Ephrin A4-ephrin receptor A10 signaling promotes cell migration and spheroid formation by upregulating NANOG expression in oral squamous cell carcinoma cells

Ephrin type-A receptor 10 (EPHA10) has been implicated as a potential target for breast and prostate cancer therapy. However, its involvement in oral squamous cell carcinoma (OSCC) remains unclear. We demonstrated that EPHA10 supports in vivo tumor growth and lymphatic metastasis of OSCC cells. OSCC cell migration, epithelial mesenchymal transition (EMT), and sphere formation were found to be regulated by EPHA10, and EPHA10 was found to drive expression of some EMT- and stemness-associated transcription factors. Among EPHA10 ligands, exogenous ephrin A4 (EFNA4) induced the most OSCC cell migration and sphere formation, as well as up-regulation of SNAIL, NANOG, and OCT4. These effects were abolished by extracellular signal-regulated kinase (ERK) inhibition and NANOG knockdown. Also, EPHA10 was required for EFNA4-induced cell migration, sphere formation, and expression of NANOG and OCT4 mRNA. Our microarray dataset revealed that EFNA4 mRNA expression was associated with expression of NANOG and OCT4 mRNA, and OSCC patients showing high co-expression of EFNA4 with NANOG or OCT4 mRNA demonstrated poor recurrence-free survival rates. Targeting forward signaling of the EFNA4-EPHA10 axis may be a promising therapeutic approach for oral malignancies, and the combination of EFNA4 mRNA and downstream gene expression may be a useful prognostic biomarker for OSCC.


EPHA10 knockdown reduced tumorigenesis and lymph node metastasis of OSCC cells. To
assess EPHA10 expression in different cancer types, we searched for data on EPHA10 expression in datasets from the Oncomine database 23 . We surveyed 4 datasets from breast, esophageal, and lung cancers and found significantly higher expression of EPHA10 in cancer versus normal tissue. In contrast, 7 datasets showed markedly lower EPHA10 expression in colorectal, brain, and central nervous system cancers compared to site-matched normal tissue (Fig. 1A). However, EPHA10 expression was not significantly different when comparing data from cancer and normal tissue in available HNSCC datasets. To investigate EPHA10 protein expression, we examined OSCC cells, a subset of HNSCC, and found variable levels of EPHA10 protein, ranging from 0.69-to 2.12-fold, in the dysplastic oral keratinocyte (DOK) and OSCC cell lines compared with human oral keratinocytes (HOKs; Fig. 1B). We also found slightly higher expression of EPHA10 in LN1-1 cells, an OSCC cell line (in vivo OEC-M1 cell derivation) with high activities of tumor growth and lymphatic metastasis 24 , compared to parental OEC-M1 cells (Fig. 1B).
To evaluate the effects of EPHA10 on OSCC tumor growth and lymphatic metastasis, we ablated EPHA10 expression in LN1-1 cells using lentiviral shRNA. We confirmed EPHA10 knockdown in LN1-1 cells by Figure 1. EPHA10 is required for tumorigenesis and metastasis of OSCC cells. (A) Expression of EPHA10 in 20 types of cancer versus corresponding normal tissues using the Oncomine database with the threshold of fold change ≥ 2, p ≤ 10 -4 , and gene rank ≥ top 10%. Red and blue, respectively, indicate the numbers of datasets with statistically significant increases and decreases in EPHA10 gene expression. (B) EPHA10 expression in human oral keratinocytes (HOK), immortalized dysplastic oral keratinocytes (DOK), and 7 OSCC cell lines was examined by western blotting. Protein levels were normalized to an internal control (α-tubulin). Relative ratios were determined by dividing the EPHA10 protein level in each cell type by that in HOK cells. (C) EPHA10 protein levels in LN1-1 cells expressing EPHA10-specific shRNA and vector control (pLKO-GFP) were determined by western blot. Protein levels were normalized to α-tubulin. Relative ratios were determined by dividing the EPHA10 protein level in each expression variant by that in the pLKO-GFP vector-expressing cells. www.nature.com/scientificreports/ immunoblot and flow cytometry (Fig. 1C,D). We performed oral cavity inoculation of immunodeficient nude mice with EPHA10 knockdown and control cells. Tumors developed in all animals regardless of EPHA10 level ( Table 1). The average tumor weight was 0.2278 ± 0.03017 g (n = 9) and 0.0475 ± 0.01191 g (n = 8) in mice receiving control cells and EPHA10 knockdown LN1-1 cells, respectively (Fig. 1E). Similarly, EPHA10 knockdown led to significantly lower tumor volume compared to controls (Figs. 1E, S1A). There were consistently, fewer Ki-67-positive cells in EPHA10 sh3 orthotopic tumors (16.63 ± 2.247%, n = 7) compared to controls (27.46 ± 3.597%, n = 9; Fig. 1F). Additionally, 67% (n = 9) of the animals injected with control cells developed lymph node metastases compared to none of the EPHA10 sh3 injected animals (0%, n = 8), as assessed by examination of H&E-stained cervical lymph nodes for evidence of tumor formation (Fig. S1B, Table 1). Quantification of LYVE-1-positive areas demonstrated reduced intratumoral lymphatics in tumors of mice injected with EPHA10 sh3 (0.9656 ± 0.139 vessels/field per specimen, n = 8) compared to those injected with parental LN1-1 cells (2.225 ± 0.454 vessels/ field per specimen, n = 9; Fig. S1C). These results indicate that EPHA10 is important for OSCC tumor growth and lymph node metastasis.

Manipulation of EPHA10 levels altered OSCC cell properties.
Knockdown of EPHA10 in LN1-1 cells showed no difference in cell proliferation compared with cells expressing the vector control (Fig. S1D). In contrast, knockdown of EPHA10 diminished the migration potential of LN1-1 cells, as indicated via transwell assay, compared to corresponding controls (Fig. S1E). LN1-1 cells expressing EPHA10 sh5 were used to confirm the inhibitory effects of EPHA10 knockdown on cell migration ( Fig. 2A). Phase contrast microscopy revealed obvious morphological differences in EPHA10 knockdown versus control cells; EPHA10 knockdown cells had a predominately polygonal shape, while the control cells showed fibroblastic morphology (Fig. 2B). In addition, immunoblot analysis showed that EPHA10 knockdown led to differential expression of epithelial proteins, such as α-catenin, E-cadherin, and β-catenin, as well as the mesenchymal protein vimentin (Fig. 2C). Sphere formation is an efficient way to assess the properties of CSCs in vitro 25 , and cells with EPHA10 knockdown showed reduced sphere formation compared to controls (Fig. 2D). Likewise, mRNA expression of EMT-and stemnessassociated transcription factors, including TWIST, SLUG, SNAIL, OCT4, and NANOG, was diminished in EPHA10 knockdown cells (Fig. 2E). These results suggest that EPHA10 knockdown alters OSCC cell properties, such as EMT, migration, and capacity for sphere formation, as well as expression of EMT-and stemnessassociated transcription factors. Additionally, OEC-M1 cells with low EPHA10 expression were infected with a retroviral vector encoding human EPHA10 (PB-EPHA10 cells) or an empty vector (PB cells), and ectopic EPHA10 expression was confirmed by immunoblot (Fig. S2A). Ectopic expression of EPHA10 did not affect OEC-M1 cell growth, division, death ( Fig. S2B-D), or E-cadherin expression; however, vimentin expression was reduced (Fig. S2A). Unexpectedly, ectopic EPHA10 expression was associated with diminished cell migration and sphere formation (Fig. S2E,F). Furthermore, ectopic EPHA10 expression significantly decreased the expression of SNAIL, SLUG, OCT4, and NANOG mRNA. Our data demonstrated that EPHA10 knockdown and ectopic EPHA10 expression results in a similar phenotype in OSCC cells.
EFNA4 promoted cell migration, spheroid formation, and induction of EMT-and stemness-related transcription factors. The most high-affinity ligands for EPHA10 are ephrin-A3 (EFNA3), A4 (EFNA4), and A5 (EFNA5) 26 . We examined the expression of EFNA3, EFNA4, and EFNA5 in OEC-M1 and LN1-1 cells via the GSE62326 dataset and found that EFNA3 and EFNA5 expression was undetectable in both cell types. Lower EFNA4 mRNA and protein expression was found in OEC-M1 cells compared to LN1-1 cells using microarray analysis and western blotting, respectively (Fig. S3A), therefore OEC-M1 cells were used to investigate the effects of EPHA10 ligands on cellular functions. When exogenous EPHA10 ligands capable of binding to the receptors and labelled with IgG-Fc tags were added to cultured cells, neither EFNA3-Fc nor EFNA5-Fc affected OEC-M1 cell migration (Fig. S3B,C). Exogenous EFNA3-Fc enhanced tumorsphere size distribution, while exogenous EFNA3-Fc and EFNA5-Fc had no effects on sphere formation activity compared to controls (Fig. S3D,E). Our data suggest that exogenous EFNA3 and EFNA5 have no marked effects on migration or sphere-forming activities of OEC-M1 cells.
Exogenous EFNA4-Fc did not have a significant effect on cell growth or division (Fig. 3A,B), but there was a small effect on cell death in OEC-M1 cells treated with low dose EFNA4 (Fig. 3C). Exogenous EFNA4 enhanced OEC-M1 cell migration in a dose-dependent manner (Fig. 3D), an effect that was also observed with other OSCC cells, including SAS and TW2.6 ( Fig. S4A). Exogenous EFNA4-Fc significantly increased sphere-forming activity in OEC-M1 cells (Fig. 3E), and qRT-PCR analysis showed that the expression of TWIST, SNAIL, OCT4, and NANOG mRNA was significantly elevated after addition of EFNA4-Fc to OEC-M1 cells (Fig. 3F).   www.nature.com/scientificreports/ PCR (Fig. 4F). NANOG knockdown suppressed EFNA4-induced cell migration in OEC-M1 cells (Fig. 4G). Similarly, EFNA4-induced sphere formation was inhibited in OEC-M1 cells with NANOG knockdown (Fig. 4H). These data suggest that NANOG is required for EFNA4-induced cell migration and sphere formation.
EPHA10 was required for EFNA4-induced cell migration and sphere formation. EFNA4 preferentially binds to EPHA2 and EPHA10 26 . To determine whether EPHA2 was required for EFNA4-induced phenotypes, OEC-M1 cells stably transduced with EPHA2 shRNA or a control vector were established. The reduced levels of EPHA2 protein in the knockdown cells (EPHA2 sh4 and sh5) were verified by immunoblot (Fig. 5A). EPHA2 knockdown OEC-M1 cells were treated with EFNA4-Fc and analyzed for cell migration and sphere formation. EPHA2 knockdown suppressed EFNA4-induced cell migration in OEC-M1 cells (Fig. 5B). In contrast, EFNA4-induced sphere formation was not disrupted in OEC-M1 cells with EPHA2 knockdown, suggesting that EPHA2 is not required for EFNA4-induced sphere formation (Fig. 5C). These data suggest that EFNA4-induced cell migration involves signaling by EPHA2. OEC-M1 cells with EPHA10 knockdown, confirmed by immunoblot assay (Fig. 5D), were stimulated with EFNA4-Fc and examined for cell migration and sphere formation. EFNA4-induced cell migration and sphere formation were inhibited by EPHA10 knockdown (Fig. 5E,F). Similarly, EFNA4-induced migration and sphere formation was disrupted by EPHA10 knockdown in TW2.6 cells (Fig. S4C,D). These data implicate EPHA10 as a critical factor for EFNA4-enhanced cell migration and sphere formation in OSCC cells. We also found that EFNA4-induced upregulation of NANOG and OCT4 mRNA was diminished by EPHA10 knockdown in OEC-M1 cells, although this effect was not observed for SNAIL mRNA (Fig. 5G). These data indicate that the forward signaling of EFNA4-EPHA10 axis modulates cellular phenotypes, such as migration, sphere formation, and expression of NANOG and OCT4 mRNA.

Co-expression of EFNA4 with NANOG and OCT4 in OSCCs.
To investigate the clinical significance of EFNA4 in OSCC, we analyzed EFNA4 expression in publicly available cDNA microarray datasets (Table S1). Two datasets from the Oncomine database demonstrated significant increases in EFNA4 mRNA in OSCC compared to normal tissues (p < 0.0001 and p < 0.05; Fig. 6A, a and b, respectively) 27,28 . Expression of EFNA4 mRNA was also significantly increased in 40 OSCC specimens compared to nontumor controls (p < 0.0001; Fig. 6A, c) 29 . EFNA4 mRNA expression was significantly higher in grade 2 OSCC than in grade 1 tumors (p < 0.05, Fig. 6A, d) 30 , and EFNA4 mRNA was significantly increased in OSCC with angiolymphatic invasion compared to tissue without invasion (p < 0.01, Fig. 6A, e) 29 . We assessed EFNA4 protein level in both tumor tissue and corresponding noncancerous epithelia in OSCC samples by immunohistochemical staining (IHC; Fig. 6B). High EFNA4 expression (intensity grade 2 or 3) was observed in 72.2% (13/18) of tumors, compared with only 11.8% of noncancerous tissues showing high EFNA4 expression (p < 0.001, Fig. 6C). Moreover, 75% of OSCC specimens demonstrated higher EFNA4 expression in tumor regions than in noncancerous regions (Fig. 6C).
To examine the relationships between EFNA4-EPHA10 signaling and expression of NANOG and OCT4 mRNA, we further investigated the cDNA expression dataset GSE37991 (gene expression profiling of OSCC) based on its complete clinical information 29 . We found that the expression of EFNA4, NANOG, and OCT4, but not EPHA10, mRNA was significantly increased in OSCC tissues compared to adjacent nontumor samples (Fig. 6D). Also, tumor/nontumor (T/N) ratios of EFNA4, NANOG, OCT4, and EPHA10 mRNA expression had averages of 1.39, 1.25, 2.37, and 1.03, respectively. Interestingly, a positive correlation existed between EFNA4 and NANOG mRNA expression in OSCC samples (compared to paired nontumor tissues) using Pearson correlation (Fig. 6E). Similarly, significant correlations in mRNA expression also occurred between EFNA4 and OCT4 and between NANOG and OCT4 (Fig. 6E). Furthermore, patients with higher EFNA4/NANOG or EFNA4/OCT4 mRNA expression levels had worse recurrence-free survival compared to patients with lower EFNA4/NANOG or EFNA4/OCT4 mRNA expression (p = 0.027, Fig. 6F). The clinical data confirmed our in vitro findings that EFNA4 mRNA expression correlated with expression of NANOG or OCT4 mRNA.

Discussion
EPHA10 is associated with tumor progression and metastasis in breast cancer 20 , and is a promising therapeutic target in triple-negative breast cancer and prostate cancer 21,22 . Based on the publicly available clinical datasets, we found that EPHA10 mRNA expression varied among different cancer types (Fig. 1A), indicating that the effect of EPHA10 was tissue-specific and likely dependent on the milieu of binding partners and related target gene regulation in a given cell type. Antibodies targeting EPHA10 significantly inhibited tumor growth in breast cancer xenograft mouse models 22 . However, the underlying mechanisms of EPHA10-mediated tumorigenesis are undefined. In this study, we first confirmed the essential role of EPHA10 in tumorigenesis and metastasis of OSCC cells with in vivo models (Figs. 1E,F, S1A,B,C, and Table 1), suggesting that EPHA10 signaling may be a useful therapeutic target in OSCC.
EPHA10 knockdown suppressed in vivo tumor growth and in vitro sphere formation (Figs. 1E,F, S1A, 2D), but did not affect cell proliferation (Fig. S1D). This discrepancy may be due to the complexity of spatial distribution and cell contact in tumor spheres 31 , as well as interplay between tumor cells and tumor-associated stromal cells, suggesting a critical role for EPHA10 in complex cell-cell interactions and the tumor-associated microenvironment. Yin et al. found that EphA receptors and co-expressed ephrin-A ligands directly interact in cis via their binding domains, and that this interaction does not seem to induce intracellular signals, but inhibits the trans interaction 15 . Similarly, our data imply that ectopic expression of EPHA10 could dramatically enhance the possibility of cis interactions with binding ligands, such as EFNA4, and that these interactions could reduce the relative level of EFNA4 for trans-interaction between EFNA4-EPHA10, thereby inhibiting the EPHA10 forward signal among cells (Figs. 2, S2). www.nature.com/scientificreports/ EFNA3 is a tumor suppressor of malignant peripheral nerve sheath tumors 32 , while EFNA5 overexpression is associated with prostate tumorigenesis 33 . No significant effects of exogenous EFNA3 or EFNA5 on OSCC cell migration or spheroid formation were observed in our study (Fig. S3). In datasets of oral cancer gene expression, EFNA4 is upregulated in OSCC tissues compared with nontumor tissues and is correlated with later stage and angiolymphatic invasion of OSCC (Fig. 6A). Similarly, EFNA4 expression was elevated in breast cancer, ovarian cancer, and hepatocellular carcinoma 34 , and may play a role in cell fate determination of mammary epithelial cells 35 . EFNA4 may be a promising target for identification of tumor-initiating cells in triple-negative breast cancer and ovarian cancer 34 . EFNA4-mediated bidirectional signaling is not well-characterized and many questions remain. Originally, our data suggested that EFNA4-Fc, like membrane-bound EFNA4, is involved in regulating TWIST, SNAIL, NANOG, and OCT4 mRNA expression, as well as inducing cell migration and spheroid formation in OSCC cells (Fig. 3). Yin et al. illustrated that Stat/Snail signaling modulated the co-expression of OCT4 and NANOG 36 . We found that EFNA4 induced expression of SNAIL, NANOG, OCT4 mRNA via ERK activation (Fig. 4), however, Stat3 activation was not addressed in this study. Additionally, our data demonstrated the inhibitory effects of NANOG knockdown on EFNA4-induced cell migration and sphere formation (Fig. 4). Similarly, Yin et al. reported that co-expression of OCT4 and NANOG induced development of CSC characteristics and enhanced EMT in hepatocellular carcinoma, and Huang et al. showed that ERK-NANOG signaling promoted CSC phenotypes and EMT in HNSCC 36,37 .
Overexpression of EPHA2 is related to malignancy and tumor angiogenesis of tongue squamous cell carcinoma 38 . EPHA2 also contributes to human glioma stem cell formation and stemness marker SOX2 expression 39 . However, the roles of EPHA10 in development of cancer stemness remain unknown. Our data showed that knockdown of EPHA10, but not EPHA2, influences exogeneous EFNA4-induced spheroid formation (Fig. 5C,F), while both EFNA4-EPHA2 and EFNA4-EPHA10 forward signaling axes regulate migration of OSCC cells (Fig. 5B,E). EPHA2 activation and EPHA2 overexpression displayed opposite roles in regulation of cell migration and invasion; EPHA2 activation by ephrin A1 suppressed chemotactic migration, whereas overexpression of EPHA2 enhanced migration in a ligand-independent manner 40 . EFNA4-EPHA2 signaling in chronic lymphocytic leukemia cells significantly reduced their adhesion and impaired cell trafficking and chemotaxis 41 . Any association between Eph receptor-ligand signaling and migration may depend on tumor subtype, microenvironmental context, and other parameters. Interestingly, we found that EFNA4-EPHA2 forward signaling affects cell migration but not spheroid formation, indicating a difference in receptor-specific functions and signaling between EPHA2 and EPHA10. Downstream proteins that interact with kinase-deficient EPHA10 and the mechanisms involved in EPHA10 phosphorylation require further investigation to clarify the differences between EPHA10 and EPHA2 signaling 16 . In OSCC cells, EPHA10 plays a critical role in linking external stimuli (EFNA4 or other ligands) to the internal signal transduction that leads to cancer cell migration and spheroid formation. When considering ephrin ligands and their receptors as potential therapeutic targets, excessive toxicity associated with pan-ephrin receptor inhibition must be considered.    34 , we demonstrated upregulated expression of EFNA4 in OSCC tissues compared to nontumor tissues (Fig. 6B,C). Interestingly, we found that EFNA4 was expressed in both tumor cells and tumor-infiltrating immune cells (unpublished data), suggesting that EFNA4 could be expressed by oral cancer cells and their associated stromal cells. Further investigation is needed to determine whether contact between tumor cells and infiltrating immune cells contributes to oral tumorigenesis. We provided data that characterizes EFNA4-EPHA10 forward signaling in OSCC cells using exogenous Eph receptor ligands, however, characterization of reverse signaling in EFNA4-expressing cells requires additional research. In this study, we demonstrated correlation of EFNA4, NANOG, and OCT4 mRNA expression in clinical OSCC specimens (Fig. 6E), and showed that the co-expression of EFNA4 with NANOG or OCT4 was associated with poor prognosis in OSCC patients (Fig. 6F). These data were consistent with the finding that the expression of NANOG and OCT4 mRNA was enhanced by induction of EFNA4-EPHA10 forward signaling in OSCC cells (Fig. 5G). Our data did not show correlation of EPHA10 expression with OSCC clinical outcome and differential expression of EPHA10 in the GSE37991 dataset (Fig. 6D). However, our data indicated inhibition of trans-interaction by ectopic EPHA10 expression (Fig. S2), suggesting that upregulation of both EPHA10 and EFNA4 in OSCC tissues increased cis-interactions, and that upregulation of EFNA4 in cancer tissues was enough to induce the downstream effects of EFNA4-EPHA10 forward signaling.
Collectively, EPHA10 supports tumor growth and lymph node metastasis of OSCC cells in vivo. Signaling by EPHA10 and its ligand, EFNA4, promotes OSCC cell migration and tumor spheroid formation through induction of NANOG mRNA expression via ERK activation (Fig. 6G). Finally, our findings are supported by clinical data showing that patients with high co-expression of EFNA4/NANOG or EFNA4/OCT4 mRNA had worse recurrence-free a survival than those with low co-expression (Fig. 6F), supporting a significant role for EHPA10 and EFNA4 in OSCC development and progression, as well as the significance of these signaling molecules as potential therapeutic targets.

Methods
Gene expression data mining. Oncomine (www.oncom ine.org), an online web-based cancer database for RNA and DNA sequences, was used to facilitate data-mining of the expression of gene transcripts in 20 types of cancer 23 . Expression of EPHA10 or EFNA4 mRNA in cancer samples was compared with expression in samples of normal tissue using the Student's t-test.
Cell lines and reagents. Human oral keratinocytes (HOK) were obtained from ScienCell Research Laboratories (Carlsbad, CA, USA) and grown in an oral keratinocyte medium (OKM) according to the manufacturer's protocols. OSCC cell lines, including the OC3 line established from an OSCC specimen 42 , the CGHNC9 line established from an oral cancer specimen 43 , the TW2.6 line established from a buccal carcinoma 44 , the OEC-M1 line established from an oral epidermoid carcinoma 45 , the SAS line established from a poorly differentiated tongue squamous cell carcinoma 46 , the DOK line established from a human dysplastic oral mucosa 47 , the HSC3 line established from tongue carcinoma with cervical metastasis 48 , and the LN1-1 line developed via in vivo selection from OEC-M1-derived tumor 24 , were obtained and kept within 20 passages in each experiment, as previously described 24,49 . These cells were authenticated using the short tandem repeat assay at the Center for Genomic Medicine, National Cheng Kung University (Tainan, Taiwan) and Mission Biotech (Taipei, Taiwan). Ephrin A3 labelled with an IgG Fc tag (EFNA3-Fc), ephrin A4 labelled with an Fc tag (EFNA4-Fc), ephrin A5 labelled with an Fc tag (EFNA5-Fc), and an IgG control were purchased from R&D Systems (Minneapolis, MN, USA).
Cell growth. Cell growth curves were developed using the MTS assay as previously described 51

Cell division analysis by carboxyfluorescein succinimidyl ester (CFSE). Cells were washed with
Dulbecco's Phosphate-Buffered Saline (DPBS, ThermoFisher Scientific, Waltham, MA, USA) twice, then stained with 10 μM CFSE (ThermoFisher Scientific) in DPBS for 10 min. Full culture medium was added to terminate the reaction, and cells were centrifuged to remove the staining solution. Cells were suspended in culture medium and seeded into 24-well plates at 2 × 10 4 cells per well. After incubation for 3 days, the cells were harvested and analyzed by flow cytometry. The experiments were performed at least twice.  Migration assay. Cell migration analyses were conducted using a transwell assay as previously described 52 .
Relative migration activity was determined by normalizing the mean number of cells that had migrated per field (n = 10) in the experimental condition to that of control cells. The experiments were performed at least twice.
Orthotopic inoculation in nude mice. The procedures for orthotopic inoculation were previously described 24 . A total of 5 × 10 5 cells in 50 μL sterile DPBS was injected into the buccal mucosa of two batches of 5-6-week-old male nude mice (BALB/cAnN.Cg-Foxn1nu/CrlNarl; n = 4-5 per batch) and mice were sacrificed at 28-31 days post-inoculation for the first and second batches, respectively. Tumor size was determined by measuring the tumor dimensions and calculating volume (mm 3 ) using the formula 1/2 × (length) × (width) 2 .
The orthotopic tumors were weighted and then processed by the Pathology Core Lab (National Health Research Institutes, Taiwan).
Immunohistochemical analysis. IHC analysis was conducted as previously described 50  Cell morphology. Imaging of actively proliferating cells was achieved using an inverted microscope with a phase-contrast lens, as previously described 50 .
Sphere formation. Cells (1 × 10 4 ) in DMEM/F12 medium (Sigma-Aldrich) with 20 ng/ml of basic fibroblast growth factor (bFGF, Abcam), 10 ng/ml of epidermal growth factor (EGF; Thermo Fisher Scientific, Waltham, MA, USA), and B27 supplement (Thermo Fisher Scientific), were cultured in Corning Costar Ultra-Low Attachment 6-Well Plates (CLS3471, Merck, Darmstadt, Germany) for 14 days. The number and size of tumorspheres were quantitatively assessed using ImageJ. Relative sphere formation activity was determined by normalizing the mean number of spheres per well (n = 2) in the experimental condition compared to that of control cells. The experiments were performed at least twice.
Quantitative reverse transcription-polymerase chain reaction (qRT-PCR). The qRT-PCR was performed as previously described 50  Diego, CA, USA). The AVOVA test and Student's t-test were used to assess statistical differences between groups. The paired t-test was used to determine differences in gene expression between tumor samples and adjacent nontumor samples. The Pearson correlation was used to evaluate the linear relation between two variants. The log rank test was used to evaluate differences in survival between stratified groups. The R base heatmap function (R Stats Package, https ://www.r-proje ct.org/) was used to generate the gene expression heatmap, which presents data from 40 OSCC patients via scaled gene expression differences (for tumor/nontumor pairs). The R chart correlation function (R Performance Analytics Package, https ://www.rdocu menta tion.org/packa ges/Perfo rmanc eAnal ytics /versi ons/2.0.4) was used to calculate the correlation between gene expression patterns and produce the correlation matrix. For all comparisons, p < 0.05 was considered statistically significant.
Ethics approval and consent to participate. All animal studies followed the guidelines for the Care www.nature.com/scientificreports/ a parent and/or legal guardian. All methods were carried out in accordance with relevant guidelines and regulations.

Data availability
The data, generated and/or analyzed, that are included in this article are available from the corresponding author upon reasonable request.