Oncogenic BARD1 Isoforms Expressed in Gynecological Cancers

BARD1 was discovered as a protein interacting with BRCA1 (1). BARD1 and BRCA1 form a stable heterodimer, which has functions in DNA repair, transcription regulation, RNA processing, and cell cycle regulation through the ubiquitin ligase function of the heterodimer (2). Depletion of BARD1 in vitro leads to genomic instability, loss of polarity, development of a premalignant phenotype (3), and embryonic lethality in knockout mice (4). As a tumor suppressor, BARD1 has also BRCA1-independent functions in mediating p53-dependent apoptosis, which is inhibited by BRCA1 (5). It binds to p53, facilitating its phosphorylation and stabilization (6).

Recently, a novel function of BARD1 in mitotic spindle assembly was described (7). The expression of BARD1 in most proliferative tissues, particularly in spleen and testis (3, 8), is consistent with its role in mitosis. Up-regulated expression of BARD1 was also observed in response to hypoxia, to genotoxic stress (5, 9), and to hormone signaling (10), suggesting that up-regulation of BARD1 might be associated with tumor suppressor pathways.

In spite of this, BARD1 mutations in tumors are infrequent. Three missense mutations, Q564H, V695L, and S761N, and five other alterations were discovered in breast, ovarian, and endometrial cancers (11). Loss-of-heterozygosity was found for Q564H and S671N mutations, substantiating the role of BARD1 as a tumor suppressor (11, 12). The V695L and S761N mutations were found in somatic breast tissue, whereas the Q564H mutation arose in the germline of a patient with clear cell ovarian carcinoma. At least six other alternations were subsequently identified (13–15), including the missense mutation C557S, which was linked to loss of p53 stability and apoptotic activity (16).

BARD1δ, an isoform presenting a deletion of exons 2 through 6, was found in a rat ovarian cancer cell line, which is resistant to apoptosis (6), and in HeLa cells (17). BARD1δ lacks the RING finger domain, required for BRCA1-interaction, and the ankyrin repeats, which are involved in apoptosis and p53 binding (6, 9). Aberrantly elevated expression and cytoplasmic localization of truncated forms of BARD1 was found in breast and ovarian cancer and correlated with factors of poor prognosis (18). Interestingly, spliced isoforms of BARD1 are expressed, temporally and spatially controlled, in human cytotrophoblasts, which have cancer-like properties in the first trimester of pregnancy (19).

Alternative splicing is a crucial mechanism for generating protein diversity and even proteins with novel antagonistic biological functions. Several cancer-associated alterations of splicing patterns have been reported for genes such as RON, RAC1, CD44, and MDM2 (20). Because differential splicing of tumor suppressor genes, such as BARD1, could have a profound effect on tumor development and progression, we characterized the expression pattern of BARD1 isoforms in ovarian, breast, endometrial, and cervical cancer cell lines, and we investigated their potential oncogenic properties in ovarian cancer.

Cancer cell lines. Cancer cell lines were a collection of the Universitätsspital Wien.

Ovarian cancer cell lines, composed of established ovarian cancer cell lines and newly generated ones from individual tumors (O1–O32), were as follows: A2780, Caov-3, ES-2, NIH:OVCAR-3, SK-OV-3, TOV-21G, TOV-112D, OV-90, OV-MZ-1a, OV-MZ-1c, OV-MZ-2, OV-MZ-2a, OV-MZ-5, OV-MZ-6, OV-MZ-8, OV-MZ-9, OV-MZ-10, OV-MZ-12, OV-MZ-12b, OV-MZ-17b, OV-MZ-18, OV-MZ-20, OV-MZ-21, OV-MZ-22, OV-MZ-26, OV-MZ-27, OV-MZ-30, OV-MZ-32, OV-MZ-33, OV-MZ-35, OV-MZ-37, and OV-MZ-38.

Breast cancer cell lines, composed of established cell lines and newly generated ones from individual tumors (B1–B26), were as follows: MCF-7, MM231, T47D, Hs578T, SKBR3, MM435s, ZR-75-1, BT549, MM453, BT474, PA1, A2780ADR, BT20, HBL100, HMEC, MCF12A, MCF10A, MCF7/6, MCF12F, MM134VI, MM157, MM175VII, MM330, MM468, UCAA812, and MM361.

Endometrial cancer cell lines (E1–E9) were as follows: KLE, RL95-2, AN3 CA, HEC-1-B, Ishikawa, Colo 684, HEC-50, EN, and EJ.

Cervical cancer cell lines (C1–C9) were as follows: HeLa, SW756, GH354, Ca Ski, C-4 I, C-33 A, HT-3, ME-180, and SiHa.

RNA isolation and reverse transcription-PCR. Total RNA from cell lines and tissue specimens were extracted after standard protocols. For reverse transcription, 0.5 μg of RNA were used in standard assays using M-MLV-Powerscript reverse transcriptase. PCR reactions were performed with 2 μL cDNA as template for 30 cycles with primers as indicated (Supplementary Data Table S1) and Taq DNA polymerase in a final volume of 50 μL. Denaturation (94°C; 3 min) and final extension (72°C; 10 min) were the same for all PCR reactions, but annealing temperature and extension times varied according to different primers. PCR products (15 μL) were analyzed on 1% agarose/Tris-acetate-EDTA gels.

Determination of BARD1 cDNA 5′ends in ovarian cancer and HeLa cells. GeneRacerTM kit (Invitrogen) was used to amplify cDNA 5′ends of RNA from ovarian tumor and HeLa cells. Total RNA of 4.5 μg was used per sample, and experiments were performed according to the instructions of the supplier. 5′nested PCR was performed with 5′nested primer 5′-GGACACTGACATGGACTGAAGGAGTA-3′ and reverse primer in exon 6 (5′-TTTTGATACCCGGTGGTGTT-3′). The PCR bands of 5′nested PCR were loaded on 1% low-melting gel, cut, and purified with the QIAEX II kit (Qiagen) followed by sequencing using 5′nested primer and reverse primer.

Western blots. BARD1 antibodies H300, directed against NH₂-terminal 300 aa (sc-7372; Santa Cruz Biotechnology), were used to detect NH₂-terminal epitopes. BARD1 JH3 (10) was used to detect epitopes encoded in exon 7. Anti-BARD1 BL518, directed against epitopes within exon 4 (Bethyl Laboratories), was used for detection of FL BARD1 and β, as described previously (21). Protein extracts from different ovarian cancer cells lines were prepared, and 40 μg of protein per lane were analyzed on 10% SDS-PAGE and blotted onto polyvinylidene difluoride filters. Membranes were blocked with 5% milk powder in TBS. Antibody incubation with purified anti-BARD1 H300 and JH3 was in a 1:500 dilution and in 1:1,000 dilution for BL518. Secondary anti-rabbit peroxidase–coupled antibodies were applied in a 1:10,000 dilution. Signal detection was performed with enhanced chemiluminescence (ECL) and/or high sensitivity ECL kit (Amersham).

Immunohistochemisty. Formalin-fixed and paraffin-embedded tissue micro arrays were deparaffinized with xylene for 48 h and rehydrated through descending ethanol (100%, 95%, 70%, and H2O). The sections were boiled for 5 min in microwave for antigen retrieval. Slides were incubated for 24 h at 4°C in a humidifying chamber with first antibody after bovine serum albumin blocking of nonspecific epitopes. The primary antibodies for BARD1 detection were N19 (sc-7373; Santa Cruz Biotechnology), WFS, described previously (3), and C-20 (sc-7372; Santa Cruz Biotechnology), which recognize epitopes at NH₂ terminus, in exon 4, and at COOH terminus of BARD1, respectively. Secondary antibodies (goat anti-rabbit or rabbit anti-goat) conjugated with horse radish peroxidase were applied in 1:100 dilutions at room temperature for 1 h. Then 3,3′-diaminobenzidine staining was permitted for 15 min at room temperature. Slides were counterstained with hematoxylin before dehydration and mounting. Sections from human lung cancers, described previously (18), were used for positive and negative control.

To quantify BARD1 expression, staining was scored for intensity and percentage of stained cells. The value of staining intensity and percentage of positive cells was used to obtain final staining score. Statistical significance of comparisons was determined by applying Student's t test.

Clinical data. The pathologic diagnoses were made by experienced pathologists and staged according to the WHO and American Joint Committee on Cancer classifications. A total of 106 cases of ovarian cancer from women of ages 32 to 87 years were analyzed, comprising 60 cases of serous carcinoma, 24 cases of endometriod carcinoma, 16 cases of mucinous carcinoma, and 6 cases of clear cell carcinoma. According to the tumor-node-metastasis staging system, there were 38 cases in T₁, 15 cases in T₂, 53 cases in T₃, 36 cases in N₀, and 63 cases in N₁ stage. There were 25, 26, and 55 cases of pathological grade 1, 2, and 3, respectively.

Cloning of and transduction with lentiviral vectors. Lentiviral vector pLVTHM (22) was used to constitutively express the BARD1 coding region under the control of cytomegalovirus promoter.

The BARD1 siRNA construct was generated by annealing of complementary oligonucleotides and inserting into the pSuperScript vector as a BglII/HindIII fragment. Primers for si78 of human and mouse BARD1 were siRNA-F GATCCCCGTGTATGCTTGGGATTCTCTTCAAGAGAGAGAATCCCAAGCATACACTTTTTGGAAA and siRNA-R AGCTTTTCCAAAAAGTGTATGCTTGGGATTCTCTCTCTTGAAGAGAATCCCAAGCATACACGGG. The pSuper fragment ClaI/EcoRI containing the BARD1 siRNA-78 were then subcloned in the lentiviral vector pLVTHM as described (22).⁴

These lentiviral vectors were then transfected into HEK293T cells with the corresponding packaging pCMV-dR8.92, pCMV-dR8.93, and envelope pMD2G plasmids. The recombinant lentiviruses were produced and purified according to standard protocols (24). HeLa-FL cells were generated by transduction with constitutive FL BARD1 for BARD1 expression control on Western blots.

SK-OV-3 and NuTu-19 cells were plated on 6-well plates (10⁴ cells per well) for 24 h and then incubated with medium containing recombinant lentiviral vectors for 48 h. To control the expression of the siRNA, cells were cotransduced with lentivirus derived from the vectors pLVTHM and pLV-tTRKRAB (22).

Structure of BARD1 isoforms in cancer cells. It was previously reported that HeLa cells express BARD1 δ, an isoform derived from differential splicing (17). We compared the expression of BARD1 and its isoforms in normal human fibroblasts and in HeLa cells by applying reverse transcription-PCR (RT-PCR) with primers intended to amplify the entire BARD1 coding region. Although full-length (FL) BARD1 was highly expressed in normal fibroblasts, in HeLa cells, smaller isoforms of BARD1 were expressed in addition to FL BARD1. We cloned and sequenced these isoforms and determined their structure and exon composition, and calculated molecular weight (MW; Fig. 1A and B). FL BARD1 translates into a protein of 777 amino acids (aa) with a calculated molecular weight of 87 kDa. Isoform α lacks exon 2 and could produce a 85 kDa protein of 758 aa. Isoform β, missing exons 2 and 3, translates into a putative protein of 680 aa and 75 kDa but would use a translation start in an alternative open reading frame (ORF) of exon 1 (Fig. 1C). Deletion of exon 4 in isoform γ disrupts the ORF. Isoform φ and δ, bearing deletions of exons 2 to 6 and 3 to 6, respectively, could produce 37 and 35 kDa proteins of 326 and 307 aa, respectively. The latter has been reported previously in HeLa (17) and ovarian cancer cells (6). Isoform ε lacks exons 4 to 9 and codes for a protein of 30 kDa, composed of 264 aa. Isoform η is composed of exons 1, 10, and 11. Splicing of exon 1 to exon 10 disrupts the ORF, but initiation of translation could occur in an alternative ORF of exon 1 and translate into a 19 kDa protein of 167 aa (Fig. 1C). Interestingly, all these isoforms lack either the RING finger or ankyrin repeats, or both, and both regions are required for the tumor suppressor functions of BARD1 (2).

BARD expression pattern in different cancer cell lines. To investigate the pattern of BARD1 isoform expression in cells from cancers of different origin, we performed RT-PCR, on RNA from breast, ovarian, endometrial, cervical, and hematologic cancer cell lines, for amplification of the entire BARD1 coding region. RNA from HeLa cells was used as a positive control.

In ovarian cancer cells, FL BARD1 was expressed as well as isoforms β, φ, δ, ε, and η, which were often more abundant than FL BARD1. Complete absence of FL BARD1 but expression of splice isoforms was found in 5 of 32 ovarian cancer cell lines. A subgroup of ovarian cancer cell lines showed no BARD1 expression at all when forward primers in exon 1 were used for amplification of FL BARD1, but in all cases, BARD1 transcripts could be amplified by using forward primers in exon 4 (Fig. 2A). Similar results were obtained for BARD1 expression in breast and endometrial cancer cell lines (Supplementary Fig. S1).

In cervical cancer cell lines, we found neither FL BARD1 nor splice isoforms expressed when forward primers within exon 1 were used for amplification of BARD1 sequences. To amplify potentially 5′truncated forms of BARD1, we used forward primers within exon 3, at different sites of exon 4, and within exon 5. BARD1 sequences could be amplified in some samples with primers in exon 4 and in all samples with a forward primer within exon 5 (Fig. 2B). This indicates that in all cervical cancer cell lines, BARD1 isoforms were initiated in exon 4 or 5. The control experiment, amplification from exon 1 to exon 6, was completely negative.

In summary (Table 1), FL BARD1 expression was absent in all cervical cancer cell lines tested, but BARD1 transcripts originating in exons 4 or 5 were present. In breast, ovarian, and endometrial cancer cells, FL BARD1 and splice isoforms were expressed, and splice isoforms alone were found in 15% of ovarian and in 11% of endometrial cancers. Loss of FL BARD1 and expression of transcripts initiating within exon 4 was found in 80% of breast cancer cell lines and was less frequent in ovarian (62.5%) and endometrial cancer cells (33.3%).

Thus, in all cancer cell lines derived from potentially hormone-controlled gynecologic cancers, FL BARD1 was either missing or was less abundant than isoforms, and in none of these cancer cells, BARD1 expression was absent.

To test whether BARD1 isoform expression was a particularity of gynecologic cancers, we performed RT-PCR in hematologic tumor cell lines that are unlikely to be hormonally controlled (Fig. 2C). Most of these samples showed FL BARD1 expression, and only weak amplification of isoform γ, an out-of-frame splice variant, was detected.

Alternative initiation of transcription in exon 4. The results described above suggested that mRNAs transcribed from exon 4 to exon 11 use an alternative site of transcription initiation, which was consistent with the loss of NH₂-terminal epitopes, which was frequently observed in ovarian cancer samples (18). Therefore, we performed RT-PCR on RNA from ovarian tumor tissue. We used forward primers located at the beginning of exon 4 (nucleotide position 428 of human BARD1 cDNA) and the end of exon 4 (nucleotide position 1280), respectively. Although in HeLa cells, both primers could amplify a BARD1 cDNA fragment of the expected length, in the ovarian cancer sample, BARD1 expression was only detected with the forward primer at the end of exon 4 (Fig. 3A).

To determine the exact start of transcription, we performed 5′race experiments to amplify the cDNA 5′ends, using RNA from ovarian cancer and from HeLa cells with reverse primers in exon 6. In HeLa cells, the amplicons corresponded to known 5′end of the human BARD1 cDNA sequence, and corresponded to FL BARD1 and isoform γ and to a cDNA derived from alternative transcription initiation within exon 4 (Fig. 3B; start 3). However, in ovarian cancer, we found two new transcription initiation sites within exon 4, mapping to nucleotide positions 458 (Fig. 3B; start 1), located at the beginning of exon 4, and 983 (start 2), in the middle of exon 4. Transcripts initiating at starts 1 and 2, designated Ω1 and Ω2, could be translated from ATGs at positions 1147, 1165, and 1183 within exon 4 and produce a protein of ∼44 kDa; and transcripts originating at start 3, designated Ψ, could produce a protein of ∼27 kDa (Fig. 3C and Supplementary Fig. S2).

These alternative 5′ends of transcripts are consistent with the results obtained by RT-PCR (Fig. 2A): Forward primer at nucleotide position 783 could amplify isoform Ω1, and forward primer at nucleotide position 1280 could amplify isofoms Ω1 and Ω2. In cervical cancer cells, Ω2 was present in five of nine samples, and Ψ was present in the remaining four samples (Fig. 2B). Interestingly, in ovarian, breast, and endometrial cancer cell lines, Ω1 or Ω2 were detected in all samples that lacked expression of FL BARD1. It is possible, but cannot be determined by RT-PCR, that Ω isoforms are also expressed in cells that express FL BARD1. These data show that gynecologic cancers express deletion-bearing BARD1 isoforms, derived from differential splicing and from alternative transcription initiation within exon 4.

Identification of BARD1 protein isoforms in ovarian cancer cell lines. Because we found different BARD1 isoforms transcribed in cancer cells, we were interested in determining whether these isoforms were translated. We analyzed protein extracts from ovarian cancer cell lines and HeLa cells overexpressing exogenous FL BARD1 (HeLa-FL) as a control on Western blots (Fig. 3D). We used anti-BARD1 H300, directed against the NH₂-terminal 300 aa of BARD1 (Fig. 3C), and antibody JH3 directed against a peptide antigen within exon 7 (10, 19, 25), to detect BARD1 and BARD1 isoforms. H300 detected FL BARD1, which migrates as a 97 kDa protein on SDS-PAGE, only in extracts from HeLa-FL cells but in none of the ovarian cancer cell lines (Fig. 3D). A 94-kDa protein was detected in a few of the ovarian cancer samples and an 82-kDa protein was present in all samples. Based on the cDNA structure of BARD1 isoforms (Fig. 1B), the 94 kDa and 82 kDa bands might be translation products of isoform α (deletion exon 2) and isoform β (deletion of exons 2 and 3), respectively, assuming that, like FL BARD1, they migrate slightly higher than predicted from their calculated molecular weight. In some of the samples, we observed two smaller bands of about 40 kDa, which might be isoforms φ (deleted of exons 3 to 6) and δ (deleted of exons 2 to 6). Western blots probed with JH3 resulted in very strong staining of a 48 kDa protein, which was not detected by H300. This protein might correspond to the protein translation product of isoforms Ω1 and Ω2 (44 kDa). Importantly, in nonmalignant tissue, JH3 recognizes FL BARD1 and BARD1β, as was shown previously for BARD1 expression in rat testis (10). JH3 did not detect FL BARD1 in any of the cell lines and only showed weak staining for BARD1β with exposure times applied for detection of Ω, indicating that Ω was more abundant that BARD1β. Isoform Ω cannot be detected by anti-BARD1 H300 because it cannot recognize in vitro generated BARD1 deletion mutants that use the same translation initiation codons as Ω isoforms (Supplementary Fig. S3). The smaller proteins of about 41 kDa detected by JH3 could be isoforms φ and δ, which were also detected by H300. Thus, RT-PCR and Western blot analyses show, at the mRNA and at the protein level, that little or no expression of FL BARD1 but high expression of isoforms is found in ovarian cancer cells.

BARD1 expression in ovarian cancer. To investigate BARD1 expression in tumors and its relevance for tumor growth, we performed immunohistochemistry on a large sample collection of ovarian tumors prepared as tissue arrays. We used antibodies detecting epitopes at the NH₂ terminus (N19), within exon 4 (WFS), and at the COOH terminus (C-20) of BARD1, which were all described before (N19 and C-20, refs. 5, 18, 19; WFS, ref. 3). A total of 106 cases of ovarian cancer were analyzed. With antibodies N19 and C-20, we observed high expression of BARD1 in some of the patient samples and weak in others, but in a third group, staining of N19 was lost or decreased (Fig. 4A). Antibody WFS was weakly expressed in most of the samples. Interestingly, loss of N19 expression mostly happened in ovarian cancer of T₃ stage or cancers with lymph node metastasis (N₁; P < 0.05; Fig. 4B). Thus, loss of NH₂-terminal epitopes is correlated with advanced stage of ovarian cancer. However, we did not find a correlation of BARD1 expression and pathologic tumor grade (data not shown). Expression of COOH terminus and loss of NH₂ terminus are compatible with the expression of Ω isoforms. Interestingly, we found strong staining for N19 and C-20 but no staining for WFS in clear cell carcinoma (Fig. 4B and C), which is the type of ovarian cancer with the worst prognosis. This staining pattern is consistent with the expression of isoforms φ, δ, and ε.

RT-PCR, performed in ovarian cancer cell lines derived from clear cell carcinoma, SK-OV-3 and TOV-21G, confirmed this hypothesis. Elevated expression of isoforms φ, δ, and ε, but not FL BARD1, was found in SK-OV-3 and TOV-21G cells (Fig. 4D). Western blots, using BARD1 H300 and JH3 antibodies on extracts from ovarian cancer cell lines, including SK-OV-3 and TOV-21G (Fig. 3D; O5 and O6), showed absence of FL BARD1 and presence of isoforms. To further support this finding, we used BARD1 antibody BL518, directed against epitopes encoded on exon 4 (26), which could only detect FL BARD1 and BARD1β, to analyze BARD1 expression in SK-OV-3 and TOV-21G cells. Although we found abundant FL BARD1 expression in HEK 293 cells, no FL BARD1 was found in SK-OV-3 and TOV-21G cells, but BARD1β was present in all (Fig. 4D). Thus, SK-OV-3 and TOV-21G cells express BARD1 isoforms but not FL BARD1.

BARD1 isoforms are essential for tumor cell growth. Because SK-OV-3 and TOV-21G cells expressed BARD1 isoforms, we were interested in elucidating the functions of these isoforms. We performed repression assays in SK-OV-3 with drug-inducible siRNAs (22). Cells were transduced with control lentiviral vector pLV or lentiviral vector si78, expressing siRNA-targeting sequences in exon 9 of human BARD1, present on isoforms β, φ, and δ (Fig. 5). Both vectors are coexpressing green fluorescent protein (GFP) as reporter gene. Cell growth was monitored and cell numbers were determined at 4 and 9 days after induction of siRNA expression. At 9 days of siRNA expression, we observed pronounced repression of cell growth and only few cells transduced with si78 were attached, whereas cells transduced with control vector proliferated normally even after 9 days (Fig. 5A). Interestingly, these differences in cell proliferation were not observed after 4 days of repression (Fig. 5B). This delay in response of siRNA repression might be explained by the slow decrease of BARD1β after siRNA induction (Fig. 5B). This observation is consistent with increased stability of in vitro generated NH₂ terminal deletion–bearing BARD1 isoforms (9). With the increased stability of BARD1β, we also observed in time curves of BARD1 repression in HeLa cells, although repression of FL BARD1 is observed after 4 days, repression of isoform β is only seen after 9 days (data not shown). Therefore, these experiments show that BARD1β, and presumably φ and δ (not detectable with anti-BARD1 BL518), are important for cancer cell growth, and that repression of these isoforms leads to proliferation block and cell death.

To further investigate this function of BARD1 isoforms in proliferation, in other than human cancer cells, we performed a similar repression experiment in a rat ovarian cancer cell line, NuTu-19, which was previously shown to abundantly express alternatively spliced isoforms BARD1β and δ but not FL BARD1 protein (ref. 6; Supplementary Fig. S4A). We transduced NuTu-19 cells with lentiviral vectors, containing BARD1 control siRNAs si34, targeting sequences in human but not rat exon 2, and si78, targeting a sequence in exon 9, which is identical in human and rat. After transduction and induction of siRNA expression, RT-PCR was performed and showed that si78 completely repressed all BARD1 isoforms (Supplementary Fig. S4B). Fluorescence microscopy, showing 4′,6-diamidino-2-phenylindole (DAPI) staining and GFP expression for transduction control, illustrates that NuTu-19 cells transduced with si78 do not proliferate and the few surviving cells became big and flat. Cells transduced with the control si34 had a normal morphology and high proliferation rate (Fig. 5C and D).

Research on BARD1 has intensified in the last few years, and evidence is accumulating that it is a tumor suppressor both as a binding partner of the breast cancer protein BRCA1 and independently of BRCA1 (2). Inherited predisposition to breast and ovarian cancer is associated in up to 50% of cases with mutations in BRCA1 and BRCA2. BARD1 mutations associated with inherited and spontaneous cases of breast and ovarian cancer, however, are very infrequent (13, 27). This could be interpreted as BARD1 being involved in pathways affecting cell viability.

Indeed, rather than absence of BARD1 due to mutations, overexpressed aberrant forms of BARD1 were found in breast and ovarian cancer (18). Here we report that cancer cells express a complex pattern of BARD1 isoforms, namely α, β, γ, δ, φ, ε, and η. All of them are likely to code for proteins, except for isoform γ, which does not have a continuous ORF. All these isoforms are also expressed in cytotrophoblasts of early pregnancy (19), which share invasive properties with cancer cells. In cytotrophoblasts, the expression of BARD1 isoforms is temporally and spatially regulated, suggesting that they play a role in proliferation and invasion (19).

In cancer cell lines derived from various types of gynecologic cancers, we found a specific pattern of differentially spliced isoforms and isoforms derived from transcription initiation in exon 4, which was in strong contrast to hematologic cancer cell lines, where FL BARD1 was expressed in all cell lines tested (Table 1). BARD1 isoforms are either deprived of the BRCA1-interacting RING finger domain, i.e., isoforms β, δ, φ, η, and Ω, or, like isoforms ε and η, have lost the ankyrin motif or part of the BRCT region, which suggests loss of other tumor suppressor functions involving binding to these regions (6, 28, 29). NH₂- terminally truncated forms of BARD1, derived from transcription initiation within exons 4 or 5, were found in all gynecologic cancer cell lines. Such transcripts could encode a 44 kDa protein, which is indeed highly expressed in all ovarian cancer cell lines, although FL BARD1 protein is absent. This finding suggests that FL BARD1, even when identified by RT-PCR, is proportionally less efficiently translated or less stable than isoforms.

BARD1 isoforms might be specifically expressed in gynecological cancers, suggesting that their expression might be linked to hormone-signaling pathways. It is well-established that exposure to female reproductive hormones is a central factor for the development of breast (30–32) and ovarian cancer (33, 34). Interestingly, a region of the BARD1 gene within intron 8 confers estrogen responsiveness, and BARD1 mRNA and protein levels can be increased by estrogen (35). Recently, the estrogen receptor α was identified as a substrate for the BRCA1-BARD1 ubiquitin ligase (36), which is in line with the observations that BRCA1 is an activator of estrogen receptor transcriptional activity (37, 38). The estrogen-regulated expression of BARD1, however, links the ubiquitin ligase activity of the BRCA1-BARD1 heterodimer to estrogen levels. The up-regulated expression of BARD1 isoforms lacking the BRCA1 interaction domain disrupts this regulatory circle and provides an explanation for BARD1 isoform expression in gynecological cancers and their correlation with poor prognosis. It is possible that BARD1 isoforms act like dominant negative mutations in an estrogen-signaling pathway and promote cancer development and progression. Further experiments will be needed to establish the estrogen-dependent expression of each particular isoform and to identify the regulatory pathways with which they interfere.

It was interesting that in all cancer cell lines tested, we always could detect BARD1 isoforms or FL BARD1 but never found cells that completely lacked BARD1 expression. In contrast to hypermethylation of BRCA1, which was observed in sporadic breast cancers (39, 40), no promoter methylation of BARD1 was found in 50 primary cell lines generated from ovarian cancer samples and 34 ovarian cancer cell lines (data not shown). This suggests that BARD1 or isoforms might be essential for cancer cell growth. Importantly, siRNA repression experiments, targeting isoforms in cell lines that lack FL BARD1, prove that isoforms are required for cell growth (Fig. 5). Hence, cancer-specific BARD1 isoforms might have lost tumor suppressor functions but retained or acquired functions that play a role in cell viability and tumor growth.

Immunohistochemistry of 106 ovarian cancers confirmed the importance of aberrant expression of BARD1 isoforms. Specifically, expression of isoforms missing NH₂-terminal epitopes, such as β and Ω, was associated with tumors of T₃ (tumor invades tissue outside of the pelvis) and N₁ (positive for lymph node metastasis) stages, suggesting that expression of β and Ω is associated with advanced stage of ovarian carcinoma and might be a negative prognostic factor in ovarian cancer.

A type of ovarian cancer with poor prognosis and poor response to platinum-based chemotherapy is clear cell carcinoma (41, 42). We found a specifically high expression of BARD1 isoforms β, φ, δ, and ε in clear cell carcinoma. In vitro repression assays, performed in clear cell carcinoma–derived cells and in cells from rat ovarian cancer, clearly show that BARD1 isoforms are essential for cell growth. Because the siRNAs were targeting sequences within exon 9, expression of β, φ, and δ was affected but not ε and η. In both cell types, the repression of these isoforms led to complete growth arrest and cell death. Thus, isoforms β φ, and δ might have essential functions in proliferation of clear cell carcinoma, the ovarian cancer with the worst prognosis.

Loss of FL BARD1 and expression of isoforms that lack the RING finger and are deficient for BRCA1 interaction (43) entails the loss of BRCA1-BARD1 ubiquitylation functions, which are important for tumor suppressor functions but will also influence BRCA1 stability and/or nuclear localization (44–46). Therefore, lack of FL BARD1 and expression of isoforms lacking the RING domain could have a similar cancer predisposition effect as mutations in BRCA1 when occurring in healthy tissue.

To summarize, BARD1 isoforms are more expressed than FL BARD1 in gynecological cancers. All isoforms lack regions that are required for BRCA1 interaction and tumor suppressor functions, but might have acquired novel functions in favor of tumor growth. Because repression of isoforms in ovarian cancer cells leads to growth arrest, this suggests that BARD1 isoforms act as tumor maintenance genes in gynecological cancers.

Grant support: Swiss National Science grant 3100A0-110038 (I. Irminger-Finger) and the Foundation for Medical Research.

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.

We thank Prof. Daniel Birnbaum for his valuable comments and suggestions, Aurelie Callon for technical help, and Professors Karl-Heinz Krause and Paul Bischof for advice and critical discussions.