Dual Src and MEK Inhibition Decreases Ovarian Cancer Growth and Targets Tumor Initiating Stem-Like Cells

Ovarian cancer is the most lethal gynecologic cancer (1). Despite introduction of targeted therapies, survival has not significantly improved in the last decade (2). Patients with most advanced ovarian cancers relapse within two years (1), and tumors become therapy resistant, underscoring the need for new treatment options. High-grade serous ovarian cancer (HGSOC), the most common and aggressive subtype, is characterized by genomic instability and few targetable genetic mutations (3). We previously showed that Src (4) and the MAPK (5) are frequently activated in HGSOC from The Cancer Genome Atlas (TCGA) by reverse-phase proteomic analysis (RPPA), identifying these as potential therapeutic targets.

Saracatinib (AZD0530), a potent Src family kinase inhibitor (6), has preclinical antitumor activity in HGSOC (4). Our prior work showed elevated expression of MAPKpT202pY204 is frequent and is an independent prognostic factor for decreased survival in HGSOC (5). Selumetinib (AZD6244), a noncompetitive MEK1/2 inhibitor, has clinical activity in low-grade ovarian cancer (7), suppresses serous and clear cell ovarian cancer xenografts (5, 8), and may prove to have clinical utility in HGSOC (5).

Stem-like or tumor-initiating cancer cells are emerging as critical mediators of drug resistance (9). The cancer stem cell (CSC) hypothesis proposes tumors are heterogeneous and contain a subpopulation of self-renewing cells that give rise to progeny with reduced proliferation (9, 10). Ascites fluid from patients with ovarian cancer contains sphere-forming cells (11) and tumor-initiating cells that are demonstrable in xenograft models (12). Various surface markers identify ovarian cancer stem cell–enriched populations (13–16). Aldehyde dehydrogenase activity (ALDH1⁺) identifies a population enriched for tumor-initiating cells in both ovarian cancer lines (16–18) and primary tumors (19). ALDH1⁺ cells are increased in populations surviving platinum chemotherapy, suggesting CSCs survive to repopulate after treatment (20). Although 50%–70% of patients with advanced HGSOC achieve a complete clinical response to initial cytoreductive surgery and chemotherapy, 70% will suffer recurrence and ultimately die of the disease (1). Thus, treatments that target resistant CSCs would be useful.

Targeted therapies have been limited by lack of drug potency, lack of target expression/activation, and by bypass pathway activation (21). Bypass activation of MEK and/or AKT limits mTOR/PI3K inhibitor therapy (22, 23). Src inhibition rapidly mediates MEK/MAPK activation in preclinical breast cancer models (24, 25). Because bypass pathway emergence limits therapeutic targeting of one signaling node, strategies targeting more than one pathway may hold promise (26). Here, we identify that both pMAPK and pSrc are elevated in nearly one-third of HGSOC. Dual Src and MEK blockade potently inhibits proliferation and promotes autophagy/apoptosis in vitro and synergistically decreases ovarian cancer xenograft growth. These effects were confirmed by genomic and proteomic analyses. In addition, ALDH1⁺ subpopulation shows Src and MAPK activation and dual kinase inhibition targets these tumor-initiating subpopulations. Finally, tumors dissociated after dual therapy showed a significant depletion of ALDH1⁺ and sphere-forming cells and decreased tumor-initiating cells upon serial xenografting.

PEO1R, an estrogen receptor (ER)-positive antiestrogen-resistant variant of PEO1 was cultured as described in ref. 4. Short tandem repeat (STR) profiling verified the unique identity of this line. PEO1R-SIR is a stable Src inhibitor resistant variant, derived by 12 weeks of continuous exposure to 1 μmol/L saracatinib. OCI-E1P was cultured from an ER-positive, primary endometrioid ovarian cancer, OCI-C5x from a primary clear cell, OCI-P5x from a primary serous, and OCI-U1a from a HGSOC in OCMI medium and have been extensively characterized as described previously (27). All were used below passage twelve. OCMI medium was obtained from Interdisciplinary Stem Cell Institute Live Tissue Culture Service Center (LTCC). Cell lines were authenticated using ATCC guidelines. Asynchronous cultures were treated with vehicle, 1 μmol/L saracatinib, 200 nmol/L selumetinib, or both for 48 hours or longer as indicated for drug assays. All in vitro cell assays described below included at least three biologic replicates and triplicate technical repeats.

Saracatinib (AZD0530) is a potent inhibitor of Src family kinases and Abl (6). Saracatinib and selumetinib (AZD6244) from AstraZeneca were dissolved in DMSO. Saracatinib did not exceed 1 μmol/L in vitro to avoid off-target effects (4). Optimal selumetinib concentrations were titrated in vitro and in vivo in xenografts (5). Dual therapy concentrations were titrated in a preliminary in vivo experiment (not shown). For xenografts, drugs were suspended in sterile 0.5% hydroxypropyl methyl cellulose with 0.1% polysorbate (Tween 80).

Cells were bromodeoxyuridine (BrdU)-labeled, stained with anti-BrdU antibodies, and propidium iodide (PI) and cell cycle assayed as described previously (28).

Three different antisense oligos to either SRC (sc-29228 for siSRC), or MEK1 (sc-29396 for siMEK1) and scrambled controls from Santa Cruz Biotechnology were transfected at 70% confluence. siRNA was added to Lipofectamine 3000 Reagent (Thermo Fisher Scientific) diluted in Opti-MEM Medium (Thermo Fisher Scientific) for 15 minutes and then added to cells for 48 hours prior to analysis.

Autophagy was evaluated by Cyto-ID Autophagy Detection Kit (ENZO) to selectively label autophagic vacuoles. Annexin V staining and flow cytometry used Apoptosis Detection Kit I (BD Biosciences) to quantitate the percentage of apoptosis (4).

Western blot analysis and densitometry were conducted as described previously (4, 28). Cells were cultured with/without drugs for 48 hours and the same cells were assayed for cell-cycle distribution, Western blot analysis, Cyclin E IP blots, and Cyclin E–associated kinase activity. Cyclin E was precipitated from 300 μg lysate and associated Cyclin E, CDK2, and p27 detected by immunoblotting (28). Cyclin E–Cdk2 precipitates were assayed for Histone H1 activity as described previously (28).

For sphere assays, cells were cultured with or without saracatinib (1 μmol/L) or selumetinib (200 nmol/L) or both for 48 hours followed by 48-hour drug washout prior to seeding as single-cell suspensions in limiting dilutions (500, 1,000, and 2,000 cells) on ultra-low attachment plates (Corning) in Mammocult media (20). Spheres > 75 μm diameter were counted after 14–21 days. Xenograft tumors from each experimental group were dissociated, pooled, and plated in triplicate sphere assays.

Drug-treated cells were also assayed for ALDH1 activity by Aldefluor kit (StemCell Technologies; ref. 20). Aldefluor-positive and -negative cells were sorted by flow cytometry (18) for Western blot analysis.

Animal experiments were approved by the Institutional Animal Care and Use Committee. Estradiol pellets 0.36 mg/90 days (Innovative Research) were implanted subcutaneously into 5-week-old female NOD/SCID mice because they improve tumorigenicity in vivo (Charles River; ref. 4). PEO1R cells (10⁶ in 100 μL Matrigel) were injected in the mammary fat pad as described in ref. 19. The mammary fat pad was chosen because ovarian CSCs are most reliably xenografted in the mammary fad pad compared with other traditional sites (peritoneal cavity and ovarian bursa) in NSG mice and recapitulated the heterogeneity of primary serous ovarian cancer, as assessed by histology, surface immunophenotype, and expression of p53, WT1, and CK7 (19). When tumors reached 70 mm³, mice were divided into four groups (n = 10/group): (i) untreated (E2 only); (ii) selumetinib [oral gavage, 5 mg/kg/day (5)]; (iii) saracatinib [oral gavage, 20 mg/kg/day (4)]; (iv) both drugs together. Animals were weighed and tumors measured using calipers and volumes were calculated twice weekly using (long-side × short-side²)/2. Mice were sacrificed when control tumors reached approximately 1,000 mm³ for tumor harvesting, dissociation, and serial transplantation.

Xenografts were immunostained after antigen retrieval by boiling in sodium citrate (10 mmol/L, pH 6.0) for 45 minutes, followed by primary antibody treatment for Ki67 (Ki-67PAb, 1:500 Abcam) and ALDH1 (ALDH1 mAb, 1:200, Cell Signaling Technology), and scored as described previously (5, 29). Four to five high-power fields (60×) from at least three tumors/treatment group were chosen and 80–100 cells/field counted for nuclear Ki67 or for ALDH1 staining. The mean score represents % positive cells from at least 400 cells counted. Cleaved caspase-3 was detected by staining with cleaved caspase-3 mAb (1:100, Cell Signaling Technology) and 546 AlexaFluor (1:200, Invitrogen) and direct immunofluorescence as described previously (4).

To quantitate drug effects on stem-like cells, primary tumors were dissociated and resuspended for sphere assays or to quantitate tumor-initiating cells in vivo in a second set of recipient mice. For tumor-initiating cell assays, 5,000 or 20,000 cells were injected in 100 μL Matrigel into the mammary fad pad of estrogen-supplemented NOD-SCID mice (n = 6/group).

The Cancer Genome Atlas/The Cancer Proteome Atlas (TCGA/TCPA) presents RPPA of 178 proteins and phospho-proteins from over 400 untreated new HGSOCs (30) at http://cancergenome.nih.gov/cancersselected/ovarian. Publicly available RPPA data for 405 primary HGSOCs was analyzed for levels of activated Src and MAPK. Kaplan–Meier curves were prepared to display overall survival dichotomized by tumors showing both pT202pY204MAPK (pMAPK) and pY416Src (pSrc) above the median versus all others in 405 HGSOC tumors. HGSOC survival distributions were compared using the log-rank test. For multivariate analysis, high pSrc/high pMAPK was evaluated together with race, age at pathologic diagnosis (as a continuous variable), year of diagnosis, no macroscopic residual disease, residual disease 11–20 mm, residual disease > 20 mm residual disease, tumor grade, clinical stage, disease subtype using Cox regression in 338 cancers with complete data.

RPPA and Illumina gene expression analyses were carried out on PEO1R with and without drug treatments for 48 hours in vitro from three biologic repeat samples. The complete method for RPPA is described at https://www.mdanderson.org/research/research-resources/core-facilities/functional-proteomics-rppa-core.html.

Gene expression analysis of these samples followed MIAME guidelines of the Microarray Gene Expression Data Society as described in ref. 3. The raw data of gene expression microarrays generated from Illumina Chips were normalized, background-corrected, and summarized using the R package “lumi.” Probes below background level (detection P < 0.01) were excluded and differential expression was identified with Bayes-adjusted variance analysis using the Bioconductor Limma package. To reduce false positives, unexpressed probes were removed. The R package “limma” was employed for gene differential expression analysis, followed by multiple test correction by the Benjamini–Hochberg procedure. Genes with adjusted P < 0.05 and fold change > 2 were claimed as significantly differentially expressed. These data are available in NCBI GEO with accession number GSE112371.

The same differential expression analysis method was applied to RPPA data. RPPA data processing used SuperCurve (SuperCurve Package. R package version 1.4.1.2011) as described previously (30, 31). Bar plots for relative RPPA for different treatments groups were generated using R package “ggplot2”, using P < 0.05 to define significant RPPA proteins. Corresponding “logFC” values of selected proteins were magnified 100X and applied as y-axis to present the log₂-fold change between treatment and control groups.

For all analyses comparing more than two conditions, results were analyzed by ANOVA. One-way or two-way ANOVA assessed difference among means. For 2 × 2 factorial experiments, interactions were tested by two-way ANOVA followed by Tukey honesty significance test, with P≤ 0.05 indicating a significant difference. Effects on cell cycle and viability were analyzed using the median-effect method of Chou and Talalay (32). Post hoc analysis by Dunnett multiple comparisons test was used to calculate differences between individual groups after the ANOVA, for example, between dual therapy and control. Student t test was used for comparisons of two conditions only. Tumor-initiating stem cell frequency was calculated using the L‐Calc software http://www.stemcell.com/tutorials/lcsetup.exe (StemCell Technologies).

To test for drug synergy in in vitro assays, the coefficient of drug interaction (CDI) was calculated as CDI = (2 drug combination divided by the control)/(drug 1 divided by the control)×(drug 2 divided by the control). CDI < 1 indicates synergistic drug interaction; CDI = 1 indicates additivity; and CDI > 1 indicates antagonism. This method was also used to evaluate differences between final tumor volumes between groups.

In vivo tumor growth curves were compared using an ANOVA for multiple comparisons tests with GraphPad PRISM (GraphPad Software) to determine differences in the growth curve slopes of dual-treated tumors compared with other groups.

The combination ratio method was used to analyze for drug synergy on xenograft growth (24, 33). Fractional tumor volume (FTV) was defined as mean-final tumor volume of drug treated animals divided by mean-final tumor volume of untreated controls (only estradiol). The combination ratio compared the FTV expected if there were no synergy, with the observed FTV. The combination ratio was calculated as (FTV of saracatinib × FTV of selumetinib)/observed FTV of combination. Observed and expected FTV are described as:

Expected FTV = (mean FTV of saracatinib) × (mean FTV of selumetinib)

Observed FTV = final tumor volume combined therapy/final tumor volume estradiol alone

Combination ratio = Expected FTV/Observed FTV.

A combination ratio > 1 indicates synergy; a ratio < 1 indicates a less than additive effect.

TCGA shows 18% of HGSOCs have activating Ras/Raf/MEK/MAPK pathway mutations (5). Src activates the Ras/Raf/MEK/MAPK pathway. Because ovarian cancers frequently express activated pSrc (4) and pMAPK (5), we determined the extent to which both kinases are activated together in the RPPA dataset of 405 HGSOCs in TCGA. pSrc (pY416) was detected in 74% (n = 300/405) and pMAPK (pT202Y204) was detected in 76% (n = 310/405) of these tumors, respectively. In nearly one-third (31% or 126/405) of HGSOCs, both pMAPK and pSrc levels were above the median, and these patients had a shorter survival (median 40.4 months) compared with all others (median survival = 48.9 months; n = 405; P = 0.004, Fig. 1A). Multivariate analysis showed elevation of both pMAPK and pSrc independently predicts poor overall survival (HR = 1.43; 95% CI, 1.05-1.95; P = 0.024; Fig. 1B).

Because Src and MAPK are frequently both activated in HGSOC, we tested the therapeutic potential of inhibition of both kinases. The HGSOC line, PEO1R, was treated with saracatinib (Src inhibitor; SI, 1 μmol/L), selumetinib (MEK inhibitor; MI, 200 nmol/L), or both for 48 hours, followed by cell-cycle analysis (Fig. 2). Saracatinib causes partial G₁ arrest at 1 μmol/L with little effect at lower doses (4). Similarly, selumetinib caused only partial G₁ arrest, with no greater effect above 200 nmol/L (5). Despite modest effects of each monotherapy (mean % S phase = 24.6% for SI, and 20.65% for MI vs. control of 45.9%), both drugs together caused synergistic cell-cycle inhibition (mean % S-phase = 4.25%). A coefficient of drug interaction or CDI < 1 indicates synergy: CDI = 0.38 for decrease in mean percent S-phase cells, %S; P = 0.0224, for dual therapy versus control Fig. 2A. Drug effects were verified by siRNA-mediated kinase knockdown. While loss of each kinase alone caused partial G₁ arrest, dual knockdown of Src and MEK together synergistically decreased the %S compared with each siRNA alone (CDI = 0.77, P < 0.0001 for dual knockdown vs. control, Supplementary Fig. S1A and S1B). After 48 hours, saracatinib decreased pY416-Src (pSrc). Although selumetinib decreased pT202pY204-MAPK (pMAPK), it increased pSrc (Fig. 2B). In addition, MEK inhibitor alone increased pS473-AKT (pAKT), but saracatinib alone and dual inhibitor treatment both decreased pAKT (see also densitometry; Supplementary Fig. S1C). The increased pMEK observed after selumetinib reflects drug action to lock MEK1/2 into a conformation that enables substrate/ATP binding but disrupts catalysis, yielding an inactive conformation that reacts with anti-pMEK antibody (34).

To determine whether p27 was mediating cell-cycle arrest, cyclin E-CDK2–bound p27 and associated kinases activities were assayed. While Cyclin E and CDK2 levels were unchanged, the levels of CDK inhibitor, p27 (Fig. 2C), and Cyclin E-bound p27 detected by immunoprecipitation/Western blot analyis (Fig. 2D) both increased modestly with each drug, and increased most significantly with combination therapy. Densitometric quantitation of mean p27 levels and of p27 bound to cyclin E are graphed on the right for data in Fig. 2C and D. Very little p21 is expressed in these cells (not shown). Cyclin E precipitates were tested for kinase activity in the same lysates recovered for Fig. 2B and C. Cyclin E-CDK2 activity was most strongly inhibited by both drugs (P < 0.000004 for dual therapy compared with control, CDI = 0.77; Fig. 2E). As little as a 2–3-fold rise in Cyclin E–bound p27 can fully inhibit cyclin E-Cdk2. Thus, p27 is a mediator of drug response in vitro.

Most available ovarian cancer lines are heavily passaged, derived from metastatic tumors, and do not always reflect the biologic phenotype of the primary cancer (27). Drug responses were thus validated in early passage, primary ovarian cancer cultures (27). In the endometrioid primary culture, OCI-E1P, saracatinib (1 μmol/L), or selumetinib (200 nmol/L) for 48 hours each modestly decreased the %S (P = 0.0002 for saracatinib and P = 0.005 for selumetinib), while dual therapy synergistically decreased %S (P < 0.00001, CDI = 0.86; Supplementary Fig. S1D). Dual Src and MEK inhibitor action was also evaluated in primary cultures from an ER-negative clear cell ovarian cancer (OCI-C5x) and from a HGSOC (OCI-P5x) using saracatinib (1 μmol/L), selumetinib (200 nmol/L), or both for 48 hours. Dual therapy caused synergistic cell-cycle inhibition in OCI-C5x, with CDI = 0.6 for mean %S (P = 0.0002; Fig. 2F) and in OCI-P5x with CDI = 0.52 (P = 0.0067; Fig. 2H). pSrc and pMAPK were decreased by their respective inhibitors, confirming drug targeting (Fig. 2G and I). p27 increased most with dual therapy in OCI-C5x and OCI-P5x (Fig. 2G and I; Supplementary Fig. S2A and S2B). Thus, for the both the established PEO1R HGSOC line and in primary cultures of HGSOC, endometrioid, and clear cell cancer origin, dual therapy caused greater cell-cycle inhibition than either monotherapy.

Despite promising preclinical antitumor activity, resistance to saracatinib monotherapy emerges rapidly in vitro and in vivo (4, 24, 25). To evaluate potential signaling pathway activation with emergence of acquired saracatinib resistance, PEO1R were cultured in the continuous presence of saracatinib (drug renewed every 3 days) and cell lysates were recovered weekly (at weeks 0, 1, 2, 3, and 4). Lysates were then evaluated for phosphorylated and total Src and MAPK. Within 2 weeks, saracatinib no longer decreased pSrc, and pMAPK increased by week 3 (Supplementary Fig. S3A). Notably, treatment of Src inhibitor-resistant PEO1R cells (PEO1R-SIR), with saracatinib (1 μmol/L) together with selumetinib (200 nmol/L) synergistically inhibited cell cycle progression within 48 hours (P < 0.00001 for % S-phase cells after dual therapy compared with control, CDI = 0.52), with little effect after each monotherapy (Supplementary Fig. S3B).

Signaling kinase activation was compared with/without drug treatment by RPPA in PEO1R and PEO1R-SIR. RPPA showed pY1068-EGFR was 1.65-fold higher (P = 0.009) and pY1248-HER2/ERBB2 was 1.3-fold higher (P = 0.05) in PEO1R-SIR than parental PEO1R while total kinase levels were similar. Western blot analysis confirmed EGFR, HER2/ERBB2, and showed Raf activation in PEO1R-SIR compared with PEO1R, while total kinase levels were unchanged. Selumetinib rapidly inhibited MAPK, despite persistent activation of these upstream kinases (Supplementary Fig. S3C and S3D). Thus, prolonged saracatinib exposure mediates bypass activation of EGFR, HER2/ERBB2, and Raf/MEK. Addition of selumetinib circumvents the MAPK activation that arises after prolonged saracatinib exposure.

PEO1R, OCI-C5x, and OCI-P5x cells were treated with saracatinib (1 μmol/L), selumetinib (200 nmol/L) or both, or with paclitaxel as positive control for 48 hours and evaluated for apoptosis. Cleaved PARP-1 levels showed little or no change with each monotherapy, but rose significantly with dual Src and MEK inhibition in all models (Fig. 3A–C, densitometry; Supplementary Fig. S4A). Similarly, Annexin V-positive cells increased by 2.5- to 4-fold with dual therapy in all models, with more modest effects with each monotherapy (Fig. 3D–F).

In autophagic cells, microtubule-associated protein light chain 3-I (LC3-I) is cleaved to LC3-II (35). LC3-II was variably increased with either kinase inhibitor but increased most significantly with dual kinase inhibition (Fig. 3G–I, densitometry; Supplementary Fig. S4B). To further evaluate autophagy, drug-treated cells were labeled with Cyto-ID-Green autophagy dye, assayed by flow cytometry (36). Autophagic cells increased most significantly with combination treatment (PEO1R P < 0.00006; OCI-C5x P < 0.0052; OCI-P5x P < 0.0029 for dual therapy vs. control; Fig. 3J–L). Autophagy is initially a survival mechanism, but leads to cell death under prolonged stress. Combination treatment appears to activate autophagy, ultimately leading to apoptosis.

Gene expression analysis and proteomic analysis by RPPA evaluated drug effects over 48 hours in PEO1R. Gene set enrichment analysis (GSEA) showed the top 10 genes from the top gene sets (cell cycle, DNA replication, and Aurora kinase) were all decreased by dual therapy compared with untreated controls. GSEA plots show effects on cell cycle, DNA replication, Aurora, and autophagy genes of combination therapy versus control, with normalized enrichment scores (NES) shown (Fig. 4A; Supplementary Fig. S5). Autophagy was the top upregulated gene set, confirming observations in Fig. 3.

RPPA showed the 20 proteins most decreased by dual treatment compared with monotherapy all promote cell cycle/proliferation (cyclin B1, FOXM1, NDRG1, CDK1, pRb, CHK1, YB1 TTF1, MDM2) or MAPK/and PI3K signaling (pS6, p70S6K, MAPK, 4EBp1, PRAS40; Fig. 4B). The top 20 upregulated proteins included apoptosis mediators (caspase 7, Bim, FOX03A), cell-cycle inhibitors (p27, p21), and TSC-1, which opposes the PI3K/mTOR pathway (Fig. 4B and C). The similarity between triplicate samples in each group is displayed in Fig. 4C.

ALDH1⁺ and ALDH1⁻ cells were sorted by flow cytometry. In PEO1R, ALDH1⁺ cells had more sphere-forming and colony-forming cells than ALDH1⁻ cells (Fig. 5A and B). ALDH1⁺ cells expressed higher pMAPK and pSrc than ALDH1⁻ cells in PEO1R and in two primary cultures (Fig. 5C–E; densitometry in Supplementary Fig. S6A–S6C).

Each drug alone decreased ALDH1 activity moderately, but both together synergistically decreased the % ALDH1⁺ cells in all models tested (P = 0.0048 for dual therapy vs. control for PEO1R, CDI = 0.96; P = 0.000001 for OCI-C5x, CDI = 0.52; P = 0.00002 for OCI-P5x, CDI = 0.57; Fig. 5F–H; and P = 0.0008 for OCI-U1a, CDI = 0.98; Supplementary Fig. S6D).

Drug effects on sphere formation were evaluated. After drug treatment for 48 hours, followed by a 48-hour drug washout, all groups had similar cell-cycle profiles, confirming similar proliferation in the bulk population prior to sphere seeding (Supplementary Fig. S6E–S6H). Combination treatment decreased sphere-forming cell abundance more than either drug alone in PEO1R, OCI-C5x, OCI-P5x, and OCI-U1a (Fig. 5I–K; Supplementary Fig. S6I). Thus, ALDH1⁺ populations in both PEO1R and in primary cultures derived from aggressive high-grade papillary serous and clear cell ovarian cancers are Src and MAPK driven, and dual therapy with saracatinib and selumetinib targets the ALDH1⁺ and sphere-forming populations.

Drug effects were next investigated in vivo. Saracatinib and selumetinib each alone modestly decreased PEO1R xenograft tumor growth, but both together not only suppressed tumor growth, but caused a modest tumor regression. Growth curve analysis revealed significant differences in the slope of dual treated tumor versus all other groups (control vs. both; P = 0.0001, SI vs. both; P = 0.0142, MI vs. both; P = 0.0027). Tests for synergy using the combination ratio (24, 33) showed a combination ratio > 1.4 at all time points after week one, indicating a synergistic drug effect (Fig. 6A). Between week 1.5 and the end of the experiment, mean tumor volume increased by nearly 5-fold in untreated controls, and by 2.05 and 2.8 fold respectively in saracatinib- and selumetinib-treated tumors, while the mean volume of dual treated tumors showed a modest decrease (P = 0.00006 for dual therapy vs. control, with a CDI = 0.42 indicating drug synergy for final tumor volume change; Fig. 6B).

In vivo, while tumor cell proliferation decreased up to 40% of dual treated tumor cells remained Ki67 positive (Fig. 6C). Apoptosis, detected as cleaved caspase on immunofluorescence (IF), increased modestly from 6% in untreated controls to 16% in dual drug-treated tumors (Fig. 6D). Tumor necrosis was similar in all tumors. Thus, in vivo, neither the antiproliferative effect, nor apoptosis could fully account for the marked decrease in xenograft growth following dual therapy. Attenuation of the stem cell population was notable in surviving tumor cells. Notably the ALDH1⁺-stained stem-like cells (29) decreased significantly with each drug alone, and most notably with dual therapy (Fig. 6E; Supplementary Fig. S7).

To further evaluate drug effects on ovarian cancer stem-like cells in vivo, treated xenografts and control tumors were excised, dissociated, and cells were either seeded for sphere formation or transplanted in limiting dilutions into a second series of recipient mice to assay tumor-initiating stem cell abundance. Sphere formation was most significantly decreased in cells surviving after combination treatment (P = 0.0002 for dual therapy vs. controls, CDI = 0.71; Fig. 6F). Tumor-initiating cells were assayed by injection of 5,000 or 20,000 dissociated tumor cells/group into each of 6 recipient mice. Tumors formed from dissociated, dual therapy–treated xenograft cells showed little growth compared with those arising from the same number of monotherapy-treated or control tumor cells (Fig. 6G). Tumor cell populations persisting after combination therapy in vivo contained fewer tumor-initiating cells and formed tumors with greater latency compared with monotherapy (shown for 5,000 cell group, Fig. 6H). Tumor-initiating stem cell (TISC) frequency decreased from 1/3,053 untreated tumor cells to 1/15,898 cells surviving dual therapy (P = 0.018, Fig. 6I). Thus, dual MEK and Src inhibitor treated primary tumors showed a significant attenuation of cells expressing ALDH1⁺, and decreased abundance of both sphere-forming cells and tumor-initiating stem cells in vivo on serial assays.

Recent targeted monotherapies directed against various signaling kinases have shown activity in HGSOC, but limited response duration (37). Because targeting single pathways is often limited by bypass pathway activation (38), targeting multiple signaling nodes might yield greater, and more durable responses. Most solid tumors appear to be initiated by a stem cell–like subpopulation that resists therapy and mediates recurrence (29, 39). In an effort to subvert drug resistance, we tested whether combined use of two rationally targeted therapies might have greater antitumor efficacy, and potentially target the ovarian cancer–initiating population.

TCGA analyses show frequent genetic aberrations that activate the Ras/Raf/MEK/MAPK pathway in HGSOCs (3). In addition, proteomic RPPA has shown that HGSOCs with high pMAPK have a worse overall survival compared with tumors without MAPK activation (5). MEK inhibitors, such as selumetinib, have activity in low-grade serous ovarian cancers but have not been fully explored in HGSOC (7). MEK/MAPK pathway gene alterations correlate with MEK inhibitor drug response in low-grade serous cancers (40).

We previously showed that most HGSOCs in the TCGA/TCPA dataset express activated pSrc (4), a major regulator of proliferation and metastasis (41). Despite the frequent activation of Src, the Src family kinase inhibitor, dasatinib has shown limited activity as a single agent for recurrent HGSOC, with only 21% of patients progression-free after 6 months (42). Another Src family kinase inhibitor, saracatinib, evaluated in combination with chemotherapy for recurrent platinum-resistant ovarian cancer, showed no added clinical benefit (43). Interestingly, saracatinib monotherapy rapidly gives rise to MEK/MAPK bypass pathway activation, and drug resistance in breast cancer xenografts (24, 25). Because Src (4) and MAPK (5) are both activated in HGSOC, we tested the potential therapeutic efficacy of combined Src/MEK blockade

Here we show that saracatinib and selumetinib synergistically induced cell cycle arrest in vitro in HGSOC, and in high-grade endometrioid and clear cell ovarian carcinoma primary cultures. Autophagy is initially a survival mechanism, but under prolonged stress, leads to cell death (44). Dual therapy upregulated both autophagy and apoptosis in all the models. Combination therapy may initially activate autophagy but ultimately lead to apoptosis. GSEA confirmed these findings. The top 20 gene sets downregulated by dual versus monotherapies govern cell-cycle progression and DNA replication, while autophagy drivers were upregulated. GSEA changes were validated further by RPPA.

Resistance to saracatinib monotherapy emerged rapidly in PEO1R, with loss of Src inhibition and activation of putative bypass mechanisms via EGFR, HER2/ERBB2, Raf, and MAPK over four weeks in culture. Treatment of saracatinib-resistant cells with selumetinib potently inhibited MAPK despite upregulation of upstream MEK activators by saracatinib. Saracatinib resistance also developed within weeks in vivo, in PEO1R xenografts, as in other models (24, 25). Addition of selumetinib to saracatinib overcame resistance to each monotherapy, yielding synergistic antitumor activity in vivo. Thus, dual treatment might circumvent acquisition of saracatinib resistance. While the MEK inhibitor, selumetinib, rapidly activated AKT, saracatinib inhibited this kinase and prevented selumetinib-mediated AKT activation. Thus, selumetinib-induced bypass AKT activation (38, 45) and saracatinib-induced bypass EGFR/MEK activation are both subverted by dual Src and MEK blockade in our ovarian cancer models.

Recent preclinical data also support combined use of Src and MEK inhibitors in other malignancies such as breast (24, 46) and lung cancer (47), and melanoma (48). Dual MEK and Src inhibition induced apoptosis and decreased breast cancer metastasis (46). In melanoma, although MEK inhibition decreased cell proliferation, it increased invasion, and dual Src/MEK blockade impaired this (48). Combined treatment with saracatinib and the MEK inhibitor, PD0325901, reversed EMT, decreased invasion in vitro and synergistically reduced lung xenograft growth (47). While Src is frequently activated in HGSOC, saracatinib is a pan-Src family inhibitor, and thus effects on other Src family kinases could contribute to results observed.

There is strong evidence that a population with ovarian CSC-like properties in vitro and in vivo is selected for by chemotherapy with platinum agents (17, 49). Ovarian cancer stem-like cells express surface markers CD133, CD44, CD24, CD117/cKit, and demonstrate ALDH1 activity (50–53). In addition, CD133⁺ cells with ALDH1 activity show higher sphere and tumor-initiating cell abundance (17). Chemotherapy enriches for ALDH1⁺ surviving cells (29, 49) and cancers recurring after chemotherapy showed increased ALDH1⁺, CD44⁺, and CD133⁺ cells compared with matched primaries (49). Preclinical strategies to target ovarian tumor-initiating cells include targeting CSC marker–positive cells with siRNA (29, 54), antibodies, or inhibitors to CD44 or CD117/cKit (52, 55). Other studies have utilized CSC-targeted dendritic cell vaccination (50–53), and epigenetic drugs to target CSCs (56). Several reports also predict successful CSC targeting by Notch (57) and Wnt (55) inhibitors.

Here, we show potential for dual Src and MEK inhibition to therapeutically target ovarian cancer initiating cells. Src is a key mediator of malignant stem cell self-renewal (58). The ALDH1⁺ populations showed higher pSrc and pMAPK than ALDH1⁻ cells. Moreover, saracatinib and selumetinib synergistically reduced ALDH1⁺ cells and sphere formation in vitro in both PEO1R and in primary cultures. After prolonged MEK and Src inhibitor exposure in vivo, apoptosis and proliferation were only modestly changed. However dual therapy decreased the ALDH1⁺ population, and dramatically reduced sphere-forming and tumor-initiating cells in xenograft tumors dissociated after in vivo treatment, suggesting Src and MEK inhibition target the tumor-initiating population. While ovarian cancers comprise multiple histologic subtypes, characterized by different molecular profiles that warrant different treatment strategies (59), current studies have potential application not only to serous, but also to other ovarian cancer subtypes. Although in vivo studies were limited to the PEO1 cell line, in vitro data in primary HGSOC and clear cell ovarian cancer cultures predict potential generalizability to other ovarian cancer subtypes.

Characterization of pathways required for CSC survival and self-renewal may yield novel more effective strategies to overcome drug resistance. Treatment with saracatinib and selumetinib together modestly decreases proliferation and induces cell death, but most critically appears to target CSCs. Both orally available drugs were well-tolerated in phase I/II studies. Application of saracatinib/selumetinib for ovarian cancer treatment will require phase I trials of dual therapy to ensure tolerability, bioavailability, and to develop clinical biomarkers of target inhibition. Current data support further clinical evaluation of dual MEK and Src inhibitor combinations.

G.B. Mills has ownership interests (including patents) at Catena Pharmaceuticals, ImmunoMet, Myriad Genetics, PTV Ventures, and Spindletop Ventures; reports receiving speakers bureau honoraria from Allostery, AstraZeneca, ISIS Pharmaceuticals, Lilly, MedImmune, Novartis, Pfizer, Symphogen, and Tarveda; is a consultant/advisory board member for Adventist Health, AstraZeneca, Allostery, Catena Pharmaceuticals, Critical Outcome Technologies, ISIS Pharmaceuticals, ImmunoMet, MedImmune, Lilly, Novartis, Precision Medicine, Provista Diagnostics, Signalchem Lifesciences, Symphogen, Takeda/Millenium Pharmaceuticals, Tarveda, and Tau Therapeutics; and reports receiving commercial research grants from Adelson Medical Research Foundation, AstraZeneca, Breast Cancer Research Foundation, Critical Outcome Technologies, Illumina, Karus, Komen Research Foundation, NanoString, Pfizer, and Takeda/Millenium Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Simpkins, K.E. Hew, D.J. Azzam, G.B. Mills, J.M. Slingerland

Development of methodology: F. Simpkins, K. Jang, H. Yoon, K.E. Hew, D.J. Azzam, D. Zhao, Z. Wei

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Simpkins, K. Jang, H. Yoon, M. Kim, D.J. Azzam, J. Sun, T.A. Ince, G.B. Mills, J.M. Slingerland

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Simpkins, K. Jang, H. Yoon, K.E. Hew, M. Kim, W. Liu, W. Guo, Z. Wei, G. Zhang, G.B. Mills, J.M. Slingerland

Writing, review, and/or revision of the manuscript: F. Simpkins, K. Jang, H. Yoon, K.E. Hew, Z. Wei, G. Zhang, G.B. Mills, J.M. Slingerland

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Simpkins, J.M. Slingerland

Study supervision: F. Simpkins, J.M. Slingerland

Saracatinib and selumetinib were provided by AstraZeneca. This work was funded by NCI K08 CA151892-01 award (to F. Simpkins) and by a grant from the Breast Cancer Research Foundation (to J. Slingerland).

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.