ScRNA-seq data of HCC patients were downloaded from NCBI with the search number GSE242889, including three patients with MVI present and two patients with MVI absent. None of the patients had significant metastasis, and they had not yet been treated. The downloaded scRNA-seq data were loaded into the “Seurat” package (Butler et al., 2018) and read using the Read10X function. Only cells with the proportion of mitochondrial genes <25% and the number of genes between 200 and 5000 were intercepted and normalized using the SCTransform function. The “Harmony” package (Korsunsky et al., 2019) was used for the elimination of batch effects between samples. The RunPCA function was used to perform principal component analysis (PCA) in Seurat, selecting the top 20 principal components, and the FindNeighbors and FindClusters functions were used to conduct cell clustering. According to the markers in the CellMarker2.0 database, the cell identity was identified and projected on UMAP. The cell identity was identified according to the markers contained in the CellMarker2.0 database (Hu et al., 2023) and projected on the UMAP.

The FindAllMarkers function was used to perform differential expression analysis between determined cell clusters in Seurat. For differentially expressed genes (DEGs), it must be in accordance with only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25, and p < 0.05 (Xu et al., 2024).

Specific genes were uploaded to the DAVID database (Dennis et al., 2003; Wang et al., 2025), which can correlate genes from the input list to biological annotations. The Functional Annotation Chart was selected in Shortcut to DAVID Tools to perform GO functional enrichment analysis, and the annotated results were presented in the form of bar charts.

The CellChat package (Jin et al., 2021) is a versatile and easy-to-use toolkit for inferring, analyzing, and visualizing intercellular communication from any given scRNA-seq data. The CreateCellChat function was applied to create objects in CellChat, while identifyOverExpressed Genes and identifyOverExpressedInteractions were employed to identify overexpressed ligand–receptor (LR) pairs of cell subsets. The projectData function projected gene expression data onto the protein–protein interaction (PPI) network. The communication probability was calculated by the computeCommunProb function and displayed as a bubble plot by the netVisual_bubble function.

Monocle 2 (Qiu et al., 2017) inferred the evolution of cells during development by conducting pseudotime analysis based on the changes in gene expression levels of cell subsets over time. In brief, the newCellDataSet function was used to construct the objects, and the highly expressed DEGs in the cell cluster were calculated by the FindAllMarkers function and submitted to reduceDimension for dimensionality reduction. Monocle 2 generated a tree-like structure after completing pseudotime modeling.

The gene lists of angiogenesis and TGFβ signaling pathways in hallmark were downloaded from the MsigDB database and entered into AUCell together with the single-cell expression matrix to calculate the enrichment scores of gene subsets in each cell, with higher scores indicating higher enrichment of gene subsets in the cell.

The TCGA-HCC cohort was selected from the public cancer genomics resource UCSC Xena to download the gene expression data of HCC genomes, and the gene expression value log2(FPKM+1) was obtained. The cohort also provided clinical follow-up data of HCC, including TNM stage and survival data, which were also included in the analysis.

Human liver immortalized cells THLE-2 (C5664) and HCC cell lines Huh-7 and HepG2 were all purchased from Baidi Biotech Ltd. (Hangzhou, China) and BeiNa Culture Bio (Xinyang, China) and cultured in Roswell Park Memorial Institute-1640 medium (90023, Solarbio LifeSciences, Beijing, China) and high-glucose Dulbecco’s modified Eagle’s Medium (C3260, Solarbio LifeSciences). The culture media were additionally supplemented with 10% bovine calf serum (S9020, Solarbio LifeSciences) and 1% penicillin–streptomycin (P1400, Solarbio LifeSciences). All cells were authenticated via short tandem repeat (STR) genotyping recently, tested negative for mycoplasma contamination, and cultured in an incubator (3311, ThermoFisher, Waltham, MA) at 37°C with 5% CO₂.

For the knockdown assay, the small interfering RNA specific to CAMK2N1 and the control small interfering RNA were all synthesized and purchased from GenePharma (Shanghai, China). Thereafter, the Lipofectamine 2000 transfection reagent (11668027, ThermoFisher) was applied for the transfection as per the manuals provided by the manufacturer. All cells were harvested after 48 h and used in the subsequent assays. The target sequences for small interfering RNA specific to CAMK2N1 are 5′-CGG​GTT​GTT​ATT​GAA​GAT​GAT​AG-3′ (si-CAMK2N1#1) and 5′-GGG​TTG​TTA​TTG​AAG​ATG​ATA​GG-3′ (si-CAMK2N1#2).

The scratch test was adopted to evaluate the migration of HCC cells. In detail, transfected HCC cells in these two groups were seeded in a 6-well plate at a density of 1 × 10⁵ cells/well and received a scratch on the monolayer via a 10-μL pipette tip once they grew fully confluent. After culturing for 48 h, the cell debris was removed, and the cell residue was visualized under the optical microscope (DP27, Olympus, Tokyo, Japan). The wound closure (%) was accordingly calculated to reflect the migration status of cells in both groups.

For the Transwell assay to determine the invasion of HCC cells, a Transwell system with a pore size of 8 μm (3422, Corning, Inc., Corning, NY) was applied. Specifically, the upper chamber was filled with HCC cells and the non-serum culture media (final volume: 200 μL) and coated with matrix gel (M8370, Solarbio LifeSciences) in advance. Thereafter, HCC cells were routinely cultured for 48 h, following which the uninvaded cells in the upper chamber were removed carefully with a cotton swab. Those cells that migrated to the lower chamber were accordingly fixed in 4% fixative (P1110, Solarbio LifeSciences) for 15 min and stained using 0.1% crystal violet (G1063, Solarbio Lifesciences) for 15 min. All cells were also visualized under the same optical microscope used for the scratch test (Zhang et al., 2024).

All cellular total RNA was extracted with TRIzol reagent (15596026, ThermoFisher), and the concentration was determined in a spectrophotometer (ND-2000, ThermoFisher). Subsequently, the complementary DNA was synthesized with a relevant assay kit (6110A, Takara, Shiga, Japan) and applied to the PCR process. A specific assay kit (204243, Qiagen, Hilden, Germany) and the ABI7500 thermocycling system (ThermoFisher) were both used to run the PCR at these parameters: 95°C for 15 min, followed by 94°C for 15 s, 60°C for 30 s, and 72°C for 30 s for 40 repeated cycles. The mRNA levels were finally calculated based on formula 2^−ΔΔc(t) with GAPDH as the reference gene (Livak and Schmittgen, 2001).

The primers applied are:

CAMK2N1: forward: 5′-GGA​CAC​CAA​CAA​CTT​CTT​CGG​C-3′; reverse: 5′-GTC​GGT​CAT​ATT​TTT​CAG​CAC​GTC-3′

GAPDH: forward: 5′-GTC​TCC​TCT​GAC​TTC​AAC​AGC​G-3′; reverse: 5′-ACC​ACC​CTG​TTG​CTG​TAG​CCA​A-3′

All statistical analysis and visualization were performed using R software (version 4.3.1) and GraphPad Prism software (version 8.0.2). The difference of continuous variables between the two groups was tested by the Wilcoxon rank sum test, analysis of variance and Student’s t-test, and the correlation was measured by Pearson correlation analysis. Grouping before survival analysis was based on the median of gene expression values, and survival differences were calculated by the log-rank test. The threshold of statistical significance was set at p < 0.05 for statistical tests without a specified p-value.

Cluster analysis and annotation of 9,828 cells in HCC patients with MVI absent and 13,402 cells in HCC patients with MVI present identified eight major cell types: they were myofibroblasts (MFs), hepatocytes, proliferative hepatocytes, endothelial cells, dendritic cells, proliferative NK/T cells, plasma B cells, and macrophages (Figures 1A,B). Apolipoprotein APOE, transporter TTR, and the acute-phase proteins fibrinogen β (FGB) and haptoglobin (HP) were significantly overexpressed in both hepatocytes and proliferative hepatocytes (Dayoub et al., 2017). However, proliferative hepatocytes specifically expressed the proliferation-related markers MKI and TOP2A relative to proliferative hepatocytes. CD163, AIF1, and C1QAM2 were significantly overexpressed in macrophages, all of which were related markers of M2 macrophages (Zhu et al., 2022). The high expression of NKG7 and CD3D specifically distinguishes proliferative NK/T cells from other cells. The DEGs for identifying plasma B cells were JCHAIN and MZB1 (Hensvold et al., 2023). The identity of MFs was confirmed based on the expression levels of BGN, COL3A1, TAGLN, and ACTA2. Dendritic cells specifically expressed CLEC9A and WDFY41, which are essential components for antigen cross-presentation in these cells (Ohara and Murphy, 2023) (Figures 1C,E). The largest difference in the proportion between MVI-absent patients and MVI-present patients was MFs, which accounted for a higher proportion in MVI-present patients (Figure 1D). This abnormal distribution of cells sent a message that MFs may play a role in the MVI of patients with HCC.

Because of the large difference in the proportion of MFs between HCC patients with MVI absent and those with MVI present, we proceeded to characterize this population in detail, focusing on the subgroup and gene expression profiles, intercellular communication networks, and differentiation trajectories of this population. The MFs were further divided into five clusters by cell clustering (Figure 2A). The proportions of MF2 and MF5 cells were significantly higher in HCC patients with MVI present than in those with MVI absent (Figure 2B). MF2 cells specifically expressed liver fibrosis-related genes AGTR1, CYGB, PIEZO2, and LPL, and MF5 cells specifically highly expressed immune-related genes TRAC, CD3D, GZMB, and FCGR3A (Figure 2C). Among the annotated biological processes, MF2 cells were linked with signal transduction, positive regulation of cell proliferation, extracellular matrix organization, angiogenesis, and cell migration (Figure 2D). Meanwhile, MF5 cells could be linked with adaptive immune response, innate immune response, and inflammatory response (Figure 2E). Therefore, MF2 cells were closely related to the promotion of angiogenesis, while MF5 cells were mainly related to the recruitment and activation of immune cells. Angiogenesis is one of the important mechanisms in the formation of MVIs. Most of the angiogenesis key genes were overexpressed in MF2 cells and underexpressed in MF5 cells (Figure 2F).

By analyzing the LR pairs of communication between MF2 and hepatocytes in HCC patients with MVI absent and MVI present, it was found that direct contact LR pairs, including OCLN-OCLN, JAG1-NOTCH3, CDH5-CDH5, CD99⁻CD99, and APP-CD74, were involved in mediating the communication between MF2 cells and hepatocytes in both HCC patients with MVI absent and MVI present. The difference was that the communication between MF2 cells and hepatocytes in MVI-present patients was also mediated by THY1 (ITGAX + ITGB2), THY1- (ITGAM + ITGB2), JAG1-NOTCH1, ITGB2-ICAM1, EFNA5-EPHA3, EFNA5-EPHA2, and EFNA5-EPHA1 (Figures 3A,B). In addition to the direct contact type LR pairs, secretory LR pairs were also involved in the communication between MF2 cells and hepatocytes and were more complex in MVI-present patients. TNFSF12-TNFRSF12A, TGFB2-(TGFBR1+TGFBR2), TGFB2-(ACVR1B + TGFBR2), TGFB2-(ACVR1+TGFBR1), PSAP-GPR37, MIF- (CD74+CXCR4), MIF-(CD74+CXCR2), MIF-(CD74⁺CD44), CXCL12-CXCR4, and BMP4-(BMPR1B + BMPR2) were secreted LR pairs specifically involved in MF2 cells communication with hepatocytes in HCC patients with MVI present (Figures 3C,D).

To characterize the dynamic changes in gene expression patterns of MF2 cells between MVI-absent and MVI-present samples, differentiation trajectories were constructed using Monocle 2. Most of the MF2 cells at the start of pseudotime were of MVI-absent origin, and MF2 cells of MVI-present origin appeared in pseudotime and were densely clustered at the end of the branch (Figure 4A). Monocle 2 also identified genes that varied with pseudotime (Figure 4B). The correlation between angiogenesis gene sets and TGF-β pathway gene sets was probed, and both angiogenesis and TGF-β pathway showed a significant positive correlation with pseudotime (Figures 4C,D). Therefore, it is speculated that MF2 cells may have a potential promoting effect on MVI formation.

The expression of marker genes extracted from MF2 cells was detected in different M stages of the TCGA cohort. The expressions of CAMK2N1 and COLEC11 were significantly higher in M1 stage samples than in M0 stage patients (Figures 5A,B). The prognosis of patients in the TCGA cohort was predicted based on CAMK2N1 and COLEC11 expression. CAMK2N1 expression showed a significant correlation with HCC prognosis as observed in the Kaplan–Meier curve. The overall survival (OS) of HCC patients with high expression was significantly shorter than that of patients with low expression (Figure 5C). However, no significant correlation between COLEC11 expression and HCC prognosis was detected in the Kaplan–Meier curve obtained by analyzing OS based on COLEC11 expression (Figure 5D). Therefore, CAMK2N1 is a potential MF2 cell marker for predicting HCC prognosis.

To explore the potential role of this CAMK2N1 in HCC, the expression levels of these metrics were first calculated in HCC cells and control cells. We observed lower CAMK2N1 expression in Huh-7 and HepG2 cells relative to THLE-2 cells (Figure 6A). In addition, to verify the potential effect of this gene on HCC cells, we validated its knockdown in two HCC cell lines (Figures 6B,C). The results of the CCK-8 assay showed that knockdown of the CAMK2N1 gene significantly increased the proliferation level of Huh-7 and HepG2 cells (Figures 6D,E). Furthermore, CAMK2N1 gene knockdown also significantly increased the migration and invasion levels of HCC cells (Figures 6F–I). These results indicated the scientific rationality of CAMK2N1, an MF2 cell marker gene significantly associated with HCC prognosis, as a potential prognostic biomarker.