Synthetic lethality between CCNE1 amplification and loss of BRCA1

Significance

Women with high-grade serous ovarian cancer (HGSC) harboring Cyclin E1 (CCNE1) gene amplification generally face a poor clinical outcome. These tumors comprise a significant group of ∼20% of HGSCs that are not associated with BRCA1/2 mutation and are unlikely to respond to standard cytotoxic or poly-ADP-ribose polymerase inhibitors. We identified a specific dependency on BRCA1 and members of the ubiquitin pathway in CCNE1-amplified tumors. The requirement for BRCA1 seems to account for the mutual exclusivity of mutations observed in primary tumors. We propose a unique therapeutic strategy involving inhibition of the proteasome and homologous recombination function with bortezomib. Our findings are likely to have relevance to the treatment of other tumor types with CCNE1 amplification, including triple negative breast cancer.

Abstract

High-grade serous ovarian cancers (HGSCs) are characterized by a high frequency of TP53 mutations, BRCA1/2 inactivation, homologous recombination dysfunction, and widespread copy number changes. Cyclin E1 (CCNE1) gene amplification has been reported to occur independently of BRCA1/2 mutation, and it is associated with primary treatment failure and reduced patient survival. Insensitivity of CCNE1-amplified tumors to platinum cross-linking agents may be partly because of an intact BRCA1/2 pathway. Both BRCA1/2 dysfunction and CCNE1 amplification are known to promote genomic instability and tumor progression. These events may be mutually exclusive, because either change provides a path to tumor development, with no selective advantage to having both mutations. Using data from a genome-wide shRNA synthetic lethal screen, we show that BRCA1 and members of the ubiquitin pathway are selectively required in cancers that harbor CCNE1 amplification. Furthermore, we show specific sensitivity of CCNE1-amplified tumor cells to the proteasome inhibitor bortezomib. These findings provide an explanation for the observed mutual exclusivity of CCNE1 amplification and BRCA1/2 loss in HGSC and suggest a unique therapeutic approach for treatment-resistant CCNE1-amplified tumors.

Epithelial ovarian cancer is complex and histologically diverse but still largely treated as a single disease with limited stratification based on histological or molecular characteristics. High-grade serous ovarian cancer (HGSC) accounts for the majority of epithelial ovarian cancer-related deaths (>60%), and almost no improvement in survival has been observed in the last 20 y (1). Widespread copy number changes are a hallmark of HGSC, including focal amplification of Cyclin E1 (encoded by CCNE1), which is associated with primary treatment failure (2) and reduced survival (3). Amplification of CCNE1 is one of very few well-defined molecular targets in HGSC.

Cyclin E1 forms a complex with cyclin-dependent kinase 2 (CDK2) to regulate G1/S transition as well as having kinase-independent functions, including in DNA replication (4). Ovarian cell lines with CCNE1 amplification show a specific dependency for maintenance of CCNE1 expression (5, 6). We have validated CDK2 as a therapeutic target by showing selective sensitivity to suppression either by gene knockdown or using small molecule inhibitors (7), consistent with findings in breast cancer (8).

Recent genomic studies have revealed a high frequency of BRCA1/2 (Breast cancer 1/2, early onset) inactivation and homologous recombination (HR) dysfunction in HGSC (9). Alterations of genes in the HR pathway include germ-line and somatic mutations of BRCA1 or BRCA2 (∼20% of patients) and epigenetic silencing of BRCA1 by hypermethylation (∼10%). Other genes inactivated by deletion, mutation, or hypermethylation include ATM, ATR, RAD51C, and PTEN (∼10%), key Fanconi anemia members (∼5%), and amplification or mutation of EMSY (∼8%). Collectively, at least 50% of HGSCs are thought to have HR pathway defects (9).

Approximately 30% of HGSC tumors have alterations in the Rb pathway or genes involved in Rb-mediated DNA repair and cell cycle control, including amplification of CCNE1 (∼20%), loss of RB1 (∼10%), or gain of RBBP8 (∼4%) (10). Strikingly, activation of the RB1/CCNE1 pathway is largely exclusive of BRCA1/2 mutation for reasons that are unclear (9, 10). Both BRCA1/2 dysfunction and CCNE1 amplification are known to promote genomic instability and tumor progression (4, 11); therefore, they may be mutually exclusive, because either change provides a path to tumor development, with no selective advantage to having both mutations (10). Insensitivity of CCNE1-amplified tumors to platinum cross-linking agents may be partly because of an intact BRCA1/2 pathway, suggesting that these patients are unlikely to respond to poly-ADP-ribose polymerase (PARP) inhibitors.

Here, we show that BRCA1 and members of the ubiquitin pathway are selectively required in cancers that harbor CCNE1 amplifications. Furthermore, we show specific sensitivity of CCNE1-amplified tumor cells to the proteasome inhibitor bortezomib. These findings provide an explanation for the observed mutual exclusivity of CCNE1 amplification and BRCA1/2 loss in HGSCs and suggest a unique therapeutic approach for treatment-resistant CCNE1-amplified tumors.

Results

CCNE1 Gene Amplification in Primary Tumors.

To better define the frequency of CCNE1 amplification in solid cancers, we used genomic data from The Cancer Genome Atlas (TCGA) to perform a pan-cancer analysis of 22 cancer types (Materials and Methods). We found that focal high-level amplification of 19q12 involving CCNE1 occurs at a frequency of ∼5% in breast, lung, and gastric cancers and that it is most frequent (∼25%) in HGSCs (Fig. 1A). Consistent with our previous findings (5), the most significant or peak region of amplification always involved CCNE1; however, genes neighboring CCNE1, such as prefoldin-like chaperone URI1, were present in peak regions in some tumor types. These findings suggest that the reported driver activity of URI (12) or other genes within the 19q12 amplicon (8) may be restricted to certain cancers.

Fig. 1.

(A) Pan-cancer copy number analysis of 6,547 tumor samples comprising 22 cancer types from TCGA. Frequency of high-level amplification of peak regions of copy number change incorporating CCNE1 is shown for cancer types with amplification. Data were obtained from the TCGA Copy Number Portal using all available data as of February of 2013. Name or total number of genes including CCNE1 within significant peak regions of amplification is indicated. (B) CCNE1 copy number assessed by qPCR in primary tumor samples from the Australian Ovarian Cancer Study (n = 193) stratified by WT, germ-line (GL), or somatic (SOM) BRCA1/2 mutation or methylation (METH) status. Bars indicate mean and SD. t test. *P value < 0.05; **P value < 0.01.

The previously reported mutual exclusivity of BRCA1/2 mutations and CCNE1 amplification (9, 10) has not been validated in an independent dataset. We, therefore, stratified tumors from the Australian Ovarian Cancer Study (n = 194) by BRCA1/2 germ-line mutation (13), somatic mutation, or methylation status (14) and accurately measured CCNE1 copy number by quantitative PCR (qPCR) (Fig. 1B). Assessment of copy number by qPCR provides a more accurate measure of the extent of copy number change (2) than microarray-based estimates used in prior studies (9). Although low-level CCNE1 gain and BRCA1/2 mutation were observed, complete mutual exclusivity was seen between high-level CCNE1 amplification (log₂ ratio > 2; approximately eight copies per genome) and BRCA1/2 germ-line mutations (Fisher test P value < 0.01). Although the study was not adequately powered for comparison with somatically mutated or methylated samples, the same pattern was observed. Our findings suggest that there is a functional difference between low- and high-level CCNE1 copy number states and that there is a threshold of CCNE1 amplification where BRCA1/2 inactivation is unlikely to co-occur.

Dependencies of CCNE1-Amplified Cell Lines.

We have recently shown oncogene addiction to Cyclin E1 and its partner kinase, CDK2, in CCNE1-amplified ovarian tumors (7), suggesting that use of CDK2 inhibitors may be effective in these cancers. In addition to CDK2, Cyclin E1 interacts with CDK1 and CDK3 and has kinase-independent functions (4). Furthermore, Cyclin E1 is regulated both positively and negatively by posttranslational proteolysis (15). To better understand the dependencies of tumor cells with CCNE1 amplification and identify other potential therapeutic targets, we analyzed data from a genome-wide shRNA screen of 102 cancer cell lines with known copy number status, including a high proportion of epithelial ovarian cancer (n = 25) (16). We included all available cell lines to obtain sufficient statistical power for the analysis. Cells infected with a pool of 54,020 shRNAs (targeting 11,194 genes) were grown for at least 16 doublings, and the abundance of individual shRNA sequences was measured relative to a reference to identify genes essential for survival (16). In two separate analyses, we compared CCNE1-amplified (n = 23) with nonamplified (n = 43) and CCNE1 high- (n = 15) with low-expressing (n = 41) cells. To improve specificity, we removed samples with CCNE1 copy number or expression that fell into the midrange of values. Using a statistical approach that considered data from either the second-best shRNA or an aggregate score from multiple shRNAs targeting the same gene (Materials and Methods), we identified 835 essential genes in either CCNE1-amplified and/or overexpressing cancer cell lines (Table S1). We then considered four additional factors as evidence of significant biological relevance to further filter candidate genes (Fig. 2A) and identified those genes that were (i) among the top-ranked shRNA hits, (ii) located in recurrent amplicons reported by TCGA (9) and therefore, likely to be tumor drivers, (iii) coexpressed with CCNE1 (Fig. S1 and Table S2), and (iv) located in pathways significantly enriched among hits (Gene Go analysis) (Fig. S2). A total of 115 genes met at least one of our additional selection criteria (Fig. S3). High confidence hits, meeting three or more filtering criteria (n = 25), are shown in Fig. 2B.

Fig. 2.

(A) Venn diagram of candidate genes essential for the survival of CCNE1-amplified (n = 474) and overexpressing (n = 486) cell lines identified in the shRNA synthetic lethal screen. Candidates were further filtered to include genes that were present in commonly amplified regions in ovarian tumors, coexpressed with CCNE1, present in significantly enriched gene pathways, or among top-ranking shRNA hits. (B) Top 25 ranking genes annotated by inclusion criteria. Statistical significance of ranking by second-best scoring shRNA or a composite score of all shRNAs (KS statistic) given (Materials and Methods). *Go term processes: cell cycle, DNA repair, or response to DNA damage.

Importantly, CCNE1 and CDK2, as well as other associated cell cycle genes, were ranked highly in our analysis, validating the experimental approach and supporting CDK2 as a key therapeutic target in CCNE1-amplified tumors (Fig. 2B). Other top-ranked hits included TPX2, a centromeric protein that maintains mitotic spindle integrity and genome stability (17). We have previously shown that the TPX2 locus, located at 20q11, is frequently coamplified with CCNE1 (5). Our findings support the view that cooperative amplification of 19q12 and 20q11 is important to the genesis and/or maintenance of CCNE1-amplified tumors. No other genes within the 19q12 amplicon, including URI1 (which was previously suggested to be a target of gene amplification) (12), were found in our analysis of the shRNA screen data. We also noted that genes involved in DNA damage response (DDR) were identified as essential genes in the screen, including BRCA1, XRCC2 (Fig. 2B), and ATR (Fig. S3). These findings suggest that chromosomal segregation and DDR mechanisms may be specific vulnerabilities of CCNE1-amplified or overexpressing cells. Finally, Ubiquitin-like modifier activating enzyme 1 (UBA1) was among the top 25 hits from the screen (Fig. 2B). We also observed that other members of the ubiquitin pathway, including Ubiquitin B (UBB), Ubiquitin C (UBC), and Ring-Box Protein 1 (RBX1), were among the top 115 ranked genes in the initial screen (Fig. S3), and they were, therefore, included in the validation studies.

Acute Effects of Gene Suppression.

We sought to validate hits from the shRNA screen using an orthogonal siRNA platform. Compared with the shRNA screen, shorter-term (5 d) siRNA experiments provide additional information on the acute effects of gene suppression on cell viability. We screened two cells lines, SK-OV-3 (CCNE1-unamplified cells) and OVCAR-3 (CCNE1-amplified cells), against a boutique siRNA library of 115 highly ranked hits from the shRNA screen (Fig. 2B and Fig. S3). Additionally, we included 27 candidate genes based on their biological relevance (Fig. S4). Selected candidates included cell cycle and DDR genes in addition to genes involved with processing and degradation of CCNE1 protein (15). Suppression of PLK1 was used as a positive (death) control (18). Cells were transfected with siRNA, and viability was measured 5 d after transfection. Volcano plots of P value significance against the effect on cell viability are shown for each cell line in Fig. S5, and all data are provided in Table S3.

The cell viability ratio between OVCAR-3 and SK-OV-3 plotted against the P value significance for OVCAR-3 cells highlights significant hits with highest specificity to the CCNE1-amplified cell line (Fig. 3A). Among these genes were CCNE1, CDK2, and BRCA1 and other genes involved with cell division and DNA damage response and repair (ATM, CHEK1, and SMC2). The greatest impact on cell viability in CCNE1-amplified cells was associated with suppression of genes involved in the ubiquitin pathway, including UBA1, UBC, and RBX1, in addition to CUL3 and FBXW7, which were included in the siRNA screen based on their roles in CCNE1 processing (15).

Fig. 3.

(A) Cells were transfected with a boutique siRNA library against 142 candidate genes, and the effect on cell viability was measured 5 d after transfection. Significance (t test P value) of hits in the OVCAR-3 (CCNE1-amplified) cell line plotted against the viability ratio of OVCAR-3 to SK-OV-3 (unamplified) highlights significant hits specific to OVCAR-3. Average data from duplicate wells across three independent experiments are shown (n = 3). The vertical dotted line is at P value = 0.05. (B) Clonogenic survival after siRNA transfection in SK-OV-3 (unamplified) and OVCAR-3 (CCNE1-amplified) ovarian cell lines. Average percentage of discrete colonies formed after 7 d relative to no siRNA controls is shown (n = 3 independent experiments performed in triplicate). Statistical significance (t test) was calculated by comparison with nonsilencing (NS) siRNA in the same cell line. *P value < 0.05; ***P value < 0.001. Error bars indicate SEM.

Although CCNE1 and CDK2 rank among the most significant hits in the OVCAR-3 cell line, the magnitude of the effect on cell viability was limited (Table S3). This finding is consistent with our previous studies that show a greater effect of gene suppression using siRNA (5) or CDK-specific inhibitors (7) in clonogenic survival assays compared with short-term viability assays. Similarly, for BRCA1, we showed a greater effect of inhibition of BRCA1 in clonogenic survival assays (∼50% reduction) than observed in the siRNA boutique screen (Fig. 3B and Fig. S5). Knockdown of each gene transcript was validated by real-time qPCR (Fig. S6).

Targeting Homologous Recombination and the Proteasome.

Dependence on BRCA1 suggests that intact HR function is required for the survival of CCNE1-amplified tumor cells. Fanconi Anemia (FA) pathway members coordinate multiple DNA repair mechanisms, including HR. Recently, a cell-based screen of over 16,000 compounds identified the proteasome inhibitor bortezomib as a potent inhibitor of the FA pathway and double-strand break repair by HR (19). It is thought that disruption of protein degradation by bortezomib either interferes with or is a requirement of FA pathway activity. The mutual exclusivity observed between HR pathway dysfunction and CCNE1 amplification and the dependence on genes involved with protein degradation suggested that CCNE1-amplified tumors may show selective sensitivity to bortezomib. We assessed ovarian cell line sensitivity to bortezomib in a panel of 10 cell lines (Fig. 4). FUOV-1 and OVCAR-3, which have high-level CCNE1 amplification and expression, showed the greatest sensitivity to bortezomib. A2780, which lacks CCNE1 amplification but strongly expresses CCNE1, was also among the most sensitive lines. By contrast, Kuramochi, with low-level CCNE1 gain, did not show heightened sensitivity to bortezomib. Interestingly, Kuramochi and IGROV-1, with reduced sensitivity to bortezomib, harbor mutations in BRCA2 and BRCA1, respectively (20). We also examined two cell lines derived from the OVCAR-3 parental line that are resistant to two CDK2-specific inhibitors PHA-533533 and dinaciclib (OVCAR-3-R1 and -RD1, respectively) (7). We found that both lines maintained sensitivity to bortezomib, suggesting a specific mechanism of resistance to CDK inhibitors in these cell lines.

Fig. 4.

Ovarian tumor cell line sensitivity to bortezomib ranked by average 72-h cytotoxicity assay IC₅₀ value (n = 3 independent experiments performed in triplicate). Error bars indicate SEM. CCNE1 copy number status was determined by qPCR, where copy number gain and amplification are defined as a log₂ ratio to normal > 0.5 and > 2.0, respectively. CCNE1 gene expression of each cell line above (high) or below (low) the median value of 10 parental lines is indicated. The OVCAR-3-R1 and -RD1 sublines were derived from OVCAR-3 and are resistant to CDK2 inhibitors PHA-533533 and dinaciclib, respectively (7).

Discussion

BRCA1/2 mutation is typically associated with platinum sensitivity and favorable clinical outcome (13). The absence of BRCA pathway disruption from CCNE1-amplified cancers may partly explain the relatively poor outcome observed in CCNE1-amplified tumors (9). Mutual exclusivity of oncogenic mutations in genes in the rat sarcoma viral oncogene homolog (RAS) signaling pathway, seen in low-grade serous ovarian cancer (21) and other solid cancers (22), seems to occur, because there is no selective advantage of compound mutations. By contrast, we found that BRCA1 suppression is not tolerated in cell lines that harbor CCNE1 amplifications. The specific requirement of BRCA1 compared with BRCA2 may relate to its wider roles in DNA repair as well as cell cycle regulation and checkpoint activation (11).

CCNE1 overexpression promotes unscheduled S-phase entry, disrupted DNA replication, and genomic instability (15), potentially rendering cells dependent on intact HR repair pathways. We also observed dependencies on genes involved in processing of CCNE1 and other components of protein degradation pathways. Synthetic lethality is seen as an important approach to the development of new cancer therapeutics, because it suggests treatments that are likely to offer a wide therapeutic index (23). Indeed, our findings suggest that the proteasome inhibitor bortezomib, either through attenuation of HR or other essential proteasome functions in CCNE1-amplified cells, offers a unique therapeutic approach in HGSCs and possibly other solid cancers. Additionally, the lack of cross-resistance to bortezomib in cells previously rendered partially resistant to CDK2 inhibitors (OVCAR3-R1/RD1) suggests that sequential treatment may be effective.

HGSC patients with high-level CCNE1 amplification represent an urgent unmet need given their high risk of treatment failure and low probability of response to PARP inhibitors because of absence of BRCA1/2 pathway dysfunction. Bortezomib is currently used in multiple myelomas and mantle cell lymphoma, but it has not shown significant activity in other solid cancers, including ovarian cancer (24). The low frequency of high-level amplification of CCNE1 in HGSCs may require specific patient selection to observe a therapeutic benefit. Here, a mutational interaction observed in patient samples is explained molecularly, and a unique treatment approach is defined.

Materials and Methods

Ethics Statement.

The Australian Ovarian Cancer Study was approved by the Human Research Ethics Committees at the Peter MacCallum Cancer Centre, Queensland Institute of Medical Research (QIMR), University of Melbourne and all participating hospitals. Written informed consent was obtained from all participants in this study.

Pan-Cancer Analysis of CCNE1 Copy Number.

Peak regions of copy number change at the CCNE1 locus identified by GISTIC (25) were obtained from the TCGA Copy Number Portal (http://www.broadinstitute.org/tcga) using all available data as of February of 2013 (dataset: 2013-02-21 stddata__2013_02_03). Analysis included a total of 6,547 tumor samples comprising 22 cancer types: breast invasive adenocarcinoma (n = 891), glioblastoma multiforme (n = 563), ovarian serous cystadenocarcinoma (n = 559), kidney renal clear cell carcinoma (n = 493), uterine corpus endometrioid carcinoma (n = 492), thyroid carcinoma (n = 430), colon adenocarcinoma (n = 413), lung adenocarcinoma (n = 403), lung squamous cell carcinoma (n = 358), head and neck squamous cell carcinoma (n = 322), stomach adenocarcinoma (n = 237), cutaneous melanoma (n = 236), brain lower-grade glioma (n = 220), prostate adenocarcinoma (n = 177), rectum adenocarcinoma (n = 162), bladder urothelial carcinoma (n = 150), kidney renal papillary cell carcinoma (n = 117), cervical squamous cell carcinoma (n = 114), liver hepatocellular carcinoma (n = 97), kidney chromophobe (n = 66), sarcoma (n = 29), and diffuse large B-cell lymphoma (n = 18).

BRCA1/2 and CCNE1 Status in Primary Ovarian Tumor Samples and Cell Lines.

We have previously published our analysis of germ-line BRCA1/2 status (13) and somatic analysis (14) of tumor samples from women enrolled in the Australian Ovarian Cancer Study. TCGA estimated the frequency of CCNE1 gain to be ∼8% and 26% in BRCA1/2-altered and WT cases, respectively (9). We calculated that analyses of ∼80 BRCA1/2-altered and 80 WT cases would give 80% power (sensitivity) to detect a difference between the two groups, where α = 0.05 (probability of a false-positive result). Our final cohort (n = 193) included 81 BRCA1/2 WT and 112 BRCA1/2-altered cases (Fig. 1B). The BRCA1/2-altered group included samples with germ-line BRCA1 mutations (n = 52), germ-line BRCA2 mutations (n = 29), somatic BRCA1 mutations (n = 5), somatic BRCA2 mutations (n = 4), and BRCA1 methylation (n = 22). CCNE1 copy number relative to normal female reference DNA (Novagen) and gene expression in cell lines was determined using qPCR and previously described methods (5).

TCGA SNP and Gene Expression Data.

Affymetrix SNP 6.0 and hthgu133a gene expression data were obtained for 157 serous tumors from TCGA (www.cancergenome.nih.gov). All SNP CEL files were normalized in a single batch using the R package aroma.affymetrix and then segmented using the circular binary segmentation algorithm to improve the signal-to-noise ratio (26). CCNE1 copy number was estimated using the mean segment value, and amplification was called for samples where the mean segment log₂ copy number ratio value was greater than 0.3. Recurrently amplified genes identified in the TCGA dataset have been previously published (9). Gene expression CEL files were normalized using the GCRMA package in R (27). Pearson correlation coefficient was computed between CCNE1 and all other probes in the genome. Genes that had an false discovery rate (FDR)-corrected P value <0.05 and a correlation coefficient >0.25 or <−0.25 were consider to be coexpressed or anticorrelated with CCNE1, respectively. We identified 501 genes that were coexpressed with CCNE1 (Table S2). Pathway analysis of genes coexpressed with CCNE1 using GeneGo (Thomson Reuters) revealed an enrichment of gene lists involved with cycle and DNA damage response pathways (Fig. S1).

shRNA Screen Data.

Cell line copy number data were obtained from the Cancer Cell Line Encyclopedia (20). CCNE1 copy number and gene expression status was assigned to each cell line with midrange samples removed from the analysis. Cell lines with a log₂ copy number ratio > 0.3 over the CCNE1 locus were defined as amplified (n = 23), and cell lines with a log₂ copy number ratio < 0 over the CCNE1 locus were defined as unamplified (n = 43). Cell lines with CCNE1 gene expression greater than median + 1 SD (n = 15) were classified as CCNE1 high expression, whereas cell lines with CCNE1 gene expression less than median (n = 41) were classified as CCNE1 low expression.

Microarray data from shRNA experiments was obtained from the Integrative Genomics Portal (http://www.broadinstitute.org/igp).

Data were analyzed using the GenePattern module ScorebyClassComp and GENE-E software (28). The weight of evidence statistic was used for class discrimination between CCNE1-amplified and unamplified cells and CCNE1 high- and low-expressing cell lines (16). Gene lists were created by collapsing shRNA ranks from each comparison, and then, they were ranked by significance to identify top hits as described previously (16). First, we selected the top 300 genes based on the second-best ranking shRNAs, and second, we selected the top 300 genes assessed using the Kolmogorov–Smirnov (KS) statistic, which uses a composite score for all shRNAs against each gene (28). The union of both analyses identified 474 essential genes in CCNE1-amplified cell lines and 486 essential genes in CCNE1 high-expressing cell lines (Fig. 2A and Table S1). Because each gene was targeted by multiple independent shRNAs (median n = 5 per gene), statistical assessment of essentiality may be affected by the number of shRNAs used per gene. For example, a higher number of shRNAs against CDK2 compared with CCNE1 and BRCA1 (n = 12, n = 4, and n = 7, respectively) may, in part, explain the higher ranked significance of CDK2 in our analysis (Table S1).

The candidate gene list was further refined using additional criteria as evidence of significant biological relevance, including if they were (i) among the top 10 shRNA hits ranked by either the second-best or KS method, (ii) located in recurrent minimal regions of amplification reported by TCGA (9), (iii) coexpressed with CCNE1 (see above), or (iv) located in pathways significantly represented among all hits (Gene Go analysis). In total, 115 genes met at least one of our additional selection criteria (Fig. 2B and Fig. S3) and were selected for validation studies.

Cell Lines.

Ovarian cell lines were obtained from the National Cancer Institute Repository and fingerprinted using short tandem repeat markers (29) to confirm identity against the Cancer Genome Project database (Wellcome Trust Sanger Institute). Cells were maintained at 37 °C and 5% (vol/vol) CO₂ in RPMI 1640 containing 10% (vol/vol) FCS, with transfection and drug sensitivity assays performed in the absence of antibiotics. Cell lines were confirmed to be mycoplasma-negative before siRNA studies.

siRNA Studies.

A boutique On-Target Plus siRNA library was obtained from Dharmacon (Thermo Fisher Scientific) in 384-well plates containing 142 candidate genes, a nonsilencing control, and a positive (death) control (PLK1). The library was hydrated and diluted to 1 µM in siRNA buffer (Dharmacon; Thermo Fisher Scientific). Cells were reverse transfected with DharmaFECT lipid reagents (Thermo Fisher Scientific) to a final concentration of 40 nM siRNA using SciClone ALH 3000 (Caliper Life Sciences) and BioTek 406 (BioTek) liquid handling robotics. Cell transfection densities were selected to achieve confluence 5 d after transfection (120 h; 800 cells per well for OVCAR-3; 500 cells per well for SK-OV-3). Selected transfection conditions allowed for efficient siRNA transfection with no impact on cell viability (OVCAR-3, 0.04 µL DharmaFECT1 per well; SK-OV-3, 0.06 µL DharmaFECT2 per well). During optimization experiments, nuclear localization of siGLO red RNA duplex (Dharmacon) was used to monitor transfection efficiency by fluorescence microscopy 24–48 h after transfection.

Cells were transfected in duplicate wells in three independent experiments. Cell viability was then assessed using the Cell Titer Glo luminescent assay (Promega) on the Synergy H4 high-throughput multimode microplate reader (BioTek). Average cell viability data for each gene were normalized to the average signal from control wells containing lipid only (n = 12 per plate). Significance of change in cell viability (log₂-transformed signal) compared with control wells was calculated using a t test (n = 3). Clonogenic survival assays and real-time qPCR were performed as described previously (5).

Sensitivity to Bortezomib.

Bortezomib was obtained from Millennium Pharmaceuticals, and drug sensitivity was determined using the CellTiter 96 Aqueous Non-Radioactive Cell Proliferation Assay (Promega) as described previously (5). IC₅₀ dose was approximated by fitting a four-parameter dose–response curve (Hill equation) using Prism 6 (GraphPad Software).