Peritoneal Spread of Ovarian Cancer Harbors Therapeutic Vulnerabilities Regulated by FOXM1 and EGFR/ERBB2 Signaling Free

Peritoneal seeding of cancer cells is the major mechanism of ovarian cancer metastasis, which ultimately leads to the death of patients with ovarian cancer. Patients with advanced ovarian cancer have a poor prognosis with an overall 5-year survival rate of less than 40% due to recurrence of peritoneal tumor after first-line therapy (1). During the process of peritoneal spread, malignant cells are often shed into the peritoneal fluid as nonadherent form where they survive as spheroid-like aggregates, which later spread through the peritoneal fluid to abdominal organs, and then attach and grow as adherent colonies.

The survival of ovarian cancer cells as tumor spheroids in the peritoneal fluid is regulated by growth factors in the peritoneal microenvironment and respective receptors on the tumor cells (2–4). These multi-cellular ovarian cancer spheroids exhibit tumor-initiating capacity, which is also known as cancer stemness, and features of resistance to standard chemotherapeutics and anoikis (5). To determine the underlying mechanisms, which are critical for the survival and colonization of ovarian cancer cells, we employed qPCR array and reverse phase protein array (RPPA) analysis using the mRNA and protein preparations collected from the adherent as well as nonadherent ovarian cancer cells. On the basis of this analysis, we demonstrated that EGFR (also known as ERBB) pathway is critical for survival of nonadherent cells, while Forkhead box M1 (FOXM1) is important for the adhesion of cancer cells and their colonization.

The EGFR (also known as ERBB) family proteins includes four ERBB proteins, that is, ERBB1 (EGFR), ERBB2, ERBB3, and ERBB4. ERBB family receptors are well studied for their role in normal development of ovarian follicles and in the regulation of growth of the ovarian surface epithelium (6). FOXM1 is a member of Forkhead family of transcription factors (also known as HFH-11, MPP-2, WIN, or TRIDENT), which contributes to mitosis and cell-cycle progression by regulating the transition from G₁- to S-phase and G₂- to M-phase (7). The Cancer Genome Atlas ovarian cancer study employing the probabilistic graphical model to search for altered pathways in the NCI Pathway Interaction Database (8, 9) identified that the FOXM1-mediated transcription network is altered in 87% of patients with ovarian cancer (10). Several studies have established the role of FOXM1 on tumor cell growth and metastasis (11, 12). However, the mechanisms underlying how FOXM1 helps the adhesion of ovarian cancer spheroids are not well understood. In this study, we demonstrated the mechanism in ovarian cancer cells tailored for adherent and nonadherent phenotypes and their transition between one form to another.

Immortalized ovarian cancer cell line, IOSE80 and fallopian tube epithelial cells, FTE187 and FTE188, were received from Jinsong Liu (MD Anderson Cancer Center, Houston, TX). OVCAR4 and OVCAR5 cell lines were purchased from NCI-DCTD Repository. HEYA8 and IGROV1 cells were received from the Characterized Cell Line core at MD Anderson Cancer Center (Houston, TX). Mouse ovarian cancer cell lines, C11 and C11-FOXC2, and BR-Luc murine cell lines were kind gift from Sandra Orsulic (University of California, Los Angeles, CA). IOSE80, FTE187, and FTE188 were maintained in cell culture medium consisting of 1:1 Medium 199 and MCDB105 Medium (Sigma-Aldrich) with 10% FBS. All other cells were cultured in DMEM supplemented with 10% FBS, 50 U/mL penicillin, and 100 μg/mL Streptomycin (Life Technologies) in a 5% CO₂ environment at 37°C following standard methods as described previously (13, 14). All cell lines were used between passages 3 and 25. Cell line authentication was performed by short tandem repeat profiling at the IDEXX Bioanalytic Laboratories Inc, and tested as Mycoplasma negative by PCR (Agilent Mycosenser Mycoplasma Assay Kit) as recent as 2 months prior to last experiments.

Ovarian cancer samples were obtained from Cancer Center and Obstetrics & Gynecology, Froedtert Hospital, Medical College of Wisconsin (Milwaukee, WI) after approval by the Institutional Review Board of Medical College of Wisconsin (Milwaukee, WI). All the human samples were collected with written informed consents from patients in accordance with recognized ethical guidelines of Belmont Report.

OVCAR4 and OVCAR5 cell lines stably expressing GFP were grown in 10% FBS containing medium on the nonadherent culture plate for 24 hours to prepare nonadherent ovarian cancer spheroids. Similarly, the above cells were grown on approximately 90% confluent red fluorescent protein (RFP)-labeled mesothelial cell lines in 10% serum containing medium for 24 hours. We used MeT-5A mesothelial cells (ATCC), which were derived from the pleural fluid of a male, and then immortalized using SV40-T antigen, for our coculture experiment. In brief, adherent cells stably expressing GFP were selected by FACS of GFP-labeled OVCAR cells that were adherent on MeT-5A cells. In a second approach, we also cultured ovarian cancer cells on regular culture plate for 24 hours to isolate mRNA and total protein.

Apoptosis was detected using a FITC Annexin V/Dead Cell Apoptosis Kit (BD Pharmingen). Trypsinized cells were washed with PBS, pelleted, resuspended in serum-free PBS, and incubated with 5 μL of FITC Annexin V and 1 μL of the 100 μg/mL propidium iodide (PI) for 15 minutes. Labeled cells were then processed and sorted using BD LSR II Flow Cytometer (BD Biosciences). The data were analyzed using FlowJo Software (FlowJo LLC). EpCAM-positive and CD45-negative cells were collected from the single-cell preparations isolated from omentum (adherent cells) or from ascites fluid (nonadherent cells) by FACS for tumor cell isolation from clinical samples.

Reverse phase protein array (RPPA) analysis was performed as described previously (15, 16) and detailed at the MD Anderson Cancer Center RPPA core facility as below: https://www.mdanderson.org/research/research-resources/core-facilities/functional-proteomics-rppa-core.html

Silencing or overexpression of genes was performed as described previously (17). siRNAs were transfected at a concentration of 5–20 nmol/L using RNAi Max (Invitrogen; Supplementary Table S1). For ectopic expression, 500 ng of plasmid pCellFree_G03 empty vector or pCellFree_G03-FOXM1 was transfected in cancer cells or in normal epithelial cells. Stable knockdown of genes was performed using small hairpin RNA (shRNA) as described previously (14). Control and target-specific shRNAs were purchased from Sigma to prepare lentiviral preparations (Supplementary Table S2). Viral particles were created by transfecting packaging vectors pLP1, pLP2, and VSVG plasmids, including control empty vector pLKO.1 (catalog no., SHC001V), or shRNA targets, EGFR or ERBB2, purchased from Sigma-Aldrich in HEK293T cells. Competent lentiviral particles were collected 48 hours after transfection and used to infect target cells. To create tetracycline-inducible ZEB1-expressing cells, we cloned ZEB1 in pTRE-Tight GFP tetracycline-inducible plasmid. GFP-positive cells were selected after tetracycline (Takara Bio Inc.) treatment.

mRNA levels of various genes were determined in a Bio-Rad CFX Connect using SYBR Green Supermix. Primers for real-time PCR were designed using DNA-star Lasergene 15.2 Core Suite (DNAstar) and listed (Supplementary Table S3). PCR was performed as: hot start for 2 minutes at 95°C, denaturation for 10 seconds at 95°C, annealing for 15 seconds according to the T_m of each primer, and extension for 10 seconds at 72°C for 15–30 cycles. Relative mRNA levels were quantitated using β-actin as the endogenous control and ΔΔC_t algorithm (18). RT² Profiler PCR Array for genes associated with tumor cell proliferation, cell death, and cell adhesion (catalog no., CAPA9696-12:CLAH36595) was purchased from Qiagen. Clustergram representing fold change of genes in nonadherent versus adherent cells was prepared by the Qiagen RT-PCR Profiler Software (https://dataanalysis2.qiagen.com/pcr) by converting the individual normalized ΔΔC_t values of “nonadherent” and “adherent” population to log₂, performed in technical triplicates as recommended by the Qiagen RT-PCR Profiler Analysis program. Genes with >1.4-fold change in expression with a P < 0.05 were selected. Five housekeeping genes, B2M, HPRT1, RPLP0, GAPDH, and ACTB, were used for normalizing data. Volcano plot was prepared from log₂ fold change plotted against −log₁₀ P value of differentially expressed genes in nonadherent and adherent cells. mRNA or protein network of differentially expressed candidate genes was performed using NetWalker pathway analysis suite as published previously (19).

Western blot analysis was performed with precast Gradient Gels (Bio-Rad) using standard methods as published previously (20) using the primary antibodies against FOXM1 (Santa Cruz Biotechnology), EGFR (Cell Signaling Technology), pEGFR Y1068 (Cell Signaling Technology), HER2 (Cell Signaling Technology), pHER2 (Cell Signaling Technology), CD44 (Abcam), CD24 (Abcam), c-Kit (Abcam), ITGAB1 (Cell Signaling Technology), ITGA-V (Cell Signaling Technology), ITGA-5 (Cell Signaling Technology), and ZEB1 (Cell Signaling Technology). The loading control was β-actin (Santa Cruz Biotechnology) followed by an incubation with horseradish peroxidase–conjugated secondary antibodies (Bio-Rad).

ELISA was performed using the kit from R&D Systems as described previously (17, 21). Briefly, culture supernatants were harvested and centrifuged for estimating the levels of EGF, amphiregulin, and TGFα using the ELISA kits according to the manufacturer's instructions. Recombinant protein of EGF, amphiregulin, and TGFα (R&D Systems) was used as the standard for quantification.

DNAStar Lasergene 8 Suite Sequence Builder and Genequest software were used to determine the transcription factor binding sites surrounding 1 kb upstream and 1 kb downstream of the transcription start points in the human genome database in GenBank. Computer-assisted search for the typical FOXM1 binding motif, which is tandem repeats of “TAACA” and the atypical binding motifs, such as “AACA” and “TAAC” or their complements, was conducted to uncover the putative FOXM1 binding sites. Chromatin immunoprecipitation (ChIP) was performed using ChIP Kit procured from Sigma-Millipore and modified as described previously (22). Immunoprecipitation was performed using anti-FOXM1 mAbs (Santa Crux Biotechnology) and anti-IgG (Santa Crux Biotechnology), or “no antibody” blanks were used as negative controls. On the basis of our in silico analysis, the putative FOXM1 binding sites were identified and qPCR was performed as described above by using the listed primers (Supplementary Table S4) surrounding the FOXM1 binding sites. AURKB promoter was selected as a positive control for FOXM1 as it was identified as direct target of FOXM1 for transcription (23). β-actin (ACTB) promoter sequence, which does not have FOXM1 binding sequences, was used as the negative control.

All the animal experiments were performed using approximately 4–6 weeks old female FVB/NJ-Homozygous mice from Jackson Laboratories, which were maintained under specific pathogen-free conditions in accordance with guidelines and therapeutic interventions approved by the Medical College of Wisconsin (Milwaukee, WI) Institutional Animal Care and Use Committee. Murine ovarian cancer cell line, Br-Luc (1 × 10⁶ cells/mouse), was trypsinized, washed, and resuspended in Hank's Balanced Salt Solution (Gibco), and injected into each mice orthotopically or intraperitoneally. Tumor-bearing mice were randomly divided into two groups (n = 7/group) after 10 days of tumor cell injection and treated with thiostrepton (20 mg/kg body weight) in 100 μL PBS intraperitoneally and lapatinib (i.e., GW2974, 30 mg/kg body weight) in 100 μL PBS orally, once in a week as indicated. Seven mice per group provides 80% power to detect a minimal effect size of 0.68 using one-way ANOVA. Treatment was continued for 7 weeks, at which point, all mice were sacrificed, necropsied, and tumors were harvested. The bioluminescence imaging of animals was performed weekly by injecting luciferin and imaging after 10 minutes using an IVIS Lumina III (PerkinElmer).

List of all reagents and software used, along with their source information and research resource identifier numbers, is included in the Supplementary Data (Supplementary Table S5).

All assays were performed in at least triplicate or more as indicated in the figure legends. Data are presented as means ± SE or means ± SD. Statistical comparisons were performed using unpaired two-tailed Student t tests or by ANOVA, where appropriate with a probability value of 0.05 considered significant.

To model nonadherent forms during peritoneal spreading of ovarian cancer cells (Fig. 1A), we cultured ovarian cancer cells, OVCAR4 and OVCAR5, and immortalized ovarian surface epithelial cell line, IOSE80, on nonadherent (low attachment) culture plates for 24 hours (Fig. 1B). To model the adhesion of ovarian cancer cells on a physiologically closer substratum in ovarian cancer, we grew the GFP-labeled ovarian cancer cells on RFP-labeled mesothelial cells, MeT-5A, for 24 hours (Fig. 1B) and then collected the GFP-positive tumor cells by FACS for gene expression profiling. We also determined the viability of tumor spheroids grown on nonadherent conditions and ensured that the retrieval of ovarian cancer cells from adherent state did not affect the viability of ovarian cancer cells largely, and we found that both OVCAR5 and IOSE80 cells exhibited more than 85% viability and OVCAR4 cells exhibited 72% viability (Q4 in Fig. 1C).

To identify the gene sets, which are critical for the adhesion and the growth of ovarian cancer spheroid, we performed a qPCR array that is comprised of genes that are for cell survival, proliferation, adhesion, and metastasis. In this assay, we found that adherent form of ovarian cancer cells expresses high levels of FOXM1, integrin-α V (ITGAV), integrin-α 5 (ITGA5), and integrin beta 1 (ITGB1), as well as the markers of active cell proliferation, such as cyclin D1, cyclin D3, cyclin-dependent kinase-2 (CDK2). We also found that fibronectin (FN1), which is the ligand of ITGA5 and ITGB1, was upregulated in the adherent ovarian cancer cells. We observed that nonadherent spheroids express high levels of ERBB2, EGFR, and their ligands, such as TGFα and EGF, as well as the markers of cancer stemness, such as prominin, c-KIT, STAT3, and MUC1 (Fig. 2A and B). Compatible with published results (24), we also found an inverse association between E-cadherin and ZEB1 in adherent populations of ovarian cancer cells.

To determine the central genes, which are critical for both cell adhesion and spheroid formation, we performed the pathway and network analysis using the genes that were upregulated in adherent and nonadherent population in our qPCR array using NetWalker pathway analysis suite (19) and found that EGFR/ERBB2 signaling and their ligands are the central node for signaling in nonadherent spheroids (Fig. 2C). Our analysis further identified that FOXM1, ITGAV, ITGA5, ITGB1, and FN1 are part of the gene network expressed in adherent cells (Fig. 2D). Next, we knocked down all the genes in the nonadherent and adherent gene network and determined their effect on 3D spheroid formation and cell adhesion ability. We found that nonadherent genes, such as EGFR or ERBB2, reduced the spheroid formation more than 70%, whereas the knockdown of EGF, which is the ligand of EGFR and an inducer of EGFR/ERBB2 dimer, as well as ZEB1 knockdown inhibited the spheroid formation approximately 40% (Supplementary Fig. S1A and S1B). We also noticed that the knockdown of genes, including ABL1, GAB2, LCK, GRB7, and RPS6, in the nonadherent gene network (Fig. 2C) reduced spheroid formation only about 30% compared with control siRNA–treated spheroids (Supplementary Fig. S1A and S1B). In our adhesion assay, we found that FOXM1 knockdown inhibited the adhesion ability of OVCAR4 and OVCAR5 cells approximately 80% as compared with the control, the knockdown of ITGAV, ITGA5, and ITGB1 individually reduced the adhesion approximately 40%, and the knockdown of CCND1 and CDH1 reduced the cell adhesion approximately 15%–20% as compared with the control (Supplementary Fig. S1C and S1D). Notably, we did not find that the knockdown of FOXM1 or EGFR/ERBB2 genes altered spheroid-forming ability or the adhesion of ovarian cancer cells, respectively (Supplementary Fig. S1E and S1F).

To further validate that the gene networks are important for adherent or nonadherent forms of cells, we performed another independent approach, called functional proteomic RPPA (15), using the cell lysates prepared from adherent and nonadherent cells of both OVCAR4 and OVCAR5 cells. Changes in total and phospho-proteins identified two distinct sets of proteins that were altered in adherent and nonadherent cells of OVCAR4 and OVCAR5 cells (Supplementary Fig. S2A–S2D; Supplementary Spread Sheet S1 and S2). We further employed NetWalker for enriched pathways (19), which identified a network comprised of FOXM1, CASP8, CASP7, ERCC5, CDK1, YWHAE, RPA2, PARP1, H2AFX, ATR, RHEB, CAV1, ABL1, COL6A1, and MUC1 as highly enriched in adherent cells, whereas FOXM1, CDK1, ERCC5, and CASP8 were enriched in adherent cells (Supplementary Fig. S2E). ERBB2, ERBB1 (EGFR), IGF1R, HSPB1, AKT1, AKT2, AKT3, MDM2, CDKN1B, RAD50, JUN, PTEN, PTGS2, CDC25C, CDK1, MAPK14, TP53, TP53BP1, MYC, SYK, GAB2, ANXA1, PRKCA, PAX8, and JAK2 formed the primary gene network in the nonadherent OVCAR4 cells (Supplementary Fig. S2F). Importantly, we found that the FOXM1 and CASP8 were the genes that were upregulated in the adherent network, whereas EGFR, ERBB2, GAB2, and PTEN were upregulated in the nonadherent gene network in both qPCR array and RPPA. Taken together, our RPPA results support the contention that FOXM1 is important for ovarian cancer cell adhesion, and EGFR and/or ERBB2 are required for nonadherent spheroid formation.

To further confirm the effects of EGFR and ERBB2 have on the growth of nonadherent cells compared with adherent cells, we collected lysates from adherent cells grown on mesothelial cells as monolayer and from nonadherent cells grown on low-attachment plate, and found that both native and phosphorylated forms of EGFR and ERBB2 were upregulated in the nonadherent cells compared with adherent cells of ovarian cancer (Fig. 3A and B). It is well known that binding of the ligands to EGFR leads to the phosphorylation of ERBB1 and ERBB2 (13, 25, 26). Therefore, we measured the levels of EGFR ligands in the culture media of adherent and nonadherent cells. Our results demonstrated that EGF, amphiregulin, and TGFα were markedly elevated in culture media of nonadherent cells compared with adherent cells (Fig. 3C–E). EGFR family ligands, EGF, amphiregulin, or TGFα, also improved the spheroid-forming ability of OVCAR4 and OVCAR5 cells (Fig. 3F). We had previously shown that breast cancer spheroids have characteristics of cancer stem cells (CSC; ref. 27). Here, we also found that ovarian cancer spheroids express high levels of CD44 and CD117 (c-KIT) proteins that are found in CSCs (Supplementary Fig. S3A). In conjunction, nonadherent cells expressed high levels of CD133 and endoglin and low levels of CD24, which are bona fide markers of ovarian CSCs (Supplementary Fig. S3B and S3C). Next, we determined the effect of EGF and TGFα on the expression of CSC markers in nonadherent cells, and found that both EGF and TGFα increased expression of CD44, CD117 (c-KIT), and prominin (CD133) and reduced expression of CD24 in favor to support the contention that EGFR ligands promote cancer stemness (Fig. 3G).

To further validate our results in patient samples, we isolated EpCAM⁺ and CD45⁻ cancer cells from tumor omentum for adherent cells and from ascites fluid for nonadherent cells from high-grade serous patients who were in advanced stage of ovarian cancer by performing FACS and performed immunoblot. Here, we showed that nonadherent cells in both patients expressed high levels of EGFR and ERBB2, as well as phosphorylated forms of EGFR (Y1068) and ERBB2 (Y1248), compared with the adherent population (Fig. 3H). In a complementary approach, we knocked down EGFR, ERBB2, or both using shRNAs in OVCAR5 cells and found that the loss of expression of either EGFR or ERBB2 reduced the number of ovarian cancer spheroids, whereas the stable knockdown of both EGFR and ERBB2 significantly inhibited the number of spheroids, as well as cancer stemness markers, as compared with the knockdown of EGFR or ERBB2 individually (Supplementary Fig. S3D and S3E). In conjunction, silencing of either EGFR or ERBB2 inhibited the levels of CD44 and cKIT, while silencing of both EGFR and ERBB2 inhibited the levels of CD44 and cKIT and upregulated the levels of CD24 more effectively than the knockdown of either EGFR or ERBB2 alone (Supplementary Fig. S3F).

We also found that the use of trastuzumab, a mAb that targets ERBB2, or the EGFR inhibitor, erlotinib, inhibited the tumor spheroid formation and the expression of cancer stemness markers (Supplementary Fig. S3G and S3H). Importantly, the treatment with GW2974 (lapatinib), which is a dual kinase inhibitor of both EGFR and ERBB2, inhibited the spheroid formation and the expression of CSC markers more effectively than inhibiting EGFR or ERBB2 alone (Supplementary Fig. S3G and S3H). Altogether, our results demonstrate that EGFR signaling is important for spheroid formation and cancer stemness of ovarian cancer cells (Fig. 3I).

We also observed that FOXM1 and integrins, such as ITGB1, ITGAV, and ITGA5, were upregulated in the adherent cells grown on MeT-5A cells as compared with the nonadherent cells of OVCAR4 and OVCAR5 cells (Fig. 4A). To confirm the role of FOXM1 on adhesion of ovarian cancer spheroids, we knocked down FOXM1 in OVCAR4 and OVCAR5 cells in ovarian cancer spheroids (Fig. 4B and C) and determined their adhesion capacity, and found that knockdown of FOXM1 siRNA reduced the number of adherent colonies, but increased the number of nonadherent spheroids as compared with control (Fig. 4D). To determine whether FOXM1 regulates integrin expression, we silenced FOXM1 using target-specific siRNAs, which again showed that the loss of FOXM1 reduced expression of ITGB1, ITGAV, ITGA5, LMNB1, and FN1 mRNA in both OVCAR4 and OVCAR5 cells (Fig. 4E). Furthermore, our immunoblot analysis also confirmed that the loss of FOXM1 decreased levels of ITGB1, ITGAV, and ITGA5 proteins (Fig. 4F). We also confirmed that adherent cells of ovarian cancer cells from tumor omentum in patients with ovarian cancer (see Fig. 3H) expressed high levels of FOXM1, ITGB1, ITGAV, and ITGA5 compared with the nonadherent cells from the ascites fluid from patients with ovarian cancer (Fig. 4G). Subsequently, we measured the expression of FOXM1, ITGB1, ITGA5, ITGAV, EGFR, and ERBB2 in adherent tumor cells isolated from the omentum and nonadherent tumor cells isolated from ascites fluid of mice bearing OVCAR4 and OVCAR5 ovarian cancer cells (Supplementary Fig. S4). In sum, our data demonstrate that FOXM1 was associated with expression of integrins and extracellular matrix (ECM) proteins, which are critical for cell adhesion to promote the adhesion of ovarian cancer cells (Fig. 4H).

To determine the effect of FOXM1 on cell adhesion, we overexpressed FOXM1 in immortalized ovarian surface epithelial cells (IOSE80) and fallopian tube epithelial cells (FTE187), which express low levels of FOXM1 (Supplementary Fig. S5A), and plated them on the culture dishes coated with different ECM, such as collagen-IV, fibronectin, vitronectin, laminin, and poly-d-lysine, and found that the cells expressing high levels of FOXM1 adhered rapidly to the ECM components compared with control cells (Fig. 5A and B). We also found that FOXM1 upregulated expression of ITGB1, ITGAV, ITGA5, LMNB1, and FN1 (Fig. 5C–E) in both IOSE80 and FTE187 cells.

To determine whether FOXM1 directly regulates transcription of ITGB1, ITGAV, ITGA5, LMNB1, and FN1, we performed ChIP by targeting FOXM1 transcription factor using mAbs specific to FOXM1. Our ChIP analysis demonstrated that FOXM1 directly binds promoters of ITGB1, ITGAV, ITGA5, LMNB1, and FN1 in both OVCAR4 and OVCAR5 cells (Fig. 5F–H). To prove that FOXM1-mediated upregulation of integrins and matrix proteins is necessary for ovarian cancer cell adhesion, we overexpressed FOXM1 in the OVCAR4 and OVCAR5 cells, which were pretransfected with either control siRNA or FOXM1 siRNA, and determined cell adhesion. Here, we found that FOXM1 overexpression rescued the cell adhesion ability of both OVCAR4 and OVCAR5 cells, which were pretransfected with FOXM1 siRNA (Supplementary Fig 5B and S5C). In conjunction with our adhesion results, our immunoblot also showed that the overexpression of FOXM1 rescued the levels of integrins ITGAV, ITGA5, and ITGB1, which were reduced by the siRNAs of FOXM1 (Supplementary Fig. S5D).

Studies have demonstrated that the arginine-glycine-aspartic acid (RGD) sequence in fibronectin, collagens, and laminin is the binding domain of integrin pairs such as αvβ3 and α5β1 during cellular adhesion (28). However, the role of integrin signaling on FOXM1 has not been established. We found that activation of integrin β1 signaling by Matrigel, laminin, and fibronectin promoted expression of FOXM1 (Supplementary Fig. S6) and that an integrin β1 blocking antibody reduced FOXM1 expression induced by Matrigel, laminin, and fibronectin (Supplementary Fig. S6). We also found that the components of ECM, such as laminin, fibronectin, and collagen-IV, upregulated FOXM1, ITGAV, ITGA5, and ITGB1 in OVCAR5 ovarian cancer cells (Fig. 5I). Taken together, our results demonstrate that FOXM1 directly regulates transcription of ITGAV, ITGA5, ITGB1, LMNB1, and FN1, which are key factors for the adhesion of ovarian cancer cells and that integrin signaling can induce FOXM1 expression supporting a feed forward signaling loop.

EGF-like growth factors upregulate epithelial-to-mesenchymal transition (EMT)-inducing transcription factors in cancer cells (24, 29). To identify the key transcription factors regulated by ERBB signaling, we determined the effect of EGF family ligands on expression of transcription factors that can induce EMT. Our results show that EGF upregulated ZEB1 more than 8-fold compared with an approximately 2-fold increase in ZEB2, SNAI1, and TWIST1 in both OVCAR4 and OVCAR5 cells (Supplementary Fig. S7A). Importantly, anti-ERBB2 antibody, trastuzumab, EGFR inhibitor, erlotinib, and the dual kinase inhibitor of EGFR and ERBB2, lapatinib, were able to reduce the levels of ZEB1 (Supplementary Fig. S3H). While we identified that ZEB1 is a key downstream effector of EGFR and ERBB2 proteins, we also found that the loss of ZEB1 expression using siRNAs downregulated the expression of EGFR and ERBB2, and secreted levels of EGF, but upregulated the expression of FOXM1 in both OVCAR4 and OVCAR5 cells (Supplementary Fig. S7B and S7C).

We also noticed that both laminin and fibronectin improved the binding of FOXM1 to ZEB1 promoter in ChIP assays (Fig. 6A–D). Furthermore, both LMNB1 and FN1 upregulated the expression of both FOXM1 and ZEB1 mRNA, where we noticed that FOXM1, which was markedly higher than ZEB1 at early timepoints, remained at higher level with a decrease in FOXM1 expression at later timepoints (Fig. 6E and F). We also observed that ectopic expression of FOXM1 upregulated ZEB1 levels in ovarian cancer cells (Fig. 6G). To determine whether ZEB1 acts as a negative feedback regulator of FOXM1, we inhibited ZEB1 expression using siRNAs and found that FOXM1 expression was sustained in cells where ZEB1 was knocked down (Fig. 6H).

Next, we used an efficient and controlled model for ZEB1 induction using tetracycline-inducible vector to validate how ZEB1 induction dysregulates EGFR/ERBB signaling and FOXM1 target genes. Here, we found that ZEB1 induction decreased adhesion of ovarian cancer cells as compared with their control groups (Fig. 6I–K) along with a decrease in the levels of FOXM1 and integrins, such as ITGAV, ITGA5, and ITGB1 (Fig. 6L). In contrast, we found that ZEB1 induction upregulated the levels of EGFR and ERBB2 modestly as compared with the zero timepoint (Fig. 6L). We also observed that ZEB1 induction resulted in an increase in the phosphorylation of EGFR and ERBB2, suggesting that the autocrine signaling activated by ZEB1 induction causes the phosphorylation potentially due to the upregulation of growth factors, like EGF (Fig. 6L; Supplementary Fig. S7D and S7E). In contrast, the stimulation of tumor cells with laminin, fibronectin, and collagen-IV resulted in a decrease in the basal level phosphorylation of EGFR and ERBB2 with a slight decrease in ZEB1 level (Fig. 6M). Taken together, our data demonstrate that ZEB1 acts as a regulator of EGFR/ERBB2, which also acts as a negative feedback regulator of FOXM1. We also found that the activation of integrins by the components of ECM can also upregulate FOXM1, and ZEB1 acts as an intermediary regulator required for the transition between adherent and nonadherent forms of cancer cells (Fig. 6N).

In this experiment, we determined whether combining FOXM1 or ERBB2 inhibitors together could control ovarian cancer growth in vivo, by combining the FOXM1 inhibitor, thiostrepton, and ERBB family inhibitor, GW2974 (lapatinib), to treat immunocompetent FVB mice bearing BR-Luc murine ovarian cancer cells, which express comparable levels of ERBB2, EGFR, and FOXM1 as in human ovarian cancer cell lines, OVCAR4 and OVCAR5 (Supplementary Fig. S8A). Here, we found that monotherapy with thiostrepton or lapatinib treatment reduced tumor nodules, peritoneal seeding to distant organ sites, ascites volume, and total tumor weight compared with controls, while the combination of both thiostrepton and lapatinib was more effective than the single-agent treatment (Fig. 7A–F). We noticed that the monotherapy decreased tumor weight by about 50%, whereas the combination therapy decreased tumor weight by 75%. We also noticed that 1 mouse in the thiostrepton or lapatinib groups did not have detectable tumor at the primary site, while 2 mice with the combination of both thiostrepton and lapatinib did not have detectable tumors at the end of experiment (Fig. 7D). In conjunction, the combination of thiostrepton and lapatinib reduced tumor weight in distant sites more effectively than the single-agent treatment. Of note, 2 mice did not have detectable tumor growth in distant sites and in 2 mice tumor growth was inhibited approximately 90% upon combination therapy (Fig. 7E).

The combination treatment also reduced the ascites volume and the number of tumor spheroids in ascites fluid more than as compared with monotherapy, where we noticed that 5 mice did not develop ascites in the combination group (Fig. 7F–H). Next, we determined the effect of lapatinib and thiostrepton on ERBB and FOXM1 protein levels, and found that lapatinib treatment reduced total and phosphorylated forms of EGFR and ERBB2, but upregulated FOXM1 levels in ovarian cancer spheroids (Fig. 7I). In contrast, thiostrepton reduced FOXM1, but induced phosphorylated EGFR and ERBB2 (Fig. 7I). Monotherapy with FOXM1 or ERBB2 inhibitors only slightly increased levels of cleaved caspase-3, but combination of both FOXM1 and ERBB2 inhibitors clearly induced cleaved caspase-3 (Fig. 7I; Supplementary Fig. S8B). As expected, monotherapy with lapatinib or thiostrepton reduced ZEB1 and proliferating cell nuclear antigen (PCNA) modestly, while the combination reduced ZEB1 and PCNA levels strongly (Fig. 7I; Supplementary Fig. S8B).

We found that knockdown of FOXM1 reduced the overall tumor growth in peritoneal cavity modestly, whereas the dual knockdown of EGFR and ERBB2 reduced tumor weight and number of tumor spheroids in the peritoneal fluid more effectively than compared with loss of FOXM1 expression alone at an early time point. These data suggest that EGFR/ERBB2 signaling is important for the peritoneal dissemination relatively in the early period of tumor spread (Supplementary Fig. S9A–S9C). We found that thiostrepton and lapatinib reduced tumor growth and number of tumor spheroids in the peritoneal fluid compared with the control BR-Luc cells at a later timepoint. Lapatinib treatment was more effective in reducing the overall tumor growth and the number of tumor spheroids of BR-Luc cells, which express control shRNA in the ascites fluid at the later timepoint. We noticed that stable knockdown of FOXM1 or stable knockdown of EGFR and ERBB2 reduced the overall tumor growth modestly at the later timepoint. Notably, the treatment of GW2974 (lapatinib) in FOXM1-knockdown group reduced tumor growth approximately 80% compare with the control group. Furthermore, the treatment of thiostrepton in EGFR/ERBB2-knockdown group inhibited overall tumor growth approximately 70% compared with the control (Supplementary Fig. S9D). While the knockdown of FOXM1 reduced the number of spheroids in the ascites fluid partially; we observed that the knockdown of EGFR and ERBB2 reduced number of tumor spheroids approximately 90% in the ascites fluid. Strikingly, the treatment of GW2974 (lapatinib) abrogated tumor spheroid formation approximately 75% in the control group, whereas GW2974 treatment markedly reduced the number of spheroids in the ascites (Supplementary Fig. S9D and S9F).

The spreading of tumor cells begins with extensive shedding of ovarian cancer cells from the ovary and fallopian tube and their growth as spheroids in the abdomen. The subsequent attachment and infiltrative pattern of ovarian cancer cells to the peritoneum and abdominal organs make the complete surgical resection and removal of ovarian cancer cells from abdomen infeasible. Our model of peritoneal spread showed that ERBB2/EGFR and FOXM1 signaling contribute to growth of cells as spheroids, as well as attachment and infiltration, and thus, represent the potential targets to prevent peritoneal spreading that we confirmed by combination therapy in animal models of ovarian cancer.

We demonstrated that FOXM1 is required for the expression of ITGB1, ITGA5, and ITGAV that are critical for adhesion of ovarian cancer cells. It has been reported that ITGB1 regulates the formation and adhesion of ovarian carcinoma multicellular spheroids (30). We identified that FOXM1 upregulates ZEB1, which in turn inhibits FOXM1 expression through a negative feedback mechanism. Given that ZEB1 can control the plasticity between FOXM1 and ERBB signaling in ovarian cancer cells, our data show that ZEB1 acts as a molecular switch that determines the characteristics of ovarian cancer cells to become adherent or nonadherent. It is known that TP53 mutations of both truncated and missense mutations are common genetic aberrations in ovarian cancer (31). In contrast, we identified the presence of p53 gene network and its connectivity with PAX-8, and ITGA5 in the adherent cells indicates that mutated p53 can regulate the expression of genes required for cell adhesion and tumor cell motility likely through its gain-of-function effects. Studies have shown that EMT-regulating transcription factors, including ZEB, SNAIL, and TWIST, are critical factors for cellular transformation, as well as stemness in breast, ovary, and lung cancer (24, 32). Furthermore, activation of integrin signaling upregulates FOXM1 in adherent ovarian cancer cells potentially representing a feed forward loop to accentuate cell adhesion. Integrins are the major mediators of adhesion between tumor cells and ECM proteins and they transduce signals required for cell survival. Integrin signaling is impaired when tumor cells detach from the primary tumor during peritoneal spreading, and the cells adopt a different mechanism to compensate the survival signaling.

We demonstrated that ERBB signaling is an important survival mechanisms of ovarian cancer cells in nonadherent form. In conjunction, others have also shown that ERBB signaling promotes the survival of tumor cells by activating ERK signaling and inhibiting the levels of apoptotic mediator, BIM (33). Amplification or overexpression of various EGFR family members and their ligands has been reported in epithelial ovarian cancer (34–36). However, anti-EGFR–targeted therapy has shown limited clinical results in ovarian cancer to date, and the reason behind the mechanism that causes limited efficacy is not well understood.

Earlier studies had shown that ERBB2 overexpression in ovarian cancer is a frequent event, but studies using datasets consisting of large number of clinical samples have shown that ERBB2 overexpression and amplification appear to be a relatively rare event in ovarian cancer (37). Importantly, overexpression of ERBB2 associates with a poor outcome in patients with ovarian cancer (38). Expression of the EGFR varies from 9% to 62% in patients with ovarian cancer, with its increased expression also correlating with poorer patient outcomes (34). We found that knocking down of either EGFR or ERBB2 has similar effect on reducing the number of nonadherent spheroids and inhibiting the levels of proteins required for cancer stemness likely because EGFR and ERBB2 activate similar downstream effectors and transcriptional program, than distinct or independent signaling pathways (39, 40). Several small-molecule inhibitors that block EGFR kinase activity, such as gefitinib and erlotinib, have only limited significant responses. Approximately 37% of the gefitinib-treated patients had stable disease for more than 2-months with no partial or complete responses (41). With erlotinib, 4 of 27 patients had progression-free disease for more than 6 months and 1 of these 4 patients had an objective response (34, 42, 43). Similarly, anti-ERBB2 therapies have not demonstrated significant effects as therapeutic agents in ovarian cancer. A phase II trial with trastuzumab (Herceptin, Genentech, Inc.), a humanized anti-ERBB2 mAb that is effective in ERBB2-overexpressing breast cancer, reported 7% partial responses with only a 2-month progression-free interval in recurrent ERBB2-expressing ovarian cancers (44). This ineffectiveness of anti-ERBB therapy may be explained partly by our model, which shows that alternate signaling mechanisms due to signaling plasticity regulated by transcription factors, ZEB1 and FOXM1, may contribute to tumor cell survival and peritoneal spread. Therefore, targeting of FOXM1 using thiostrepton and concurrently inhibiting both ERBB2 and EGFR using lapatinib or other irreversible inhibitors, such as neratinib or afatinib, could be a promising strategy to eliminate peritoneal spreading and metastasis of ovarian cancer and improve the quality of life of patients with ovarian cancer.

G.B. Mills reports personal fees from AstraZeneca, Chrysalis, Ionis, PDX Pharma, Symphogen, Turbine, and Zentalis, personal fees and other from ImmunoMet, Signalchem Lifesciences, Lilly USA, LLC, and Tarveda, other funding from Catena Pharmaceuticals, Myriad Genetics, NanoString, Ionis, Genentech, and GlaxoSmithKline, and grants from Adelson Medical Research Foundation, Breast Cancer Research Foundation, Komen Research Foundation, Ovarian Cancer Research Foundation, and Prospect Creek Foundation during the conduct of the study. No disclosures were reported by the other authors.

D. Parashar: Investigation, visualization, methodology, writing-original draft, project administration. B. Nair: Validation, investigation, methodology, writing-original draft, writing-review and editing. A. Geethadevi: Data curation, investigation, methodology, writing-original draft, writing-review and editing. J. George: Methodology, writing-review and editing. A. Nair: Formal analysis, methodology. S.-W. Tsaih: Formal analysis, methodology. I.P. Kadamberi: Methodology. G.K. Gopinadhan Nair: Methodology. Y. Lu: Methodology. R. Ramchandran: Writing-review and editing. D.S. Uyar: Methodology. J.S. Rader: Formal analysis, writing-review and editing. P.T. Ram: Formal analysis, methodology. G.B. Mills: Methodology, writing-review and editing. S. Pradeep: Conceptualization, resources, validation, investigation, methodology, writing-original draft. P. Chaluvally-Raghavan: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing-original draft, project administration, writing-review and editing.

P. Chaluvally-Raghavan was supported by the NCI (1R01CA229907), the Ovarian Cancer Research Fund Alliance (OCRFA), DoD Breast Cancer Research Program (W81XWH-18-1-0024), the Women's Health Research Program (WHRP) at MCW, American Cancer Society's Institutional Research Grant (ACS-16-183-31), and research funds from MCW's Cancer Center. S. Pradeep was supported by the OCRFA and research funds from WHRP at MCW. Y. Lu was supported by NIH R50CA221675. G.B. Mills was supported by the NCI (CA217685, CA217842, and R01CA123219-01) and OCRFA.

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.