The published HCC cohort dataset, including gene expression profiles and relevant clinical information, can be downloaded from TCGA data portal (18). The clinical information of patients includes age, gender, preoperative diagnosis, Child–Pugh score of liver function, radical resection, postoperative pathological diagnosis, pathological grading and staging, survival time, last follow-up time, and so on. Only the survival time-related data in the complex clinical data are used, so only the data needed in the article are shown in ( Supplementary Tables 1 and 2 ). Data acquisition and application were conducted in accordance with TCGA release guidelines and data access policy without additional approval from the local ethics committee.

The limma package in R language (version 3.6.1) was subsequently used for the calculation of DEGs. Gene counts >0, the adjusted p value <0.05, and |log₂ fold change|>1 were set as the cutoff criteria. Similarly, Differentially Expressed microRNAs (DEmiRs) were selected by the three cutoff values, Gene Counts >0, the adjusted p value <0.05, and |log₂ fold change|>1. Heat maps and volcanic maps were drawn using R language. Then, the principal component analysis (PCA) figure about the samples was performed.

In order to understand the biological functions of selected DEGs and DEmiRs, we performed the enrichment analysis of DEGs and DEmiRs in Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) Database Pathways. The clusterProfiler (version 3.12.0) (19) package in R language was used for revealing the roles of DEGs in biological process (BP), cellular component (CC), and molecular function (MF). The adjusted p value was less than 0.01 and considered statistically significant.

R language GDCRNATools package was used to search and match on StarBase database and find out the common miRNAs targeting mRNA and lncRNA (20, 21). The mRNA and lncRNA are negatively correlated with miRNA, but mRNA and lncRNA present a positive correlation. Hypergeometric distribution test was used to assess the importance between each pair of them. We used false discovery rate (FDR) to correct p values while FDR <0.05 is the cutoff value. Meanwhile, the regulation of similarity correlation with miRNAs (similarity of correlation between miRNA and lncRNA expression and correlation between miRNA and mRNA expression) is not equal to zero. All the competitive lncRNA and mRNA were mixed together after we identified them under the conditions above. Finally, ceRNA network diagram was established in the Cytoscape (3.7.2) for visualization.

In order to find out the survival-related lncRNA–miRNA–mRNA pathways, we conducted a single-factor survival analysis on each lncRNA, mRNA, and miRNA node in the ceRNA network. The node was marked in the ceRNA network diagram, and the survival-related ceRNA network was constructed through the survival-related node. Then, samples were divided into the high-expression group and the low-expression group according to the median expression of each gene. The Kaplan–Meier survival was used to evaluate the difference of overall survival (OS) time between the two groups. Finally the SNHG1–miR-326–LMNB2 pathway was selected, and the effects of this pathway were experimentally verified in our study.

The human hepatocellular cancer cell lines (Huh7 and PLC) were obtained from the Cell Bank of Chinese Academy of Sciences, Shanghai Branch. The Huh7 and PLC cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum in 5% CO₂ and 90% humidity at 37°C.

The lentivirus vectors (sh-SNHG1, sh-LMNB2) and small interfering RNAs (siRNAs) against human SNHG1 or LMNB2 were synthesized by GenePharma Co. Ltd. The stable Huh7 cells with SNHG1 and LMNB2 knocked down were generated using lentiviral vectors. Infected cells were then treated with puromycin (2 µg/ml) for 2 days, and surviving cells were maintained in complete medium with puromycin (0.5 µg/ml). The siRNAs were transfected into hepatocellular cancer cells using Lipofectamine 2000 (Thermo Fisher Scientific, Inc.) according to their instructions. Besides, miR-326 mimics, miR-326 inhibitors, and negative controls were purchased from GenePharma Co. Ltd. When the cell confluence reached 50%, oligonucleotide transfection was performed using Lipofectamine 2000 according to the manufacturer’s protocol.

We constructed the wild-type plasmid SNHG1-WT and the mutant plasmid SNHG1-MUT. PLC and HuH7 cells that were seeded in 24-well plates were cotransfected with miR-326 mimic or negative control and wild-type or mutant plasmids using Lipofectamine 2000. The luciferase intensity on the microplate was measured with the Dual-Luciferase Reporter Assay System (Promega Corp.), and Renilla luciferase activity was normalized to firefly luciferase activity.

The total protein was extracted from the cells using radioimmunoprecipitation assay lysis buffer with protease inhibitors. Each lane is loaded with the same amount of total protein (20 μg), and the sample is separated by 10% sodium dodecyl sulfate polyacrylamide gel electropheresis (SDS PAGE) and then transferred to a polyvinylidene fluoride membrane (Roche Diagnostics). After blocking with 5% skim milk at room temperature for 1 h, primary antibody LMNB2 (Abcam) or glyceraldehyde 3-phosphate dehydrogenase (GAPDH; Abcam) was used at 4°C overnight. Subsequently, the membranes were incubated with anti-rabbit (1:3,000, cat. no. 7074, Cell Signaling Technology, Inc.) horseradish peroxidase-conjugated secondary antibodies at 37°C for 1 h. Finally, ECL Western blot substrate (Promega corp.) and FluorChem E system (Protein Simple) were used to observe the immune response zone.

According to the manufacturer’s protocol, TRIzol reagent (Thermo Fisher Scientific, Inc.) was used to extract total RNA from cells and tissues. The RNA purity was evaluated based on the A260/280 ratio. RNA was reverse transcribed into cDNA using miRNA first-strand cDNA synthesis kit [Accurate Biotechnology (Hunan) Co., Ltd.]. SYBR Green Premix Pro Taq HS qPCR Kit II [Accurate Biotechnology (Hunan) Co., Ltd.] was used for qPCR to detect the relative expression of the target gene. Thermal cycling conditions are as follows: Initial denaturation at 95°C for 30 sec; 40 cycles of 5 sec at 95°C; 1 min at 60°C and 72°C for 15 sec; with a final extension cycle at 72°C for 5 min. Finally, it step in the dissociation stage. The relative levels were calculated using the 2^-ΔΔCq method. The endogenous control gene is GAPDH. The primer sequence is shown in Table 1 .

Cells (1 × 10⁵) were seeded in the upper chamber of a Boyden chamber (8 μm aperture) in 200 μl of serum-free medium (Corning Corporation). The lower chamber was filled with 700 μl containing 10% fetal bovine serum (Gibco) as a chemical attractant. After 24 h of incubation at 37°C, cells remaining on the upper side of the membrane were removed with a cotton swab, and the cells that had migrated to the lower side of the membrane were fixed with 70% ethanol for 20 min and stained with 0.1% crystal violet for 20 min at room temperature. Then, the cells were counted with an optical microscope. The method of the invasion test is similar to cell migration assay except that the Boyden chamber is covered with a matrix before seeding the cells.

The cells were seeded in six-well plates and cultured to 80% confluence before we scratched the cell layer with a 20-μl pipette tip. The cells were then incubated in fresh medium containing 10% fetal bovine serum for 48 h. Scratches were observed under a fluorescence microscope after 24 and 48 h.

Female nude mice (aged 5–6 weeks, weight 18–22 g) were purchased from Hangzhou Ziyuan Biology and bred in a 12-h light/dark cycle and sterile conditions (temperature 26°C–28°C, humidity 40%–60%) with free access to water and food. We injected 3 × 10⁶ cells into the fore limbs of nude mice to generate transplanted tumors and measure the tumor size with a caliper every 3 days. After 19 days, the mice were sacrificed, and tumor images were captured. Tumor volume was calculated with the formula: maximum diameter × (minimum diameter) 2 × 0.5.

Statistical analysis was performed using SPSS 19.0 software (SPSS Inc., Chicago, IL, USA). The summary data are expressed as the mean ± standard error of the mean (SEM). We use the χ² test, Student’s t-test, or one-way analysis of variance with the least significant difference correction to assess the differences between groups. Spearman rank correlation analysis was used to evaluate linear regression. p value less than 0.05 was considered to be statistically significant in all cases.

The study protocol was approved by the research ethics committee of the Qilu Hospital of Shandong University.

Firstly, 421 HCC patient samples with para-tumor tissue available from TCGA database were analyzed using R language tools. We used p value <0.05 and |log₂ fold change|>1 as cutoffs to identify differential gene profiles. As a result, a total of 2,416 differentially expressed genes (DEGs), 2,181 differentially expressed mRNAs, and 148 differentially expressed lncRNAs were identified ( Figures 1A–C ). The green dots represent downregulated genes, and the red dots represent upregulated genes. Black dots represent genes without statistically significant change. The heatmap shows the DEGs between the primary tumor (red) and solid tissue normal (blue). Moreover, 131 differentially expressed miRNAs were screened using both R language and GDCRNATools ( Figures 1D–F ).

GO and KEGG enrichment analysis on these 2,416 DEGs was carried out to explore their functions (p value <0.01 was the cutoff value). The results showed that ( Figure 2A ) the differential genes mainly focus on the catabolic process of biological procedure extracellular matrix of cells and small molecular compounds bindings with the oxidative respiratory chains. Otherwise, KEGG enrichment analysis demonstrated that the differential genes concentrated upon the pathways participating in complement and coagulation cascades, cell cycle regulation, and cell metabolism ( Figure 2B ).

Based on the 2,181 differentially expressed mRNAs and 148 differentially expressed lncRNAs mentioned above, we constructed the ceRNA network containing 115 mRNAs, 12 lncRNAs, and 37 miRNAs, in which 343 different lncRNA–miRNA–mRNA pathways were also described ( Figure 2C ). Then, we intersected the selected 37 miRNAs with DEmiRs obtained previously and only four differentially expressed miRNAs including miR-326, miR-154-5p, miR-21-5p, and miR-93-5p were found, which means that these four miRNA-associated lncRNA–miRNA–mRNA pathways may function in HCC development.

In order to evaluate the prognostic significance of the obtained ceRNA network, we used univariate survival analysis, Cox regression, and Kaplan–Meier survival analysis and found that only miR-326 was associated with HCC patient survival among the four potential miRNAs (p < 0.01) ( Figure 3C ). So, the lncRNA–miRNA–mRNA pathways that associated with miR-326 were extracted from ceRNA as a whole ( Figure 2D ). And the heatmap of correlation matrix was drawn to visualize the gene expression correlation quantitatively ( Figure 2E ). We evaluated the prognostic significance of SNHG family members, such as SNHG1 SNHG3, SNHG12, and SNHG20; although the p value for SNHG1 survival analysis is slightly greater than 0.05, the survival analysis p values for 25% and 75% of the quartile subgroups were both less than 0.05 ( Figures 3A, B ). Furthermore, previous studies have demonstrated that lncRNA SNHG1 predicted a poor prognosis in HCC (16, 22), and the diagram showed that lncRNA SNHG1 is a potential lncRNA in HCC progression. Although three candidate targets ANKRD13B, LMNB2, and FANCE, which were regulated by miR-326, were all associated with HCC patient survival ( Figure 3D ), only LMNB2 has been investigated in a hepatocellular cancer study before (23). Previous studies showed that ANKRD13B and FANCE were associated with epidermal growth factor receptor (EGFR) activation (24) and Fanconi anemia (25, 26), respectively. Furthermore, correlation analysis indicated both lncRNA SNHG1 and LMNB2 were negatively correlated with miR-326, but SNHG1 correlated with LMNB2 positively in HCC ( Figures 4A–C ).

Firstly, SNHG1 level were examined in five HCC cell lines. As shown in Figure 5A , Huh7 and PLC have a higher expression of SNHG1, so we chose the two cell lines for further research. Three independent siRNAs were designed to silence SNHG1 expression. The results showed that SNHG1 expression decreased significantly compared with control group ( Figure 5B ). MTT assay and colony formation demonstrated that SNHG1 silencing inhibited cell proliferation significantly ( Figures 5C–F ). The transwell assay indicated that knockdown of SNHG1 inhibited the migration and invasion of Huh7 and PLC cells ( Figures 5G, H ), and these results were confirmed in the wound healing experiment using both Huh7 and PLC cells ( Figure 5I ). The above results showed that SNHG1 promoted the proliferation, migration, and invasion of hepatocellular cancer cells.

To investigate the role of SNHG1 in HCC growth in vivo, we cultured Huh7 cells stably expressing shSNHG1 using lentivirus ( Figure 6A ). As shown in Figures 6B, C (six mice per group), the knockdown of SNHG1 inhibited tumor growth significantly compared with control group ( Figures 6A, B ). Furthermore, decreased tumor size and weight were observed after SNHG1 was knocked down ( Figures 6D, E ). These results indicated that SNHG1 knockdown could influence tumor growth significantly in vivo. The miR-326 level of tumor increased markedly in the shSNHG1 group compared with control, whereas the expression of LMNB2 decreased ( Figure 6F, G ).

In order to confirm the regulatory axis of SNHG1–miR-326–LMNB2 in HCC mentioned above, we firstly perform the subcellular localization assay of SNHG1. Nucleocytoplasmic separation and RT-qPCR assay revealed that SNHG1 was expressed in both nucleus and cytoplasm of HuH7 and PLC cells, whereas a larger proportion of SNHG1 was located in the cytoplasm ( Figure 7A ). Many lncRNAs acted as ceRNAs through binding miRNAs (27). SNHG1 has also been reported to be involved in tumor development including colorectal cancer and osteosarcoma (28–30). Therefore, we used dual-luciferase reporter assays to make sure whether SNHG1 can directly interact with miR-326 in Huh7 and PLC cells. The results showed that the miR-326 mimics could reduce the luciferase activity in both Huh7 and PLC cells ( Figure 7B ). Meanwhile, wild-type and mutant SNHG1 luciferase reporter vectors were constructed using binding sites predicted by StarBase ( Figure 7C ), and it was found that miR-326 mimics reduced the luciferase activity significantly of wt-SNHG1 but showed no effect on mut-SNHG1. Furthermore, addition of wt-SNHG1 will attenuate the inhibition effect of miR326 on the LMNB2 expression level, while addition of mut-SNHG1 will not ( Figure 7D ). Actually, miR-326 level increased markedly after SNHG1 was knocked down in HuH7 and PLC cells ( Figure 7E ). These results were consistent with the correlation described in Figure 4 . Additionally, miR-326 mimics could inhibit cell proliferation significantly, while miR-326 inhibitor could accelerate HuH7 cell proliferation ( Figure 7F ). Bioinformatics analysis has revealed that LMNB2 is the potential target of miR-326 in HCC. To test the effect of miR-326 on LMNB2 expression, as the results shown in Figure 7G , LMNB2 level decreased markedly with miR-326 mimic treatment in Huh7 cells compared with control, whereas the expression of LMNB2 increased significantly with miR-326 inhibitor treatment. In a word, miR-326 could downregulate LMNB2 expression in HCC. Moreover, LMNB2 was observed to be upregulated in the HCC tissues with low expression of miR-326 ( Figure 7H ). Furthermore, the LMNB2 expression was also investigated after SNHG1 silence in HCC cells ( Figure 7I ). Moreover, wild-type and mutant LMNB2 luciferase reporter vectors were constructed using binding sites predicted by StarBase ( Figures 7J, K ), and it was found that miR-326 mimics significantly reduced the luciferase activity of wt-LMNB2 but showed no effect on mut-LMNB2. The data showed that LMNB2 RNA expression was reduced significantly after SNHG1 siRNA treatment. To sum up, LMNB2 is the target of miR-326 in HCC, and SNHG1 regulates LMNB2 expression via sponging miR-326 in HCC.

In order to explore the biological effects of LMNB2 in HCC, we also knocked down LMNB2 expression using siRNAs or lentivirus-mediated shRNA. The results showed that LMNB2 expression was reduced in HuH7 cells significantly from RNA to protein level ( Figures 8A, B ). It is worth noting that the proliferation ability of Huh7 cells decreased markedly with the silence of LMNB2 expression ( Figure 8C ). Furthermore, we further studied the role of LMNB2 on tumor growth in vivo (three mice per group). Twenty days after transfected HuH7 cells were injected into nude mice, the volume and weight of formed tumor were significantly smaller than those of control group ( Figures 8D, E ), also indicating that LMNB2 knockdown could inhibit the growth of HCC cells in vitro and in vivo. Besides, SNHG1 knockdown reduced LMNB2 expression compared with control in Huh7 cells, but SNHG1-induced downregulation of LMNB2 was reversed or enhanced by miR-326 inhibitor or mimic treatment, respectively. The similar results were also confirmed by Western blotting assays ( Figures 8F, G ). In addition, MTT assay proved that miR-326 inhibitor could rescue the role of SNHG1 knockdown on HuH7 cell proliferation. ( Figure 8H ). These data demonstrated that the oncogenic SNHG1 upregulated LMNB2 expression and promoted tumor growth via suppressing miR-326 level in HCC.