A total of 130 lung squamous cell carcinoma (SCC) samples were analyzed in this study. The patient characteristics were described in a previous publication (23). Genome-wide mRNA expression profiles of the tumor samples were quantified with the Affymetrix U133A Gene Chip (23). Microarray data were extracted and calculated using the Affymetrix MAS 5 software after global scaling of the average intensity to 600. The mRNA microarray data were available from Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) with accession no. GSE4573. Out of 130 SCC tumors, 57 samples were screened for miRNA expression profiles with TaqMan miRNA assays (Applied Biosystems Inc., Foster City, CA) (24). The miRNA expression data was available from GEO with accession no. GSE16025. The mRNA and miRNA profiles of these matched 57 tumor samples were used for further analysis.

The mRNA expression levels of the lung cancer prognostic markers identified in our previous studies (19–21) were retrieved from the genome-wide expression profiles of the 57 SCC tumor samples. Pearson’s correlation coefficient between each mRNA and all available miRNAs was computed. Significant mRNA-miRNA pairs (|r|≥0.258; p≤0.05) were selected for target prediction. Four Bioinformatics tools were used in the study for miRNA target prediction, including TargetScan (http://www.targetscan.org/) (10), PicTar (http://www.pictar.mdc-berlin.de/) (25), miRDB (http://www.mirdb.org/miRDB/) and microRNA.org (http://www.microrna.org/microrna/home.do) (26). These computational methods use sequence complementary base-pairing between miRNA and mRNA in target prediction. Files and databases containing miRNA and predicted target genes were downloaded from the online websites of these toolsets. These files were then analyzed with in-house software script written in R to identify miRNA-mRNA gene pairs showing significant expression correlation in SCC tumor samples. In addition, TarBase (http://www.diana.cslab.ece.ntua.gr/tarbase/) (27) was used to retrieve experimentally validated miRNA targets.

The miRNA targets predicted by TargetScan are based on the presence of conserved 8mer, 7mer and 6mer sites that match the seed region of each miRNA (10). TargetScan also predicts non-conserved sites, additional types of seed matches that are preferentially conserved in different species, and sites with mismatches in the seed region that are compensated by conserved 3′ pairing (28). PicTar predicts miRNA targets by searching 3′ UTR alignments with predicted sites (25). The miRDB uses an algorithm, MirTarget2, based on Support Vector Machine, to predict miRNA targets. The microRNA.org uses miRanda algorithm to predict miRNA targets based on sequence and contextual features of the predicted miRNA-mRNA duplex (26).

The binding sites between each miRNA-mRNA pair was retrieved from the Homo sapiens microRNA.org database with the ‘Target mRNA’ module. The searching parameters were set with mirSVR score ≤0 and PhastCons score ≥0.

Small airway epithelial cells (SAEC) and normal human bronchial/tracheal epithelial cells (NHBE) were obtained from Lonza Walkersville Inc. (Walkersville, MD). Human non-small cell lung cancer cell line H1299 and human immortalized lung epithelial cell line BEAS-2B were purchased from the American Type Culture Collection (ATCC, Manassas, VA). SAEC cells were cultured according to the supplier’s recommendations in SABM medium supplemented with 52 μg/ml bovine pituitary extract, 0.5 ng/ml human recombinant epidermal growth factor (EGF), 0.5 μg/ml epinephrine, 1 μg/ml hydrocortisone, 10 μg/ml transferrin, 5 μg/ml insulin, 6.5 ng/ml triiodothyronine, 50 μg/ml gentamicin/amphotericin B (GA-1000) and 50 μg/ml fatty acid-free bovine serum albumin. NHBE cells were cultured according to the supplier’s recommendations in BEBM media supplemented with bovine pituitary extract (52 μg/ml), hydrocortisone (0.5 μg/ml), human epidermal growth factor (hEGF, 0.5 ng/ml), epinephrine (0.5 μg/ml), insulin (5 μg/ml), triiodothyronine (6.5 ng/ml), transferrin (10 μg/ml), gentamicin (50 μg/ml), amphotericin B (50 ng/ml), bovine serum albumin (1.5 μg/ml). H1299 and BEAS-2B cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with penicillin, streptomycin, L-glutamine and 10% fetal bovine serum. All cells were cultured at 37°C in humidified incubator with 95% air and 5% CO₂.

Human miRIDIAN shMIMIC lentiviral miRNA particles (hsa-miR-200a: UAACACUGUCUGGUA ACGAUGU, hsa-miR-200b: UAAUACUGCCUGGUAAUG AUGA, hsa-miR-200c: UAAUACUGCCGGGUAAUGA UGGA, and control scrambled microRNA were purchased from Open Biosystems (Huntsville, AL) and used for infection of target cells in the presence of 4 μg/ml of polybrene.

Cells were lysed in 1X SDS lysis buffer (50 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol). Total protein was quantified by the BCA method. β-mercaptoethanol was added to lysates to a final concentration 100 mM. Equal amounts of total protein were separated by 4–12% SDS-PAGE and transferred to a PVDF membrane. Membranes were blocked 1 h with 5% non-fat milk in 1X PBS containing 0.05% Tween-20. Membranes were then incubated for 1 h at room temperature with primary antibodies. After incubation with the primary antibody, membranes were washed thrice in 1X PBS with 0.05% Tween-20 for 5 min each and blocked for 7 min in blocking solution. Membranes were incubated for 1 h at room temperature with horseradish peroxidase (HRP) conjugated donkey anti-mouse IgG or donkey anti-rabbit IgG in 1X PBS with 0.05% Tween-20. Membranes were then washed five times for 5 min in PBS-Tween-20 and finally developed with HyGLO Western Blotting Substrate (Denville Scientific) according to the instructions of the manufacturer. Protein band intensity was determined using FluorChem^® Q software (AlphaInnotech, Santa Clara, CA). Relative protein level was determined after normalization to tubulin and relative to negative control (miR-scr) samples. The following antibodies were used: ATRX (Santa Cruz Biotechnology, catalog no. SC-55584), DLC1 (BD Biosciences, catalog no. 612020), HFE (Santa Cruz Biotechnology, catalog no. SC-130375), ZEB1 (Sigma, catalog no. HPA027524), HNRNPA3 (Santa Cruz Biotechnology, catalog no. SC-133665), E-cadherin (BD Biosciences, catalog no. 610181), GAPDH (Millipore, catalog no. MAB374) and tubulin (Sigma, catalog no. T9026).

Total RNA was extracted using the mirVana^® kit (Ambion Inc., Austin, TX) according to the manufacturer’s protocol. To ensure a good RNA quality, the quality and integrity of the total RNA was evaluated using 28S/18S ratio and a visual image of the 28S and 18S bands were evaluated on the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). RNA isolated using this method yielded a very good quality, with a RIN number ≥9. Concentration of the total RNA was assessed using the NanoDrop-1000 Spectrophotometer (NanoDrop Technologies, Germany).

Complementary DNA (cDNA) was generated using total RNA according to the TaqMan^® MicroRNA Reverse Transcription protocol (Applied Biosystems Inc.). Quantitative RT-PCR for microRNA was performed using TaqMan MicroRNA assays (Applied Biosystems Inc.). Human U47 small nuclear RNA was used as an endogenous control. The expression levels of miRNAs were quantified using ABI 7500 quantitative real-time instrument and SDS software (Applied Biosystems Inc.). The abundance of miRNA is expressed as Ct (threshold fluorescence) which gives the number of cycles required to reach threshold fluorescence. Real-time PCR for target genes was determined using total RNA and cDNA was generated using a High-Capacity cDNA Reverse Transcription kit and TaqMan gene expression assays (Applied Biosystems Inc.). E-cadherin (CDH1) mRNA was measured using SYBR-Green Master mix and CDH1 specific primers according to manufacturer’s protocol (Applied Biosystems Inc.). All qRT-PCR reactions were performed on 7500 instrument (Applied Biosystems Inc.). In the qRT-PCR analysis of E-cadherin, the dissociation curve showed the absence of a secondary peak, indicating no presence of primer dimer. Specificity of the PCR product obtained from SYBR-Green reactions was verified by sequencing. The expression level of each gene was determined by following formulas: fold change = 2^−ΔΔCt, where ΔCt (cycle threshold) = Ct_{target gene} - Ct_{endogenous control gene}, and ΔΔCt = ΔCt_{treated sample} - ΔCt_{control sample}. The expression level of the analyzed genes is reported as fold change relative to negative miR-scrambled (-src) infected samples. The human UBC gene was used as an endogenous control gene.

In this study, a predicted gene was considered a confirmed target if the mRNA level was significantly downregulated or the protein level was downregulated at least 15% relative to negative control samples. Not all of the predicted targets were analyzed at the protein level due to the lack of specificity of commercially available antibodies.

Ingenuity pathway analysis (IPA) software (Ingenuity Systems, Redwood City, CA) was used to derive curated molecular interactions reported in the scientific literature. These interactions included both physical and functional interactions, as well as interactions representing pathway relevance. In this study, in order to delineate molecular networks of genes interacting with the miR-200 family and novel molecular targets, a core analysis was employed to identify the most relevant canonical pathways, biological functions and physiological processes from the interactions reported in the IPA database. We then selected pathways that were statistically significant with a p<0.05 in adjusted Benjamini-Hochberg tests.

The statistical significance of the difference between groups was determined by un-paired t-tests at p≤0.05. The qRT-PCR expression data are presented as mean ± SEM.

To screen for potential miRNA regulators of our previously identified lung cancer prognostic gene signatures (19–21), the correlation between the mRNA expression of these biomarkers and all available human miRNA expression profiles in 57 squamous cell lung cancer tumors was computed. For all miRNA-mRNA gene pairs showing significant correlation (|r|>0.258, p≤0.05, Pearson’s correlation analysis) in the lung SCC tumors, 4 bioinformatics toolsets (TargetScan, PicTar, miRDB and microRNA.org) were used to determine whether or not a given gene is a predicted target of the corresponding miRNA (Fig. 1A). A total of 233 miRNA-mRNA gene pairs were predicted as a target pair by at least one bioinformatic method. The correlation analysis and the prediction results may be viewed on our website (http://www.wvucancer.org/guoLab/Publications). Due to alternative splicing, each gene may have multiple probes in DNA microarray data. This will lead to discrepant gene expression among different probes for the same gene and discordant correlation between mRNA and miRNA. The mRNA-miRNA pairs with a negative correlation for at least one probe set were selected for further analysis.

Several miRNAs, including the miR-200 family, were predicted to target multiple prognostic biomarkers. We focused on the miR-200 family because of its reported role in tumor metastasis. The miRNA-200 family is represented by miR-200a, miR-200b, miR-200c, miR-141 and miR-429, based on their genomic location and primary sequence. Based on sequence similarity, the miR-200 family is divided into two subclasses: one class includes miR-200b, -200c and -429, and the other class includes miR-200a and miR-141 (29). The members in each subclass share the same seed sequence. We analyzed the expression levels of the miR-200s in squamous cell lung cancer patient primary tumors and normal lung tissues in the cohort from Raponi et al(24). The results show that miR-200a had a 2.56-fold overexpression (p<1.35e-7; unpaired t-tests) in the tumors, miR-200b exhibited a 2.94-fold overexpression (p<1.32e-6; unpaired t-tests) and miR-200c showed a 3.16-fold overexpression (p<0.001; unpaired t-tests) in the patient tumors (Fig. 1B). However, during metastasis, previous studies showed that the miR-200 family expression is lost in mesenchymal subtypes of epithelial cancers and negatively correlates with cancer cell invasion (15,16).

The predicted targets for hsa-miR-200a include deleted in liver cancer 1 gene (DLC1), E2F transcription factor 4 (E2F4), and AHNAK nucleoprotein (desmoyokin) (AHNAK); for hsa-miR-200b: DLC1, ubiquitin-like modifier activating enzyme 6 (UBA6), ubiquitin-conjugating enzyme E2I (UBE2I) and heterogeneous nuclear ribonucleoprotein A3 (HNRNPA3); for hsa-miR-200c: alpha thalassemia/mental retardation syndrome X-linked gene (ATRX), hereditary hemochromatosis (HFE), DLC1 and thrombospondin 1 (THBS) (Tables I and II). To confirm the regulation of miR-200a, -200b, -200c on the predicted targets, expression of these prognostic biomarkers genes at the mRNA and protein levels were examined following re-expression of miR-200a, -200b, -200c in H1299 and BEAS-2B cells.

We analyzed the expression levels of miR-200a, -200b -and 200c in a metastatic human NSCLC model, H1299 cells. Normal human small airway epithelial cells (SAEC) were used as control cells. The expression level of miR-200a, -200b and -200c in H1299 cells was at the detection limit (Fig. 2). In contrast, SAEC expressed higher levels of miR-200a, -200b and -200c (data not shown; http://www.wvucancer.org/guoLab/Publications) consistent with their normal epithelial phenotype. In order to identify potential molecular targets of miR-200a, -200b, -200c, H1299 cells were stably infected with lentiviral vectors expressing miR-200a, -200b, -200c or negative control miRNAs. The infected H1299 cells expressed high levels of exogenous miR-200a, -200b, -200c, which were comparable with the levels exhibited in normal lung epithelial SAEC cells (data not shown; http://www.wvucancer.org/guoLab/Publications).

ZEB1 and ZEB2 genes are the most extensively characterized targets of the miR-200 family (14,15) and they were used as positive controls in the qRT-PCR experiments. In H1299 cells over-expressing miR-200a, -200b and -200c, ZEB1 and ZEB2 exhibited significant (p<0.05; unpaired t-tests) downregulation at the mRNA expression level (Fig. 2A). Following overexpression of miR-200a, DLC1 showed significantly decreased mRNA expression. Restoring miR-200b resulted in significant downregulation of DLC1 and HNRNPA3 at the mRNA level. The overexpression of miR-200c resulted in a significant mRNA downregulation of ATRX and HFE (Fig. 2A).

At the protein level, ZEB1 was reduced in H1299 cells overexpressing miR-200a, -200b and -200c, with the most downregulation (about 30%) in cells re-expressing miR-200b (Fig. 3B and C). A considerable downregulation of DLC1, ATRX and HFE was observed in H1299 cells overexpressing miR-200a. It is noteworthy that DLC1 and HFE had about 40% downregulation in H1299 cells infected with miR-200a mimic, exhibiting a more significant downregulation than ZEB1. In H1299 cells with restored expression of miR-200b, the protein levels for DLC1, HNRNPA3 and HFE were reduced by 15–30%. The protein expression of DLC1 was decreased about 20% in H1299 cells that overexpressed miR-200c (Fig. 3B and C).

The results of the present study show that miRNA-200a regulates ATRX, DLC1 and HFE; miRNA-200b regulates DLC1, HFE and HNRNPA3; miRNA-200c regulates ATRX, DLC1 and HFE (Table II). The target prediction results from each of the four bioinformatics algorithms are provided on our website (http://www.wvucancer.org/guoLab/Publications). There is a concordant downregulation of these potential targets at both mRNA and protein expression levels in the corresponding H1299 cells overexpressing the miR-200 family. All of the potential targets contain binding sites for the corresponding miR-200 family members according to the microRNA.org database (26), except for ATRX/miR-200a and HFE/miR-200a. The 3′ UTR region of DLC1 contains 2 binding sites for miR-200a, 3 for miR-200b and 3 for miR-200c. ATRX contains 1 binding site for miR-200c. HFE contains 1 binding site for miR-200b and 1 for miR-200c. HNRNPA3 contains 3 binding sites for miR-200b (Fig. 3A). HFE is a predicted target of miR-200c, not miR-200a or -200b, in NSCLC. Although miR-200b and -200c have the same seed sequence, the correlation between HFE and miR-200b was not statistically significant (p=0.057) in SCC patient samples. Therefore, it was not initially selected as a predicted target of miR-200b in NSCLC. Nevertheless, HFE was shown to be regulated by miR-200a, -200b and 200c in metastatic H1299 cells. AHNAK and E2F4 are predicted targets of miR-200a. However, miR-200a did not suppress the expression of these two genes in H1299 cells (Fig. 2A). Similarly, UBA6 and UBE2I were not regulated by miR-200b in metastatic lung cancer cells as they were predicted to be. These results were summarized in Table II.

To further substantiate the regulatory effects of miR-200 on these lung cancer prognostic markers, SCC patient samples (n=57) (23,24) were screened to investigate the correlation between the miR-200 family and the mRNA expression of its predicted target genes as well as ZEB1 and ZEB2. The results showed that all the downregulated genes had a significant negative correlation with the corresponding miR-200 family member in SCC patient tumor tissues, except ATRX/miR-200a and HFE/miR-200a (Table III). The results in the patient samples further strengthened the in vitro findings.

The regulation of miR-200 on these predicted target genes was also evaluated in human immortalized lung epithelial cells BEAS-2B. The overexpression of miR-200b in these cells resulted in significantly downregulated mRNA level of HFE (Fig. 4A). The overexpression of miR-200a, -200b and -200c in BEAS-2B cells caused approximately 60, 40 and 70% downregulation of HNRNPR3 protein, respectively (Fig. 4B and C). The overexpression of miR-200b in BEAS-2B resulted in a 20% downregulation of ATRX at the protein level and the re-expression of miR-200c resulted in a 70% downregulation of HFE at the protein level. These results indicate that miR-200 family downregulates HNRNPR3, HFE and ATRX in normal lung epithelial cells. Together, these results substantiate the role of miR-200 family and its regulated genes in lung cancer initiation and progression.

After the potential molecular targets of the miR-200 family were shown in the present study, we sought to explore the effect of miR-200s on EMT in the metastatic NSCLC cells. Re-expression of miR-200c induced a 1.53-fold upregulation of E-cadherin (CDH1) through the downregulation of the E-cadherin repressor transcriptional factors ZEB1 and ZEB2, which is consistent with previous studies in breast cancer or NSCLC cells (16,30,31) (Fig. 2B). E-cadherin protein was highly expressed in normal lung SAEC cells, but not in metastatic NSCLC H1299 cells (data not shown; http://www.wvucancer.org/guoLab/Publications). These results indicate that re-expression of miR-200 may reverse EMT process.

Molecular network interactions and significant canonical signaling pathways associated with miR-200s and their predicted molecular targets were retrieved using IPA. The molecular network map shows interactions between the miR-200s and their known target genes, ZEB1 and ZEB2, as well as potential targets identified from the present study (Fig. 5; the regulation identified in the present study is shaded in orange font). Among the genes in the molecular network, Histone H3 and E-cadherin (CDH1) were major focal points in the miR-200 network. Histone H3 is a component of the nucleosome and CDH1 is a cell adhesion protein and epithelial phenotype marker. These results suggest that miR-200s and their potential target genes participate in molecular interactions involved in gene transcription regulation, either during the regulation of gene expression at the chromatin level or in the regulation of cell-cell interactions as mediated by E-cadherin. Increasing evidence indicates that chromatin remodeling induced by DNA damage or epigenetic changes are responsible for carcinogenesis.

The IPA functional analysis found a total of 69 canonical pathways associated with the miR-200 network, of which 13 canonical pathways were statistically significant (adjusted p<0.05 with Benjamini-Hochberg tests; Table IV). The top signaling pathways include virus entry via endocytic pathways, allograft rejection signaling, OX40 signaling pathway, caveolar-mediated endocytosis signaling, communication between innate and adaptive immune cells, chronic myeloid leukemia signaling, molecular mechanisms of cancer and DNA double-strand break repair by homologous recombination, among others (Table IV). Furthermore, the IPA functional analysis found 25 significant diseases and disorders related to the miR-200 network. The top 3 diseases included genetic disorders, metabolic diseases and cancer (Table V). At the molecular level, beta-2-microglobulin (B2M), CDH1, ZEB1, ZEB2, ATRX, HFE and the miR-200s are involved in genetic disorders; B2M, transferrin receptor 2 (TFR2), HFE and angiotensin II receptor (AGTR1) are involved in metabolic diseases; B2M, E2F, miR-200s, ATRX, DLC1, ZEB1, ZEB2, CDH1 and caveolin 1 (CAV1) are involved in cancer (Table V). These results indicate that miR-200 network involves complex signaling pathways and mechanisms, and has implications in numerous human diseases and disorders.