MiR-34a suppresses amphiregulin and tumor metastatic potential of head and neck squamous cell carcinoma (HNSCC).

MiR-34a is a well-known tumor metastasis inhibitor, but only a few target genes involved in metastasis have been identified. In HNSCC, the role of miR-34a in metastasis has not been fully elaborated, and the target gene of miR-34a is still blind. Here we addressed that, the relative lower expression of miR-34a is associated with HNSCC lymphatic metastasis. HNSCC metastasis was found to be strongly suppressed in vitro and in vivo by over-expressing miR-34a. In order to screen the possible target genes of miR-34a in HNSCC, a microarray-based differential mRNA profiling mediated by miR-34a over-expression was performed, and AREG was identified as a pivotal target. We demonstrated that the mRNA and protein levels of AREG were greatly reduced when forcing miR-34a expression. The correlation between AREG mRNA levels and HNSCC metastatic phenotype was also significant in HNSCC tissues (p < 0.01). Moreover, the results of luciferase assay provided the further evidence that miR-34a degraded AREG mRNA through targeting the 3'-UTR site. Restoration of AREG expression partially rescued miR-34a-mediated cell invasion defects in vivo and in vitro. Additionally, Over-expressing miR-34a greatly reduced EGFR and uPA, which were reversed by re-expression of AREG. Taken together, these findings indicate that miR-34a targets AREG, and is essential in inhibition of HNSCC metastasis.


INTRODUCTION
Head and neck squamous cell carcinoma (HNSCC) is an invasive epithelial neoplasm with a propensity to early and extensive lymph node metastases [1]. Epidemiological statistics show that HNSCC represents the sixth most common lethal malignancy by incidence which affects approximately 650,000 people worldwide annually and leads to above 350,000 deaths [2]. Because of the multifactorial etiopathogenesis of HNSCC, the development of multimodality treatments is suboptimal in recurrent and/or metastatic disease and only half of the patients are alive in 5 years [3,4]. Despite of the advances in primary HNSCC treatment, the lack of good molecular targets and biological heterogeneity hinder the development of gene therapy in metastatic HNSCC [4,5]. Therefore the investigation of new biomarkers, based on the biology of the disease, is of major importance to HNSCC metastasis.
MicroRNAs, a small non-coding RNAs consisting of 20-22 nucleotides in length, have emerged as a new hotspot for anti-cancer therapy [6,7]. By base-pairing complimentary or uncomplimentary sequences of 3'UTR of target mRNAs, microRNAs regulate gene expression in the post-transcription stages via mRNAs degradation or transcription inhibition. It has been suggested that miRNAs are capable to modulate a number of oncogenes or tumor suppressor genes and participate in the regulation of malignant transformation, progression of disease, and metastatic colonization [8]. Recently, a group of www.impactjournals.com/oncotarget microRNAs have been identified altered in HNSCC and served as biomarkers for tumorigenesis, clinical prognosis, and metastasis [9][10][11].
MiR-34a, the most prevalent form of microRNA-34 family, is located in the region of chromosome lp36. 23 and acts as a tumor-suppressor in variety of tumors [12]. It has been shown to represent direct target of p53 and repress several oncogenes directly or indirectly, resulting in cell-cycle arrest, senescence, apoptosis and reducing chemoresistance [13][14][15], and a miR-34 mimic has become the first microRNA to reach phase 1 clinical trials [12]. Recently, miR-34a is also reported as a tumor metastasis inhibitor in some solid cancers, but the target genes involved in metastasis have not been fully elucidated, and those targets vary in different cancer types [16][17][18][19][20][21][22][23]. To our knowledge, only one report has observed that miR-34a decreased HNSCC cell mobility, which might indicate its suppressing function in HNSCC metastasis [24]. However the role of miR-34a in metastasis of HNSCC has not been clearly elaborated, and until now, no target gene of miR-34a is determined in HNSCC. Therefore, exploring the function role and the possible target genes of miR-34a in HNSCC is an essential way to understand the molecular mechanism of miR-34a in tumor metastasis.
Human amphiregulin (AREG), synthesized as a 252-amino acid transmembrane precursor, pro-Areg, is an important ligand of the epidermal growth factor receptor [25]. A large number of information was available to support a role for AREG in tumor development and cancer cell biology. Studies have been demonstrated that over-expression of AREG participated in the maintenance of the oncogenic and metastatic properties in several solid malignancies through activation of EGFR families [26][27][28][29]. However, despite numerous identified stimuli, the epigenetic alterations responsible for the post-transcriptional regulation of AREG are not fully clarified [30], and information on the regulation of AREG expression by miRNAs is lacking [31].
In this study, we have identified that miR-34a acted as an essential metastatic suppressor in HNSCC with AREG as a novel target. Suppression of AREG by ectopic miR-34a expression prevented tumor invasion, and inhibited EGFR, which might be the potential molecular targets for future HNSCC therapy.

The expression of miR-34a in HNSCC samples
In order to evaluate the expression of miR-34a in HNSCC tissues, real-time RT-PCR was performed in the human fresh specimens obtained from 40 primary HNSCC tumors and the adjacent normal epithelial tissues. About 42.0-fold to 1.3-fold reduction of miR-34a expression was found in 29 of 40 (72.5%) HNSCC tissues, compared with the adjacent normal epithelia ( Figure 1A).
To determine the association between miR-34a expression and HNSCC metastatic potential, we investigated miR-34a expression in lymphatic metastasis samples and non-metastatic samples. Statistically, in the 29 cases of miR-34a reduction samples, 19 of which (65.5%) were associated with lymph node metastasis; 10 of which (34.5%) were associated with non-metastasis. In contrast, in the 11 cases of miR-34a increase samples, only 3 of which (27.3%) were associated with lymph node metastasis; 8 of which (72.7%) were associated with non-metastasis  Figure 1A). To normalize the relative expression value of miR-34a in cancer tissues, the miR-34a expression level in the corresponding matched adjacent normal epithelial tissues was defined as 1. Statistical analysis showed that the relative expression of miR-34a in metastatic samples (0.5685 ± 0.1847, mean ± SEM) was significantly lower than that in non-metastatic samples (1.274 ± 0.4602 mean ± SEM) ( Figure 1B, P = 0.0006).

MiR-34a suppresses migration and invasion of HNSCC cell lines in vitro
To further elucidate the function of miR-34a in the metastatic process of HNSCC, four HNSCC cell lines (Fadu, SCC-15, UM-SCC-23, and Cal27) were stably over-expressed miR-34a using lentiviral vector. The miR-34a expression was enhanced dramatically in miR-34a transfected cells compared with the control ( p < 0.01, Supplementary Figure 1). The ectopic miR-34a expression caused a significant reduction of cell invasion capability ( Figure 2A). Statistically, over-expression of miR-34a resulted in 7.64-fold decrease in Fadu cell invasion, 2.23-fold decrease in SCC-15 invasion, 5.84-fold decrease in UM-SCC-23 invasion, and 2.81-fold decrease in Cal27, compared with the cells transfected with control vector (P < 0.01, Figure 2B).

MiR-34a reduces metastatic potential of HNSCC cells in vivo
To testify whether ectopic miR-34a could inhibit cell metastasis in vivo, Fadu cells over expressing miR-34a or transfected with control vector were injected into the tail vein of nude mice. Two months after tail vein injection, a dramatic effect on the size of eventually formed lesions and the weight of lung with metastases were observed. Fadu-miR-34a generated fine and scattered metastatic nodes, while the control cells (Fadu-miR-control) resulted in massive and confluent metastatic nodes ( Figure 3A). Statistical analysis showed that the weights of mice lungs with metastatic nodes from Fadu-miR-34a cells shrunk about 1.5-fold compared with that from control cells ( Figure 3B). The pathological changes of the bilateral lungs with HE staining were observed through light microscope at the same time. The tumor nests derived from control cells exhibited large area of lung tissue destruction and/or necrosis, while cells over-expressing miR-34a formed smaller and fewer tumor nests ( Figure 3C).

AREG is a direct target of miR-34a
To identify whether the expression of AREG is suppressed by miR-34a, the mRNA expression level of AREG gene was screened by qRT-PCR in four HNSCC cell lines. The results showed that AREG-transcript levels had a general reduction in HNSCC cell lines with ectopic miR-34a expression compared with control (9.1-fold decrease in Fadu-miR-34a, 2.2-fold decrease in SCC-15-miR-34a, 4.9-fold decrease in UM-SCC-23-miR-34a, and 3.4-fold decrease in Cal27-miR-34a) ( Figure 5A). Furthermore, western blot analysis confirmed the ectopic miR-34a expression causing a remarkable reduction of AREG protein amounts in those cell lines ( Figure 5B).

Re-expression of AREG reverses miR-34aimposed metastasis defects in vivo
To ascertain whether restored AREG reversed in vivo tumor metastasis defects caused by miR-34a, stable clones of Fadu-miR-34a-AREG and Fadu-miR-34a-control cells were injected into the tail vein of nude mice, respectively. Two months later, the macro and microscopic changes of the mice bilateral lungs were evaluated. In consonance with our earlier findings, ectopic miR-34a generated fine and scattered metastatic nodes; in contrast, Fadu-miR-34a-AREG cells generated masses of metastatic nodes in mice lungs ( Figure 7A and 7C). Statistical analysis showed that the weights of mice lungs with metastatic nodes from Fadu-miR-34a-AREG cells were about 2.8-fold higher than those from control cells ( Figure 7B).

EGFR are inhibited by ectopic miR-34a through suppressing AREG
The 32 co-down-regulated genes affected by forced miR-34a expression were performed by a KEGG pathway analysis. The result showed that the differentially regulated genes involved in P53 signaling pathway (p = 1.43E-05) and ErbB signaling pathway (p = 2.86E-05) were overrepresented among the down-regulated mRNAs (log2 fold change ≤ -0.3, p < 0.01). Besides the well-known P53 signaling pathway, the influence of miR-34a in ErbB pathway is also important ( Figure 8A).
To identify which of the ErbB family members were inactivated by ectopic miR-34a expression through inhibiting AREG, four EGF receptors (EGFR, ErbB2, ErbB3 and ErbB4) were detected. In HNSCC cell lines, the expression level of EGFR (ErbB1) was extremely   high, in contrast, ErbB4 were fairly low. Forced miR-34a expression significantly decreased EGFR and ErbB3 mRNA levels in Fadu cell line, while only EGFR was remarkably suppressed in UM-SCC-23 cell line ( Figure 8B and 8C). The changes of ErbB2 and ErbB4 mRNA levels were very slight in response to miR-34a over-expression. Interestingly, re-expression AREG in miR-34a over-expression cell lines not only dramatically elevated EGFR, but also ErbB3 and ErbB4 mRNA levels ( Figure 8B and 8C).

Expression of AREG is partially inversely correlated with miR-34a in HNSCC samples and associated with metastatic phenotypes
To investigate whether AREG expression was inversely correlated with miR-34a in HNSCC tissues, the mRNA expression of AREG was evaluated by realtime RT-PCR in the 40 primary HNSCC tumors and the corresponding adjacent normal epithelial tissues. There were 29 of 40 (72.5%) HNSCC samples showed opposite expression trend between AREG and miR-34a ( Figure 10A). Statistically, the relative AREG expression was negatively correlated with miR-34a in the 29 samples ( Figure 10B, P = 0.0132), although general inverse correlation between miR-34a and AREG expression in all 40 tumor tissues was not observed ( Figure 10C, P = 0.0745).
To determine the association between AREG expression and HNSCC metastatic potential, we investigated AREG mRNA level in lymphatic metastasis samples and non-metastatic samples. The relative   expression level of AREG in metastatic samples (2.147 ± 0.4063, mean ± SEM) was much higher than that in non-metastatic samples (0.9086 ± 0.2493, mean ± SEM) ( Figure 10D, P = 0.0169). Furthermore, the reversed expression pattern between miR-34a and AREG was found to be closely related with HNSCC lymph node metastasis. It was shown that miR-34a level was decreased and AREG mRNA level was simultaneously increased in 18 samples, 13 of which (72.2%) showed lymph node metastasis. On the other hand, 11 samples co-expressed increased miR-34a and decreased AREG, 3 of which (27.3%) exhibited lymph node metastasis ( Figure 10A). The results indicate that patient who loses miR-34a expression meanwhile over-expresses AREG, tends to exhibit highly potential of lymphatic metastasis, and vice versa.

DISCUSSION
Head and neck squamous cancer is a heterogeneous oncological setting typically appears in the oral cavity, oropharynx, hypopharynx, or larynx. Local invasion and lymph node metastases are the major causes of recurrence and mortality of HNSCC [32]. For those patients in recurrent/distant metastases stages, multimodality therapy approaches with surgery, radiotherapy, and chemotherapy, have reached their upper limit of tolerability. The mortality rate of patients presented with advanced disease is higher than expected [33]. To date, although cetuximab has been approved for commercial use for molecular-targeted therapy in HNSCC, only a minority of patients with stage III/IV HNSCC have meaningful tumor regression [4,5]. Hence, the investigation of new molecular targets, based on the biology of HNSCC, will be important to improve treatment efficacy. Here, we demonstrated that miR-34a through targeting amphiregulin, a ligand of EGFR, was a distinct signaling pathway in inhibition of HNSCC metastasis, which might be a potential therapeutic target of HNSCC.
MiR-34a is best understood as a pleiotropic anticancer microRNA, which acts majorly on cell-cycle arrest, differentiation, senescence and apoptosis [12,34]. In the last few years, studies further discovered that ectopic overexpression of miR-34a results in the inhibition of metastasis in some solid tumors, including colon cancer [16,35], prostate cancer [17], hepatoma [19], non-small cell lung cancer [18], endometrial carcinoma [36] and osteocarcinoma [37]. In head and neck cancer, one previous study observed that overexpression of miR-34a may decrease the cell migration motility in vitro [24]. This result may suggest a potential effect of miR-34a on HNSCC invasion. However, more detailed evidences should be given to fully elaborate whether the miR-34a has an effective function in blocking HNSCC metastasis. Our present study demonstrated that the miR-34a expression was reduced in HNSCC when compared with the adjacent normal epithelial tissues. Moreover, the downregulation of miR-34a was mainly found in tumors with lymph node metastasis, suggesting an association between the loss of miR-34a and the potential of HNSCC lymph node metastasis. Moreover, forced expression of miR-34a dramatically decreased HNSCC cells invasion in vitro, and strikingly impaired their capacity to seed lung metastases in vivo. These results further confirmed the effective function of miR-34a in suppressing HNSCC invasion and metastasis.
Identification of miRNA target gene is of important approach to understand the molecular mechanisms of miRNA-mediated tumorigenesis. Up to now, miR-34a is reported to target over fifty genes directly, most of which are involved in inhibition of cell cycle and apoptosis [38][39][40]. There are several targets of miR-34a related to metastasis have been identified, including Fra-1 [16], AXL [18], CD44 [17], c-Met [19], SNAIL [41], ARHGAP1 [21], AXIN2 [22], c-Myc [23] and YY1 [42]. It is noted that miR-34a usually targets different genes in different tumor types, suggesting miR-34a may involve in distinct signaling pathway in specific tumor. In our study, in order to screen the possible target genes of miR-34a in HNSCC, a microarray-based differential mRNA profiling using cells transfected with miR-34a or control vector was performed. Then, AREG was selected and finally confirmed to be a crucial target of miR-34a through a series of verified experiments. We provide evidence for decreased expression of the AREG gene at the mRNA and protein levels when forced expressing of miR-34a in four HNSCC cell lines. In particular, miR-34a reduced the luciferase reporter activity of the pSiCheck™-2-AREG-3'-UTR, while had a minimal effect on the reporter activities of the pSiCheck™-2-AREG-mutant 3'-UTR, confirming that miR-34a degrades AREG gene through binding on the seed sequence in AREG 3'-UTR. Recently M.A. Avila has predicted microRNAs might involved in regulation of AREG mRNA, however the clear information on the miRNAs in AREG mRNA regulation has not been ever reported [31]. To our knowledge, this is the first time that miR-34a is demonstrated to inhibit tumor metastasis in HNSCC by directly targeting AREG, which provides the evidence that AREG could be posttranscriptionally regulated by microRNA. On the other hand, based on our microarray data, there are several other candidate genes have potential miR-34a binding site on the 3'-UTR, such as AXL, MAFF, FUT1, CAMTA2 and PCK2. Among these, AXL and FUT1 were previously determined as the targets of miR-34a in non-small cell lung cancer, colorectal cancer and breast cancer [18,43]. Unfortunately, none of them have a universal decrease like AREG in four HNSCC cell lines over-expressing miR-34a (data not shown). Since microRNA could posttranscriptionally regulate gene expression by mRNA degradation or translation inhibition or both, loss of miR-34a may activate more than one pathway involved in HNSCC metastasis. Future efforts on identifying more targets of miR-34a should be developed by combining more advance approaches to clarify the mechanisms of miR-34a in HNSCC progression, like using highthroughput mRNA and Proteomics sequencing combined with computational algorithms or bioinformatics.
AREG, short for amphiregulin, synthesized as a 252-amino acid transmembrance precursor pro-Areg, is one of a member of the epidermal growth factor family [44]. The family pairs with and induces the homo-or heterodimerization of four receptors: ErbB1 (EGFR), ErbB2, ErbB3 and ErbB4 [45][46][47]. The extracellular domain is the least conserved among the 4 ErbB members, allowing AREG specifically bind to the EGFR resulting in phosphorylation of C-terminal tail, and subsequent downstream signaling [30]. Upon binding to EGFR, AREG can also induce the activation of ErbB2, ErbB and ErbB4 [31]. In accordance with previous researches, our data showed that EGFR was significantly inhibited by abundant miR-34a or defective AREG, besides which ErbB3 and ErbB4 were also remarkably activated by restoration of AREG in miR-34a over-expressed cells. It suggested that miR-34a may play an important role in regulation of ErbB pathway though inactivating ErbBs via targeting AREG. In addition, AREG is found to be less effective in inducing ErbB receptors at low concentrations, but has an affinity for the receptors at higher concentrations [30]. These might partially explain why re-expression AREG in miR-34a over-expression cell lines not only dramatically elevated EGFR, but also ErbB3 and ErbB4 mRNA levels.
Recent decade, studies have demonstrated AREG can indeed behave as a pro-oncogenic factor and highlighted the multiple functional roles of AREG overexpression in inhibition of apoptosis and promoting proliferation, motility, angiogenesis, and tumor invasion and metastasis in many solid tumors [31,48,49]. Besides the direct effects that autocrine or paracrine, a recent study has identified exosome-mediated AREG signaling in human breast and colorectal cancer cells, which activates EGFR resulting cell invasion [50]. Most importantly, AREG was also identified to participate in the tumor resistance to PI3K inhibition and sorafenib therapy, and as the sole targeted agent in the treatment of advanced solid carcinomas [51,52]. In our research, the cellular effects of AREG in metastasis were identified in HNSCC that AREG over-expression was strongly associated with HNSCC metastatic phenotype in the tissues and cell lines. Attenuated AREG expression significantly decreased HNSCC invasion in vitro, and restoration of AREG rescued miR-34a-imposed cell invasion and metastasis defects in vitro and in vivo. The results agree with the previous study that knockdown of pro-AREG expression by SiRNAs inhibited the migration of SCC-9 head and neck cancer cells toward fibronectin [53]. Moreover, for the first time, AREG was identified to promote in vivo tumor metastasis and reverse miR-34a-imposed metastasis defects, indicating that AREG is the key determinant in the maintenance of metastatic properties of HNSCC, and a potential therapeutic target for metastatic tumor.
The urokinase plasminogen activator (uPA) is one of the serine proteinase systems involved in ECM degradation, which are overexpressed in several malignant tumors. For decades, the urokinase plasminogen activator (uPA) system has been thought to drive tumor progression by mediating directed extracellular proteolysis on the surface of migrating or invading cells. Intervention with this proteolysis by targeting uPA system has been proposed to represent a novel approach for inhibiting tumor progression and a potential value as molecular prognostic marker prostate cancer, melanoma, breast cancer, pancreatic cancer, gastric cancer, thyroid cancer, and etc [54][55][56][57][58]. Previous researches also showed that the downstream factors of ErbB pathway, such as MMP-9, uPA, EMMPRIN, and PAI-1, could be induced by AREG in head and neck cancer [28,29], breast cancer [59,60], and malignant mesothelioma cell lines [61]. Although both of MMP-9 [28,62] and uPA [29] were demonstrated previously activated by AREG in HNSCC invasion and metastasis, based on our microarray data and qRT-PCR results, only uPA showed significant decreases in this research, when AREG was suppressed by ectopic miR-34a. Our results indicate that degradation of uPA by inhibition of AREG might be an essential pathway of miR-34 modulating HNSCC invasion defects.
In conclusion, our data show that miR-34a has a pivotal role in suppressing HNSCC invasion and metastasis. AREG is one of essential miR-34a targets, over-expression of which could rescue miR-34a mediated tumor invasion defects and EGFR inhibition. Patients, who lose miR-34a expression and over-express AREG, tend to exhibit highly potential of lymphatic metastasis. Therefore, miR-34a/AREG might be promising molecular targets in anti-invasion/metastasis of HNSCC.
For each plasmid to be transfected, plate HEK-293T cells were seeded in a 10 cm tissue culture plate and incubated at 37°C, 5% CO 2 overnight. When HEK-293T cells get 50%-70% confluent in the next day, a cocktail for each transfection was made by putting the constructed plasmid, packaging plasmid, and envelope plasmid to serum-free OPTI-MEM. Drop wise the cocktail media to HEK-293T cells and incubate cells at 37°C, 5% CO2 for 24 hours and change the media in the next day. 2 days later, harvest the virus and store in the -80°C. To stably transfect the target cells, lentiviral particle solution with polybrene was added to the 80-90% confluent target cells. Change to fresh media 24 hours after infection and use GFP fluorescence sorting or puromycin to select transfected cells.

Cell invasion assay
BD Matrigel™ Basement Membrane Matrix (BD Biosciences Inc. San Jose, CA, USA) was thawed on ice at 4°C and diluted with the ratio of 1:8 with DMEM. 30 ul of the diluted BD Matrigel matrix coating solution was added to the 6.5 mm transwell with 8.0 um pore polycarbonate membrane insert (BD Biosciences Inc. San Jose, CA, USA) and incubated at 37°C for 5-6 hours. For the cell invasion assay, 1.5 × 10 5 cells were seeded onto a Matrigelcoated chamber with 8.0 mm pores (BD Biosciences Inc. San Jose, CA, USA). Cancer cells were seeded using serum-free media to the upper chamber pairing with the lower chamber filled with 10% serum DMEM media. At 24 hrs after allowing cells to invade, cells on the lower side of the chamber were fixed and stained with crystal violet (Promoter biotechnology, Wuhan, China). The total number of cells per high-power microscopic field on the lower side of matrigel-coated chamber was counted and scored for migrating or invading cells. The mean number of cells in five high-power microscopic fields was calculated with standard deviations.

In vivo metastatic assay
Fadu cells infected with lentiviruses carrying control vector or miR-34a expression construct, or miR-34a-AREG expression construct were used for vivo pulmonary metastasis assays. 6-8 weeks old female BALB\c-nu nude mice (animal experiment center, Wuhan University, China) were injected with 1 × 10 6 cells (resuspended in 100 μL PBS) via the tail vein. Eight weeks after the injection, the mice were sacrificed and the lung tissues were isolated. Metastasis was quantified by evaluating the weights of lung tissues. The lung tissues were made into serial sections before HE staining. The pathological changes were observed under a light microscope. Mice were manipulated and housed in accordance with the Guide for the Care and Use of Laboratory Animals (Ministry of Science and Technology of China, 2006), and were approved by the animal ethics committee of Wuhan University. All efforts were made to minimize suffering.

MRNA microarray analysis
Total RNA was extracted from Fadu and UM-SCC-23 cells expression of miR-34a and control miRvector, and was further purified with an RNA clean-up kit (Macherey-Nagel, Germany). RNA was quantified by using of spectrophotometer and then was assessed by visualization of the 28S/18S ribosomal RNA ratio on 1.2% formamide agarose gel. Acceptable values were defined as: A260/A280 ratio in the range of 1.8-2.2, rRNA ratio (28S/18S) > 0.9 and RIN value > 8.0. cRNA (2 mg) was reversely transcribed by CbcScript II reverse transcriptase and 9-nt random primer. The products of reverse transcription were labeled with KLENOW polymerase and 9-nt random primer. Labeled cDNA with nucleotides coupled to a fluorescent dye (either Cy3 or Cy5) dCTP (GE Healthcare Cat. No. PA 55021/ PA 53021) was hybridized to 22k human Genome Array Genechips (CapitalBio, Beijing, China) using two-channel microarray technology. LuxScan 3.0 (CapitalBio, Beijing, China) was used to analyze the image of microarray, and the two channel ratio values were normalized by the method of Lowess. Cy3/Cy5 ratios were log-transformed (base 2), median centered by arrays and genes, and hierarchically clustered (average linkage correlation metric). Gene products with at least 2-fold decreases (log2 ≤ -0.3) were determined as the significant differences of expression. Statistical comparisons were done by using the one class method in significance analysis of microarray (SAM) software 3.0 (Stanford University, USA).

Protein extraction and western blot analysis
The total proteins were isolated from the HNSCC cell lines using RIPA buffer (50 mM Tris/HCl, pH 8.0, 250 mM NaCl, 1% Nonidet P-40, 0.5% (w/v) sodium deoxycholate, 0.1% sodium dodecylsulfate, Protease Inhibitor Cocktail Tablets (Roche, Mannheim, Germany)). Fifty micrograms of protein of different groups were boiled for 5 min in sample buffer and were then separated in sodium dodecyl sulphate-polyacrylamide gel electrophoresis and transferred on to a polyvinylidene fluoride membrane (Bio-Rad Laboratories, Hercules, CA, USA). Nonspecific reactivity was blocked using 5% bovine serum albumin in a TBST buffer (100 ml 10 × TBS, 10 ml 10% Tween20 final 0.1% v/v, dissolve with 890 ml double distilled-H2O) and then subsequently and probed with Human AREG MAb (R&D, Minneapolis, USA) overnight at 4°C. Membranes were also probed with anti-β-actin (Santa Cruz Biotechnology, Texas, USA) to assure equal amounts of protein. Bound antibodies were detected using horseradish peroxidase-conjugated secondary antibody (Santa Cruz Biotechnology, Texas, USA). Western blot experiments were repeated three times to confirm the results. The protein amounts were estimated through densitometry as the ratio detected protein/β-actin.

Bioinformatic and statistical analysis
TargetScan5.2 (http://www.targetscan.org/) and Pict-Tar (http://pictar.mdc-berlin. de/) were used to predict the target genes of miR-34a. According to TargetScan, highly conserved predictions were included as the potential targets. All statistical analyses were performed using 13.0 SPSS software system. All data are presented as the mean ± SD; groups were compared using two tailed Student's t-test. p values < 0.01 were considered statistically significant.
The microarray data was deposited in the Gene expression omnibus (GEO) database. The following link has been created to allow reviewing: http://www.ncbi.nlm. nih.gov/geo/query/acc.cgi?token=spkteomuttuvnqd&acc= GSE59617.