Integrated analysis of long noncoding RNA interactions reveals the potential role in progression of human papillary thyroid cancer

2.1 Patients and samples

Our study was granted approval by the ethics committee of Nanjing Drum Tower Hospital. Firstly, we collected 503 PTC cases and 59 normal thyroid cases from the The Cancer Genome Atlas (TCGA) database (https://cancergenome.nih.gov/). Then, we set the exclusion criteria as below: (a) pathologic diagnosis is not PTC; (b) suffering from other cancers except PTC; (c) patients samples without integrated clinical information. Overall, 461 PTC tumor cases and 55 normal thyroid cases were identified in the study. On the basis of the 7th American Joint Committee on Cancer (AJCC) TNM staging system, among these 461 patients with PTC, there were 256 patients with stage I PTC, 51 patients with stage II PTC, 104 patients with stage III PTC, and 50 patients with stage IV PTC. Additionally, there are 219 PTC cases with lymphatic metastasis and 242 PTC cases without lymphatic metastasis.

Additionally, 28 paired frozen PTC samples (thyroid tumor samples and adjacent normal thyroid samples) were obtained from Nanjing Drum Tower Hospital for qRT-PCR. Tissues were immersed in RNAlater (GenStar BioSolutions, Beijing, China) immediately after surgical resection and stored at −80°C until use. Pathology reports and quality assessment reports were required to verify the collected tumor tissues and normal thyroid tissues. Demographic and clinical features of these patients were shown in Table 1. Informed consent agreements were signed by all patients in this study.

2.2 RNA sequence data analysis

Firstly, we obtained all RNA expression profiles (level 3) of patients with PTC from the TCGA data portal (up to May 05, 2016). The raw data of lncRNAs and mRNAs were processed and normalized by TCGA RNASeqv2 system. The PTC level 3 miRNA data were normalized and provided by Illumina HiSeq 2000 microRNA sequencing platforms (Illumina Inc., Hayward, CA, USA). No further normalization was needed as all the RNA sequencing data have already been normalized already. Then, all samples collected in TCGA were divided into three groups to perform the differential analysis of lncRNAs, miRNAs, and mRNAs. Thus, all collected samples were grouped as: (a) PTC tumor samples and normal thyroid samples; (b) PTC samples with lymphatic metastasis and PTC samples without lymphatic metastasis; (c) PTC samples with stages I-II and PTC samples with stages III-IV. Four groups were listed as follows: stages I-II with/without lymphatic metastasis and stages III-IV with/without lymphatic metastasis, and these groups contained 112, 195, 107, and 47 cases, respectively. The intersected lncRNAs, miRNAs, and mRNAs were then chosen to construct the ceRNA network which shows the genetic regulation related to PTC. The bioinformatics analysis is illustrated in the flowchart (Figure 1).

2.3 Gene functional enrichment analysis

The intersected mRNAs were selected following the above methods and were analyzed to detect the potential biologic functions and pathways. The Database for Annotation, Visualization, and Integrated Discovery (DAVID) Bioinformatics resources (http://david.abcc.ncifcrf.gov/) were employed to analyze the potential functions and pathways. The biological functions and pathways were considered significant when an enrichment score of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis exceeds 1.5 (P < 0.05).

2.4 The ceRNA network in PTC

The interactions among lncRNAs, miRNAs, and mRNAs contribute to the post-transcriptional regulation of mRNAs. The ceRNA network is built based on miRNA sponge25 in our study. This theory indicates that lncRNAs conjunct with miRNAs to regulate mRNAs expression. Furthermore, the miRNA-lncRNA and miRNA-mRNA relationships were collected in the starBase v2.0 database (http://starbase.sysu.edu.cn/),26 and only differentially expressed intersection lncRNAs, miRNAs, and mRNAs with fold change (FC) >2.0 or FC <0.5 and P < 0.05 were selected. Subsequently, miRanda tools (http://www.microrna.org/microrna/home.do), the miRNAs target prediction tool, were employed to detect the lncRNA-miRNA interactions. Then, the mRNA-miRNA interactions were analyzed by miRTarBase (http://mirtarbase.mbc.nctu.edu.tw/) and TargetScan (http://www.targetscan.org/). The summary results were identified to determine the specific interacted lncRNAs and mRNAs. Then, according to ceRNA theory, the negatively regulated lncRNA-miRNA and mRNA-miRNA pairs were analyzed in the ceRNA network in PTC.27 Finally, the ceRNA network was built using Cytoscape v3.0.28 The ceRNA network construction is illustrated in a flowchart (Figure 2).

2.5 Clinical features analysis and qRT-PCR verification

Specific lncRNAs involved in ceRNA network and bioinformatics analysis were selected to assess the clinical characteristics including age, race, gender, lymphatic metastasis, TNM staging system, and patient outcome in TCGA. In addition, qRT-PCR validation of eight randomly selected lncRNAs in 28 paired frozen tissues was performed to examine the credibility of the bioinformatics analysis results.

Following the manufacturer's protocol, total RNAs were extracted and purified using TRIzol reagent (Pufei, Shanghai, China). Then, concentration and purity of total RNA were detected by the spectrophotometer from BioTek (Shoreline, WA, USA). According to the manufacturer's protocol, a total of 1000 ng RNA was used to perform cDNA synthesis, and reverse transcription reaction was conducted using PrimeScript™ from Takara (Dalian, China). Candidate lncRNA expression was amplified with TAKARA SYBR^® Premix Ex Taq™ (Dalian, China) on an Applied Biosystems ViiA™ 7 System (Foster City, CA, USA) using specific primers in duplicate. All primers were designed and synthesized by Generay Biotech (Shanghai, China), and the sequences of selected lncRNAs primers were presented in Table 2. qPCR primers were determined to anneal at 60°C according to the instructions. The comparative CT method was used to measure the relative expression of candidate lncRNAs.29 β-actin is used as the internal control in our experiment. FC values were shown as 2^−∆∆Ct, where ∆∆Ct = (Ct_RNAs − Ct_β-actin)_tumor − (Ct_RNAs − Ct_β-actin)_{adjacent normal samples}.

2.6 Statistical analysis

The Student's t test was performed to detect the statistical significance for two groups, and ANOVA for multiple groups using the GraphPad Prism 6.0. All data are expressed as mean ± SD (standard deviation). The threshold of false discovery rate (FDR) was set as 0.001 to reduce the false positives in multiple tests in bioinformatics analysis, and the threshold of P value was set as 0.05 to evaluate the null hypothesis. Box plots were used to present the FC and clinical relevance of the specific lncRNAs in donor samples. The upper and lower boundaries of the boxplots are at 25th and 75th percentiles. The vertical lines below and above the box represent the minimum and the maximum. And the band inside the box is always the median.

3.1 PTC-specific lncRNAs

Totally, 1806 lncRNAs were identified from TCGA database (Data S1). We compared the expression of lncRNAs in 461 patients with PTC tumor tissues with adjacent normal tissues and identified 199 differentially expressed lncRNAs (FC > 2, P < 0.01). Further analysis was carried out between tumor tissues and adjacent normal thyroid tissues on tumor stages and lymph node status. For patients without lymphatic metastasis, there are 99 aberrant expression lncRNAs between stages I-II PTC tumors samples and normal thyroid samples, and 122 aberrant expression lncRNAs between stages III-IV PTC tumors samples and normal thyroid samples. For patients with lymphatic metastasis, 160 aberrant expression lncRNAs were identified between stages I-II PTC tumors samples and normal thyroid samples, and 177 aberrant expression lncRNAs were identified between stages III-IV PTC tumors samples and normal thyroid samples (Figure 3A). The intersected 83 specific intersected lncRNAs were selected initially for further analysis on ceRNA network to improve the credibility of bioinformatics analysis results (Table 3).

3.2 Functional enrichment analysis

DAVID bioinformatics resources were employed to predict the biological functions and pathways of aberrantly expressed intersection mRNAs. A total of 18 633 mRNAs was identified from TCGA database, and 3531 mRNAs were aberrantly expressed in PTC tumor samples compared with normal thyroid samples (FC >2, P < 0.01). For patients without lymphatic metastasis, 2209 aberrant expression mRNAs were identified between stages I-II PTC tumors samples and normal thyroid samples, and 2457 aberrant expression mRNAs were identified between stages III-IV PTC tumors samples and normal thyroid samples. For patients with lymphatic metastasis, 2903 aberrant expression mRNAs were identified between stages I-II PTC tumors samples and normal thyroid samples, and 3128 aberrant expression mRNAs were identified between stages III-IV PTC tumors samples and normal thyroid samples. Finally, we identified 1822 mRNAs from the intersection of the four groups for further functional analysis.

The upregulation and downregulation of these intersected mRNAs were further analyzed, respectively. GO analysis revealed 531 biological processes corresponding to these upregulated genes and 692 biological processes corresponding to these downregulated genes. Cell adhesion (GO:0007155) was the most enriched function for both upregulation and downregulation transcripts (Figure 4). KEGG pathway indicated 100 pathways matching with upregulated genes, and the top pathway was neuroactive ligand-receptor interaction. Furthermore, pathway analysis showed 101 pathways matching with downregulated genes and the top pathway was pathways in cancer (Figure 5). Among these pathways, “MAPK signaling pathway” has already been demonstrated as the main cause of genetic alterations and PTC development. Targeting “MAPK signaling pathway” has shown an effective antitumor effect in preclinical studies and ongoing clinical trials.30

3.3 Prediction of miRNAs targets and construction of ceRNA network

A total of 1030 miRNAs were obtained from TCGA database to identify potential differentially expressed miRNAs. And it was found that 87 miRNAs were aberrantly expressed in PTC tumor samples compared with normal thyroid samples (FC > 2, P < 0.05). According to the comparisons of four groups described above, 36 miRNAs were selected from the intersections of 87 PTC-associated miRNAs to build the ceRNA network. Consequently, miRNAs target prediction including lncRNAs and mRNAs were conducted following the instructions described above. Firstly, 83 intersected lncRNAs and 36 specific miRNAs were collected to analyze the potential relationship in PTC. The potential miRNA-lncRNA interactions were predicted using StarBase v2.0 which could explore the candidate MREs. Finally, according to the analysis results, there were 29 specific miRNAs interacting with 54 specific lncRNAs (Table 4).

In the further analysis, the miRNA-targeted mRNAs were predicted using TargetScan and miRTarBase. It was found that 30 miRNAs interacted with 130 mRNAs (Table 5). Then, we built a ceRNA network based on the above results (Tables 4 and 5). Cytoscape 3.0 was performed to depict the ceRNA network (Figure 6). The ceRNA network contains 45 lncRNAs, 21 miRNAs, and 78 mRNAs finally.

3.4 The association between clinical characteristics and PTC-specific lncRNAs

In the following analysis, the 45 specific lncRNAs identified in ceRNA network were evaluated to detect their clinical relevance, such as age, gender, race, lymphatic metastasis, TNM staging system, and patient outcome in TCGA database. It was found that a total of 37 specific lncRNAs expressed differentially correlated with PTC clinical features (P < 0.05, Table 6). Seven lncRNAs (ATP6V0E2-AS1, FOXD2-AS1, LINC01257, EGFEM1P, MIR9-3HG, MIR31HG, and FER1L4) were differentially expressed in age, 4 lncRNAs (YWHAEP7, LINC00887, MIR4697HG, and MIR9-3HG) were differentially expressed in race, 1 lncRNA (LINC00887) was differentially expressed in gender, 30 lncRNAs (ATP6V0E2-AS1, LINC00982, NR2F1-AS1, LOC100126784, DPY19L2P4, FLJ23867, TINCR, ANKRD20A8P, EGOT, FAM95B1, LOC143666, MIR4697HG, LRRC37A6P, LOC100130238, MIR31HG, TPTE2P1, FAR2P1, LINC01257, FAM181A-AS1, FLJ16779, EGFEM1P, LOC93429, GGT8P, FER1L4, GGT3P, ABCC6P1, LINC00284, KRTAP5-AS1, FOXD2-AS1, and HAND2-AS1) were differently expressed in lymphatic metastasis, 31 lncRNAs (LINC00982, FLJ23867, ABCC6P1, FAM181A-AS1, FOXD2-AS1, NR2F1-AS1, EGOT, TINCR, LRRC37A6P, LOC100126784, TPTE2P1, LOC100130238, FLJ16779, DPY19L2P4, GGT3P, MIR4697HG, MIR9-3HG, ANKRD20A8P, SMIM10L2A, GGT8P, FAM95B1, KRTAP5-AS1, LOC441204, GVINP1, LINC00887, MIR31HG, HAND2-AS1, EGFEM1P, LOC93429, ATP6V0E2-AS1, and LINC01257) were aberrantly expressed in different tumor sizes, and six lncRNAs (PLEKHA8P1, LOC100130238, GGT3P, MIR9-3HG, LINC00284, and EGFEM1P) were differentially expressed in prognosis.

The univariate Cox proportional hazards regression model was performed to detect the prognostic value of 45 specific lncRNAs. We found that three specific lncRNAs closely related with PTC patient overall survival (OS, log-rank P < 0.05). There are two lncRNAs (GGT3P and KRTAP5-AS1) positively relating to OS, and only one lncRNA (DPY19L2P4) negatively relating to OS (Figure 7).

3.5 qRT-PCR validation

Evaluation of randomly selected eight lncRNAs (LOC100130238, HAND2-AS1, MIR9-3HG, LOC143666, EGFEM1P, LINC00284, TINCR, and ABCC6P1) was performed to prove the reliability of the above bioinformatics analysis (Table 7). Firstly, the expression status of the eight specific lncRNAs was detected in 28 newly diagnosed patients with PTC using qRT-PCR (Data S2). Compared with paired normal thyroid tissues, four lncRNAs (LOC100130238, HAND2-AS1, MIR9-3HG, and LOC143666) were downregulated in PTC tumor tissues, whereas the remaining four lncRNAs (EGFEM1P, LINC00284, TINCR, and ABCC6P1) were upregulated in PTC tumor tissues (Figure 8). All the eight lncRNAs were aberrantly expressed with the same trend and reached statistical significance (P < 0.05). Thus, the results were coherent with the above bioinformatics analysis. Subsequently, we performed the analysis of these eight lncRNAs and clinic features, and we found that HAND2-AS1 was significantly related to tumor size (Figure 9, P < 0.05). The result was identical with the analysis in TCGA database. Thus, our bioinformatics analysis is convincing based on the evaluation results.