Integrated genomic analyses of lung squamous cell carcinoma for identification of a possible competitive endogenous RNA network by means of TCGA datasets

Computational analysis of RNA sequence data for clinical variables of LUSC

We employed all the available RNA expression profile data (level 3) for 504 LUSC tumors and 46 samples of adjacent non-tumorous lung tissues from TCGA in 2016. TCGA data are divided into three levels, and level 3 data are public, containing high-level summaries such as expression quantifications of genes (Zhu, Qiu & Ji, 2014; Alcaraz et al., 2017). All of datasets in the present work were obtained in their processed level 3 form. The exclusion criteria were the following: (i) clinical stage not clear (four cases), (ii) cancer diagnosis other than LUSC (71 cases), and (iii) overall survival longer than 2,000 days (54 cases). In total, 375 LUSC samples were included in this study. Among these 375 LUSC datasets from TCGA, the available tumor data were subdivided into an early-stage group (Stages I, IA, and IB according to histopathological analysis), mid-stage group (Stages II, IIA, and IIB), and late-stage group (Stages III, IIIA, IIIB, and IV), and the 46 samples of adjacent non-tumorous lung tissues were designated as the control group. For each group, the data distribution of the obtained RNA sequencing data for both small RNAs and long RNAs is shown in Table 1. This study was conducted in full compliance with the publication guidelines of TCGA. The data were retrieved from the database of TCGA, and therefore approval of an ethics committee was not required. Clinical characteristics of patients in this study are shown in Table S1 in the Supplement. Their corresponding mRNA, miRNA, and lncRNA expressions including associated adjacent non-tumorous lung tissue samples were listed in Tables S2–S4, respectively.

At the next step, we compared the differentially expressed mRNAs, miRNAs, and lncRNAs: the early-stage group vs. control group, mid-stage group vs. control group, and late-stage group vs. control group. In this study, significant differences in gene expression were defined by two criteria: false discovery rate (FDR ≤ 0.05 and fold change ≥3). A flow chart for bioinformatics analysis was depicted in Fig. S1.

Gene ontology and pathway analysis

Statistically significant differentially expressed mRNAs were examined in the Gene Ontology (GO) database (http://www.geneontology.org), where significantly enriched GO terms were identified to analyze their biological function (Gene Ontology Consortium, 2006). With the help of Kyoto Encyclopedia of Genes and Genomes (KEGG; http://www.kegg.jp/) (Kanehisa et al., 2008; Draghici et al., 2007), the pathway enrichment analysis uncovered the metabolic pathways and signal transduction pathways significantly enriched in LUSC (P ≤ 0.05 and FDR ≤ 0.05) (Benjamini & Hochberg, 1995). Both upregulated and downregulated genes were analyzed.

Seed match analysis and construction of the ceRNA network

The miRNA seed sequences were determined by mapping the TCGA miRNA identifiers to miRBase (www.miRBase.org, release_21). The mRNA target genes of differentially expressed miRNAs in this study were predicted using miRanda (http://www.microrna.org/) and Targetscan (http://www.targetscan.org/). The miRanda (http://www.microrna.org/) was also applied to predict the lncRNAs targeted by miRNAs. The corresponding miRNA–mRNA and miRNA–lncRNA paired libraries were listed in Tables S5 and S6, respectively.

According to the theory that lncRNAs can act as a miRNA sponge by sequestering and binding them to further regulate mRNA activity, the miRNAs negatively regulated by the competing expression levels of lncRNAs and mRNAs were selected to construct a lncRNA–miRNA–mRNA ceRNA network (upregulated or downregulated fold change ≥3, FDR ≤ 0.05, and P < 0.05) (Li et al., 2016). Cytoscape v3.0 was used to construct the interactive and visual ceRNA network.

Clinical features of key members of the ceRNA network

Using the obtained ceRNA network, we then analyzed the clinical features for assessment of patients’ outcomes. The Cox proportional hazards regression model was employed to analyze the association among the mRNAs, miRNAs, and lncRNAs from the ceRNA network and LUSC patient survival periods obtained from TCGA. Statistically significant mRNAs, miRNAs, and lncRNAs affecting the survival period (P < 0.05) were then determined by the Cox regression univariate analysis to subsequently construct the Kaplan–Meier survival curve for patients with LUSC.

Cell culture

Human lung squamous cell carcinoma SK-MES-1 cells were purchased from the Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China). Human bronchial epithelial 16-HBE-T cells were acquired from MssBio Co., Ltd. (Guangzhou, China). SK-MES-1 cells were cultured in the Minimum Essential Medium (Grand Island, New York, NY, USA) supplemented with 10% (v/v) of fetal bovine serum (FBS), Glutamax, nonessential amino acids, and a sodium pyruvate solution (0.1 mol/L). 16-HBE-Tcells were cultured in the RPMI-1640 medium (Grand Island, New York, NY, USA) supplemented with 10% of FBS. All the cell lines were grown in a humidified incubator (5% CO₂) at 37 °C.

RNA extraction and quantitative PCR

Total RNA was extracted from the cells using the TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). Single-strand cDNA was synthesized from 1 μg of total RNA using the Prime-ScriptTM Reagent Kit with gDNA Eraser (Takara, Dalian, China). Real-time quantitative PCR (RT-qPCR) primers were purchased from the Beijing Genomics Institute. The primers were as follows: PLAU sense, 5′-TCACCACCAAAATGCTGTGT-3′, and antisense, 5′-CCAGCTCACAATTCCAGTCA-3′ (Xu et al., 2015). The qPCR was conducted on a 7300 Real-Time PCR system (Applied Biosystems, Life Technologies, Singapore). The following cycling conditions were applied: 95 °C for 5 min, followed by 40 cycles of 95 °C for 20 s and 60 °C for 30 s. For each sample, qPCR assays were conducted in triplicate in a 10 μl reaction volume. β-Actin served as an internal control to normalize the expression of PLAU. The 2^−ΔΔCt method was employed to calculate the relative expression of PLAU mRNA (Livak & Schmittgen, 2001).

Statistical analysis

These analyses were performed using the SPSS 21.0 software. All the data are presented as mean ± SD. The statistical analysis involved Student’s t test to compare two groups, and ANOVA was conducted for multiple comparisons. Differences with P-value <0.05 were considered statistically significant. The significance level was set to 0.05 as a default to control the FDR. The univariate Cox proportional hazards regression was applied to determine the ceRNAs correlating with overall survival.

Differentially expressed genes and their GO and pathway analysis

The differentially expressed genes obtained from the TCGA data were analyzed among LUSC groups and controls. We found 3,564 mRNAs to be differentially expressed between the early-stage group and control group, 3,864 mRNAs were differentially expressed between the mid-stage group and control group, and 3,818 mRNAs were differentially expressed between the late-stage group and control group (Fig. 1A). All significantly expressed genes are listed in Table S7. A Venn diagram (Fig. 1B) was constructed to illustrate the distribution of genes whose expression was common among groups or unique to each LUSC group. In total, 3,366 common differentially expressed genes were found in each library when compared with the control. Finally, the 3,366 differentially expressed genes were further analyzed using the GO database (http://www.geneontology.org). Enrichment of these differentially expressed genes represents a measure of significance of a function; for instance, the most enriched GO was “signal transduction” and other significant GO terms included “cell adhesion,” “blood coagulation,” and “immune response” (Fig. 2). These data provided a definitive functional description of the genes differentially expressed in LUSC.

We next subjected the 3,366 differentially expressed genes to pathway analysis, where the most enriched network corresponding to the upregulated transcripts turned out to be “cell cycle” and the most enriched network corresponding to the downregulated transcripts was “complement and coagulation cascades.” Moreover, pathway analysis showed that the “p53 signaling pathway,” “Fanconi anemia pathway,” “transcriptional misregulation in cancer,” “microRNAs in cancer,” and “small cell lung cancer” correspond to the upregulated transcripts (Fig. 3A) whereas “cell adhesion molecules (CAMs),” “hematopoietic cell lineage,” “PI3K-Akt signaling pathway,” “Ras signaling pathway,” and “Jak-STAT signaling pathway” correspond to the downregulated transcripts (Fig. 3B). In addition, “pathways in cancer” was identified as the significant gene category involved in the development of LUSC corresponding to both the upregulated and downregulated transcripts (Fig. 3).

Cancer-specific miRNAs and lncRNAs in LUSC and their ceRNA network

We then identified the miRNAs and lncRNAs differentially expressed between the LUSC tumor tissues and the adjacent non-tumorous tissues from the TCGA database (absolute fold change >2, P < 0.05). In the analysis of differentially expressed miRNAs, 88 miRNAs were found to be differentially expressed between the early-stage group and control group, 87 miRNAs were differentially expressed between the mid-stage group and control group, and 101 were differentially expressed between the late-stage group and control group. Figure 4A shows the distribution of miRNAs expressed in all three libraries, and 79 miRNAs that were coexpressed in the three libraries (Fig. 4B) were selected to build the ceRNA network.

We also studied the distribution of lncRNAs to illustrate which lncRNAs are differentially expressed between the LUSC tumor tissues and adjacent non-tumorous tissues. We found that 167, 183, and 177 lncRNAs were differentially expressed between the groups “early-stage” and “control,” between the groups “mid-stage” and control, and between the groups “late-stage” and control (Fig. 5), respectively. Finally, 151 lncRNAs (68 upregulated and 83 downregulated) coexpressed in the three libraries were selected to build the ceRNA network (Fig. 6).

Further, we search for mRNAs targeted by miRNAs from the dataset between the Figs. 1 and 4B, and also predicted miRNAs targeted lncRNAs from the dataset between the Figs. 4B and 6. The ceRNA network of LUSC was built in Fig. 7 through the mRNA–miRNA–lncRNA of negative regulation. 86 mRNAs, 39 miRNAs and 88 lncRNAs were integrated into the ceRNA network of LUSC (Fig. 7).

Key ceRNAs and their associated clinical features

By analysis of the association between key ceRNAs and LUSC patients’ survival periods, ceRNAs were chosen according to the bioinformatics analysis and the ceRNA network to analyze the effects significant for survival (P value less than 0.05) in order to identify the specific ceRNAs with prognostic characteristics. Among the differentially expressed ceRNAs, one mRNA (PLAU), two miRNAs (miR-31-5p and miR-455-3p), and three lncRNAs (FAM83A-AS1, MIR31HG, and MIR99AHG) were found to be associated with the overall survival of patients with LUSC by univariate Cox regression analysis. Kaplan–Meier survival curves indicated that the lncRNA MIR99AHG positively correlated with overall survival, whereas PLAU, miR-31-5p, miR-455-3p, FAM83A-AS1, and MIR31HG were negatively associated with overall survival (Fig. 8).

Characteristics of PLAU in LUSC

We examined the expression data on PLAU in different groups of LUSC. The result showed that PLAU was significantly upregulated at the early, middle, and late stage of LUSC relative to the adjacent non-tumorous tissues (Fig. 9A). According to GO analysis, anomalous expression of PLAU is related to aberrant regulation of gene function groups such as pulmonary squamous cell carcinoma, proteolysis, and fibrinolysis (Fig. 9B). Protein–protein interaction analysis revealed that urokinase plasminogen activator (uPA, a.k.a. PLAU) can interact with several proteins including serine protease inhibitor1 (SERPINE1), SERPINE2, plasminogen activator inhibitor2 (SERPINB2), vitronectin (VTN), plasminogen activator receptor (PLAUR), very low density lipoprotein receptor (VLDLR), matrix metallopeptidase 9 (MMP9), epidermal growth factor receptor (EGFR), FOS protein, and mitogen-activated protein kinase 1 (MAPK1; Fig. 9C).

We also evaluated the expression of PLAU in SK-MES-1 cells compared with 16-HBE-T cells. As shown in Fig. 9D, uPA transcript levels were significantly higher in SK-MES-1 cells than in 16-HBE-T cells (up to 9.99-fold).