Glucose transporter 4 promotes head and neck squamous cell carcinoma metastasis through the TRIM24-DDX58 axis

The human head and neck squamous cancer cell lines FaDu, Detroit-562, HSC-2, HSC-3, HSC-M3, HSC-4, RPMI-650, and Ca-922 were grown in MEM supplemented with 10% FBS (Invitrogen, Carlsbad, CA, USA). All cells were incubated in a humidified atmosphere of 5% CO₂ at 37 °C. All cell lines were purchased from the JCRB cell bank. The pGIPZ lentiviral shRNAmir system (Thermo, Waltham, MA, USA), virus-backboned short hairpin RNA (shRNA) clones, and the GLUT4 sequence were used to establish stable cell lines (Additional file 1: Table S5). Lentiviruses were used to infect the cells for 2 days. Stable clones were selected by treating the cells with 1 μg/ml puromycin (Sigma, St. Louis, MO, USA) for 2 weeks.

HNSCC cell pellets were lysed in RIPA buffer with protease/phosphatase inhibitors on ice. The protein content was quantified using a BCA assay kit (Thermo, Waltham, MA, USA), and equal protein amounts (30 μg) of each sample were used for western blot analysis. PVDF membranes (Millipore, Bedford, MA, USA) were blocked with 5% fat-free milk and then incubated with primary antibodies directed against GLUT4 (Epitomics, Cambridge, MA, USA), GLUT1 (GeneTex, Hsinchu, Taiwan), DDX58 (GeneTex, Hsinchu, Taiwan) or OASL (GeneTex, Hsinchu, Taiwan), and α-tubulin (Sigma, St. Louis, MO, USA). Immunoreactive bands were visualized using an enhanced chemiluminescence (ECL) system (Amersham ECL Plus™, GE Healthcare Life Sciences, Chalfont St. Giles, UK).

Total RNA was extracted and purified using an RNeasy Mini kit (Qiagen, Valencia, CA, USA) and qualified with a model 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). All RNAs were labeled using a GeneChip 3′IVT Expression Kit & Hybridization Wash and Stain Kit (Affymetrix, Santa Clara, CA, USA) and analyzed using Affymetrix GeneChip Human Genome U133 plus 2.0 arrays (Affymetrix, Santa Clara, CA, USA). The gene expression levels were normalized as log2 values using GeneSpring software (Agilent Technologies, Palo Alto, CA, USA). Genes that were up- or downregulated with greater than 1.5-fold changes in response to GLUT4 overexpression were further subjected to computational simulation by Ingenuity Pathway Analysis (IPA; QIAGEN, Valencia, CA, USA) online tools to predict potential upstream regulators and canonical pathways. The microarray data were uploaded to the National Center for Biotechnology Information Gene Expression Omnibus (GEO, NCBI) (GSE89631).

Glucose consumption and lactate production were measured using colorimetric glucose and lactate assay kits (BioVision, Milpitas, CA, USA) according to the manufacturer’s protocols. Briefly, cells from the designated experiments were incubated with assay buffer containing enzyme and glucose/lactate probes. Then, the optical densities were measured at 570/450 nm wavelengths. The glucose analog 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG; Sigma, St. Louis, MO, USA) was also used to analyze glucose uptake. In addition, cells were treated with the GLUT4 transport inhibitors indinavir or ritonavir (Sigma, St. Louis, MO, USA) at 100 and 50 μM, respectively, for 60 min, and the uptake of 2-NBDG was measured using Vector 3 (Bruker, MA, USA) to detect relative fluorescence counts.

Three representative 1-mm-diameter cores from each tumor, taken from formalin-fixed paraffin-embedded tissues, were selected for morphology typical of the diagnosis. Assessable cores were obtained in 90 cases. The histopathological diagnoses of all samples were reviewed and confirmed by a pathologist, Michael Hsiao. IHC staining was performed on serial 5-μm-thick tissue sections cut from the tissue microarray (TMA) using an automated immunostainer (Ventana, Tucson, AZ, USA). Briefly, the sections were first dewaxed in a 60 °C oven, deparaffinized in xylene, and rehydrated in graded alcohol. Antigens were retrieved by heat-induced antigen retrieval for 30 min in Tris-EDTA buffer. The slides were stained with a polyclonal rabbit anti-human GLUT4 antibody (1:750, Epitomics, Cambridge, MA, USA). The sections were subsequently counterstained with hematoxylin, dehydrated, and mounted. The IHC staining intensity was scored by two pathologists as follows: no cytoplasmic staining or cytoplasmic staining in <10% of tumor cells was defined as score 0; faint/barely perceptible partial cytoplasmic staining in >10% of tumor cells was defined as score 1+; moderate cytoplasmic staining in >10% of tumor cells was defined as score 2+; and strong cytoplasmic staining in >10% of tumor cells was defined as score 3+. Scores of 0 and 1+ were defined as low GLUT4 expression, while scores of 2+ and 3+ were defined as high GLUT4 expression.

Age-matched, nonobese diabetic-severe combined immunodeficient gamma (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ JAX®, NOD-SCID γ) male mice (6–8 weeks old, 20–25 g body weight) were used. To evaluate lung colony-forming ability, 1 × 10⁶ cells were resuspended in 100 μL of PBS and injected into the lateral tail vein. Lung nodule formation was quantified after H&E staining using a dissecting microscope at the endpoint. To evaluate in vivo tumorigenicity ability and establish an orthotopic model, 5 × 10⁶ cells were resuspended in 100 μL of PBS and then subcutaneously injected into the flanks of the mice, and 5 × 10⁶ cells were resuspended in 10 μL of PBS and injected into the buccal submucosa. All animal experiments were conducted in accordance with a protocol approved by the Academia Sinica Institutional Animal Care and Utilization Committee (IACUC).

In total, 90 patients diagnosed with head and neck squamous cell carcinoma at the Taipei Medical University Hospital in Taiwan from 1991 to 2010 were included in this study. Patients who received preoperative chemotherapy or radiation therapy were excluded. Clinical information and pathology data were collected via a retrospective review of patient medical records. All cases were staged according to the 7th edition of the Cancer Staging Manual of the American Joint Committee on Cancer (AJCC), and the histological cancer type was classified according to the World Health Organization (WHO) 2004 classification guidelines. Follow-up data were available in all cases, and the longest clinical follow-up time was 190 months. Overall survival and disease-free survival were defined as the intervals from surgery to death caused by head and neck squamous cell carcinoma and recurrence or distant metastasis, respectively. The study was performed with the approval of the Institutional Review Board and with permission from the ethics committee of the institution involved (TMU-IRB 99049).

The nonparametric Mann–Whitney U test was used to analyze the statistical significance of results from three independent experiments. Statistical analyses were performed using SPSS (Statistical Package for the Social Sciences) 17.0 software (SPSS, Chicago, IL, USA). A paired t test was performed to compare the GLUT4 IHC expression levels in cancer tissues and in the corresponding normal adjacent tissues. The association between clinicopathological categorical variables and the GLUT4 IHC expression levels were analyzed by Pearson’s chi-square test. Estimates of the survival rates were calculated using the Kaplan–Meier method and compared using the log-rank test. The follow-up time was censored if the patient was lost during follow-up. Univariate and multivariate analyses were performed using Cox proportional hazards regression analysis with and without an adjustment for GLUT4 IHC expression level, tumor stage, lymph node stage, and recurrence status. For all analyses, a P value of <0.05 was considered significant.

To determine the clinical association between glucose transporters (GLUTs) in HNSCC patients, we utilized a previously developed HNSCC microarray database to examine and compare the expression of 10 major GLUTs using the Oncomine website. GLUT4 was found to be the only GLUT family member to have a significant correlation with metastatic status compared with other GLUT family members in the clinical cohort (Fig. 1a, 3.59-fold change, P = 5.20E-5). We then compared the correlations of all the GLUT family members with the prognosis of patients in the Petel HNSCC cohort (E-MTAB-1328, n = 89) in the SurvExpress database. The number of cases was divided approximately in half based on the expression (low or high) of the GLUT family member of interest, and a Kaplan–Meier survival analysis was performed on both groups using the SurvExpress website. The results showed that GLUT4 is the only GLUT family member whose RNA expression is significantly correlated with HNSCC overall survival (Fig. 1b, HR = 3.37, P value =0.043, other GLUT family data in Additional file 1: Figure S1). Forest plots of GLUT family members and their corresponding hazard ratios and Cox-P values were generated for another HNSCC microarray cohort (GSE2837, n = 40), and these results also showed GLUT4 to be the strongest prognosis marker with the highest hazard ratio. The Cox-P value for GLUT4 was calculated to be 0.07 by the Pronoscan website (Additional file 1: Figure S2). Together, these data show that of the GLUT family members, GLUT4 is the most significantly correlated with the clinical outcomes of HNSCC.

We next validated these findings by examining GLUT4 protein expression using our own clinical HNSCC tissue cohort. The immunohistochemical staining results showed stronger staining of the GLUT4 protein in tumor tissues than in the adjacent normal tissues (Fig. 1c). After scoring, we determined the correlation between patient survival and either low-level GLUT4 staining (Fig. 1d, IHC scores 0 and 1) or high-level GLUT4 staining (Fig. 1d, IHC scores 2 and 3). The results indicated that high-level GLUT4 staining was significantly correlated with the poor overall and disease-free survival probabilities (Fig. 1e, P = 0.017, P = 0.001, respectively). A clinicopathological analysis showed that high GLUT expression is significantly correlated with recurrence (Table 1, P = 0.001). The patient demographic features are shown in Additional file 1: Table S1. Univariate and multivariate analyses of the disease-free survival probability showed that high-level GLUT4 expression served as the strongest independent prognostic marker in both the univariate analysis (Table 2, HR = 3.35, P = 0.001) and the multivariate analysis (Table 2, HR = 3.76, P < 0.001). These data indicate that the upregulation of GLUT4 is significantly associated with the distant metastasis and disease-related progression in HNSCC patients.

To determine the functional attributes of GLUT4 in promoting HNSCC cellular migration and invasion, we first examined the GLUT4 protein expression levels in HNSCC cell lines. Our results showed varied expression levels of the GLUT4 protein in the eight HNSCC cell lines examined (Fig. 2a). We then determined the migration and invasion potentials of these HNSCC cell lines (Fig. 2b, c). The migration and invasion potentials of these cell lines were compared with their respective GLUT4 protein expression levels. Our results showed that GLUT4 expression appeared to be causally associated with metastatic potentials in HNSCC cells (Fig. 2d, Spearman rho = 0.81, P = 0.015). The GLUT4 gene was ectopically overexpressed in the low GLUT4-expressing cell lines HSC-3 and FaDu to determine whether GLUT4 overexpression induces HNSCC cell migration and invasion. The results in the left panel of Fig. 2e show the overexpression of the GLUT4 protein in the HSC-3 and FaDu cells. GLUT4 overexpression indeed significantly promoted the migration and invasion capabilities of the low-metastatic FaDu and HSC-3 cells (Fig. 2e, right panel, P < 0.01). In a complementary model, GLUT4 gene silencing significantly reduced the GLUT4 protein levels in HSC-3-M3 and HSC-2 cells, which expressed a high level of endogenous GLUT4 (Fig. 2f, left panel). GLUT4 knockdown in these two cell lines significantly inhibited the migratory/invasive capabilities of the highly metastatic HSC-3-M3 and HSC-2 cells (Fig. 2f, right panel).

Next, we examined the role of GLUT4 in the promotion of metastasis in vivo using xenograft mouse models by intravenously injecting GLUT4-overexpressing and vector-control FaDu cells into mice. Six weeks after injection, the lungs were removed and examined for metastatic foci. Mice injected with GLUT4-overexpressing FaDu cells exhibited significantly higher numbers of metastatic foci compared to the vector control group by gross and histopathological examinations (Fig. 3a). There was a 4-fold increase in foci number in the GLUT4 overexpression group compared to the vector control group (Fig. 3b, P < 0.001). To mimic clinical HNSCC metastasis, we established an orthotopic xenograft HNSCC mouse model by injecting mice intrabuccally with luciferase-expressing FaDu cells that expressed either the GLUT gene or a vector control. Our results showed that stronger bioluminescence could be observed in 4 out of 5 mice injected with the GLUT4-overexpressing cells compared to only 1 mouse exhibiting weak bioluminescence in the vector control group (Fig. 3c). The average bioluminescence counts were obtained from the neck lymph nodes of all 10 mice, and the results showed that the GLUT4-overexpressing group had significantly higher counts compared to the vector control group (Fig. 3d, P < 0.05). In addition, we also established a xenograft model by subcutaneous injection of GLUT4-overexpressing FaDu cells. We observed that GLUT4 did not significantly increase the tumorigenicity of FaDu cells in vivo, and this result is consistent with the cell proliferation rate in vitro (Additional file 1: Figure S3). These data suggest that GLUT4 overexpression promotes HNSCC metastasis in vivo and in situ.

To determine whether the GLUT4-mediated promotion of HNSCC cell migration and invasion requires glucose transporter activity, we screened glucose uptake and lactate production in a panel of HNSCC cells. The data showed that metabolic events may be correlated with metastasis ability in several cell lines (Additional file 1: Figure S4), but no significant P values were obtained. Therefore, we added the glucose transport inhibitors ritonavir and indinavir to block glucose transport efficiency in a GLUT4-overexpressing cell model. We first used 2-NBDG treatment to demonstrate that ritonavir did indeed block transporter function. We observed the uptake of the glucose analog 2-NBDG by its autofluorescence. The GLUT4-overexpressing FaDu and HSC-3 cells treated with ritonavir had lower fluorescence counts (Fig. 4a, P < 0.01, left panel and P = 0.016, right panel) than did the vector-control FaDu and HSC-3 cells (Fig. 4a, P = 0.021, left panel and P < 0.01, right panel). We further confirmed the decrease in glucose uptake after inhibitor treatment by analyzing the culture medium (Fig. 4b). However, the results showed that ritonavir/indinavir did not significantly reduce the GLUT4-induced migration and invasion abilities of FaDu and HSC-3 cells compared to control cells (Fig. 4c). These results suggested that GLUT4 promotes HNSCC cell migration and invasion only partially through the transportation of glucose to the cancer cells.

To determine whether another novel pathway or network by plays a transporter function-independent role in the GLUT4-mediated promotion of HNSCC cell migration and invasion, we next performed a microarray analysis using the low-metastatic HNSCC FaDu cells with or without GLUT4 overexpression. The normalized data from the microarray database analysis were subjected to Ingenuity Pathway Analysis (IPA) to identify molecules that are activated upon GLUT4 overexpression in FaDu cells. The results showed that the transcription factor TRIM24 is the top predicted candidate to be activated in response to GLUT4 overexpression, as verified by the transcriptional activity of its target genes with a Z-score of 2.868 and P value =1.55E-05 (Fig. 5a and Additional file 1: Table S2). The top 11 activated transcription factors with Z-scores higher than 2 and their respective downstream genes are shown in Additional file 1: Table S2. Similarly, the top 7 inhibited transcription factors and their respective downstream genes are shown in Additional file 1: Table S3.

Because TRIM24 was the most activated transcription factor upon GLUT4 overexpression in FaDu cells, we compared our GLUT4 microarray datasets with the TRIM24-related signature obtained by IPA analysis. Our results showed that DDX58 and OASL were the most significantly downregulated transcriptional targets of TRIM24 (Fig. 5b; −2.42- and −2.99-fold for DDX58 and −2.96- and −2.34-fold for OASL in Additional file 1: Table S4). We further validated the expression levels of DDX58 and OASL in cell models of GLUT4 overexpression and knockdown. Our western blot results showed that DDX58 and OASL were downregulated in GLUT4-overexpressing FaDu and HSC-3 cells (Fig. 5c, left panel) and that the knockdown of GLUT4 expression in HSC-2 cells resulted in the higher expression of the DDX58 and OASL proteins (Fig. 5c, right panel). We further knocked down the gene expression of DDX58 and OASL by their respective shRNAs in the GLUT4-knockdown highly metastatic HSC-2 cells. A western blot analysis showed that the DDX58 and OASL protein expression was significantly reduced (Fig. 5d). The subsequent knockdown of DDX58 could significantly restore the migration potential of GLUT4-knockdown HSC-2 cells by 1.5-fold (Fig. 5e, P < 0.001); however, OASL knockdown did not restore the migration capability (Additional file 1: Figure S5). These data suggested that DDX58 may be the primary negatively regulated downstream target of TRIM24 mediating the GLUT4-induced HNSCC cell migration. This in vitro inverse relationship of GLUT4 and DDX58 expression was then validated in in silico clinical HNSCC cohorts. The results showed that high GLUT4 RNA expression in combination with low DDX58 RNA expression levels was significantly correlated with the worst HNSCC patient survival (Fig. 5f and Additional file 1: Figure S6, P = 0.029, P < 0.001, respectively).