Transforming Growth Factor β1 Signal is Crucial for Dedifferentiation of Cancer Cells to Cancer Stem Cells in Osteosarcoma§

Authors

  • Haixia Zhang,

    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    Search for more papers by this author
  • Haotong Wu,

    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    Search for more papers by this author
  • Junheng Zheng,

    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    Search for more papers by this author
  • Pei Yu,

    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    Search for more papers by this author
  • Lixiao Xu,

    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    Search for more papers by this author
  • Pan Jiang,

    1. Department of Oral and Maxillofacial Surgery, Guanghua School of Stomotology, Hospital of Stomotology, Sun Yat-sen University, People's Republic of China
    Search for more papers by this author
  • Jin Gao,

    1. School of Medicine and Discipline of Dentistry, James Cook University, Cairns, Queensland, Australia
    Search for more papers by this author
  • Hua Wang,

    1. Department of Oral and Maxillofacial Surgery, Guanghua School of Stomotology, Hospital of Stomotology, Sun Yat-sen University, People's Republic of China
    Search for more papers by this author
  • Yan Zhang

    Corresponding author
    1. Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, People's Republic of China
    • Key Laboratory of Gene Engineering of the Ministry of Education, State Key Laboratory of Biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou Higher Education Mega Center North, Guangzhou 510006, People's Republic of China
    Search for more papers by this author
    • Telephone: 86-2039332955; Fax: 86-2039332955


  • Author contributions: H.Z.: conception and design, data analysis and interpretation, manuscript writing, and final approval of manuscript; H.W.: collection and assembly of data and data analysis; J.Z.: collection and/or assembly of data, interpretation; P.Y., L.X., and P.J.: collection and/or assembly of data; J.G.: provision of study materials and final approval of manuscript; H.W.: collection and/or assembly of data and final approval of manuscript; Y.Z.: conception and design, financial support, provision of study materials, manuscript writing, and final approval of manuscript.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLSEXPRESS November 15, 2012.

Abstract

Human osteosarcoma harbors a small subpopulation of cancer stem cells (CSCs) that is believed to be associated with tumor metastasis, radioresistance/chemoresistance, local invasion, and poor clinical outcome. In this study, we found that transforming growth factor β1 (TGF-β1) signaling and a hypoxic environment dramatically induced self-renewal capacity in non-stem osteosarcoma cells, which in turn promoted chemoresistance, tumorigenicity, neovasculogenesis, and metastatic potential. Furthermore, blocking the TGF-β1 signaling pathway resulted in the inhibition of the dedifferentiation and clonogenicity of osteosarcoma cells, and the reduction of CSC self-renewal capacity and hypoxia-mediated dedifferentiation. These findings demonstrate that stem cells and non-stem cells exist in a dynamic equilibrium within the osteosarcoma cell population, and that CSCs may develop de novo from differentiated cancer cells. Hierarchical models of mammalian CSCs, therefore, should be considered to serve as bidirectional interconversion between the stem and non-stem cell components of the tumor. STEM CELLS2013;31:433–446

INTRODUCTION

Osteosarcoma is the second highest cause of cancer-related death in children and adolescents. Despite advanced multimodality and multiagent cancer therapy, long-term survival rates for high-grade osteosarcomas remain low, due to pulmonary metastasis via the hematogenous route [1]. Thus, an appropriate therapy is urgently required as the top prioritized treatment strategy. Therefore, it is most important to understand the molecular and cellular mechanism of the development of osteosarcoma.

Osteosarcomas are not only diverse in terms of their pathological nature and clinical distribution but also in the aspect of the cellular heterogeneity within the tumors. Osteosarcoma most commonly develops around the knee joint, distal femur, and proximal tibia; the vast majority of the patients have already shown micrometastasis at the time of diagnosis [2]. It is believed to originate from bone mesenchymal cells or osteoprogenitors [3]. The etiology of osteosarcoma is currently rather limited, while the pathogenesis remains unknown. The unique properties of osteosarcoma may be related to either the cell of origin or components in the bone marrow microenvironment, such as the large amount of transforming growth factor β1 (TGF-β1) [4] and low oxygen tension [5].

Once an osteosarcoma germinates in the bone, tumor cells secrete factors that initiate osteoclast-mediated bone destruction, and matrix-derived growth factors, especially TGF-β1, are released from bone matrix. At this time, osteosarcoma cells also release TGF-β1 directly [6]. TGF-β1 is a pleiotropic cytokine that acts as a mediator upon the tumor to promote further tumor expansion, metastasis, and cytokine production [7]. Recent findings suggest that genetic and epigenetic events mediate the acquisition of oncogenic activity by TGF-β1, as do the aberrant alterations within the tumor microenvironments. These events coalesce to enable TGF-β to directly play a role in the metastatic progression via stimulating the epithelial–mesenchymal transition (EMT), which permits carcinoma cells to abandon polarized epithelial phenotypes in favor of apolar mesenchymal-like phenotypes [8, 9]. In addition to its ability to enhance carcinoma cell invasion and metastasis, EMT also endows transitioned cells with stem cell-like properties, including the acquisition of self-renewal and tumor-initiating capability coupled to chemoresistance [10–12].

Hypoxia is a common condition in solid tumors, particularly in those that grow rapidly [13, 14]. The reduced cell division detected in areas of low oxygenation in tumors results in resistance to both radiotherapy and chemotherapy [15, 16]. Cellular responses to hypoxia are commonly regulated by the hypoxia inducible factor (HIF) family of transcriptional factors. HIFs function as heterodimers which consist of an oxygen-sensitive HIF-α subunit and a HIF-β subunit. The bone marrow niche is relatively hypoxic in comparison with other tissues (≤2% O2) [17]. This, in conjunction with a rather high proliferate capacity of resident cells, results in the hypoxia observed in osteosarcoma. An increased level of tumor hypoxia is often associated with poor prognosis, while hyperbaric oxygen has been shown to enhance the efficacy of chemotherapy for osteosarcoma [18].

Recent studies in several types of human cancers have suggested that tumors are organized in a hierarchy of heterogeneous cell populations. Cancer stem cells (CSCs) are a subpopulation of cancer cells that share the characteristics of self-renewal ability and multipotency with stem cells [19]. CSCs could generate daughter cells with relative limited replicative ability and thus contribute to tumor. The presence of CSCs is associated with radioresistance/chemoresistance, local invasion, and poor prognosis [20–25]. Previous studies confirmed the presence of osteosarcoma stem cells (OSCs) by the sphere forming assay or cell sorting according to cell surface markers (CD133+, CD117+/Stro-1+) [26–28]. OSCs were capable of initiating tumors and had adipogenic lineage potential in vitro.

In this study, we demonstrate that CSCs and cancer non-stem cells are in a dynamic equilibrium in the total cell populations in osteosarcoma. The microenvironment and intracellular context of osteosarcoma cells determines the direction of dynamic equilibrium movement. TGF-β1 and hypoxia are two important elements that induce osteosarcoma cells toward CSC phenotypes.

MATERIALS AND METHODS

Cell Culture

Osteosarcoma cell lines MNNG/HOS and MG63 were obtained from the cell bank of the Chinese Academy of Sciences (Shanghai, China, http://www.cellbank.org.cn), Saos-2 cell line was purchased from ATCC (Manassas, VA, http://www.atcc.org). The cells were confirmed to be free from bacteria and mycoplasma contamination or crosscontamination with other kinds of cell. The adherent cells were cultured in Dulbecco's modified Eagle's medium/F12 (DF; 1:1 mixture; Sigma, Missouri http://www.sigmaaldrich.com/united-states.html) supplemented with 5% FBS (Hyclone, Logan, UT, http://www.hyclone.com), 90 μg/ml ampicillin, and 90 μg/ml kanamycin. The sarcospheres were cultured in serum-free DF medium supplemented with 10 μg/ml human insulin, 5 μg/ml human transferrin, 10 μM 2-aminoethanol, 10 nM sodium selenite, and 10 μM mercaptoethanol.

Cell Cycle Analysis

The cells were fixed in 70% ethanol at −20°C and then treated with 100 μg/ml RNase A. After staining with 50 μg/ml propidium iodide, the cells were analyzed in a flow cytometry (FACScalibur; BD Biosciences, NJ, http://www.bdbiosciences.com). Data were collected from 10,000 single cell events, and cell cycle phase distributions were calculated using MODFIT software (Verity Software House, http://www.vsh.com).

Colony Forming Assay

Three hundred single living adherent cells or iOSCs cells were immediately plated onto 0.66% solidified agar-based six-well plate. Soft agar cultures were incubated for 10 days and fed with 1 ml of the assay medium twice a week. The presented colonies were fixed with 4% paraformaldehyde for 20 minutes at room temperature, and stained with crystal violet.

Chemosensitivity Assessment

The chemosensitivity to cisplatin and adriamycin were tested using CellTiter 96 AQueous One Solution Cell Proliferation Assay kit (Promega, Madison, WI, http://www.promega.com). 5 × 103 cells per well were seeded onto 96-well plates, and drugs were added for 72 hours, the absorbance at 490 nm was measured using a Microplate Reader 680 (Bio-Rad, Hercules, CA, http://www.bio-rad.com).

In Vitro Differentiation

The cells were cultured in adipogenic differentiation medium (Invitrogen, Carlsbad, CA, http://www.invitrogen.com) for 9 days. Mature adipocytes were identified by Sudan III staining. For vascular endothelial cells differentiation, the cells mixed with Matrigel (R&D, Minneapolis, MN, http://www.rndsystems.com) were cultured in differentiation media for 3 weeks.

Evaluation of Tumorigenicity

Following the approval of the research protocol by Sun Yat-sen University Animal Ethics Committee, mice were purchased from National Resource Center for Rodent Laboratory Animal and maintained in accordance with the institutional guidelines. The single cells were suspended in 100 μl phosphate buffered saline, and i.m. injected into the leg muscle of 5-week-old athymic nude mice (BALB/c). Mice were monitored every 2 days until 7 weeks at the end point. Data were collected from four or six independent experiments. Tumor nodules >1 mm in diameter were taken for pathologic examination. When tumor nodules were identified at secondary sites apart from a primary injection site and diagnosed as osteosarcoma, the nodules were determined as metastases.

Histological Analysis and Immunostaining

The cells or tissues were fixed in 10% buffered formalin for 24 hours and embedded in paraffin. Paraffin-embedded sample sections (4-μm-thick) were deparaffinized, rehydrated, and stained with H&E. For immunostaining, the cells were incubated with primary antibodies (Supporting Information Table S3) overnight at 4°C, and then were incubated with secondary antibodies for 1 hour at room temperature. Nuclei were stained with 1 μg/ml 4,6-diamino-2-phenyl indole (DAPI) (Invitrogen). Stained cells were observed with a confocal microscope (LSM510, Carl Zeiss, Oberkochen, Germany, http://www.zeiss.com/explore). Images were analyzed with Leica Microsystems CMS GmbH software (http://las.software.informer.com).

EdU Retention Assay

Dissociated cells were stained with EdU using Cell-Light 5-ethynyl-2′-deoxyuridine (EdU) DNA Cell Proliferation Kit (RiboBio, Guangzhou, China, http://www.sirna.cn) and visualized under a confocal microscope (LSM510, Carl Zeiss). Images were analyzed with Leica Microsystems CMS GmbH software.

RNA Isolation, Semiquantitative RT-PCR and Real Time RT-PCR

Total RNA was isolated using the Trizol reagent (Invitrogen) and was treated with RNAase-Free DNase (Qiagen Sciences, Valencia, CA, http://www.qiagen.com/default.aspx) to remove genomic DNA contamination. First-strand cDNA was made using total RNA (0.5 μg), Oligo(dT)20 primers, and ReverTraAce-a Reverse Transcriptase (Toyobo, Osaka, Japan, http://www.toyobo.cn), according to the manufacturer's recommendations. The target cDNA was amplified using rTaq (Takara, Shiga, Japan, http://www.clontech.com/takara) or KOD (Toyobo, Osaka, Japan) for 25–30 cycles. The primers are provided in Supporting Information Table S4. For real-time RT-PCR, a 10 μl of real-time PCR reaction containing Lightcycler 480 SYBR Green I Master (Roche, Basel, Switzerland, http://www.roche.com) was prepared according to the manufacturer's instructions. Real-time quantitation was performed using LightCycler 480 (Roche). The fluorescence threshold value was calculated using Lightcycler 480 series software. The calculation of relative change in mRNA was performed using the melting curve analysis, with normalization using the housekeeping gene GAPDH.

Western Blot

Cells were lysed on ice for 30 minutes in radioimmuno-precipitation assay buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% NP-40, 0.1% SDS) containing a protease inhibitor cocktail (Roche). Cellular proteins were determined using the Protein Quantitative Analysis kit (K3000-BCA Shenergy Biocolor, Shanghai, China, http://www.biocolors.com.cn/home.asp). Samples were separated in 10% SDS gels, transferred to polyvinylidene fluoride Immobilon-P membranes (GE Healthcare, Buckinghamshire, U.K., http://www3.gehealthcare.cn/zh-CN), and blocked using Tris-buffered saline with Tween-20 containing 5% nonfat dried milk. After incubation with primary antibodies, the membranes were incubated with corresponding horseradish peroxidase-conjugated secondary antibodies for 1 hour, and developed with SuperSigna®west Pico chemiluminescent Substrate (Pierce, Rockford, IL, http://www.piercenet.com).

DNA Microarray

Isolated total RNA (5 μg) was submitted for microarray analysis using the Human Expr 12 × 135K Arr Del (Roche-NimbleGen, http://www.nimblegen.com) containing 135K probes in triplicate. Clustering analysis was performed by Cluster 3.0. ANOVA statistical modeling was used to categorize differentially expressed genes. The microarray data was deposited at http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GSE38135. In general, all genes regulated at p < 0.05 were considered statistically significant and were categorized by >2 SDs from the mean. Biological process analysis by GO classification was performed on this gene set using these criteria, and biological process regulated as a whole (p < 0.05) was considered significant.

GSEA

From GEO, we downloaded three osteosarcoma-related microarray datasets GSE33382, GSE14827, and GSE19276 according to clinicopathological characteristics. We designated gene sets (Supporting Information Table S5) in the base of our microarray analyses (GSE38135), which were used for Gene Set Enrichment Assay (GSEA) analyses. False discovery rate (FDR) q-value is used to account for multiple comparisons. The FDR < 0.25 was considered significant.

Statistical Analyses

Each experiment was performed independently at least triplicate. Values were expressed as mean ± SD. Statistical significance (p < 0.05) was determined using Student's t test. Pearson product-moment correlation coefficient (Pearson's correlation coefficient, R) was used to measure the relationships in three groups of sarcosphere: iOSCs and OSCs, hypoxia-induced spheres and OSCs, or iOSCs and hypoxia-induced spheres. Only Pearson's correlation coefficients with an accompanying p-value of < 0.01 were considered statistically significant.

RESULTS

Osteosarcoma Cells Contain a Subpopulation with Sarcosphere Forming Capacity

As CSCs have the ability to form sarcospheres, a sphere-forming assay was undertaken to enrich a subpopulation showing stem cell characteristics. When human osteosarcoma cell line MNNG/HOS cells were cultured in DF medium supplemented with 5% fetal bovine serum (FBS), a monolayer of polygonal and spindle-shaped cells formed (Fig. 1A). When the cells were transferred to serum-free DF medium, a small portion of this monolayer of cells gradually changed to morphological sarcospheres as the bulk cells reached 60%–70% confluence. The spontaneously formed sarcospheres (OSCs) could be separated from the adherent cells and suspended in the culture medium (Fig. 1B). Immunofluorescent (IF) analyses showed that the spheroid cells were enriched in the expression of CD117+/Stro-1+ when compared to the adherent cells (Fig. 1C). The maximal diameter of OSCs cultured for 5 days ranged from 100 μm to 150 μm in a cell cluster. The biggest OSCs comprised approximately 1,000 cells and could be passaged continuously on low adhesion plates.

Figure 1.

Morphology of MNNG cells in different culture conditions. (A): Representative image of MNNG bulk cells cultured in DF medium supplemented with 5% fetal bovine serum (FBS). (B): MNNG bulk cells cultured in serum-free DF medium for 5 days. (C): Immunocytochemical study of Stro-1 and CD117 expression counterstained with DAPI (blue) in MNNG bulk cells (left) and sarcospheres (right). Scale bar = 10 μm. (D): Scheme of experimental procedure. (E): MNNG residual adherent cells (after sarcosphere removing) suspended cultured in serum-free DF medium. Hochest 33342 staining indicates nuclei (inset). (F): MNNG residual adherent cells cultured in DF medium supplemented with 5% FBS. (G): MNNG residual adherent cells monolayer cultured in serum-free DF medium. (H): MNNG bulk cells cultured in serum-free DF medium supplemented with 4 ng/ml TGF-β1. (I): MNNG residual adherent cells cultured in serum-free DF medium supplemented with 4 ng/ml TGF-β1. (J): MNNG bulk cells cultured in serum-free DF medium. (K): MNNG residual adherent cells cultured in serum-free DF medium; scale bar = 100 μm. (L): Statistical analysis of the sarcosphere numbers during dedifferentiation process. The sarcosphere number was counted under ×4 microscope (**, p < .01; *, p < .05). Abbreviations: DAPI, 4,6-diamino-2-phenyl indole; TGF-β1, transforming growth factor β1.

When the residual adherent cells were suspended under the same culture conditions (Fig. 1D), cells did not form sarcospheres but underwent apoptosis (Fig. 1E). When the residual adherent cells of the monolayer were cultured in DF medium containing 5% FBS, these cells grew quickly and never form spheres (Fig. 1F). However, when these adherent cells were monolayer cultured in serum-free DF medium, sarcospheres reappeared and once again resembled the original bulk cell culture (Fig. 1G). We then treated the monolayer bulk cells and residual adherent cells with TGF-β1 to mimic the bone marrow microenvironment. We found that most of the monolayer bulk cells (Fig. 1H) and residual adherent cells (Fig. 1I) changed to sarcospheres within 5 days. The sarcospheres derived from TGF-β1-stimulated residual adherent cells were designated iOSCs. Compared with untreated monolayer bulk cells (Fig. 1J) and residual adherent cells (Fig. 1K), TGF-β1-treated cells formed sarcospheres quickly, and the number of induced sarcospheres was much greater than the control group (Fig. 1L, p ≤ 0.05). In addition, TGF-β1 had no effect on the proliferation rate of osteosarcoma cells (Supporting Information Fig. S1); the size of iOSCs was similar to that of OSCs. We speculated that TGF-β1 could be involved in the process of sarcosphere formation.

TGF-β1 Signaling and a Hypoxic Microenvironment Were Crucial for Maintaining an Immature State in Osteosarcoma

Since we found that exogenous TGF-β1 had no effect on proliferation and morphology of pre-existing OSCs or iOSCs (Fig. 2A), we speculated that TGF-β1 was secreted by this type of cell to maintain the sarcosphere potential. TGF-β1 assembles a transmembrane receptor serine/threonine kinase (TβR2), which induces transphosphorylation and the activation of the type I receptor (TβR1). Once the TβR1 inhibitor, SB431542, was added to pretreated cells, adherent cells formed very few sarcospheres, even in the presence of TGF-β1 (Fig. 2B). Most iOSCs were adherent to the culture dish bottom and began to differentiate after addition of SB431542 in a dose-dependent manner (Fig. 2C). Moreover, SB431542 had no effect on cell survival or proliferation of adherent cells (Supporting Information Fig. S2). These results suggest that the TGF-β1 signaling pathway is essential to maintain osteosarcoma cells in an immature or undifferentiated state.

Figure 2.

TGF-β1 signal was crucial for maintaining an immature state in osteosarcoma. (A): TGF-β1 had no effect on the proliferation and morphology of iOSCs or OSCs. (B): The adherent cells were pretreated with 0.01% DMSO (as control) or indicated concentrations of SB431542 for 3 hours and then incubated with 4 ng/ml TGF-β1 for 5 days in serum-free DF medium. (C): iOSCs were treated with 0.01% DMSO (as control), 47 nM SB431542, 94 nM SB431542, 188 nM SB431542, or 376 nM SB431542 for 5 days, respectively, in serum-free medium. (D): Representative images of adherent cells cultured in serum-free DF medium or serum-free DF medium supplemented with 2 μM CoCl2 for 5 days. (E): Adherent cells were pretreated with 0.01% DMSO or 188 nM SB4315425 for 3 hours and then incubated with 2 μM CoCl2 for 5 days. Scale bar = 100 μm. (F): Sarcosphere numbers were counted from three separate experiments (*, p < .05). (G): Real time reverse transcription polymerase chain reaction (RT-PCR) analysis of TGF-β1 expression during CoCl2 induced dedifferentiation (*, p < .05). Total RNA was isolated from adherent cells cultured with 2 μM CoCl2, and then RT-PCR was performed. Primers used in this study are listed in Supporting Information Table S4. (H): Western blot analysis of hypoxia inducible factor 1α (HIF1α) expression. MNNG adherent cells were pretreated with 0.01% DMSO or 376 nM SB4315425 for 3 hours and then cultured with 4 ng/ml TGF-β1 or 2 μM CoCl2 in serum-free DF medium for 5 days. (I): Saos-2 adherent cells were pretreated with 0.01% DMSO (as control) or 188 nM SB431542 for 3 hours and then incubated with 4 ng/ml TGF-β1 or 2 μM CoCl2 for 5 days in serum-free DF medium. Scale bar = 100 μm. (J): Western blot analysis of HIF1α expression of Saos-2 cells in (I). (K): MG-63 adherent cells were pretreated with 0.01% DMSO (as control) or 376 nM SB431542 for 3 hours and then incubated with 4 ng/ml TGF-β1 or 20 μM CoCl2 for 5 days in serum-free DF medium. Scale bar = 100 μm. (L): Western blot analysis of HIF-1α expression of MG-63 cells in (K). All images obtained by Western blot analysis were captured with ImageQuant RT (GE Healthcare, Buckinghamshire, U.K., http://www.gehealthcare.com/), and densitometry analysis was performed using ImageQuant TL 7.0 Image Analysis Software (GE Healthcare). Only samples run in the same gel were compared. Abbreviations: DMSO, dimethyl sulfoxide; HIF-1α, hypoxia inducible factor; OSCs, osteosarcoma stem cells; TGF-β1, transforming growth factor β1.

It has been well documented that low oxygen tension is necessary to maintain pluripotency without compromising cell proliferative ability. Therefore, we evaluated the impact of hypoxia on osteosarcoma cells. To mimic the hypoxic bone marrow environment, cobalt chloride (CoCl2) [29] was added to treat the adherent cells. Most of the adherent cells were observed to form sarcospheres during a 5-day culture (Fig. 2D). Interestingly, following the treatment of cells with SB431542 prior to CoCl2 administration in the cell culture, the number of sarcospheres was significantly reduced (Fig. 2E, 2F), while the expression levels of TGF-β1 in sarcospheres was increased when CoCl2 was added (Fig. 2G). These results suggest that the TGF-β1 signal pathway is essential for hypoxia inducing the dedifferentiation of osteosarcoma cells. To understand the mechanism of this phenomenon, we further examined the expression of HIF-1α and HIF-2α after various treatments. The increased expression of HIF-1α protein mediated by either TGF-β1 or CoCl2 was abrogated by the addition of SB431542, indicating that HIF-1α expression was likely to require the involvement of TβR1 kinase activity in the formation of CSCs within osteosarcoma (Fig. 2H). In addition, we were unable to detect the expression of HIF-2α in MNNG cells (data not shown).

To confirm our results, we next analyzed the effect of TGF-β1 or CoCl2 on Saos-2 and MG-63 cells. When Saos-2 cells were transferred to serum-free DF medium, a small portion of the monolayer cells gradually formed sarcospheres. MG-63 monolayer cells could not form sarcospheres in serum-free DF medium. TGF-β1 or CoCl2 observably promoted the process of sarcosphere formation in Saos-2 cells (Fig. 2I) and MG-63 cells (Fig. 2K), whereas, the dedifferentiation process was obviously blocked by SB431542. Furthermore, the expression of HIF-1α mediated by TGF-β1 or CoCl2 was inhibited by SB431542 (Fig. 2J, 2L).

Induced Spheroid Cells Displayed Characteristics of Stem Cells

MNNG spheroid cells showed a higher nuclear:cytoplasm ratio in comparison with adherent cells (Fig. 3A). As such, OSCs and iOSCs expressed human embryonic stem cell (ESC)-specific surface antigens, including sex determining region Y-box 2 (Sox2), stage-specific embryonic antigen 1 (SSEA-1), stage-specific embryonic antigen 4 (SSEA-4), tumor rejection antigen 1-60 (TRA-1-60), and tumor rejection antigen 1-81 (TRA-1-81) at much higher levels (Fig. 3A). However, both adherent cells and spheroid cells expressed human ESC markers Nanog and Oct4 at very low levels. When 5% FBS or SB431542 was added to the culture medium, the expression of ESC markers in the attached iOSCs was decreased sharply (Fig. 3B). Quantitative reverse transcription polymerase chain reaction (qRT-PCR) analyses revealed that the expression of Sox2 from immature or undifferentiated cells was clearly induced, whereas, the expression of Nanog and Oct4 was marginally induced during the process of TGF-β1-initiated sarcosphere formation (Fig. 3C). Furthermore, IF analyses showed that iOSCs were CD117+/Stro-1+ (Fig. 3D). In addition, cell cycle analyses showed that 28.495% ± 0.335% of iOSCs derived cells were in the mitotic division phase, while 71.5% ± 0.33% of iOSCs derived cells were in G0/G1 phase. In total, 65.32% ± 1.12% of the adherent cells were in mitosis, and 34.68% ± 1.12% were in G0/G1 phase (Fig. 3E). We also stained the cells with Ki67 and EdU to confirm cell status. In the adherent cells, 46% of the cells were EdU positive, and most of the cells expressed Ki67. In contrast, 53% of iOSCs derived cells did not express Ki67, and the rest of the spheroid cells were in mitosis (Fig. 3F). We then examined chemosensitivity to adriamycin or cisplatin in cultures of OSCs, iOSCs, and adherent cells. The spheroid cells showed more resistance to adriamycin and cisplatin than adherent cells (Fig. 3G). The qRT-PCR analysis also showed that the expression of the ATP-binding cassette, subfamily A1 (ABCA1) and ATP-binding cassette, subfamily G1 (ABCG1) was higher in spheroid cells than that in adherent cells (Fig. 3H). However, pretreatment of SB431542 resulted in increased chemosensitivity (Fig. 3I) and significant downregulation of the expression of ABCA1 and ABCG1 (Fig. 3J) in spheroid cells.

Figure 3.

Spheroid cells displayed characteristics of stem cells. (A): Immunofluorescent (IF) analysis of Sox2, Oct4, Nanog, SSEA-1, SSEA-4, TRA-1-60, and TRA-1-81 expression counterstained with DAPI (blue). Scale bar = 20 μm. H&E staining and phase contrast of adherent cells and spheres are also shown. The voltage used in this study was 830 V, at this point, just blue fluorescence of DAPI could be detected in the negative control for each antibody. (B): Sarcospheres were cultured with 5% fetal bovine serum or 376 nM SB431542 for 3 days and then were analyzed by IF. Scale bar = 20 μm. (C): Quantitative real time polymerase chain reaction (qRT-PCR) analysis of Nanog, Oct4, and Sox2 expression during dedifferentiation (**, p < .01; *, p < .05). (D): IF analysis of Stro-1 and CD117 expression in iOSCs counterstained with DAPI (blue). Scale bar = 10 μm. (E): Cell cycle analysis of spheroid cells and adherent cells. (F): IF analysis of Ki-67 and EdU in spheroid cells and adherent cells (scale bar = 20 μm). Nuclei were stained with DAPI (blue). Positive cells were counted in five microscopic fields, and results from three separate experiments are averaged. (G): The cells were treated with increasing concentrations of adriamycin (left) or cisplatin (right) for 72 hours, and then cell viability were calculated. (H): qRT-PCR analysis of ABCA1 and ABCG1 in spheres and adherent cells. Results from three separate experiments are averaged (*, p < .01). (I): OSCs were pretreated with 0.01% DMSO (as control) or 376 nM SB431542 for 3 days. Then the cells were treated with adriamycin (left) or cisplatin (right) for 72 hours. (J): qRT-PCR analysis of the expression of ABCA1 and ABCG1 in OSCs pretreated with 0.01% DMSO or 384 nM SB431542. Results from three separate experiments are averaged (*, p < .01). All iOSCs/OSCs used in Figure 3 were obtained at day 5 of TGF-β1 treatment/serum-free culture and subsequently subjected to further assay. Abbreviations: ABCA1, ATP-binding cassette, sub-family A1; ABCG1, ATP-binding cassette, sub-family G1; DAPI, 4,6-diamino-2-phenyl indole; DMSO, dimethyl sulfoxide; EdU, 5-ethynyl-2−-deoxyuridine; OSCs, osteosarcoma stem cells; Oct4, octamer-binding transcription factor-4; sox2, sex determining region Y-box 2; SSEA, stage-specific embryonic antigen 1; TRA, tumor rejection antigen.

Spheroid Cells Increased Anchorage-Independent Growth and Tumorigenicity

To test colony forming efficiency in vitro, spheroid cells and adherent cells were placed into soft agar. Over a 10-day culture period, 33% of iOSCs derived cells formed colonies, sized from 80 to 200 μm in diameter. In comparison, less than 1.5% of the adherent cells formed small-sized colonies around 50 μm in diameter (Fig. 4A). However, following the treatment of spheroid cells with SB431542, only 15% of the spheroid cells formed colonies (Fig. 4B). Thus, the clonogenicity of spheroid cells may be significantly inhibited via the blockage of the TGF-β1 signaling pathway. To evaluate tumorigenicity, spheroid cells and adherent cells were intramuscularly transplanted into BALB/C athymic mice. As few as 1 × 103 iOSCs derived cells or 1 × 104 OSCs derived cells were required to confer tumor formation. Thus, all mice inoculated with 1 × 104 iOSCs derived cells developed tumors and lung metastases were detected in these mice. In contrast, 1 × 106 bulk cells or 1 × 106 residual adherent cells were required to initiate tumors, which did not lead to distant metastasis. Interestingly, all of the mice injected with 1 × 105 hypoxia-induced spheroid cells developed tumors associated with lung metastases in two-third of these mice. These results suggest that the spheroid cells have a strong tumor-initiating capacity, TGF-β1 signaling and a hypoxic environment promote the tumorigenecity and metastatic potential (Fig. 4C, 4D).

Figure 4.

The tumorigenic potential of adherent cells and spheroid cells. (A): Colony formation capacity of adherent cells versus iOSCs derived cells in serum-free DF medium (left). Analysis of colony-forming ratio from five separate experiments is shown (right; *, p < .05). (B): Colony formation capacity of spheroid cells in serum-free DF medium supplemented with 376 nM SB431542 or 0.01% DMSO (left). Analysis of colony-forming ratio from five separate experiments is shown (right; *, p < .05). (C): H&E staining of primary tumor initiated by 1 × 106 monolayer bulk cells, 1 × 106 residual adherent cells, 1 × 104 OSCs derived cells, 1 × 103 iOSCs derived cells, or by 1 × 105 hypoxia-induced spheroid cells. Metastasis was detected in spheroid cell injected mouse. Scale bar = 100 μm. (D): In vivo tumorigenicity of osteosarcoma cells. Spheroid cells (1 × 103, 1 × 104, or 1 × 105), bulk cells (1 × 104, 1 × 105, or 1 × 106), or adherent cells (1 × 104, 1 × 105, or 1 × 106) were transplanted into BALB/C athymic mice, tumor formation and metastasis was analyzed 7 weeks postinjection. White oblique dotted line represents metastasis and the number above bar of white dotted lines represents the percentage of metastasis. All sarcospheres used in Figure 4 were obtained at day 5 of serum-free culture of NA/TGF-β1/CoCl2 treatment and subsequently subjected to further assay. Abbreviations: DMSO, dimethyl sulfoxide; OSCs, osteosarcoma cells.

Spheroid Cells Had Endothelial and Adipogenic Lineage Potential

At 6 days postseeding of cells onto Matrigel, some vessel-like sprouts appeared around the outermost region of the iOSCs, followed by the appearance of numerous vascular-like branches. After 18 days in culture, long tentacle-like structures extended out from the spheres, and they extended gradually to create more branches. Meanwhile, a large number of adherent tumor cells exited and spread rapidly along the branches. Up to day 21, the tentacle-like structures formed a large web-net that resembled a capillary network (Fig. 5A). To induce the differentiation pathway, 5 ng/ml of vascular endothelial growth factor (VEGF)-A was added to the 3D cultures. Vascular sprouts were observed by day 4 following the addition of VEGF-A. The transdifferentiation of cells into endothelial-like cells was observed in both iOSCs (Fig. 5B) and OSCs (Fig. 5C) cultures by 21 days. Hypoxia-induced sarcospheres also demonstrated transdifferentiation into endothelial-like cells in 3D Matrigel. Compared with normal oxygen tension (Fig. 5D), both the hypoxic microenvironment (Fig. 5E) and the presence of VEGF-A (Fig. 5F) promoted hypoxia-induced sarcospheres forming vascular-like networks. These endothelial-like cells not only formed tubular structures but also expressed CD31 (Fig. 5G). During capillary network formation, a few spheroid cells and the vast majority of tubular structures under differentiation expressed CD31 (Fig. 5H). However, CD31 was unable to be detected in the spheres without induction (Fig. 5I). In the case of adherent cells, a number of single cells proliferated to form small clones, in which small vascular sprout-like structures were observed but did not extend further. However, the structures showed signs of atrophy and apoptosis at week 3 of culture (Supporting Information Fig. S3). OSCs (Fig. 5J), iOSCs (Fig. 5K), and hypoxia-induced sarcospheres (Fig. 5L) pretreated with SB431542 could not form tentacle-like structures and gradually entered apoptosis by week 3 of culture. In the mouse model, CD31 positive staining (CD31+) was detected in vascular vessels of xenografted tumors originated from iOSCs, OSCs as well as bulk cells. CD31+ vessels were mainly observed in the areas near necrosis, the frequency of CD31+ vessels was rather high. The monoclonal anti-human CD31 antibody, which exclusively recognizes the human protein, was used for these studies. Therefore, the specific immunostaining verified that the CD31+ microvascular endothelial cells were of human origin. The frequency of CD31 expression in vascular vessels of xenografted tumors originated from iOSCs (Fig. 5M, 5N) was higher than that of bulk cell-derived tumors. These results indicated that spheroid-derived cells could transdifferentiate into vascular endothelial cells in vivo. Hypoxic microenvironment and TGF-β1 signaling promoted the process of transdifferentiation.

Figure 5.

Multipotency of spheroid cells. (A): iOSCs were cultured in 3D Matrigel containing 5% fetal bovine serum (FBS) for 21 days. (B): iOSCs cultured in 3D culture supplemented with 5 ng/ml vascular endothelial growth factor A (VEGF-A) for 21 days. High magnification of the image is shown as an inset. (C): OSCs cultured in 3D culture supplemented with 5 ng/ml VEGF-A for 21 days; hypoxia-induced spheroid cells cultured in 3D Matrigel in normal oxygen tension (D), containing 2 μM CoCl2(E) and containing 5 ng/ml VEGF-A (F) for 21 days. Scale bar = 100 μm. (G): 3D cultures were dehydrated and embedded in paraffin blocks, 4-μm-thick sections were analyzed by H&E staining and immunohistochemistry using anti-CD31 antibody. Scale bar = 50 μm. (H): IF study of CD31 (red) in vessel-like branches that extended out from induced spheres. White dotted lines were used to indicate the borderline of vessel-like branches and induced spheres. (I): CD31 was unable to be detected in the spheres without induction. Nuclei were stained with DAPI (blue). Scale bar = 100 μm. OSCs (J), iOSCs (K), and hypoxia-induced sarcospheres (L) pretreated with 376 nM SB431542, then cultured in 3D Matrigel for 21 days. Scale bar = 100 μm. Immunohistochemical study of CD34 (M) and CD31 (N) in primary tumors derived from iOSCs. Adipogenic differentiation of OSCs (O), iOSCs (P), hypoxia-induced spheroid cells (Q), and adherent cells (R). Adipogenic differentiation of OSCs (S), iOSCs (T), and hypoxia-induced spheroid cells (U), all which were pretreated with 376 nM SB431542. Scale bar = 50 μm. (V): Quantitative reverse transcriptase polymerase chain reaction analysis of the expression of PPARγ and AP2 in adherent cells, spheroid cells, and adipogenic differentiated spheroid cells with or without SB431542 pretreatment (*, p < 0.05). H-OSCs represents hypoxia-induced spheroid cells; OSCs (induced), iOSCs (induced), and H-OSCs (induced) represent adipogenic differentiation of OSCs, iOSCs, and hypoxia-induced spheroid cells, respectively; OSCs (SB + induced), iOSCs (SB + induced), and H-OSCs (SB + induced) represent adipogenic differentiation of OSCs, iOSCs, and hypoxia-induced spheroid cells all which were pretreated with SB431542, respectively. Abbreviation: OSCs, osteosarcoma cells.

Figure 6.

Alteration of gene expression during dedifferentiation. (A): Biological process analysis of differentially regulated genes in iOSCs and adherent cells. The 1,031 genes identified as differentially regulated from the hypergeometric distribution were subjected to biological process analysis. The 23 biological processes at the p < 1 × 10−7 level are presented as decreasing p values (−ln), and descriptions of each biological process are indicated. (B): Cluster analysis of gene expression in residual adherent cells, mesophase, and iOSCs. (C): Hypothetic transcriptional crosstalk between TGF-β, MAPK/ERK, Notch, PDGF, IGF, and Wnt signaling pathways in the process of dedifferentiation. Box A and Box B represent upregulated genes, Box C represents downregulated genes upon TGF-β1 stimulation; the dotted line boxes represent hypothetical signal pathways, and the dotted lines represent translocation. (D): Semiquantitative reverse transcription polymerase chain reaction (RT-PCR) analysis of β-catenin, Hey1, Jag1, IGF1R, PDGFRB, Snail1, Snail2, LTBP1, and Shc1. Total RNA was isolated from residual adherent cells, mesophase, and iOSCs and then RT-PCR analyses were performed. Primers used in this study are listed in Supporting Information Table S4. Images were captured with ImageQuant RT. (E): The Scatter-Plot and cluster analysis of gene expression of OSCs, iOSCs, and hypoxia-induced sarcospheres (R [mt]0.9). Abbreviations: OSCs, osteogenic stem cells; TGF-β1, transforming growth factor β1.

During the adipogenic differentiation process, the morphology of spheroid cells was prismatic after adhesion, and lipid droplets accumulated in the cytoplasm gradually. Finally, the cells were oval in shape, and lipid droplets were more readily observed in the cytoplasm. Sudan III staining verified the mature adipogenic cells differentiated from the spheroid cancer cells (Fig. 5O–5Q). Although some monolayer cells could form lipid droplets in adipogenic differentiation medium, the number of lipid droplets in monolayer cells was limited (Fig. 5R). We detected the expression of PPARγ and AP2 in monolayer cells, OSCs, iOSCs, and hypoxia-induced sarcospheres before and after adipogenic differentiation by qRT-PCR. Spheroid cells which have differentiated into adipogenic cells expressed PPARγ and AP2 at much higher levels in comparison with untreated spheroid cells and monolayer cells (Fig. 5V). To testify whether lineage potential is dependent on TGF-β1 activity, sarcospheres pretreated with SB431542 were cultured in adipogenic differentiation medium. After day 9 of culture, few lipid droplets were found in SB431542-pretreated spheroid cells (Fig. 5S, 5T, 5U). In addition, the expression of PPARγ and AP2 tended to decrease (Fig. 5V).

Analysis of Gene Expression

To evaluate gene alteration during the process of dedifferentiation, we detected gene expression in residual adherent cells, mesophase (adherent cells treated with 4 ng/ml TGF-β1 for 3 days), and iOSCs. The 1,031 probe sets were differentially expressed in iOSCs compared with the adherent cells. Among those cells, 481 probe sets were obviously upregulated, and 550 probe sets were downregulated in iOSCs. Functional annotation of the 1,031 probe sets using the Gene Ontology (GO) analysis tool predicted that the differential expression primarily encompassed genes involved in a variety of biological processes including the development, cell differentiation, signal transduction, transcription, cell adhesion, negative regulation of cell proliferation, induction of apoptosis, cell cycle arrest, and response to hypoxia and angiogenesis (Fig. 6A, 6B). We then performed gene set enrichment analyses (GSEA) to evaluate the association of the gene expression mentioned above with clinicopathological characteristics. GSEA showed that upregulated genes significantly enriched in the patients with poor chemotherapeutic response and metastasis. Downregulated genes closely related with good chemotherapeutic responders (Supporting Information Table S1). We further analyzed genes significantly altered during the process from adherent cells to mesophase and the process from mesophase to iOSCs. GSEA showed that upregulated genes in mesophase relative to adherent cells significantly enriched in the patients with poor chemotherapeutic response, and downregulated genes in iOSCs relative to mesophase significantly enriched in the patients with good chemotherapeutic response and nonmetastasis (Supporting Information Table S1, FDR < 0.25). These results suggested that osteosarcoma cells gain the ability of chemoresistance and metastasis during the progression of TGF-β1 induced dedifferentiation. Furthermore, signaling pathways were activated by TGF-β1 in the process of dedifferentiation, including Jun and Fos of the MAPK/ERK signaling pathway, TGF-β1 and MMP2 of the TGF-β signaling, PDGFRA and PDGFRB of the PDGF signaling, IGF1R and IGFBP5 of the IGF signaling, Notch1, Jagged1, HEY1, and HEY2 of the Notch signaling and β-catenin, Wnt5A and LEF1 of the Wnt signaling. In addition, some genes were downregulated by TGF-β1, such as LTBP1, ID3, and ADAM-9. All these signaling pathways have reported roles in tumor invasion, EMT, metastasis, and angiogenesis. Of note, Snail1 and Snail2, which are two important transcriptional targets of EMT were upregulated (Fig. 6C; Supporting Information Table S2). To confirm the data of the microarray, the related genes were selected and evaluated by semiquantitative RT-PCR analysis (Fig. 6D). Using GSEA analysis, the upregulated genes appeared to be significantly enriched in poor chemotherapeutic responders (Supporting Information Table S1).

We also analyzed the gene expression of OSCs, iOSCs as well as hypoxia-induced sarcospheres. Correlation coefficient (R) analysis indicated that the three kinds of sarcospheres have roughly similar expression pattern (Fig. 6E, R >0.9). Four thousand and seventy-one probe sets (10.414%) were differentially expressed in iOSCs compared with OSCs, 2,014 probe sets (5.22%) were differentially expressed in hypoxia-induced sarcospheres compared with OSCs. Functional annotation of altered probe sets using the GO analysis predicted that hypoxia-induced sarcospheres were more similar to OSCs. Upregulated gene expression in iOSCs relative to both OSCs and hypoxia-induced sarcospheres primarily encompassed genes involved in positive regulation of cell cycle (Supporting Information Fig. S4). Compared with OSCs, the upregulated genes in iOSCs and hypoxia-induced sarcospheres primarily related with metastasis and chemoresistance in the GSEA analysis (Supporting Information Table S1).

DISCUSSION

The CSC hypothesis asserts that only a small subset of cells within a tumor is capable of tumor initiation and growth. CSCs have been identified in a variety of solid tumors. Recent studies delineated CD133+ or CD117+/Stro-1+ as markers for isolating OSCs, which can form the sarcospheres [26–28]. In addition, OSCs were enriched through sarcosphere formation without the use of cell surface markers. The implantation of a few OSCs was sufficient to initiate tumorigenesis in mice, while non-OSCs failed to do so within the same experimental time frame. Treatments targeting OSCs therefore seem to be the next step in combating this disease; several therapies targeting CSCs have been investigated. In this study, we enriched OSCs using a sarcosphere forming assay, in which the enriched OSC subpopulation was found to be CD117+/Stro-1+ and have multilineage potential. However, a portion of non-OSCs could also dedifferentiate to exhibit stem cell properties, including efficient tumor-initiating ability, self-renewal capacity, multipotency, high metastatic potential, and chemoresistance. It was presumed that the dedifferentiation process was likely promoted when non-OSCs were exposed to TGF-β1 or hypoxic stress. On the other hand, dedifferentiated cells could again differentiate into mature cells upon the stimulation with exogenous factors, such as serum. Our results argue against the unidirectional CSC model and favor a model of tumor heterogenicity. Among osteosarcoma cells, those exhibiting stem cell properties can be driven by the microenvironment. In the total cell population, stem cells and non-stem cells coexist in a dynamic equilibrium, which can, in turn, be influenced by the tumor microenvironment. A previous study used CD117+/Stro-1+ to identify CSCs in osteosarcoma [28]. The animal studies therein showed slow-growing tumor formation using a large number of CD117/Stro-1 cells. It was speculated that this was due to contamination of CD117+/Stro-1+ cells in the CD117/Stro-1 population during cell sorting. An alternative explanation is that other populations with stem cell potential exist within the CD117/Stro-1 cell culture. The process of non-stem cell dedifferentiation into stem cells is another important issue to consider. Although OSCs are attractive targets for cancer therapy, the phenomenon of dedifferentiation from non-OSCs to OSCs should not be ignored.

TGF-β1 is one of the critical growth factors that are secreted by osteosarcomas, and also one of the most important cytokines within the bone modulating osteosarcoma cell behavior. In our study, non-OSCs dedifferentiated into the subpopulation with stem cell characteristics, which was specifically promoted by TGF-β1. The sarcospheres from non-OSCs also had strong tumorigenic ability and multilineage potential. Therefore, following the blockage of TGF-β1 signaling pathways, the initiation of dedifferentiation from non-OSC and stem cell self-renewal were inhibited in our studies. Additionally, blocking the TGF-β1 signaling pathway in spheres increased chemosensitivity significantly. GSEA analysis indicated that the genes upregulated by TGF-β1 closely related with chemoresistance and metastasis of osteosarcoma. Yang and Alessandro et al. also found higher TGF-β1 expression in the patients with high-grade osteosarcoma and lung metastasis [30, 31], indicating that TGF-β1 promoted the chemoresistance, tumorigenecity, and metastatic potential of osteosarcoma.

We propose that the TGF-β1 signaling pathways might crosstalk with Wnt, Notch, IGF, and PDGF during the dedifferentiation of osteosarcoma cells (Fig. 6C). Previous studies have demonstrated that TGF-β1 induces EMT through a Smad-independent signaling pathway via the activation of ERK1/2 [32], JNK, and p38 MAPK [33], which then induce Jun and Fos to heterodimerize and activate AP-1 [34]. Our cDNA microarray results show that expression of AP-1, TGF-β1, PDGFRA, and PDGFRB increased significantly during the process of dedifferentiation. Thus, it is likely that TGF-β1 is activated in an autocrine loop via the MAPK/ERK signaling pathway. Once the signaling pathway was activated, osteosarcoma cells could express TGF-β1 in an autocrine manner to further promote the dedifferentiation process. Wnt signaling regulates development, cell proliferation, and cell motility. Recent studies have suggested that canonical Wnt signaling in stem cells is crucial for the maintenance of the self-renewal capacity [35]. The expression of genes related to the canonical Wnt signaling pathway, such as β-catenin, is upregulated in the process of dedifferentiation. We speculated that canonical Wnt signaling plays an important role in promoting osteosarcoma cell dedifferentiation and sustaining osteosarcoma cells in an undifferentiated state. Noncanonical Wnt signaling pathways are believed to be initiated by the activation of FZD and coreceptors [36]. In this study, Wnt5A and FZD8 transcription levels were strikingly upregulated during dedifferentiation, suggesting that noncanonical Wnt signaling pathways also participate in dedifferentiation. Notch1 signaling plays an important role in tumor invasion, EMT, metastasis, and angiogenesis [37]. Previous research suggests that Notch1 binds to the transcription factor to form a complex that activates the transcription of target genes, such as HES1, HEY1, HEY2, Snail1, and Slug [38, 39], our results from microarray experiments show that expression of Notch1, Jagged1, HES1, HEY1, HEY2, Snail1, and Slug are increased significantly. Thus, TGF-β1 may activate the Notch signaling pathway during dedifferentiation. The IGF signal is composed of a dynamic network, including IGF1, IGF2, IGFR1, IGFR2, IGF-binding proteins (IGFBPs), and IGFBP proteases [40]. IGF1R is the main receptor for IGF2, which induces a redistribution of β-catenin from the plasma membrane to the nucleus [41]. IGFBP5 is an important regulator of bone formation in IGF-dependent and IGF-independent mechanisms, which activate EMT in a manner similar to TGF-β1 [42]. Results from microarray analyses show that the expression of IGF2, IGF1R, and IGFBP5 are dramatically increased during TGF-β1 induced dedifferentiation, while the expression of ADAM-9, an IGFBP-5 protease, is downregulated. Based on recent reports coupled with our findings, we speculate that the IGF signaling pathway plays an important role in TGF-β1-induced dedifferentiation. Moreover, TGF-β1 strongly increases the expression of PDGFRA and PDGFRB, suggesting that TGF-β1 activates the PDGF signaling pathway in osteosarcoma. Further experiments are needed to explore TGF-β1 signaling pathways in osteosarcoma.

Osteosarcomas frequently have areas of hypoxic or anoxic tissue, and the oxygen consumption in these highly metabolic tissues cannot match the oxygen delivery to these sites. Measurements of HIF-1α levels, as a marker of tumor hypoxia, are correlated with poor outcomes and highly aggressive disease in various malignancies including bone, breast, ovarian, colorectal and lung cancers. Furthermore, the bone marrow microenvironment is inherently hypoxic [43], indicating that the adaptations to hypoxia used for cell survival may additionally help select for highly aggressive and increasingly malignant cells. Generally, hypoxia is assumed to maintain an undifferentiated state in stem cell populations. It has been reported that hypoxia is a feature of the CSC niche and that the targeted inhibition of factors involved in tumor hypoxia attenuate tumor initiation potential [44]. HIFs, in particular HIF-1α, have been shown to regulate signal pathways that regulate CSC self-renewal and multipotency [45]. Recent studies have found that high expression of HIF-1α had a close relationship with angiogenesis, metastasis, and poor outcome of human osteosarcoma patients [46–49]. In our study, we found that hypoxia was a critical component of the OSC niche and could induce osteosarcoma cell dedifferentiation. In the process of the dedifferentiation, hypoxic stress and TGF-β1 stimulation promoted the expression of HIF1α. In addition, inhibiting the TGF-β1 signaling pathway not only reduced the expression of HIF1α but also blocked the hypoxia-induced dedifferentiation of osteosarcoma cells. Therefore, TGF-β1 signaling pathway might be involved in the hypoxia-induced dedifferentiation of osteosarcoma cells via the activation of HIF-1α.

Tumor vasculature is indispensable for tumor growth, invasion, and metastasis, and is generated via angiogenesis and vasculogenesis. Traditionally, it was thought that tumor vasculature was derived from pre-existing blood vessels, which were stimulated by various angiogenic growth factors secreted by cancer cells. However, increasing evidence suggests that the tumor vasculature is formed through a process of vasculogenesis, whereby some cancer cells can form blood vessels in tumors via transdifferentiation into vascular endothelial cells, for example, vasculature in human melanoma and brain tumors may be derived from cancer cells rather than pre-existing vessels [50–52]. In our study, we provided additional evidence showing the sarcosphere-derived vasculature in osteosarcoma to be an important source for tumor vasculogenesis. Importantly, our results show that the spheroid cells generated from non-stem cells could differentiate into vascular endothelial-like cells and several types of cancer cells. The series of phenotypic changes observed in 3D culture suggest that neovasculogenesis, as initiated by OSCs or iOSCs, may occur in the early stages of tumor formation. These results indicate that dedifferentiation is important for tumor initiation and vasculogenesis. In xenografted tumors, CD31-positive vascular endothelial cells were detected in some areas, indicating that vasculogenesis might be initiated by osteosarcoma cells. This observation suggests that both vasculogenesis and angiogenesis are required during tumorigenesis. Osteosarcoma cell-derived endothelial-like cells bearing the same genomic profile as tumor cells have different sensitivities to conventional therapy. Moreover, osteosarcoma cell-derived vessel provides an opportunity for tumor metastasis via the hematogenous route. Thus, targeting the process that OSCs differentiate into endothelial cells may provide new therapeutic options for osteosarcoma. In addition, hypoxia-induced sarcospheres could form vascular endothelial-like cells in 3D culture, while hypoxia accelerated the transdifferentiation process. It was reported that hypoxic stress enhanced the expression of VEGF-A, a major regulator of angiogenesis, in osteosarcoma [53]. The expression of VEGF-A is upregulated in response to microenvironmental cues such as hypoxia. Therefore, microenvironment targeting could block cancer cell dedifferentiation and further block neovasculogenesis in osteosarcoma.

CONCLUSIONS

In summary, our results suggest that TGF-β1 is a crucial cytokine for osteosarcoma progression. TGF-β1 signaling significantly promoted the tumorigenicity, neovasculogenesis, chemoresistance, and metastatic potential of osteosarcoma. TGF-β1-targeting therapies may offer new inroads to cure the patients with osteosarcoma.

Acknowledgements

We are grateful to Dr. Derek Kennedy (Griffith University, Australia) and Dr. Guokai Chen (Morgridge Institute for Research, Madison, WI) for critical reading the manuscript and Linxuan Huang for administrative supports. This study was supported in part by a grant from the program of the State High-Tech Development Project (No. 2008AA092604), National Basic Research Program (No. 2009CB945400) and Guangdong Planning Project of Science and Technology (No. 2009B030803037).

DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST

The authors indicate no potential conflict of interest.

Ancillary