Pooled Genomic Screens Identify Anti-apoptotic Genes as Targetable Mediators of Chemotherapy Resistance in Ovarian Cancer

This article is featured in Highlights of This Issue, p. 2143

High-grade serous ovarian cancer (HGSOC) comprises ∼70% of ovarian cancers and is frequently diagnosed at an advanced stage; the five-year survival is <50% (1). Most patients with HGSOC respond to surgery and combination chemotherapy with platinum and taxanes, but >80% eventually develop recurrent disease and chemotherapy resistance, for which few effective therapies exist (1). HGSOC is characterized by TP53 mutations (nearly 100%) and defects in homologous recombination DNA repair (HRR), including BRCA1/2 mutations (1). HGSOC with HRR defects are more sensitive to platinum chemotherapy and poly (ADP-ribose) polymerase (PARP) inhibitors (1).

Numerous resistance mechanisms to platinum and taxanes have been reported in ovarian cancer, although their clinical significance is often unclear. Reversion mutations in BRCA1/2 and other genes involved in HRR have been reported to confer clinical resistance to platinum and PARP inhibitors (1, 2). In addition, recurrent fusions driving ABCB1 overexpression occur in platinum-resistant HGSOC (3); ABCB1 encodes MDR1 (multidrug resistance-1, P-glycoprotein), which mediates efflux of drugs including paclitaxel and some PARP inhibitors, leading to drug resistance (4).

Anti-apoptotic proteins have also been linked to chemotherapy resistance in ovarian cancer. Platinum and taxanes cause cell death primarily via the intrinsic pathway of apoptosis (5); activity of this pathway is restrained by BCL-2 family anti-apoptotic proteins (BCL-2, BCL-XL, BCL-W, MCL1, and BFL1; ref. 5). Increased BCL-XL protein expression was observed in recurrent compared with primary ovarian cancers (6) and was associated with clinical resistance to chemotherapy (7) and decreased survival (6, 7). BCL-2 overexpression correlated with poor responses to primary chemotherapy and decreased survival in ovarian cancer patients (8, 9), and MCL1 expression was also associated with poor prognosis (10). In ovarian cancer cell lines (including non–high-grade serous subtypes; ref. 11), enforced overexpression of BCL-XL conferred resistance to cisplatin or paclitaxel (6, 12, 13), and modulating MCL1 levels altered sensitivity to chemotherapy and targeted drugs (14–18). The role of BCL-W in ovarian cancer is unknown, though in other solid cancers BCL-W protects cells from drug-induced apoptosis (19). Targeting anti-apoptotic proteins with genetic knockdown of BCL-XL or with small-molecule inhibitors of BCL-2/BCL-XL or BCL-XL enhanced sensitivity to platinum or paclitaxel in ovarian cancer cell lines (7, 17, 20–24) and ex vivo patient samples (23, 24).

Despite the clinical use of platinum and taxanes for decades, and known mechanisms of resistance including reversion of HRR gene mutations, overexpression of ABCB1, and increased levels of certain anti-apoptotic proteins, the diverse landscape of chemotherapy resistance mechanisms in HGSOC remains incompletely understood. Correspondingly, few therapies are available to effectively target resistance pathways.

To investigate mechanisms of resistance to standard-of-care chemotherapy in HGSOC, we performed a pooled, near-genome-scale open reading frame (ORF) overexpression screen and a pooled genome-scale CRISPR-Cas9 knockout screen for mediators of resistance to cisplatin, paclitaxel, or cisplatin plus paclitaxel in HGSOC cell lines. In these unbiased screens, anti-apoptotic genes were among the most significant genes promoting resistance across all drug treatments. Based on this result, we systematically investigated the role of individual BCL-2 family anti-apoptotic proteins in HGSOC chemotherapy resistance and the effects of targeting these proteins on HGSOC survival.

Additional details are available in Supplementary Methods.

Kuramochi and OVSAHO cell lines were infected with a library of 17,255 individually barcoded human ORF clones in lentiviral vector pLX317 under an EF1a promoter (25, 26), at a multiplicity of infection of 0.5 to 1. Following puromycin selection, cells were collected for an early time point sample of initial library representation, then divided into drug treatment arms: DMSO, cisplatin (0.5 μmol/L), paclitaxel (10 nmol/L), or cisplatin + paclitaxel combination (0.5 μmol/L + 10 nmol/L, respectively) in quadruplicate and treated with drug every 3 to 7 days for 21 days. Surviving cells were collected for DNA extraction, PCR, and sequencing of barcodes. The total number of sequencing reads for each clone was determined, normalized to reads per million, and log₂ transformed. The log₂ fold change (LFC) of each ORF was determined relative to the early time point, and statistical significance (q-value) of candidate resistance ORFs was assigned using differential representation methods (see detailed methods). Candidate genes meeting the indicated criteria for LFC and q-values were further validated in a mini-pool secondary screen of 376 ORFs including candidate resistance genes and neutral and candidate sensitizing genes. Cells were infected with the mini-pool library and treated with drugs for 10 to 14 days, and collected for PCR and sequencing to identify resistance genes based on LFC and Z-score relative to the population.

Kuramochi and OVSAHO cells stably expressing Cas9 were infected with a library of 73,687 barcoded sgRNAs targeting 18,454 genes and 1,000 nontargeting guides (26, 27). Following selection and collection of cells for the early time point, cells were divided into four drug arms in quadruplicate [cisplatin (1 μmol/L), paclitaxel (10 nmol/L), or two combination doses (0.5 μmol/L + 10 nmol/L or 0.3 μmol/L + 5 nmol/L cisplatin + paclitaxel)], treated with drug every 3 to 4 days, and harvested at 14 days for PCR and sequencing of barcodes. The enrichment or depletion of each sgRNA was determined relative to the initial time point.

Lentiviral supernatant was produced by transient transfection of 293T cells with viral constructs and FuGene6 reagent. Cells were transduced with viral supernatant using polybrene and spin infection by centrifugation, followed by selection with puromycin (2 μg/mL) for 4 to 5 days. Protein overexpression was confirmed by Western blot using standard methods and protein-specific primary antibodies. For drug experiments, drug stocks were diluted in media and added to cells manually or by HP/Tecan d300e digital dispenser. Drugs included cisplatin solution 1 mg/mL in 0.9% NaCl; and 10 mmol/L solutions in DMSO of paclitaxel, PARP inhibitor olaparib, or anti-apoptotic protein inhibitors ABT-199/venetoclax, ABT-263/navitoclax, WEHI-539 hydrochloride, S63845, and A1331852. Cells were treated with a single dose of drug unless otherwise indicated, at physiologically relevant concentrations (28). For viability assays, cells were seeded at 10% to 20% confluence and 24 hours later drugs or vehicle were added at multiple doses in duplicate or triplicate; viability was assessed at 5 days using the CellTiter-Glo luminescent reagent. For apoptosis assays, cells were treated with drugs 24 hours after seeding, and 24 to 72 hours later, caspase-3/7 activity was evaluated by Caspase-Glo luminescent reagent. For colony formation assays, cells were seeded at low density and treated with vehicle or drug in triplicate once weekly for 14 to 21 days, fixed with 4% formaldehyde, and stained with 0.5% crystal violet. For population doubling assays, cells were seeded in duplicate, and counted and treated with drugs every 3 to 7 days for 14 to 21 days. Cell lines were genotyped and routinely tested for Mycoplasma. Data were analyzed using GraphPad Prism.

BH3 profiling was performed as previously described using flow cytometry (29). Briefly, permeabilized cells were incubated with BH3 domain proapoptotic peptides for 60 minutes and fixed in 4% to 8% formaldehyde followed by neutralization. Cells were stained with AlexaFluor 647-conjugated anticytochrome c antibody and a viability marker and analyzed by flow cytometry. For dynamic BH3 profiling, the same procedure was performed but was preceded by incubation of the cells with drug for 24 hours prior to collection.

HGSOC genomic data were obtained from the Cancer Genome Atlas (Pan-Cancer; TCGA; ref. 30), Australian Ovarian Cancer Study Group (3), and Cancer Cell Line Encyclopedia (31). TCGA genomic data were analyzed with R packages or the cBioPortal. Copy-number alterations were analyzed by GISTIC 2.0. Statistical differences in parameters between groups were analyzed by t test. Survival curves were generated by Kaplan–Meier analysis, and significance was assessed by log-rank test.

To identify genes whose overexpression affects response to cisplatin and paclitaxel treatment in ovarian cancer cell lines, we performed an unbiased, near-genome-scale, pooled ORF overexpression screen using a barcoded lentiviral library of 17,255 human cDNAs (Fig. 1A; ref. 25). We selected Kuramochi and OVSAHO cell lines due to genomic similarity to HGSOC (11) and relative sensitivity to cisplatin and paclitaxel (Supplementary Fig. S1A). Kuramochi has a BRCA2 mutation and BRCA1 copy loss, and OVSAHO has BRCA2 copy loss (11, 31); both are deficient in HRR (32).

After lentiviral infection and selection titrated to introduce a single barcoded cDNA to each cell, the pooled cells were cultured with DMSO, cisplatin (0.5 μmol/L), paclitaxel (10 nmol/L), or cisplatin plus paclitaxel (0.5 μmol/L + 10 nmol/L) for 21 days (Supplementary Fig. S1B). We collected the surviving cells posttreatment and performed sequencing of barcodes to determine representation of each cDNA clone in the post-treatment condition compared with the pre-treatment pool (Supplementary Table S1). We evaluated the statistical significance of the resistance conferred by each ORF by comparing the relative representation of each ORF clone at the pre-treatment and post-treatment time points to generate a log-fold change value and q-value (Supplementary Methods; Fig. 1B; Supplementary Table S2). Genes whose ORF-specific barcode showed higher abundance following drug treatment, compared with pre-treatment, were nominated as candidate resistance genes.

Among individual ORFs scoring highly in single or multiple drug treatments, ABCB1 was an outlier for paclitaxel treatment but not cisplatin treatment (Fig. 1B), consistent with the known role of MDR1 in paclitaxel resistance (4). Anti-apoptotic genes BCL2L1 (BCL-XL) and BCL2L2 (BCL-W) scored consistently across all three drug arms in both cell lines (Fig. 1B; Supplementary Fig. S1C). BCL2L1 and BCL2L2 also showed a stronger positive enrichment following drug treatment compared with DMSO controls, suggesting a drug-specific survival advantage (Supplementary Table S2). Additional high-scoring genes (Fig. 1B; Supplementary Fig. S1C) were identified, which have not been previously associated with drug resistance, e.g., OOSP2 (PLAC1L), encoding oocyte-specific protein 2.

We next aimed to test whether individual overexpression of top resistance genes from the pooled ORF screen conferred chemotherapy resistance in orthogonal assays, and whether inhibiting resistance genes could mitigate chemotherapy resistance. As proof-of-principle, we evaluated the known paclitaxel resistance gene ABCB1, which is a clinically validated resistance mechanism (3) and the top-scoring ORF in our paclitaxel screen. HGSOC cell lines infected with the ABCB1 ORF and overexpressing MDR1 (Supplementary Fig. S2A) showed increased viability following paclitaxel treatment (Supplementary Fig. S2B and S2C) and increased cell proliferation in the presence of paclitaxel in a population doubling assay (Supplementary Fig. S2D) and a colony formation assay (Supplementary Fig. S2E). However, treatment of ABCB1-overexpressing ovarian cancer cells with the MDR1 inhibitor elacridar (1 μmol/L) resensitized the HGSOC cells to paclitaxel (Supplementary Fig. S2E and S2F). Although ABCB1-mediated drug resistance has been difficult to exploit with clinical therapeutics (4), this example suggests that our unbiased overexpression screen could detect clinically relevant, targetable mechanisms of chemotherapy resistance.

To validate the primary overexpression screen, we performed a secondary screen with a mini-pool of 376 ORFs consisting of top candidate resistance genes balanced by neutral genes and candidate sensitizer genes. We selected 191 candidate resistance ORFs with log-fold change (LFC) >0.5 (enrichment after drug treatment) and q-value < 0.001 in any cell line plus drug condition (Supplementary Tables S2 and S3), plus 48 ORFs that scored in multiple cell line/drug conditions (LFC > 0.5 and q-value < 0.1 in ≥2 screen arms; Supplementary Table S3). ABCB1 was excluded from the mini-pool, enabling more complete assessment of other paclitaxel resistance genes. A lentiviral pool of selected candidate resistance, sensitizer, and neutral ORFs was transduced into Kuramochi and OVSAHO cells. Following selection, cells were treated with DMSO, cisplatin (0.5 μmol/L), paclitaxel (10 nmol/L), or cisplatin/paclitaxel (0.3 μmol/L + 5 nmol/L) for 10 to 14 days (Supplementary Fig. S3A). The surviving cells were collected for barcode sequencing to identify genes enriched (positive LFC) or depleted (negative LFC) compared with the pretreatment pool (Supplementary Table S4). Genes for which barcode abundance increased following drug treatment, relative to the abundance prior to drug exposure, were considered candidate resistance genes.

In this overexpression validation screen, anti-apoptotic genes BCL2L1, BCL2L2, and/or BCL2 were again enriched following drug treatment compared with pretreatment (Supplementary Fig. S3B; Supplementary Table S4). These three genes were among the top-ranked enriched genes by LFC in each drug condition. Notably, BCL2L1 and BCL2L2 were the top two enriched genes by sum of the ranks across all conditions (Supplementary Fig. S3C; Supplementary Table S4). Additional candidate resistance ORFs such as OOSP2 were enriched in multiple drug conditions, whereas many candidate sensitizing ORFs were depleted following drug treatment (Supplementary Fig. S3B and S3C; Supplementary Table S4). Together, the initial pooled ORF screen and secondary mini-pool screen identified consistent candidate genes whose overexpression increases viability of HGSOC cell lines following treatment with cisplatin and/or paclitaxel, including multiple BCL-2 family anti-apoptotic genes.

To complement the ORF gain-of-function screen, we performed a loss-of-function screen to identify genes whose loss affects cisplatin and paclitaxel resistance in HGSOC. We performed a genome-scale CRISPR-Cas9 knockout screen with the same HGSOC cell lines and drugs, using a pooled library of 73,687 barcoded sgRNAs targeting 18,454 genes (27). HGSOC cells stably expressing Cas9 were infected with the sgRNA lentiviral pool titrated to target one sgRNA per cell, then treated with cisplatin (1 μmol/L), paclitaxel (10 nmol/L), or cisplatin/paclitaxel (0.5 μmol/L + 10 nmol/L or 0.3 μmol/L + 5 nmol/L) for 14 days (Supplementary Fig. S4A). Using the sequencing reads of each uniquely barcoded sgRNA, we calculated LFC in abundance for each sgRNA in the surviving cells compared with the pretreatment pool, and the q-value for statistical enrichment or depletion compared with the population (Fig. 1C; Supplementary Fig. S4B; Supplementary Table S5). We identified candidate genes whose sgRNA-mediated knockout promoted cell survival (positive LFC) and candidate genes whose knockout correlated with cell death (negative LFC; Fig. 1C). Candidate genes were also ranked with the STARS algorithm (http://portals.broadinstitute.org/gpp/public/software/stars; Supplementary Table S6).

Genes whose knockout were associated with increased cell survival, and therefore may normally mediate drug-induced killing, included proapoptotic caspase enzymes CASP2 (caspase-2), CASP9 (caspase-9), and APAF1 (apoptotic peptidase activating factor; Fig. 1C; Supplementary Fig. S4B; Supplementary Tables S5 and S6). Conversely, BCL2L1 was among the top genes whose knockout promoted cell death (and thus may be associated with chemotherapy resistance when intact; Fig. 1C; Supplementary Fig. S4B; Supplementary Tables S5 and S6). This independent, genome-scale loss-of-function screen supports the significance of the intrinsic pathway of apoptosis in HGSOC resistance to chemotherapy. Other genes for which knockout was associated with response to individual drugs included spindle assembly checkpoint genes associated with paclitaxel response (33) and cisplatin importer LRRC8A associated with cisplatin resistance (34), confirming that the screen identified genes relevant to chemotherapy response.

Taken together, our systematic gain-of-function and loss-of-function screens revealed numerous known and novel genetic mediators of cisplatin and paclitaxel response in HGSOC cell lines, and supported anti-apoptotic signaling as a candidate pathway that could be targeted to eliminate chemotherapy-resistant HGSOC.

Because our overexpression and knockout screens converged upon anti-apoptotic proteins in the intrinsic pathway of apoptosis as key mediators of both platinum and taxane resistance in HGSOC, and because these proteins are targets of clinically relevant small-molecule inhibitors, we decided to further analyze BCL-2 family anti-apoptotic proteins in HGSOC resistance to chemotherapy.

First, to evaluate the role of individual anti-apoptotic proteins in driving HGSOC chemotherapy resistance, we tested the effects of overexpression of anti-apoptotic genes in HGSOC cell lines. We first focused upon BCL2L1 and BCL2L2 as the top-ranked resistance candidates (Supplementary Fig. S3C). BCL2L1 encodes a primary anti-apoptotic isoform, BCL-XL, and a minor proapoptotic isoform, BCL-XS. BCL2L2 encodes BCL-W, which was not previously implicated in HGSOC drug resistance. We lentivirally infected HGSOC cell lines with ORF constructs encoding BCL2L1 and BCL2L2 and confirmed protein overexpression (Supplementary Fig. S5A). Cells overexpressing BCL2L1 or BCL2L2 showed decreased apoptosis by caspase-3/7 activation at 48 hours after cisplatin or paclitaxel treatment, though this difference was not statistically significant for BCL2L2 in Kuramochi cells (Fig. 2A). Overexpression of BCL2L1 or BCL2L2 increased cell viability at 5 days after paclitaxel treatment, and modestly increased viability after cisplatin treatment, in Kuramochi, OVSAHO, and other HGSOC cell lines (Fig. 2B; Supplementary Fig. S5B).

We next applied BH3 profiling as a functional assay of mitochondrial priming for apoptosis, in which cytochrome c release from mitochondria is measured following stimulation with BH3-domain proapoptotic peptides (29, 35). In prior BH3 profiling studies of primary ovarian cancers, greater pretreatment apoptotic priming correlated with improved chemotherapy response and progression-free survival (35), and increased priming following carboplatin treatment correlated with clinical response to carboplatin and paclitaxel (29), supporting that apoptotic priming may affect chemotherapy response in HGSOC. Overexpression of BCL2L1 markedly decreased apoptotic priming following BIM proapoptotic peptide treatment of Kuramochi and OVSAHO cells, and overexpression of BCL2L2 slightly diminished priming (Supplementary Fig. S5C), indicating that overexpression of anti-apoptotic proteins can reduce mitochondrial priming in HGSOC.

We also tested genes encoding other key BCL-2-family anti-apoptotic proteins, BCL2 and MCL1, using N-terminal FLAG-tagged lentiviral ORF constructs to overexpress each protein in our HGSOC cell lines (Supplementary Fig. S5D). BCL2 or MCL1 overexpression decreased caspase-3/7 activity after paclitaxel treatment, which was statistically significant in OVSAHO cells (Supplementary Fig. S5E). With cisplatin treatment, BCL2 overexpression decreased caspase-3/7 activity, whereas MCL1 overexpression had minimal effect (Supplementary Fig. S5E). BCL2 or MCL1 overexpression increased viability following paclitaxel treatment, but did not markedly affect response to cisplatin (Supplementary Fig. S5F). In the ORF screen, a BCL2 ORF scored highly in the primary screen and mini-pool (Fig. 1B; Supplementary Fig. S3B; Supplementary Tables S1–S4), whereas MCL1 did not score in the screen nor mini-pool (potentially a false negative due to screen conditions or translation efficiency of the ORF construct, especially as MCL1 has a short half-life). The effects on cell viability did not seem to be mediated by slower cell-cycle progression, as untreated cells overexpressing each anti-apoptotic protein had similar cell-cycle profiles as parental cells (Supplementary Fig. S6).

Collectively, these data suggest that overexpression of specific anti-apoptotic proteins confers chemotherapy resistance via decreased apoptosis. Of note, overexpression of anti-apoptotic genes occurs frequently in HGSOC tumors, including focal amplifications and copy gains of BCL2L1 and MCL1 in patient samples and cell lines (Supplementary Fig. S7 and S8; ref. 36), although such events were not significantly increased in a small cohort of chemotherapy-resistant HGSOC patients (Supplementary Fig. S9). Hence, overexpression of anti-apoptotic proteins—particularly BCL-XL and MCL1—may contribute to cancer cell survival in HGSOC patients.

Because increased levels of anti-apoptotic proteins appear to promote platinum and taxane resistance in HGSOC cell lines, we next investigated whether, conversely, inhibiting anti-apoptotic proteins increases cell death, either with or without concomitant chemotherapy. Inhibitors of anti-apoptotic proteins are in preclinical and clinical development, including BCL-2 inhibitor venetoclax (ABT-199), BCL-2/BCL-XL inhibitor navitoclax (ABT-263), MCL1 inhibitors (AMG176, S64315, and AZD5991), and BCL-XL inhibitors (WEHI-539 and A1331852; refs. 21, 37, 38).

First, we examined the effects of inhibition of individual BCL-2 family anti-apoptotic proteins on survival of ovarian cancer cell lines in the absence of chemotherapy. In our panel of ovarian cancer cell lines, individual inhibitors including ABT-199, ABT-263, A1331852, and S63845 had little effect on viability at 5 days after treatment with concentrations less than 2 μmol/L (Supplementary Fig. S10A).

Next, we addressed whether inhibiting anti-apoptotic proteins may enhance chemotherapy-induced apoptosis. We tested whether treatment with specific anti-apoptotic protein inhibitors alters the sensitivity of Kuramochi or OVSAHO cells to chemotherapy by treating cells with a range of doses of cisplatin or paclitaxel, combined with each anti-apoptotic protein inhibitor at a series of concentrations (0–2 μmol/L) that do not significantly affect viability as single agents (Supplementary Fig. S10A). Adding BCL-XL, BCL-2/BCL-XL, or MCL1 inhibitors to cisplatin or paclitaxel treatment decreased viability of HGSOC cells, whereas inhibition of BCL-2 alone had little effect (Figs. 3 and 4). Hence, inhibition of BCL-XL or MCL1, but not BCL-2, increased cell killing of HGSOC in combination with cisplatin or paclitaxel, and may represent an effective therapeutic strategy.

We applied an orthogonal approach, dynamic BH3 profiling, to probe the effects of combining anti-apoptotic protein inhibition by specific proapoptotic peptides with chemotherapy treatment (29, 35). We applied BH3 profiling with promiscuous peptides that inhibit all BCL-2 family anti-apoptotic proteins (BIM, PUMA), or selective peptides that only inhibit a subset of proteins [HRK (BCL-XL), MS1 (MCL1), BAD (BCL-2, BCL-XL, BCL-W)], to interrogate anti-apoptotic dependencies in HGSOC cell lines. Kuramochi cells showed cytochrome c release upon exposure to BIM, PUMA, and BAD but not MS1, indicating mild dependence on BCL-2, BCL-XL, and/or BCL-W, whereas OVSAHO cells had very low cytochrome c release with all peptides, suggesting inherent apoptosis resistance (Supplementary Fig. S10B). Treatment with carboplatin (up to 100 μg/mL for 24 hours) slightly increased BIM- and BAD-induced cytochrome c release in Kuramochi but not OVSAHO cells (Supplementary Fig. S10B), demonstrating that proapoptotic signaling is being induced by this treatment and thereby increasing dependence on prosurvival BCL-2 family proteins. We next compared these responses to those of the HGSOC cell line OVCAR3, which is primed for apoptosis. We found that cytochrome c release increased in response to BIM, BAD, and HRK peptides when cells were treated with carboplatin for 24 hours (Supplementary Fig. S10B). These BH3 profiles support that HGSOC cell lines can exhibit a functional apoptotic dependency upon specific anti-apoptotic proteins, and that this dependency can be further increased by carboplatin treatment in some contexts. Combination treatment strategies with chemotherapy and specific anti-apoptotic protein inhibitors can potentially exploit these induced dependencies.

To investigate factors determining response to combination treatment with chemotherapy plus anti-apoptotic protein inhibitors, we tested the impact of overexpressing each anti-apoptotic protein on the efficacy of the combinations. Upon overexpression of the target protein of each inhibitor (e.g., BCL-XL overexpression with A1331852), sensitivity to the combination treatment was maintained or enhanced (Supplementary Fig. S11). However, overexpression of a nontargeted anti-apoptotic protein mitigated the sensitizing effect of the inhibitor on chemotherapy response (e.g., MCL1 overexpression reduced the efficacy of a chemotherapy plus BCL-XL inhibitor combination; Supplementary Fig. S11). These data support that baseline expression levels and/or genomic alterations in different members of the BCL-2 anti-apoptotic protein family (Supplementary Fig. S7–S9) may moderate the response to combination regimens with anti-apoptotic protein inhibitors, and therefore may be candidate biomarkers of effectiveness of these treatments (16, 18).

Taken together, these data suggest that combining anti-apoptotic protein inhibition, particularly specific inhibitors of BCL-XL or MCL1, with platinum or taxane chemotherapy may promote apoptotic cell death in HGSOC and may represent a promising therapeutic strategy in certain patients.

Apoptosis is the final common pathway of cell death induced by multiple classes of drugs beyond platinum and taxanes. To examine correlations between anti-apoptotic gene expression and drug response, we used gene expression and drug-sensitivity data for a large panel of cancer cell lines (39). Interestingly, BCL2L1 expression level was the top predictor of resistance to a variety of drugs in a panel of solid tumor cell lines (Fig. 5A). Drugs with high correlations between BCL2L1 expression and resistance included microtubule inhibitors such as paclitaxel and vincristine, DNA crosslinkers (mitomycin C), anthracyclines (doxorubicin), and inhibitors of cyclin-dependent kinases (CDK). The correlation between BCL2L1 expression and multidrug resistance was maintained when considering only ovarian cancer cell lines. These data suggest that BCL-XL may mediate resistance to multiple clinically relevant therapeutics in ovarian cancer.

Due to the clinical significance of PARP inhibitors in ovarian cancer, we asked whether inhibition of BCL-XL or other anti-apoptotic proteins may sensitize to these drugs. Prior studies showed that ABT-263 is synergistic with the PARP inhibitor BMN-673 (talazoparib; ref. 40), and BCL-2/BCL-XL inhibitors ABT-263 and ABT-737 also synergize with several other targeted agents in HGSOC cell lines, including the CDK inhibitor dinaciclib (41), PI3K/mTOR inhibitors (16, 42), and MEK inhibitors (43). We interrogated the effects of overexpression or inhibition of each BCL-2 family anti-apoptotic protein on the response to the PARP inhibitor olaparib. In our HGSOC cell lines, overexpression of several anti-apoptotic proteins resulted in a small increase in viability following olaparib exposure compared with parental cells, as did ABCB1 (olaparib is an MDR1 target; Supplementary Fig. S12A). Similar to the data for cisplatin and paclitaxel, inhibitors of BCL-XL, MCL1, or BCL2/BCL-XL, but not BCL-2 alone, increased sensitivity of HGSOC cells to olaparib (Fig. 5B), though concomitant overexpression of anti-apoptotic proteins mitigated this effect (Supplementary Fig. S12B). These data support that BCL-2 family anti-apoptotic proteins may mediate resistance to multiple clinically relevant drugs, and that combination strategies with BCL-XL or MCL1 inhibition may enhance drug efficacy in HGSOC.

Chemotherapy resistance is an important determinant of poor outcomes in HGSOC, but the mediators of chemotherapy resistance are incompletely understood. Large-scale functional genomic screens are an efficient, unbiased approach to defining the landscape of drug-resistance mechanisms in cancer. Several genomic screens have explored chemotherapy resistance factors in a variety of cancers, revealing candidate mediators of taxane resistance such the spindle assembly checkpoint (33), and mediators of platinum resistance such as DNA repair proteins. Screens focusing on ovarian cancer platinum and taxane resistance include an shRNA screen in BRCA2-mutant HGSOC identifying loss of CHD4 as a cisplatin resistance factor (44), a CRISPR-Cas9 screen identifying loss of DYNLL1 as a mediator of platinum and PARP inhibitor resistance in BRCA1-mutant HGSOC cell lines (45), a kinome siRNA screen identifying loss of ATR and CHK1 checkpoint proteins as cisplatin sensitizers (46), and an siRNA screen showing that microtubule regulators affect paclitaxel responses (47).

In this study, we performed parallel near-genome-scale pooled ORF and genome-wide pooled CRISPR-Cas9 screens in BRCA2-altered HGSOC cell lines to identify genes whose overexpression or knockout enhances survival following treatment with cisplatin, paclitaxel, or cisplatin plus paclitaxel. We identified candidate genes promoting resistance to platinum and taxane chemotherapy in HGSOC. BCL-2 family anti-apoptotic proteins were among the most strongly enriched resistance genes in each screen, highlighting the critical role of this pathway in mediating HGSOC response to chemotherapy. Enforced overexpression of BCL2L1 and BCL2L2, and to a lesser extent BCL2 and MCL1, decreased cell death in HGSOC cells following cisplatin and paclitaxel treatment, suggesting that elevated levels of anti-apoptotic proteins can promote chemotherapy resistance in HGSOC. The differences in viability in short-term assays were relatively modest, possibly due to baseline or enforced expression levels of anti-apoptotic proteins, intrinsic resistance to platinum and taxanes, the balance of proapoptotic and anti-apoptotic proteins, or the involvement of other cell death pathways. Nonetheless, the pooled screen over a longer time frame may better model survival of HGSOC cells following repeated exposure to chemotherapy than the short-term viability assay. Our study provides the first evidence for BCL2L2 (BCL-W) as a potential resistance factor in ovarian cancer. Though BCL2L2 is expressed at lower levels than BCL2L1 and MCL1 in ovarian cancers, it may affect drug responses if aberrantly overexpressed, and further study of its role in apoptosis and drug resistance may be warranted. Our study and others suggest that anti-apoptotic proteins comprise a clinically relevant chemotherapy resistance pathway that can potentially be targeted in HGSOC to mitigate resistance.

From a therapeutic perspective, a subset of ovarian cancers may be vulnerable to inhibition of individual anti-apoptotic proteins, particularly BCL-XL and, less frequently, MCL1. A phase II study of single-agent BCL-2/BCL-XL inhibitor ABT-263 (navitoclax) showed modest activity in heavily pretreated patients with recurrent ovarian cancer (48). To more specifically target the key anti-apoptotic proteins in HGSOC, recently developed inhibitors selective for BCL-XL or MCL1 may merit preclinical investigation in ovarian cancer, along with identification of biomarkers to select patients likely to respond to single-agent inhibitors. For the majority of ovarian cancer patients, however, strategies combining anti-apoptotic protein inhibitors with chemotherapy or targeted agents may be the most efficacious approach to increase apoptotic cell death in ovarian cancer cells. Previous studies showed that BCL-2/BCL-XL and BCL-XL inhibitors sensitize ovarian cancer cells to platinum or taxanes (7, 17, 20–24) but BCL-2 inhibition does not (20). We used a panel of specific inhibitors to show that inhibition of BCL-XL, MCL1, or BCL-2/BCL-XL can increase the response of HGSOC cells to cisplatin and paclitaxel. However, BCL-2 inhibition did not have this effect, consistent with the fact that BCL2 is not highly expressed in ovarian cancer, thus BCL-2 is likely not an optimal therapeutic target for combination treatment of HGSOC.

Although we initially focused upon chemotherapy resistance, our data and others suggest that anti-apoptotic proteins may also prevent cell death due to targeted agents, and that inhibitors of anti-apoptotic proteins can enhance sensitivity to targeted drugs. In ovarian cancer, effective combination strategies have been demonstrated between BCL-2/BCL-XL inhibition and PARP inhibitors (40), CDK inhibitors (41), PI3K/mTOR inhibitors (16, 42), and MEK inhibitors (43); in some cases, the combinations were effective against cell lines resistant to the targeted drugs. Our data supported that specific inhibition of BCL-XL or MCL1 enhanced response to the PARP inhibitor olaparib in HGSOC cell lines, suggesting a potential therapeutic combination, if hematologic toxicities can be managed. For example, thrombocytopenia related to ABT-263 may be improved by pretreatment with an anti-apoptotic protein inhibitor or by staggered administration of two agents (49).

Synergy between chemotherapy and anti-apoptotic protein inhibitors may be limited by overexpression of other anti-apoptotic proteins in the BCL-2 family, which can confer cross-resistance, as observed in our study and others (16, 18). Combinations of two anti-apoptotic protein inhibitors may overcome the resistance conferred by elevated levels of anti-apoptotic proteins. For example, a combination of ABT-263 and an MCL1 inhibitor was effective in sensitizing HGSOC patient–derived xenograft cells to PI3K/mTOR inhibition, especially in cells with higher MCL1 expression (16), and combined siRNA to BCL-XL– and MCL1-sensitized ovarian cancer cell lines to cisplatin (50). Further studies are needed to identify the best biomarkers for response to anti-apoptotic protein inhibitors and combinations, including anti-apoptotic protein levels and functional assays such as BH3 profiling.

In summary, functional genomic screens for mediators of cisplatin and paclitaxel resistance highlighted the importance of the intrinsic pathway of apoptosis in HGSOC response to chemotherapy. Systematic overexpression and inhibition of BCL-XL, BCL-W, MCL1, or BCL-2 demonstrated that overexpression of anti-apoptotic proteins can increase chemotherapy resistance, whereas inhibition of BCL-XL or MCL1 may increase apoptosis in HGSOC cells treated with cisplatin, paclitaxel, or PARP inhibitors. BCL-XL and MCL1 appear to be the most critical anti-apoptotic proteins in HGSOC cell lines and the genes encoding these proteins are amplified and overexpressed in a subset of HGSOC; BCL-2 or BCL-W may be contributory in some contexts. Targeting anti-apoptotic proteins, particularly BCL-XL and/or MCL1, in combination with chemotherapy or other agents may be a promising therapeutic strategy in ovarian cancer.

P.D. Bhola has ownership interest in a patent related to BH3 profiling. J. Ryan is a consultant for Vivid Bioscience, has ownership interest (including patents) in Dana-Farber Cancer Institute. J.L. Guerriero is a consultant for GlaxoSmithKline and reports receiving a commercial research grant from the same. J. Montero is a consultant for Oncoheroes Biosciences and Vivid Biosciences. A.D. Cherniack reports receiving other commercial research support from Bayer AG. A. Letai is a consultant for AstraZeneca, reports receiving a commercial research grant from Novartis and AbbVie, has ownership interest (including patents) in Vivid Biosciences, Flash Therapeutics, Dialectic Therapeutics, and Zeno, has provided expert testimony for AbbVie. L.A. Garraway is Senior Vice President at Eli Lilly, has ownership interest (including patents) in Eli Lilly and Tango Therapeutics, and is a consultant/advisory board member for Tango Therapeutics. C.M. Johannessen is employed at Novartis Institutes of Biomedical Research and is a consultant/advisory board member for Pediatric Low Grade Glioma Foundation, Scientific Advisor. M. Meyerson is Chair, Scientific Advisory Board for OrigiMed and reports receiving a commercial research grant from Bayer, Ono, and Janssen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E.H. Stover, G. Wei, F. Piccioni, A. Letai, C.M. Johannessen

Development of methodology: E.H. Stover, M.B. Baco, J. Ryan, J.L. Guerriero, F. Piccioni, C.M. Johannessen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.H. Stover, M.B. Baco, E.L. Christie, A. Goodale, Y. Lee, S. Pantel, G. Wei, A.G. Presser, M.K. Gelbard, W. Zhang, J. Ryan, J.L. Guerriero, F.J. Liang, F. Piccioni, U.A. Matulonis, D.D.L. Bowtell, K.A. Sarosiek, L.A. Garraway

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.H. Stover, M.B. Baco, O. Cohen, Y.Y. Li, E.L. Christie, M.G. Rees, G. Wei, A.G. Presser, I.K. Zervantonakis, J. Ryan, J.L. Guerriero, F.J. Liang, A.D. Cherniack, F. Piccioni, U.A. Matulonis, D.D.L. Bowtell, K.A. Sarosiek, M. Meyerson

Writing, review, and/or revision of the manuscript: E.H. Stover, M.B. Baco, Y.Y. Li, E.L. Christie, M.G. Rees, G. Wei, I.K. Zervantonakis, J.L. Guerriero, A.D. Cherniack, F. Piccioni, U.A. Matulonis, D.D.L. Bowtell, K.A. Sarosiek, A. Letai, C.M. Johannessen, M. Meyerson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Bagul, P.D. Bhola, J. Montero, F. Piccioni, U.A. Matulonis, C.M. Johannessen

Study supervision: E.H. Stover, A. Letai, L.A. Garraway, C.M. Johannessen, M. Meyerson

We acknowledge collaborators for helpful discussions: Joan Brugge, Dipanjan Chowdhury, Claudia Iavarone, Whitney Henry, Colles Price, Kentaro Ito, Vasanthi Viswanathan, Alex Spektor, Daniel Stover, Panagiotis Konstantinopoulos, Joyce Liu, Kyle Davis, Uri Ben-David, Justin Hwang; John Daley, Suzan Lazo, and the DFCI Flow Cytometry Core Facility; and members of the Garraway, Meyerson and Letai laboratories, the Broad Institute Cancer Program, and the Boston HGSOC working group. This work was supported by the KL2/Catalyst Medical Research Investigator Training award (an appointed KL2 award) from Harvard Catalyst [The Harvard Clinical and Translational Science Center (National Center for Advancing Translational Sciences, NIH Award KL2TR001100)], Rivkin Center for Ovarian Cancer Scientific Scholar Award, American Society of Clinical Oncology Young Investigator Award, Kathryn Fox Samway Foundation, and Susan F. Smith Fellowship to E.H. Stover; Victorian Cancer Agency Early Career Seed Grant (ECSG15012) and National Health and Medical Research Council Project Grant (APP1124309) to E.L. Christie; Stand Up To Cancer (SU2C) and The V Foundation (TVF) SU2C-TVF Convergence Scholar Awards (D2015-037) and Ramon y Cajal Programme, Ministerio de Economia y Competitividad (RYC-2015-18357) to J. Montero; Ruth L. Kirschstein Postdoctoral Individual National Research Service Award F32 (F32CA180733) to J.L. Guerriero; NCI K99 (K99CA222554) to I.K. Zervantonakis; Breast Cancer Research Foundation to U.A. Matulonis; National Health and Medical Research Council of Australia Program Grant (APP1092856) and NHMRC Research Fellowship Grant (APP1117044) to D.D.L. Bowtell; NCI Clinical and Translational Science Award U54 (1U54CA224068) and R01 (1R01CA219943) to C.M. Johannessen; NCI R35 (1R35CA197568) and an American Cancer Society Research Professorship to M. Meyerson. This work was conducted as part of the Slim Initiative for Genomic Medicine, a project funded by the Carlos Slim Foundation in Mexico.

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.