An integrated analysis was conducted on the hepatocellular carcinoma single-cell RNA sequencing dataset GSE242889. For quality control, cells were excluded if they met any of the following criteria: detection of fewer than 200 genes, detection of more than 5,000 genes (indicative of potential doublets), or mitochondrial gene content exceeding 10% of total counts. The Seurat package was employed for data normalization, dimensionality reduction, and clustering. Data normalization, dimensionality reduction, and clustering were performed using the Seurat package. To address batch effects, correction was applied to the top 2,000 most variable genes with the Harmony package under default parameters. Cell clustering and dimensionality reduction were carried out using the ‘FindClusters’ and ‘RunUMAP’ functions, respectively, with the resolution parameter optimized to 0.3 through systematic evaluation. Cell type annotation was achieved by manual mapping against liver-specific marker genes obtained from the ACT database (http://xteam.xbio.top/ACT/index.jsp) (29, 30), and supplemented by automated scoring via the UCell algorithm based on cell-type signature gene sets (31). Interactions among cell types were analyzed with CellChat, focusing on receptor-ligand interactions as defined in the database, using default parameters. Cell cycle scoring was performed with the CellCycleScoring function in Seurat, which calculates S-phase and G2/M-phase scores from predefined gene sets and assigns each cell to G1, S, or G2M phase based on phase-specific marker expression. The resulting scores and phase classifications were stored in the metadata slot for subsequent analyses.

The expression data, represented as HTseq-Counts, were categorized into high and low expression cohorts based on the median expression level of SRGN. Subsequently, these groups underwent further analysis utilizing the unpaired Student’s t-test within the DESeq2 R package (version 1.36.0). A significance threshold was established, with adjusted p-values <0.05 and |log2-fold change (FC)| >1 deemed as indicators for DEGs. Gene Set Enrichment Analysis (GSEA) was performed using the clusterProfiler package to further explore the biological functions and pathways associated with the DEGs (32, 33). Before GSEA, gene ID conversion was conducted to ensure the compatibility of gene identifiers with the gene sets in the MSigDB collections database.

Immune microenvironment characterization was performed using single-sample gene set enrichment analysis (ssGSEA) implemented via the GSVA package (34), utilizing 24 gene sets corresponding to immune cell signatures (35). These gene sets enabled the calculation of immune infiltration levels for each sample. Gene-immunocyte correlations were assessed using Spearman’s rank correlation, and the results were visualized as lollipop plots using the ggplot2 package. Additionally, we referred to previous studies (31, 32) to identify specific immune checkpoint genes. The correlation between SRGN and these immune checkpoint genes was analyzed using Spearman’s rank correlation, and the results were visualized using a rose diagram (36).

The differential expression and prognosis of SRGN in pan-cancer were studied using Gene Expression Profi;ling Interactive Analysis (GEPIA, http://gepia.cancer-pku.cn/index.html) (37), Kaplan–Meier plotter databases (http://kmplot.com/analysis/(registration-freeKM-plotter)) (38), Tumor Immune Estimation Resource (TIMER, https://cistrome.shinyapps.io/timer/) (39) Clinical Proteomic Tumor Analysis Consortium (CPTAC, https://proteomics.cancer.gov/programs/cptac) (40) or Human Protein Atlas (HPA, https://www.proteinatlas.org/) (41). The correlations between SRGN and TME cells or gene markers were studied by Spearman’s correlation analysis via TIMER which was determined using the following criteria: 0.1–0.3, weak correlation, 0.3–0.5, moderate correlation, and 0.5–1.0, strong correlation (42).

SRGN sequences were designed following the gene (Homo sapiens (human) Gene, variant 3, NM, NCBI132 Reference 105 Sequence 3.2, mRNA, NCBI132 ID: 5552). SRGN XhoI F: 5’ccgctcgag ccaccATGATGCAGAAGCTACTCAAATGCAGTC3’, BamHI R: 5’cgcggatccTTATAACATAAAATCC TCTTCTAATCCATG 3’. RNA was extracted by Trizol, amplificated by PCR. SRGN and pLVX- IRES-Neo vector 15μL each was digested with XhoI/BamHI, purified and ligated. The product was transformed to DH5α competent cells. The positive clones were cloned into pLVX-IRES-Neo vector. The shuttle plasmid containing the target sequence along with the packaging plasmids pVSV-G, pRev, and pGag/Pol were constructed and prepared by Guangzhou Yeshan Biotechnology Co., Ltd., and 293T cell was co-transfected with the transfection reagent LipofectamineTM 2000. After 72 hours of culture, the lentiviral particles were collected and infected HepG2 cells with 400μg/mL G418 to screen for one month. HepG2 cells stably overexpressing SRGN or a blank pLVX-IRES-Neo vector were constructed, and named HepG2SG and HepG2-NC, respectively.

The primers were as follows: SRGN 150 bp, F: CTGCAAACTGCCTTGAAGAA, R: GTGGGAA GATACGATTCAAGTC; β-actin 275 bp, F: TGGATCAGCAAGCAGGAGTA, R: TCGGCCACATT GTGAACTTT. qPCR assays were performed in HepG2-NC group and HepG2SG. Following the manufacturer’s instructions, RNA was extracted with Trizol (Invitrogen Corp, Carlsbad, CA, USA), and cDNA was synthesized with a first-strand cDNA synthesis kit (Takara Inc., Dalian, P. R. China). SYBR Green qPCR SuperMix (Invitrogen Corp) was used for qPCR (ABI, PRISM^® 7500 Sequence Detection System). The reaction conditions were 50°C 2 min, 95°C 2 min, 95°C 15 s, 60°C 32 s, 40 cycles, melting curve analysis at 60°C–95°C. The independent experiment was repeated in triplicate, similarly for the following methods. Relative mRNA expression was analyzed by the 2^−△△Ct method (43). Cell-cycle qPCR: HepG2-NC and HepG2SG cells were incubated with 10 μg/mL Hoechst-33342 (37 °C, 20 min), and G1/S/G2/M populations were sorted on a BECKMAN COULTER CytoFLEX STR (≥ 1 × 10⁵ cells/phase). RNA was extracted immediately, reverse-transcribed, and SRGN mRNA quantified as above.

The proteins were extracted in RIPA buffer (Thermo Scientific™, USA), quantified by the bicinchoninic acid assay, separated by 12% SDS-PAGE, and transferred to polyvinylidene difluoride membranes (Immobilon-P Transfer Membrane, Millipore, IPVH00010). After blocking in TBST buffer, membranes were incubated with Rabbit Anti-SRGN antibody (bs-6789R, BIOSS, Beijing, China ratio 1:1000), Anti-NF-κB(D14E12, CST, Danvers, MA, USA)and p65 antibody (Ser468, #8242 and #3039, CST), Phospho-Stat3 (Ser727, #9134, CST), and Anti-STAT3 antibody (EPR787Y, ab68153, Abcam, Boston, MA, USA) with a GAPDH antibody serving as the loading control (Catalog No.KC-5G5, Kangcheng Bio, Shanghai, China, ratio 1:10,000). After incubation with peroxidase-labeled rabbit anti-rat IgG (Bode Biotech Co., Ltd., Wuhan, China, Cat. no BA1058) secondary antibody at 1:2000 and 37 °C for 1 h, the bands were read with a Pro-light HRP Chemiluminescence Kit (TIANGEN, Beijing, China), and Image J software (Gel Image Analyzer, Tianneng Technology Co., Ltd, Shanghai, China, Tanon 1220) (44).

For the cell viability assay, the cell cultures were allocated to a HepG2-NC group and HepG2SG groups, and treated with 2, 5, 10, 15, 20, 25μm sorafenib (S7397, Selleck, Houston, TX, USA) diluted in DMSO (S1209, Selleck), respectively. Cell viability was evaluated by a CCK8 assay (CellTiter 96 AQueous One Solution Cell Proliferation Assay, Cat. No. G3582, Promega, Madison, USA). The cell density was 1×10⁴ cells/100μL per well into a 96-well plate. The cells were collected at each time point (0, 1, 3, and 5 days), and 10 μL CCK-8 solution was added according to the manufacturer’s instructions. The results were obtained with five replicates each and read at an absorbance of 490 nm using multiscan MK3 plate reader (Thermo Fisher Scientific Ltd, Waltham, MA, USA). The effect on cell viability at each assay time was reported as Proliferation rate= (mean OD value at time point÷ mean OD value at primary time point-1) ×100%.

Invasiveness was assayed in HepG2-NC and HepG2SG groups. The methods were performed as before (45). The results were read at absorbance of 570 nm using multiscan MK3 plate reader.

The Matrigel (BD Company, 356234, USA) was added 150-200uL to each well of 48-well plates, and solidified at 37°°C for more than 2h. Human umbilical vein endothelial cells (HUVECs) were incubated with serum-free medium of respective HepG2SG and HepG2-NC for 12 h, then added 2×10⁴ cells/200uL medium to each well, the tube formation was observed after 4-6h, randomly selected 5 fields of the images, analyzed the results with Image J software, and the number of tubular structures was counted. For vasculogenic mimicry, the procedures as above, while the HepG2-NC and HepG2SG cells were used, and the tube formation was observed after 3–5 days.

BALB/c nude mice (male, 5-weeks-old, weight 18~20 g) were purchased from the Guangzhou University of Chinese Medicine Laboratory Animal Center (Guangzhou, China). Tumor cells (5×10⁶ in 0.1 mL phosphate-buffered saline, PBS) were injected into the right axillary region randomly. For tumor formation, the mice were monitored two times per week for 4 weeks and then sacrificed by carbon dioxide asphyxiation. After the animal experiments, carcasses were returned to the Laboratory Animal Center for harmless treatment. The study was conducted in accordance with the ARRIVE guidelines (https://arriveguidelines.org). Animal experiments were approved by Rulge Biotechnology Committee for the Institutional Animal Care and Use (ethical approval number: 20230201002).

The paraffin-embedded tissue sections were subjected to a series of washes for rehydration and clearing following dehydration. The sections were immersed in xylene twice, each for 20 minutes, followed by a graded ethanol series: processing twice with anhydrous ethanol (10 minutes each), 95% ethanol (5 minutes), 90% ethanol (5 minutes), 80% ethanol (5 minutes), and 70% ethanol (5 minutes). Subsequently, the sections were rinsed thoroughly with distilled water. For histological staining, the nuclei were stained using Harris hematoxylin solution for approximately 5 minutes. The sections were then differentiated with hydrochloric acid alcohol, blued in ammonia water, and rinsed again under running water. Cytoplasmic staining was performed by treating the sections with eosin solution for 1 to 3 minutes. After final dehydration steps, the sections were mounted with neutral balsam for microscopic examination.

After dehydration, the sections were processed through xylene and gradient ethanol, then transferred to a retrieval box containing citrate antigen retrieval buffer, allowed to cool naturally, placed in PBS and washed three times. An autofluorescence quencher was applied to the sections for 5 min, and rinsed with running water for 10 min, added BSA in the circle and incubated for 30min, dropped the primary antibody (anti-CD206 antibody, 18704-1-AP, Proteintech, Rosemont, Illinois, USA; Rabbit anti-CD80 antibody, abs137159, Absin, Shanghai, China; Rabbit anti-SRGN antibody, bs-6789R, BIOSS, Beijing, China) on the slice. In a humid box, the slice was incubated at 4°C overnight, then added the secondary antibody in the circle to cover the tissue, and incubated for 50 min at room temperature in the dark. The slides were washed three times. DAPI staining solution was added to the circle and incubated for 10 min, and then the slides washed as above. After air-drying, the sections were added with anti-fluorescence quenching mounting medium and then examined under a fluorescence microscope, with images captured for further analysis.

Sections were sequentially deparaffinized and hydrated. For antigen retrieval, tissue sections, placed in a retrieval cassette filled with citrate-based antigen retrieval buffer, underwent heat-induced epitope retrieval. For endogenous peroxidase blocking, sections were incubated in 3% hydrogen peroxide solution at room temperature. After washing, 3% BSA was applied to for serum blocking. Then sections were incubated with PBS-diluted anti-CD34 antibody (YT0757, ImmunoWay, San Jose, CA, USA) overnight at 4°C in a humidified chamber. After washing, a horseradish peroxidase (HRP, BA1060, Bode)-conjugated secondary antibody was applied and incubated. Freshly prepared 3,3’-Diaminobenzidine (DAB, ab64238, Abcam) substrate solution was applied, and color was monitored under a microscope. For periodic acid-schiff (PAS, G1285, Solarbio, Beijing, China) staining, sections were treated with periodic acid solution, then incubated with PAS reagent. Nuclei were counterstained with hematoxylin. Stained sections were examined under a microscope, and images were acquired and analyzed. Microvessel density (MVD) was analyzed following the protocol of Weidner et al (46).

THP-1 cells at a density of 5×10⁵ cells/mL per well were added in a six-well plate and cultured in RPMI1640 complete medium supplemented with 100 ng/mL of Phorbol 12-Myristate 13-Acetate (PMA, 16561-29-8, Solarbio) and 10% fetal bovine serum (FBS). After 48h of stimulation, the macrophages were successfully differentiated. M1 macrophages were induced by stimulation with 100 ng/mL LPS (L2880, Sigma-Aldrich, St. Louis, Missouri, USA) and 10 ng/mL IFN-γ (4Abiotech, Beijing, China) for 48h. For M2 macrophages, the cells were stimulated with 25 ng/mL IL-4(#200-04, PeproTech, Cranbury, NJ, USA) and 25 ng/mL IL-13(P5178, Beyotime, Haimen District, Jiangsu, China).

M0 cells in the logarithmic growth phase were seeded into the lower chamber (0.3 μm) of the Transwell. HepG2-NC and HepG2SG cells were cultured in the upper chamber, respectively, maintaining the same serum concentration in both chambers. After 48 h, cells collected from the lower chamber were designated as tumor-associated macrophages TAM1 (co-cultured with HepG2-NC) and TAM2 (co-cultured with HepG2SG).

TAM1 and TAM2 (in the upper chamber) were co-cultured in triplicates with HepG2 and Huh7 hepatoma cells (in the lower chamber) for 48h. Subsequently, HepG2 and Huh7 cells were seeded in a 96-well plate at a concentration of 1×10⁴ cells/100 μL per well. Other procedures were followed as step 2.8.

Hepatoma cells at a density of 1×10⁵ were seeded in the upper chamber, added 100 μL serum-free DMEM. TAM1 or TAM2 with 600 μL DMEM were added to the lower chamber, followed by incubation at 37°C and 5% CO2 for 12h. After the chamber was removed, the cells on the upper chamber were carefully wiped away. The cells on the membrane were fixed with 4% paraformaldehyde for 20 min, followed by staining with crystal violet for 10 min. Images of the stained cells were captured for statistical analysis.

For invasion assay, HepG2 and Huh7 cells added to the upper chamber were co-cultured with TAM1 and TAM2 in the lower chamber, respectively, then other steps as step 2.9.

Standards and samples were added 100 μL per well to the corresponding wells, with sample diluent serving as blank. After 48 h, the supernatant was collected and ELISA was performed according to the manufacturer’s instructions. Finally, 50 μL stop solution was added to each well, the results of SRGN, arginase1(Arg1), inducible nitric oxide synthase(iNOS2), interleukin 1β(IL-1β), vascular endothelial growth factor A (VEGF-A) and matrix metalloproteinase-9 (MMP9) were read at the absorbance of 450 nm, with 620 nm as the calibration wavelength. ELISA kits were purchased from Elabscience (Wuhan, China).

For surface marker staining, M2 macrophages, TAM1, and TAM2 cells were incubated with fluorescently-labeled antibodies against CD80 (FITC-conjugated, clone 2D10.4, 11-0809-42, eBioscience™, San Diego, CA, USA) and CD206 (APC-conjugated, clone 19.2, 17-2069-42, eBioscience™) at room temperature for 20 minutes in the dark. After washing with PBS by centrifugation at 1000rpm for 10min, the cells were fixed using 100 μL of Fix&Perm Reagent A for 15 minutes. Finally, the stained cells were resuspended in 0.2 mL PBS for immediate flow cytometry analysis.

Measured data were compared using t-tests or ANOVA using SPSS 22.0 (IBM Corp. Armonk, NY, USA), and the counted data were compared using χ² tests. Differences between means for data with skewed distribution and variance were analyzed using rank-sum tests. Statistical significance was set at P < 0.05.

Single-cell sequencing data of HCC and corresponding normal tissue samples from five patients were retrieved from the GEO database. After rigorous quality control, dimensionality reduction, and clustering analysis, 17 distinct cell subpopulations were identified (Figure 1A). Further cell annotation, along with umap visualization, classified these cells into 12 types: B cells, dendritic cells, endothelial cells, epithelial cells, fibroblast, hepatocyte, macrophages, mast cells, plasma cells, monocytes, T cells, and tumor-associated macrophages (TAMs) (Figure 1B). SRGN gene expression was nearly ubiquitous across all cells, being especially prominent in dendritic cells, endothelial cells, macrophages, fibroblast, mast cells, plasma cells, monocytes, T cells, and TAMs (Figure 1C). Comparative analysis of normal and tumor tissues revealed significantly higher SRGN expression in B cells and macrophages of tumor samples than in normal tissue cells (Figure 1D). Subsequently, we utilized the CROST database (47) to analyze SRGN’s spatial localization and gene expression in different HCC tissue samples. The cell types of VISDS000514, VISDS000454, VISDS000507, and VISDS000513 are shown in Supplementary Figures S1A, C, E, G, respectively. Our findings demonstrated high SRGN expression in HCC tissues, particularly in macrophages of VISDS000514 (Supplementary Figure S1B), VISDS000454 (Supplementary Figure S1D), VISDS000507 (Supplementary Figure S1F), and VISDS000513 (Supplementary Figure S1H).

Building on the observation of SRGN’s prominent expression in macrophages within the TME of HCC, we further explored the relationship between SRGN expression and specific macrophage subsets, along with its prognostic implications in LIHC. Figure 2 provides a more in-depth analysis of SRGN expression across different macrophage subsets in LIHC. GEPIA analysis revealed that in LIHC tumor tissues, SRGN expression was highest in M2 macrophages, followed by M0 and M1 macrophages, when compared to liver tissues and LIHC-adjacent normal tissues (Figures 2A, B). This pattern highlights SRGN’s potential role in modulating macrophage polarization within the TME, potentially promoting a pro-tumorigenic M2 phenotype.

To determine whether macrophage-intrinsic SRGN is functionally relevant, we re-annotated macrophages from the single-cell data into SRGN_High and SRGN_Low subsets (median split; Supplementary Figure S6A). SRGN_High macrophages exhibited a striking up-regulation of pro-angiogenic pathways (VEGF, EGFR, MTORC1 and VEGFR2-mediated permeability) in GSEA (Supplementary Figures S6B, C), suggesting that SRGN expression within macrophages themselves may also contribute to neovascularisation.

Differential expression analysis across pan-cancer via TIMER showed that SRGN RNA levels in tumor tissues were significantly lower than those in adjacent normal tissues in LIHC, breast invasive carcinoma (BRCA), lung squamous cell carcinoma, and LUAD (Supplementary Figure 1). However, GEPIA analysis of RNA-Seq data showed no significant difference in SRGN expression between LIHC and normal tissues, and no significant prognostic association in LIHC (Supplementary Figure 2; Supplementary Table 1). TIMER analysis suggested that SRGN expression was closely related to TME cells (Supplementary Tables 2, 3), which should be considered in multivariate Cox regression models. Notably, Kaplan–Meier plotter multivariate analysis, adjusted for clinical confounders such as stage, grade, AJCC_T stage, demographic factors, treatment history, and vascular invasion, suggested SRGN mRNA as a favorable prognostic indicator for overall survival (OS) and progression free survival (PFS), but an unfavorable one for disease-specific survival (DSS) and relapse-free survival (RFS) (Supplementary Table 4).

TIMER analysis using the Cox proportional hazards model further elucidated the relationship between SRGN expression and monocyte/macrophage subsets in LIHC (Figures 2C–F). The analysis indicated that SRGN expression was significantly associated with the infiltration of various immune cells, including macrophages. In the cohort with high SRGN expression, high infiltration of MDSCs, M1 macrophages, and monocyte_MCPCOUNTER was associated with unfavorable OS, whereas high infiltration of CD8+ T cells, resting memory CD4+ T cells, ECs, and HSCs was associated with favorable prognosis for OS. In the cohort with low SRGN expression, high infiltration of macrophage_TIMER, M0_CIBERSORT, and M2_CIBERSORT-ABS were unfavorable factors for OS. These findings suggest that SRGN may influence the tumor microenvironment by modulating the infiltration and activity of different immune cell subsets, particularly macrophages, thereby impacting patient prognosis.

We stratified the TCGA-LIHC cohort into high-risk and low-risk categories based on the median expression of SRGN. Following this stratification, we performed a differential expression analysis, comparing the high-risk and low-risk groups using thresholds of logFC > 1 or < -1 and p < 0.05. The results of this analysis are illustrated in the volcano plot shown in Figure 3A. Subsequently, we conducted GSEA on the differentially expressed genes, revealing six upregulated pathways (Figure 3B) and six downregulated pathways (Figure 3C). The upregulated pathways included FCERI-mediated MAPK activation, cell surface interactions at the vascular wall, PD-1 signaling, the CTLA4 pathway, the JAK-STAT signaling pathway, and the PI3K-AKT signaling pathway. Conversely, the downregulated pathways encompassed selenoamino acid metabolism, ribosome biogenesis, oxidative phosphorylation, steroid metabolism, and cholesterol biosynthesis.

In addition, we performed enrichment analysis on the six upregulated pathways within the single-cell dataset of HCC. Our findings indicate that FCERI-mediated MAPK activation is upregulated across various cell types, including TAMs, macrophages, plasma cells, mast cells, B cells, dendritic cells, and monocytes. The pathway of cell surface interactions at the vascular wall showed upregulation in macrophages, plasma cells, B cells, endothelial cells, and monocytes. PD-1 signaling was found to be elevated in TAMs, macrophages, dendritic cells, T cells, and monocytes. The CTLA4 pathway exhibited increased expression in TAMs, macrophages, dendritic cells, T cells, and monocytes. Likewise, the JAK-STAT signaling pathway was upregulated in macrophages, dendritic cells, endothelial cells, T cells, and monocytes, while the PI3K-AKT signaling pathway showed upregulation specifically in fibroblasts and endothelial cells. Notably, macrophages were involved in nearly all of the aforementioned pathways (Figures 3D–I).

We subsequently analyzed the infiltration of 24 immune cell types in hepatocellular carcinoma tissues using single-sample Gene Set Enrichment Analysis (ssGSEA). The correlation between the expression levels of SRGN in TPM format and the levels of immune cell infiltration was assessed using Spearman correlation tests. The results revealed a negative correlation between SRGN expression and Th17 cells, while a positive correlation with the majority of immune cell types was observed. Notably, the highest correlation was found with macrophages, yielding a correlation coefficient of 0.7 (Figure 4A). Furthermore, stratifying by the median expression of SRGN, we discovered that the high SRGN expression group exhibited significantly higher infiltration scores for most immune cells, including macrophages, compared to the low SRGN expression group (Figure 4B). Given that the GSEA analysis identified PD-1 signaling and the CTLA4 pathway as upregulated in the high SRGN expression cohort, we proceeded to analyze the correlation between SRGN and immune checkpoint genes. Our findings indicated a positive correlation between SRGN and 40 immune checkpoint genes. The specific immune checkpoint genes and their respective correlation coefficients are detailed in Figure 4C. Notably, while the P-value for EGFR was 0.001, the P-values for all other immune checkpoint genes were less than 0.001. This indicates that high expression of SRGN is closely associated with an immune activation status in hepatocellular carcinoma.

Cell cycle activity was quantified using the CellCycleScoring function in Seurat, which computed S-phase and G2/M-phase gene set scores to classify cells into distinct cycle phases. PCA revealed clear separation along PC1, with S-phase cells clustering in the upper quadrant and G2/M-phase cells in the lower quadrant (Figure 5A), while UMAP demonstrated partial segregation of these populations (Figure 5B). Cell-type–specific analysis showed heterogeneous distributions: macrophages, endothelial cells, and dendritic cells were enriched in G1/S phases, whereas B cells and hepatocytes were predominantly G2/M-phase–enriched; T cells exhibited increased G2/M and S-phase proportions, and TAMs were enriched in G1/S phases (Figure 5C). Critically, SRGN exhibited consistently high expression across all three phases (G1, S, and G2/M), indicating its potential role in cell cycle regulation independent of phase-specific transcriptional programs. These findings collectively suggest SRGN as a cell cycle–associated factor in the HCC microenvironment. To experimentally verify the single-cell observation that SRGN transcript levels remain constant in G1, S and G2/M, we flow-sorted HepG2-NC and HepG2SG cells after Hoechst-33342 staining (Figures 5E, F). qPCR of three independent sorts showed that SRGN mRNA was barely detectable and essentially unchanged across the cycle phases in NC cells, whereas it was markedly elevated in HepG2SG cells without significant cycle-dependent fluctuation (Figure 5G). Thus, SRGN abundance remained uniformly high across G1, S and G2/M phases, mirroring the cycle-independent pattern observed in single-cell data.

Based on the relationship between SRGN and macrophages, we initiated our analysis from macrophages and employed Monocle3 for single-cell trajectory analysis (Supplementary Figure S7A). Subsequently, we conducted a bulk differential gene analysis of time-series macrophage genes. The heatmap demonstrated six clusters of differentially expressed genes (Supplementary Figure S7B). The trend line diagram indicated that the expression dynamics of SRGN gradually increased over time, with macrophages from the normal group exhibiting higher expression levels during the early developmental stages, while those from the tumor group showed enhanced expression in the later developmental stages (Supplementary Figure S7C).

To explore potential intercellular interactions, we utilized CellChat to assess cell communication within the hepatocellular carcinoma dataset GSE242889. The results indicated that overall signaling activity was significantly lower in the normal group compared to the tumor group (Figures 6A, B). Analysis of regulatory relationships revealed that within the tumor microenvironment, macrophages exhibited heightened interactions with T cells, tumor-associated macrophages (TAMs), and endothelial cells. Plasma cells in the tumor group also showed increased communication with other cell types. Given that Gene Set Enrichment Analysis (GSEA) identified enrichment for cell surface interactions at the vascular wall, we focused on the VEGF signaling network from a cell communication perspective. VEGF comprises a family of signaling proteins that promote angiogenesis. In the normal group, communication probability among macrophages, TAMs, and endothelial cells was negligible, while fibroblasts displayed minimal involvement (Figure 6C). In contrast, the tumor group demonstrated enhanced communication probabilities between macrophages, TAMs, fibroblasts, and endothelial cells, particularly targeting endothelial cells (Figure 6D). We visualized the VEGF signaling pathway using chord diagrams for both the normal (Figure 6E) and tumor (Figure 6F) groups. Further dissection of the VEGF network revealed that macrophages in the normal group did not contribute significantly (Figure 6G), whereas in the tumor group, macrophages displayed high expression levels and elevated importance as both Sender and Influencer nodes (Figure 6H).

To functionally validate the enhanced VEGF signaling network predicted by CellChat, we conducted co-culture experiments using HepG2 cells and M0-THP-1 macrophages. We first measured VEGFA mRNA levels in macrophages recovered from the co-culture system. VEGFA transcript levels were more than threefold higher under SRGN-overexpressing HepG2 (HepG2SG) conditions compared to control (HepG2-NC) co-cultures (Figure 6I). Consistent with the mRNA data, ELISA analysis of 48-hour supernatants confirmed a significant increase in VEGF-A protein secretion in SRGN-OE-HepG2 + M0-THP-1 co-cultures relative to control (Figure 6J). Collectively, these data demonstrate that SRGN-high hepatoma cells actively drive VEGF-A production, particularly in macrophages, thereby confirming the computationally predicted upregulation of VEGF-mediated intercellular communication in the tumor microenvironment.

SRGN promoted the proliferation of SRGN-overexpressing HepG2SG cells compared to HepG2-NC cells (transfected with blank control vector) and HepG2 cells (Figures 7A–C, ^*p< 0.05, ^**p < 0.01). HepG2SG cells showed sorafenib resistance (resistance index 1.15, Figure 7D). HepG2SG cells also showed increased cell invasion compared to HepG2-NC cells in the transwell assay (Figure 7E, ^***p< 0.001). The number of tubular structures was higher in human umbilical vein ECs (HUVECs) cultured with HepG2SG supernatant than in those cultured with HepG2-NC supernatant (Figure 7F, ^***p< 0.001). SRGN promoted vascular mimicry tube formation in HepG2SG cells compared to that in HepG2-NC cells (Figure 7G, ^**p< 0.01). Hematoxylin and eosin staining showed that the cells were distributed in clumps or strands in subcutaneous tumor tissues (Figure 7H). IHC staining for CD34 and PAS staining demonstrated enhanced tumor angiogenesis, as evidenced by a higher microvessel density in the HepG2SG group compared to the HepG2-NC group (Figure 7I, ^**p< 0.01). Subcutaneous xenograft tumors derived from HepG2SG cells were heavier than those derived from HepG2-NC cells (Figure 7J, ^**p< 0.01). SRGN and CD206 were upregulated, whereas CD80 was downregulated in subcutaneous tumor tissues in HepG2SG xenograft mice compared with HepG2-NC mice (Figure 7K, ^*p < 0.05, ^**p < 0.01).

The SRGN mRNA expression was examined in THP-1 cells, M0 macrophages, M1 and M2 macrophages, as well as tumor-associated macrophages (TAMs) differentiated by HepG2SG or HepG2-NC cells. Through pairwise comparison, SRGN mRNA levels were found to be significantly different (Figure 8A, ^**p < 0.01), except between TAM1 and M2 (Figure 8A, NS: no significant difference). Among all the cell types, SRGN levels were highest in TAM2, with M0 macrophages having the second-highest increase, yet it was ten times lower than that in TAM2, but higher than THP-1 monocytes. This indicates that SRGN levels are regulated according to the differentiation/polarization of monocytes/macrophages, and SRGN-overexpressing HepG2 may enhance SRGN transcription in macrophages through a positive-feedback loop. Under the microscope, M0 macrophages derived from THP-1 cells induced by PMA exhibited an adherent and irregular shape. These M0 macrophages polarized into M1 and M2 macrophages. M1 macrophages displayed pseudopods and distinct bifurcations, while M2 macrophages exhibited a more uniform morphology, lacking apparent bifurcations. Tumor-associated macrophages 1 (TAM1, co-cultured with HepG2) and TAM2 (co-cultured with SRGN- overexpressing HepG2) demonstrated morphologies characteristic of both M1 and M2 macrophages (Figure 8B).

By flow cytometry, CD206 expression was significantly higher in TAM2 than in TAM1, but lower than in M2 macrophages. The level of CD80 showed no significant difference between TAM2 and TAM1, although CD80 expression was hardly indicated in M2(Figure 9A, ^*p < 0.05, ^** p < 0.01). On day 1 and 2, the proliferation rate of HepG2 and Huh7 cells co-cultured with TAM2 cells was significantly higher than that with TAM1 cells (Figure 9D, ^**,##p ≤ 0.01, ^# p<0.05). The number of migration and invasion HepG2 and Huh7 cells co-cultured with TAM2 cells was significantly greater than that with TAM1 cells, respectively (Figures 9E, F, ^**,##p<0.01).

The levels of Arg1 and SRGN in TAM2 were significantly higher than that in TAM1. Conversely, iNOS2 and IL-1β levels were significantly lower in TAM2 compared to TAM1. However, MMP9 levels showed no significantly difference between TAM1 and TAM2 (Figure 9B, ^**p <0.01, ^*p < 0.05). NF-κB and phosphorylation p65, phosphorylation Stat3 were mostly activated in M2 macrophages, which were secondly increased in TAM2 (Figure 9C, ^**p <0.01, ^* p <0.05). This suggested that NF-κB and STAT3 signaling pathways might be involved in SRGN-promoted TAMs.