Hypoxia‐inducible factor‐2α directly promotes BCRP expression and mediates the resistance of ovarian cancer stem cells to adriamycin

Ovarian cancer stem cells (OCSCs) are sources of tumor chemoresistance and recurrence. A hypoxic microenvironment contributes to the chemoresistance of cancer stem cells (CSCs), but the underlying mechanism is not fully understood yet. Here, we show that increased HIF‐2α expression is associated with enhanced stemness of OCSCs and poor outcomes in ovarian cancer patients. OVCAR‐3 and CAOV‐3 sphere‐forming (OVCAR‐3 S and CAOV‐3 S) cells with OCSC‐like properties showed strong resistance to adriamycin (ADR). Hypoxia (1% O2) induced high expression of both HIF‐1α and especially HIF‐2α, and increased the resistance of OVCAR‐3 S and CAOV‐3 S cells to ADR. Notably, treatment with ADR further increased the expression of HIF‐2α, but not that of HIF‐1α. Knockdown of HIF‐2α expression substantially attenuated the resistance of OVCAR‐3 S and CAOV‐3 S cells to ADR, and the HIF‐2α overexpression had the opposite effect. Furthermore, in mouse models xenografted with OCSCs, HIF‐2α depletion significantly inhibited tumor growth and sensitized OCSCs to ADR in vivo. Mechanistically, HIF‐2α directly promotes transcription/expression of BCRP, a gene encoding a transporter protein responsible for pumping drugs (e.g., ADR) out of cells, which in turn increases drug resistance due to increased drug transportation. Collectively, our studies reveal a novel drug‐resistant mechanism in ovarian cancer by which hypoxia (and ADR treatment)‐induced HIF‐2α overexpression endows OCSCs with resistance to ADR by promoting BCRP expression and ADR transportation. Therefore, targeting the HIF‐2α/BCRP axis holds therapeutic potential for treating drug‐resistant ovarian cancer.


Introduction
Ovarian cancer is the most lethal gynecological malignancy and one of the leading causes of cancer-related mortality in women in the USA, resulting in an estimated 22 440 new cases and 14 080 deaths in 2017 according to the American Cancer Society (Siegel et al., 2017). It was estimated that about 52 100 new cases of ovarian cancer were diagnosed and 22 500 women died of ovarian cancer in China in 2015 (Chen et al., 2016). Although most ovarian tumors initially respond to chemotherapy, refractory tumors often emerge as a result of expansion of clones with either innate or acquired resistance, which then ultimately develop into recurrent tumors (Ottevanger, 2017). It is increasingly evident that heterogeneous ovarian cancers contain a subpopulation of cancer stem cells (CSCs) with enhanced self-renewal, differentiation, and chemoresistance capabilities (Jung et al., 2016). These ovarian CSCs (OCSCs), with their increased chemoresistance, are believed to contribute to tumor maintenance and recurrence; therefore, targeting OCSCs may provide a more effective novel therapeutic strategy for ovarian cancer treatment.
Hypoxia within tumors is emerging as a critical niche for CSCs. The microenvironment associated with hypoxia contributes to chemoresistance of the tumor through the generation of CSCs (Baran and Konopleva, 2017;Maugeri-Sacca et al., 2011). Hypoxia-inducible factor-2a (HIF-2a), a new recently identified member of the HIF family, complexes with HIF-1b to form a heterodimer that binds to the conserved hypoxia-response elements (HREs) in the promoters of targeted genes as a transcription factor under hypoxic conditions, similar to HIF-1a; HIF-2a is degraded in an oxygen-dependent manner via the von Hippel-Lindau protein pathway (Koh and Powis, 2012). Previous studies have shown that HIF-1 is associated with the chemotherapy failure in cancer cells including ovarian cancer (Ai et al., 2016;Heddleston et al., 2010). Accumulating evidence in recent years suggests that HIF-2 plays important roles in the chemoresistance of tumor cells, especially in promoting a more stem-like phenotype in CSCs (Pietras et al., 2009;Zhao et al., 2014). It was reported that silencing HIF-2a expression reduced in vitro self-renewal ability and in vivo tumor initiation in ALDH-positive breast cells (Kim et al., 2013). However, nothing has yet been reported regarding the regulatory effects and mechanisms of HIF-2a on the drug resistance in OCSCs, which is worthy of further investigation.
Human breast cancer resistance protein (BCRP), also known as ATP-binding cassette subfamily G member 2 (ABCG2), is a vital member of the ATP-binding cassette transporter family. BCRP expression is closely correlated with CSC-like properties; as such, it is also an important CSCs markers (Gilani et al., 2012;Szotek et al., 2006). BCRP was first identified in the multidrug resistance (MDR) breast cancer cell line MCF-7/AdrVp in 1998 (Doyle et al., 1998). BCRP normally functions to detoxify and protect normal cells from xenobiotics. Increased BCRP expression confers the MDR on CSCs, including OCSCs (Hu et al., 2010), by pumping chemotherapeutic agents such as adriamycin (ADR) and mitoxantrone (MX) out of the cell, thereby effectively reducing the intracellular concentrations of these drugs (Natarajan et al., 2012). It has been reported that the inhibition of BCRP expression effectively reduces 'stemness' and chemotherapy resistance in ovarian cancer cells (Januchowski et al., 2016;Jung et al., 2015). Reversing the chemotherapy resistance in tumors caused by OCSC via targeting BCRP might lead to the development of new strategies to cure ovarian cancer clinically.
In the present study, we found that HIF-2a expression was elevated by both hypoxia and ADR treatment, and silencing HIF-2a increased the sensitivity of the OVCAR-3 and CAOV-3 sphere-forming cells (OVCAR-3 S and CAOV-3 S) to ADR. Moreover, we demonstrated that HIF-2a overexpression enhances drug resistance of OCSCs to ADR through directly promoting BCRP transcription and increasing ADR transportation. These findings provide evidence for a novel mechanism by which HIF-2a regulates the resistance of OCSCs to ADR by directly promoting BCRP expression and thereby enhancing ADR transportation.
For exposure to hypoxic conditions, OVCAR-3 S and CAOV-3 S cells were incubated at 1% O 2 for 48 h in a controlled hypoxia incubator (HERACELL 150i, Thermo Fisher Scientific, Waltham, MA, USA) at 37°C before the indicated analysis. For the analysis of sphere formation, after OVCAR-3 S and CAOV-3 S cells were treated with ADR (60 nM) alone, 1% O 2 alone, or ADR (60 nM) + 1% O 2 for 48 h, the cells were collected, digested into single cells, and plated in 6-well ultra-low adherent plates with 2000 cellsÁwell À1 in above sphere formation medium (2 mL). After culture for 7-14 days, the number of the spheres/2000 cells was counted under an inverted microscope (Nikon TE2000-U, Tokyo, Japan).

Flow cytometry
We utilized flow cytometry in order to compare CD133 expression in OVCAR-3 vs. OVCAR-3 S cells and CAOV-3 vs. CAOV-3 S cells. Cells were suspended at a density of 1 9 10 6 cellsÁmL À1 in PBS and incubated with fluorescence isothiocyanate (FITC)-conjugated antibodies against CD133 (1 : 20, BD Pharmingen, San Diego, CA, USA) for 30 min at 4°C in darkness. For detection of ALDH activity in OVCAR-3 vs. OVCAR-3 S cells and CAOV-3 vs. CAOV-3 S cells using the ALDEO-FLUOR assay according to the manufacturer's instructions (Stem Cell Technologies, Vancouver, Canada), the ALDH-positive population was defined as cells with increased FITC fluorescence. The gates were determined using diethylaminobenzaldehyde (DEAB)-treated cells. Single-cell suspensions were analyzed by flow cytometry.
An ADR accumulation assay was performed as previously described (Zhao et al., 2016). Briefly, OVCAR-3 S and CAOV-3 S cells transfected with sh-HIF-2a or sh-NC lentiviral transduction particles or OVCAR-3 and CAOV-3 cells transfected with EPAS1-cDNA or NC-cDNA lentiviral transduction particles were incubated with 5 lM ADR at 37°C for 2 h in darkness. Next, cells were collected and washed with ice-cold PBS. Intracellular ADR levels were determined by measuring mean fluorescence intensity (MFI) using a flow cytometer (Becton-Dickinson, Franklin Lakes, NJ, USA).

Soft agar colony formation assay
The soft agar colony formation assay was performed as described previously (Mullendore et al., 2009). In brief, 1.2% SeaPlaque low melting temperature agarose (Lonza Rockland, ME, Thomaston, USA) in phenol red-free medium supplemented with 20% FBS was coated in 6well plates as a bottom layer. 2000 OVCAR-3 or CAOV-3 cells were mixed in 0.6% agarose and the same medium as a middle layer, and a top layer was 600 lL medium only. The colonies were stained with 100 lL MTT (5 mgÁmL À1 ) after incubating at 37°C in 5% CO 2 for 3 weeks. Colonies were counted using the analysis software QUANTITY ONE (Bio-Rad, Hercules, CA, USA).

ChIP assay
ChIP assays were performed using a ChIP assay kit (Beyotime, Haimen, China) following the manufacturer's instructions. Briefly, OVCAR-3 or OVCAR-3 S cells were cultured under hypoxic conditions (1% O 2 ) for 48 h in a controlled hypoxia incubator at 37°C in a 90-mm culture dish. Next, cells were resuspended in SDS lysis buffer for sonication and further immunoprecipitation with anti-HIF-2a antibody (1 : 100, abcam). Normal rabbit IgG was used as a negative control. Immune complexes were then collected, washed, and eluted. The presence of HIF-2a on the promoter region was analyzed by amplifying the DNA fragment (199 bp) corresponding to upstream HRE motifs in the BCRP gene using the following set of primers (forward: 5 0 -AATGAGCGCCTG GTGATTCT-3 0 , and reverse: 5 0 -CGATAAGCGCCCT GCGA-3 0 ) in the PCR product. To study the tumorigenic ability of OVCAR-3 and OVCAR-3 S cells, equal numbers (1 9 10 6 ) of cells were suspended in 200 lL PBS and Matrigel (1 : 1, BD Biosciences, Franklin Lakes, NJ, USA) and subcutaneously inoculated into the right flank of nude mice (n = 6 per group). Tumor volume (V) = length 9 width 2 /2. Mice were euthanized 35 days after the initial injection of the cells, and then, xenograft tumors were excised for the study.

In vivo xenograft experiments
To study the effects of silencing HIF-2a on the response of the xenograft tumors to ADR, 1 9 10 6 OVCAR-3 S cells stably transfected with sh-EPAS1 lentiviral transduction particles or sh-NC as a control were subcutaneously inoculated into nude mice as described above. Thirteen days after inoculation, the transplanted nude mice were randomly divided into four groups: sh-NC alone, sh-NC + ADR, sh-EPAS1 alone, sh-EPAS1+ ADR groups (n = 5 per group). The mice in the sh-NC + ADR and sh-EPAS1+ ADR groups were intraperitoneally injected with ADR (1.5 mgÁkg À1 ), while the mice in the sh-NC and sh-HIF-2a groups were intraperitoneally injected with DMSO once every other day. Tumor volume was calculated and recorded. Mice were sacrificed 4 weeks after the initial injection, and tumors were weighted and harvested for further study.

High-performance liquid chromatographytandem mass spectrometry (HPLC-MS/MS)
The procedure was carried out according to methods with the modifications (Ma et al., 2015). For the preparation of the tumor tissues, xenograft tumors from the sh-NC + ADR and sh-EPAS1+ ADR groups were weighed, gradually added to 1.5 mL of 0.9% sodium chloride solution, and homogenized. For preparation of transfected cells, cells were cultured in 6-well plates (3 9 10 5 /well). Twenty-four hours after transfecting OVCAR-3 S and CAOV-3 S cells with sh-EPAS1 lentiviral vector particles and OVCAR-3 and CAOV-3 cells with EPAS1-cDNA lentiviral vector particles, ADR was added to cells at the final concentration of 8 lM for 2 h. After washing cells three times with PBS, cells were cultured in fresh medium without ADR at 37°C for 1 h in darkness. Next, cells were lysed by repeated cycles of freezing and thawing and centrifuged at 13 680 g for 5 min at 4°C. One hundred microliters of the supernatant was mixed with 10 lL of pioglitazone solution as an internal standard (IS) at a final concentration of 1 lgÁmL À1 , mixed with 50 lL methanol water (1 : 1), and then vortexed for 30 S. The mixtures were shaken for 2 min after the addition of another 250 lL methanol. After centrifugation at 18 630 g for 10 min at 4°C, the supernatant was separated and transferred to an auto-sample vial and an aliquot of 5 lL was used for HPLC-MS analysis. Three independent experiments of each sample were analyzed. The biological samples were analyzed with an Agilent series 1290 UHPLC system (Agilent Technologies, Santa Clara, CA, USA) coupled to an AB 3500 triplequadrupole mass spectrometer (Agilent Technologies) via an electrospray ionization (ESI) interface. All acquisition data were analyzed using the ANALYST software version 1.6.3 package (Agilent Technologies).

Patients
Ovarian cancer tissues were obtained from 115 ovarian cancer patients who underwent surgery at Shengjing Hospital of China Medical University, Liaoning Province, China, between 2010 and 2012. The study has been approved by the Institutional Review Board of China Medical University, and all subjects gave their informed consent prior to their inclusion in the study.

Immunohistochemistry
Immunohistochemistry staining was performed as previously described . Briefly, sections from paraffin-embedded tumor tissues from ovarian cancer patients or transplanted nude mice were incubated with primary antibodies against HIF-2a (1 : 100), HIF-1a (1 : 100), CD133 (1 : 100), aldehyde dehydrogenase isoform 1A1 (ALDH1A1; 1 : 200), and BCRP (1 : 100). The immunoreactivity for ovarian cancer patient sections was evaluated by multiplying the intensity score with the score for the percentage of positively stained cells. Five random fields without overlaps from each section were counted. The intensity scores were as follows: 0 for no staining, 1 for weak staining, 2 for moderate staining, and 3 for strong staining. The scores for the percentage of positively stained cells ranged between 0% and 100%. The integrated optical density (IOD) of the immunoreactivity of transplanted nude mice sections was analyzed using IMAGE-PRO PLUS 6.0 software (Georgia Avenue Silver Spring, MD, USA).

Statistical analysis
Data were analyzed using GRAPHPAD PRISM 7 (Graph-Pad Software Inc., San Diego, CA, USA). Quantitative data are presented as the mean AE standard deviation (SD) of at least three experiments. A Student's t-test was used to compare the differences between two groups. One-way analysis of variance (ANOVA) was used for multiple comparisons. P < 0.05 was considered statistically significant.

Results
3.1. Ovarian cancer sphere-forming cells with OCSC-like characteristics are resistant to chemotherapeutic drugs and overexpress the CSC marker, BCRP Serum-free suspension culture is a common and reliable method for the culture of ovarian cancer stem cells (OCSCs). We isolated the sphere-forming cells (OVCAR-3 S and CAOV-3 S) from OVCAR-3 and CAOV-3 ovarian cancer cell lines by culturing cells in CSC culture medium according the methods described by Vermeersch et al. (2015) and Wang et al. (2012). Both CD133 expression and ALDH activity were higher in OVCAR-3 S and CAOV-3 S cells than their parental cells, as determined using flow cytometry (Figs. 1A,B and S1A). The expression levels of Nanog and OCT4, stem cell markers, were significantly increased in OVCAR-3 S and CAOV-3 S cells via immunoblot analysis (Fig. 1C). Moreover, we found that OVCAR-3 S and CAOV-3 S cells formed more colonies than OVCAR-3 and CAOV-3 cells in a soft agar colony formation assay (Fig. S1B). Transwell invasion assays showed that the number of OVCAR-3 S and CAOV-3 S cells that invaded into the lower transwell chamber was significantly greater than the number parental cells that were able to invade (Fig. S1C). We also found that OVCAR-3 S and CAOV-3 S cells re-differentiated into their parental OVCAR-3 and CAOV-3 cells in DMEM culture medium supplemented with 10% FBS for 72 h (Fig. S1D) by referring to Ponti et al.'s (2005) and Wu et al.'s (2010) reports. In addition, we observed that injection of OVCAR-3 S cells (1 9 10 3 ) could form tumors in xenograft mice, whereas 1 9 10 5 parental OVCAR-3 cells were required to generate tumors in vivo (Fig. 1D). Five weeks after 1 9 10 6 OVCAR-3 S cells were injected, the average tumor volume and weight of the resulting xenograft tumors were much higher than those of parental OVCAR-3 xenograft tumors ( Fig. 1E-G). The expression levels of CD133 and OCT4 proteins were also higher in OVCAR-3 S xenograft tumors (Fig. 1H). These data suggest that OVCAR-3 S and CAOV-3 S cells obtained from serum-free suspension culture possess ovarian CSC-like properties such as self-renewal, strong invasion capacity, differentiation potential, and high tumorigenicity.
Chemoresistance is one of the most important characteristics of OCSCs. We assessed the sensitivity difference of OVCAR-3 vs. OVCAR-3 S and CAOV-3 vs. CAOV-3 S cells to ADR, MX, PTX, VP-16, and DDP treatment for 48 h using CCK-8 assays. We found that OVCAR-3 S and CAOV-3 S cells exhibited resistance to all these drugs with higher IC 50 values as compared with their parental cells (Figs 1I,J and S2A-D). OVCAR-3 S and CAOV-3 S cells were most strongly resistant to ADR; the resistant indices were 103.6 and 24.7 relative to OVCAR-3 and CAOV-3 cells, respectively. Thus, we utilized ADR to evaluate the mechanisms of OCSC resistance in subsequent experiments.
Breast cancer resistance protein, as a major CSC marker, plays an important role in the MDR of CSCs. To evaluate whether BCRP confers ADR resistance on OVCAR-3 S and CAOV-3 S cells, we measured the BCRP protein expression levels in OVCAR-3 S and CAOV-3 S as compared to their parental cells. Western blot analysis showed that the expression levels of BCRP in OVCAR-3 S and CAOV-3 S cells were 3.61-fold and 3.35-fold higher than in OVCAR-3 and CAOV-3 parental cells, respectively (Fig. 1K). These data suggest BCRP may contribute to ADR resistance in OCSCs.

Hypoxia mediates the resistance of OCSCs to ADR by inducing high expression of HIF-2a
Hypoxia can promote 'stemness' maintenance of CSCs and is a functional component of CSC niches (Takakura, 2012). In order to evaluate the association HIF-1a or HIF-2a expression with stem cells markers in clinical ovarian cancer patients, we first used the Cancer Genome Atlas (TCGA) database to analyze the relationship between mRNA expression of HIF-1A or EPAS1 and ALDH1A1, a CSC marker, in 379 ovarian cancer cases. The mRNA expression of EPAS1 (but not HIF-1A) showed a significantly positive correlation with mRNA expression of ALDH1A1 ( Fig. 2A). Next, we used the online tool, KM plotter (http://www.km plot.com), to analyze the relationship between mRNA expression of HIF-1a or HIF-2a and the outcomes of 1657 ovarian cancer cases. We found that higher mRNA expression of EPAS1 was significantly positively associated with shorter OS of patients, but higher mRNA expression of HIF-1A had no positive association with worse outcomes of patients (Fig. 2B). In addition, we tested the protein expression of HIF-1a or HIF-2a and ALDH1A1 or CD133, another CSC marker, in 115 ovarian tumor tissues using immunohistochemical staining. The clinicopathological features were shown in Table S1. A Pearson correlation OVCAR-3 S cells and CAOV-3 vs. CAOV-3 S cells. Data are presented as mean AE SD from three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001. (K) The protein expression levels of BCRP were analyzed by western blot in OVCAR-3 S and CAOV-3 S cells versus their parental cells. b-actin was used as a loading control. The expression level of BCRP was normalized to that of b-actin. The BCRP expression level in OVCAR-3 cells or CAOV-3 cells was set as 1. Data are presented as mean AE SD from three independent experiments. **P < 0.01. Statistical significance was evaluated by Student's t-test.
analysis indicated that expression of HIF-2a was significantly positively correlated with the expression of ALDH1A1 and CD133, but the expression of HIF-1a was only positively correlated with the expression of ALDHA1 (Figs 2C and S3). These data suggest HIF-2a may contribute to the 'stemness' of OCSCs and poor outcomes of ovarian cancer patients.
To study the effects of hypoxia on the sensitivity of OCSCs to ADR, we analyzed the cell survival rate via CCK-8 assay after culturing OVCAR-3 S and CAOV-3 S cells with different concentrations of ADR under hypoxic conditions (1% O 2 ) for 48 h. We found that hypoxia decreased the response of OVCAR-3 S and CAOV-3 S cells to ADR; ADR IC 50 values were higher than when cells were grown under normoxia (Fig. 2D). In addition, sphere formation assays demonstrated that either ADR treatment (60 nM) alone or hypoxia treatment alone increased sphere numbers of OVCAR-3 S  The relationship between mRNA expression of HIF-1A or EPAS1 and ALDH1A1 was analyzed in 379 ovarian cancer cases from TCGA data. (B) The relationship between mRNA expression of HIF-1A or EPAS1 and the outcomes of 1657 ovarian cancer cases was analyzed using the online tool, KM plotter (http://www.kmplot.com). (C) The protein expression correlation between HIF-1a or HIF-2a and ALDH1A1 or CD133 in 115 ovarian tumor tissues was analyzed using Pearson correlation analysis. (D) Cell survival rate was analyzed by CCK-8 assays, and the IC 50 values were compared 48 h after OVCAR-3 S and CAOV-3 S cells were treated with different concentrations of ADR under normoxic or hypoxic conditions (1% O 2 ). Data are presented as the mean AE SD from three independent experiments. **P < 0.01, compared with the cells under normoxia. (E) Sphere formation assays were carried out in OVCAR-3 S and CAOV-3 S cells 48 h after treatment with ADR (60 nM) alone, 1% O 2 alone, or ADR (60 nM) + 1% O 2 . The length of the scale bars is 200 lm. Data are presented as the mean AE SD from three independent experiments. *P < 0.05, **P < 0.01. (F) Protein expression levels of HIF-1a and HIF-2a were analyzed by western blot in OVCAR-3 S and CAOV-3 S cells 48 h after treatment with 60 nM ADR or hypoxia (1% O 2 ). b-actin was used as a loading control. The expression levels of HIF-1a or HIF-2a were normalized to that of b-actin. The HIF-1a or HIF-2a expression levels in OVCAR-3 cells or CAOV-3 cells without ADR or exposure to hypoxia (1%O 2 ) were set as 1. Data are presented as the mean AE SD from three independent experiments. *P < 0.05, **P < 0.01. Statistical significance was evaluated by one-way ANOVA.

410
and CAOV-3 S cells as compared to untreated cells. Also, compared with cells treated with ADR under normoxic conditions, OVCAR-3 S and CAOV-3 S cells treated with ADR under hypoxic conditions formed more spheres (Fig. 2E), suggesting that hypoxia can induce the resistance of OCSCs to ADR.
To study whether HIF-1a or HIF-2a plays a role in the OCSC resistance to ADR induced by hypoxia, we measured protein expression levels of HIF-1a and HIF-2a in OVCAR-3 S and CAOV-3 S cells 48 h after hypoxic treatment (1% O 2 ) or ADR (60 nM) by western blot. We found that hypoxia induced higher expression of HIF-1a and especially HIF-2a in OVCAR-3 S and CAOV-3 S cells as compared to untreated cells. Importantly, we observed that treatment with ADR significantly upregulated HIF-2a 412 expression, but not HIF-1a expression in OVCAR-3 S and CAOV-3 S cells (Fig. 2F). We also found the increased HIF-1a, especially HIF-2a expression in OVCAR-3 S and CAOV-3 S cells relative to their parental cells (Fig. S4A). Additionally, hypoxia culture for 48 h induced the increased HIF-1a, especially HIF-2a expression in OVCAR-3 and CAOV-3 cells (Fig. S4B).
Collectively, these findings suggest that hypoxia may mediate the resistance of OCSCs to ADR by inducing high expression of HIF-2a.
A CCK-8 assay demonstrated that HIF-2a silencing in OVCAR-3 S and CAOV-3 S cells dramatically reduced the cell survival rate after treatment of different concentrations of ADR for 48 h as indicated by significantly reduced IC 50 values relative to the sh-NC-transduced cells (Fig. 3A,B). Though HIF-1A silencing also decreased the cell survival rate, the effects were not as pronounced as EPAS1 silencing. Meanwhile, we found that HIF-2a, but not HIF-1a overexpression, resulted in an elevated cell survival rate after treatment with ADR as indicated by increased IC 50 values in OVCAR-3 cells and CAOV-3 cells (Fig. 3C,D). Furthermore, sphere formation assays indicated that EPAS1 suppression significantly reduced sphere numbers of OVCAR-3 S and CAOV-3 S cells, whereas HIF-1A suppression did not (Fig. 3E). In addition, overexpression of HIF-2a, but not HIF-1a, significantly increased the sphere numbers of OVCAR-3 and CAOV-3 cells, relative to the NC-cDNA-transduced cells in colony formation assays (Fig. 3F). Western blot analysis also showed that overexpression of HIF-2a, but not HIF-1a, notably increased the protein expression of OCT4 in OVCAR-3 and CAOV-3 cells (Fig. S7). Thus, these findings suggest that HIF-2a, as opposed to HIF-1a, plays an important role in regulating the resistance of OCSCs induced by hypoxia to ADR.

HIF-2a regulates the resistance of OCSCs to ADR by affecting the expression and transport function of BCRP
How does HIF-2a regulate the resistance of OCSCs to ADR? Considering the expression of BCRP, a CSC marker, was upregulated in OCSCs, we considered whether the high expression of BCRP contributed to the

(B) Correlation between protein expression of HIF-2a and BCRP in 115 ovarian cancer tissues was analyzed by linear regression analysis. (C) The protein expression of BCRP was analyzed by western blot in OVCAR-3 S and CAOV-3 S cells transduced with sh-EPAS1 or sh-NC lentivirus under hypoxic conditions (1% O 2 ) for 48 h and (D) in the OVCAR-3 and CAOV-3 cells transduced with EPAS1-cDNA or NC-cDNA lentivirus. b-actin was a used an endogenous control. BCRP expression levels in the OVCAR-3 S cells and CAOV-3 S cells transduced
with sh-NC lentivirus and OVCAR-3 cells and CAOV-3 cells transduced with NC-cDNA were set as 1. Data are presented as the mean AE SD from three independent experiments. *P < 0.05, **P < 0.01. (E) Changes in the intracellular accumulation of ADR were determined by flow cytometric analysis in OVCAR-3 S and CAOV-3 S cells transduced with sh-EPAS1 or sh-NC lentivirus under hypoxic conditions (1% O 2 ) for 48 h, and (F) in the OVCAR-3 cells or CAOV-3 cells transduced with EPAS1-cDNA or NC-cDNA lentivirus. The MFI of the intracellular accumulation of ADR in OVCAR-3 S cells or CAOV-3 S cells transduced with sh-NC lentivirus and in OVCAR-3 cells or CAOV-3 cells transduced with NC-cDNA was set as 1. Data are presented as mean AE SD from three independent experiments. **P < 0.01. (G) Changes in intracellular accumulation of ADR in EPAS1-silenced OVCAR-3 S and CAOV-3 S cells and (H) EPAS1-overexpressing OVCAR-3 and CAOV-3 cells were analyzed by mass spectrometry. Data are presented as the mean AE SD from three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001. Statistical significance was evaluated by Student's t-test. regulation of the HIF-2a and thus the resistance of OCSC to ADR. To test this hypothesis, we first used the TCGA data to analyze the relationship between HIF-2a mRNA expression and BCRP in 379 ovarian cancer cases. The results showed that the HIF-2a mRNA expression was significantly positively associated with EPAS1 mRNA expression (Fig. 4A). We also analyzed the correlation between protein expression of HIF-2a and BCRP in 115 ovarian cancer tissues from our case bank using linear regression analysis. Protein expression of HIF-2a was also positively correlated with protein expression of BCRP (Fig. 4B). The evolutionarily conserved hypoxia-response element (HRE) sequence (CACGTG), located between À483 and À478 nucleotides upstream of the transcription start site of the human BCRP gene. (B) Luciferase reporter activity was measured in transient co-transfections with EPAS1-cDNA or NC-cDNA plasmids and BCRP promoter vectors (GV238-BCRP-WT) or vectors containing mutated HRE sequence binding sites (GV238-BCRP-mut) in 293T and OVCAR-3 cells. The luciferase activities in the cells co-transfected with NC-cDNA and GV238-BCRP-WT or GV238-BCRP-mut plasmids were set as 1. Data are presented as the mean AE SD from three independent experiments. (C) ChIP assays were performed to verify HIF-2a binding to the BCRP gene in OVCAR-3 and OVCAR-3 S cells cultured under hypoxic conditions (1% O 2 ) for 48 h. (D) ChIP-qPCR shows the enhanced binding HIF-2a on BCRP promoter in OVCAR-3 S cells. Antibody enrichment was quantified relative to the amount of input DNA. Antibody directed against IgG was used as a negative control. (E) The mRNA expression of BCRP was analyzed by qRT-PCR in the OVCAR-3 and CAOV-3 cells transduced with EPAS1-cDNA or NC-cDNA lentivirus, and in the OVCAR-3 S and CAOV-3 S cells transduced with sh-EPAS1 or sh-NC lentivirus under hypoxic conditions (1% O 2 ) for 48 h. BCRP expression levels in OVCAR-3 cells or CAOV-3 cells transduced with EPAS1-cDNA lentivirus and in OVCAR-3 S cells or CAOV-3 S cells transduced with sh-NC lentivirus were set as 1. Data are presented as the mean AE SD from three independent experiments. *P < 0.05, **P < 0.01. Statistical significance was evaluated by Student's t-test.
To evaluate whether HIF-2a regulates the resistance of OCSCs to ADR by affecting the expression and transport function of BCRP, we first analyzed BCRP protein expression in the OVCAR-3 S and CAOV-3 S cells transduced with sh-EPAS1 lentivirus or EPAS1-cDNA lentivirus by western blot. HIF-2a depletion significantly reduced the BCRP protein levels in the OVCAR-3 S and CAOV-3 S cells (Fig. 4C), while EPAS1 overexpression significantly upregulated the protein expression of BCRP in OVCAR-3 and CAOV-3 cells (Fig. 4D).
Next, flow cytometric analysis was performed to measure changes in the intracellular accumulation of ADR in EPAS1-silenced OVCAR-3 S and CAOV-3 S cells and EPAS1-overexpressing OVCAR-3 and CAOV-3 cells. The MFI of ADR in EPAS1-silenced OVCAR-3 S and CAOV-3 S cells was increased relative to control cells (Fig. 4E). Conversely, the MFI of ADR in EPAS1-overexpressing OVCAR-3 and CAOV-3 cells was reduced compared to control cells (Fig. 4F). We also used mass spectrometry to analyze the changes in intracellular accumulation of ADR in the EPAS1-silenced OVCAR-3 S and CAOV-3 S cells and in EPAS1-overexpressing OVCAR-3 and CAOV-3 cells. Consistent with the results from flow cytometric analysis, mass spectrometry also showed that silencing EPAS1 led to increased cellular accumulation of ADR in the OVCAR-3 S and CAOV-3 S cells (Figs 4G and S8A-C), while overexpressing HIF-2a resulted in decreased cellular accumulation of ADR in the OVCAR-3 and CAOV-3 cells (Figs 4H and S8D). Therefore, these findings indicate that HIF-2a might regulate the resistance of OCSCs to ADR by affecting the expression and transport function of BCRP.

HIF-2a directly activates expression of the BCRP gene in ovarian cancer cells
To further study the exact mechanism by which HIF-2a regulates the expression and function of BCRP, we first analyzed sequences in 2-kb upstream of the BCRP gene. We found that there was an evolutionarily conserved hypoxia-response element (HRE) sequence (CACGTG) located between À483 and À478 bases upstream of the transcription start site of the BCRP gene (positions 88158941-88158936 in BCRP gene sequence (GenBank accession number NM_001257386.2, Fig. 5A)). Next, we utilized BCRP promoter vectors (GV238-BCRP-WT) or vectors with mutated HRE sequence binding sites (GV238-BCRPmut) to drive a luciferase reporter gene in transient cotransfections with HIF-2a-cDNA plasmids in 293T and OVCAR-3 cells. We found that ectopic expression of HIF-2a strongly increased luciferase activity in cells transfected with GV238-BCRP-WT plasmids, but not in cells transfected with GV238-BCRP-mut plasmids (Fig. 5B).
We also determined whether HIF-2a bound to the BCRP promoter in vivo using ChIP assays. As shown in Fig. 5C, chromatin solutions prepared from OVCAR-3 and OVCAR-3 S cells were used to immunoprecipitate the HIF-2a-DNA complex using anti-HIF-2a serum or control IgG serum followed by PCR amplification of the BCRP gene sequence. Using primers designed to amplify the HRE, we were able to amplify product from the DNA bound to protein precipitated with the anti-HIF-2a serum, but not from DNA precipitated using control antibodies, indicating the specificity of HIF-2a binding to the BCRP promoter. Moreover, we found that HIF-2a highly bound BCRP gene promoter region in OVCAR-3 S cells compared to that in parental cells (Fig. 5C) because OVCAR-3 S cells have higher expression of HIF-2a than OVCAR-3 cells (Fig. S4A). ChIP-qPCR further demonstrated the enhanced binding HIF-2a on BCRP promoter in OVCAR-3 S cells (Fig. 5D). Additionally, we also performed ChIP assay of HIF-1a binding with VEGF-HRE and BCRP-HRE in OVCAR-3 and OVCAR-3 S cells, and found the binding of HIF-1a with VEGF-HRE, but no binding with BCRP-HRE (Fig. S9A,B). Thus, the results suggest that BCRP promoter is targeted by HIF-2a but not by HIF-1a in OVCAR-3 cells.
Next, we determined the BCRP mRNA expression in the HIF-2a-overexpressing OVCAR-3 and CAOV-3 cells and in the OVCAR-3 S and CAOV-3 S cells transduced with sh-EPAS1 lentivirus using qRT-PCR. HIF-2a overexpression significantly upregulated the BCRP mRNA expression in the OVCAR-3 and CAOV-3 cells, while HIF-2a depletion significantly reduced BCRP mRNA expression levels in OVCAR-3 S and CAOV-3 S cells (Fig. 5E). Collectively, the data confirm that HIF-2a can directly activate BCRP gene expression in ovarian cancer cells.

Silencing HIF-2a inhibits the growth of OCSCs in vivo
To further investigate whether HIF-2a could mediate the response of OCSCs to ADR by regulating BCRP in vivo, we used the BALB/c (nu/nu) mouse model to evaluate the effects of HIF-2a expression changes on the sensitivity of OCSCs to ADR. OVCAR-3 S cells (1 9 10 6 ) stably transfected with sh-EPAS1 lentiviral transduction particles, or sh-NC as a control, were subcutaneously injected into nude mice. Thirteen days after inoculation, transplanted nude mice were intraperitoneally injected with ADR (1.5 mgÁkg À1 ) or DMSO as a control once every other day, respectively (Fig. 6A). We observed that the tumor volumes and tumor weights in the xenografted mice transplanted with sh-EPAS1-transduced OVCAR-3 S cells were both significantly reduced compared with those in sh-NC xenografted mice. Additionally, ADR treatment showed weakly inhibitory effects on the tumor volumes and tumor weights in sh-NC xenografted mice (P < 0.05), and tumor growth was notably decreased in sh-EPAS1 xenografted mice treated with ADR (P < 0.001, Fig. 6B-D). These findings demonstrate that decreased HIF-2a could inhibit the growth of OCSC in vivo.
To further evaluate the association between the expression of HIF-2a and BCRP in vivo, we measured HIF-2a and BCRP mRNA expression using RT-PCR and protein expression using immunohistochemical staining and western blot of the transplanted tumors. As shown in Figs 6E,F and S10A,B, the mRNA and protein expression levels of HIF-2a and BCRP were significantly reduced in the transplanted tumors transduced with sh-EPAS1 lentivirus as compared with those transduced with sh-NC lentivirus. In addition, mRNA and protein expression levels of HIF-2a and BCRP were significantly decreased in the sh-EPAS1transduced mice treated with ADR as compared to those in the sh-NC-transduced mice treated with ADR. We also evaluated the effects of silencing HIF-2a on the ability of BCRP to transport ADR in vivo using mass spectrometry. As expected, our results indicated that the ADR accumulation in sh-EPAS1-transduced OVCAR-3 S xenograft tissue was much higher than the sh-NC-transduced OVCAR-3 S xenograft tissue (Fig. 6G). Thus, these data suggest that silencing HIF-2a can inhibit the growth of OCSC by decreasing the expression and function of BCRP in vivo.

Discussion
OCSCs are believed to be the cause of ovarian cancer chemoresistance and recurrence (Jung et al., 2016). However, the mechanisms underlying OCSC-mediated chemoresistance remain unclear. OCSCs were first reported as multilayered spheroids, displaying heterogeneity and stem-like qualities, and were derived from the ascites of a patient with advanced ovarian cancer (Bapat et al., 2005). In the present study, we obtained OVCAR-3 and CAOV-3 sphere-forming cells (OVCAR-3 S and CAOV-3 S) by culturing OVCAR-3 and CAOV-3 cells in suspension in serum-free medium as previously reported by Wang et al. (2012). OVCAR-3 S and CAOV-3 S cells exhibited OCSC-like characteristics such as high expression of the stem cell markers OCT4 and nanog, strong colony formation ability, increased invasion capability, and strong tumorigenicity in vivo. CD133, a cellular surface protein, has been reported as a potential OCSC marker (Curley et al., 2009). In addition, the high enzymatic activity of aldehyde dehydrogenase (ALDH) was also defined as a marker for various CSCs, including OCSCs (Silva et al., 2011). In our study, OVCAR-3 S and CAOV-3 S cells exhibited increased expression of CD133 and higher enzymatic activity of ALDH relative to the parental OVCAR-3 and CAOV-3 cells, which was consistent with Wang et al.'s (2012) previous reports. Importantly, we found that OVCAR-3 S and CAOV-3 S cells exhibited reduced sensitivity to the anticancer drugs ADR, MX, PTX, VP-16, and DDP as compared with their parental cells. The RI values of OVCAR-3 S cells relative to OVCAR-3 cells were 103.6, 20.9, 17.3, 24.9, and 3.55, respectively, and the RI values of CAOV-3 S cell relative to CAOV-3 cells were 24.7, 14.1, 12.8, 15.9, and 3.33, respectively. Overall, OVCAR-3 S and CAOV-3 S cells showed the strongest resistance to ADR. Meirelles et al. reported that the OCSCs isolated from OVCAR-7 ovarian cancer cell lines were resistant to ADR (Meirelles et al., 2012). Therefore, the OVCAR-3 S and CAOV-3 S cells in the present study can be reliably used to investigate the effects and mechanisms of the resistance to ADR on OCSCs.
Hypoxic microenvironments have been identified in many solid tumors and increase CSC characteristics of tumor cells, including the resistance to chemotherapeutic agents (Axelson et al., 2005). In this study, we used Error bars indicate SD. *P < 0.05, ***P < 0.001. (E) Representative immunohistochemical staining for the expression of HIF-2a and BCRP proteins and statistical results for IOD in the xenograft tumors. Error bars indicate SD. *P < 0.05, **P < 0.01, ***P < 0.001. The length of the scale bars is 20 lm. (F) Representative images for the expression of HIF-2a and BCRP proteins in the xenograft tumors by western blot and the relative protein expression of HIF-2a and BCRP normalized to those of b-actin were analyzed. The expression levels of HIF-2a/BCRP in the xenograft tumors derived from cells transduced with sh-NC and without ADR treatment were set as 1. Error bars indicate SD. *P < 0.05, **P < 0.01, ***P < 0.001. (G) ADR accumulation in xenograft tumors derived from sh-EPAS1 or sh-NC transduced OVCAR-3 S cells by mass spectrometry. The representative chromatograms were for ADR, internal standard (IS) in the sh-NC xenograft tumors with ADR treatment, and ADR, IS in the sh-EPAS1 xenograft tumors with ADR treatment. RT, Retention Time. Error bars indicate SD. ***P < 0.001. Statistical significance was evaluated by one-way ANOVA. target for reversing OCSC-mediated chemoresistance and HIF-2a inhibitors in combination with conventional chemotherapy are expected to improve therapeutic outcomes in ovarian cancer patients.
Elimination of head and neck cancer initiating cells through targeting glucose regulated protein78 signaling. Mol Cancer 9, 283. Xiang L, Liu ZH, Huan Q, Su P, Du GJ, Wang Y, Gao P and Zhou GY (2012)