Identification of invasion-metastasis-associated microRNAs in hepatocellular carcinoma based on bioinformatic analysis and experimental validation

In the discovery step, we only included datasets that compared the miRNA expression in invasive/metastatic HCC cell or tissue with non-invasive/metastatic HCC cell or tissue. Besides, only datasets containing more than twenty samples were included. The titles and abstracts of these datasets were screened, and the full information of the datasets of interest were further evaluated. Finally, only GSE67140 dataset was selected to further study. The dataset GSE67140 based on the platform of GPL8786 (Affymetrix Multispecies miRNA-a Array) contained 91 human HCC tumor samples without vascular invasion and 81 samples with vascular invasion was downloaded from the National Center for Biotechnology Information (NCBI) GEO database (https://​www.​ncbi.​nlm.​nih.​gov/​geo).

Data were normalized using the normalizeBetweenArray function from R package ‘LIMMA’ from the bioconductor project [13]. Data before and after normalization were shown in Fig. 1a, b, respectively. The miRNA differential expression analysis was conducted using the limma software package in the Bioconductor package (http://​www.​bioconductor.​org/​). The related codes were put into R, and the DE-miRNAs in HCC tumor samples with vascular invasion compared to HCC tumor samples without vascular invasion were analyzed through the limma package. P-value < 0.05 and |fold change (FC)| > 2 were set as the thresholds for identifying DE-miRNAs.

The potential target genes of the top three most upregulated and downregulated miRNAs were predicted by miRTarBase (http://​mirtarbase.​mbc.​nctu.​edu.​tw/​php/​index.​php), which is an experimentally validated microRNA-target interactions database [14].

The database for annotation, visualization and integrated discovery (DAVID 6.8, http://​david-d.​ncifcrf.​gov/​) was introduced to perform functional annotation and pathway enrichment analysis for the predicted targets of the selected 6 DE-miRNAs, including GO and KEGG pathway analysis [15, 16]. P-value < 0.05 was considered as statistically significant.

PPI network and miRNA-gene network were successively constructed. The target genes were first mapped to the STRING database (http://​string-db.​org) to assess functional associations among these target genes [17]. Only the interactions with a combined score > 0.4 were considered as significant. Next, to obtained the hub genes, the degree of connectivity in the PPI network was analyzed by Cytoscape software (version 3.6.0), after which miRNA-hub gene network was established.

The human HCC cell lines HepG2, Bel-7402, HCCLM3 and SMMC7721 and normal cell line HL7702 were kindly provided by the First Affiliated Hospital of Medical College, Zhejiang University (Hangzhou, China). HepG2, Bel-7402, HCCLM3 and HL7702 were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco,12430047) supplemented with 10% fetal bovine serum (FBS; Biological Industries, 04-0101-1, Cromwell, CT, USA) and SMMC7721 was maintained in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco, 31800105, Life Technologies, Carlsbad) containing 10% FBS under a humidified atmosphere of 5% CO₂ at 37 °C. HCC tumor tissues and matched noncancerous tissues were also obtained from the First Affiliated Hospital of Medical College, Zhejiang University (Hangzhou, China).

The miRNA mimics and negative control (purchased from RiboBio, Guangzhou, China) were transfected into HCC cell lines using Lipofectamine™ 3000 according to the manufacturer’s instruction.

RNAiso plus Reagent (TaKaRa, Kusatsu, Japan) was introduced to extract total RNA from HCC cell lines and clinical samples. Subsequently, miRNAs reverse transcription primers and q-PCR primers were purchased from Ribobio Co. Ltd (Guangzhou, China). Q-PCR was performed in triplicates by SYBR Premix Ex Taq (TaKaRa, RR420A). The U6 small nuclear RNA was used as the internal control. The 2^−ΔΔCT method was utilized to determine fold change in the RNA level of each sample compared with the reference sample.

The expression levels of miR-494-3p and miR-126-3p in HCC tissue from the Cancer Genome Atlas (TCGA) database (https://​genome-cancer.​ucsc.​edu/​) were determined. Firstly, the miRNA-sequencing data in patients of liver cancer was downloaded from the database. Then, miR-494-3p and miR-126-3p expression values were extracted, after which unpaired Student’s t-test was employed to conduct statistical analysis between normal group and tumor group. A P-value < 0.05 was considered as significant.

Cells were plated in six-well plates with 1 × 10⁵ cells per well. After transfection (50 nM), the cells were grown to 100% confluence in six-well plates. Then, a micropipette tip was applied to make a cross wound. Photographs were taken by a microscopy immediately or 24 h after wounding.

The cell invasion assay was performed using 24-well transwell chambers (Corning, USA). Firstly, the inserts were coated with Matrigel (BD Bioscience, USA) on the upper surface. After transfection (50 nM), 1 × 10⁵ cells were suspended in 0.2 ml serum-free medium and added to the inserts. 0.6 ml medium with 20% FBS was added to the lower compartment as a chemoattractant. After incubation at 37 °C for 48 h, the cells on the upper surface of the membrane were carefully removed using a cotton bud and cells on the lower surface were successively fixed with 100% methanol and stained with 0.1% crystal violet. Five visual fields of 200× magnification of each insert were randomly selected and counted under a light microscope (Olympus, Japan).

UALCAN, a portal for facilitating tumor subgroup gene expression and survival analyses, provides easy access to publicly available cancer transcriptome data including TCGA [18]. The database was utilized to analyze the expression levels of potential targets of miR-494-3p and miR-126-3p between normal tissue and tumor tissue. In brief, gene symbols were firstly pasted in the text area, and liver hepatocellular carcinoma dataset was chosen. Then, the expression information of each target gene could be acquired by linking to gene expression analysis result. Statistical analysis of comparison between normal group and tumor group was performed and log-rank P-value was observed in the database.

The prognostic values of miR-494-3p and miR-126-3p in HCC were analyzed using Kaplan–Meier plotter (KM plotter) database [19]. In brief, the two miRNAs were entered into the database, after which Kaplan–Meier survival plots were generated and hazard ratio (HR), 95% confidence intervals (CI), log rank P-value were displayed on the webpage. P-value of < 0.05 was regarded as statistically significant.

The results were shown as mean ± SD. Differences between two groups were estimated using unpaired Student’s t-test. A two-tailed value of P < 0.05 was considered as statistically significant.

To identify DE-miRNAs of miRNA array GSE67140 downloaded from GEO, we conducted a differential expression analysis using limma software package. Based on this analysis, a total of 138 miRNAs were found to be significantly differentially expressed in HCC tumor samples with vascular invasion more than two-fold when compared to HCC tumor samples without vascular invasion, including 57 upregulated and 81 downregulated miRNAs. For better visualization, the top ten most upregulated miRNAs and top ten most downregulated miRNAs were presented in Tables 1 and 2, respectively. According to fold change (FC), miR-494-3p, miR-1207-5p and miR-1268a were the top three most upregulated miRNAs; and miR-126-3p, miR-199a-3p and miR-199b-3p were the top three most downregulated miRNAs. 485 potential target genes were predicted for the three upregulated miRNAs and 277 genes for the three downregulated miRNAs by using miRTarBase. A volcano plot of these DE-miRNAs was provided in Fig. 2.

Three categories of GO functional annotation analysis were performed on these potential target genes mentioned above, including biological process (BP), cellular component (CC) and molecular function (MF). As shown in Fig. 3a1–a3, the enriched GO functions for target genes of the three upregulated miRNAs included the regulation of transcription, transcription, regulation of programmed cell death, regulation of cell death and regulation of apoptosis in the BP category; intracellular non-membrane-bounded organelle, non-membrane-bounded organelle, membrane-enclosed lumen, organelle lumen and intracellular organelle lumen in the CC category; and DNA binding, transcription regulator activity, transcription factor binding, RNA binding and sequence-specific DNA binding in the MF category. The enriched GO functions for target genes of the three downregulated miRNAs were presented in Fig. 3b1–b3, involving the regulation of transcription, regulation of RNA metabolic process, positive regulation of cellular biosynthetic process, positive regulation of macromolecule metabolic process and phosphorylation in the BP category; membrane-enclosed lumen, organelle lumen, intracellular organelle lumen, nuclear lumen and cytosol in the CC category; and transcription regulator activity, adenyl ribonucleotide binding, ATP binding, purine nucleoside binding and nucleoside binding in the MF category.

To further analyze the enriched pathways of these target genes, we subsequently conducted KEGG pathway enrichment analysis. For upregulated miRNAs (Fig. 4a), the enriched KEGG pathways contained pathways in cancer, focal adhesion, glioma, neurotrophin signaling pathway and insulin signaling pathway. And the enriched KEGG pathways for downregulated miRNAs (Fig. 4b) included pathways in cancer, focal adhesion, MAPK signaling pathway, neurotrophin signaling pathway and chemokine signaling pathway. Interestingly, both target genes of upregulated miRNAs and targets of downregulated miRNAs were enriched in the process, focal adhesion. Accumulating evidence has demonstrated that dysregulation of focal adhesion plays a central role in cell invasion and migration [20]. Therefore, 17 genes (PRKCA, HRAS, TLN1, MAP2K1, PPP1CC, PTEN, AKT1, VEGFB, MAPK1, IGF1R, CCND1, ARHGAP5, BCL2, RAC1, PPP1R12A, RAP1A, RAP1B) of upregulated miRNAs and 21 genes (PIK3CG, CAV2, FLT1, ROCK1, MET, IGF1, ITGA3, HGF, KDR, AKT1, MAPK1, CRKL, PAK4, BCL2, VEGFA, MAPK9, MAPK8, CRK, PIK3R1, AKT2, PIK3R2) of downregulated miRNAs which are involved in focal adhesion may interact with the selected 6 miRNAs, thereby modulating invasion and metastasis in HCC.

The PPI networks of target genes of the six selected DE-miRNAs were constructed using the STRING database. Data from this database revealed that lots of these target genes could interact with each other. The top ten hub genes listed in Table 3 were screened out according to the node degree. For the upregulated miRNAs, the hub genes were TP53, AKT1, MAPK1, MYC, BCL2, HRAS, PTEN, CALM3, CDKN1A, CYCS. For the downregulated miRNAs, the hub genes were MYC, AKT1, VEGFA, MAPK8, PIK3CG, BCL2, FGF2, KRAS, PIK3R1, SIRT1. Among these genes, TP53 and MYC demonstrated the highest node degrees, which were 80 and 42, respectively. The results suggest that TP53 and MYC may be two key targets correlated with HCC invasion and metastasis.

Subsequently, the miRNA-hub gene network was constructed by cytoscape software as showed in Fig. 5. From Fig. 5a, we found that six (CYCS, MYC, AKT1, MAPK1, BCL2 and PTEN) in ten hub genes could be potentially modulated by upregulated miR-494-3p. 4 hub genes and 2 hub genes could be potentially regulated by miR-1207-5p and miR-1268a, respectively. Additionally, miR-126-3p could potentially target nine (PIK3R1, PIK3CG, SIRT1, BCL2, AKT1, MAPK8, VEGFA, KRAS and MYC) in ten hub genes. And 4 hub genes and 1 hub gene could be potentially modulated by miR-199a-3p and miR-199b-3p. These data supported that miR-494-3p and miR-126-3p might be two potential regulators of invasion and metastasis in HCC.

Next, we further evaluated the expression of potential targets of miR-494-3p and miR-126-3p in HCC using the UALCAN database as shown in Fig. 6. The analytic results suggested that the expression of five (CYCS, MAPK1, PTEN, AKT1, BCL2) in six targets of miR-494 was significantly upregulated in HCC tissue than normal tissue whereas MYC gene was markedly downregulated. It is widely acknowledged that there is an inverse relationship between miRNA expression and target gene expression. Thus, among these detected genes, the significantly downregulated MYC may be the most potential target for the upregulated miR-494-3p. Similarly, these significantly upregulated genes including AKT1, BCL2, MAPK8, VEGFA and KRAS may be the most potential targets for the downregulated miR-126-3p in theory.

Based on our previous findings, we intended to further determine the expression and prognostic roles of miR-494-3p and miR-126-3p in HCC. Firstly, we evaluated the expression of miR-494-3p and miR-126-3p in four HCC cell lines (HepG2, Bel7402, SMMC7721 and HCCLM3) compared with that in normal liver cell line (HL7702). The results presented in Fig. 7a, d demonstrated that miR-494-3p expression was markedly upregulated and miR-126-3p expression were significantly downregulated in all four HCC cell lines. Of note, expression of miR-494-3p and miR-126-3p in HCCLM3 is the most upregulated and downregulated compared with other three cell lines. HCCLM3 is a high metastatic HCC cell [21], further suggesting miR-494-3p and miR-126-3p may regulate HCC metastasis. The expression of miR-1207-5p, miR-1268a, miR-199a-3p and miR-199b-3p in HCC cell lines was also detected (Additional file 1 : Figure S1). However, the expression of them was only significantly dysregulated in one or two HCC cell lines as shown in Additional file 1 : Figure S1. To some extent, the result hinted that the functional roles of miR-1207-5p, miR-1268a, miR-199a-3p and miR-199b-3p in HCC may be not as good as miR-494-3p and miR-126-3p. Besides, miR-494-3p and miR-126-3p expression in 70 pairs of clinical tumor and normal samples also demonstrated a higher expression of miR-494-3p and a lower expression of miR-126-3p in HCC tissues than normal tissues (Fig. 7c, f), further hinting promoting and suppressive roles of miR-494-3p and miR-126-3p in HCC, respectively. Moreover, for miR-126-3p, similar results were obtained by downloading and analyzing HCC data from the Cancer Genome Atlas (TCGA) (Fig. 7e). However, for miR-494-3p, there was no significant difference between tumor tissue and normal tissue (Fig. 7b). To summarize, miR-494-3p was upregulated and miR-126-3p was downregulated in HCC, suggesting that they may play a major role in the onset and progression of HCC, especially in HCC invasion and metastasis.

KM plotter database was utilized to assess the prognostic values of the two potential miRNAs, miR-494-3p and miR-126-3p in HCC. As shown in Fig. 7g, high expression of miR-494-3p was significantly correlated with worse OS, further implying that miR-494-3p is an oncogene in HCC. Figure 7h indicated a favorable prognostic role of miR-126-3p in HCC. In short, all these findings suggest that miR-494-3p and miR-126-3p are aberrantly expressed in HCC and can serve as two prognostic biomarkers. Therefore, the two miRNAs, miR-494-3p and miR-126-3p were selected for further experimental excavation and research.

In the first place, miR-494-3p and miR-126-3p expression in HepG2 and HCCLM3 after transfected with miRNA mimics at final concentrations of 20, 50 and 100 nM were detected using qPCR (Additional file 2 : Figure S2). To determine the roles of miR-494-3p and miR-126-3p in HCC migration, we first performed a wound healing assay. As shown in Fig. 8, wound healing assay suggested that miR-494-3p overexpression markedly enhanced HepG2 and HCCLM3 migration whereas overexpression of miR-126-3p could significantly attenuate cell migration, when compared with negative control cells. Next, a transwell invasion assay was employed to measure the effects of miR-494-3p and miR-126-3p on HCC invasion. The results presented in Fig. 9 demonstrated that upregulation of miR-494-3p and miR-126-3p could evidently boost and suppress cell invasion, respectively. All these data confirmed that miR-494-3p and miR-126-3p were two key regulators of HCC invasion and metastasis. Targeting miR-494-3p and miR-126-3p may represent novel approaches in treating HCC patients with vascular invasion or/and distant metastasis, thus improving the prognosis of patients with HCC.