We downloaded gene expression profiling data from the Gene Expression Omnibus database (GEO, http://www.ncbi.nlm.nih.gov/gds/). The GSE113850 dataset, which consists of 39 tissue samples including 32 HCC and 7 normal liver tissues, was used (10). We also downloaded gene expression profiling data from The Cancer Genome Atlas database (TCGA, http://portal.gdc.cancer.gov/). This dataset consisted of 424 tissues, including 371 primary HCC, 3 recurrent HCC, and 50 normal liver tissues.

We used the GEO2R statistical analysis tool (https://www.ncbi.nlm.nih.gov/geo/geo2r/) and Student's t-test to identify DEGs in the GSE113850 dataset. The adjusted P-value was set at <0.05, and the logFC value was set at >2 (upregulated genes) or <-2 (downregulated genes). Data from the TCGA database were analysed using the R packages, DESeq and edgeR (https://www.bioconductor.org/), and Student's t-test to identify DEGs. Meanwhile, heat maps and volcano plots were created by a hierarchical clustering method.

Gene Expression Profiling Interactive Analysis (GEPIA, http://gepia.cancer-pku.cn/index.html) is an online tool used for analysing RNA sequencing (RNA-Seq) data from TCGA and GTEx databases. It includes RNA sequencing expression data of 9,736 tumors and 8,587 normal samples and related clinical datasets of TCGA and the GTEx projects. It provides key interactive and customizable functions including differential expression analysis, profiling plotting, correlation analysis, patient survival analysis, similar gene detection and dimensionality reduction analysis (11). Using GEPIA, we compared the gene expression levels of HCC and normal liver tissues. We used log-rank test to analyse the correlation between the levels of gene expression and the duration of patient survival. Moreover, we analysed the gene expression levels according to the tumour stage of the HCC patients. The adjusted P-value was set at <0.05, and the datasets were set at LIHC (liver hepatocellular carcinoma).

Forty patients (age range, 20–76 years; mean age, 54.78±10.64 years; males, 33; females, 7), who attended the Zhongnan Hospital of Wuhan University from April 2018 to April 2019, were enrolled in this study. All patients were diagnosed with HCC by histological examination, and they provided written informed consent for their tissues to be used for medical research. Ethical approval for this study was obtained from the Zhongnan Hospital of Wuhan University. Liver cancer cell lines (SK-Hep1, HCCLM9 and HepG2) were obtained from the Cell Bank of the Type Culture Collection (Chinese Academy of Sciences, Shanghai, China). Cells were cultured in Dulbecco's modified Eagles medium (DMEM; Invitrogen; Thermo Fisher Scientific, Inc.), supplemented with 10% foetal bovine serum (FBS) and 1% penicillin-streptomycin, at 37°C in a 5% CO₂ humidified atmosphere.

The sequence of lncRNA LINC01554 was sub-cloned into the lentiviral expression vector Lv105 (GeneCopoeia, Guangzhou, China). 293T cells were plated into 6-well plates. After 293T cell confluence reached 80%, cells were co-transfected with LINC01554-Lv105 (or control vector EX-EGFP-Lv105) and Lenti-Pac CMV Expression Packaging vectors (gag-pol-RRE vector, REV vector and VSV-G vector) (GeneCopoeia). The mass of each vector in each well is 1.0 µg lentiviral plasmid, 1.2 µg gag-pol-RRE vector, 1.0 µg REV vector, and 1.0 µg VSV-G vector. The transfection was performed by using LipoFectMax™ Transfection Reagent (ABP Biosciences, USA). After incubation for 48 h, the lentiviral particles were harvested from the cell medium supernatant by centrifuging at 500 × g for 10 min. The SK-Hep1 and HCCLM9 cells were then transduced with Lv-LINC01554 lentivirus controlled with the Lv-NC lentivirus.

Total RNA was isolated from the clinical specimens and cells using TRIzol reagent (Yeasen Biotech). cDNA was generated using a Hieff First Strand cDNA Synthesis Super Mix for RT-qPCR kit (Yeasen Biotech). RT-qPCR analysis was performed using the Hieff qPCR SYBR Green Master Mix kit (Yeasen Biotech) and a StepOnePlus instrument (Applied Biosystems; Thermo Fisher Scientific, Inc.). Relative lncRNA expression was analysed using the 2^−ΔΔCq method (12), with normalisation to β-actin expression. Primer sequences used for RT-qPCR were as follows: LINC01554 forward, 5′-TGTGGCAAACGCAAACGA-3′ and reverse, 5′-GCCCAGAGTAAAGGGAAATGTA-3′; β-actin forward, 5′-TGGCACCCAGCACAATGAA-3′ and reverse, 5′-CTAAGTCATAGTCCGCCTAGAAGCA-3′.

Total protein was isolated from the HCCLM9 and SK-Hep1 cells using RIPA lysis buffer (Wuhan Servicebio Biotechnology). Protein concentrations were determined using the bicinchoninic acid assay (Wuhan Servicebio Biotechnology). The mass of protein loaded per lane was 25 µg. Proteins were separated using 10% SDS-PAGE gels and electro-transferred to PVDF membranes. The membranes were then blocked by TBST (with 5% skim milk powder) for 2 h at room temperature and incubated overnight, respectively, with primary antibodies at 4°C. After washing with TBST three times, the membranes were incubated, respectively, with goat anti-rabbit secondary antibody at 37°C for 2 h. The dilutions and suppliers of the primary antibodies and secondary antibodies are documented in Table I. Protein blots were detected using a chemiluminescence kit (NCM Biotech) and Tanon 4500 Immunodetection System (Tanon). The protein bands were visualised with chemiluminescence ECL reagent and recorded with X-ray film. The signal intensity was analysed by ImageJ v1.42q software (National Institutes of Health). GAPDH content was used to standardise sample loading.

After transfection, the cells were collected and washed twice with phosphate-buffered saline (PBS). For cell cycle analysis, the cells were stained with a cell cycle staining kit (KeyGEN) for 15 min. Apoptosis was detected using an Annexin V-FITC and Propidium Iodide Apoptosis Detection Kit (KeyGEN). Finally, the proportion of cells in different phases of the cell cycle and the apoptosis rate were measured using a FACSCalibur flow cytometer (BD Biosciences).

Briefly, the SK-Hep1 and HCCLM9 cells were first fixed with 4% formaldehyde. They were then placed on slides, and hybridisation reactions were performed using LINC01554-, U6-, and 18S rRNA-specific probes. Cells were then counterstained with DAPI. Fluorescence images were captured using a confocal microscope. The FAM-labelled LINC01554, TAMRA-labelled 18S rRNA, and TAMRA-labelled U6 probe sequences were 5′-ACTTCCCGACTGCTCATCAACCGACCTCCCTGGGGCCGAGAGGCA-3′, 5′-CGCTGAGCCAGTCAGTGTAGCGCGCGTGCAGCCCCGGACATCTAAGGGCATCACAGACCTGTTATTGCTCAATCTCGGGTGGCTGAACGCCACTTGTCCCTCTAAGAAGTTGGGGGACGCCGACCGCTCGGGGGTCGCGTAACTAGTTAGCATGCCAGAGTCTCGTTCGTTATCGGAATTAACCAGACAAATCGCTCCACCAACTAAGAACGGCCAT-3′ and 5′-AAAAATATGGAACGCTTCACGAATTTGCGTGTCATCCTTGCGCAGGGGCCATGCTAATCTTCTCTGTATCGTTCCAATTTTAGTATATGTGCTGCCGAAGCGAGCAC-3′, respectively. The probes were purchased from AxlBioCo., Ltd.

Cells in the logarithmic growth phase were digested with trypsin and seeded (2×10³) in a 96-well plate. CCK-8 reagent (10 µl) (Dojindo Molecular Technologies) was added to each well at 0, 24, 48, and 72 h. After incubating the cells with the CCK-8 reagent for 4 h, we measured the optical density (OD) at 450 nm.

Cells were seeded in 6-well plates at 50% density of the cells. When they were approximately 90% confluent, the tip of a sterile 1,000-µl pipette was used to scratch a line wound. Cells that migrated to the wound region were observed under a microscope at 0 and 24 h.

A 24-well Transwell chamber with Matrigel was used to evaluate the cell invasion rates. Cells (5×10⁴) from each group were added to the upper Transwell chamber, and DMEM (600 µl), containing 10% FBS, was added to the lower Transwell chamber. Cells were then incubated at 37°C in 5% CO₂ for 48 h. Finally, after fixation with 4% paraformaldehyde for 20 min and crystal violet staining for 15 min at room temperature, the number of cells passing through the membrane was observed under a light microscope. Cells were counted in five random fields, and the average number of cells per field was calculated.

Thirty-two HCC samples and 7 normal liver samples from GSE113850 were divided into two groups (HCC vs. NC). In order to investigate the further function of LINC01554, GSEA (http://software.broadinstitute.org/gsea/index.jsp) was carried out between the two groups. Annotated gene were selected as the reference gene sets. False discovery rate (FDR) <0.25, and gene size ≥100 were regarded as the cut-off criteria.

Co-LncRNA (http://bio-bigdata.hrbmu.edu.cn/Co-LncRNA/) is a web-based computational tool that predicts the CEGs of single or multiple lncRNAs. We analysed the same RNA-Seq datasets to identify the CEGs. The regression/correlation method was set as Linear Regression. The corresponding coefficient threshold values were set at >2, and the adjusted P-value was set at <0.01.

The GO resource (http://www.geneontology.org/) describes gene functions and groups them into three categories: Molecular function, cellular component, and biological process. KEGG (http://www.genome.jp/kegg/) contains a set of genomes and biological pathways that provide information on genome functions and the relationships among genes. We visualised all known GO functional annotations and KEGG pathways of the CEGs using DAVID (http://david.ncifcrf.gov). GO and KEGG enrichment plots were also created.

Statistical analysis was performed using the SPSS 22.0 software (IBM Corp.). All experiments were performed in triplicate, and all data are presented as means ± standard error of the mean. All data are presented as a comparison between the groups and analysed by Student's t-test or analysis of variance (ANOVA). Receiver operating characteristic (ROC) curves were used to determine the correlation between the true- and false-positive rates, and this was important for assessing the diagnostic value of the genes. Results were regarded as statistically significant at P<0.05.

After analysing the GSE113850 dataset, we identified 229 DEGs, of which 151 were downregulated and 78 were upregulated, as shown in a volcano plot (Fig. 1C). Hierarchical clustering analysis of the top 10 DEGs with a fold change >2 is presented as a heat map (Fig. 1B). Three lncRNA genes (DIO3OS, LINC01554, and LINC01093) were among these DEGs. After analysing TCGA data, we identified 148 differentially expressed lncRNAs, and hierarchical clustering analysis of the top 10 lncRNAs with a fold change >1 is presented as a heat map (Fig. 1A). The Venn plot shows that LINC01554 was the only differentially expressed lncRNA in both the GSE113850 and TCGA datasets (Fig. 1D).

Using GEPIA, we compared the expression levels of the three lncRNAs, DIO3OS, LINC01554 and LINC01093, in the HCC and normal liver tissues. The results showed that the lncRNA expression levels were lower in HCC tissues than these in the normal liver tissues (Fig. 2A-C). The area under the curve (AUC) values for DIO3OS, LINC01554 and LINC01093 were 0.793, 0.729 and 0.969, respectively (Fig. 2D-F), indicating a moderate diagnostic value of their expression levels in HCC. These results were based on the original data from TCGA. We also analysed the lncRNA expression levels according to the tumour stage of HCC patients. Among the three lncRNAs, LINC01554 showed expression levels that significantly differed according to the HCC stage (Fig. 2G-I). We further explored the effects of these lncRNAs on the survival of HCC patients. These results showed that decreased DIO3OS LINC01554 and LINC01093 were significantly correlated with disease-free survival (DFS) of all HCC patients (Fig. 2J-L), and decreased levels of all three lncRNAs were significantly correlated with overall survival (OS) of all HCC patients (Fig. 2M-O). Meanwhile, we also explored the correlation between LINC01554 expression and the clinicopathological features in 40 HCC patients to evaluate the clinical role of LINC01554 in HCC (Table II). Lower LINC01554 expression levels were strongly linked to certain clinical characteristics, including HBV infection (P=0.0044), pathological stage (P=0.0079), tumour size (P=0.0252), portal vein tumour thrombus (P=0.0471), and TNM stage (P=0.0022).

LINC01554 expression is downregulated in human HCC samples. We analysed LINC01554 expression in 40 paired HCC and normal tumour-adjacent tissue samples by RT-qPCR. The results showed that LINC01554 expression was significantly lower in the HCC samples (P<0.01, Fig. 3B), and downregulated LINC01554 expression was observed in 70.0% (28/40) of the HCC samples (Fig. 3A). We also investigated LINC01554 expression in the liver cancer cell lines (SK-Hep1, HepG2, and HCCLM9). RT-qPCR results showed that LINC01554 expression was relatively lower in the SK-Hep1 and HCCLM9 cells than in the HepG2 cells (Fig. 3C). Therefore, SK-Hep1 and HCCLM9 cells were used for subsequent experiments. To study LINC01554 functions in HCC cell lines, a recombinant lentivirus that carried the LINC01554 DNA sequence (Lv-LINC01554) and a recombinant lentivirus without the LINC01554 DNA sequence (Lv-NC), as a control, were transfected into HCC cells. Transfection efficiencies were confirmed by RT-qPCR (P<0.01; Fig. 3D).

According to the CCK-8 assay results, the OD450 values of the SK-Hep1 and HCCLM9 cell lines in the Lv-LINC01554 group at 0, 24, 48 and 72 h were significantly lower than those in the Lv-NC group (P<0.01; Fig. 4A). We examined the distribution of cells at different stages of the cell cycle using flow cytometry. The results revealed that LINC01554 overexpression promoted G0/G1 arrest (P<0.05; Fig. 4B), but it did not significantly affect apoptosis (Fig. 4C).

Wound-healing and Transwell invasion assays were used to investigate the role of LINC01554 in HCC cell migration and invasion. Scratch assays showed that the migration rate of the Lv-LINC01554 transfection group was significantly lower than that of the Lv-NC transfection group (P<0.01; Fig. 5A). Transwell experimental results showed that the number of migrated cells in the Lv-LINC01554 transfection group was significantly lower than that in the Lv-NC transfection group (P<0.01; Fig. 5B). Next, we assessed EMT-related gene expression in the HCCLM9 and SK-Hep1 cells. The results of the western blot analysis showed that LINC01554 overexpression led to increased ZO-1 and E-cadherin expression, but decreased N-cadherin and vimentin expression (Figs. 5C and S1). Moreover, FISH assays showed that LINC01554 was mainly located in the nuclei of the SK-Hep1 and HCCLM9 cells (Fig. 5D).

GSEA was used to map into 6 functional gene sets. A total of 6 functional gene sets were enriched, which were mainly ‘Epithelial mesenchymal transition’, ‘G2M checkpoint’, ‘Kras signaling up’, ‘IL6/JAK/STAT3 signaling’, ‘Bile acid metabolism’ and ‘Fatty acid metabolism’ (Fig. 6).

We analysed the GO functions and KEGG pathways of the genes co-expressed with LINC01554 using the DAVID online database. Only the top 10 pathways for each category are listed (Fig. 7A-D). GO analysis showed that the CEGs were mainly enriched in ‘intracellular signal transduction’, ‘transcription, DNA-templated’, ‘negative regulation of transcription by the RNA polymerase II promoter’, ‘negative regulation of cell proliferation’, ‘response to hypoxia’, ‘canonical Wnt signalling pathway’, ‘cell migration’, and ‘cell cycle arrest’. KEGG pathway analysis showed that these genes were related to PI3K-Akt, Ras, Wnt, chemokine, and tumour necrosis factor (TNF) signalling pathways; viral carcinogenesis; and toxoplasmosis, hepatitis B, and non-alcoholic fatty liver disease pathways.

To examine whether LINC01554 exerts its function through Wnt and PI3K-Akt signalling in HCC cells, the expression levels of proteins related to these pathways (Akt, p-Akt, Gsk3β, p-Gsk3β and β-catenin) were measured in the LINC01554-overexpressing HCCLM9 and SK-Hep1 cells (Figs. 8A and B and S2). The results showed that LINC01554 overexpression inhibited Akt, p-Akt, β-catenin, and p-Gsk3β expression.