Ceramide Glycosylation by Glucosylceramide Synthase Selectively Maintains the Properties of Breast Cancer Stem Cells*

Background: Glucosylceramide synthase catalyzes ceramide glycosylation that regulates the synthesis of glycosphingolipids. Results: Increased globo-series glycosphingolipids in breast cancer stem cells activate c-Src signaling and β-catenin-mediated transcription up-regulating stem cell factors. Conclusion: Ceramide glycosylation maintains the stemness of cancer stem cells. Significance: Glycosphingolipids in cell membrane actively participate in maintaining cancer stem cells. Cancer stem cells are distinguished from normal adult stem cells by their stemness without tissue homeostasis control. Glycosphingolipids (GSLs), particularly globo-series GSLs, are important markers of undifferentiated embryonic stem cells, but little is known about whether or not ceramide glycosylation, which controls glycosphingolipid synthesis, plays a role in modulating stem cells. Here, we report that ceramide glycosylation catalyzed by glucosylceramide synthase, which is enhanced in breast cancer stem cells (BCSCs) but not in normal mammary epithelial stem cells, maintains tumorous pluripotency of BCSCs. Enhanced ceramide glycosylation and globotriosylceramide (Gb3) correlate well with the numbers of BCSCs in breast cancer cell lines. In BCSCs sorted with CD44+/ESA+/CD24− markers, Gb3 activates c-Src/β-catenin signaling and up-regulates the expression of FGF-2, CD44, and Oct-4 enriching tumorigenesis. Conversely, silencing glucosylceramide synthase expression disrupts Gb3 synthesis and selectively kills BCSCs through deactivation of c-Src/β-catenin signaling. These findings highlight the unexploited role of ceramide glycosylation in selectively maintaining the tumorous pluripotency of cancer stem cells. It speculates that disruption of ceramide glycosylation or globo-series GSL is a useful approach to specifically target BCSCs specifically.


Cancer stem cells are distinguished from normal adult stem cells by their stemness without tissue homeostasis control. Glycosphingolipids (GSLs), particularly globo-series GSLs, are important markers of undifferentiated embryonic stem cells,
but little is known about whether or not ceramide glycosylation, which controls glycosphingolipid synthesis, plays a role in modulating stem cells. Here, we report that ceramide glycosylation catalyzed by glucosylceramide synthase, which is enhanced in breast cancer stem cells (BCSCs) but not in normal mammary epithelial stem cells, maintains tumorous pluripotency of BCSCs. Enhanced ceramide glycosylation and globotriosylceramide (Gb3) correlate well with the numbers of BCSCs in breast cancer cell lines. In BCSCs sorted with CD44 ؉ /ESA ؉ /CD24 ؊ markers, Gb3 activates c-Src/␤-catenin signaling and up-regulates the expression of FGF-2, CD44, and Oct-4 enriching tumorigenesis. Conversely, silencing glucosylceramide synthase expression disrupts Gb3 synthesis and selectively kills BCSCs through deactivation of c-Src/␤-catenin signaling. These findings highlight the unexploited role of ceramide glycosylation in selectively maintaining the tumorous pluripotency of cancer stem cells. It speculates that disruption of ceramide glycosylation or globo-series GSL is a useful approach to specifically target BCSCs specifically.
Cancer stem cells (CSCs), 3 which have been characterized with surface markers in tumors, possess the malignant capacities of self-renewal and pluripotency, thus initiating tumorigenesis and driving tumor progression (1)(2)(3)(4). In human breast cancer, the CD44 ϩ /ESA ϩ /CD24 Ϫ/low cells have been tested as breast cancer stem cells (BCSCs), because they are able to differentiate into cells with diverse phenotypes and have tumorous pluripotency to generate mammary tumors and metastases in vivo (3,5). BCSCs, like other CSCs, give rise to tumor resistance to chemotherapy and radiation (6 -8). As a cause of tumor metastasis and recurrence, CSC is an adverse prognostic factor for cancer patients (9 -11) and a critical target to eradicate cancers (12)(13)(14)(15). CSCs are distinguished by loss of tissue homeostasis control and often display unlimited proliferation and growth from normal stem cells, even though both share the similar properties in self-renewal, pluripotency, and resistance to cytotoxins (2,16). Targeting CSCs pharmacologically for therapeutic purpose requires understanding by which mechanisms CSCs maintain their tumor behaviors.
Several cellular signals including Wnt, Notch, and Hedgehog have been reported to be implicated in the self-renewal ability and pluripotency of normal stem cells in mammary gland development or remodeling and of BCSCs in cancer pathogenesis (5, 16 -19). It is not clear how BCSCs, like other CSCs, maintain tumor pluripotency without tissue homeostasis control. Glycosphingolipids (GSLs), particular globo-series GSLs, are common markers of undifferentiated embryonic stem cells (ESCs) (20 -22). As essential components of lipid rafts, particularly GSL-enriched microdomains in plasma membrane, GSLs actively mediate cellular signals, gene regulation, and functions (23)(24)(25)(26). GSLs may play a vital role in maintaining ESCs, because deletion of GSLs induces apoptosis of ESCs and stops embryo development in ugcg (encoding glucosylceramide synthase, GCS) knock-out mouse (27). GCS catalyzes ceramide glycosylation and is a limiting enzyme controlling the synthesis of GSLs (28,29). Enhanced ceramide glycosylation by GCS converts ceramide to GSLs, conferring multidrug resistance of cancer cells (30,31). It has been reported that GCS is overexpressed in metastatic breast tumors (32,33) and inhibition of GCS decreases lung tumor metastasis (34). These studies suggest that ceramide glycosylation by GCS is associated with CSC behaviors. We examined the correlation of ceramide glycosylation with BCSCs in drug resistance and tumorigenesis.

EXPERIMENTAL PROCEDURES
Cell Culture and Treatments-Human MCF-7 breast cancer cells and the doxorubicin-selected subline MCF-7/Dox were kindly provided by Dr. Kapil Mehta (M. D. Anderson Cancer Center, Houston, TX). MCF-7/Dox cells were derived from MCF-7 cells by stepwise culture with doxorubicin (35). Human MCF-12A mammary epithelial cells were purchased from American Type Culture Collection (Manassas, VA). The MCF-7 and MCF-7/Dox were cultured in RPMI 1640 medium containing 10% FBS, 100 units/ml penicillin, 100 g/ml streptomycin, and 584 mg/liter L-glutamine. MCF-12A cells were cultured in Dulbecco's modified Eagle's medium/F-12 (1:1) with 5% horse serum, insulin (5 g/ml), hydrocortisone (500 ng/ml), human epidermal growth factor (20 ng/ml), and cholera toxin (100 ng/ml). All of the cells were maintained in an incubator humidified with 95% air and 5% CO 2 at 37°C. Cell lines were authenticated in November 2010 at the John Hopkins University Fragment Analysis Facility (Baltimore, MD) by profiling short tandem repeats for breast cells. FBS was purchased from HyClone (Logan, UT), medium was from Invitrogen, and other reagents from Sigma-Aldrich.
A mixed backbone oligonucleotide (MBO) was designed to target the ORF 18 -37 of human GCS and designated as MBO-asGCS (36,37). For MBO-asGCS treatment, cells (2 ϫ 10 6 cells/ 100-mm dish) were grown in 10% FBS RPMI 1640 medium overnight, and MBO-asGCS (100 nM) were introduced into cells using Lipofectamine 2000 (Invitrogen). The cells were cultured with MBO-asGCS in 10% FBS RPMI 1640 medium for 7 days. An additional transfection of MBO-asGCS under the same conditions was conducted on day 4. MBO-asGCS was synthesized and purified by reverse phase HPLC and desalting in Integrated DNA Technologies (Coralville, IA).
Cell Viability Assay-Cell viability was analyzed by quantification of ATP, an indicator of metabolically active cells using the CellTiter-Glo luminescent cell viability assay (Promega, Madison, WI), as described previously (38). Briefly, cells (4,000 cells/well) were grown in 96-well plates with 10% FBS RPMI 1640 medium for 24 h. MBO-asGCS was introduced into cells by Lipofectamine 2000 (vehicle control) in Opti-MEM reduced serum medium for 4 h of incubation. The cells were then incubated with increasing concentrations of doxorubicin in 5% FBS medium for another 72 h. Cell viability was determined by the measurement of luminescent ATP in a Synergy HT microplate reader (BioTek, Winooski, VT), following incubation with CellTiter-Glo reagent (Promega).
Colony Formation Assay-The cells (10,000 cells/800 l) were suspended in 10% FBS RPMI 1640 medium containing 0.2% agarose and overlaid onto 24-well plates containing a solidified bottom layer of 0.3% agarose (600 l) in 10% FBS RPMI 1640 medium. Once the top layer solidified, 600 l of 10% FBS RPMI 1640 medium was placed on the top to keep the wells moist. For MBO-asGCS treatment, the cells were pretreated with MBO-asGCS as described above. In addition, the MBO-asGCS (200 nM) was added with the 0.2% agarose medium (top layer). The plates were incubated for 1 week until the colonies were visible. The cell numbers of colonies were determined by the measurement of luminescent ATP in a Synergy HT microplate reader, following incubation with CellTiter-Glo reagent for 30 min, at room temperature with gentle shaking.
Flow Cytometry-Flow cytometry was performed as described previously (15). For analysis of BCSCs, the CD24 Ϫ cells sorted from each cell line were incubated with Alexa Fluor 488 anti-CD44 mouse antibody (5 l/10 6 cells; BioLegend, San Diego, CA) and Alexa Fluor 647 anti-ESA antibody (5 l/10 6 cells; BioLegend) in blocking buffer (PBS containing 5% serum) at 4°C for 30 min. After washing off unbound antibodies, the cells were resuspended in 1 ml of PBS and immediately analyzed on a BD FACSCalibur with the BD CellQuest Pro program (BD Bioscience, San Jose, CA) and FlowJo v10 (Tree Star, Ashland, OR). To identify CD44 ϩ /ESA ϩ cells, each sample was incubated in RPMI medium containing serum to determine the autofluorescence, as negative control. To analyze BCSCs in tumors, cell suspensions were prepared immediately after resection (Ͻ10 min). Tumor tissues (ϳ60 mg) were cut into small pieces (Ͻ1 mm) and incubated with collagenase IV (500 units/ml; purchased from Sigma) in RPMI 1640 and incubated at 37°C for 2 h with shaking (100 rpm). Further, the cells were passed through a 70-m cell strainer and washed twice with PBS.
The endogenous GSL were extracted and analyzed as described previously (25,42). Briefly, cellular lipids were extracted with chloroform/methanol/water (1:1:1, v/v/v) from each cell line or subpopulation after vehicle or MBO-asGCS treatments. Extracted lipids were resuspended in chloroform/ methanol (1:1, v/v) and applied to Partisil HPTLC plates with fluorescent indicator. Lipids were resolved using the solvent systems of chloroform/methanol/water (65:35:8, v/v/v) for GSLs. HPTLC plates were dipped for 10 s in 0.02% primuline (w/v; purchased from Sigma) in acetone/water (4:1, v/v). Fluorescence TLC profile graph was visualized under long wave UV light (360 nm) and captured using AlphaImager HP system (Alpha Innotech). Neutral GSL Qualimix (Matreya) were used as TLC standards. Globotriosylceramide (Gb3) of each sample was quantified by fluorescence intensity against the standard curve established using Gb3 lipids (Matreya) and normalized against cellular proteins.

MALDI-MS and MS Analysis of GSLs-MALDI-MS
profiling of permethylated GSLs was performed as described previously (43). Briefly, the extracted GSLs (from 7.5 ϫ 10 5 cells) were incubated with leech ceramide glycanase (50 milliunits/100 l) (44) in 50 mM sodium acetate, pH 5.5, containing 0.1% sodium cholate at 37°C for 16 h. The released glycans were separated from the ceramides and detergent by passing through a Sep-Pak C18 cartridge (Waters, Milford, MA), and further desalted by a porous graphitized carbon solid phase extraction cartridge. Permethylation was carried out by the NaOH/Me 2 SO slurry method as described previously (45). MALDI-MS analysis of the derivatives were carried out in positive ion mode on a ABI 4700 MALDI-TOF/TOF mass spectrometer (Applied Biosystems, Foster City, CA) using 2Ј,4Ј,6Ј-trihydroxyacetophenone as the MALDI matrix. Gb3 mass intensity was normalized against cellular proteins of each sample.
Western Blot Analysis-After treatment, cells or tissue homogenates were lysed using Nonidet P-40 cell lysis buffer (BIOSOURCE, Camarillo, CA). The amount of total proteins was measured using a BCA protein assay kit (Pierce). Equal amounts of detergent-soluble proteins (50 g/lane) were resolved using 4 -20% gradient SDS-PAGE (Invitrogen). After transferring, the blots were blocked in 5% fat-free milk in PBS for 1 h at 26°C. The blots were then incubated with specific primary antibodies against GCS, FGF-2, Oct-4, CD44, c-Src, phosphorylated c-Src, ␤-catenin, and phosphorylated ␤-catenin (1:200 to 1:5000 dilution) overnight at 4°C and then with respective horseradish peroxide-conjugated secondary antibodies (1:2500 dilution) for 1 h at 26°C after washing. The proteins were detected using enzyme-linked SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific, Rockford, IL) as described previously (25,30). Endogenous GAPDH was used as a loading control for each sample.
Tumor-bearing Mice and Treatments-All of the animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Louisiana at Monroe and were handled in strict accordance with good animal practice as defined by National Institutes of Health guidelines. Athymic nude mice (Foxn1 nu /Foxn1 ϩ , 4 -5 weeks, female) were purchased from Harlan (Indianapolis, IN) and maintained in the vivarium at University of Louisiana at Monroe. Animal studies were conducted as described previously (25,36). Briefly, cells of each subpopulation (BCSC, CD44 Ϫ /CD24 Ϫ ) were cultured 24 h after sorting and resuspended in serum-free RPMI 1640 medium (20 l). After anesthesia, cell suspensions (2.5 ϫ 10 4 cells/20 l or 1 ϫ 10 5 cells/20 l) were inoculated into the second mammary gland fat pad just beneath the nipple of each mouse implanted with 17␤-estradiol tablets (0.72 mg, 90 days release; Innovative Research of America, Sarasota, FL) to generate orthotopic breast tumors. The mice were monitored by measuring tumor volume, body weight, and clinical observation. Tumor volume (V) was calculated by V ϭ L/2 ϫ W 2 , where L was the length, and W was the width of tumors. Tumors and metastasis of mice were examined by a pathologist with hematoxylin and eosin stain of tissues sections at Louisiana State University Health Science Center (Shreveport, LA).
Statistical Analysis-All of the experiments were conducted in triplicate and repeated two times. The data were analyzed by using Prism version 4 (GraphPad software, San Diego, CA) or SAS 9.2 (SAS Institute, Gary, NC) and presented as the means Ϯ S.D. Two-tailed Student's t tests were used to compare the continuous variables between groups, and Fisher's extract test was used to compare the proportion between groups. All p Ͻ 0.05 was considered statistically significant.

GCS Is Associated with BCSCs in Drug Resistance and
Tumorigenicity of Cancer Cells-Different cancer cell lines display their own behavior in response to anticancer drugs and in tumor formation. In assessment of cell response to doxorubicin, we found that the half-maximal inhibitory concentration (IC 50 ) value for doxorubicin in MCF-7/Dox cells was 30-fold (9.8 M versus 0.33 M; p Ͻ 0.001) higher than the MCF-7 breast adenocarcinoma cells or human MCF-12A mammary epithelial cells (Fig. 1A). These results are consistent with previous reports, because MCF-7/Dox cells, which have been selected with doxorubicin culture from MCF-7 cells, are resistant to doxorubicin (35). Furthermore, MBO-asGCS treatment (100 nM) to silence GCS expression (36) in the present study significantly decreased the IC 50 for doxorubicin by 13-fold (9.8 M versus 0.75 M; p Ͻ 0.001) in MCF-7/Dox cells (Fig. 1A). Examining colony formation in soft agar, we found that MCF-7/Dox cells formed 160% more colonies (9,325 cells versus 5,715 cells; p Ͻ 0.001) than MCF-7 cells or MCF-12A cells (Fig. 1B); in contrast, MBO-asGCS treatment reduced the colonies by 2-fold (9,315 cells versus 4,702 cells; p Ͻ 0.001) in MCF-7/Dox cells (Fig. 1B). These results show that GCS expression is associated with drug resistance and tumorigenicity, which are characteristics of CSCs in these cancer cell lines.
Ceramide Glycosylation Determines Stem Properties of BCSCs in Tumor Formation and Metastasis-To determine whether or not BCSCs rely on ceramide glycosylation, we sorted BCSCs and other subpopulations from MCF-7/Dox cells and examined their tumorigenicity and ceramide glycosylation. Consistent with previous reports (6), it was found that the clonogenicity of BCSCs (CD44 ϩ /ESA ϩ /CD24 Ϫ ) was 3-fold greater (19,540 cells versus 5,717 cells; p Ͻ 0.001) than the CD44 Ϫ /CD24 Ϫ subset and significantly higher than other nonstem cell subsets (CD44 ϩ /CD24 Ϫ /ESA Ϫ , CD44 ϩ /CD24 ϩ , and CD44 Ϫ /CD24 Ϫ ) (data not shown here). Conversely, silencing GCS by MBO-asGCS reduced the number of colonies to 45% (8,724 cells versus 19,540 cells; p Ͻ 0.001) in BCSCs (Fig. 5A). These results further were confirmed by tumor sphere assay (Fig. 5A, bottom panels). Furthermore, the tumorigenicity of BCSCs and CD44 Ϫ /CD24 Ϫ cells were examined in athymic nude mice. It was found that BCSCs formed significantly more tumors, and the tumors were bigger than the non-stem cell subset (CD44 Ϫ /CD24 Ϫ ) ( Table 1 and Fig. 5B). The tumor incident detected in the BCSC group (1 ϫ 10 5 cells/mouse inoculation; 12 of 12 mice) is statistically significant, 6-fold higher (p Ͻ 0.0071) than CD44 ϩ /CD24 Ϫ group (one of six mice). Tumors from BCSCs grew aggressively and were 6-fold larger than the non-stem cell subset (1,858 versus 274 mm 3 ; p Ͻ 0.001) (Fig. 5B). Moreover, lung metastases were detected in mice injected with BCSCs (two of 12 mice), but not in mice inoculated with CD44 Ϫ /CD24 Ϫ cells, as indicated in the hematoxylin and eosin staining of lung section (Fig. 5B, bottom panels). However, the difference of metastasis incidence was not significant (p Ͻ 0.99) between the two groups, because of small sample size.

DISCUSSION
Our results clearly indicate that ceramide glycosylation catalyzed by GCS is important for CSCs in drug resistance and tumorigenesis. BCSCs have been identified with surface markers, such as CD44 ϩ /ESA ϩ /CD24 Ϫ , in established breast cancer cell lines and in tumor specimens (3,5,6,46). In addition to tumorigenesis, the accumulated BCSCs in tumors drive tumor progression to disseminated metastasis and poor response to chemotherapy (8,10,58). Among human breast cancer cell lines, the tumorigenesis and drug resistance are reported highly associated with the numbers of BCSCs (5,6,49). In the present study, MCF-7/Dox cells that have 3-fold more BCSCs displayed marked resistance to doxorubicin and were aggressive in tumorigenesis (Figs. 1, 2, and 4). Despite GCS overexpression in metastatic breast tumors and causing drug resistance (31,33), there is no clear experimental evidence linking these to CSCs. Ceramide glycosylation by GCS found in the present study correlated well with BCSCs, as well as their tumorous property in drug resistance and tumorigenicity. MCF-7/Dox cells overexpressed GCS protein and produced 2-fold more Gb3 following enhanced ceramide glycosylation, as compared with MCF-7 breast cancer cells. In contrast, the levels of GCS protein and Gb3, which were consistent with the BCSC numbers, were substantially lower in normal MCF-12A mammary epithelial cells (Fig. 3). Disruption of ceramide glycosylation by silencing of GCS decreased BCSCs in cultured cells and in tumors generated from MCF-7/Dox cells (Figs. 2 and 4); consequently, silencing of GCS reversed drug resistance and eliminated tumorigenesis. These results together indicate that enhanced ceramide glycosylation caused by GCS overexpression is essential for BCSCs retaining drug resistance and tumorigenicity.
Ceramide glycosylation by GCS is a unique process by which CSCs maintain their tumorous stemness. Ceramide glycosylation, which is known for reducing cellular ceramide biochemically and protecting cell from ceramide-induced apoptosis, can endow CSC resistance to cytotoxins and anticancer drugs. It has been reported that CSCs (CD55 hi ) have high tolerance to ceramide-induced apoptosis, and this resistance is attributed to abundance of sphingomyelin synthase 1 that can converts ceramide to sphingomyelin (67). Because GCS level is higher in BCSCs versus lower in bone marrow stem cells, doxorubicin can enhance BCSC number and decrease the number of bone marrow stem cells (68). Numerous studies show that GSLs, such as Gb5 and MSGb5, are markers of stem cells; however, no cutting evidence links their expression to a role in pluripotency, and their molecular functions remain unclear. Gb5 and MSGb5, as markers of human ESCs, disappear with differentiation (64,69). Mass spectrometry analysis that can distinguish glycans shows a switching of the core structures of glycosphingolipids from globo-series and lacto-series to ganglio-series when human ESCs differentiate into embryoid body outgrowth with three germ layers (69). Recent study shows that ganglioside GD2 can be used as a marker to identify CSCs (GD2 ϩ ) from breast cancer cell lines and patient samples, and interference with GD3 synthase can reduce CSC population and CSC-associated properties (70). The present study found GCS and Gb3 overproduced in BCSCs, not normal adult stem cells, such as mammary epithelial stem cells (Fig. 3) and bone marrow stem cells reported by another study (68). It is not clear by which molecular mechanism CSCs regain ability in ceramide glycosylation that exists in ESCs. We now know that ceramide generated in cells exposed to doxorubicin can activate GCS promoter and up-regulate its expression (38).
A major contribution of the present work is that we identified the role of Gb3 in maintaining stem properties of BCSCs. It has been reported that Gb3 (CD77), a precursor of Gb5 is associ-ated with Src family Yes kinase on GSL-enriched microdomains (25,71). Gb3 can modulate c-Src kinase in GSL-enriched microdomains and up-regulate mdr1 expression through ␤-catenin signaling (25). Wnt/␤-catenin signaling maintains hematopoietic stem cells and neuronal stem cells and the epithelial to mesenchymal transition for BCSCs (72); Wnt inhibitors can reduce the size of tumor spheres and self-renewal in prostate cancer stem cells (73). FGF-2 is critical for the selfrenewal and maintenance of normal and cancer stem cells (47,48), and Oct-4 is a transcription factor that is involved in maintaining the pluripotency of ESCs (74) and BCSCs (59,75). This study has found that the levels of GCS, Gb3, phosphorylated c-Src, ␤-catenin, FGF-2, and Oct-4 are substantially higher in BCSCs (Fig. 7). Stepwise inhibition of GCS, c-Src kinases, and ␤-catenin indicates that these molecules are critical for the maintenance of BCSCs (Fig. 7). Although stem cell factors including ␤-catenin, FGF-2, Oct-4, and CD44 are involved in regulating stemness of either normal stem cells or CSCs, this study demonstrates that GSLs in BCSCs activate c-Src/␤catenin signaling and up-regulate these factors to maintain BCSCs selectively. In addition to insight regulating stemness of stem cells, this finding opens the possibility of disrupting ceramide glycosylation or GSL synthesis to target CSCs specifically.