16 pairs of HCC tissue samples and background liver tissue (BL) were obtained from patients suffering HCC who underwent partial hepatectomy between January 2019 and December 2020. After surgical resection, specimens were immediately snap-frozen in liquid nitrogen. RNA-later (#AM7201, ThermoFisher, USA) was added to the cryogenic vials to prevent RNA degradation for subsequent extraction. Collection and usage of these samples followed ethical and institutional guidelines; all protocols were implemented after obtaining written informed consent from donors. This study was approved by the local Zhongnan Hospital Ethics Committee of Wuhan University (Approval No.2018078).

Cell lines (L02, Huh7, HepG2, HCCL-M3) involved in this experiment were purchased from Stem Cell Bank, Chinese Academy of Sciences. A mixture of Dulbecco’s Modified Eagle’s Medium (ThermoFisher), 10% fetal bovine serum (ThermoFisher), and 1% penicillin–streptomycin were used in cell culturing. All cells were grown in a humidified environment at 5% CO₂ and 37°C.

Microarray data adopted for analysis have been described in detail before (23). Data were uploaded to NCBI’s Gene Expression Omnibus (GEO) public database http://www.ncbi.nlm.nih.gov/geo/, accession number, GSE20077).

Total RNA extraction was performed using Trizol Reagent (Invitrogen, USA) according to manufacturer instructions. Detection of RNA quality and concentration was accomplished using the Nanodrop 2000 spectrophotometer (ThermoFisher); cDNA was obtained via a reverse transcription reaction with relevant kits (Toyobo, Japan). Gene and miRNA expression were examined using UltraSYBR Mixture (CWbio, China) and the CFX96Touch RT-PCR system (Biorad, USA). The mRNA levels of miR-21-3p and relevant genes were normalized to U6 (Ribobio, China) and GAPDH, respectively. The relative expression ratio was calculated using the 2^−ΔΔCq method. Primers used are listed in Table S1 .

Gain and loss of function of miR-21-3p were accomplished by transfecting miRNA mimics and inhibitors, respectively. Mimic- (mimic-NC) and inhibitor- (in-NC) negative controls were purchased from Ribobio. The complementary strand of miR-21-3p served as an inhibitor of miR-21-3p and competed for binding sites between miR-21-3p and target genes, but would not decrease the expression levels of miRNAs. The final concentration of these reagents was 50 nM. The SMAD7 expression vector (P-SMAD7) was constructed using a GV141 vector (GeneChem, China) with plasmids-NC (P-NC) serving as an experimental control. All transfection experiments were performed using Lipofectamine 2000 (Invitrogen, USA) according to manufacturer guidelines.

Proteins were collected using lysis buffer (Beyotime Biotechnology, China) containing protease inhibitor (PMSF, 1:100) and were centrifuged at 20627g for 20 min at 4°C; the supernatant was subsequently transferred to new tubes. Protein concentrations were detected using a BCA protein kit (Beyotime). Lysis buffer was mixed with protein loading buffer and protein was denatured at 100°C for 5 min. An equal amount of proteins was separated using SDS-polyacrylamide gels and then transferred to polyvinylidene fluoride (PVDF) membranes (Millipore, USA). Membranes were blocked for 2 h with 5% skimmed milk and incubated with primary antibodies overnight at 4°C. Secondary antibodies were subsequently added for 1.5 h at room temperature. Protein bands were obtained using an enhanced chemiluminescence (ECL) kit (Servicebio, Wuhan, China) using GENESys (Synoptics Ltd, China). anti-SMAD7 (#25840-1-AP) was purchased from Proteintech (China). anti-YAP1 (#14074), anti-E-cad (#3195), anti-N-cad (#13116), anti-vimentin (#5741) and anti-CTGF (#86641); these antibodies were purchased from Cell Signaling Technologies (USA). Anti-Bcl2 (#ab182858), anti-Bax (#ab32503) and anti-GAPDH (#ab8245) antibodies were purchased from Abcam (UK) while Goat Anti-Rabbit IgG HCS (#A25222) and goat anti-mouse IgG-horseradish peroxidase (#A25012) were purchased from Abbkine (USA).

A pmirGLO-SMAD7 vector was constructed, consisting of predicted binding sites that were mutated and ligated between the PmeI and XbaI restriction enzymes sites of the pmirGLO Dual-Luciferase miRNA Target Reporter Vector (#E1300, Promega, USA). The SMAD7 3′-UTR region contains two putative binding sites for miR-21-3p, with seed regions at 1,428–1,434 and 1,445–1,451. The mutant 3 (Mut 3) strand mutated two binding sites simultaneously. The wild-type (WT) and mutant SMAD7 3′-UTR luciferase reporter plasmids (0.25 μg/well) were co-transfected into huh-7 cells with miR-21-3p mimics or miR-NC (50 nM, 0.15 μl/well), respectively. A Dual-Glo Luciferase Assay System (#E2920, Promega) was employed to investigate firefly luciferase and Renilla luciferase activity 24 h after transfection, respectively. Firefly luciferase activity was normalized to Renilla luciferase activity. Data were collected using Enspire 2300 (PerkinElmer). Six repetitions per group were calculated.

A total of 4 × 10⁵ cells were transfected with miR-21-3p mimics with/or SMAD7 plasmids. After 24 h, cells were collected (including dead cells in culture medium); Huh-7 and Hep-G2 cells were processed according to Annexin V-FITC/PI Double stain apoptosis detection kit manufacturer instructions (#4101-2, BestBio, China). Fluorescence intensity on the flow cytometer was promptly collected to investigate the rate of apoptosis (Cytoflex Beckman, China).

The protocol for processing HCC cell lines was identical to the above cell apoptosis assay. Cells were harvested at 24 h post-transfection, suspended in 5% FBS culture medium and seeded at 5 × 10⁴ cells per upper chamber (6.5 mm. in diameter, 8.0 um pore size, Corning, USA). The lower chamber was filled with 600 μl of medium (15% FBS). Cells were fixed with 4% paraformaldehyde for 20 min at room temperature and stained with 0.1% crystal violet (Sigma-Aldrich, USA). The upper cells were wiped with cotton swabs. Random photographs of 200× fields were taken using an inverted microscope (Olympus IX3). The difference between cell invasion and migration assays was that in the latter, one upper chamber required pretreatment with 0.3% Matrigel matrix (#356234, Corning); this was performed at 37°C for 4 h. Three independent experiments were performed.

A Wistar rat model was established as previously described and rats were grouped according to the Metavir score system (24, 25). In brief, Wistar rats (male, aged 7–8 weeks and weighing 200–220 g) were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd (China). All animal handling methods and experimental procedures were approved by the Animal Care and Use Committee of Wuhan University in accordance with the Animal Experiment/Animal Biosafety Level-III Laboratory Guidelines. Rats were injected with 40% carbon tetrachloride (CCl4) dissolved in maize oil (1.5ml/kg) twice per week. At 8, 14, and 18 weeks, liver fibrosis, cirrhosis, and HCC manifested, respectively. Hematoxylin–eosin and Masson staining images of rat tissues are shown in Figure S1A .

Immunohistochemical staining of patient and animal tissues was performed by the Servicebio Company. Tissue photographs were taken using an inverted microscope (Olympus IX3). Positively stained cells were counted in at least five fields from each area at 400× magnification. Primary antibodies were:

Anti-SMAD7: (#25840-1-AP, Proteintech): 1:200; anti-YAP1 (#14074, Cell Signaling Technology): 1:400.

GSVA, an open-source software package for R (3.5.2), was adopted to explore the potential biological relevance of genes according to their levels of expression. Findings are presented in the form of the intuitive volcano and specific heat maps. Gene terms with |logFC| ≥0.1 and P <0.05 were considered statistically significant. The KEGG gene sets (c2.cp.kegg.v6.2.symbols.gmt) downloaded from the Molecular Signatures Database–MsigDB (http://www.broad.mit.edu/gsea/msigdb/index.jsp) were used for miRNA enrichment analysis.

MicroRNA Target Prediction Database (miRDB), PicTar and TargetScan databases were used for searching miRNA targets and predicting binding sites. Cohort data, including RNA sequencing and clinical data of 376 HCC patients were downloaded from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/). Data with incomplete clinical information were excluded during analysis. Clinical characteristics of HCC patients from the TCGA database are detailed in Table 1 .

Data were presented as mean ± standard deviation (SD) and analyzed using the Student’s t-test or one-way ANOVA. Quantitative data were representative of three experiments. Wilcoxon signed-rank and Wilcoxon rank-sum tests were adopted to analyze gene expression in paired and non-paired tissue samples. The relationship between clinicopathological characteristics and gene expression was evaluated using the Wilcoxon rank-sum test. Prognosis analysis was performed using the Kaplan–Meier method and univariate Cox regression. All statistical analyses and data plotting were performed using R (v.3.5.2); P< 0.05 was considered statistically significant.

Based on our earlier chip analysis results, which were sorted by P-value and differential expression, miR-21-3p was found to be significantly (P = 0.0278) upregulated in HCC as compared to normal liver tissue ( Table 2 ). To further verify chip analysis results, miR-21-3p mRNA expression was examined in both human HCC as well as Wistar rat model tissues, respectively. Results revealed miR-21-3p to have been significantly enriched in human HCC tissues as compared to background liver tissues [ Figure 1A(a) ]. Such upregulation was also observed in tissues obtained from rats with late-stage liver fibrosis and cirrhosis [ Figure 1A(b) ]. Relative to normal human L02 cells, miR-21-3p levels were increased in Huh-7, Hep-G2, and HCCL-M3 HCC cell lines [ Figure 1A(c) ]. Considering expression consistency in human tissues, rat liver disease models and cell lines, miR-21-3p upregulation was further investigated. Data downloaded from TCGA, including the clinicopathological characteristics of 376 HCC patients ( Table 1 ), were divided into high and low miR-21-3p expression groups. Patients with high miR-21-3p expression were significantly correlated with advanced clinical stages (P = 0.029), especially higher T staging (P = 0.026), while no significant differences were observed in patient age or gender ( Figure 1B ). Higher miR-21-3p expression was also correlated to shorter 10-year overall survival (OS) time (log-rank P = 0.026) ( Figure 1C ). GSVA analysis of miR-21-3p displayed that high miR-21-3p expression was mainly enriched among 20 pathways in the progression of HCC ( Figure 1D ). The top three relevant gene sets were involved in nitrogen metabolism, fatty acid metabolism and primary bile acid biosynthesis, underscoring the vital influence miR-21-3p exerts on cellular metabolic activity ( Figure 1E ). KEGG pathway analysis of miR-21-3p targets revealed that downstream targets were strongly associated with metabolic pathways, the Hippo signaling pathway and TGF-β transduction pathway ( Table 3 ). The intersections of miR-21-3p potential targets from three authoritative databases, namely TargetScan, PicTar, and miRDB, were SMAD7, HBP1, and FBXO11, respectively ( Figure 1F ). Of these, SMAD7 scored highest ( Table 4 ). The relationship between miR-21-3p and SMAD7, however, requires further elucidation.

The KEGG enrichment results of miR-21-3p and prediction of potential target scores were considered when evaluating expression of SMAD7 in human and rat model tissue samples, respectively. Compared to background liver tissue, SMAD7 was decreased in HCC (9/10) [ Figure 2A(a) ]. In addition, lighter brown staining was observed in HCC as compared to background liver tissue [ Figure 2B(a) ]. Along with the progression of liver disease in rat models, SMAD7 deletion in rats with advanced fibrosis and cirrhosis (F3, F4) (F: fibrosis) was observed in terms of protein expression levels when compared to normal rats [ Figure 2A(b) ]; immunohistochemical staining confirmed this trend [ Figure 2B(b) ]. Considering the apparent up-regulation of miR-21-3p and down-regulation of SMAD7 in HCC, the Dual-Luciferase assay was performed in Huh-7 cells to confirm the linear regulatory association among miR-21-3p and SMAD7. Binding site prediction revealed two high-scoring potential binding sites in the 3′-UTR region of SMAD7. In the miR-21-3p and reporter vector containing wild-type SMAD7 co-transfection group, luciferase activity was found to be significantly inhibited, while in the group containing mutant SMAD7 reporter vectors, the inhibition efficiency was not found to be statistically significant between miR-NC and miR-21-3p groups ( Figure 2C ). The above findings indicate that the predicted binding sites likely both inhibit SMAD7.

Expression of SMAD7 increased after transfection of miR-21-3p inhibitors and decreased following transfection of miR-21-3p mimics in Huh-7 [ Figure 3A(a) ] and Hep-G2 [ Figure 3A(b) ] cells. To investigate the effects of the miR-21-3p/SMAD7 axis on HCC malignant phenotypes, biomarkers associated with cell apoptosis, migration, and invasion were studied. The pro-apoptotic protein Bax and epithelial signature protein E-cadherin (E-cad) were found to be upregulated, while the anti-apoptotic protein Bcl-2, mesenchymal characteristic protein N-cadherin (N-cad) and vimentin were downregulated after down-regulating miR-21-3p in Huh-7 and Hep-G2 cells, respectively ( Figure 3A ). Effects of miR-21-3p on malignant cellular biomarkers were partly attenuated by SMAD7 co-transfection in both cell lines ( Figure 3B ). Flow cytometry results revealed that rates of early apoptosis, as evidenced by green staining, were increased after SMAD7 transfection (P < 0.01) ( Figure 4A ). Migration [ Figure 4B (a-b) ] and invasion [ Figure 4B(c-d) ] were enhanced by transfection of miR-21-3p mimics and partly reversed by SMAD7 restoration.

KEGG pathway enrichment analysis revealed that miR-21-3p associates significantly with the Hippo/YAP1 pathway ( Table 3 ). Notably, a prior study reported cytoplasmic YAP1 to enhance SMAD7 binding to activated TβRI (22). To further investigate potential associations among, miR-21-3p, SMAD7 and YAP1 in HCC, YAP1 expression was detected in both human and rat model HCC tissues. Compared to background liver tissue, YAP1 was found to be enriched in human HCC tissues (8/10) ( Figure 5A ), as well as in advanced liver fibrosis and cirrhosis tissues compared to early stages (F1, F2) (F: fibrosis) [ Figure 2A(b) ]. Darker brown staining on immunohistochemistry was observed in HCC tissues [ Figure 5B (a) ] as well as F3 and F4 stages in rat tissues as compared to F1 [ Figure 5B (b) ]. In both Huh-7 and Hep-G2 cells, YAP1 expression was upregulated by miR-21-3p and inhibited after reducing miR-21-3p [ Figures 5C(a, b) ]. On the protein level, promotion of YAP1 expression was partly attenuated by co-transfecting SMAD7 [ Figure 5C(b) ]. However, neither miR-21-3p nor SMAD7 had any significant influence on YAP1 mRNA levels [ Figures 5C (c, d) ]. Besides, mRNA ( Figure S2A ) and protein [ Figure 5C(b) ] expression trends of connective tissue growth factor (CTGF), previously demonstrated to be the direct nuclear target of YAP1, were consistent with YAP1 here (26).

SMAD7 was decreased (P = 4.578e-04) and YAP1 (P = 3.6e-05) was increased in HCC compared to adjacent normal liver tissues based on data of 376 HCC patients downloaded from TCGA database ( Figure 6A ). Analysis of RNA sequencing and patient clinicopathological data revealed that higher levels of SMAD7 associate with lower grades (P = 0.047) [ Figure 6B (a) ]. In addition, higher YAP1 levels were found to associate with higher disease stages (P = 0.037), in particular, M stages (P = 0.045) [ Figure 6B(b) ]. Co-survival analysis revealed that lower miR-21-3p/higher SMAD7 (P = 0.0494) and lower miR-21-3p/lower YAP1 (P = 0.0379) ratios associate with a better five-year OS rate ( Figure 6C ). GSVA data, presented as a volcano map, revealed SMAD7 to be mainly involved in 18 pathways ( Figure 6D ). Among them, the TGF-β signaling pathway and the Notch signaling pathway were most relevant in HCC ( Figure 6E ).