We sourced HCC-related miRNA expression datasets (GSE6857, GSE36915, GSE67140) and a gene expression dataset (GSE33006) from the GEO database. GSE6857 comprised 241 tumor samples alongside paracancer normal controls; GSE36915 included 68 tumor samples and 21 non-tumor liver tissues; GSE67140 consisted of 91 HCC samples without vascular invasion and 81 with vascular invasion; GSE33006 encompassed 3 normal controls and 3 HCC samples. Additionally, HCC-related miRNA, gene expression profiles, and survival data were obtained from TCGA via the UCSC Xena platform (https://xena.ucsc.edu/), including data from 375 HCC and 50 normal tissues. For differential gene expression analysis in HCC, we utilized the “limma” package in R (Bioconductor, Buffalo, NY, USA), setting a significance threshold of p 0.05 for filtering. Kaplan–Meier survival analysis was conducted using the “survival” package in R, with genes exhibiting a p-value of less than 0.05 deemed prognostically significant. To identify potential miRNA target genes, we accessed databases including miRDB (http://mirdb.org/), mirDIP (http://ophid.utoronto.ca/mirDIP/), RNAInter (http://www.rnainter.org/), and miRmap (https://mirmap.ezlab.org/).

Our study focused on 80 early-stage HCC patients who underwent comprehensive HCC surgery at Shenzhen Hospital, Southern Medical University, from December 2017 to December 2022 (HCC group). The postoperative pathological staging adhered to the criteria set by the Korean HCC Study Group (KLCSG) and the National Cancer Center (NCI) [28].

The inclusion criteria were (1) patients who had undergone radical HCC surgery at Shenzhen Hospital, Southern Medical University, with post-surgical pathological confirmation of HCC; (2) no preoperative anti-cancer treatments such as immunotherapy; (3) complete medical documentation for each patient; (4) no significant diseases or dysfunctions in major organs, and normal results in routine blood, urine, feces, liver function, kidney function, and cardiac function tests.

The exclusion criteria included: (1) Patients with non-alcoholic fatty liver-related HCC; (2) patients with secondary HCC; (3) individuals with other malignant tumors; (4) patients who had received radiotherapy prior to their radical HCC surgery; (5) patients recently treated with molecularly targeted therapies.

The preoperative Barcelona Clinic Liver Cancer (BCLC) staging for the patients is detailed in Supplementary Table 1. Patient follow-ups commenced in January 2022. Additionally, serum samples from 120 healthy individuals, randomly selected from those who underwent health check-ups at Shenzhen Hospital, Southern Medical University in the same period and showed normal results, were used as control samples. HCC tissues and corresponding paracancerous tissue wax blocks from enrolled patients were preserved for future analysis. All participants provided written informed consent, and the study protocol was approved by the Ethics Committee of Shenzhen Hospital, Southern Medical University.

In our study, venous blood was collected from the HCC and the control groups. Following collection, the blood samples were allowed to clot naturally at 37 °C, then refrigerated at 4 °C overnight to ensure proper clot formation. Subsequently, the samples were centrifuged at 3000 rpm for 10 minutes at 4 °C to separate the serum, discarding any insoluble material. Then, the clear serum was carefully transferred to sterile tubes, aliquoted into smaller volumes, and stored in an ultra-low temperature freezer for future use. During the entire process of serum collection and storage, efforts were made to minimize the exposure time of the specimens to ambient conditions to prevent RNA degradation, which could potentially impact the accuracy of the experimental results.

For RNA extraction, 200 µL of serum previously stored in ultra-low temperature conditions was processed using the miRNeasy Mini Kit (217084, Qiagen, Germany). We diluted 5 µL of the RNA sample 20 times in RNase-free water (N003-1-1, Beijing Lingbao Technology Co., Ltd., Beijing, China) and measured its absorbance at 260 nm and 280 nm using a UV spectrophotometer. This step was crucial to ascertain the concentration and purity of the RNA. We considered samples with an OD${}_{260}$/OD${}_{280}$ ratio ranging from 1.7 to 2.1 to be high in purity and suitable for further analysis.

PolyA-tailed cDNA was synthesized using the PolyA Tailing Assay Kit (B532451, Sangon, Shanghai, China) in a PCR thermocycler (Applied Biosystems, Foster City, CA, USA). This process involved the addition of reverse transcription reaction components and a set of universal PCR primers (R), resulting in cDNA that included miRNAs with a polyA tail.

We conducted real-time PCR using the ABI 7500 Real-time PCR System (Applied Biosystems, USA). The 20 µL PCR mixture contained 10 µL of 2$\times$ real-time PCR buffer (A44941, Invitrogen, Foster City, CA, USA), 0.32 µL of specific primer pairs (5 mmol/µL), 0.2 µL of Taq DNA polymerase (5 $\times$ 10${}^6$ U/L), 2 µL of cDNA template, and sufficient RNase-free water to bring the final volume up to 20 µL. The PCR cycling conditions included an initial denaturation at 95 °C for 3 minutes, followed by cycles of denaturation at 94 °C for 30 seconds, annealing at 55 °C for 30 seconds, and extension at 72 °C for 30 seconds, for a total of 30 cycles. Each sample was tested in triplicate. The Ct value (threshold cycle) was determined by selecting the threshold value at the lowest point of the parallel rise in the logarithmic amplification curve. Data analysis was performed using the 2${}^{-\mathrm{\Delta }\mathrm{\Delta }\text{Ct}}$ method, and details of the primer sequences are provided in Supplementary Table 2.

We prepared tissue sections from wax-embedded HCC and adjacent normal tissues using a microtome. These sections were carefully floated on warm water before being mounted onto slides and dried in preparation for staining. The staining process began with dewaxing the sections, followed by staining with hematoxylin. Then, we differentiated the sections using hydrochloric acid ethanol and dehydrated them through a series of graded ethanol solutions before, finally, mounting them using neutral gum. Thereafter, these prepared slides were ready for examination under a light microscope (globalspec, Waltham, MA, USA).

Initially, CD34 immunohistochemical staining was applied to the tissue sections. Subsequently, these sections underwent a series of treatments: Immersion in distilled water for 2–3 minutes, a 10-minute periodate solution soak, and a rinse in distilled water. This was succeeded by PAS staining for 20 minutes, another rinse, and then a 1-minute hematoxylin application for light nuclear staining. After a brief rinse in distilled water, we used hydrochloric acid ethanol for differentiation. Next, the sections were subjected to the same series of steps as H&E staining: Bluing, dehydration, clearing, and mounting. The final step involved observing the stained slides under a light microscope to identify CD34 and PAS signals and count the formation of tube-like structures, as referenced in [29].

H&E stained sections were first observed microscopically to determine VM roughly. VM was surrounded by tumor cells without hemorrhage, necrosis, or inflammatory cell infiltration. The luminal structure of CD34-positive ducts showed endothelium-dependent vasculature. VM was surrounded by plasticized tumor cells rather than endothelial cells and lined with PAS-positive rings. Erythrocytes may or may not be present in the lumen. Determination of endothelium-dependent vessels included looking for PAS-stained vessels under 100 $\times$ light microscopy. CD34-positive microscopy showed brownish-yellow granular staining of the vascular endothelial cell cytosol. The microvessels were counted in a microscopic 100 $\times$ field of view, which covered the entire tissue. The area of concentrated endothelium-dependent vessels was selected randomly, and the sum of vessels in three high-magnification fields was counted as the microvascular density (MVD). Microvessels comprised single or clusters of brown CD34-positive endothelial cells [30].

In our study, IHC was performed on a set of 80 HCC samples. For this, we used anti-ZEB2 antibodies (Solarbio, K106922P, 1:100, Beijing, China) in accordance with a standardized staining protocol. To quantitatively analyze the protein expression indicated by the IHC staining, we utilized an Olympus BX51 image analysis system, complemented by Olympus cellSens Dimension 1.5 Imaging software (Olympus, Tokyo, Japan). Each stained section was scrutinized at a 200$\times$ magnification, and from each sample, 10 representative fields of view were carefully selected and analyzed.

To predict potential complementary binding sites between miR-215-5p and the 3${}^{\prime }$UTR in the ZEB2 gene, we employed the TargetScan online software (https://www.targetscan.org/vert_40/). Then, we designed mutagenic primers for the potential binding site of miR-215-5p using a targeted mutation kit. The construction of vectors involved treating pGEMT-ZEB2 3${}^{\prime }$UTR-WT and pGEMT-ZEB2 3${}^{\prime }$UTR-MUT, along with the pmirGLO vector, with restriction endonucleases Pme I and Sal I in a water bath at 37 °C overnight. Post-digestion, the products underwent purification via agarose gel electrophoresis and gel extraction. The purified pmir-GLO vector was then ligated with the ZEB2 3${}^{\prime }$UTR-WT/MUT fragment in the presence of T4 DNA ligase at 4 °C overnight. These steps led to the amplification of the wild-type and mutant recombinant pmir-GLO vectors, named ZEB2-WT and ZEB2-MUT, respectively. PCR amplification was performed on a small volume of the amplified bacterial broth, followed by identification through 1.5% agarose gel electrophoresis. Positive clones of the recombinant plasmids were selected and subsequently sent to Nanjing Kingsray Biotechnology Co.

For the cellular assays, HEK293T cells (Item No. ACS-4500, ATCC, MA, USA) were seeded in 24-well plates and cultured in DMEM (Gibco, USA) supplemented with 10% fetal bovine serum (16140089, Gibco, USA), 10 µg/mL streptomycin, and 100 U/mL penicillin. All cell lines were validated by STR profiling and tested negative for mycoplasma. The cells were incubated overnight at 37 °C with 5% CO${}_2$ in a humidity-controlled incubator. At 70% confluency, the cells were transfected with either miR-215-5p mimics or Negative Control (NC) miR-215-5p along with the recombinant pmirGLO vector (pmirGLO ZEB2 3${}^{\prime }$UTR-WT or pmirGLO-ZEB2 3${}^{\prime }$UTR-MUT), using Lipofectamine 2000 transfection reagent (11668019, Invitrogen, USA). After 48 hours post-transfection, the cells were lysed and centrifuged at 12,000 g for 1 minute, and the supernatant was collected for analysis. The Dual-Luciferase Reporter Assay System (E1910, Promega, Beijing, China) was utilized to detect luciferase activity. In each assay, 100 µL of firefly luciferase working solution and sea kidney luciferase working solution were added consecutively to measure firefly and Renilla luciferase activities, respectively. The results are expressed as the ratio of firefly luciferase to Renilla luciferase activity.

We acquired HCC cell lines HCCLM3, Huh-7, MHCC97L, and Hep G2, as well as the normal hepatocyte THLE-2 cell line from ATCC, USA. All cell lines were validated by STR profiling and tested negative for mycoplasma. Cells were all cultured in a humidified incubator at 37 °C and 5% CO${}_2$. The HCC cell lines (HCCLM3, Huh-7, MHCC97L, Hep G2) were maintained in DMEM (11960044, Gibco, USA) enriched with 10% fetal bovine serum (Gibco, USA), 100 µg/mL streptomycin, and 2100 U/mL penicillin. The normal hepatocyte line THLE-2 was cultured in RPMI-1640 (11875101, Gibco, USA) supplemented with 10% fetal bovine serum (51023126, Gibco, USA), 100 µg/mL streptomycin, and 100 U/mL penicillin. Cells in the logarithmic growth phase were trypsinized and seeded in 6-well plates at a density of 1 $\times$ 10${}^5$ cells per well for 24 hours. Upon reaching about 75% confluency, the cells underwent transient transfection, following the protocol provided alongside the Lipofectamine 2000 reagent (Invitrogen).

The experimental groups included the NC mimic group, miR-215-5p mimic group, oe-NC group (overexpression of null control plasmid), oe-ZEB2 group (overexpression of ZEB2 plasmid), and the miR-215-5p mimic + ZEB2 group. Transfection plasmids and mimics were sourced from Sino Biological (Beijing, China). Post-transfection (after 6 hours), the medium was changed, and cells were cultured for an additional 48 hours before collection for downstream analyses. The relevant gene expressions in each group were assessed 48 hours post-transfection using qPCR and Western blot techniques, as detailed in references [31, 32].

Protein extraction from the cultured cells was performed for subsequent electrophoresis and Western blotting analysis. The protein concentrations were determined using a Bicinchoninic Acid (BCA) Protein Assay Kit (20201ES76, Yisheng Biotechnology Co., Ltd., Shanghai, China). The proteins were then separated by standard Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE) and transferred onto Polyvinylidene Difluoride (PVDF) membranes. These membranes were blocked at room temperature using 5% Bovine Serum Albumin (BSA) for 1 hour. Subsequently, we added the primary antibodies, ZEB2 (ab214425, 1:1000, Abcam, Cambridge, UK) and Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (ab8245, 1:5000, Abcam, UK) and incubated the membranes overnight at 4 °C with gentle shaking. Post-incubation, the membranes were washed three times with Tris-buffered saline with 0.1% Tween® 20 Detergent (TBS-T), each for 5 minutes. Detection was carried out using an enhanced chemiluminescence assay following another three 5-minute TBS-T washes. For protein quantification, we used ImageJ 1.48v software (National Institutes of Health, Bethesda, MD, USA), employing the ratio of the gray value for each protein to that of the GAPDH internal reference protein.

The plasmids were left to transfect for 48 h before the fused cells were scraped off using a sterile 200 µL pipette tip. Separated cells and debris were removed by PBS washing. The scratch distance was photographed microscopically and set as 0 h. After incubation in a serum-free DMEM medium for 48 h, the same observation position was taken. The scratch distance was photographed again and compared with the 0 h cell scratch distance to observe the migration distance.

For the invasion assay, we coated the upper surface of the Transwell membranes with BD Matrigel (356234, Bedford, MA, USA) and allowed it to gel at 37 °C for 30 minutes. Prior to use, the basement membrane was hydrated. Cells were serum-starved for 12 hours to facilitate the assay. Afterward, these cells were collected and resuspended in a serum-free medium at a concentration of 1 $\times$ 10${}^5$ cells/mL. The lower chamber of the Transwell apparatus was filled with a medium containing 10% fetal bovine serum. We then added 100 µL of the cell suspension to the upper chamber.

Following a 24-hour incubation at 37 °C, cells that did not invade through the Matrigel membrane were carefully removed with cotton swabs. The cells that penetrated the membrane were fixed with 100% methanol and stained with 1% toluidine blue (89640, Sigma, Kawasaki, Japan). We then counted the stained, invading cells under an inverted light microscope, selecting five random fields for analysis, as detailed in reference [33].

Post-transfection (24 hours), tumor cells were incorporated into a Matrigel matrix (356234, Bedford, MA, USA) to induce polymerization. During a 10-day incubation period in the 3D culture, we employed both pretreatment and continuous treatment protocols using conditioned media supplemented with 10% fetal bovine serum (Hyclone). For RNA and protein extraction, cells embedded in Matrigel were retrieved using trypsin digestion. Following this, we used Trizol and RIPA buffer to isolate total RNA and protein from the cells, respectively, as outlined in reference [34].

For our data analysis, we utilized SPSS statistical software. Quantitative data were presented as mean $\pm$ standard deviation, and differences between the two groups were assessed using the independent samples t-test. Qualitative data, expressed in percentages or rates, were analyzed using the chi-square test. We employed two-way stepwise regression and the Akaike information criterion (AIC) for variable selection. Logistic regression models were used to investigate the association between VM expression and the risk of vascular invasion and recurrent metastasis in HCC tissues [4]. Additionally, the predictive accuracy of these models was evaluated using receiver operating characteristic (ROC) curves.

To identify miRNAs potentially linked to vascular invasion in HCC, we analyzed miRNA expression data from the GEO database, specifically datasets GSE6857 and GSE36915. In the GSE6857 dataset, we found 81 miRNAs exhibiting differential expression in HCC cases, with 49 miRNAs being upregulated and 32 downregulated (as shown in Fig. 1A and detailed in Supplementary Table 3). Similarly, analysis of the GSE36915 dataset revealed 386 differentially expressed miRNAs in HCC, comprising 160 upregulated and 226 downregulated miRNAs (Fig. 1B, Supplementary Table 4).

Fig. 1.

Bioinformatics analysis to screen miRNAs associated with vascular invasion in hepatocellular carcinoma (HCC). Note: (A) Volcano plot of differentially expressed miRNAs between HCC samples (n = 241) and normal liver tissue samples (n = 241) in GSE6857. (B) Volcano plot of differentially expressed miRNAs between HCC samples (n = 68) and normal liver tissue samples (n = 21) in GSE36915. (C) Volcano plot of differentially expressed miRNAs between HCC samples without vascular invasion (n = 91) and HCC samples with vascular invasion (n = 81) in GSE67140. Red dots indicate high expression, and green dots indicate low expression. (D) Overlapping differentially expressed miRNAs in the three datasets.

Further analysis was conducted on the HCC vascular invasion-associated miRNA dataset, GSE67140. This revealed that, in HCC samples exhibiting vascular invasion, 156 miRNAs were upregulated, while 126 miRNAs were downregulated compared to HCC samples without vascular invasion (Fig. 1C, Supplementary Table 5). By intersecting data from these three datasets, we identified eight potential candidate miRNAs: hsa-miR-335, hsa-miR-340, hsa-miR-215-5p, hsa-let-7c, hsa-miR-203, hsa-miR-214, hsa-miR-224, and hsa-miR-181c (Fig. 1D, Supplementary Table 6).

In our quest to pinpoint miRNAs that can predict patient outcomes for HCC, we investigated the link between the eight identified miRNAs and overall patient survival, utilizing data from the TCGA database. Kaplan–Meier survival analysis revealed a notable correlation: HCC patients with elevated levels of hsa-miR-215-5p demonstrated significantly improved overall survival compared to those with lower expression levels. In contrast, the remaining seven miRNAs (hsa-miR-335, hsa-miR-340, hsa-let-7c, hsa-miR-203, hsa-miR-214, hsa-miR-224, and hsa-miR-181c) showed no substantial impact on the overall survival of HCC patients (as depicted in Fig. 2). Consequently, miR-215-5p was chosen for further investigation.

Fig. 2.

Kaplan–Meier curve analysis of the correlation between eight miRNAs and the overall survival prognosis of HCC patients. Note: analysis of hsa-miR-215-5p (A), hsa-miR-335 (B), hsa-miR-340 (C), hsa-let-7c (D), hsa-miR-203 (E), hsa-miR-214 (F), hsa-miR-224 (G) and hsa-miR-181c (H) correlated with the overall survival prognosis of HCC patients.

Furthermore, as depicted in Fig. 3, miR-215-5p was markedly downregulated in HCC tissues compared to normal tissues in TCGA, GSE6857, and GSE36915 datasets. Specifically, in the GSE67140 dataset, miR-215-5p expression was significantly lower in HCC samples with vascular invasion than in those without vascular invasion.

Fig. 3.

miR-215-5p expression in HCC-related datasets in The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. Note: expression levels of miR-215-5p in TCGA datasets (A), GSE6857 (B), GSE36915 (C), and GSE67140 (D). vi, vascular invasion.

To elucidate the mechanism through which miR-215-5p regulates vascular invasion in HCC, we utilized a range of bioinformatics resources, including miRDB, mirDIP, RNAInter, and miRmap, to predict the miR-215-5p target genes. Our analysis identified six potential target genes that were common across these databases (as shown in Fig. 4A and listed in Supplementary Table 7). Additionally, we conducted a differential gene expression analysis on the GSE33006 dataset with the criteria of $|$logFC$|$ $>$1 and p 0.05. This analysis revealed 2806 genes with differential expressions in HCC, including 1526 genes that were upregulated (Fig. 4B, Supplementary Table 8). By intersecting these upregulated genes with the predicted targets of miR-215-5p, we pinpointed ZEB2 as a gene of interest (Fig. 4C). ZEB2 a factor known for regulating EMT, plays a crucial role in tumor invasion and metastasis. Moreover, existing studies suggest that ZEB2 could be instrumental in promoting angiogenesis in HCC [35]. Therefore, we hypothesize that miR-215-5p may affect vascular invasion in HCC by targeting ZEB2.

Fig. 4.

Prediction of miR-215-5p target genes in HCC. Note: (A) The miRDB, mirDIP, RNAInter, and miRmap databases were used to predict the miR-215-5p target genes. (B) The Volcano plot shows the differentially expressed genes between HCC and normal samples in GSE33006. (C) The overlap between the miR-215-5p target genes and the upregulated genes in HCC.

Our study revealed no significant differences in gender, age, smoking habits, or alcohol consumption between the HCC group and the control group, with all p-values exceeding 0.05. To assess the expression levels of miR-215-5p and ZEB2, PCR analysis was conducted on serum samples from both HCC patients and healthy individuals. The results indicated a notable downregulation of miR-215-5p and an upregulation of ZEB2 in the serum of HCC patients compared to the control group. The variations in the expression levels of both miR-215-5p and ZEB2 were statistically significant, with all p-values falling below 0.05, as detailed in Supplementary Table 9.

H&E staining revealed distinct cellular changes in HCC tissues compared to adjacent normal liver tissue. Notably, HCC cells were polygonal, densely packed, and disorganized, disrupting the hepatic lobular structure. Infiltration of inflammatory cells and tumor cells forming tubular structures containing erythrocytes were also observed (Fig. 5A). Moreover, CD34/PAS double staining indicated VM presence in 38 (47.5%) out of 80 HCC samples, while 42 samples were VM negative (Fig. 5B). A significant increase in VM formation was noted in VM-positive tissues compared to VM-negative ones.

Fig. 5.

Validation of miR-215-5p targeting relationship with zinc finger E-box binding homeobox 2 (ZEB2) in HCC and its correlation with vasculogenic mimicry (VM) expression. Note: (A) Hematoxylin and eosin (H&E) staining map of HCC tissues and normal tissues next to cancer ($\times$200). (B) CD34/PAS double staining and quantification of VM in HCC tissues—black arrows indicate CD34 (+) and red arrows PAS (+). (C) RT-PCR detection of miR-215-5p and ZEB2 in different tissues expression level changes in different tissues. (D) Western blot detection of ZEB2 protein expression level changes in different tissues. (E) Immunohistochemistry (IHC) to detect ZEB2 positive expression levels in different tissues. (F) RNA22 to predict the binding site of ZEB2 and miR-215-5p. (G) Dual luciferase reporter assay to detect whether miR-215-5p could bind ZEB2, In the connected group data, *** indicates p 0.001. GAPDH, Glyceraldehyde 3-Phosphate Dehydrogenase.

RT-PCR assays demonstrated that miR-215-5p expression was significantly lower in both VM-positive and VM-negative HCC samples than in normal tissue, with the levels even lower in VM-positive samples. Conversely, ZEB2 expression levels were elevated in both VM groups, particularly in VM-positive samples (all p 0.05, Fig. 5C). Western blot and immunohistochemistry (IHC) results further confirmed that ZEB2 expression was significantly higher in HCC tissues, especially in the VM-positive group (Fig. 5D,E).

To explore the interaction between miR-215-5p and ZEB2, the TargetScan database was used to identify the potential binding site between them (Fig. 5F). Dual-luciferase reporter assays indicated a significant reduction in luciferase activity of ZEB2-WT with miR-215-5p overexpression, whereas the mutant group showed no change. This suggests a direct interaction between miR-215-5p and ZEB2 (Fig. 5G).

Research has demonstrated that MVD is a primary indicator of tumor neovascularization, while VM significantly contributes to tumor blood supply [36]. VM is closely related to higher tumor grade and stage, and MVD is associated with increased tumor size, grade, and stage [37]. Moreover, a notable link exists between tumor cell proliferation, VM formation, and MVD [38].

In our study, as detailed in Supplementary Table 10, notable differences were observed between VM-positive and VM-negative HCC groups in terms of tumor size, differentiation, Tumor, Node, Metastasis (TNM) stage, cirrhosis level, MVD, preoperative Alpha-Fetoprotein (AFP), BCLC stage, and expressions of miR-215-5p and ZEB2 (all p 0.05). VM positivity was significantly associated with vascular invasion and recurrent metastasis in HCC (all p 0.05).

Logistic regression analysis evaluated the relationship between VM expression, vascular invasion, and recurrent metastasis as dependent variables, and factors, such as age, gender, smoking and drinking history, miR-215-5p and ZEB2 expressions, tumor size, differentiation, TNM stage, cirrhosis, MVD, and preoperative AFP levels as independent variables.

In analyzing the VM-positive rate, miR-215-5p expression, ZEB2 expression, and tumor size were significant (p 0.05). Specifically, miR-215-5p expression was inversely correlated with VM positivity, while ZEB2 expression showed a positive correlation (Supplementary Table 11). In assessing factors linked to vascular invasion in HCC, ZEB2 expression and tumor size were significant (p 0.05), with a notable positive correlation found for ZEB2 expression (Supplementary Table 12). For factors associated with recurrence and metastasis in HCC, significant variables included miR-215-5p expression, tumor size, and preoperative AFP levels (p 0.05), with miR-215-5p expression negatively correlated with recurrent metastasis (Supplementary Table 13).

ROC curve analysis was conducted for two final models to predict the likelihood of vascular invasion and recurrent metastasis. As illustrated in Fig. 6A and Supplementary Table 14, the model predicting vascular invasion, which included gender and ZEB2 expression, showed an area under the curve (AUC) of 0.940 (95% CI: 0.886–0.993) for tumor size and 0.900 (95% CI: 0.824–0.977) for TNM stage. The AUC for the model incorporating all predictor variables was 0.959 (95% CI: 0.915–1.000).

Fig. 6.

Predictive accuracy of receiver operating characteristic (ROC) curve evaluation model for HCC vascular invasion (A) and recurrent metastasis (B). Note: In the figure, “tumor size” indicates the model that includes tumor size, “TNM” indicates the model that includes the TNM stage, and “all” indicates the model that includes all predictive variables at the same time. The horizontal coordinates are the false positive rate, and the vertical coordinates are the true positive rate. In the area under the curve (AUC) comparison by the Delong test.

In the model predicting recurrent metastasis, including alcohol consumption history, miR-215-5p expression, ZEB2 expression, and TNM stage (Fig. 6B and Supplementary Table 15), the AUC was 0.941 (95% CI: 0.889–0.994) for tumor size, 0.888 (95% CI: 0.807–0.968) for cirrhosis, and 0.772 (95% CI: 0.664–0.881) for preoperative AFP. The AUC for the model with all predictor variables was 0.965 (95% CI: 0.930–1.000). These results underscore the strong association between VM positivity and the incidence of vascular invasion and recurrent metastasis in HCC.

Several studies have indicated that the formation of VM in tumor cells mirrors the migration and invasion processes of endothelial cells [35, 39, 40]. To explore the impact of miR-215-5p on the invasive and proliferative capabilities of HCC cells, we first employed RT-PCR to assess the expression levels of miR-215-5p and ZEB2 in the normal hepatocyte line THLE-2 and four HCC cell lines (HCCLM3, Huh-7, MHCC97L, and Hep G2). Our findings revealed a significant reduction in miR-215-5p and an increase in ZEB2 expression across all HCC cell lines compared to THLE-2, with Hep G2 showing the most pronounced changes; thus, it formed the primary focus for further functional assays (Fig. 7A).

Fig. 7.

Effect of miR-215-5p upregulation or ZEB2 overexpression on invasion, migration, and VM formation abilities by HCC cells in vitro. Note: (A) RT-qPCR was used to detect the expression of miR-215-5p and ZEB2 in human HCC cell lines (HCCLM3, Huh-7, MHCC97L, Hep G2) and human normal liver cell lines (THLE-2). (B) RT-qPCR was used to detect the mRNA expression levels of miR-215-5p and ZEB2 in each group of cells. (C) Western blot was used to detect the level of ZEB2 protein expression in cells in each group. (D) The cell migration ability of each group was detected by scratch test. (E) Transwell invasion test was used to detect the invasion ability of cells in each group. (F) In the in vitro vascular simulation test, the tubular structure diagram of each group was obtained after 48 hours and 7 days of 3D culturing. The experiment was repeated three times.Compared with the Negative Control (NC) mimic group, ** indicates p 0.01, and *** indicates p 0.001. Compared with the oe-NC group, ${}^{\mathrm{\#}\mathrm{\#}}$ indicates p 0.01, and ${}^{\mathrm{\#}\mathrm{\#}\mathrm{\#}}$ indicates p 0.001. Compared with the oe-ZEB2 group, indicates p 0.01, and indicates p 0.001.

Subsequently, Hep G2 cells were transfected with different plasmids, resulting in groups: NC mimic, miR-215-5p mimic, oe-NC, oe-ZEB2, and miR-215-5p mimic+oe-ZEB2. Post-transfection, RT-PCR analysis of miR-215-5p and ZEB2 expression (Fig. 7B) indicated a significant increase in miR-215-5p and a decrease in ZEB2 expression in the miR-215-5p-mimic group compared to Negative Control (NC) mimic; in the oe-ZEB2 group, ZEB2 expression was significantly higher than in the oe-NC group. The miR-215-5p mimic+oe-ZEB2 group showed a marked increase in miR-215-5p and a decrease in ZEB2 expression compared to the oe-ZEB2 group. Western blot analysis of ZEB2 protein levels corroborated these PCR findings (Fig. 7C), confirming successful plasmid transfection.

We employed scratch and Transwell invasion assays to evaluate cell migration and invasion, respectively (Fig. 7D,E). Compared to the NC mimic group, the miR-215-5p mimic cells exhibited significantly reduced migration and invasion capabilities; the oe-ZEB2 group showed enhanced migration and invasion compared to the oe-NC group; the miR-215-5p mimic+oe-ZEB2 group displayed reduced migration and invasion compared to the oe-ZEB2 group.

Additionally, we utilized a Matrigel 3D culture, a standard in vitro model, to assess VM formation (Fig. 7F). The results demonstrated that HepG2 cells in the miR-215-5p upregulated group did not form typical tubular structures on the 3D Matrigel, whereas cells with upregulated ZEB2 did. In cells with simultaneous overexpression of miR-215-5p and ZEB2, the ability to form tubular structures was diminished. These findings suggest that overexpression of miR-215-5p inhibits ZEB2, subsequently reducing the capacity of HCC cells to invade, migrate, and form VMs.