Combined MEK and BCL-2/X_L Inhibition Is Effective in High-Grade Serous Ovarian Cancer Patient–Derived Xenograft Models and BIM Levels Are Predictive of Responsiveness

This article is featured in Highlights of This Issue, p. 495

High-grade serous ovarian cancer (HGSOC) is the most commonly diagnosed subtype of ovarian cancer and accounts for the majority of deaths from this disease (1, 2). Despite an initial response to platinum-based chemotherapy regimens, most patients with late-stage disease will experience cycles of relapse and development of treatment resistance ultimately leading to death. Improvement of HGSOC survival has been minimal over the past decade (3, 4). Therefore, the identification of new effective therapies and validated predictive biomarkers for patient selection represents an urgent, unmet clinical need.

The identification of targetable genetic drivers for HGSOC has been difficult due to the fact that somatic point mutations in oncogenes or tumor suppressors other than TP53 are relatively uncommon in HGSOC. Instead, HGSOC is largely characterized by chromosomal instability with a large burden of copy-number gains and losses (5, 6). Genetic and proteomic characterization of HGSOC has identified aberrantly activated protumorigenic signaling pathways that represent attractive targets for therapeutic intervention (7, 8). However, the effectiveness of any single targeted therapy might be limited by the ability of tumor cells to develop adaptive mechanisms of resistance (9). This possibility is particularly true in HGSOC due to its genomic complexity (10–13). Thus, one promising approach is to identify and block adaptive responses to targeted agents through rationally designed combination therapies that can anticipate and overcome drug resistance (14).

The RAS/MEK/ERK pathway is frequently activated in HGSOC through copy-number alterations in key members of the pathway, including NF1 deletions or gene breakages, and KRAS and BRAF amplifications (5, 6, 15). In addition, proteomic analyses of patient tumors have revealed that a subset of HGSOC exhibit MEK pathway activation as measured by the levels of phosphorylated forms of key nodes of the pathway, such as phospho-ERK (pERK). MEK pathway activation has also been implicated as an independent prognostic factor and predictor of early platinum-resistant relapse (16–18). These findings implicate the MEK/ERK pathway as a target in HGSOC.

MEK inhibition has been evaluated as a therapeutic strategy in several tumor types characterized by activating mutations in KRAS or BRAF such as melanoma but also including low-grade serous ovarian cancer (LGSOC; refs. 19–21). The modest clinical activity of MEK inhibitors as single agents in these studies is, at least in part, attributable to activation of compensatory survival pathways that lead to cell adaptation and drug resistance (22, 23). Of note, clinical responses were observed in both wild-type and mutant KRAS or BRAF tumors in a phase II clinical trial of the MEK inhibitor selumetinib in LGSOCs (24), suggesting that MEK inhibition can be effective in tumor types such as HGSOC that lack strong oncogenic drivers of this pathway. Despite the growing evidence of activation of the MEK pathway in HGSOC and its potential as a therapeutic target in this disease, a comprehensive assessment of MEK inhibitors and the adaptive responses they induce remains to be addressed in robust preclinical models of HGSOC.

Preclinical studies of therapeutic agents for HGSOC have largely relied on established ovarian cancer cell lines as model systems. However, significant differences in the molecular profiles have been reported between the most commonly used ovarian cancer cell lines and HGSOC tumors (25), calling into question the suitability of many of these lines as clinically relevant models for HGSOC. Alternatively, patient-derived xenografts (PDX) faithfully recapitulate the tumor of origin with respect to clinical pathology, degree of genetic heterogeneity and, importantly, response patterns to therapeutic regimens (26–28). Thus, PDXs represent a valuable clinical tool to assess efficacy of novel targeted therapies and identify predictive biomarkers to guide patient selection.

Here, we evaluated the effect of inhibiting the MEK/ERK signaling pathway in preclinical PDX models of HGSOC and identified a drug combination that exploited an adaptive response to MEK/ERK inhibition that primed cells for inhibition of BCL-2 and BCL-X_L. In addition, we utilized detailed proteomic and correlation analyses to define markers that correlate with drug combination sensitivity in different subsets of PDX models.

Ascites cells were harvested from PDX models and were allowed to recover for 4 days in vitro in primary ovarian growth medium previously described (29). Cells were seeded in 96-well plates in MCDB105 + M199 medium, supplemented with 2% HI-FBS and 1% penicillin/streptomycin. Cells were treated with the indicated doses of GDC-0973, ABT-263, PD0325901, or MEK162 for 96 hours. Dose-response experiments were performed starting from a drug dose of 3 μmol/L followed by 3-fold serial dilutions for five dose ranges (dose range, 0.03–3 μmol/L). In the drug combination studies, the same dose range was used, and the two drugs were combined at 1:1 ratio. DMSO was used as control. At the end of the treatment period, the relative cell number was determined by luciferase assay. D-luciferin (Fisher Scientific) was added to the cells in the 96-well plate, cells were incubated for 3 minutes at room temperature, and luciferase signal was measured using bioluminescence imaging (Envision 3). To determine growth rate, a parallel plate was read on the day of drug addition to calculate the luciferase signal before treatment (Day 0). For cell death analysis, after 96 hours of treatment, cells were incubated with NucGreen dye (Invitrogen) to identify dead cells, and the 96-well plate was read on a laser scanning cytometer (TTP LabTech). The percentage of cell death was calculated by dividing the number of dead cells (green) by total cells (red plus green). Dose-response curves were generated in GraphPad Prism by plotting the relative cell number (normalized to the DMSO-treated condition 96 hours after drug treatment) for each drug concentration. IC₅₀ was determined from dose-response curves using GraphPad Prism by fitting a four-parameter, variable slope dose-response curve. A Bliss-independence model was used to evaluate combination effects. The Bliss expectation was calculated with the equation (A + B) − A × B, in which A and B are the fractional growth inhibitions induced by agents A (i.e., GDC-0973) and B (i.e., ABT-263) at a given dose, respectively. The difference between the Bliss expectation and the observed growth inhibition induced by the combination of agents A and B at the same dose is the Bliss excess and is indicated as Bliss score. Genentech and AbbVie provided the compounds GDC-0973 and ABT-263, respectively. The compounds MEK162, PD0325901, and Z-VAD-FMK were purchased from Selleckchem. The established ovarian cell lines OVCAR3, OV90, OAW28, OVCAR8, OVSAHO, OVCAR5, TKYNU, JHOM1, and RMUGS were obtained from Dr. Dennis Slamon (UCLA) and were maintained in MCDB105/M199 medium supplemented with 10% HI-FBS and 1% penicillin/streptomycin (Invitrogen). Cell lines were reauthenticated using STR profiling. The in vitro drug dose-response experiments for these cell lines were performed as described for the PDX models. PDX models and established ovarian cancer cell lines tested negative for Mycoplasma contamination using luminescence assay.

All mouse studies were conducted through Institutional Animal Care and Use Committee (IACUC)–approved animal protocols in accordance with Harvard Medical School institutional guidelines. For drug efficacy studies, 8- to 10-week-old female NSG mice (Jackson labs) were injected intraperitoneally with approximately 5 × 10⁶ cells/200 μL of PBS. Tumor burden was monitored using bioluminescence imaging (BLI; IVIS Lumina) 1–3 weeks after tumor cell injections. Mice were randomized based on BLI into four groups of eight mice each and were treated by oral gavage daily with GDC-0973 (7.5 mg/kg) or ABT-263 (100 mg/kg), or the combination of these drugs at the indicated doses. We used our pilot experiments with four PDX models (DF68, DF101, DF118, and DF20) to perform power calculation. With eight animals per group, the study was powered to detect a minimum difference of 36% in tumor growth inhibition between conditions, given 80% power, α = 0.05, and a standard deviation of 20%. Tumor response was monitored weekly during treatment using BLI. Tumor-growth rate analysis under different treatment arms was blinded from the investigators. At the end of the treatment, ascites cells were harvested from the peritoneal cavity, collected, lysed with RBC buffer (BioLegend), and pellet volume was measured before snap freezing. To monitor treatment tolerability, body weights were monitored every 4 days. Solid tumors and ascites cells from all the mice were also formalin fixed for IHC analysis.

IHC for cleaved caspase-3 was carried out using paraffin sections of formalin-fixed tissue by deparaffinization, treatment with antigen retrieval buffer (Citrate pH6, Sigma) and incubation with anticleaved caspase-3 primary antibody (Cell Signaling Technology, CST 9661). Bound antibody was detected using a secondary anti-rabbit (SignalStain Boost, Cell Signaling Technology) followed by 3,3′-diaminobenzidine staining (DAB, Sigma), and sections were counterstained with hematoxylin. A customized Cell Profiler pipeline was written to automatically identify the percentage of cleaved caspase-3–positive tumor cells. The RGB images were split in different channels generating two images to identify nuclei (hematoxylin) and cleaved caspase-3–positive cells (DAB). Each image was converted into a binary image, and a threshold was applied to select pixels with a defined intensity. The area occupied by pixels in each image was measured. Cell death index was calculated as the ratio of cleaved caspase-3–positive area over nuclei area. We used at least 20 locations per PDX model per treatment arm across different mice. For the scoring of the solid tumor implants in the peritoneal cavity of PDXs, sections were stained with hematoxylin and eosin. A pathologist blinded to the experiments reviewed the slides. A score was assigned to each tumor based on the size of the tumor: 1 for small, 2 for medium, and 3 for big tumor. A score of 1 was also assigned to tumor cellular aggregates visible in the section. A sum of the scores was calculated and indicated as tumor implants score for each treatment group.

Copy-number variation analysis for the MEK/ERK pathway was performed by quantifying alterations in BRAF, KRAS, and NF1 of the TCGA tumors. Copy-number values were reported as ratios of the PDX sample to a reference normal value and were log₂ transformed for further analysis. Thresholds were used to define alterations as follows: (i) amplification ≥ 0.5; (ii) low-level gain ≤ 0.25; (iii) heterozygous deletion ≤ −0.25; (iv) homozygous deletion ≤ −0.5. Genomic sequencing of the PDX models, including sample library construction, targeted capture, next-generation sequencing, and bioinformatics analyses were performed as previously described (27, 30, 31). In brief, fragmented genomic DNA from the PDX models were used for targeted regions using the Agilent SureSelect 50 Mb kit according to the manufacturer's instructions (Agilent). DNA sequencing data will be available from the European Genome-phenome Archive, (Accession number: EGAS00001003427). Captured DNA libraries were sequenced with the Illumina HiSeq 2000 Genome Analyzer (Illumina). Sequence reads were analyzed and aligned to the human genome sequence (hg18) with Eland v.2 algorithm in CASAVA 1.7 software (Illumina). Potential somatic mutations and copy-number alterations were identified using VariantDx software as previously described (PMID: 25877891, 26416732, and 27573169). Potential alterations were compared with mouse sequences from experimentally obtained mouse targeted sequence data as well as the reference mouse genome (mm9) to remove mouse-specific variants. Mutations of interest were visually inspected in PDX sequences using Integrative Genomics Viewer, version 2.3.23. Protein lysates were extracted from short-term in vitro and orthotopic in vivo PDX samples. For the RPPA assay, 40 μg of protein lysate per sample was analyzed. Protein lysates were extracted from short-term in vitro and orthotopic in vivo PDX samples. The list of measured proteins included 308 unique antibodies, and protein measurements were assessed at MD Anderson RPPA Core. Heat maps were generated using GraphPad Prism and Java TreeView. In the different experiments, log₂ transformation and median centering were performed whenever indicated. For the analysis of TCGA samples, data were collected from the TCPA portal (http://tcpaportal.org/). Log₂-transformed and median centered values were used, and unsupervised hierarchical clustering was performed using Cluster software. For the analysis of the PDX models, MEK/ERK pathway score was calculated by summing all the values of the indicated proteins. Western blotting experiments were performed using BIM antibody (CST, 2933S), p-BIMS69 (CST, 4585S), p-ERKT202/T204 (Sigma, M9692), total ERK (CST, 9102S), cleaved caspase-3 (CST, 9661L); NOXA (CST, 14766S); PUMA (CST, 4976S); β-actin (Sigma, A1978). For BIM knockdown experiments, siRNA oligos from Dharmacon were used (L-004383-00-0005). For qPCR, RNA was extracted from short-term in vitro PDX cells treated with GDC-0973. cDNA was synthesized using the Quanta Qscript cDNA synthesis kit (VWR), and primers for BIM and GAPDH, as housekeeping gene, were used in Power SYBR Green PCR Mix (Life Technology): BIM R-TGGCAAAGCAACCTTCTGA; F-GGCCCCAGGTCTGAGC; GAPDH R-ATGGGGAAGGTGAAGGTCG; F-GGGGTCATTGATGGCAACAATA.

Correlation between IC₅₀ values and protein expression levels of each of the 308 proteins assayed by RPPA either before or after drug treatment was assessed using Pearson correlation in GraphPad Prism. Pearson r coefficient was used to identify proteins with positive (negative r) and negative (positive r) correlation with drug sensitivity. To adjust for multiple comparisons we used the false discovery rate approach (method of Benjamini–Hochberg; ref. 32) and reported q values. Linear regression models were generated using IC₅₀ values as the outcome variable and RPPA expression values as predictors with the lm function in R 3.4.2. For multiple regression, statistical significance was assessed with likelihood ratio tests of nested models from the lrtest package and a threshold of P < 0.05. For all the other experiments, significance was assessed using two-tailed Student t test unless otherwise stated. P values below 0.05 were considered significant; their magnitude is given in the text and/or figure legends and asterisks in the figures mark level of significance.

Proteomic analysis of TCGA tumor samples by RPPA revealed that the MEK/ERK pathway is commonly activated in a subset of HGSOC (7). To further assess activation of the ERK pathway in HGSOC, we compared the relative activation of ERK in HGSOC TCGA tumors by assessing its state of phosphorylation. The data show that there are relatively higher levels of phosphorylated ERK (pERK) in a large subset of HGSOC samples (Fig. 1A). Importantly, in this analysis, the levels of pERK strongly correlate with the levels of phosphorylation of its upstream regulators, MEK1 and RAF, and its downstream target p90RSK and its target pYB1, providing strong evidence of pathway activation. Because mutations in Ras pathway genes are not common in HGSOC, the elevated levels of MEK/ERK pathway activation in HGSOC is most likely due to upstream activation of this pathway through other pathway drivers, e.g., receptor tyrosine kinases.

To assess whether the MEK/ERK pathway is a potential vulnerability in HGSOC, we systematically analyzed the extent of pathway activation and the effect of its inhibition in a panel of well-annotated HGSOC PDX models. These models were generated from ascites/pleural effusions from patients with advanced disease and have been profiled for copy-number alterations, RNA expression and RPPA proteomic analysis (27). All the PDX models carried mutations in TP53, which is mutated in the vast majority of HGSOC (Supplementary Table S1). Although most of the PDX models do not harbor amplifications or deletions that would be predicted to be associated with activation of the MEK/ERK pathway, two PDX models (DF20 and DF118) harbor homozygous deletion of NF1 (a RAS GTPase activating protein), one has a mutation in KRAS (DF216), some have gains of KRAS and/or BRAF or heterozygous deletions of NF1 (Supplementary Fig. S1A). Because mutations in the genes of the MEK/ERK pathway that occur in HGSOC do not predict the activation of the pathway in tumor samples, we measured the relative activation status at the protein level using RPPA on lysates of short-term in vitro cultures of ascites cells from the PDX models. We previously described a workflow for the harvest and culture of these cells (referred to here as PDX ascites cells) for in vitro drug-sensitivity studies and molecular characterization (29). We measured the protein levels of several members of the MEK/ERK pathway, including upstream positive activators (pRAF, pMEK1, and pERK) and downstream targets (pp90RSK, pYB-1, and pElk1), and calculated a pathway activation score by adding the values of all these proteins(as described previously; ref. 7). Figure 1B shows a heat map of the 14 PDX models, ranked based on pathway score and the relative levels of phosphorylation of MEK/ERK pathway components, and demonstrates the extent to which pathway activation varies across the models. The pathway activation score correlates with the level of activated pERK measured by RPPA (Fig. 1C). In addition, relative levels of pERK detected by Western blotting are similar to those measured by RPPA (Fig. 1D). The pathway activation scores of short-term in vitro cultures were generally similar to those of their corresponding xenografts grown orthotopically in mice (in vivo; Supplementary Fig. S1B). These results confirm that the MEK/ERK pathway is activated in a subset of HGSOC PDX models and that PDX ascites cells can be utilized to interrogate the importance of this pathway in HGSOC.

To determine the response of the 14 PDX models to MEK pathway inhibition, we performed dose-response studies with the potent and highly selective MEK inhibitor GDC-0973 (cobimetinib), which is FDA approved for BRAF-mutated melanoma in combination with the BRAF inhibitor vemurafenib (33, 34). The PDX models were previously engineered to express a firefly luciferase reporter to monitor cell numbers in vitro and tumor growth in vivo (27). The PDX ascites cells were treated with clinically relevant doses of GDC-0973 (35, 36) for 96 hours in vitro, and luciferase assay was used as a readout of cell number. We found that GDC-0973 treatment had minimal effects on cell number (Fig. 2A and B; Supplementary Fig. S2A), with only two PDX models, DF106 and DF216, displaying a substantial reduction. These two models showed the most significant GDC-0973-induced increase in cell death, indicating that the reduction in cell number is predominantly due to cell death (Fig. 2C). Interestingly, DF216 is the model with a KRAS mutation.

To assess the extent of MEK pathway inhibition and subsequent adaptive response, we performed RPPA on samples before and after treatment with GDC-0973. The 14 PDX ascites cells were treated with a fixed concentration of GDC-0973 (0.5 μmol/L), and 308 proteins were analyzed by RPPA after 48 hours of treatment. Figure 2D shows a heat map of the 30 most downregulated (blue) and upregulated (red) proteins across all 14 PDX models. Treatment with GDC-0973 significantly reduced the phosphorylation of ERK and its target p90RSK, and this was confirmed by Western blot analysis (Fig. 2E). The fold change in pERK correlated with the extent of baseline activation of the pathway (Supplementary Fig. S2B). The RPPA analysis showed a slight reduction in cell proliferation markers (p-Rb and Cyclin B1), consistent with a reduction in cell number in some models. The most upregulated phosphoprotein was MEK, consistent with inhibition of its catalytic activity, its phosphorylation by RAF proteins, and activation of feedback loops (37). Upregulation of the proapoptotic protein BIM was detected by RPPA (Fig. 2D) and Western blotting in almost all of the models (Fig. 2E). The MEK pathway regulates BIM expression at both the transcriptional and protein stability levels. Activated ERK1/2 phosphorylates BIM on serine 69, leading to its ubiquitination and degradation (38, 39). In addition, ERK1/2-mediated phosphorylation of FOXO3A promotes its degradation and thus inhibits FOXO3A-dependent transcription of BIM (40).

We also measured BIM mRNA before and after MEK inhibition and found that BIM transcription is increased compared with untreated cells consistent with stabilization of FOXO3A (Supplementary Fig. S2C). Because the balance of pro- and antiapoptotic proteins dictates whether a cell undergoes apoptosis, we also assessed the fold changes in protein levels of other BCL-2 family members after MEK inhibition and found that, with the exception of BIM, the vast majority of the other pro- or antiapoptotic proteins were not affected by MEK inhibition (Supplementary Fig. S2D). Because PUMA and NOXA were not included in the RPPA analysis, we analyzed their levels by Western blotting and found no significant changes in protein levels after GDC-0973 treatment (Supplementary Fig. S2F).

Taken together, these findings demonstrate that MEK inhibition upregulates the level of the proapoptotic protein BIM, suggesting that MEK inhibition may increase priming of tumor cells to undergo apoptosis.

The inability of MEK inhibition to affect cell number suggested that the induction of BIM alone is not sufficient to induce cell killing and that concomitant suppression of one or more antiapoptotic proteins is necessary to induce tumor cell death. The antiapoptotic members of the BCL-2 family, including BCL-2 and BCL-X_L, exert their survival functions by binding and neutralizing the apoptotic effectors BAX and BAK. We have previously demonstrated that the 14 PDX models express different levels of the BCL-2 family proteins and show a wide range of sensitivity to the dual BCL-2/X_L inhibitor ABT-737 (29). To assess whether the apoptotic priming induced by MEK inhibition could be exploited therapeutically, we assessed the effects of GDC-0973 in combination with the orally available BCL-2/X_L inhibitor ABT-263 (navitoclax), which has been clinically evaluated in several different tumor types, including ovarian cancer (41–43). We performed in vitro dose-response studies in the 14 PDX models with ABT-263 alone or in a 1:1 fixed ratio with GDC-0973. As we previously reported, ABT-263 alone caused a variable reduction in cell number, with half of the models showing IC₅₀s in the range of 0.1–0.3 μmol/L (Fig. 3A; Supplementary Fig. S3A). Interestingly, combination of ABT-263 and GDC-0973 was more effective in decreasing cell number than single agents in 10 of the 14 PDX models (Figure 3B; Supplementary Fig. S3A). These results were not specific to GDC-0973, as other MEK inhibitors, such as PD0325901 and MEK162 (44, 45), showed similar efficacy in reducing cell numbers when combined with ABT-263 (Supplementary Fig. S3B). IC₅₀ calculations revealed that there is up to 8-fold increased sensitivity to the GDC-0973 and ABT-263 drug combination compared with ABT-263 alone (e.g., DF101; Fig. 3C). Drug synergy was assessed using the Bliss-independence model (46, 47), which does not require IC₅₀ values for all samples since treatment with GDC-0973 did not yield IC₅₀ values for most PDX models. The Bliss score is the difference between the calculated inhibition value if the two agents act independently and the observed combined inhibition values. Figure 3D shows the Bliss scores for each of the five dose combinations tested and the average across all combinations. Positive Bliss scores indicate that the combination effect is greater than additive (synergistic). The results show that GDC-0973 and ABT-263 combination is synergistic in the majority of the PDX models. Among the models that showed the lowest average Bliss score were those that were very sensitive to ABT-263 as a single agent (i.e., DF68, DF216, and DF86) and those with the least reduction in cell number after treatment with the drug combination (DF181, DF59, and DF149; Fig. 3B).

We next evaluated the cell killing effects of the single agents and the drug combination after 96 hours of treatment. As described in Fig. 2, GDC-0973 increased cell death significantly only in two models (DF106 and DF216). The GDC-0973 and ABT-263 combination induced a significant increase in cell death compared with single-agent ABT-263 in most of the responsive models (Fig. 3E). Cell death induction was validated in a subset of PDX ascites cells that exhibit a range of sensitivities to the drug combination by immunoblotting for cleaved caspase-3 after 48 hours of treatment in vitro (Supplementary Fig. S3C). Concurrent treatment with a pan-caspase inhibitor in a subset of PDX models reduced the sensitivity of the drug combination and rescued the cell killing effect, confirming that cell death is occurring through induction of the intrinsic apoptosis pathway (Fig. 3F; Supplementary Fig. S3D).

Although ABT-263 targets both BCL-2 and BCL-X_L, we have previously reported greater dependency on BCL-X_L versus BCL-2 for survival in these HGSOC PDX models (29). Therefore, we investigated the sensitivity of these PDX ascites cells to specific inhibitors of BCL-2 or BCL-X_L as single agents and in combination with GDC-0973. We found that treatment with the BCL-X_L-specific inhibitor (A-1155463) alone or in combination with GDC-0973 phenocopied the effects of ABT-263, whereas treatment with the BCL-2-specific inhibitor (ABT-199/venetoclax) had no effect on any of the PDX models tested (Supplementary Fig. S3E and S3F), confirming that BCL-X_L, and not BCL-2, plays a major role in cell survival in these PDX models.

We have previously shown that the PDX models most resistant to GDC-0973 and ABT-263 (DF149, DF83, DF181, and DF59) express the high levels of MCL-1 (29). To examine the contribution of MCL-1 to the resistance to combined GDC-0973 and ABT-263 treatment, we concurrently treated the cells with GDC-0973, ABT-263, and an MCL-1 inhibitor (A-1210477). MCL-1 inhibition significantly increased the efficacy of combined GDC-0973 and ABT-263 treatment, indicating that the antiapoptotic protein MCL1 plays a role in the survival of these PDX models (Supplementary Fig. S3G).

Taken together, these results demonstrate that inhibition of the MEK pathway results in apoptotic priming of HGSOC PDX models and increases their dependency specifically on BCL-X_L for survival. As a result, combined inhibition of MEK and BCL-X_L acts synergistically to decrease cell number and increase apoptotic cell death in HGSOC PDX ascites cells in vitro.

We next investigated whether the combined inhibition of MEK and BCL-2/X_L is effective in reducing tumor burden in HGSOC PDX models in vivo. First, we selected four PDX models that displayed sensitivity to the drug combination in vitro based on IC₅₀ (DF68, DF118, DF20, and DF101) to conduct a pilot survey of the in vivo efficacy of the combination therapy. Ascites cells were injected intraperitoneally into immunocompromised mice, allowed to grow for 2 to 3 weeks depending on the PDX model, and then treated daily for 21 days with GDC-0973 (7.5 mg/kg) and ABT-263 (100 mg/kg), a dosing schedule that was tolerated in non–tumor-bearing mice. Tumor burden was measured by BLI during the course of treatment and by harvesting ascites cells and measuring cell volume at the endpoint. In all PDX models tested, we observed a reduction in tumor growth by at least 50% when GDC-0973 was combined with ABT-263 compared with vehicle control mice (Supplementary Fig. S4A). This result was in concordance with the in vitro sensitivity data.

Based on these preliminary results, we compared the effects of the GDC-0973 and ABT-263 drug combination to single agents in two PDX models (DF118 and DF20) that were sensitive to the drug combination in the in vivo efficacy pilot experiment described above and in which a synergistic effect was observed in vitro (Fig. 3D). Single-agent treatment with either GDC-0973 or ABT-263 reduced tumor growth by 50% in DF118, but did not affect tumor growth in DF20. Combined GDC-0973 and ABT-263 treatment significantly reduced tumor growth in both models compared with single agents or vehicle (Fig. 4A–C; Supplementary Fig. S4B). The drug treatments were tolerated as indicated by minimal changes in animal body weight in both PDX models (Fig. 4D).

We next evaluated whether the combination treatment was effective in reducing the burden of solid peritoneal tumor masses in the abdominal cavity of mice. Histologic analysis revealed that the drug combination caused nearly complete eradication of tissue-disseminated solid tumor implants in the DF118 mice compared with vehicle and single-agent treatments (Fig. 4E). Although we did observe some residual tissue-disseminated solid tumor implants in DF20 mice after treatment with the drug combination, they displayed larger areas of necrosis compared with other treatments groups (Supplementary Fig. S4C), suggesting that these remaining solid tumor implants are unlikely to be viable. To evaluate whether the combination treatment induced apoptotic cell death in vivo in DF20 solid tumor implants, we performed IHC of cleaved caspase-3 (CC3). Quantification of representative images revealed that the combination treatment significantly increased cell death compared with single-agent treatments (Fig. 4F).

These data demonstrate that the GDC-0973 and ABT-263 combination substantially reduces tumor burden in HGSOC PDX models compared with untreated controls and to each agent alone at tolerable doses. These results also corroborate the use of short-term in vitro cultures of the HGSOC PDX models for predicting drug sensitivity in vivo. Although tumor regression was not observed in this short-term tumor model, this reduction in tumor growth is significant for the chemoresistant tumor models.

The range of responses elicited by the GDC-0973 and ABT-263 combination across the 14 PDX models (Fig. 3B and C) presented a valuable opportunity to identify markers of drug responsiveness. We performed linear correlation analysis to evaluate correlations between the IC₅₀ of the drug combination and the 308 proteins measured by RPPA in the PDX models before treatment. Using a false-discovery rate approach to correct for multiplicity analysis, we identified a total of 16 proteins that significantly correlate with drug response (Table 1). One of the top positive predictors was BIM (R² = 0.51, P = 0.004; q value = 0.15; Fig. 5A). This result indicates that PDX models with high basal expression of BIM respond better to the drug combination than those with low BIM expression. Indeed, all six PDX models with high BIM expression had lower IC₅₀ values for the drug combination indicating greater sensitivity. However, among the eight PDX models with low expression of BIM, there was a subset of four PDX models (DF09, DF118, DF172, and DF101) that also displayed lower IC₅₀ values and were sensitive to the drug combination, suggesting that, in this specific subset of PDX models, BIM baseline levels are insufficient to fully explain the response.

Because BIM was identified as one of the most upregulated proteins following MEK inhibition, we next investigated whether the changes in BIM can predict response to the drug combination to a greater extent than baseline BIM levels. We analyzed the correlation between BIM levels after treatment with GDC-0973 (measured by RPPA) and IC₅₀ values for the drug combination. We found that the correlation between IC₅₀ and BIM levels after GDC-0973 treatment is better than that of baseline BIM, with an R² of 0.71 (P = 0.0001; Fig. 5B). Interestingly, BIM was the only protein among the positive predictors whose predictive power increased after treatment with GDC-0973 (Supplementary Fig. S5A). Of the eight PDX models with low basal BIM, the four sensitive ones (with low IC₅₀; DF09, DF118, DF172, and DF101) show upregulation of BIM following MEK inhibition (Supplementary Fig. S5B), providing mechanistic insights into the responsiveness of these models. In contrast, the four resistant PDX models (with high IC₅₀; DF181, DF59, DF83, and DF149) show minimal upregulation of BIM after treatment with GDC-0973 (Supplementary Fig. S5B). These results were confirmed in an independent set of established HGSOC ovarian cancer cell lines (n = 9). We performed dose-response experiments with the GDC-0973 and ABT-263 combination, calculated IC₅₀ values, and performed correlation analysis between IC₅₀ and BIM expression after treatment with GDC-0973. We found that BIM levels after MEK inhibition correlate with response to the drug combination (R² = 0.54; P = 0.02; Supplementary Fig. S5C).

To investigate the functional role of BIM in drug responsiveness, we performed siRNA knockdown experiments in two PDX models, DF14 and DF101, that express high BIM protein levels following MEK inhibition and display two different levels of synergy with combined MEK and BCL-2/X_L inhibition (Figs. 2E; 5A and B). siRNA knockdown decreased BIM protein levels by 40% and 80% in DF14 and DF101, respectively (Fig. 5C), and reduced the sensitivity to the GDC-0973 and ABT-263 combination as indicated by the increase of the IC₅₀ values (Fig. 5D). These results were validated in the ovarian cancer cell line OAW28 that expresses high levels of BIM (Supplementary Fig. S5D).

The ability to upregulate BIM in response to MEK inhibition likely contributes significantly to the differential sensitivity of GDC-0973 and ABT-263 drug combination in those PDX models with low basal BIM levels (DF09, DF118, DF172, DF101 vs. DF181, DF59, DF83, and DF149). To identify proteomic markers that distinguish these PDX models with low basal BIM, we examined the proteins that were differentially expressed between the responders (DF09, DF118, DF172, and DF101) and the nonresponders (DF181, DF59, DF83, and DF149). We identified pERK as one of the differentially expressed proteins (P = 0.04; Fig. 5E). Indeed, the PDX models with low basal BIM expression that are sensitive to the drug combination also show higher expression of pERK compared with the models that are poor responders (Fig. 5E). We also found that the basal levels of pERK can predict the extent to which BIM is upregulated following MEK inhibition (R² = 0.62; P = 0.02; Fig. 5F), indicating that PDX models displaying higher activation of ERK respond by upregulating BIM to a greater extent than the PDX models with low ERK activation. Finally, multiple regression analysis revealed that a model that includes both baseline levels of BIM and pERK correlates better with response to dual inhibition than a model that includes just BIM (model improvement, P = 0.04; Supplementary Fig. S5E and S5F), suggesting that both baseline markers could be used to correlate response to combination therapy.

Together, these results demonstrate that baseline BIM expression represents a candidate marker of responsiveness to combined inhibition of MEK and BCL-2/X_L. However, BIM expression after treatment with single-agent MEK inhibitor has a higher predictive power than basal BIM expression, suggesting that BIM upregulation is a critical component of responsiveness to this combination therapy. Importantly, we also show that the extent of BIM upregulation after MEK inhibition can be predicted by the expression of baseline pERK in HGSOC PDX models.

In this study, we examined adaptive responses to inhibition of the MEK/ERK pathway, which is upregulated in a large portion of HGSOC (5, 6, 15–17), in a panel of HGSOC PDX models. We identified vulnerabilities of drug-treated tumor cells that could be targeted to enhance the efficacy of inhibitors of this pathway and demonstrate the effectiveness of a drug combination targeting MEK and the antiapoptotic proteins BCL-2 and BCL-X_L. In addition, we identify protein markers that correlate with the effectiveness of this drug combination in chemoresistant HGSOC PDX models and established ovarian cancer cells. Given the high frequency of the development of chemoresistance in HGSOC and the need for orthogonal approaches to treat refractory disease, this drug combination warrants evaluation for the treatment of chemoresistant HGSOC.

Despite strong activation of the MEK/ERK pathway in multiple PDX models, MEK pathway inhibition elicited minimal effects on cell survival in vitro in the majority of the models. Our results are consistent with published reports showing modest preclinical and clinical activity of MEK inhibitors when used as single agents in other tumor types (24, 48). However, in this study, we successfully exploited the apoptotic priming induced by MEK inhibition upregulation of BIM to promote cytotoxicity of the HGSOC models through combined BCL-2/X_L inhibition by ABT-263. This drug combination effectively enhanced cell death in vitro and in vivo compared with single-agent treatments. The mechanism of BIM regulation by the MEK pathway and the cytotoxic effects of combined MEK and BCL-2/X_L inhibition has been demonstrated in other tumor types characterized by point mutations in RAS/MEK pathway components (49–51), prompting the evaluation of this combination therapy in clinical trials comprising different types of KRAS- or NRAS-mutated tumors (Clinicaltrial.gov identifiers NCT02079740 and NCT01989585). Our study provides a rationale for inclusion of HGSOC, where activation of the MEK signaling pathway and apoptotic priming following its inhibition is apparent despite the lack of recurrent activating MEK/ERK pathway mutations.

In addition to BIM upregulation, we also detected increased levels of γ-H2AX following MEK inhibition. This finding is consistent with a recent report that showed regulation of DNA-damage repair pathway by the MEK/ERK pathway in ovarian cancer cell lines, supporting the development of MEK and PARP inhibitor combinations for ovarian cancer (52). We also detected activation of receptor tyrosine kinase signaling, specifically activated HER2, which is in agreement with a dynamic reprogramming of the kinome in response to MEK inhibition that has been reported in triple-negative breast cancer (48). Finally, the AKT pathway was also upregulated following MEK inhibition in the PDX models. Numerous preclinical studies in other tumor types have demonstrated that multiple points of cross-talk, negative feedback, and redundancy exist between the RAS/MEK and the PI3K/AKT pathways and that inhibition of both is crucial to induce tumor cell death (35, 53, 54). However, in initial clinical trials, combining MEK and AKT inhibitors elicited overlapping monotherapy toxicities (55), limiting the clinical utility of this combination. Because we were most interested in new combination therapies that could be imminently translated into the clinic, we did not further pursue this drug combination.

The GDC-0973 and ABT-263 combination therapy that we evaluated was well tolerated in mice, even with concomitant continuous daily dosing of both drugs. Furthermore, both therapeutic agents are currently being evaluated in clinical trials, facilitating immediate translation of our findings to the clinic. GDC-0973 (cobimetinib) is a highly selective, allosteric MEK1/2 inhibitor with generally acceptable tolerability and safety profile in clinical trials. Based on the promising results of a phase III clinical trial (coBRIM), the FDA approved GDC-0973 in 2015 for its utility in combination with vemurafenib in patients with advanced BRAF^V600-positive melanomas (34). The oral BCL-2/X_L dual inhibitor ABT-263 (navitoclax) exhibits an acceptable safety profile in clinical trials of patients with solid tumors, with dose-dependent thrombocytopenia as the major adverse effect (56). In this study and a previous one from our lab, we demonstrate that inhibition of BCL-2/X_L as a single agent elicited a wide range of responses in the panel of HGSOC PDX models (29, 57), but when combined with either MEK or PI3K/mTOR inhibitors, it effectively increased cell killing, providing a strong rationale for further evaluation of these combinations in dedicated HGSOC trials. Additionally, by using selective inhibitors of BCL-2 and BCL-X_L inhibitors, we show that BCL-X_L inhibition is sufficient to mediate the cell killing in both these combination therapies, uncovering specific apoptotic vulnerabilities of HGSOC PDX models that can be therapeutically exploited.

The PDX models used here exhibit a range of responses to the combined inhibition of MEK and BCL-2/XL, highlighting the utility of this model system to identify proteins that can be further evaluated as predictors of response. Overall, the results of our correlation analyses are consistent with the model depicted in Fig. 5G. High pretreatment levels of BIM protein predict a strong response to MEK and BCL-2/X_L combined inhibition. However, PDX models in which BIM is low basally, but induced by MEK inhibition, also respond well to MEK plus BCL-2/X_L inhibition. PDX models with low BIM before and after MEK inhibition are poorly responsive.

BIM represents a critical regulator of apoptosis induced by a variety of targeted therapies in both hematologic and solid tumors, including ovarian cancer, as previously demonstrated by our group and others (58). These studies have demonstrated that many therapies that inhibit oncogenic kinase signaling cascades ultimately lead to the inhibition of downstream pathways, including MEK/ERK and PI3K/AKT, and modulate the expression of BIM, suggesting that this mechanism is indeed critical for the therapy-induced cell killing. Our group has demonstrated that direct inhibition of either MEK or PI3K pathways induced expression of BIM in PDX models of HGSOC and increased priming of the cancer cells to apoptotic cell death regulated by the BCL-2 family proteins (29). These studies corroborate the importance of the BIM regulatory axis in maintaining the survival state of HGSOC. Our work here demonstrates that BIM levels not only correlate with therapy response, but also play a functional role in the response, in that depletion of BIM is able to reduce the sensitivity of PDX cells to the drug combination. Furthermore, our results suggest that there is a threshold level of BIM upregulation after MEK inhibition that the PDXs with low baseline BIM must reach in order to be committed to death and therefore respond to the combination therapy.

The evidence that BIM correlates with response to treatment with GDC-0973 and ABT-263 has important implications for the design of clinical trials to evaluate the MEK and BCL-2/X_Lcombination therapy. Measuring BIM levels after the start of treatment would be useful not only to select a larger number of patients who are likely to benefit from this combination therapy but also to mitigate unnecessary treatment for those patients predicted to respond poorly. Applying this approach in the clinic remains a challenge, as it would require measuring BIM protein levels in tumor biopsies before and during therapy. Although the identification of posttreatment biomarkers and their incorporation into clinical trials are at their infancy, this represents a promising approach to improve treatment response prediction in patients.

Considering the challenges of evaluating posttreatment biomarkers in the clinical setting, we performed additional analyses to identify other molecular features that can be evaluated before treatment and can predict the extent of BIM upregulation after MEK inhibition. We found that the baseline level of activated ERK (pERK) correlates with BIM fold induction after MEK inhibition in PDX models with low baseline BIM levels. Additionally, through a multiple regression analysis, we demonstrate that a predictive statistical model that includes both baseline BIM and pERK levels has a better predictive power than one that includes BIM alone.

Additional work is needed to uncover the molecular mechanism that mediates BIM suppression in the PDX models that show poor response to MEK and BCL-2/X_L combination therapy. Epigenetic mechanisms have been implicated in BIM suppression in different tumor types, suggesting that a combination with epigenetic agents might be effective in these models. Furthermore, because MCL-1 contributes to resistance to combined MEK–BCL2/XL inhibition, it is important to develop therapeutic strategies to inhibit MCL-1 for treatment of tumors with high levels of this protein.

In conclusion, our study sets the stage to evaluate the effectiveness of combined inhibition of MEK and BCL-2/X_L and whether MEK inhibition–induced BIM or pretreatment levels of BIM and pERK correlate with response in HGSOC patients.

J.F. Liu is a consultant/advisory board member for AstraZeneca, Tesaro, Merck, Clovis, and Mersana. R. Drapkin is a consultant/advisory board member for Repare Therapeutics and Siamab Therapeutics. U.A. Matulonis is a consultant/advisory board member for Mersana, Geneos, Myriad, Clovis, Merck, 2X, Eli Lilly, Genentech, Immunogen, Fujifilm, Cerulean, and AstraZeneca. V.E. Velculescu has ownership interest (including stock, patents, etc.) in Personal Genome Diagnostics and Ignyta and is a consultant/advisory board member for Personal Genome Diagnostics and Ignyta. J.D. Leverson has ownership interest (including stock, patents, etc.) in AbbVie. G.B. Mills reports receiving a commercial research grant from Pfizer Pharmaceuticals and AstraZeneca, reports receiving other commercial research support from Adelson Medical Research Foundation, Breast Cancer Research Foundation, Tesaro, Immunomet, Ionis, Karus Therapeutics, Komen Research Foundation, Nanostring, Ovarian Cancer Research Foundation, Prospect Creek Foundation, and Takeda/Millennium Pharmaceuticals, has ownership interest (including stock, patents, etc.) in Catena Pharmaceuticals, ImmunoMet, SignalChem, Spindletop Ventures, and Tarveda, is a consultant/advisory board member for AstraZeneca, Catena Pharmaceuticals, Tesaro, Critical Outcome Technologies, ImmunoMET, Ionis, Signalchem Lifesciences, Symphogen, Takeda/Millennium Pharmaceuticals, Tarveda, and Nuevolution, has received and expert testimony from Myriad Genetics (Licensed Technology—HRD Assay). J.S. Brugge is a consultant/advisory board member for Genentech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Iavarone, S. Palakurthi, J.D. Leverson, D. Sampath, G.B. Mills, J.S. Brugge

Development of methodology: C. Iavarone, S. Palakurthi, J.F. Liu, V.E. Velculescu, D. Sampath

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Iavarone, S. Palakurthi, J.F. Liu, R. Drapkin, U.A. Matulonis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Iavarone, I.K. Zervantonakis, L.M. Selfors, S. Palakurthi, U.A. Matulonis, D. Hallberg, V.E. Velculescu, G.B. Mills, J.S. Brugge

Writing, review, and/or revision of the manuscript: C. Iavarone, I.K. Zervantonakis, J.F. Liu, R. Drapkin, U.A. Matulonis, V.E. Velculescu, J.D. Leverson, D. Sampath, G.B. Mills, J.S. Brugge

Study supervision: S. Palakurthi, J.S. Brugge

We thank the Neurobiology Imaging Facility/HNDC Enhanced NeuroImaging Core and the ICCB-Longwood Screening Facility at Harvard Medical School for providing access to instruments; Qing Zheng, Shan Zhou, and Vilmos Adleff for technical assistance; Geoffrey Shapiro, Dennis Slamon, and Gottfried Konecny for helpful scientific discussions; and Angelica Martinez-Gakidis for scientific editing. This work was supported by a grant from the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (J.S. Brugge, R. Drapkin, G.B. Mills, D. Hallberg, and V.E. Velculescu), Breast Cancer Res Foundation (U.A. Matulonis), Ovarian Cancer Res Fund Alliance (U.A. Matulonis), and NIH grant CA121113 (V.E. Velculescu), and K99CA222554 (I.K. Zervantonakis).

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.