Dysregulation of RUNX2/Activin-A Axis upon miR-376c Downregulation Promotes Lymph Node Metastasis in Head and Neck Squamous Cell Carcinoma Free

Head and neck squamous cell carcinoma (HNSCC) is one of highly prevalent and lethal cancers in the world, including Taiwan (1, 2). More than 90% of HNSCCs are oral squamous cell carcinomas, which arise in the oral cavity (3). The initiation and progression of HNSCCs are highly correlated with the exposure to carcinogens, such as cigarette and alcohol. In southeast Asia and Taiwan, betel chewing has been considered as a major risk factor for HNSCC (3). Although cancer therapies have tremendously improved over the past 20 years, the prognosis of patients with HNSCC remains unfavorable (4, 5). Several recent studies have reported single-molecule biomarkers, such as IGF1R (6), Axl (7), and EpCAM (8) for predicting HNSCC progression; however, their practical usefulness remains to be determined. By using genome-wide analysis, this study was thus attempted to identify the critical molecule(s) as a diagnostic biomarker or even therapeutic target for metastatic HNSCC.

RUNX2, a Runt family transcription factor, is a key regulator physiologically associated with osteoblast differentiation and chondrocyte maturation (9, 10), as well as cytoskeleton remodeling and cellular movement via activating its downstream genes. Recently, RUNX2 plays as a pivotal molecule in forcing breast and prostate cancer progression, for example, bone metastasis, through the transcriptional regulation of MMPs, VEGFs, SNAIL, SLUG, and PTHrP expression (11). Nevertheless, the functional consequences of RUNX2 upregulation and its downstream transcriptome are still unclear in HNSCC progression.

Homodimer INHBA forms a polypeptide hormone, known as activin A, which is a pluripotent growth and differentiation factor (12). In addition to its physiologic functions, activin A appears to display multiple roles in tumorigenesis and cancer progression. Upregulation of activin A has been reported in esophageal, gastric, colon, and lung cancer (13–16) and promotes in vitro invasion and epithelial–mesenchymal transition (EMT) in the oral cancer (17). However, the mechanism for INHBA regulation remained uncertain in HNSCC.

miRNAs are small noncoding RNAs that posttranscriptionally control gene expression. miRNAs use seeding regions to match miRNA response elements (MRE) of its target mRNAs and fine-tune target expression through mRNA degradation or translational inhibition (18). In mammals, approximately 30% of protein-coding genes are regulated by miRNAs and a single miRNA can regulate hundreds to thousands of targets (19). Although miRNAs have fine-tuning roles in cell, it can make a huge change when targeted to key biological molecules, such as transcription factors, upstream receptors, or hub proteins. Aberrant miRNA expression was considered to be a common phenotype in tumorigenesis and cancer progression. More than 50 miRNAs are located in 14q32.31 behind the long intervening noncoding RNA (lincRNA) meg3 and are regulated coordinately (20). Here we find that miR-376c-3p, which is located in chromosome 14q32.31, is profoundly downregulated in HNSCC and thereby leads to the dysregulation of RUNX2/activin-A axis. Significantly, this scenario strongly correlates with the mechanism for lymph node metastasis and poor prognosis in HNSCC.

Cell cultures were prepared and maintained according to a standard protocol. HOKs were obtained from ScienCell and passaged according to the manufacturer's instructions. 293T and Cal-27 cells were obtained from ATCC. Ca9-22 and SAS cells were obtained from JCRB Cell Bank in 2014 and maintained according to the manufacturer's instructions. All culture media were supplemented with 10% FBS (Invitrogen), 1,000 U/mL penicillin and 100 μg/mL streptomycin, and 2 mmol/L GlutaMAX at 37°C in a humidified atmosphere of 95% air and 5% CO₂. All cells were routinely authenticated on the basis of morphologic and growth characteristics as well as by short tandem repeat (STR) analysis and confirmed to be free of mycoplasma.

Paired RNAs from tumor specimens and adjacent noncancerous epithelia were obtained from surgeries performed between 1999 and 2010 at the National Cheng Kung University Hospital (Tainan, Taiwan). Fresh/frozen tissues were preserved in liquid nitrogen until use. The study protocol was reviewed and approved by the Institutional Human Experiment and Ethics Committee of National Health Research Institutes (HR-97-100). HNSCC tissue microarrays were collected from Taipei Medical University Hospital (Taipei, Taiwan) with Institutional Review Board (IRB) approval (TMU-IRB 99049), and then fixed in formalin and embedded in paraffin before archiving. Archived specimens were spotted onto tissue microarrays on Dako-coated slides before being used for immunohistochemical staining or in situ hybridization. The histologic diagnosis of the type of head and neck cancer was performed according to the recommendations of the WHO classification. Tumor size, local invasion, lymph node involvement, distal metastasis, and the final disease stage were determined according to the definition of the American Joint Committee on Cancer TNM staging system of HNSCC (21). Follow-up was done for up to 100 months.

Head and neck cancer cDNA microarrays were performed according to our previous studies (20) and analyzed with the core transcription factor of gene set enrichment analysis (GSEA) and upstream regulator of Ingenuity Pathway Analysis (IPA). Briefly, the HNSCC patient cDNA microarrays were divided into two groups, adjacent normal tissues or HNSCC tumors, and analyzed by C3 TFT module of GSEA; then, the core oncogenic transcription factors were ranked by normalized enrichment score (NES). The upstream regulators of IPA were analyzed from the significant differently expressed genes (>2-fold) between adjacent normal tissues and HNSCC tumors, and then analyzed by the upstream regulator of IPA. The prognostic value of single gene or predefined gene set among The Cancer Genome Atlas (TCGA) HNSCC cohort was analyzed through SurvExpress database (23).

The miR-376c-3p expressing, shRUNX2, shINHBA, and a nontargeting silencing miRNA or shRNA control vector were purchased from OpenBiosystems and Gateway donor cDNAs, such as RUNX2, INHBA, and PTHLH, were purchased from DNasu and then recombinated into plenti6.3-DEST (Invitrogen) vector by Clonase LR (Invitrogen). All lentiviral vectors were transfected into the packaging cell line 293T along with the pMD.G and pCMVΔR8.91 plasmids using a calcium phosphate transfection kit (Invitrogen). After 48-hour incubation, the viral supernatants were transferred to the target cells, and the infected cells were cultured in the presence of different concentrations of puromycin or blasticidin (Calbiochem), depending on the cell line and vector backbone. The RUNX2 regulation targets in HNSCC were performed by Affymetrix U133 microarray assays and uploaded onto GEO database (GSE74137).

All of the gene detection assays were conducted as described in our previous publication (20). In the pairwise miR-376c-3p and RUNX2 expression assay, patients' RNAs were reverse-transcribed using the miScript II RT-Kit (Qiagen). Chromatin immunoprecipitation (ChIP) was performed using both the EZ-Zyme chromatin prep kit (Millipore) and ab500 ChIP kit (Abcam) with 10 μg primary RUNX2 antibodies (sc-10758X, Santa Cruz Biotechnology) and equal level of rabbit IgG controls (Millipore). The specific gene and amplicon expression was detected with OmicsGreen. The primer sequences and antibodies are shown in Supplementary Table S1.

The migration and invasion ability of HNSCC cells was determined in a Boyden Chamber Invasion assay (GE Healthcare). The polyvinylidene difluoride (PVDF) membrane pore size was 8 μm. For the invasion assay, PVDF was coated with 1 μg/mL Matrigel Basement Membrane Matrix (BD Biosciences) on the upper side, and 10 μg/mL fibronectin was used to coat the opposite site. In the migration assay, where only fibronectin coating was used, 1.5 × 10⁴ cells were seeded into the upper medium. After incubation for 16 hours at 37°C, the invaded cells were stained with Giemsa stain (Sigma-Aldrich). At least five phase-contrast images were finally obtained through microscopy, and the invasive cells were quantified. For each treatment, the cells were seeded in at least triplicate, and the experiments were repeated at least twice.

Ago2 immunoprecipitation (IP) was performed following the protocol described in ref. 20.

The RUNX2-3′UTR-reporter was cloned into the pmiRGLO vector (Promega) at the SacI and XhoI sites, and the 1.5-kb INHBA promoter was cloned into the pGL4.22 vector at the XhoI and HindIII sites. The miRNA-binding site mutation reporter pmiRGLO-RUNX2-mt-376c, which carries the seeding sequence mutation of miRNA-binding sites and harbors a deletion in the RUNX2-binding site, was produced through site-directed mutagenesis, and its sequence was subsequently confirmed.

All animal experiments were performed in strict accordance with the recommendations in the guidelines for the Care and Use of Laboratory Animals of the National Health Research Institutes (Miaoli, Taiwan). The protocol was approved by the Institutional Animal Care and Use Committee of the Genomic Research Center, Academia Sinica (Taipei, Taiwan; protocol no.: AS-IACUC-15-06-833). Male NOD-SCID mice aged 5–6 weeks were bred in the Genomic Research Center. The animals were housed in a climate-controlled room (12:12 dark–light cycle, with constant humidity and temperature) with food and water provided ad libitum. All efforts were made to minimize suffering.

For the lymph node metastasis assay, 1.5 × 10³ HNSCC GL-subline cells were injected into mice through an intrabuccal route. In vivo tumor images were captured with an IVIS Imaging System (Caliper Life Sciences) to measure the signal intensity from the GFP⁺ luciferase vector.

In situ hybridization was performed with the Ishyb In Situ Hybridization Kit (Biochain) following the manufacturer's protocol.

A paired t test was performed to compare the RUNX2 and INHBA IHC expression levels and miR-376c-3p or RUNX2 mRNA level in cancer tissues, and in the corresponding normal adjacent tissues. The association between clinicopathologic categorical variables and the IHC expression levels were analyzed by Pearson χ² test. Estimates of the survival rates were calculated using the Kaplan–Meier method and compared using the log-rank test. Patient follow-up time was censored until the patient was lost during follow-up. For all experiments, bar graphs represent the mean (± SEM) from three independent experiments and statistical analyses were performed using SPSS (Statistical Package for the Social Sciences; SPSS) 17.0 software. Unless otherwise stated, statistical differences between means were determined using a Student t test. A P value of <0.05 was considered significant for all of our analyses.

To investigate the key transcription factor in HNSCC progression, we used the c3 transcription factor targets (TFT) module of GSEA (22) to dissect the 40 pairwise HNSCC cDNA microarrays (GSE37991; ref. 23). We identified 10 key transcription factors that are highly upregulated in patients with HNSCC (Fig. 1A and B). We further analyzed the prognostic significance of the whole predefined gene set of those 10 transcription factors from Molecular Signatures Database in Taiwanese cohort and TCGA HNSCC cohort. SPI1, BACH2, TFCP2, NFE2, JUN, and BACH2 had higher HRs and significant P values in both cohorts (Table 1). We further used the IPA to analyze upstream regulators in Taiwanese HNSCC patients, and we found that RUNX2 was the most activated upstream regulator in patients with HNSCC (Supplementary Table S2). RUNX2 was also the only transcription factor that was presented in both analyses. Thus, the activation of RUNX2 might be important for HNSCC progression. Immunocytochemistry (ICC) revealed that RUNX2 massively accumulates in the nucleus probably implying a constitutively activated status of RUNX2 in HNSCC cells (Supplementary Fig. S1). Remarkably, our results showed that most of HNSCC cells express RUNX2 much higher than primary oral keratinocytes (HOKs; Fig. 1C and D). Kaplan–Meier analysis revealed that RUNX2 protein serves as a poor prognostic marker in an independent validation cohort (Fig. 1E). Clinical pathologic results also elucidated that RUNX2 significantly correlates with perineural invasion (P = 0.001) and marginally associates with N status (N1–3 vs. N0, P = 0.06; Supplementary Table S3). Accordingly, RUNX2 mRNA level also appeared to be a poor prognostic marker in TCGA HNSCC cohort deposited in the SurvExpress database (24) (Supplementary Fig. S2). These results demonstrated that RUNX2 is likely hyperactivated/upregulated and acts as a poor prognostic marker in HNSCC.

To understand the correlation between RUNX2 expression and metastatic potential in HNSCC cells, we determined the endogenous RUNX2 levels and cellular migration/invasion abilities in a panel of HNSCC cells. The data showed that RUNX2 level causally associates with in vitro migration and invasion abilities of HNSCC cells (Fig. 2A and B) and lung colony formation ability (Supplementary Fig. S3; Supplementary Table S4). Next, we enforcedly expressed the exogenous RUNX2 gene in Ca9-22 cells with less endogenous RUNX2 to validate its effects on promoting HNSCC metastatic progression. Our results revealed that RUNX2 effectively promotes cellular migration/invasion abilities in Ca9-22 cells (Fig. 2C). In contrast, we performed RUNX2 knockdown in Cal-27 and SAS cells that highly express RUNX2. Silencing RUNX2 significantly inhibited in vitro cellular migration/invasion abilities in both Cal-27 and SAS cells (Fig. 2D and E) and suppressed the in vivo lymph node metastasis judged by the reduced luminescent signals in tumor-bearing mice (Fig. 2F and G). Moreover, RUNX2 knockdown also decreased the lung colonization ability of HNSCC cells in mice (Fig. 2H and I). These results demonstrated that RUNX2 overexpression fosters the metastatic potential of HNSCC cells.

To further investigate the role of RUNX2 transcription targets in HNSCC metastasis progression. We used three different approaches to identify specific RUNX2 targets in HNSCC (Fig. 3A). First, to analyze the transcriptome of RUNX2, we performed a microarray analysis against RUNX2-overexpressing Ca9-22 cells and then found 538 genes with 2-fold changes after RUNX2 overexpression. In addition, we employed the RUNX2 ChIP sequence data obtained from ChEA, in which 3,423 genes were validated to be regulated by RUNX2 (25). We also identified 2,884 genes that display 2-fold upregulation with a high concordance (>50%) in HNSCC tumors. Combining these approaches, we identified 26 RUNX2 downstream genes that are obviously upregulated in HNSCC (Supplementary Table S5). Then, we used the TCGA HNSCC cohort to examine the prognostic power of those genes. The data showed that INHBA, as well as PTHLH, exhibited as the most powerful predictor for poor prognosis in TCGA HNSCC cohort. However, INHBA, but not PTHLH, effectively promoted the in vitro migration and invasion abilities in Ca9-22 cells (Supplementary Fig. S4). Thus, the next experiments were designed to examine the metastasis-promoting effect of RUNX2/INHBA axis in HNSCC cells. Enforced ectopic expression of RUNX2 gene was found to promote INHBA expression in both Ca9-22 and SCC-9 cells (Fig. 3B and C) and also regulated INHBA promoter activity assay in Ca9-22 cells (Fig. 3D). Chromatin immunoprecipitation quantitative PCR (ChIP-qPCR) experiment was performed and the results showed the direct binding of RUNX2 to INHBA promoter (Fig. 3E). The deletion of RUNX2-binding site in the promoter region of INHBA suppressed the INHBA promoter activity, which indicates a direct regulation of RUNX2 on the transcriptional activity of INHBA gene at the RNA level (Fig. 3D). Restoring INHBA expression in RUNX2-silencing SAS cells rescued the cellular migration and invasion abilities (Fig. 3F). Conversely, INHBA knockdown in RUNX2-overexpressing Ca9-22 cells compromised the RUNX2-promoted cellular migration and invasion abilities (Fig. 3G). In Ca9-22 cells, the activin/inhibin family genes, only INHBA, not INHA or INHBB, were stimulated by RUNX2 overexpression. Thus, we used the homodimer INHBA, activin-A recombinant protein, which is able to stimulate Ca9-22 migration and invasion (Fig. 3I). Moreover, in Cal-27 and SAS cells that have high endogenous RUNX2 expression, INHBA knockdown prevented the in vitro migration and invasion (Fig. 3H), as well as in vivo cervical lymph node metastasis of SAS cells (Fig. 3J and K). Furthermore, we found a positive correlation of RUNX2 with INHBA in the 2016 HNSCC TCGA dataset that high-level of RUNX2 and INHBA expression strongly predicted a worst clinical outcome (Supplementary Fig. S5) and might correlate to HPV-negative head and neck cancer (Supplementary Table S6). These findings demonstrated that RUNX2/INHBA axis potentially dominates the metastatic progression in HNSCC.

To ascertain the possible mechanism for RUNX2 dysregulation, we surveyed the configuration of miRNA-binding sites within the 3′-untranslated region (3′UTR). We found that miR-376c-3p is significantly silenced in clinical HNSCC samples (Supplementary Table S7; refs. 20, 26) and aligned within the 3′UTR of the RUNX2 gene (Fig. 4A). Dramatically, the miR-376c-3p expression was shown to negatively correlate with the malignancy of HNSCC cells (Fig. 4B). Moreover, the mutation of miR-376c-3p–binding site predominantly restored the transcriptional activity of RUNX2 in SAS cells (Fig. 4C; Supplementary Fig. S6). Other HNSCC transcription factors, such as SMAD4, BACH2, and HIF-1A, also have the miR-376c–binding sequences in its 3′-UTR regions (Supplementary Table S8). The ago2-dependent immunoprecipitation results also revealed that the Ago2–RUNX2–miR-376c-3p complex was enriched in miR-376c–expressing cells compared with scramble miRNA–expressing cells (61.15 ± 1.97-fold, P < 0.001) and HIF-1A was also significant enriched (2.08 ± 0.05-fold, P < 0.01) compared with SMAD4 and BACH2. Moreover, miR-376c-3p expression showed an inverse correlation with the RUNX2 mRNA levels in HNSCC patients (Fig. 4E). These findings demonstrated that miR-376c-3p directly regulates RUNX2 expression in HNSCC cells.

Next, we enforcedly expressed an ectopic miR-376c and then found a dramatic decrease in RUNX2 (Fig. 5A) and INHBA (Fig. 5B) expression, as well as the cellular migration/invasion abilities (Fig. 5C and D) in Cal-27 and SAS cells. Notably, miR-376c overexpression prevented the lymph node metastasis of SAS cells in an orthotopic animal model (Fig. 5E and F). In contrast, after treating Ca9-22 cells with miR-376c antisense oligonucleotides, we found significantly enhanced expression levels of RUNX2 and HIF-1A proteins at 24 hours and the levels sustained till 72 hours. However, INHBA protein expression was found to be subsequently expressed 48 hours after the anti-miR-376c treatment (Fig. 5G). The cellular migration/invasion abilities in Ca9-22 cells were also affected correspondingly (Fig. 5H and I). Taken together, our data showed that enforced expression of 3′UTR-free ectopic RUNX2 gene strongly restores the miR-376c–suppressed cellular migration/invasion abilities in SAS cells (Fig. 5J and K). Anti-miR-376c treatment could increase the cellular migration/invasion abilities in Ca9-22 cells and then subsequent knockdown of RUNX2 in those cells could deplete the migration/invasion abilities from anti-miR-376c treatments (Supplementary Fig. S7). These results indicated that RUNX2 is an important downstream target of miR-376c–induced HNSCC progression.

Significantly, IHC results revealed that INHBA expression inversely correlates with miR-376c-3p levels (Fig. 6A) and predicts poor prognosis in HNSCC (Fig. 6B). Therefore, miR-376c downregulation likely leads to the dysregulation of RUNX2/INHBA axis and eventually promotes HNSCC metastatic progression.

Here we demonstrate that the loss of miR-376c-3p in HNSCC promotes lymph node metastasis because of the dysregulation of RUNX2/INHBA axis. Although the posttranscriptional regulation of miR-376c toward RUNX2 in malignancies other than HNSCC remains to be verified, RUNX2 has been shown to be the key transcription factor in controlling the cancer metastasis in breast and prostate cancer (11, 27). Other well-known RUNX2 downstream genes, such as MMP-9, TGFB1, and Snail, appeared to enhance cellular metastatic ability; however, these genes were not markedly stimulated by RUNX2 in Ca9-22 cells. This indicated that the RUNX2 may have specific transcription target in HNSCC. On the other hand, we determined that INHBA acts as a master downstream gene of RUNX2 in promoting cancer metastasis, even though PTHLH also displayed as a stronger poor prognostic marker in HNSCC. RUNX2 is a multipotent transcription factor that regulates hundreds of genes (25) and also has direct protein–protein interaction with other HNSCC oncogenic transcription factors, such as JUN (28), HIF-1A (29), and TP53 (30) in other cancer cell types. In HNSCC, RUNX2 may also have a cooperated role or cross-talk with other transcription factors to control oncogene functions and cancer progression (Supplementary Table S9). Our study provides a prometastatic function of RUNX2 that may contribute to HNSCC cervical metastasis but also identifies its novel transcriptional targets INHBA in HNSCC.

INHBA has been found to interplay both oncogenic and tumor suppressor roles in cancer (31). INHBA was appeared to associate with a mechanism for promoting cancer initiation, cachexia, and drug resistance (32–34). Upregulation of INHBA also promotes EMT process (35, 36) via suppressing E-cadherin mRNA level, but stimulating mesenchymal molecule expression, such as N-cadherin and Vimentin mRNA expression in HNSCC cells (17). Here we further show that INHBA acts as a key effector in HNSCC lymphatic metastasis. A similar aspect that INHBA is referred to as a predictor for lymph node metastasis and poor overall survival in HNSCC was reported recently by Kelner and colleagues (37). Although the practical application of activins has been proposed to treat cancers previously, none of activin agents are available in current cancer therapy (38). Our findings may offer a feasibility of activin-A agents for combating the metastatic HNSCC.

Previously, miRNAs have been developed for several types of translational applications, such as prognostic, diagnostic, and predictive biomarkers (39). Recently, a miR-34a–based cancer therapy was entered into a phase I clinical trial for the treatment of liver cancer (40). In addition to act as a negative regulator of RUNX2/INHBA axis, miR-376c-3p also controls the cellular expression of activin receptors, ALK-5, and co-SMADs, Smad4 (41, 42). It demonstrates that downregulation of miR-376c-3p promotes the cancer progression in HNSCC probably due to the loss of regulation in several oncogenic signaling cascades. Although, RUNX2, HIF-1A, SMAD4, and BACH2 have the miR-376c–binding elements on its 3′-UTR regions, however, RUNX2 was the most predominant transcription factor in HNSCC cells. Ago2–HIF-1A–miR-376c complex was found to have a significant increase in HNSCC cells (Fig. 4D). HIF-1A has been reported to promote INHBA expression under hypoxia in chondrocytes (43). There is a possibility that HIF1A can transactivate INHBA expression in HNSCC. Indeed, in our experimental results, similar to RUNX2, HIF-1A protein was found to increase, to a lesser extent, at 24 hours after anti-miR-376c treatment. These results suggest that both RUNX2 and HIF1A transcription factors may be able to increase the transcriptions of INHBA RNA (Fig. 5G).

Betel nut exposure induces meg-3 promoter hypermethylation in chromosome 14q32.31 and coordinately downregulates the expression of miRNAs including miR-376c (20). Here we find that the loss of miR-376c-3p expression increases RUNX2 and its transcriptional target, INHBA, in HNSCC. The overexpression of miR-376c-3p was capable of inhibiting the in vitro cellular migration/invasion abilities and in vivo lymphatic metastasis in HNSCC cells. Moreover, the downregulation of miR-376c was extensively found in HNSCC tumor compared with the adjacent normal tissues as determined by IHC analysis and RNA-sequencing results from the TCGA database (44). Similar views were also observed in metastatic melanoma and prostate cancer cells (45, 46). Therefore, these findings demonstrate the tumor-suppressive function of miR-376c, particularly in preventing metastatic progression in HNSCC.

Directly targeting transcriptional dysregulation is an emerging approach bypassing cancer clonal heterogeneity in modern cancer therapy (47). On the basis of our findings, we suggest that the replenishment of miR-376c, therapeutic targeting of RUNX2 transcriptional activity or the immunoneutralization of extracellular activin A could be a new strategy to combat metastatic evolution in HNSCC.

No potential conflicts of interest were disclosed.

Conception and design: W.-M. Chang, J.-Y. Chang, Y.-S. Shieh, M. Hsiao, S.-G. Shiah

Development of methodology: W.-M. Chang, C.-Y. Su, T.-C. Lai, J.-R. Hsiao, Y.-S. Shieh

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W.-M. Chang, C.-Y. Su, Y.-C. Chang, J.-R. Hsiao, C.-L. Chen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W.-M. Chang, Y.-F. Lin, C.-Y. Su, H.-Y. Peng, T.-C. Lai, G.-H. Wu, Y.-M. Hsu, L.-H. Chi, Y.-S. Shieh, M. Hsiao, S.-G. Shiah

Writing, review, and/or revision of the manuscript: W.-M. Chang, Y.-S. Shieh, M. Hsiao, S.-G. Shiah

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.-L. Chen

Study supervision: Y.-S. Shieh, M. Hsiao, S.-G. Shiah

The authors would like to acknowledge the great help and assistance of Experimental Animal Imaging and Molecular Pathology Core Facilities of Genomic Research Center, Academia Sinica (Taipei, Taiwan).

This work was supported by National Health Research Institutes (NHRI) grants from Taiwan (NHRI-CA103-PP-02, NHRI-CA104-PP-03) and the Ministry of Science and Technology (MOST) grant from Taiwan (MOST-104-2314-B-400-018 to S.-G. Shiah). This study was also supported by Academia Sinica and Ministry of Science and Technology grants MOST 104-0210-01-09-02 and MOST 105-0210-01-13-01 to M. Hsiao. The head and neck cancer tissue array construction and related works were supported by Health and Welfare surcharge on tobacco products (DOH102-TD-C-111-008) from the Ministry of Health and Welfare to Comprehensive Cancer Center of Taipei Medical University.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.