Disulfiram Eradicates Tumor-Initiating Hepatocellular Carcinoma Cells in ROS-p38 MAPK Pathway-Dependent and -Independent Manners

Tumor-initiating cells (TICs) play a central role in tumor development, metastasis, and recurrence. In the present study, we investigated the effect of disulfiram (DSF), an inhibitor of aldehyde dehydrogenase, toward tumor-initiating hepatocellular carcinoma (HCC) cells. DSF treatment suppressed the anchorage-independent sphere formation of both HCC cells. Flow cytometric analyses showed that DSF but not 5-fluorouracil (5-FU) drastically reduces the number of tumor-initiating HCC cells. The sphere formation assays of epithelial cell adhesion molecule (EpCAM)+ HCC cells co-treated with p38-specific inhibitor revealed that DSF suppresses self-renewal capability mainly through the activation of reactive oxygen species (ROS)-p38 MAPK pathway. Microarray experiments also revealed the enrichment of the gene set involved in p38 MAPK signaling in EpCAM+ cells treated with DSF but not 5-FU. In addition, DSF appeared to downregulate Glypican 3 (GPC3) in a manner independent of ROS-p38 MAPK pathway. GPC3 was co-expressed with EpCAM in HCC cell lines and primary HCC cells and GPC3-knockdown reduced the number of EpCAM+ cells by compromising their self-renewal capability and inducing the apoptosis. These results indicate that DSF impaired the tumorigenicity of tumor-initiating HCC cells through activation of ROS-p38 pathway and in part through the downregulation of GPC3. DSF might be a promising therapeutic agent for the eradication of tumor-initiating HCC cells.


Introduction
Accumulating evidence has revealed that a minor population of tumor cells, called cancer stem cells or tumor-initiating cells (TICs), organizes a cellular hierarchy in a similar fashion to normal stem cells and shows pronounced tumorigenic activity in xenograft transplantations [1]. Recent progress in stem cell biology and technologies has contributed to the identification and characterization of TICs in various cancers including hepatocellular carcinoma (HCC) [2]. In HCC, side population cells and cells expressing several surface molecules such as epithelial cell adhesion molecule (EpCAM), CD133, CD90, and CD13 have been reported to function as TICs [3]. Besides the identification of tumor-initiating HCC cells, cancer-related molecules and signaling pathways, such as the polycomb group proteins, NANOG, AKT/ PKB signal, and Wnt/b-catenin, have been shown to play an important role in maintaining or augmenting of tumor-initiating capability of TICs [4]. Although inhibitors of these molecules and signaling pathways may be potent TIC-targeting drugs, no effective therapy targeting TICs has been developed.
Disulfiram (DSF) is an irreversible inhibitor of aldehyde dehydrogenase and has been clinically used in the treatment of alcohol dependence for roughly 70 years [5]. DSF is a potent therapeutic agent in a wide range of human cancers. In addition, recent reports showed that DSF reduced the number of tumorinitiating cells and attenuated their sphere-forming abilities in breast cancer and glioblastoma [6,7]. Although these findings indicate that DSF could eradicate TICs, the molecular machinery of its effect against TICs still remains largely unknown.
In the present study, we examined the effects of DSF on tumorinitiating HCC cells in vitro and in vivo. We found that DSF impaired their tumor-initiating ability and induced apoptosis by activating the reactive oxygen species (ROS)-p38 pathway. Furthermore, the downregulation of Glypican3 (GPC3) expression, which is caused independently of the ROS-p38 pathway, appeared to also be responsible for the anti-TIC effect of DSF.

DSF inhibited tumorigenicity of HCC cells in vitro and in a xenograft transplantation model
As shown in a variety of cancer cells [8][9][10], DSF treatment inhibited cell growth in both a time-dependent and dosedependent manner in HCC cells ( Figure S1A). Immunostaining of active caspase-3 (CASP3) showed that the DSF treatment induced apoptosis dose-dependently ( Figure S1B). The percentage of apoptotic cells was roughly ten-fold higher among HCC cells treated with DSF (1 mM) than among control cells ( Figure S1C). To examine whether DSF affected the tumorigenic ability of HCC cells, we conducted a non-adherent sphere assay, a standard assay for evaluating tumorigenic capacity. Sphere-forming ability was significantly impaired in DSF-treated HCC cell lines in a dosedependent manner ( Figure 1A and 1B). Subsequently, we determined the effects of DSF using a xenograft nonobese diabetic/severe combined immunodeficient (NOD/SCID) mouse model. After the implantation of 2610 6 Huh1 and Huh7 cells into NOD/SCID mice, DSF was administered intraperitoneally every other day. Tumor initiation and growth were apparently suppressed by the DSF treatment in a dose-dependent manner ( Figure 1C and 1D). Together, these results indicate that DSF reduced the tumorigenicity of HCC cells.
Loss-of-function assays of ALDH1 and ALDH2 DSF and its metabolites were shown to suppress ethanol metabolism mainly through the inhibition of cytosolic aldehyde dehydrogenase 1 (ALDH1) and mitochondrial ALDH2 [11]. It has been reported that ALDH-knockdown reduced proliferation and motility of lung cancer cells [12]. Because we previously showed that there was no association between the expression of ALDH1 and EpCAM or CD13 and that ALDH1-knockdown affected neither cell growth nor tumorigenicity in HCC cells [13], we conducted loss-of-function assays on ALDH2. We achieved the stable knockdown of ALDH2 in Huh1 and Huh7 cells with lentivirus-mediated short hairpin RNA (shRNA) against ALDH2 using enhanced red fluorescent protein (ERP) as a marker for infection ( Figure S2A). No significant differences in cell growth and sphere formation were observed between ALDH2-knockdown cells and control cells expressing shRNA against luciferase (sh-Luc) ( Figure S2B and S2C). Additionally, double-knockdown of ALDH1 and ALDH2 in the culture produced similar results to the singleknockdown of ALDH2 ( Figure S2D-F). Taken together, the effects of DSF on HCC cells appeared to be independent of its inhibitory function toward ALDH1 and ALDH2.

Decrease in the number of tumor-initiating HCC cells after DSF exposure
We then examined the expression of various markers of tumorinitiating HCC cells such as CD13, epithelial cell adhesion molecule (EpCAM), and CD133 using flow cytometry. The DSF treatment appeared to decrease the number of HCC cells expressing these markers (Figure 2A). Among them, the EpCAM-high fraction markedly decreased from 44.4% to 9.8% in Huh1 cells and from 36.7% to 12.5% in Huh7 cells. Concordant with this, real-time RT-PCR analysis showed decreased expression of E-cadherin (CDH1) and alfa-fetoprotein (AFP), hepatic stem/ progenitor cell markers, in DSF-treated cells ( Figure 2B). In clear contrast, the 5-FU treatment resulted in the enrichment of TIC fractions ( Figure S3). These results indicate that the biological effect of DSF differs from that of 5-FU, and is promising for the eradication of tumor-initiating HCC cells.
DSF activated p38 MAPK in response to increased intracellular ROS levels in tumor-initiating HCC cells Consistent with previous reports [6,7], the present flow cytometric analyses showed that intracellular ROS levels were higher in DSF-treated HCC cells than in control cells ( Figure 3A). However, co-treatment with NAC canceled this increase in ROS levels ( Figure 3A). Western blotting showed increased levels of phosphorylated p38 after DSF exposure, which indicates p38 MAPK activation in HCC cells ( Figure 3B). It has been well established that TICs maintain ROS at levels as low as normal stem cells [14,15]. ROS levels were higher in EpCAM 2 HCC cells than in EpCAM + cells ( Figure 3C). Notably, the co-treatment of sorted EpCAM + cells with the antioxidant, NAC, canceled the phosphorylation of p38 induced by DSF ( Figure 3D). Although EpCAM 2 HCC cells generated only a small number of spheres, DSF treatment further reduced the number of spheres ( Figure S4A and S4B). Approximately 90% of EpCAM + cells treated with DSF was positive for phosphorylated p38 ( Figure 3D), but the rate for EpCAM 2 cells positive for phosphorylated p38 was nearly 25% ( Figure S4C). The cell growth of EpCAM + HCC cells was greatly restored by the additional NAC treatment ( Figure 3E). Together, DSF caused activation of the ROS-p38 MAPK pathway in tumorinitiating HCC cells.

p38 MAPK activation impaired self-renewal capability of tumor-initiating HCC cells
To examine the impact of p38 MAPK activation on tumorinitiating HCC cells, we conducted sphere formation assays on EpCAM + HCC cells treated with DSF and/or SB203580, a specific inhibitor of p38 ( Figure 4A). The co-treatment of cells with SB203580 largely abrogated the cell growth inhibition and apoptosis observed following the DSF treatment ( Figure S5). Consistent with this, additional SB203580 treatment significantly restored the sphere-forming ability of EpCAM + HCC cells ( Figure 4B). Additionally, subsequent analyses for secondary sphere formation after replating showed results similar to those for the primary spheres ( Figure 4C). These results indicate that activated p38 MAPK restricts the self-renewal of tumor-initiating HCC cells. We then conducted immunocytochemical analyses of the spheres and examined the expression of EpCAM and afetoprotein (AFP), a hepatic stem/progenitor cell marker [16]. Although the DSF treatment decreased the number of cells positive for AFP or EpCAM, co-treatment with DSF and SB203580 restored the number of positive cells ( Figure 4D and 4E). Taken together, DSF impaired the tumor-initiating capability of HCC cells in part in a p38-dependent manner.

Gene expression profiles of EpCAM + HCC cells treated with DSF
EpCAM + HCC cells treated with DSF or 5-FU for 48 hours were subjected to oligonucleotide microarray experiments. Concordant with the results presented in Figures 3 and 4, gene set enrichment analysis (GSEA) showed that EpCAM + HCC cells treated with DSF, but not 5-FU were significantly enriched for genes involved in p38-MAPK signaling ( Figure 5A) [17,18]. The DSF treatment altered the expression of several genes involved in cell cycle regulation ( Figure S6A and S6B). In particular, striking upregulation of p57KIP2 was observed in Huh1 EpCAM + cells. The gene set for the proteasome pathway showed a higher enrichment score in DSF-treated EpCAM + HCC cells than in 5-FU-treated cells, although there was no significant difference ( Figure S6C) [19].
We identified DSF-responsive genes (698 upregulated genes and 605 downregulated genes) and 5-FU-responsive genes (717 upregulated genes and 1,350 downregulated genes) ( Figure 5B and 5C). Of interest, the DSF treatment causes no marked changes in the gene expression of the ROS scavenger pathway ( Figure S6D). Furthermore, functional annotation analysis revealed different gene expression profiles between EpCAM + HCC cells treated with DSF and 5-FU (Table S1 and S2). In particular, gene ontology terms enriched for downregulated genes were different. Additionally, 23 genes categorized into ''liver cancer'' were downregulated after exposure to DSF, but not 5-FU ( Figure 5D). Among them, Glypican3 (GPC3) was shown to be specifically overexpressed in human HCC and GPC3-knockdown induced apoptosis in HCC cells [20,21]. Quantitative RT-PCR showed that GPC3 expression was downregulated in EpCAM + HCC cells treated with DSF as shown in the microarray analyses ( Figure 5E). However, the downregulation of GPC3 was not observed in EpCAM 2 HCC cells after DSF treatment (data not shown).

Regulation of GPC3 gene expression
To examine whether activation of the ROS-p38 MAPK pathway was crucial to the downregulation of GPC3 expression by DSF, we examined GPC3 expression in EpCAM + HCC cells co-treated with NAC or SB203580. Neither NAC nor SB203580 restored the expression of GPC3 ( Figure S7A). In addition, proteasome inhibition by the MG132 treatment had no effect on GPC3 expression ( Figure S7B). These findings indicate that neither ROS-p38 MAPK pathway activation nor proteasome inhibition contributed to the downregulation of GPC3 expression.

Loss-of-function and gain-of-function assays of GPC3 in EpCAM + HCC cells
Dual immunostaining analyses showed that GPC3 and EpCAM were frequently co-expressed in HCC cells ( Figure 6A). Moreover, quantitative RT-PCR revealed a higher level of GPC3 expression in the EpCAM + fraction than in the EpCAM 2 fraction ( Figure 6B). Stable HCC cell lines expressing shRNA against GPC3 or luciferase were successfully obtained by cell sorting with enhanced green fluorescent protein (EGFP) as a marker for viral infection. Western blot analysis of these cells showed that both shRNAs against GPC3 (sh-GPC3-1 and sh-GPC3-2) markedly repressed GPC3 expression, although sh-GPC3-1 was more effective than sh-GPC3-2 ( Figure 6C). GPC3-knockdown suppressed cell growth and induced apoptosis relative to sh-Luc ( Figure S7C and S7D). Similarly, GPC3-knockdown markedly impaired primary sphere formation by EpCAM + cells and EpCAM 2 cells and more severely impaired secondary sphere formation ( Figure 6D-F). Immunocytochemical analyses of the large spheres showed a decrease in the number of cells expressing AFP or EpCAM ( Figure S7E and S7F). In contrast, the stable overexpression of GPC3 promoted cell growth and sphere formation of tumor-initiating HCC cells ( Figure S8). Together, these results indicate that GPC3-knockdown suppresses tumorigenicity of HCC cells by directly affecting the cell growth and the self-renewal of TIC.

Discussion
High levels of ALDH activity are characteristic of normal stem cells in a variety of organs. The human ALDH superfamily consists of 19 putatively functional genes [22]. ALDH1 is a major isoform in mammalian tissues and functions as a stem cell marker in liver and mammary stem cells [23,24]. Recent reports have indicated ALDH1 to be a useful marker for the enrichment of TICs from various cell lines and primary tumors. It has been shown that a high level of ALDH1 expression correlates with malignant phenotypes and an unfavorable prognosis in a range of cancers [24].
In this study, we first showed that DSF inhibited the proliferation and sphere-forming ability of HCC cells in a dosedependent manner. In addition, DSF suppressed tumor growth in xenograft transplant experiments using NOD/SCID mice. Our flow cytometric analysis showed that the DSF treatment caused a significant decrease in the number of tumor-initiating HCC cells expressing surface markers such as CD13, CD133, and EpCAM. Knockdown of ALDH1 and ALDH2 in HCC cells had no effect on cell proliferation and sphere-forming ability in the culture. Our findings suggest that DSF exerts its anti-HCC function in an ALDH-independent fashion.
HSCs have been shown to tightly control intracellular ROS levels to maintain long-term self-renewal and survival [25]. Conversely, activation of p38 MAPK upon an elevation in ROS levels resulted in the exhaustion of HSCs [26]. Similarly, TICs in a wide range of tumors exhibited lower concentrations of ROS than corresponding non-TICs. In addition, lower ROS levels in TICs were shown to be closely associated with both chemo-sensitivity and radio-sensitivity [15]. In the present study, we confirmed that EpCAM + HCC cells contained lower ROS levels than EpCAM 2 cells. Because previous studies reported that DSF activated the ROS-p38 MAPK pathway and thereby suppressed the sphereforming ability of TICs [6,7], we examined whether exposure to DSF activated the ROS-p38 MAPK pathway in tumor-initiating HCC cells. As expected, the treatment of EpCAM + HCC cells with NAC canceled p38 activation. Moreover, the SB203580 treatment largely restored the tumorigenicity of EpCAM + HCC cells. These findings indicate that the ROS-p38 MAPK pathway is directly associated with cell growth and tumor-initiating capability of HCC cells. Low levels of ROS in TICs have been attributable to the activation of the ROS scavenger pathway [27]. The present microarray results showed comparatively high expression levels of ROS scavenger genes such as GCLM and GSS in purified EpCAM + HCC cells. However, the DSF treatment caused no marked changes to the ROS scavenger genes. Considering that not only H2DCFDA staining but also MitoSOX staining showed a high ROS level in DSF-treated EpCAM + HCC cells, DSF might increases mitochondrial ROS production rather than impairs the scavenging of ROS. Further analysis is required to clarify this point.  Of interest, our microarray analyses revealed that DSF acted in a manner different from 5-FU. The GSEA results support the present biological finfings and implicate the activation of p38 in the anti-TIC activity of DSF. Importantly, the 23 genes in the ''liver cancer'' category were significantly downregulated after the DSF exposure, but none of them was significantly altered after the 5-FU treatment. One of these genes, GPC3, was frequently overexpressed in HCC and increased GPC3 expression was correlated with a poor prognosis among HCC patients [20,21]. A clinical trial using a GPC3 peptide vaccine in patients with advanced HCC has also been carried out [28]. While GPC3 functions as a marker for normal hepatic stem/progenitor cells [29], the immunostaining analyses showed an association between the expression of EpCAM and GPC3 in both HCC cell lines and downregulated genes annotated as ''liver cancer'' in DSF-treated EpCAM + HCC cells. (E) The expression of GPC3 in DSF-treated EpCAM + cells was compared to that in control cells. The data obtained by microarray analyses and quantitative RT-PCR analyses are presented. doi:10.1371/journal.pone.0084807.g005 HCC surgical specimens (data not shown) and the higher basal expression of GPC3 in EpCAM + cells than EpCAM 2 cells. Lentiviral knockdown of GPC3 significantly reduced the sphereforming ability of EpCAM + HCC cells. Additionally, replating assays and immunocytochemical analyses of EpCAM and AFP indicated that GPC3 regulated tumor-initiating HCC cells. Although it appears that DSF suppresses the tumorigenicity of tumor-initiating HCC cells in part by downregulating GPC3 expression, further analyses would be of importance to clarify the mechanisms underlying the downregulation of GPC3 by DSF.
Finally, our findings successfully demonstrated that DSF significantly reduced the number of tumor-initiating HCC cells through apoptosis induction and the conversion to non-TICs. These effects appeared to be attributable to the activation of the ROS-p38 MAPK pathway and gene silencing with GPC3 ( Figure 6G). Further analyses of the genes listed here are necessary to determine the effects of DSF. Recent reports showed that TICs of brain tumors reside in vascular niches in which endothelial cells maintain the TICs in an undifferentiated state [30]. Bevacizumab, a vascular endothelial growth factor (VEGF)-specific inhibitor, causes a drastic decrease in the number of TICs in vascular niches by inhibiting the self-renewal of TICs [31]. Although the niche for TICs in HCC remains to be elucidated, combination therapy using DSF and the anti-angiogenic multi-kinase inhibitor sorafenib might be effective in the eradication of tumor-initiating HCC cells.

Ethics statement
All experiments using the mice were performed in accordance with our institutional guidelines for the use of laboratory animals and approved by the review board for animal experiments of Chiba University (approval ID: 22-187).

Mice
Nonobese diabetic/severe combined immunodeficiency (NOD/ SCID) mice (Sankyo-Lab Service, Tsukuba, Japan) were bred and maintained in accordance with our institutional guidelines for the use of laboratory animals.

Non-adherent sphere culture
For the sphere formation assay of Huh1, Huh6 and Huh7 cells, 1,000 cells were plated onto ultra-low attachment 6-well plates (Corning, Corning, NY). For the assay of PLC/PRF/5 cells, 500 cells were plated onto NanoCulture 24-well plates (Scivax, Kawasaki, Japan). The number of spheres (.100 mm in diameter) was counted on day 14 of culture. For the secondary sphere formation, a single cell suspension derived from primary colonies was obtained using a Neurocult chemical dissociation kit (StemCell Technologies, Vancouver, BC). Paraffin-embedded sections of the spheres were subjected to hematoxylin & eosin (H&E) staining and immunohistochemical staining with anti-EpCAM (Cell Signaling Technology, Beverly, MA) and anti-AFP (Dako Cytomation, Carpinteria, CA) antibodies.

Cell sorting and analysis
Single-cell suspensions were stained with allophycocyanin (APC)-conjugated anti-EpCAM antibody and anti-CD13 antibody (Biolegend, San Diego, CA) or APC-conjugated anti-CD133/1 antibody (Miltenyi Biotec, Auburn, CA). After the incubation, 1 mg/ml of propidium iodide was added to eliminate dead cells. Flow cytometirc cell sorting and analyses were performed using FACSAria or FACSCanto (BD Biosciences, San Jose, CA). Intracellular ROS levels were determined by flow cytometry using H2DCFDA (Sigma) and MitoSOX (Molecular Probes, Eugene, OR) staining.
Xenograft transplantation using NOD/SCID mice A total of 2610 6 Huh1 and Huh7 cells were suspended in DMEM and Matrigel (BD) (1:1). The cells were implanted into the subcutaneous space of the backs of NOD/SCID mice. DSF (10 or 50 mg/Kg) was administered intraperitoneally every other day.

Western blotting
DSF-treated HCC cells were subjected to Western blot analysis using anti-p38 (Santa Cruz Biotechnology, Santa Cruz, CA), antiphospho-p38 (Cell Signaling Technology), and anti-tubulin (Oncogene Science, Cambridge, MA) antibodies. ALDH2-knockdown cells and ALDH1-and ALDH2-double knockdown cells were subjected to Western blotting using anti-ALDH1 (BD Biosciences) and anti-ALDH2 (Abcam, Cambridge, MA) antibodies. GPC3knockdown cells selected by cell sorting for enhanced green fluorescent protein (EGFP) expression were also subjected to Western blot analysis using anti-GPC3 antibody (Santa Cruz Biotechnology).

Generation of stable GPC3-expressing cells
Human GPC3 cDNA was cloned into a site upstream of IRESneomycin in the pLP-IRESneo vector (Clontech, Palo. Alto, CA). Stable transfection into Huh1 cells with G418 selection was performed.

Reverse transcription-polymerase chain reaction (RT-PCR)
Quantitative RT-PCR was performed with an ABI PRISM 7300 Sequence Detection System (Applied Biosystems) using the Universal Probe Library System (Roche Diagnostics) according to the manufacturer's directions. The sequences of primers are listed in Table S3. Relative quantification was conducted by using the comparative cycle threshold (Ct) method.

Microarray analysis
Cy3-labeled complementary RNA was hybridized to a Sur-ePrint G3 Human GE 8660 K microarray (Agilent Technologies, Santa Clara, CA). Array images were scanned using a DNA Microarray Scanner (Agilent) and analyzed using Feature Extraction version 10.27.1.1. (Agilent). Normalization was performed using GeneSpring GX11.5.1 (Agilent). The expression value (Signal) for each probe set was calculated using GeneSpring GX 12.0 (Agilent). Data were obtained for triplicate samples from three independent experiments. The data were subjected to normalization using GeneSpring normalization algorithms (Agilent). Only gene expression levels with statistical significance (p, 0.05) were recorded as being ''detected'' above background levels, and genes with expression levels below this statistical threshold were considered ''absent.'' To identify differentially expressed genes in EpCAM + cells, we selected probe sets that exhibited gene expression changes with statistical significance as follows: (i) genes exhibiting a change greater than 1.5-fold (p,0.05), (ii) genes exhibiting a change from 1.0 to 1.5-fold (p,0.01), and (iii) switchon type (upregulated from the ''absent'' to ''present'' level) and switch-off type genes (downregulated from the ''present'' to ''absent'' level) exhibiting a change greater than 4.0-fold (p, 0.01). Moreover, functional analyses were performed using Ingenuity Pathway Analysis (IPA) version 12402621 (Ingenuity Systems). To identify gene signatures after DSF or 5-FU treatment, gene set enrichment analysis (GSEA) was also conducted [33]. The raw data are available at http://www.ncbi. nlm.nih.gov/geo/(accession number; GSE 42318).

Statistical analysis
Data are presented as the mean 6 SEM. Statistical differences between 2 groups were analyzed using the Mann-Whitney U test. P values less than 0.05 were considered significant.