Loss of Cellular Retinol-binding Protein 1 Gene Expression in Microdissected Human Ovarian Cancer¹

Ovarian cancer is the most frequently fatal gynecological malignancy with an estimated 23,300 new cases and 13,900 deaths in the United States in 2002 (1). The epithelial ovarian carcinomas, which make up 90% of all human ovarian malignant tumors, arise in the surface epithelium, or serosa, of the ovary (2). Clinically, about two-thirds of serous neoplasms of the ovary present de novo as advanced-stage tumors, reflecting their propensity for intra-abdominal/peritoneal spread (3). Recent statistics suggest some improvement in mortality rate from ovarian cancer in the United States (from 10.4 cases/100,000 women in 1969 to 8.8 cases/100,000 women in 1999; Ref. 1). This change is modest and certainly not satisfying. The ovarian cancer mortality has not significantly improved in the past three decades because of our poor understanding of the underlying biology, lack of reliable biomarkers for disease detection, late stage of presentation, and inaccessible location of the ovary. There are still few appropriate animal models for the study of human ovarian tumors, and the methodology to culture HOSE³ has become available only recently (4, 5, 6).

Retinoids, metabolites of vitamin A (retinol), play an important role in fundamental physiological processes such as vision, reproduction, hematopoiesis, and differentiation of epithelial tissues. They have been shown to prevent mammary carcinogenesis in rodents (7), inhibit the growth of human cancer cells in vitro (8, 9), and be effective chemopreventive and chemotherapeutic agents in ovarian and in a variety of other human epithelial and hematopoietic tumors (10, 11).

In the cytoplasm, retinol and retinoic acid are bound to CRBPs (types 1, 2, and 3) and to cellular retinoic acid-binding proteins (types 1 and 2), highly conserved among mammals (12). CRBP1 (or CRBP) is widely distributed throughout the body, whereas CRBP2 is restricted to the small intestine and CRBP3 to the hearth and skeletal muscle (13). This specific tissue distribution defines different functions between CRBP types. CRBP1, a M_r 15,000 cytosolic protein, is essential for vitamin A homeostasis. It serves as a chaperone that protects retinol from the cellular milieu and interacts with various retinoid-metabolizing enzymes (14). Importance of this gene product in vivo is documented in CRBP1 knockout mice (15). When fed a vitamin A-deficient diet, these animals fully exhaust their liver retinyl ester stores within 5 months and develop abnormalities characteristic of postnatal hypovitaminosis A, including squamous keratinizing metaplasia.

We have previously described an in vitro ROSE cell transformation model that may mimic ovarian cancer initiation in women (16). In this model, two genes whose products are involved in vitamin A transport and metabolism, RBP and CRBP, were found to be down-regulated (17). In an initial effort to translate these animal model data to human disease, we have also examined several human ovarian cancer cell lines for the expression of CRBP1 and determined that it is often not expressed (17).

Kuppumbatti et al. (18) have recently reported the association of human breast carcinogenesis with the loss of CRBP1 expression in roughly one fourth of breast cancer cases.

In this study, we have investigated clinical relevance of our previous findings in the rat model and human ovarian cancer cells by evaluating CRBP1 expression in a series of fresh frozen ovarian cancer samples by real-time quantitative PCR and in paraffin sections by immunohistochemistry. We were interested whether CRBP1 expression is lost early in the course of ovarian cancer, i.e., occurs in stage I as frequently as in later stages, and if it can be correlated with the degrees of tumor differentiation. We hypothesize that the loss of CRBP1 gene expression is a contributing factor in the development of ovarian carcinoma via alteration of retinol metabolism.

Cultures of HOSE cells were established by scraping the surface of grossly and microscopically confirmed normal ovaries. These specimens were obtained from women who had surgery for nonmalignant gynecologic diseases and were referred through Fox Chase Cancer Center Family Risk Assessment Program. No BRCA1/2 gene mutations were detected among these individuals (19). HOSE cells were isolated and maintained in Medium 199 and Medium MCDB105 (Sigma, St. Luis, MO), mixed 1:1, and supplemented with 5% fetal bovine serum and 0.2 units/ml insulin as described previously (4, 6).

All experimental protocols involving usage of human normal and tumor tissues were examined and approved by the Fox Chase Cancer Center Institutional Review Board. Ovarian tissue specimens were collected from the Gynecologic Oncology Group Ovarian Tissue Bank and Fox Chase Cancer Center Ovarian Tumor Bank. Specimens were obtained from 48 patients who underwent cytoreductive surgery for primary cancer of the ovary. According to the FIGO stage grouping for primary carcinoma of the ovary, 12 tumors were stage I, 6 were stage II, 29 were stage III, and 1 was stage IV. On the basis of the WHO criteria of histological classification, all of the tumors were identified as serous papillary carcinomas. The tumor grade grouping was as follows: 4 cases were borderline (grade 0); 4 well differentiated (grade 1); 14 moderately differentiated (grade 2); and 26 poorly differentiated (grade 3). The age of the patients ranged from 28 to 82 years, with a mean age of 58.6 years. The characteristics of examined ovarian cancer specimens are summarized in Table 1. Total RNA was isolated from whole ovarian tumor tissues, as well as HOSE cells, by homogenization of the cell pellets or minced tissues with QIAshredder Kit, followed by extraction with the RNeasy Mini Kit (Qiagen, Valencia, CA).

Frozen ovarian cancer tissue samples were embedded in OCT medium, cut in a cryostat at 5-μm thickness, and mounted on nonadhesive glass slides. Fixation was performed in 70% ethanol for 30 s. As an example and shown in Fig. 1, invasive ovarian cancer is visualized by H&E staining and then subjected to LCM. H&E-stained frozen sections were dehydrated for 5 s in 70, 90, and 100% ethanol with a final 5-min dehydration step in xylene. Air-dried sections were laser capture microdissected by a PixCell II LCM system (Arcturus Engineering, Mountain View, CA). The ovarian malignant epithelial cells to be selectively microdissected away from stroma were identified and targeted through a microscope, and a 15-μm laser beam pulse activated the film on a CapSure LCM Cap (Arcturus Engineering). In each case, ∼1 × 10⁴ carcinoma cells were captured. LCM cells were pooled from multiple caps, which were stored on dry ice until dissection was complete. LCM was performed for no longer than one-half h for each procurement. On the basis of careful review of the histological sections, each microdissection is estimated to contain 90% of desired cells.

After microdissection of each sample, 100 μl of guanidinium isothiocyanate-containing lysis buffer with 0.7 μl β-mercaptoethanol was applied directly to the microdissected cells adhered on the CapSure LCM cap, placed into a 0.5-ml microfuge tube, and vortexed vigorously. Total RNAs were extracted using the Strata Prep Total RNA Microprep Kit (Stratagene, LaJolla, CA). A DNase treatment was performed according to the manufacturer’s recommendations. The RNA was resuspended in 30 μl of RNase-free water. Two-thirds (20 μl) of total RNA from each LCM sample were reverse transcribed in a 40-μl reaction also containing 25 μg/ml oligo (dT)₁₅ primer (Promega, Madison, WI), 100 μm deoxynucleoside triphosphate mix, 1× RT (reverse transcriptase) buffer, and 200 units of Moloney murine leukemia virus reverse transcriptase (GenHunter, Nashville, TN). The reaction conditions were 5 min at 65°C, 1 h at 37°C, and 10 min at 70°C. Quantitative real-time PCR analysis was performed in the Smart Cycler apparatus (Cepheid, Sunnyvale, CA) to simultaneously amplify CRBP1 and the housekeeping gene GAPDH. GAPDH is our endogenous reference to control for differences in RNA extraction and cDNA synthesis. Real-time PCR is defined as ability to monitor or visualize accumulation of PCR products using fluorescence. Real-time PCR products were detected from 10 μl of microdissected tumor cDNA; 25 ng of HOSE cell cDNA, or 100 ng of bulk tumor cDNA in a final volume of 25 μl, also containing 0.2 μm primer, 0.3 μm probe, as well as 1× reaction mix, including 1.5 units of TaqDNA polymerase, 20 mm Tris-HC1 (pH 9.0), 50 mm KCl, 1.5 mm MgCl₂, and 200 μm of each deoxynucleoside triphosphate and stabilizers (Ready-To-Go PCR Beads; Amersham Pharmacia Biotech, Piscataway, NJ). Applied Biosystems (Foster City, CA) predeveloped GAPDH PCR primers and fluorescently labeled probe. We designed gene specific primers for CRBP1 detection to amplify a 196-bp portion of the cDNA. This sequence was derived from exons two and three of the CRBP1 gene, hence, genomic DNA would not contribute to the PCR product. The forward primer was 5′-CCAGACAAAGAGATCGTGCAG-3′ and the reverse 5′-ACACACTGGAGCTTGTCTCCG-3′. The sequence of the fluorigenic Blackhole1 probe, located 5 bp from the 3′ end of the forward primer, was 5′-FAMTGACCATATGATCATCCGCACGCBHQ1–3′. The PCR cycling conditions were performed for all of the samples as follows: 60 s at 95°C for initial denaturation; and 55 cycles for the melting (95°C, 15 s), annealing (55°C, 30 s), and extension (60°C, 45 s) steps. PCR reactions for each template were done in duplicate. Initially, each PCR run included a five-point standard to generate standard curves for CRBP1 and GAPDH, plus a no template control. Quantitative aspects of real-time PCR detection were based on C_t, the value at which the amplification enters into the log-linear phase of the PCR growth curve. All of the experiments were optimized such that the threshold cycle (C_t) from duplicate reactions did not span more than two cycle numbers.

We chose to compare tumor CRBP1 mRNA expression to HOSE cell expression. Differences in gene expression between ovarian tumor samples were calculated using the following formula (17):

To ascertain that different HOSE cell lines express CRBP1 uniformly, three HOSE cell lines (HOSE 5044, 5047, and 5057) were analyzed for CRBP1 expression.

A representative set of formalin-fixed, paraffin-embedded tissue sections from 15 ovarian tumors and five normal ovaries was reacted with polyclonal affinity-purified rabbit antibody against CRBP1. CRBP1 antibody was a gift from Dr. Ulf Eriksson and was generated as previously described using peptide GKEFEEDLTGIDDRKC that corresponded to residues 68–83 of human CRBP1 (20, 21). The paraffin blocks used for immunohistochemistry corresponded to the frozen tumor tissues analyzed by real-time PCR. The tissue sections were deparaffinized and hydrated through xylenes and graded alcohol series. After rinsing in PBS solution and blocking in 3% hydrogen peroxide for 20 min, the sections were washed in PBS, and nonspecific binding was blocked for 30 min in normal 10% goat serum (Biogenex, San Ramon, CA). This was followed by 15 min of avidin block and 15 min of biotin block. In between all of the steps, slides were washed with PBS. No antigen retrieval method was applied before immunostaining. Tissue sections were then incubated with 0.4 μg/ml of the CRBP1 antibody overnight at 4°C. In negative controls, primary antibody was replaced by PBS. After washing, the antigen was detected with the Super Sensitive Detection Kit (Biogenex), which uses horseradish peroxidase as conjugated enzyme and 3,3′-diaminobenzidine as chromogen. The sections were incubated for 30 min with the secondary biotinylated antibody (goat antirabbit IgG) and for 30 min with streptavidin-horseradish peroxidase reagent, finally with 3,3′-diaminobenzidine for 4 min, and washed with deionized water. Gill’s hematoxylin was used for counterstaining. Stained sections were mounted with aqueous medium (Dako, Carpenteria, CA) and covered with glass slips. All of the incubations were performed in humidified chamber. Two independent observers who were blinded to the stage and grade carried out the analysis of CRBP1 staining in ovarian tumor specimens. The intensity of immunostaining was graded based on a visual assessment of the intensity of brown reaction product as described previously (21). Slides were examined by light microscopy using an Olympus BX50 microscope. Photographic images were produced using a Roper Color CoolSNAP camera.

CRBP1 gene expression was dichotomized as detectable and undetectable. Fisher’s exact test and the exact test of Cochran-Armitage trend test were used to study whether the presence of detectable levels of CRBP1 was associated with the tumor stage, grade, or patient age.

To validate and extend our previous findings of genes differentially expressed in ovarian cancer (17), we chose real-time PCR, a highly reproducible and sensitive technique. This method allows for quantitative comparisons of message levels in tissue samples. Because real-time PCR does not require large amounts of starting RNA, we decided to analyze CRBP1 gene expression in a cohort of fresh frozen ovarian cancers of serous histological subtype. Serous tumors were chosen because they comprise ∼80% of all epithelial ovarian cancers.

To evaluate CRBP1 gene expression, we defined real-time PCR protocols to minimize C_t value variance, produce the best linearity with respect to input RNA (1 pg to 1 μg), and allow for GAPDH/CRBP1 multiplexing. Using a predeveloped GAPDH real-time PCR primer/probe set, we determined that the manual selection of C_t resulted in a lower intra-assay variance than the Cepheid’s software-selected value (data not shown).

In an initial experiment on whole ovarian tumor tissue, we have found by real-time PCR that the CRBP1 gene had uniformly higher expression in all four tumors, compared with HOSE cells (Table 2). We reasoned that the amplification of CRBP1 mRNA from stromal ovarian tissue adjacent to the malignant epithelium contributed to these findings. Because the heterogeneous nature of tissue samples often makes interpretation of the results difficult, we decided to use RNA prepared from laser-captured tumor cells instead of bulk tumor tissue RNA. When the same set of ovarian tumors was microdissected and subjected to real-time PCR, significant variation in CRBP1 mRNA expression of the malignant epithelium was found (Table 3). One specimen had nondetectable gene transcript, and in the other three, expression ranged from 14-fold less to 1771-fold greater than expression seen in primary HOSE cells. These findings confirmed the necessity for LCM for our additional analysis of gene expression. Microdissection ensured isolation of homogenous populations of ovarian cancer cells from clinical tissue specimens for real-time PCR. We reasoned that this approach would allow an accurate determination of CRBP1 gene expression.

We have normalized our results to expression in HOSE obtained from scrapings from normal ovaries. Although it has been confirmed that these samples are comprised mainly of epithelial cells (19), we wanted to minimize the possibility of stromal contamination and possible variations in CRBP1 expression levels among individuals. Therefore, we analyzed three independent HOSE cell lines (HOSE 5044, 5047, and 5057) for CRBP1 expression and found no statistical difference.

We extended our analysis to 48 human serous ovarian tumor samples of different FIGO stages and grades of differentiation. Table 4 summarizes these results. In 17 of 48 cases (35%), there was no detectable CRBP1 mRNA expression by quantitative PCR. In the remaining 31 specimens, the expression varied from 14-fold less to 2826-fold greater than expression seen in HOSE cells. When the percentages of CRBP1-negative specimens between tumor subsets of stage and grade were evaluated by Cochran-Armitage trend test and between patient age subsets by Fisher’s exact test, no statistically significant differences were found (P = 0.8746, P = 0.7384, and P = 0.2471, respectively). Therefore, the loss of CRBP1 expression in microdissected ovarian tumors was not related to tumor stage, grade, or patient age. Although borderline tumors have genetic alterations distinct from invasive ovarian cancers, four cases were included in our cohort.

Table 5 shows the comparison of CRBP1-negative tumors among the combinations of stage subsets (I+II, I+II+III, II+II+IV, and III+IV), whereas Table 6 shows the comparison of CRBP1-negative tumors among the combinations of grade subsets (0 + 1, 0 + 1 + 2, 1 + 2 + 3, and 2 + 3). Again, there was no statistically significant difference in the loss of CRBP1 when pooled tumor stage or grade subsets were compared among each other. Ps reported were two-sided and not adjusted for multiple comparisons.

Next, we examined the expression of CRBP1 protein by immunohistochemistry. A representative set of serous ovarian carcinomas was selected. The formalin-fixed, paraffin-embedded tumor blocks were sectioned and analyzed with H&E staining for histological diagnosis and by immunohistochemical staining for CRBP1. The stroma in tumor samples showed macrophages and fibroblasts staining, serving as an internal positive control. Examples of the typical staining patterns are shown in Fig. 2. The immunohistochemical analysis verified the quantitative PCR results of CRBP1 expression in ovarian cancer. The microdissected tumors that had lost CRBP1 by real-time PCR, showed no CRBP1 immunostaining in the malignant epithelium (Fig. 2,A). On the other hand, tumors that had expressed CRBP1 by real-time PCR, stained positive for CRBP1 (Fig. 2,B). Fig. 2,A shows the absence of CRBP1 staining in tumor cells in this serous papillary ovarian carcinoma. Only macrophage/fibroblast cells in the adjacent stroma are stained brown. Invasive ovarian tumor in Fig. 2 B is an example of positive CRBP1 staining in epithelial tumor cells, as well in stroma.

In Fig. 2 C, surface epithelium of the normal ovary stained positive for CRBP1, confirming the real-time PCR data on HOSE cell line.

We have previously developed and applied the consolidative suppression subtractive hybridization technique to identify genes differentially expressed in normal ROSE cells and their transformed counterparts. Northern blot analysis using 14 of 28 nonredundant cDNA fragments from this difference library showed that the mRNA transcripts were present in normal ROSE cells but lost or markedly reduced in four related transformed cell lines (17). The two genes down-regulated in this ovarian cancer model, RBP and CRBP1, were chosen to be studied because of their involvement in retinol metabolism and transport. As suggested by Kuppumbatti et al. (18), changes in vitamin A metabolism could be an early event in breast tumorigenesis. To ascertain if the observation in our rat ovarian cancer model had potential relevance to human disease, we investigated the expression of RBP and CRBP1 in human ovarian cancer cell lines by Northern blot analysis, qualitative PCR, and real-time PCR (Ref. 17; Figs. 3 and 4, Table 2]. The expression of RBPs was lost or markedly reduced relative to expression in HOSE cells. On the basis of the more established role of CRBP1 in cellular vitamin A metabolism compared with RBP, we have focused our additional studies on this gene.

Hence, in the present study, we attempted to more comprehensively examine CRBP1 expression in ovarian cancer patient samples using real-time PCR, a highly sensitive and reproducible technique. In addition, we have used microdissected specimens to maximize the proportion of tumor cells in the samples under investigation. The combination of real-time PCR and microdissected clinical tissue specimens allowed a highly quantitative study of our gene of interest in a relatively large number of clinical ovarian tumors.

Here we show that CRBP1 is lost in roughly one-third of ovarian cancer cases. Our results are in good agreement with those of a previous in situ hybridization study of CRBP1 expression in human breast cancer (18). Twenty-four percent CRBP1 loss rate in breast tumors was observed in a subset of cells already lacking retinoic acid receptor β expression. The authors hypothesized that compound loss of CRBP1 and retinoic acid receptor β hinders the bioactivity of endogenous vitamin A more extensively than the single loss of either gene. We consider the rate of CRBP1 expression loss in our study to be comparable with loss of other genes in ovarian cancer (22).

The most important prognostic factor for ovarian cancer is tumor stage or extent of disease at diagnosis (2). We reasoned that if stage I disease is a precursor of advanced ovarian cancer, examining clinical tumor specimens for alteration in gene expression and identifying molecular differences between early- and late-stage ovarian carcinomas should provide insight into developing strategies for early detection. In our study, 18 early-stage (stage I/II) and 30 advanced stage tumors (stage III/IV) were analyzed. The loss of CRBP1 gene expression by real-time PCR was as frequent in stage I ovarian cancer as in stages II, III, and IV, suggesting that it is an early event in ovarian carcinogenesis and is not associated with tumor stage. Likewise, there was no statistically significant difference in the percentage of CRBP1-negative specimens across grades of tumor differentiation. Although borderline tumors are believed to be genetically distinct from invasive cancers of the ovary, four borderline cases were included in our experimental cohort.

Because of the complexity of vitamin A metabolism, the precise role of CRBP1 in retinoid signaling remains controversial, despite numerous studies conducted over the past three decades on its binding properties, three-dimensional structure, tissue localization, regulation of expression, involvement in in vitro retinal metabolism, and null mutation in mouse (12, 15, 23, 24, 25). The function of the CRBP1 gene in controlling the availability of vitamin A to cells suggests that its product has special relevance to inhibition of early steps in transformation. Nevertheless, in human cancer, the presence and role of the specific binding proteins for retinol and retinoic acid have not been extensively investigated. In a case-controlled CRBP1 study in gynecological cancers, significant differences were found between the concentrations of CRBP1 in the dysplastic lesions of the uterine cervix and the concentrations in the normal cervix of the control subjects (26). As the histopathological severity of the dysplasia increased, decreasing amounts of CRBP1 were detectable. In some instances, CRBP1 was absent. When a spectrum of human cancers was screened for CRBP1 in tissue homogenates by high-pressure liquid chromatography with a reverse-phase elution system, CRBP was detected at a reduced level in the carcinoma of endometrium, ovary, and breast compared with normal tissue aliquots. In the ovarian carcinomas, the mean CRBP1 value was one-half of that found in normal tissue (27).

As stated earlier, retinoids, vitamin A and its derivatives and metabolites, can act as cancer chemopreventive and/or chemotherapeutic agents. With regard to their potential therapeutic benefit in ovarian cancer, it seems counterintuitive that use of agents predicted to have their influence in early steps of carcinogenic process would have any benefit against established invasive tumors. On the basis of this concern, we believe that the likely benefit might be in prevention rather than active therapy. There is experimental data indicating the following effects of retinoids in ovarian cancer cell lines: growth inhibition and promotion of cellular differentiation (28); increased induction of cytokeratins (29, 30); and increased apoptosis (31). When administered i.p., synthetic retinoid fenretinide causes a significant increase in the survival time of nude mice transplanted with a human epithelial ovarian adenocarcinoma line (32). A Phase III clinical trial of fenretinide for prevention of secondary breast cancers in 2972 women suggested the ability of retinoids to protect against the development of ovarian carcinoma (33, 34). Significantly lower incidence of ovarian tumors was demonstrated during the intervention period, but the protective effect of fenretinide seemed to cease after discontinuation of drug administration. All of the described studies have helped define the basis for an ongoing clinical trial by the Gynecologic Oncology Group. Gynecologic Oncology Group 190 trial examines the tissue effects of 4–6 months preoperative fenretinide administration in women undergoing prophylactic oophorectomy because of high familial risk of ovarian cancer (35).

In our current study, we have demonstrated that CRBP1 is frequently lost in human ovarian cancer. This provides validation of our earlier report showing CRBP1 underexpression in rat in vitro transformation model and in human ovarian cancer cell lines. The frequency of CRBP1 loss that was observed by real-time PCR in fresh microdissected specimens corresponds to the frequency observed in vitro. Three (33.3%) of nine human ovarian cancer cell lines were CRBP1 negative by Northern blot analysis; four (44.4%) of nine cell lines were CRBP1 negative by qualitative PCR; and two (33.3%) of six were negative for CRBP1 by real-time PCR (17).

We verified the results of quantitative PCR analysis of CRBP1 expression in ovarian cancer by immunohistochemistry. This is the first CRBP1 immunolocalization analysis in human malignant tumors of the ovary. Others have looked at the expression of CRBP1 by the means of immunohistochemistry in mouse-embryo tissues (20), rat and human liver, kidney, respiratory tract, testis (21), uterus (36), and rat ovary, including focal expression in surface epithelium (37).

In conclusion, this study is notable because it identifies a tumor-associated abnormality in a protein believed to be critical to vitamin A homeostasis. Here we have shown that the loss of CRBP1 is an early event in ovarian carcinogenesis. Therefore, if proper vitamin A usage is necessary for the prevention of oncogenesis, it is possible that the loss of CRBP1 would enhance the development of the tumor.

We thank Dr. Samuel Litwin and Hao Wang for statistical analysis of the data related to this study. We also thank Dr. Ulf Eriksson (Ludwig Institute for Cancer Research, Stockholm, Sweden) for kindly providing antibodies against CRBP1.