Blizard Institute Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London United Kingdom
BDS, FDSRCS, Ph.D., Blizard Institute, Barts and The London School of Medicine and Dentistry, Queen Mary University of London, 4 Newark Street, London E1 2AT, U.K. E-mail: firstname.lastname@example.org Telephone: 20-7882-7159; Fax: 20-7882-7172
Author contributions: H.S.: conception and design, financial support, administrative support, collection and/or assembly of data, data analysis and interpretation, manuscript writing, and final approval of manuscript; A.B. and L.G.: conception and design, data analysis and interpretation, manuscript writing, and final approval of manuscript; H.E., C.O.R., E.G., and B.F.: data analysis and interpretation and final approval of manuscript; M.S. and N.K.: conception and design, financial support, administrative support, and final approval of manuscript; I.C.M.: conception and design, financial support, data analysis and interpretation, manuscript writing, and final approval of manuscript.
Cells sorted from head and neck cancers on the basis of their high expression of CD44 have high potency for tumor initiation. These cells are also involved in epithelial to mesenchymal transition (EMT) and we have previously reported that cancer stem cells (CSCs) exist as two biologically distinct phenotypes. Both phenotypes are CD44high but one is also ESAhigh and maintains epithelial characteristics, the other is ESAlow, has mesenchymal characteristics and is migratory. Examining CD44-regulated signal pathways in these cells we show that CD44, and also RHAMM, act to inhibit phosphorylation of glycogen synthase kinase 3β (GSK3β). We show that inhibitory phosphorylation reduces the formation of both “tumor spheres” and “holoclone” colonies, functional indicators of stemness. GSK3β inhibition also reduces the expression of stem cell markers such as Oct4, Sox2, and Nanog and upregulates expression of the differentiation markers Calgranulin B and Involucrin in the CD44high/ESAhigh cell fraction. Transition of CSCs out of EMT and back to the epithelial CSC phenotype is induced by GSK3β knockdown. These results indicate that GSK3β plays a central role in determining and maintaining the phenotypes and behavior of CSCs in vitro and are likely to be involved in controlling the growth and spread of tumors in vivo.
Various reports indicate that many cancers contain a subpopulation of cells that is endowed with the stem cell properties of self-renewal and tumor-initiating capacity [[1, 2]]. Based on their high levels of expression of CD44, enriched populations of these cancer stem cells (CSCs) have been isolated and enriched from a range of solid tumors, including head and neck squamous cell carcinomas (HNSCC) [[3-7]]. Cells that have high expression of CD44 and show stem cell properties are also present in cell lines derived from HNSCC and other malignancies [[8-10]]. CD44 is a multifunctional and ubiquitously expressed glycoprotein adhesion molecule derived from a gene with 18 exons, 9 of which are expressed in the standard form (CD44s) with alternative splicing of the remainder generating a great many variant (CD44v) isoforms . CD44 expression potentially influences stem cell behavior by a wide range of mechanisms and interactions with hyaluronan, its principal ligand, activates several signaling pathways influencing tumor growth, motility, and metastasis [[12, 13]]. Although CD44 provides a consistent marker for some CSCs, the functional significance of its expression, and particularly of its roles in CSC self-renewal and differentiation, remain uncertain. Like CD44, RHAMM (receptor for hyaluronic acid-mediated motility) binds hyaluronan [[14, 15]]. Oncogenic expression of RHAMM has been reported for HNSCC and it has also been implicated in promoting proliferation of tumor cells . One of the actions of RHAMM is to co-operate with CD44 in forming complexes that coordinately activate the MAP/ERK1,2 pathway . Such complexes sustain high motility of breast cancer cells but, while RHAMM acts partly as a nonintegral cell surface hyaluronan receptor, it is also found as an intracellular protein that binds to mitotic spindles and has less certain functions .
Several reports have indicated a role for epithelial to mesenchymal transition (EMT) in metastasis [[19, 20]] and in generating cells that express marker patterns characterizing breast CSCs  and show enhanced resistance to therapeutic killing . We have reported that these phenomena are related to the presence of two biologically distinct CSC phenotypes both of which have high levels of expression of CD44 but differ in their levels of expression of epithelial specific antigen (ESA) . One phenotype has a CD44high/ESAhigh marker pattern, is proliferative, forms cohesive colonies in vitro, and has epithelial characteristics. The other has a CD44high/ESAlow marker pattern, is migratory, has mesenchymal characteristics, and a pattern of gene expression indicative of EMT. Brabletz  has suggested a role for EMT in metastasis during which CSCs undergo EMT to escape from the parent tumor, invade locally, and then migrate to distant sites where they undergo mesenchymal to epithelial transition (MET) to generate secondary tumors . Although CSCs switch both in and out of the EMT phenotype, only CD44high/ESAlow cells that also express aldehyde dehydrogenase 1 (ALDH1) are able to switch back to the CD44high/ESAhigh phenotype to reconstitute the cellular heterogeneity typical of the original tumor . Interestingly, in head and neck cancers, only CD44high cells that also express ALDH are involved in EMT and show high potency for tumor initiation [[23, 24]].
HNSCC cell lines plated at low density form a range of heterogeneous colony morphologies that correspond to the holoclone, meroclone, and paraclone colonies that result from the presence of hierarchies of cells at different stage of maturation [[10, 25]]. After single cell cloning, the small, tightly packed cells of holoclones have a persistently high proliferative potential, corresponding to stem cell self-renewal . It has also been reported that self-renewing stem cells have the ability to form tumor spheres under suspension culture conditions  but we previously reported that, in HNSCC, the ability to form tumor spheres in suspension cultures is associated with EMT CSC whereas the ability to form holoclones under adherent condition is mainly a property of epithelial CSC .
Glycogen synthase kinase 3β (GSK3β) is known to regulate cell cycle progression and cell proliferation . Several oncogenic signaling pathways, for example, Wnt/AKT, MAP kinase, and phosphoinositide 3-kinase (PI3K) pathways, act to inhibit the activity of GSK3β [[28, 29]]. AKT-mediated inactivation of GSK3β leads to nuclear translocation of β-catenin , and inactivation of GSK3β is thought to drive oncogenic progression in oral SCC through an accelerated cell cycle progression and enhanced tumor invasion and metastasis . However, β-catenin becomes vulnerable to destruction if E-cadherin is lost via EMT , and degradation of membranous β-catenin is necessary for the invasion and metastasis of oral SCC . These recent observations suggest that GSK3β remains active in CSCs that have undergone EMT and imply an oncogenic role of GSK3 acting through the EMT CSC phenotype. In this study, we investigated the influences of GSK3β on the self-renewal, switching, and differentiation of CSCs and report a GSK3β-based mechanism that influences these events.
Materials and Methods
The CA1 and LUC4 cell lines were established from oral SCCs , and the MET2 line was established from a cutaneous SCC . All cell lines were grown in a highly supplemented epithelial growth medium (termed FAD) with 10% FBS under 5% CO2 in air at 37°C. For replating and for assays, cells were released from flasks using Accutase (PAA Laboratories, Linz, Austria, http://www.paa.at).
Colony and Sphere Formation Assays
To test the ability of cells to form holoclone colonies , 5 × 102 cells in 0.5 mL of medium were added to each well of 24-well plates, and the number of holoclones was counted 7 days after cell plating. For growth as tumor spheres in suspension cultures, 0.75 cm2 wells were coated with 12 mg/mL PolyHEMA (2-hydroxyethyl methacrylate, Sigma) in 95% ethanol prior to seeding cells at a density of 2 × 103 cells per well in 0.5 mL medium with addition of 1% methylcellulose (Sigma) to prevent cell aggregation. After 2 weeks, plates were assayed visually for the formation of floating tumor spheres, and the number of spheres was counted. Results were expressed as the mean ± SD for more than three independent cultures.
In Vitro Scratch Assay
To assess cell motility, scratch assays were performed as described previously . A scratch was made on the confluent cell layers using 200 μL micropipette tip and the wound monolayer was washed to remove the dislodged cells. The images of the wounded area were captured at 0 hour and 24 hours by phase contrast microscopy and a digital camera. Adobe Photoshop Elements version 4.0 software was used to quantify migration of cells by subtracting the area of the scratch remaining at 24 hours from the original scratch area which provided the percentage of scratch occupied by newly migrated cells as an index of migration.
Fluorescence-Activated Cell Sorting
Fluorescence-activated cell sorting (FACS) analyses used an anti-CD44-phycoerythrin (PE)-conjugated antibody, an anti-ESA-allophycocyanin (APC)-conjugated antibody (both from BD Biosciences, San Diego, CA, http://www.bdbiosciences.com), and an anti-human RHAMM (CD168) rabbit monoclonal antibody (OriGene Technologies, 1:150). An anti-human phosphorylated-GSK3β rabbit polyclonal, which detects inactivating phosphorylation at ser9, was used to detect inactivated GSK3β (1:200, Cell Signaling Technology, Beverly, MA, http://www.cellsignal.com). An anti-rabbit fluorescein isothiocyanate (FITC)-conjugated antibody (1:1,000, BD Biosciences) was used as a secondary antibody at dilution. Diamidophenylindole (DAPI) was used at a concentration of 1 μg/mL to exclude dead cells. Samples were assayed on a Becton Dickenson LSRII FACS (Oxford, U.K.) and sorted using Becton Dickenson FACSAria equipment. FACS Diva version 6.1.1. (BD Biosciences) software was used to analyze the data. Intensities of FITC fluorescent signals for RHAMM and phosphorylated-GSK3β were measured using FACS Diva version 6.1.1 software (BD Biosciences) and shown as mean ±SEM. For multiple staining procedures, cells were fixed with 4% formaldehyde in phosphate buffered saline (PBS), stained for CD168 (RHAMM) for 1 hour, washed and then stained with an anti-rabbit secondary antibody (Alexafluor 488) at 1:1,000 for 30 minutes. Anti-CD44 and ESA antibodies were then added.
For CD44v staining, cells were detached using enzyme-free cell dissociation buffer (Invitrogen, Carlsbad, CA, http://www.invitrogen.com). They were then stained with antibodies at 1:100 dilution in PBS (PAA). The DAPI nuclear dye (Sigma) was used at 1 μg/mL to exclude dead cells. The antibodies used were as follows: PE-CD44 (clone G44-26) was from BD Biosciences; APC-ESA (clone HEA-125) was from Miltenyi Biotec (Bergisch Gladbach, Germany, http://www.miltenyibiotec.com); PE-CD44v3 (clone 3G5) was from R&D Systems (Minneapolis, MN, http://www.rndsystems.com); FITC-CD44v6 (clone VFF7) was from Bender Medsystems (Vienna, Austria, http://www.biolab21.com); CD44v5 (clone VFF8) was from AbD Serotec (Kidlington, UK, http://www.abdserotec.com/); the FITC rabbit-anti-mouse secondary antibody was from Invitrogen (Paisley, UK, http://www.invitrogen.com/). The FITC- and PE-conjugated mouse IgG isotype control antibodies were from BD Biosciences.
RNA was extracted by the RNAeasy micro kit (Qiagen, Hilden, Germany, http://www1.qiagen.com) and reverse transcription into cDNA was conducted using the superscript III first strand synthesis supermix (Invitrogen, Paisley UK, http://www.invitrogen.com). For quantitative polymerase chain reaction (qPCR) of Sox2, Nanog, Oct4, Calgranulin B, Involucrin, and G3PDH the quantification of mRNA levels was carried out using the ABI 7500 real-time PCR system (Applied Biosystems, Warrington, U.K. http://www.appliedbiosystems.com) and Power SYBR green mix (Applied Biosystems). The reaction mixture contained 1.0 μg of cDNA, 12.5 μL of SYBR Green Mix, and 10 μmol of each pair of oligonucleotide primers. GAPDH was used as a reference mRNA control. The PCR program was as follows: initial melting at 95°C for 10 minutes followed by 40 cycles at 95°C for 15 seconds, 60°C for 30 seconds, and 72°C for 40 seconds. Reverse transcribed Human Total Reference RNA (Stratagene, Cheshire, U.K., http://www.stratagene.com) was used to make a standard curve. PCR for the standard and variant forms of CD44 was performed essentially according to Rajarajan et al.  combining a CD44 standard sense primer with an appropriate exon-specific antisense primer. The reaction mixture contained 1.0 μg of cDNA, 2.0 μL of 10× buffer (Toyobo, Japan, http://www.toyobobiologics.com), 2 mM dNTPs, 10 μM of each primer, and 2.5 units/μL of Taq DNA polymerase (Toyobo, Japan). The PCR program was as follows: initial melting at 95°C for 2 minutes followed by 35 cycles at 95°C for 1 minute, 57°C for 1 minute, and 72°C for 2 minutes. PCR products were separated by 1.5% agarose gel electrophoresis. DNA ladder (100 bp DNA Ladder, Toyobo, Japan) was used for the PCR marker. Primer details are provided in the Supporting Information, and the results are expressed as the mean ± SD for three independent experiments.
The cells were lysed in RIPA buffer [2.5 mM Tris pH 7.3, 152 mM NaCl, 0.0005% SDS, 1% Nonidet P40, CompleteMini protease inhibitor (Roche Diagnostic Ltd., Burgess Hill, U.K., http://www.roche-applied-science.com)]. Protein concentrations were measured using a protein assay reagent (Bio-Rad, Hercules, CA, http://www.bio-rad.com). Protein samples (15 μg) were solubilized in sample buffer by boiling and then run in a 10% polyacrylamide gel and blotted onto a nitrocellulose. The bands on Western blotting were detected using an enhanced chemiluminescence Western blotting reagent (GE Healthcare, Amersham, UK., http://www3.gehealthcare.co.uk). The antibodies, all used at 1:1,000 dilution, were an anti-human CD44 mouse monoclonal anti-body, an anti-human GSK3β mouse monoclonal antibody, and an anti-human phosphorylated GSK3β (Ser9) rabbit polyclonal antibody (all from Cell Signaling Technology, Hitchin UK, http://www.neb.uk.com), an anti-human RHAMM (CD168) rabbit monoclonal anti-body (OriGene Technologies. Cambridge UK, http://www.origene.com), an anti-human extracellular regulated kinase (ERK1/2) rabbit polyclonal antibody, an anti-human phosphorylated-ERK1/2 mouse monoclonal antibody, and anti-human GAPDH mouse monoclonal antibody (all from Abcam, Cambridge, U.K., http://www.abcam.com). Also used were an anti-human Vimentin mouse monoclonal antibody (BD Pharmingen, San Diego, http://www.bdbiosciences.com) and an anti-human E-cadherin rabbit polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA, http://www.scbt.com). GAPDH was used as a loading control in all blots.
Inhibition of GSK3β CD44 and RHAMM Functions
To inhibit GSK3β, cells were sorted as needed into subpopulations and were replated in culture medium containing the inhibitor N-(4-methoxybenzyl)-N′-(5-nitro-1,3-thiazol-2-yl) urea (Calbiochem 361549) at a concentration of 200 nM. Stealth siRNAs were used for RHAMM and GSK3β knockdown (Invitrogen). siGLO RISC-Free Control siRNA (Thermo Fisher Scientific Inc. Loughborough UK, http://www.fisher.co.uk/) was used as control. CD44 proteins were knocked down using siRNA ON-TARGET-plus SMARTpool siRNA for human CD44 (Thermo Fisher Scientific Inc. Loughborough UK, http://www.fisher.co.uk/). Cells were transiently transfected with the indicated combinations of the siRNAs using HiPerFect transfection reagent (Qiagen. Manchester, UK. http://www.qiagen.com), according to the manufacturer's recommendations.
Immunofluorescence was performed as described previously [[16, 25]]. Cells were seeded onto glass Lab-Tek II Chamber Slides (Thermo Fisher Scientific Inc. Loughborough UK, http://www.fisher.co.uk/) and incubated for a day. The growth medium was then removed, and cell monolayers were washed three times with a 1% PBS solution and fixed with 3.5% paraformaldehyde for 10 minutes at room temperature. Cells were washed three times with PBS and permeabilized by Triton X-100 (0.2%, 10 minutes at room temperature). Nonspecific binding sites were blocked by treatment at room temperature for 30 minutes with PBS containing 1% BSA. The cells were washed three times with PBS before incubating with the following antibodies: (a) anti-RHAMM rabbit monoclonal antibody (1:200, OriGene Technologies. Cambridge UK, http://www.origene.com), (b) anti-GSK3β mouse monoclonal antibody (1:200, Cell Signaling Technology, hitchin UK, http://www.cellsignal.com/), and (c) anti-phosphorylated GSK3β rabbit polyclonal antibody (1:200, Cell Signaling Technology). Staining for RHAMM and phosphorylated GSK3β used an Alexa-Fluor-labeled goat anti-rabbit or anti-mouse secondary antibodies (Invitrogen, Paisley UK, http://www.invitrogen.com). Cells were counterstained with DAPI for visualization of nuclear morphology.
At least three independent samples were available for all data points, and statistical analyses were performed using a paired Student's t test or a Welch's t test for samples with unequal variance. p values less than .05 were regarded as statistically significant.
CD44high/ESAhigh and CD44high/ESAlow Phenotypes are Present in SCC Cell Lines
SCC cells were fractionated by FACS using antibodies against CD44 and ESA (Fig. 1A; Supporting Information Fig. S1A) and, as previously reported, CSCs include two biologically distinct phenotypes . Each of the cell lines examined contained cells with high expression of CD44 that were either (a) ESAlow and exhibited a spindle-like appearance, had high expression of Snail, Vimentin, and Axl and low expression of E-cadherin (Fig. 1B–1D; Supporting Information Fig. S1B, S1C) or (b) ESAhigh and formed holoclones, had high expression of E-cadherin, low expression of Snail and Vimentin, and grew faster in adherent culture conditions than CD44high/ESAlow cells (Fig. 1C, 1D, 1F; Supporting Information Fig. S1C). The CD44high/ESAlow cell type had mesenchymal features, generated more motile cells in “scratch” assays (Fig. 1E; Supporting Information Fig. S1D), and was designated as “EMT CSC.” The CD44high/ESAhigh phenotype preserved epithelial characteristics and was designated as “EPI CSC.” The CD44low cells formed only paraclone-like cells and were unable to form self-renewing holoclones or grow extensively (Fig. 1B; Supporting Information Fig. S1B).
GSK3β Is Necessary for the Self-Renewal of CSCs
Examining the number of tumor spheres and holoclones formed by the EMT and EPI CSCs of the CA1, MET2, and LUC4 cell lines, we found that significantly greater numbers of tumor spheres were formed by EMT CSCs, and of holoclones by EPI CSCs. The number of tumor spheres formed by the EMT CSCs (Fig. 2A) appeared to correlate with the size of the fraction of EMT CSCs present in the unsorted parental population (7.5%, 16.5%, and 21.0% in CA1, MET2, and LUC4, respectively).
Western blotting to assess total GSK3β and phosphorylated GSK3β indicated a relationship between the self-renewal ability of CSCs and GSK3β activity. In the high sphere-forming MET2 and LUC4 lines, the CD44high/ESAlow cells had low levels of phosphorylated GSK3β whereas in the low sphere-forming CA1 line the CD44high/ESAlow cells exhibited higher expression of phosphorylated GSK3β (Fig. 2B). FACS analyses of EMT CSCs, EPI CSCs, and CD44low cells for their levels of phosphorylated GSK3β used a triple combination of antibodies against CD44, ESA, and phosphorylated GSK3β. These showed that within each line, the EMT CSCs had the lowest expression of phosphorylated GSK3β and CD44low cells the highest (Supporting Information Fig. 2A). Western blots showed high levels of CD44 and low levels of phosphorylated GSK3β in EMT cells (Fig. 2B). Quantitative analyses indicated that these differences were significant and that sphere forming ability correlated negatively with levels of phosphorylated GSK3β (Supporting Information Fig. S2B, S2C). Given that GSK3β is inactivated by phosphorylation at ser9, it appears that tumor sphere formation in suspension culture, a property indicative of maintenance of a stem cell state, is correlated with highly active GSK3β.
To clarify potential roles of GSK3β in the self-renewal of CSCs, the formation of tumor spheres and of holoclones was examined after GSK3β inactivation using either chemical inhibition or siRNA knockdown. After GSK3β inactivation, no significant changes in the overall growth rates of CD44high/ESAlow or CD44high/ESAhigh cells were observed after 7 days of adherent culture (Supporting Information Fig. S3A). However, inactivation of GSK3β clearly decreased tumor sphere formation by CD44high/ESAlow cell fractions (Fig. 2C, 2D; Supporting Information Fig. S3B–S3D). After inhibition of GSK3β, the number of holoclones formed by CD44high/ESAhigh cells decreased for each of the LUC4, MET2, and CA1 cell lines indicating their loss of self-renewal abilities (Fig. 2E; Supporting Information Fig. S3E, S3F). Collectively, these results indicate that active GSK3β is necessary for the self-renewal of both EMT and EPI CSCs.
Self-Renewing CD44High/ESALow Cells Show Less GSK3β Phosphorylation
Immunofluorescence indicated the uniform cytoplasmic presence of GSK3β in nearly all CD44high/ESAlow and CD44high/ESAhigh cells (Fig. 2F). However, staining for phosphorylated GSK3β showed the lowest expression in CD44high/ESAlow cells, higher expression in CD44high/ESAhigh cells, and the highest expression in CD44low cells (Fig. 2F). Nuclear staining for GSK3β was high in ESAlow cells, intermediate in ESAhigh cells, and lowest in CD44low cells but nuclear staining for phosphorylated GSK3β was generally absent (Fig. 2G). Very few of the cells present in tumor spheres showed expression of phosphorylated GSK3β (Fig. 2G). These observations are consistent with the results of Western blotting for phosphorylated GSK3β (Fig. 2B) and provide further support for the notion that nonphosphorylated GSK3β (i.e., active GSK3β) plays a central role in maintaining the self-renewing state of EMT and EPI CSCs.
CSC Expression of Stem Cell Markers Is Reduced by GSK3β Inhibition
To further clarify the role played by GSK3β in the maintenance of CSCs, mRNA expression of the stem cell markers Sox2, Oct4, and Nanog was evaluated by reverse transcriptase PCR (RT-PCR). Except for Sox2 in CD44high/ESAhigh cells, inactivation of GSK3β significantly reduced each of these markers in both CD44high/ESAlow and CD44high/ESAhigh cells (Fig. 3A, 3B). These findings further indicate a critical role played by GSK3β in the maintenance of the stem cell state of CSCs.
Inactivation of GSK3β Induces Cell Differentiation
To examine roles of GSK3β in maintaining the EMT and EPI CSC phenotypes, changes in the proportions of CD44high/ESAlow, CD44high/ESAhigh, and CD44low cells were examined after GSK3β inhibition or knockdown. For all cell lines, the percentage of both CD44high/ESAlow and CD44high/ESAhigh cells decreased, but the percentage of CD44low cells consistently increased (Fig. 3C; Supporting Information Fig. S4A). The increase in CD44low cells suggested a shift of CSCs into differentiation, a change that would be expected to increase expression of epithelial differentiation markers such as Involucrin and Calgranulin B. Inactivation of GSK3β produced significantly increased levels of expression of these genes in CD44high/ESAhigh cells (Fig. 3D, 3E) indicating that GSK3β acts to suppress entry of CSC into differentiation.
To examine more closely how knockdown of GSK3β influences the differentiation of EPI CSCs, CD44high/ESAhigh cells were sorted for single cell cloning, and clones derived from individual CD44high/ESAhigh cells were examined after 4 weeks of growth in culture. All clones produced CD44low populations (Supporting Information Fig. S5A). For all CD44high/ESAhigh clones, knockdown of GSK3β reduced the formation of holoclones (Supporting Information Fig. S5A). It also increased the size of the CD44low population identified by FACS analysis (Supporting Information Fig. S5B), supporting the concept that inhibition of GSK3β attenuates self-renewal and induces differentiation of EPI CSCs.
Inactivation of GSK3β Promotes MET of CD44High/ESALow Cells
Lack of induction of differentiation markers in CD44high/ESAlow cells after GSK3β inhibition suggested that EMT CSCs do not directly enter terminal differentiation. However, GSK3β inhibition was found to lead to a reduced percentage of EMT CSCs (Fig. 3C; Supporting Information Fig. S4A), less expression of Snail and Vimentin, and increased expression of E-cadherin (Supporting Information Fig. S4B). This indicated a shift of EMT CSCs into the CD44high/ESAhigh phenotype, suggesting that maintenance of cells in the CD44high/ESAlow cell compartment requires active GSK3β.
To examine the phenotypic plasticity of EMT CSCs in terms of their ability to switch back to the EPI CSC phenotype, CD44high/ESAlow cells were sorted for single cell cloning. Initially all developing clones formed cells with a spindle-like appearance. After 8 weeks of culture, the clonal populations were examined by FACS and this showed that of the 17 single cell clones examined, eight maintained an entirely CD44high/ESAlow identity without any cells switching into the EPI CSC phenotype (Fig. 4A). These were termed type 1 clones. The formation of tumor spheres by type 1 clones was significantly suppressed by knockdown of GSK3β. However, FACS analyses indicated that this was not associated with a switch into the EPI CSC phenotype (Fig. 4A, 4B). Nine of the 17 single cell clones derived from CD44high/ESAlow cells were able to switch and gave rise to both CD44high/ESAlow and CD44high/ESAhigh cell populations. These clones initially displayed a spindle-like appearance (Fig. 4C, left) but by 8 weeks of culture had visibly generated a mixture of both EMT and epithelial cells (Fig. 4C, right). Interestingly, for all of these clones, termed type 2 clones, inactivation of GSK3β greatly accelerated the transition into CD44high/ESAhigh cells and produced a marked shift toward the EPI CSC phenotype within 5 days (Fig. 4D). Inactivation of GSK3β in type 2 clones significantly decreased expression of Snail and Vimentin and upregulated expression of E-cadherin (Fig. 4E), findings consistent with loss of GSK3β activity promoting a switch of EMT CSCs back towards the EPI CSC phenotype.
CD44 and RHAMM Are Required for Self-Renewal of CSCs and Regulate GSK3β
CD44high/ESAhigh cells showed considerably higher expression of RHAMM than CD44high/ESAlow cells (Fig. 5A; Supporting Information Fig. S4A) and FACS analysis, using triple staining for CD44, ESA, and FITC-labeled RHAMM, confirmed that RHAMM was most highly expressed on CD44high/ESAhigh cells (Fig. 5A). Immunofluorescent staining confirmed expression of both RHAMM and CD44 in CD44high/ESAhigh holoclone cells (Fig. 5B). Both RHAMM and CD44 knockdown resulted in phosphorylation (inactivation) of GSK3β and phosphorylation (activation) of ERK1/2 (Fig. 5C). Holoclone formation by CD44high/ESAhigh cells was significantly inhibited by both RHAMM and CD44 knockdown (Fig. 5D). Knockdown of CD44 caused a significant decrease in the number of tumor spheres formed by CD44high/ESAlow cells but, in contrast, knockdown of RHAMM did not (Fig. 5E). CD44 and RHAMM knockdown also significantly decreased expression of Sox2, Nanog, and Oct4 in CD44high/ESAhigh cells (Fig. 5F) and it also upregulated expression of the differentiation markers Involucrin and Calgranulin B (Fig. 5G). These results support the notion that CD44 is required for self-renewal of both EPI and EMT CSCs, whereas RHAMM is required for self-renewal only of EPI CSCs. Both CD44 and RHAMM contribute toward maintenance of active GSK3, potentially through blocking the activation of ERK1/2, which is a known inhibitor of GSK3β .
Due to alternatively spliced exon products, CD44 exists as a standard form and as range of variant isoforms whose differential expression has been linked to the behavior of head and neck cancers and to EMT [[11, 13, 37]]. For the Ca1 and Met2 cell lines, EPI CSCs have higher expression of variant isoforms than EMT CSCs . Therefore, to assess the expression pattern for the LUC4 cell line used in this study, cell subpopulations were examined for the expression of standard and variant isoforms by flow cytometry and by qPCR. Cytometry of cells stained with antibodies against total CD44 and ESA showed similar levels of CD44 surface expression on both ESAhigh and ESAlow cells. However, staining for the CD44v3, v5, and v6 isoforms indicated markedly less expression of these isoforms on the ESAlow EMT cell fraction. qPCR for the v3, v4, v5, v6, and v7 isoforms similarly indicated their low expression in the CD44high/ESAlow EMT cell fraction compared with the EPI CSC, parental, or CD44low cell fractions, whereas the standard CD44 isoform was more highly expressed in the EMT and EPI CSC fractions than in the parental and CD44low fractions (Supporting Information Fig. S6).
The use of CD44 to identify a population of highly tumorigenic cells has been described for HNSCC as well as for breast, colon, prostate, and other cancers [[3, 7]]. However, the mechanisms maintaining the balance between self-renewal and differentiation of CD44high CSCs have been uncertain and recently this problem has been further complicated by work indicating the presence of two biologically distinct phenotypes of CSCs in HNSCC and breast cancers [[23, 38]]. In HNSCC, both CSC phenotypes are relatively CD44high but one is ESAhigh and the other ESAlow. Consequently, three distinct cell subpopulations can be identified: CSCs with an EMT phenotype, CSCs with an epithelial phenotype, and CD44low cells that have entered the differentiation pathway and lost self-renewal ability . In this study, we have begun to examine transitions occurring between these cell types and have focused on potential roles of GSK3β in mediating the choice between their self-renewal and differentiation. Assessment of sphere formation and clonogenicity after GSK3β inhibition demonstrated that GSK3β is required to maintain both CD44high/ESAlow and CD44high/ESAhigh cells in a self-renewing state. This finding is strengthened by the lack of significant levels of inactive (phosphorylated) GSK3β in either of the self-renewing sphere-forming or holoclone-forming populations. The few holoclones that continued to grow in the presence of GSK3β inhibitor showed lack of phosphorylation (Fig. 5B), and inactivation of GSK3β reduced the population of CD44high/ESAhigh cells and shifted cells into the CD44low compartment with loss of self-renewal ability.
Under various pathological conditions, upstream signal pathways such as PI3K/AKT, Raf/MEK/ERK, and Wnt induce phosphorylation and inactivation of GSK3β [[29, 30, 39]]. Our observations indicate that the preservation of functionally active GSK3β is required for CSC self-renewal and that this is promoted by signaling pathways initiated by CD44 and RHAMM. One action of CD44 is to inhibit phosphorylation of AKT, thus preventing AKT from phosphorylating GSK3β at Ser9 and inhibiting its activity [[40, 41]]. In oral SCC, RHAMM has an important role in promoting tumor proliferation  and, like CD44, RHAMM have been shown to act together in a hyaluronan-dependent autocrine mechanism to co-ordinate signaling that sustains cancer cell motility . Knockdown of CD44 induced phosphorylation of GSK3β and reduced the holoclone forming ability of EPI CSCs and the sphere forming ability of EMT CSCs. Phosphorylation of GSK3β induced by knockdown of RHAMM similarly reduced holoclone forming ability but had less effect on sphere formation. These results indicate that CD44 and RHAMM act upstream to prevent GSK3β phosphorylation and maintain activities necessary to promote CSC self-renewal. High levels of expression of CD44 are associated with stem cell properties such as self-renewal  and also commonly mark CSCs isolated both from fresh tumors and from malignant cell lines [[3-10]]. A switch from variant CD44 isoforms to the standard isoform appears necessary for EMT in breast cancer cells where the epithelial phenotype is maintained by the expression of epithelial splicing regulatory protein-1 which promotes alternative splicing . The low expression levels of variant CD44 isoforms by EMT CSCs of our oral cancer cell lines suggests that the standard CD44 isoform, which is expressed by both CSC phenotypes, may be required for the self-renewal of stem cell compartments while the variant isoforms are associated with maintenance the epithelial phenotype.
Brabletz  has proposed that invasion and metastasis are dependent both on EMT and on the reverse process of MET: whereas EMT initially provides cells with the ability to escape from a tumor and migrate to distant sites, MET is subsequently required to restore the epithelial nature of developing secondary tumors. A great deal is now known about EMT and about the cytokines and growth factors in the tumor environment that induce EMT [[20, 44, 45]]. However, much less is known about factors that induce MET [[46, 47]]. Some EMT CSCs have the ability to switch spontaneously into the EPI CSC phenotype  and GSK3β, in addition to its roles in self-renewal, also appears to influence this transition. When the high levels of active GSK3β normally present in CD44high/ESAlow cells are knocked down, cells that are capable of shifting into the CD44high/ESAhigh phenotype do so more rapidly. Signaling through GSK3β may thus be a key regulator of the MET shift that enables metastatic tumor cells to produce new tumors at secondary sites.
Figure 6 illustrates the proposed roles of active GSK3β in promoting the self-renewal of both EMT and EPI CSCs and the effects of loss of its activity on the MET of EMT CSCs and the differentiation of EPI CSCs. Typically, the bulk of tumor cells, both in vivo and in cell lines, is not self-renewing and consists of CD44 low differentiating cells . In the presence of active GSK3β, the EPI CSC fraction is self-renewing but with loss of GSK3β activity this fraction generates differentiating CD44low cells. The EPI CSC fraction is also able to undergo EMT to generate the EMT CSC fraction, a transition influenced by autocrine and paracrine actions of cytokines such as TGFβ [[20, 44]]. EMT CSCs are also able to self-renew and to switch back to the EPI CSC phenotype and, although the mechanism inducing this change is unclear, it is associated with loss of GSK3β activity. GSK3β is thus key to the regulation of the choice between CSC self-renewal and differentiation. Inducing loss of GSK3β activity might therefore be used therapeutically to enhance differentiative loss of the EPI CSC phenotype. However, blocking GSK3β activity may have conflicting results in terms of tumor behavior. Its loss, by reducing the self-renewal ability of CSCs, should reduce the CSC population, but the promotion of MET by inactivation of GSK3β may aid development of secondary metastatic tumors. However, it now seems apparent that tumor growth depends on self-renewal of CSCs, and that tumor invasion and metastasis is related to EMT and the reverse process of MET. A better understanding of how these processes are controlled and sustained seems necessary to enable future therapies to successfully manipulate loss of both EMT CSC and EPI CSC, an effect apparently required for eradication of an entire tumor.
We have demonstrated that in squamous cell carcinoma active GSK3β is required to maintain self-renewal of cancer stem cells and the expression of stem cell markers. Inactivation of GSK3β through phosphorylation results in cell differentiation and a reduction in CD44 expression. In addition, a population of cancer stem cells that has undergone EMT also requires GSK3β activity in order to maintain its mesenchymal phenotype. CD44 and RHAMM act upstream of GSK3β to control these functions.
This work was supported by grants to Dr. Shigeishi from the Scientific Research Fund of Sugiyama Chemical Industrial Laboratory (2011), the Satake Fund for Scientific Research from Hiroshima University Supporters’ Association (2011), and a Grant-in-aid for Scientific Research (C) (No.11008667) from the Japanese Ministry of Education, Culture, Sports, and Technology. It was also supported by grants from the NC3Rs, the Fanconi Anemia Research Fund, Bart's and The London Charitable Foundation, and the Saving Faces Research Foundation.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.