Gene Expression Profiles of Luteal Phase Fallopian Tube Epithelium from BRCA Mutation Carriers Resemble High-Grade Serous Carcinoma

Purpose: To identify molecular alterations potentially involved in predisposition to adnexal serous carcinoma (SerCa) in the nonmalignant fallopian tube epithelium (FTE) of BRCA1/2 mutation carriers, given recent evidence implicating the distal FTE as a common source for SerCa.

Experimental Design: We obtained and compared gene expression profiles of laser capture microdissected nonmalignant distal FTE from 12 known BRCA1/2 mutation carriers (FTEb) and 12 control women (FTEn) during the luteal and follicular phase, as well as 13 high-grade tubal and ovarian SerCa.

Results: Gene expression profiles of tubal and ovarian SerCa specimens were indistinguishable by unsupervised cluster analysis and significance analysis of microarrays. FTEb samples as a group, and four individual FTEb samples from the luteal phase in particular, clustered closely with SerCa rather than normal control FTE. Differentially expressed genes from these four samples relative to other FTEb samples, as well as differentially expressed genes in all FTEb luteal samples relative to follicular samples, were mapped to the I2D protein-protein interaction database, revealing a complex network affecting signaling pathways previously implicated in tumorigenesis. Two candidates, disabled homolog 2 mitogen-responsive phosphoprotein (DAB2) and Ski-like (SKIL), were further validated by real-time reverse transcription–PCR and tissue arrays. FTEb luteal and SerCa samples expressed higher levels of oncogenic SKIL and decreased levels of tumor suppressor DAB2, relative to FTEb follicular samples.

Conclusions: These findings support a common molecular pathway for adnexal SerCa and implicate factors associated with the luteal phase in predisposition to ovarian cancer in BRCA mutation carriers.

High-grade serous carcinoma (SerCa) is the most common histologic type of both ovarian and fallopian tube carcinoma, presents at an advanced stage of disease, and has a high mortality rate. Approximately 15% of SerCa cases are considered hereditary, arising in association with a germline mutation in the BRCA1 or BRCA2 genes (1, 2), which confer an estimated 50% to 60% or 18% to 23% lifetime risk, respectively (3, 4).

Until recently, no reproducible histologic cancer precursor of high-grade SerCa had been identified (5–7), but the unexpected finding of occult SerCa in the fallopian tubes of BRCA mutation carriers undergoing prophylactic surgery has led to the description of putative cancer precursors in the distal fallopian tube epithelium (FTE; refs. 8–11). These discoveries have led investigators to postulate that the distal FTE is an alternative source of SerCa designated as of ovarian origin by pathologists (11–13). Comparison of SerCa and coexisting tubal intraepithelial carcinomas has shown identical mutations of the tumor suppressor p53 in each lesion, indicating that mutation of p53, a frequent event in SerCa, occurs before the development of invasive disease and supports the idea that they are causally related (14). Although it is not possible to definitively prove the tubal origin of tubal intraepithelial carcinomas, these are likely primary lesions given the previously observed resistance of tubal mucosa to direct implantation (14, 15). This concept is further supported by the finding that gene expression profiles of ovarian SerCa are more similar to normal FTE than the ovarian surface epithelium (16), as well as by the similar risk factors (17, 18) and genomic alterations of ovarian and fallopian tube cancers (19, 20).

To identify molecular alterations that may be involved in predisposition to SerCa in the nonmalignant FTE of BRCA1/2 mutation carriers, we generated and compared gene expression profiles of laser capture microdissected epithelial cells from the nonmalignant distal fallopian tube of women at a baseline risk for SerCa (FTEn) and known BRCA1/2 mutation carriers (FTEb), as well as from high-grade tubal and ovarian SerCa specimens (fallopian tube SerCa and ovarian SerCa, respectively). To our knowledge, this is the only study to date of gene expression profiles in the histologically normal FTE of BRCA1/2 mutation carriers and control patients.

Study samples. The study was approved by the University Health Network Research Ethics Board, and all patients provided informed consent. Snap-frozen tissues were selected from the University Health Network Ovarian Tissue Bank and Database, including histologically normal fallopian tubes from 12 BRCA mutation carriers (FTEb) and from 12 control women undergoing salpingo-oophorectomy for reasons other than adnexal malignancy or family history (FTEn). Consequently, these women had not been tested for the presence of a BRCA1/2 mutation. All women were premenopausal, and patients undergoing treatment for breast carcinoma, on hormonal therapy, or with a diagnosis of endometrial carcinoma were excluded. Although 3 of 12 mutation carriers had a history of breast carcinoma, they were not taking antiestrogenic therapy at the time of surgery, with an average time since last treatment of 45.7 mo [95% confidence interval (95% CI), 6.6-84.7]. Nonmalignant FTEn and FTEb specimens were age-matched, with an average age of 44.5 y (95% CI, 42.3-46.7) for control patients and 41.3 y (95% CI, 39.4-43.2) for mutation carriers. Each group of FTE included six samples obtained during the luteal phase and six from the follicular phase of the ovarian cycle, determined after review of histologic sections of ovaries and endometrium by a gynecologic pathologist (P.A.S.), to account for potential effects of the endocrine milieu on gene expression.

Thirteen SerCa samples designated as of either tubal (n = 6) or ovarian (n = 7) origin from 11 patients were included. Two independent samples were obtained from two of the patients for direct comparison of tubal and ovarian SerCa; one of these patients did not have a known family history or BRCA1/2 mutation. The remaining five ovarian SerCa patients had either a strong family history or identified mutation, whereas two of four remaining tubal SerCa patients fulfilled these criteria. The other two of four tubal SerCa patients did not have a documented family history or identified mutation, but were selected based on a similar age of cancer onset as the ovarian cases. Overall, tubal and ovarian specimens were age-matched, with an average age of 54.8 y (95% CI, 49.4-60.3) and 49.6 y (95% CI, 42.6-56.5), respectively. In addition, 12 of 13 of the SerCa specimens showed accumulation of p53 protein by immunohistochemistry (not shown). Relevant clinical information on all study patients is summarized in Table 1.

Laser capture microdissection and RNA extraction. To eliminate potential bias and minimize inclusion of stromal cells in all profiled cases, materials for gene expression analysis of FTEn, FTEb, and SerCa specimens were obtained using the same protocol of laser capture microdissection followed by linear amplification. Shortly before laser capture microdissection, 7-μm frozen sections were cut, adhered onto Superfrost slides (Fisher Scientific Company), and immediately stored on dry ice. Additional 5-μm sections were cut at the beginning and end of each sample and stained with H&E for review to ensure proper tissue orientation and histology. Epithelial cells were then obtained from all specimens using the Arcturus PixCell IIe Laser Capture Microdissection system (Arcturus) following manufacturer's instructions. Collected cells were immediately stored at −80°C before extraction of total RNA using a Stratagene Absolutely RNA MicroPrep kit (Stratagene). Sufficient quality and quantity of RNA was confirmed using an Agilent 2100 Bioanalyzer RNA 6000 Pico LabChip kit (Agilent Technologies, Inc.) and NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies), respectively, before official inclusion in the study.

Linear amplification and hybridization to GeneChip arrays. Ten nanograms of total RNA from study cases were reverse transcribed and linearly amplified using the NuGEN Biotin Ovation kit (NuGEN Technologies) in collaboration with the University Health Network Microarray Center. This method has previously been shown to provide sufficient material for global gene expression analysis from limited total RNA, while accurately and reproducibly maintaining differential transcript abundance in nonamplified samples (21). Briefly, first-strand cDNA synthesis was done using a unique DNA/RNA primer, followed by second strand cDNA synthesis and SPIA isothermal linear amplification. Amplified cDNA was then purified, enzymatically fragmented, and labeled with biotin. Quality and quantity of the purified labeled cDNA product were confirmed before hybridization to Affymetrix GeneChip U133A Plus 2.0 arrays (Affymetrix, Inc.) according to manufacturer's instructions.⁷

Amplification and hybridization of samples was done in six separate runs, with each group represented in each run (FTEn from luteal and follicular phase, FTEb from luteal and follicular phase and tubal and ovarian SerCa). Profiling data for all samples has been deposited in the National Center for Biotechnology Information Gene Expression Omnibus⁸ and are accessible through Gene Expression Omnibus Series accession number GSE10971.

Statistical analysis of gene expression profiles. All CEL files were imported into ArrayAssist (Stratagene), and a master table of all hybridizations was created using the robust multiarray average algorithm. Acceptable array performance was confirmed by review of generated RPT files. Robust multiarray average normalized data were then used as the basis for all subsequent clustering and statistical analysis. Unsupervised binary tree-structured vector quantization clustering was done using all SerCa samples. This method combines tree-structured vector quantization and partitive k-means clustering to separate samples into homogenous groups based on similar levels of gene expression (22). ArrayAssist was used to perform unsupervised hierarchical clustering of all SerCa cases following definition of tubal SerCa as baseline. Hierarchical clustering was subsequently done on all hybridizations following definition of FTEn samples as baseline, as well as on data grouped by major class (FTEn, FTEb, or SerCa) or subgroup (FTEn luteal, FTEn follicular, FTEb luteal, FTEb follicular, fallopian tube SerCa, ovarian SerCa) after defining FTEn and FTEn luteal as baseline, respectively.

Robust multiarray average normalized data were then imported into Microsoft Excel for identification of differentially expressed genes using significance analysis of microarrays (SAM version 3.0).⁹

For each comparison, two-class unpaired analysis was done using unlogged data, and the δ value was selected based on an acceptable corresponding median false discovery rate (FDR). To find overlapping significant probe sets, relevant significance lists were imported into ArrayAssist and compared using the Venn diagram function. Additional information for probe sets of interest, including gene ontology, was obtained using the batch query function on the NetAffx Analysis Center portion of the Affymetrix Web site.¹⁰

Identification of candidate genes using protein-protein interaction networks. To visualize the interactive networks of genes found significantly altered, as well as to aid in the selection of the most promising candidates for follow-up studies, a combination of bioinformatic tools was used. Lists of significant probe set IDs separated by analysis and direction of change in the group of interest were submitted to the I2D (Interlogous Interaction Database) Web site¹¹

(version 1.6) to obtain first-degree interaction information for the corresponding proteins (23). Next, known and predicted protein-protein interactions for each list were visualized using NAViGaTOR (Network Analysis, Visualization, and Graphing TORonto) software, version 1.1.0, to elucidate the connectivity of the significantly altered genes.¹² The extent of overlap of individual networks of interest was visualized using the Compare Networks (union) option. Proteins of extreme interest (p53 and BRCA1) were then located by Swiss Prot ID, and all noninteracting proteins were deleted.

Real-time quantitative reverse transcription–PCR. Three representative samples from each nonmalignant group (FTEn luteal, FTEn follicular, FTEb luteal, FTEb follicular) and six SerCa samples were selected for validation of gene expression profiles. Ten nanograms of RNA were linear amplified using the NuGEN Ovation RNA Amplification System (NuGEN Technologies). RNA was also isolated from SKOV3 ovarian cancer cells using TRIzol reagent (Invitrogen Corporation) for use as a calibrator sample. Contaminating DNA was removed (DNA-free kit, Ambion, Inc.), and first-strand cDNA was generated using SuperScript III Reverse Transcriptase and Oligo(dT)₂₀ primers (Invitrogen). Quality and quantity of all amplified cDNA were confirmed, and all samples were diluted to 1.6 ng/μL using sterile double distilled water.

Primer Express Software version 2.0 (Applied Biosystems) was used to select specific primers for human disabled homolog 2 mitogen-responsive phosphoprotein (DAB2), Ski-like (SKIL), and reference gene β-actin (ACTB) based on the sequence of the relevant probe set(s). All amplicons were situated within the first 1,500 bp from the poly(A) tail, given the use of oligo(dT)-primed reverse transcription of study and calibrator samples. Selected primer sequences included DAB2 forward 5′-CTAGCTATTGCAAATGAGGGAAG-3′, DAB2 reverse 5′-GGTAATACTACTTGAACCCAGGAGCA-3′, SKIL forward 5′AAAACTGTCCTCTACCTCACGTG-3′, SKIL reverse 5′-AAGCTCAGAGCACAGTATGTCATG-3′, ACTB forward 5′-GCATTGTTACAGGAAGTCCCTTG-3′, and ACTB reverse 5′-CTATCACCTCCCCTGTGTGGA-3′. End-point PCR was done for all primer sets, and the reaction product was sequence verified.

Real-time quantitative PCR was done using the QuantiTect SYBR Green PCR kit (Qiagen, Inc.) according to manufacturer's instructions. PCR was done using the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems), with use of the ROX internal reference dye. A dissociation reaction was done at the end of the PCR program to confirm amplification of a single product at the expected melting temperature, and the product was run on a 2.5% agarose gel to confirm correct size. All experiments included triplicate wells of each sample for both target and reference gene. The comparative C_T method for relative quantitation was done using ABI PRISM Sequence Detection Software version 2.0 (Applied Biosystems). Target gene C_T values were normalized to ACTB. Statistical analysis was done using one-way ANOVA followed by a least significant difference test for post hoc comparisons (P < 0.05).

Immunohistochemistry. A tissue microarray consisting of a larger subset of nonmalignant fallopian tube specimens from control patients and mutation carriers was constructed using the semiautomated TMArrayer (Pathology Devices, Inc.). Patient samples were identified according to history and diagnosis from the University Health Network Ovarian Tissue Bank and Database. Routine H&E sections were examined and a 1.5 mm² area of histologically normal FTE was selected before transfer of the corresponding formalin-fixed paraffin-embedded tissue core to the master block. The final array consisted of a total of 33 nonmalignant specimens from confirmed BRCA1/2 mutation carriers (including 11 luteal, 16 follicular, 3 early follicular, and 3 perimenopausal), 25 normal control specimens (including 11 luteal, 12 follicular, and 2 perimenopausal), and 2 cores each of normal kidney and liver for orientation purposes. All nonmalignant FTEn and FTEb cases used for gene expression profiling were included. Duplicate 5 μm sections of the array were included on each slide to control for potential variability during immunohistochemistry. In addition, a separate array containing duplicate 0.6 mm² cores obtained from 59 cases of high-grade SerCa was also used to assess the expression of candidates in both hereditary and sporadic carcinoma tissue specimens (average age, 61.8 y; 95% CI, 59.1-64.5). All carcinoma specimens were obtained from the Ovarian Tissue Bank and Database and were naive to chemotherapy.

Immunohistochemistry was done using standard procedures following microwave antigen retrieval using goat polyclonal anti-SKIL antibody [SnoN (K-20), Santa Cruz Biotechnology] at a dilution of 1 in 600, mouse monoclonal anti-DAB2 (disabled-2/p96 antibody, BD Biosciences) at a dilution of 1 in 50, or mouse monoclonal anti-cdc2 [cdc2 (P0H1), Cell Signaling Technology, Inc.] at a dilution of 1 in 50. Antibody concentrations were optimized on normal or malignant breast tissue sections before staining tissue array slides.

After immunohistochemistry, the ScanScope CS slide scanner (Aperio Technologies, Inc.) was used to create digital slide images at 20× magnification. Tissue cores were then visualized and manually scored using ImageScope software (Aperio Techologies), version 6.25. Scoring was independently done by two observers (A.A.T. and P.A.S.) blinded to patient information. Scores were given for percentage of epithelial cells stained (0, 0%; 1, 1-24%; 2, 25-50%; 3, >50%), as well as intensity of staining (0, negative; 1, light; 2, medium; 3, dark). The combined histoscore was obtained by adding these two individual scores, and nuclear and cytoplasmic staining was reviewed independently (24). Statistical analysis was done using ANOVA followed by least significant difference (P < 0.05). Comparison of unpaired proportions was by Fisher's exact test (P < 0.05).

High-grade ovarian SerCa and fallopian tube SerCa exhibit indistinguishable gene expression profiles. An initial critical question was whether SerCa designated as of ovarian origin, according to current pathology practice, is molecularly similar to that apparently derived from the fallopian tube. A high degree of similarity would support the idea that the FTE is the cell of origin for both tubal and ovarian SerCa. Unsupervised hierarchical (agglomerative) clustering of the individual SerCa samples after robust multiarray average normalization showed that the tubal and ovarian specimens had closely related global gene expression profiles (Fig. 1A). SerCa of presumed ovarian origin clustered together, with many of the tubal cancers clustering more closely with the ovarian cancers than with other tubal cancers. To examine this relationship more closely, an unsupervised partitive clustering method, binary tree-structured vector quantization, was used. This method iteratively partitioned the 13 cases by a k-means algorithm (k=2), considering the partitioning of all probe set responses by a self-organizing map algorithm. This method clearly revealed that samples partitioned together irrespective of presumed origin or known mutation status (Fig. 1B). Although similar, identical results were not obtained using the two clustering methods. This is likely due to the distinct approaches (agglomerative versus partitive), as well as the differential sensitivity of each method to data normalization (22). Two-class paired SAM analysis of the two tubal and ovarian cancers from the same patients revealed no differentially expressed genes at a minimum FDR of 40%, as did two-class unpaired analysis of the remaining carcinoma specimens at an FDR of 21%. Thus, cluster analysis, as well as SAM, indicates that SerCas have similar molecular profiles whether of presumed ovarian or tubal origin.

In contrast, comparison of all FTEn samples with all cancers revealed 4,263 probe sets with decreased expression and 7,311 probe sets with increased expression in the cancers at an FDR of 4.3%. Of the down-regulated probe sets, 2,235 (52%) were shared between the two malignancies, as were 3,287 (45%) of the up-regulated probe sets (Fig. 1C). Given the high degree of similarity between fallopian tube SerCa and ovarian SerCas, all subsequent comparisons to malignant samples were done combining the SerCas as a single group.

Differential expression of genes in a subset of histologically normal FTEb samples. Hierarchical clustering using all probe sets was done to determine the overall similarity of FTEb and SerCa samples. Analysis using grouped data revealed that FTEb samples, which were histologically indistinguishable from FTEn, clustered with SerCa samples rather than normal controls (Fig. 2A), demonstrating that gene expression altered in mutation carriers is consistent with changes that have occurred in malignant cells. These changes likely reflect and contribute to the overall increased risk for malignant transformation in mutation carriers, consistent with the documented roles of BRCA1 and BRCA2 proteins in maintaining genomic integrity through participation in DNA repair, transcriptional regulation, and cell cycle control (25–27).

Unsupervised clustering using individual data revealed that four BRCA1-mutated FTEb samples clustered with all of the carcinomas, indicating that global gene expression in this subset of samples most closely resembles that of malignant epithelial cells (Fig. 2B). Whereas it is not possible to determine if these cases would have been more likely to progress to SerCa had prophylactic surgery not occurred, it is feasible that these samples have acquired gene expression changes involved in serous carcinogenesis that have not yet resulted in atypical histology. SAM two-class unpaired analysis identified 288 probe sets with decreased expression and 598 with increased expression in these four samples relative to the remaining FTEb samples that were more similar to normal controls at an FDR of 4.3%. Gene ontology analysis revealed that many differentially expressed probe sets correspond to genes with known roles in BRCA1/2-dependent processes (transcriptional regulation, cell cycle control, ubiquitin cycle; refs. 25, 27), as well as others involved in tumor initiation and progression in general (apoptosis, cell adhesion, and cell motility; Fig. 2C; see Supplementary Table S1 for a complete list of significantly altered genes arranged by gene ontology). In addition, 181 of 288 (63%) of the probe sets with decreased expression and 172 of 598 (29%) with increased expression overlapped in the same direction as those differentially expressed between nonmalignant FTEn and SerCa samples as a group at the same FDR (4.3%), further supporting the idea that these may include some of the earliest events in serous carcinogenesis (Fig. 2D).

Differential effect of the ovarian cycle in BRCA1/2 mutation carriers. Given the unexpected finding that all four FTEb samples grouping with SerCa were obtained during the luteal phase, we further explored the association of the ovarian cycle with FTE gene expression. Hierarchical clustering of all samples using all probe sets was repeated using data grouped by both BRCA mutation and ovarian cycle status. The dendrogram obtained confirmed that FTEb samples from the luteal phase (FTEb luteal) more closely resembled SerCa samples compared with FTEb samples from the follicular phase (FTEb follicular; Fig. 3A).

Ovarian steroids are known to affect the growth and differentiation of the FTE. Estrogens act during the follicular phase to promote ciliogenesis and secretory cell hypertrophy, whereas progesterone, which is predominant during the luteal phase, induces dedifferentiation (28, 29). However, these hormonal influences are not widespread within the tube and can be quite variable among individuals; thus, both ciliated and secretory cells are present in the FTE throughout the cycle. Interestingly, SAM two-class unpaired analysis revealed that the number of differentially expressed probe sets between the luteal and follicular phases was far greater in mutation carriers than in normal controls. At an FDR of 9.8%, only 21 probe sets were found to be decreased in the luteal relative to the follicular phase in normal control patients (Fig. 3B). In contrast, 322 probe sets were decreased and 70 were increased in the luteal relative to follicular phase in FTEb at the same FDR, suggesting that hormonal influences on the FTE may be altered as a result of reduced functional BRCA levels. Accordingly, the average BRCA1 level in FTEb luteal samples was 50% lower than that seen in FTEb follicular samples, and an 80% decrease was observed in the four FTEb luteal samples clustering with SerCa relative to the remaining FTEb specimens (not shown). Previous studies have implicated both BRCA1 and BRCA2 in the regulation of gonadal steroid hormone receptor activity (30, 31), and this regulation may contribute to the differences in gene expression between control patients and BRCA1/2 mutation carriers.

The 389 probe sets specific to mutation carriers were subjected to gene ontology analysis and were found to correspond to BRCA1/2-dependent and tumorigenesis-involved processes similar to that observed in the previous analysis (Fig. 3C; see Supplementary Table S2 for a complete list of significantly altered genes arranged by gene ontology). Interestingly, 114 of 319 of the probe sets with decreased expression (36%) and 26 of 70 with increased expression (37%) in FTEb luteal samples overlapped in the same direction with probe sets differentially expressed between FTEn and SerCa samples at a similar FDR (8.3%; Fig. 3D). This is in contrast to an overlap of 9 of 70 with decreased expression (13%; P = 0.0001) and 59 of 319 with increased expression (18%; P = 0.001) in FTEb follicular samples. That the profiles obtained during the luteal phase of mutation carriers most closely resembles the gene expression pattern of SerCa raises the possibility that the hormonal milieu associated with the luteal phase may contribute to cancer predisposition in some individuals, such as BRCA1/2 mutation carriers.

A primary hormonal difference between the follicular and luteal phases is the elevation of circulating progesterone during the luteal phase. Studies have frequently suggested that progesterone is a protective rather than exacerbating factor in ovarian cancer development (32, 33). Epidemiologic data generally show that women with conditions in which progesterone is elevated have a lower incidence of ovarian cancer. For instance, the protective effect of increased parity and twin pregnancies has been attributed to the greatly increased level of maternal circulating progesterone (17, 18, 32, 34). In addition, many studies have linked the use of combination oral contraceptives to a decreased ovarian cancer risk in mutation carriers and the general population. A few studies have further suggested that formulations with high progestin potency offer greater protection than those with low progestin potency (30, 32, 35, 36). Finally, treatment of normal ovarian surface epithelium and ovarian cancer cells with progesterone has a growth inhibitory effect (32, 34). Accordingly, some authors have suggested that exposure to progesterone may eliminate cells genetically damaged by incessant ovulatory events through induction of apoptosis, thus providing an exfoliation effect (32, 34). Evidence has emerged, however, that the effects of progesterone may be more complex than previously appreciated. For instance, whereas high concentrations of progesterone suppressed tumorigenesis in nude mice inoculated with ovarian cancer cells (37), lower concentrations increased the proliferative capacity of ovarian surface epithelium and ovarian cancer cells in culture (38). This is consistent with the postulate that luteal-phase levels of progesterone have a growth promotional effect, whereas the higher levels achieved during pregnancy or oral contraceptive use induce cell cycle arrest or apoptosis of epithelial cells (32). Similar studies have not yet been reported for FTE cells. However, it is possible that exposure to luteal-phase levels of progesterone may not be protective against malignant transformation of the epithelial cells of the distal fallopian tube, similar to the effect observed in the breast (39–41), particularly in cells in which the DNA repair pathways may be compromised. It is of course possible that other hormonal changes associated with the luteal phase may affect gene expression in the FTE, and further studies are required using FTE cells in culture. Finally, while all nonmalignant cases in this study have been age-matched, there is evidence that mutation carriers initiate menopause at an earlier age than normal controls (42), suggesting that the nature of the hormonal milieu of the luteal phase may be different in the two populations.

Identification of protein-protein interaction networks potentially involved in initiation of SerCa. To select genes for validation and further investigation, we queried the I2D protein interaction database with our set of differentially expressed genes from both SAM analyses. Of the 886 probe sets differentially expressed in the four FTEb sample grouping with SerCa compared with the remaining FTEb, 417 (47%) mapped to I2D, as well as 218 of 389 (56%) of the probe sets differentially expressed in FTEb luteal compared with FTEb follicular samples. Because of the known involvement of p53 in high-grade SerCa, we focused our attention on differentially expressed genes encoding for proteins predicted to directly or indirectly interact with BRCA1/2 or p53 proteins. Because mutation of BRCA1 or BRCA2 leads to activation of the damage response pathway, loss of this pathway by somatic mutation of p53 and presumably other members is required for malignant transformation (43). Consequently, accumulation of p53 protein is frequently observed in the histologically normal FTE of mutation carriers and control patients, and decreased expression of cell cycle arrest genes p21 and p27 have been observed in prophylactic tubal specimens relative to normal controls (8, 13). Supplementary Figs. S1 and S2 show the networks obtained through I2D visualized with NAViGaTOR software. One gene found in networks of both comparisons, DAB2, was found to be decreased in FTEb luteal and SerCa samples. DAB2 is an adaptor molecule that exerts its tumor suppressive function largely through its role in facilitating transforming growth factor-β (TGF-β)–induced growth inhibition (44), a pathway that must be circumvented in early stages of epithelial carcinogenesis and which has previously been shown to be dysfunctional in ovarian cancer (45). In addition, we selected SKIL, which was up-regulated in FTEb luteal samples and acts to inhibit Smad2 and Smad3 (46), targets that are enhanced by DAB2 (Fig. 4). The implication of expression changes in the TGF-β pathway regulatory genes DAB2 and SKIL in FTEb samples during the luteal phase is highlighted by the observation that 15% of the significantly altered probe sets found by both SAM analyses (185 of 1,201 unique probe sets) correspond to genes known to be downstream TGF-β targets (47–50), consistent with an altered TGF-β response (see Supplementary Tables S1 and S2). Such genes include DAB2 (47, 49) and SKIL (48, 49) themselves, in addition to well-established TGF-β targets ID2, ETS2, and PIM1 (50).

Differential expression of SKIL and DAB2 in FTEb samples during the luteal phase. In agreement with our Affymetrix data, FTEb luteal samples were found to express higher levels of SKIL mRNA compared with FTEb follicular samples as determined by real-time quantitative reverse transcription–PCR, although this did not reach statistical significance (Fig. 5A, left). A further increase in SKIL mRNA was observed in SerCa samples compared with FTEb follicular (P < 0.05). Immunohistochemistry on tissue microarrays revealed a similar average intensity of nuclear staining in FTEb luteal and SerCa samples, as well as a trend toward increased staining in FTEb luteal compared with FTEb follicular samples (Fig. 5A, right). Representative images of SKIL immunohistochemical staining are shown in Fig. 5B.

Also consistent with our profiling data, FTEb luteal samples expressed a level of DAB2 mRNA similar to SerCa samples and a greatly reduced expression compared with FTEb follicular samples (P < 0.05; Fig. 5C, left). Immunohistochemistry for DAB2 on tissue microarrays revealed that FTEb luteal and SerCa samples express a decreased level of cytoplasmic DAB2 protein compared with FTEb follicular samples (P < 0.05; Fig. 5C, right). A lower proportion of FTEb luteal (one of nine, 11%) and SerCa (1 of 48, 2%) samples had a histoscore of >4 compared with FTEb follicular specimens (13 of 16, 81%). The decrease in DAB2 during the luteal phase could partly be explained by altered cdc2 protein levels. Previous studies have indicated that functional BRCA1 indirectly inhibits cdc2 kinase activity (51), which then negatively regulates DAB2 through cell cycle–specific phosphorylation (52). Absence of functional BRCA proteins would therefore lead to accumulation of active cdc2 and subsequent reduction of DAB2. Accordingly, in direct opposition to DAB2, there was a trend toward an increased nuclear histoscore for cdc2 in FTEb luteal compared with FTEb follicular samples, and a similar proportion of FTEb luteal (4 of 11, 36%) and SerCa (18 of 51, 35%) samples showed intense nuclear staining. Representative images of DAB2 and cdc2 immunohistochemical staining are shown in Fig. 5D. Most intriguingly, whereas FTEb luteal and SerCa samples showed an overall loss of cytoplasmic DAB2 staining, FTEb follicular samples typically showed retention of staining in the secretory cells of the mucosal lining resulting in an increased overall intensity and histoscore. The specific loss of DAB2 expression in secretory cells during the luteal phase may explain the modest difference in mRNA and protein expression between luteal and follicular samples when including both secretory and ciliated cell populations and further emphasizes the importance of this cell type in serous carcinogenesis (9, 15).

Many studies have observed a similar loss of DAB2 in ovarian carcinomas compared with normal ovarian surface epithelium (53–55), although this has not previously been found in the histologically normal ovarian surface epithelium from BRCA1/2 mutation carriers. Interestingly, mice heterozygous for DAB2 frequently develop epithelial dysplasia on the ovarian surface but not malignant ovarian tumors, suggesting that loss of DAB2 is necessary but not sufficient for ovarian carcinoma development (53); however, the oviductal epithelium was not examined. Whereas functional DAB2 exerts its tumor-suppressive effect by mediating TGF-β–induced growth inhibition through transmission of signals from the TGF-β receptors to the Smad transcriptional activators (44), this is opposed by SKIL, which directly represses Smad gene transcription and enhances cytoplasmic sequestration of active Smad protein complexes (46, 56, 57). Although SKIL has not previously been implicated in ovarian cancer development, it lies within a chromosomal region (3q26) previously found to be amplified in a majority of serous fallopian tube and ovarian carcinomas by comparative genomic hybridization (19). SKIL protein is highly elevated in many human cancer cell lines, including ovarian, and has been found to promote tumor growth in nude mice (58). High expression of SKIL protein is also a negative prognostic factor in ER+ breast carcinomas (56). It is therefore likely that the combined effect of decreased DAB2 and increased SKIL in FTEb would promote malignant transformation through potent suppression of TGF-β–induced cell cycle arrest and apoptosis.

In conclusion, gene expression changes potentially involved in the earliest events of tubal and ovarian SerCa have been identified in histologically normal FTE from BRCA1/2 mutation carriers. These expression changes seem to be influenced by reproductive hormones, with components of the luteal phase inducing changes similar to those observed in SerCa specimens. Increased expression of SKIL, coupled with decreased expression of DAB2 in mutation carriers during this phase, could represent some of the earliest initiating or predisposing events of SerCa. Finally, specific loss of DAB2 in the secretory cells of the tubal epithelium during the luteal phase further highlights the relevance of this cell type in SerCa development.

No potential conflicts of interest were disclosed.

We thank Dr. Susan Done and Dr. Igor Jurisica for their critical comments, the University Health Network Biobank for facilitating this research, Kelvin So for his assistance in immunohistochemistry, Neil Winegarden for his guidance in microarray experiments, and the Jurisica group for their support in network analysis of gene expression data.