Generation of Prostate Tumor–Initiating Cells Is Associated with Elevation of Reactive Oxygen Species and IL-6/STAT3 Signaling

Prostate cancer is the most frequently diagnosed malignancy among males in Western countries and is intimately associated with aging (1). Reactive oxygen species (ROS) is one of the major aging-associated influences on prostate carcinogenesis (2). Although recent epidemiologic and clinical studies have linked prostate cancer risk and ROS, the underlying mechanism remains to be elucidated (3).

In prostate cancer research, one of the big obstacles is the lack of relevant preclinical models to understand the mechanism of human prostate carcinogenesis and to develop effective preventive and therapeutic interventions. In the past 40 years, several malignant transformations of human prostate cells have been established by radiation (4) or chemical treatment, such as cadmium (5), N-nitroso-N-methylurea (6), or introduction of virus elements (7). In patients, however, prostate cancer is unlikely to be caused by exposure to such strong external carcinogens but is considered mainly a disease of aging.

Previously, we have attempted to establish a prostate carcinogenesis model based on physiologic selection and adaption. We started with a human primary prostate basal cell line EP156T that was derived from human prostate benign tissue and immortalized using the human telomerase catalytic subunit (hTERT; ref. 8). Subsequently, nonmalignant mesenchymal EPT1 cells and premalignant EPT2-D5 cells were derived from EP156T cells in a stepwise manner by selection of cells with loss of contact inhibition and evasion of quiescence, respectively (9, 10). EPT2-D5 cells showed several in vitro malignant features such as colony formation in soft agar, resistance to apoptosis, and independence of exogenous growth factors. However, EPT2-D5 cells failed to form xenograft tumors following repeated attempts (9), suggesting that additional change is needed to acquire full malignant transformation.

Self-sufficiency in growth signals is a classic oncogenic feature of cancer cells (11–13). No type of normal cell can proliferate in culture without supplement of serum or growth factors, whereas cancer cells invariably show greatly reduced dependence on exogenous growth stimulation and uncontrolled proliferation (12, 13). In addition, chemically defined serum-free medium is widely used to enrich and culture tumor-initiating cells (TIC; refs. 14–16). In this work, we attempted to obtain TICs by adapting EPT2-D5 cells in chemically defined medium without any protein component. Surprisingly, numerous tight prostate spheres were generated from single cells in monolayer culture and efficiently established xenograft tumors.

The prostate cell line EPT2-D5 was grown in MCDB153 medium (Biological Ind. Ltd.) supplemented with 1% Minimum Essential Media (MEM) nonessential amino acids solution, 200 nmol/L hydrocortisone, 10 nmol/L triiodothyronine, 5 μg/mL insulin, 5 μg/mL transferrin, 5 μg/mL sodium selenite, 100 ng/mL testosterone, 5 ng/mL EGF, 50 μg/mL bovine pituitary extract (Invitrogen Life Sciences), 100 U/mL penicillin, 100 μg/mL streptomycin, and 5% fetal calf serum (FCS). Prostate cancer PC3 and DU145 cell lines were grown in Ham's F-12 medium and Dulbecco's Modified Eagle's Medium (DMEM; Lonza) containing 10% FCS, respectively. The medium was changed every 3 days. All the cell lines have been tested and authenticated using forensic grade DNA microsatellite analyses, karyotyping, and copy number analyses in parallel with the experiments reported in this work. Lentiviral pGF-STAT3 reporter was a gift from Dr. Yong-Joon Chwae at Ajou University School of Medicine, Korea. Interleukin (IL)-6 protein (H7416) and chemical reagents cryptotanshinone (C5624), FLLL31 (F9057), AG490 (T3434), N-acetyl-l-cysteine (A9165), and melatonin (M5250) were products from Sigma-Aldrich. A neutralizing antibody against IL-6 was from Abcam (ab6672).

EPT2-D5 cells were seeded in T75 tissue culture flasks (TPP) to reach 30% confluence. On the second day, the complete medium was removed, and cells were washed twice with PBS and cultured in basic Ham's F-12 medium (Lonza) without serum and any other protein components (protein free medium). The medium was changed every 3 days. When the culture reached confluence, cells were trypsinized and the reaction was quenched by soybean trypsin inhibitor (Invitrogen), the cells were pelleted, resuspended, and split to 2 new flasks. The protein-free medium was changed every 3 days until the spheres were generated about 2 weeks later. To passage the prostate spheres, culture medium was collected in 15-mL centrifuge tubes and kept at room temperature for 10 minutes to let spheres settle down to the bottom of tubes. The upper nine tenths of medium was gently removed, and the remaining lower part contained most of the big spheres (>50 μm in diameter) and was gently centrifuged at 1300 × g for 2 minutes. The pelleted spheres were dissociated enzymatically (0.05% trypsin, 0.53 mmol/L EDTA) and mechanically (pipetting). The dissociated cells were analyzed microscopically for single cellularity. The single cells were transferred to and attached in new T75 tissue culture flask. The protein medium was changed every 3 days until new spheres again were generated after around 10 days.

Cells were grown in complete medium or protein-free medium. Production of ROS at various conditions or treatments was measured by incubation with membrane permeable dye 2′,7′-dichlorofluorescein diacetate (DCFDA; Sigma-Aldrich; final concentration 1 μmol/L) for 10 minutes at 37°C. The ROS levels in the cells were determined by fluorescent microscopy or measurement of the fluorescence (excitation 500 and emission 520 nm) using AccuriC6 flow cytometer (Accuri Cytometers, Inc.).

Expression levels of IL-6 in cell supernatants were determined using a commercial sandwich ELISA kit (Fisher Thermo Scientific). Cells were cultured for 3 days and the medium was harvested and centrifuged at 1300 × g for 5 minutes. The supernatant was used for determination of IL-6 levels according to the manufacturer's protocol.

The levels of expression and phosphorylation of STAT3 protein were determined by Western blotting following the procedures described previously (10). All the antibodies used in Western blotting were products of Abcam with the dilutions as following: anti-phospho-specific STAT3 (Tyr705) (ab76315) 1:10,000; total STAT3 (ab119352) 1:4,000; endogenous control β-actin (ab11003) 1:1,000.

Total RNA was converted to Cy3-labeled cRNA targets and hybridized to Agilent Whole Human Genome Microarrays 44k (Cat.no. G4112F and G4845A, Agilent Technologies) according to Agilent's instructions. Arrays were scanned using the Agilent Microarray Scanner Bundle. Raw data were imported and analyzed in J-Express software (Molmine, http://www.molmine.com). Mean spot signals were used as intensity measure, the expression data were quantile normalized over the entire arrays and log₂-transformed. Differentially expressed genes were identified using the feature subset selection (FSS) method. Only genes that changed more than 3.0 fold with P < 0.01 were considered as differentially expressed.

Reverse transcription and real-time quantitative PCR (qPCR) were done as described (17). The TaqMan assays used for quantification of human IL-6 (Hs00985639_m1), IL-6R (Hs00169842_m1), and β-actin (Hs99999903_m1) were obtained from Applied Biosystems.

Proteins harvested from serum-free culture were digested with trypsin and analyzed by mass spectrometry following a protocol described previously (18). Details of mass spectrometric analysis are described in Supplementary Methods.

For subcutaneous tumor formation, BALB/c nude mice were purchased from Shanghai Experimental Animal Center, Chinese Academy of Sciences and maintained under pathogen-free conditions. About 5 × 10⁶ EPT2-D5 or D5H cells or 4 × 10³ D5HS spheres were resuspended in 200 μL PBS and injected subcutaneously into BALB/c nude mice. Five mice were used for each cell type. The tumor was removed from sacrificed nude mice when the diameter was about 70 mm. After mechanical (cutting) and enzymatic (collagenase IV for 30 minutes at 37°C) dissociation, the cells were suspended in Ham's F-12 medium containing 5% FCS and plated in tissue culture flasks. The subcutaneous tumor experiment was approved by the Second Military Medical University Committee on Animal Care.

For orthotopic xenograft tumor experiments, NOD/SCID IL2rγnull mice (male, 6–10 weeks old, the Gade Institute, University of Bergen, Bergen, Norway) were used and experiments were approved by the Norwegian Animal Research Authority and conducted according to The European Convention for the Protection of Vertebrates Used for Scientific Purposes. Bioluminescent EPT3 or EPT3-M1 cells were suspended in 50 μL PBS and injected in the left or right ventrolateral lobes of the prostate. Bioluminescence images were conducted 10 minutes following administration of 150 mg/kg d-luciferin (Promega) intraperitoneally. All Optix images were acquired using Optix MX2 Time-Domain Molecular Imager (ART Inc.). or In Vivo MS FX PRO (Carestream Health Inc.) as previously described (19) or Kodak Molecular imaging software (Version 5.0 Carestream Health Inc).

Mice were euthanized and sacrificed and tissues were collected by a certified veterinary. Tumor tissues were fixed in neutral-buffered 4% formaldehyde. Paraffin embedding, preparation of sections, staining, and immunohistochemical (IHC) examination were conducted using standard procedures. Briefly, paraffin-embedded tissue slides were deparaffinized and processed for antigen retrieval using PTLink and pretreatment buffer at pH 9.0 (Dako, DK, S2375) and 100°C for 20 minutes. Finished pretreated slides were washed in TBS. Staining was conducted using specific antibodies and a single staining rabbit/mouse kit (Dako Real Envision K5007) using peroxidase/DAB+. Counter staining was carried out by a hematoxylin staining kit. Histologic images were captured using the Qcapture Suite software with a Qimaging Exi Blue camera attached to a Leica DMRBE microscope. The following primary antibodies and dilutions were used for antigen detection in IHC: AE1/AE3 (ab961, Abcam) prediluted; vimentin (ab8545) prediluted; IL-6 (ab6672) 1:400; phospho-STAT3 Y705 (ab76315) 1:100; STAT3 (ab119352) 1:1,600; cyclin D1 (ab10540) 2 μg/mL, and human cell–specific anti-mitochondrion antibody (ab92824, 1:1,000). Inclusion of only the secondary and not the primary antibody in parallel provided negative controls, in addition to differential staining of different neighboring tissue types in the sections.

Two-sample t tests were used for differences of sphere formation, IL-6, or ROS levels between different groups. ANOVA was used to identify the differentially expressed genes in microarray data.

The gene microarray data are accessible in database ArrayExpress (ID: E-MTAB-1521) according to MIAME guidelines.

EPT2-D5 cells showed malignant features in vitro but failed to form xenograft tumors (9). To further promote malignant cells, we cultured EPT2-D5 cells in chemically defined medium without any protein components. Most EPT2-D5 cells can survive well in protein-free medium, but the growth speed was much slower than under standard culture conditions (9). When cells were split to new flasks after reaching confluence, cells kept growing in monolayer and surprisingly, multiple tight and large spheres (>50 μm in diameter) were generated and detached from the surface in 2 weeks (Fig. 1A and B). The floating spheres were collected and dissociated to single cells (Fig. 1C). These single cells attached efficiently to the surface and expanded well in new tissue culture flasks (Fig. 1D). New spheres were visible again after 5 days and floating spheres could be collected in 10 days (Fig. 1E and F). We named this process attachment sphere cycles and the cells generating these spheres D5HS (Fig. 1G). This D5HS cycle continued for up to 1 year. The attachment of growing spheres to adherent individual cells in monolayer culture showed that these spheres were generated from single cells (Fig. 1H). The remaining cells in the flask continued forming spheres for up to 3 months even when the floating spheres were harvested every third day (Fig. 1I). This indicates that the spheres were indeed generated by cell proliferation and not by cell aggregation and the whole-cell population did not start to generate spheres simultaneously. In the following study, all the D5HS spheres used were newly generated (passage 2). When D5HS cells were cultured in the presence of serum, cells grew only in monolayer and no sphere was generated, showing the indispensability of protein deprivation for sphere generation.

The formation of spheres is one important feature of TICs in vitro (16). We tested the tumorigenicity of newly generated D5HS by subcutaneous injection in BALB/c nude mice. As controls, EPT2-D5 cells grown in complete medium (D5) and EPT2-D5 cells that were adapted to protein-free medium for 3 days (D5H) were also injected in parallel. Significantly, only D5HS spheres efficiently formed large subcutaneous tumors within 12 weeks (Fig. 2A), indicating the tumor initiation ability of D5HS spheres (tumor spheres). Cells recovered from the subcutaneous tumors were named EPT3. To obtain metastatic tumors, luciferase-expressing EPT3 cells were inoculated orthotopically into the prostate of mice. Large primary tumors and extensive abdominal metastasis were found within 6 weeks (Fig. 2B). We confirmed the human origin of the local tumor and metastases by IHC staining using an antibody specific for human mitochondrion (Supplementary Fig. S1). Cells recovered from the primary tumor and the metastases were named EPT3-PT1 and EPT3-M1, respectively. Thereby, we have established a complete and stepwise model of prostate carcinogenesis, which includes prostate primary epithelial EP156T cells, mesenchymal nontransformed EPT1 cells, premalignant EPT2 cells, primary tumor derived EPT3-PT1, and metastasis-derived EPT3-M1 cells (Fig. 2C). DNA microsatellite analysis and copy number analysis verified their common genetic origin and a stepwise increase of DNA aberrations in these cells (Supplementary Table S1, Supplementary Data S1 and S2, and Supplementary Fig. S2). The phenotypes that were acquired step by step in the model from EP156T to EPT3-M1 cells are summarized (Fig. 2D).

To explore the mechanism of sphere formation, we profiled the gene expression of newly generated D5HS using DNA microarrays (Supplementary Data S3). A number of growth factors and cytokines were upregulated in D5HS (Fig. 3A and B). The top 1 increased (98-fold) growth factor gene IL-6 is particularly interesting, as it is a well-recognized marker of prostate cancer (20–22). The importance of its signaling pathway in D5HS spheres is further supported by the increase of the IL-6 receptor (IL-6R; Supplementary Data S3). Real-time qPCR verified the elevated transcription levels of both IL-6 and IL-6R (Fig. 3C). At the protein level, proteomic analysis of the secretome of the D5HS cells detected IL-6 and soluble IL-6 signal transducer IL-6ST (also known as gp130; Fig. 3D and Supplementary Table S2), indicating that the IL-6 signaling is indeed activated in D5HS. ELISA confirmed up to 200-fold higher IL-6 in the culture supernatant of D5HS than in EPT2-D5 and D5H cells (Fig. 3E).

To test whether elevation of IL-6 is important to sphere formation, a neutralizing antibody against IL-6 was used to treat sphere-forming D5HS cells. Significant decrease of the spheres in treated cells showed the requirement of IL-6 in D5HS generation (Fig. 4A). STAT3 has been reported important in IL-6-type signaling in prostate cancer (19–22). Activated STAT3 signaling in D5HS was verified by Western blotting using an antibody against phosphorylated (Y705) STAT3 (pSTAT3; Fig. 4B). Blockade of IL-6 reduced the pSTAT3 level, showing IL-6-dependent activation of STAT3 in D5HS (Fig. 4C). To track the STAT3 signaling activity, EPT2-D5 cells were stably transduced with the fluorescence (GFP)-based STAT3 pathway reporter pGF-STAT3. Very few cells expressing GFP showed low activities of STAT3 signaling in EPT2-D5 and D5H cells. A small but distinct subpopulation of GFP-positive cells was found when EPT2-D5 cells started to form spheres in protein-free medium (D5HS; Fig. 4D). Importantly, most of these positive cells were dividing and tending to grow spheres (Fig. 4E). In addition, some of the GFP positive cells divided into one positive and one negative cell (Fig. 4F), suggesting asymmetric division that is a hallmark of self-renewal of cancer stem cells (CSC; ref. 23). D5HS cells with top high (STAT3^high) and bottom low (STAT3^low) levels of STAT3 activities were isolated using flow cytometry and sphere formation in STAT3^high cells was much higher than in STAT3^low cells (Fig. 4G). STAT3 inhibitors significantly reduced the sphere formation in D5HS cells (Fig. 4H), showing the critical role of the STAT3 signaling in D5HS sphere formation. However, the IL-6 level in the supernatant was not significantly affected by these STAT3 inhibitors (Fig. 4I), suggesting that the activation of STAT3 is IL-6-dependent but the activation of IL-6 is not caused by STAT3 signaling activity.

It has been reported that IL-6/STAT3 signaling can be activated by ROS (24) and that ROS can be generated during serum deprivation (25). To test the possibility that the activation of IL-6/STAT3 in D5HS is due to ROS induced by growth factor withdrawal, endogenous ROS were detected using redox sensitive dye DCFDA. As presented in Fig. 5A, very few and weak signals were detected in EPT2-D5 cells growing in complete medium, whereas signals were significantly stronger (5-fold) when cells were adapted in protein free medium for 3 days (D5H; Fig. 5A and B). Strikingly, even higher (24-fold) ROS level was found in D5HS (Fig. 5A and B). Treatment of D5HS cells with anti-oxidants N-acetyl-l-cysteine (NAC) and melatonin significantly reduced DCFDA levels (Fig. 5C), showing that ROS indeed were induced in EPT2-D5 cells upon growth factor withdrawal. Similar to STAT3-positive cells, heterogeneous intensities of ROS were found among D5HS cells and most cells with higher ROS status tended to generate spheres (Fig. 5A). D5HS cells with top high and bottom low ROS levels were isolated and named ROS^high and ROS^low cells, respectively. Significantly, ROS^high cells generated numerous spheres, whereas ROS^low cells continuously grew in monolayer without sphere formation even at day 18 (Fig. 5D and E). Treatment of ROS^high cells with antioxidants NAC and melatonin efficiently decreased the sphere numbers (Fig. 5E), showing the indispensable role of ROS in D5HS formation. Examination of secreted IL-6 showed much higher IL-6 in ROS^high cells than in ROS^low cells (Fig. 5F), and the IL-6 level in ROS^high cells can be repressed by NAC and melatonin (Fig. 5F), indicating the requirement of ROS for IL-6 activation in D5HS sphere formation. Consistently, higher STAT3 phosphorylation was found in ROS^high cells than in ROS^low cells (Fig. 5G), and the level can also be reduced by antioxidant melatonin (Fig. 5G). All these data showed the requirement of ROS for spheres formation and IL-6/STAT3 activation.

It has been reported that ROS were induced in a few hours upon serum starvation and the level was relatively stable or lower afterwards (24). The much higher (4.8-fold) ROS level in D5HS than in D5H cells suggested an additional factor inducing ROS in D5HS (Fig. 5A and B). We tested whether it was due to the highly activated IL-6 considering that cytokines can stimulate intracellular ROS (26). As expected, the ROS level in D5HS was reduced (1.7-fold) by neutralizing antibody against IL-6 (Fig. 5H), and conversely, adapting EPT2-D5 cells in protein-free medium containing IL-6 protein at 2 ng/mL induced significant higher (2.1-fold) ROS level than control cells (Fig. 5I), indicating a positive feedback loop between ROS and IL-6 in D5HS formation. A ROS/IL-6/STAT3 regulation cascade in prostate tumor sphere formation is consequently proposed (Fig. 5J).

To evaluate the IL-6/STAT3 signaling activity in EPT3 tumor formation in vivo, the expression of IL-6 and STAT3, as well as a putative STAT3 target cyclin D1, were examined in the xenograft tumors by IHC staining. Clear staining of IL-6 with a pericellular and patchy intercellular pattern was detected in both primary and metastatic tumors (Fig. 6A). pSTAT3 was detected in the nuclei of a small proportion of prostate tumor cells (Fig. 6A). Total STAT3, on the other hand, stained the cytoplasm of a larger proportion of cells in addition to several nuclei (Fig. 6A). Nuclear staining of cyclin D was evident in a proportion of cells (Fig. 6A), all together indicating activation of IL-6/STAT3 signaling in the EPT3 primary tumor and metastasis.

Autocrine IL-6 and phosphorylation of STAT3 were also measured in EPT3 tumor–derived EPT3-PT1 and -M1 cells. The levels of secreted IL-6 in both cell lines were in the same range as for D5HS cells and more than 200-fold higher than in EPT2-D5 cells (Fig. 6B). The autocrine effect of IL-6 in EPT3 tumor–derived cells was extremely high considering that these cells were cultured in complete medium without the cell stress experienced by D5HS cells due to protein depletion. Consistently, ROS level in EPT3-M1 cells was also much higher (5-fold) than in EPT2-D5 cells (Fig. 6C and D). In addition, antioxidant melatonin efficiently blocked the production of IL-6 in EPT3-M1 cells (Fig. 6E), indicating that the high level of IL-6 in EPT3 tumors is due to high endogenous ROS, the same as in D5HS. The high level of IL-6/STAT3 signaling in EPT3 tumor cells is also evident compared with DU145 and PC3 cell lines that are widely used for studying the IL-6/STAT3 signaling in prostate cancer (27, 28). The IL-6 level in EPT3-M1 cells are 4- and 2.5-fold higher than in PC3 and DU145 cells, respectively (Fig. 6F). Western blotting also showed higher phosphorylation of STAT3 in EPT3-PT1 and EPT3-M1 cells (Fig. 6G).

To further test the role of IL-6/STAT3 activity in EPT3 tumor initiation, EPT3-M1 cells were transduced with the STAT3–GFP reporter and the top and bottom 5% of GFP expressed cells (STAT3^high and STAT3^low) were examined in mice. As few as 500 STAT3^high cells efficiently generated large tumors in 7 weeks, whereas many more STAT3^low cells were needed to form tumors and metastasis (Fig. 6H), showing the high tumorigenicity of STAT3^high cells.

In this work, we have generated prostate TICs via adapting premalignant EPT2 cells in protein-free medium. Chemically defined serum-free medium is widely used to culture tumor spheres or TICs (14–16). However, these media are often supplemented with growth factors such as EGF and fibroblast growth factor (FGF), and all cells are cultured in nonadherent conditions (14–16). In the current study, EPT2-D5 cells efficiently generated tumor spheres in medium completely free of serum or any other growth factor, suggesting an even higher independence of exogenous growth stimulation. In addition, generation of D5HS spheres from single adherent cells provides a unique model to study the cell division and metabolism of TICs.

Although ROS and cellular oxidant stress have long been associated with cancer, and elevated ROS level is a hallmark of many highly invasive cancers (29, 30), there is a hypothesis that keeping ROS levels low within stem cells or TICs is an important feature of “stemness” and offers protection against the cell toxicities of ROS (31, 32). Recent studies suggest that ROS not only are byproducts of cellular metabolism but also play roles as second messengers in cell signaling (33). In this work, high ROS level is indispensable to tumor sphere formation, autocrine activity of IL-6, and activation of STAT3 signaling, showing the central role of ROS as second messenger in tumor initiation. The different requirements of ROS in stem cells and D5HS cells may due to their different status: the relatively quiescent state of stem cells under normal circumstances is in contrast to the uncontrolled proliferation of tumor spheres with high ROS status advantageous for high proliferation. Similarly, high ROS supported proliferation, self-renewal, and neurogenesis of neural stem and progenitor cells (34), suggesting the context-dependent and complex function of ROS in stem cells and TICs. Actually, tumor spheres and TICs were usually maintained in serum-free medium in vitro (14–16). The absence of serum inevitably induces high ROS status in these cultures, which indicates that higher ROS level is common in tumor spheres and TICs.

IL-6 is elevated in the sera of patients with metastatic prostatic cancer and it is a widely recognized marker of prostate cancer (20–22, 35). Extensive studies have shown the central role of STAT3 in IL-6-type cytokine signaling in prostate cancer (27, 28, 36). Using a STAT3 reporter, we found that the IL-6-dependent STAT3-positive cells are enriched in tumor sphere–forming cells, and the top STAT3-positive cells isolated from xenograft tumors showed much stronger tumorigenicity than negative control cells in vivo, showing the essential role of IL-6/STAT3 activity in prostate tumor initiation. The central role of IL-6 in tumor initiation is also found in recent reports on prostate and breast cancer. Expression of IL-6 is much higher in prostate CSCs than non–stem cancer cells (NSCC) and NSCCs can be converted to CSCs by IL-6 treatment (37). Although these prostate CSCs were defined by cell surface markers CD44+/CD133+ (37), whereas our prostate TICs were defined by the activity of STAT3 signaling, both showed high dependence on IL-6. Activation of an IL-6 inflammatory loop is required for expanding the trastuzumab-resistant breast CSC population (38). However, it should be noted that other pathways or factors are also important to the D5HS generation and tumor initiation as not all the sphere-forming cells scored positive for STAT3 signaling (Fig. 4G) and chemical blockade of STAT3 signaling did not abolish all sphere formation (Fig. 4H), and finally, STAT3^low cells still generated tumors even though many more cells were needed (Fig. 6H). It will be interesting to evaluate the roles of other growth factors and cytokines that were also significantly increased in D5HS, such as bone morphogenetic protein (BMP), platelet-derived growth factor (PDGF), TGFBs, VEGFs, IL8, and IL33 (Fig. 3A and B).

As discussed above, there are abundant epidemiologic and clinical studies indicating the correlation of prostate cancer to ROS and IL-6/STAT3 signaling, respectively. In this study, for the first time, we linked an intrinsic association of ROS and IL-6/STAT3 in prostate carcinogenesis, which allows an improved therapeutic strategy regarding prostate cancer. Importantly, we observed a positive feedback loop between ROS and IL-6 during prostate tumor sphere formation. Induction of IL-6 by ROS was previously reported in many contexts (39–41). The mechanisms involved were transcriptional activation of the IL-6 gene through an NF-κB–dependent pathway (39, 40) or mitogen-activated protein kinase (MAPK) activity (41). On the other hand, IL-6 has been reported to induce ROS by activating the transcription and enzyme activity of spermine oxidase (SMO), which oxidizes spermine into spermidine, 3-aminopropanal, and H₂O₂ (42). Spermine can also be first acetylated by spermidine/spermine N1-acetyltransferase (SAT) and then oxidized by N1-acetylpolyamine oxidase (APAO), producing H₂O₂ as a byproduct (42). The expression of the SAT gene is significantly (5.5-fold) increased in D5HS (Supplementary Data S3), which can be a potential mechanism of the induction of ROS by IL-6 in D5HS.

No potential conflicts of interest were disclosed.

Conception and design: Y. Qu, A.M. Oyan, E. McCormack, K.-H. Kalland, X.-S. Ke

Development of methodology: Y. Qu, A.M. Oyan, M. Popa, W. Zhang, E. McCormack, K.-H. Kalland, X.-S. Ke

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.M. Oyan, R. Liu, Y. Hua, J. Zhang, R. Hovland, K.A. Brokstad, R. Simon, A. Molven, W. Zhang, E. McCormack, K.-H. Kalland, X.-S. Ke

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Qu, A.M. Oyan, R. Hovland, M. Popa, A. Molven, B. Lin, E. McCormack, K.-H. Kalland, X.-S. Ke

Writing, review, and/or revision of the manuscript: Y. Qu, A.M. Oyan, R. Hovland, K.A. Brokstad, A. Molven, B. Lin, W. Zhang, E. McCormack, K.-H. Kalland, X.-S. Ke

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Popa, X. Liu, K.A. Brokstad, W. Zhang, K.-H. Kalland, X.-S. Ke

Study supervision: E. McCormack, K.-H. Kalland, X.-S. Ke

The authors thank Dr. Yong-Joon Chwae for providing the STAT3 reporter pGF1-STAT3; Beth Johannessen for cell culture work; Hua My Hoang for DNA microarray profiling; and Anne Aarsand, Marianne Eidsheim, Kjerstin Jakobsen, and Dagny Ann Sandnes for tissue embedding, sectioning, and immunohistochemistry. They also thank Marianne Enger for flow cytometry and cell sorting, Solrun Steine for DNA microsatellite analyses, and Kjetil Solland and Atle Brendehaug for chromosome analyses and the Molecular Imaging Centre (MIC) in University of Bergen for advanced microscopy. They thank Dr. Huarong Zhang and Yingying Ma for help in the mass spectrometry analysis and Kjell Petersen at the Elixir.no bioinformatics helpdesk, funded by the Research Council of Norway through the Large Infrastructure program, for microarray data management and export to the public data repository Array Express.

This work was supported by grants from Helse Vest (911626, 911555, 911747, 911582), the Bergen Medical Research Foundation, Bergen Research Foundation, the Norwegian Cancer Society, Chinese NSFC (81230090), MOST of China (2012AA022705), EU FP7- PEOPLE-IRSES-2008 (TCMCANCER Project 230232 and PPI-MARKER 247097).