
63
:
2913
–22.
31
Zhai Y, Wu R, Schwartz DR, et al. Role of β-catenin/T-cell factor-regulated genes in ovarian endometrioid adenocarcinomas.
Am J Pathol
2002
;
160
:
1229
–38.
32
Shedden KA, Taylor JM, Giordano TJ, et al. Accurate molecular classification of human cancers based on gene expression using a simple classifier with a pathological tree-based framework.
Am J Pathol
2003
;
163
:
1985
–95.
33
Ramaswamy S, Tamayo P, Rifkin R, et al. Multiclass cancer diagnosis using tumor gene expression signatures.
Proc Natl Acad Sci U S A
2001
;
98
:
15149
–54.
34
Su AI, Welsh JB, Sapinoso LM, et al. Molecular classification of human carcinomas by use of gene expression signatures.
Cancer Res
2001
;
61
:
7388
–93.
35
van de Vijver MJ, He YD, van't Veer LJ, et al. A gene-expression signature as a predictor of survival in breast cancer.
N Engl J Med
2002
;
347
:
1999
–2009.
36
Cadigan KM, Nusse R. Wnt signaling: a common theme in animal development.
Genes Dev
1997
;
11
:
3286
–305.
37
Polakis P. Wnt signaling and cancer.
Genes Dev
2000
;
14
:
1837
–51.
38
Saegusa M, Okayasu I. Frequent nuclear β-catenin accumulation and associated mutations in endometrioid-type endometrial and ovarian carcinomas with squamous differentiation.
J Pathol
2001
;
194
:
59
–67.
39
Wright K, Wilson P, Morland S, et al. β-Catenin mutation and expression analysis in ovarian cancer: exon 3 mutations and nuclear translocation in 16% of endometrioid tumours.
Int J Cancer
1999
;
82
:
625
–9.
40
Fukuchi T, Sakamoto M, Tsuda H, Maruyama K, Nozawa S, Hirohashi S. β-Catenin mutation in carcinoma of the uterine endometrium.
Cancer Res
1998
;
58
:
3526
–8.
41
Saegusa M, Hashimura M, Yoshida T, Okayasu I. β-Catenin mutations and aberrant nuclear expression during endometrial tumorigenesis.
Br J Cancer
2001
;
84
:
209
–17.
42
Gamallo C, Palacios J, Moreno G, Calvo de Mora J, Suarez A, Armas A. β-Catenin expression pattern in stage I and II ovarian carcinomas: relationship with β-catenin gene mutations, clinicopathological features, and clinical outcome.
Am J Pathol
1999
;
155
:
527
–36.
43
Schlosshauer PW, Ellenson LH, Soslow RA. β-Catenin and E-cadherin expression patterns in high-grade endometrial carcinoma are associated with histological subtype.
Mod Pathol
2002
;
15
:
1032
–7.
44
Lackman FD, Craighead PS. Therapeutic dilemmas in the management of uterine papillary serous carcinoma.
Curr Treat Options Oncol
2003
;
4
:
99
–104.
45
Holloway RW. Treatment options for endometrial cancer: experience with topotecan.
Gynecol Oncol
2003
;
90
:
S28
–33.
46
Pere H, Tapper J, Seppala M, Knuutila S, Butzow R. Genomic alterations in fallopian tube carcinoma: comparison to serous uterine and ovarian carcinomas reveals similarity suggesting likeness in molecular pathogenesis.
Cancer Res
1998
;
58
:
4274
–6.
47
Morin PJ, Sparks AB, Korinek V, et al. Activation of β-catenin-Tcf signaling in colon cancer by mutations in β-catenin or APC.
Science
1997
;
275
:
1787
–90.
48
Rubinfeld B, Robbins PD, El-Gamil M, Albert I, Porfiri E, Polakis P. Stabilization of β-catenin by genetic defects in melanoma cell lines.
Science
1997
;
275
:
1790
–2.
49
Koch A, Denkhaus D, Albrecht S, Leuschner I, von Schweinitz D, Pietsch T. Childhood hepatoblastomas frequently carry a mutated degradation targeting box of the β-catenin gene.
Cancer Res
1999
;
59
:
269
–73.
50
Voeller HJ, Truica CI, Gelmann EP. β-Catenin mutations in human prostate cancer.
Cancer Res
1998
;
58
:
2520
–3.
51
Chesire DR, Ewing CM, Sauvageot J, Bova GS, Isaacs WB. Detection and analysis of β-catenin mutations in prostate cancer.
Prostate
2000
;
45
:
323
–34.
52
Schlosshauer PW, Pirog EC, Levine RL, Hedrick-Ellenson L. Mutational analysis of the CTNNB1 and APC genes in uterine endometrioid carcinoma.
Mod Pathol
2000
;
13
:
1066
–71.
53
Palacios J, Gamallo C. Mutations in the β-catenin gene (CTNNB1) in endometrioid ovarian carcinomas.
Cancer Res
1998
;
58
:
1344
–7.