Single-cell dissection of the multicellular ecosystem and molecular features underlying microvascular invasion in HCC

Abstract

Background and Aims:

Microvascular invasion (MVI) is a crucial pathological hallmark of HCC that is closely associated with poor outcomes, early recurrence, and intrahepatic metastasis following surgical resection and transplantation. However, the intricate tumor microenvironment and transcriptional programs underlying MVI in HCC remain poorly understood.

Approach and Results:

We performed single-cell RNA sequencing of 46,789 individual cells from 10 samples of MVI+ (MVI present) and MVI- (MVI absent) patients with HCC. We conducted comprehensive and comparative analyses to characterize cellular and molecular features associated with MVI and validated key findings using external bulk, single-cell, and spatial transcriptomic datasets coupled with multiplex immunofluorescence assays. The comparison identified specific subtypes of immune and stromal cells critical to the formation of the immunosuppressive and pro-metastatic microenvironment in MVI+ tumors, including cycling T cells, lysosomal associated membrane protein 3+ dendritic cells, triggering receptor expressed on myeloid cells 2+ macrophages, myofibroblasts, and arterial i endothelial cells. MVI+ malignant cells are characterized by high proliferation rates, whereas MVI- malignant cells exhibit an inflammatory milieu. Additionally, we identified the midkine-dominated interaction between triggering receptor expressed on myeloid cells 2+ macrophages and malignant cells as a contributor to MVI formation and tumor progression. Notably, we unveiled a spatially co-located multicellular community exerting a dominant role in shaping the immunosuppressive microenvironment of MVI and correlating with unfavorable prognosis.

Conclusions:

This study provides a comprehensive single-cell atlas of MVI in HCC, shedding light on the complex multicellular ecosystem and molecular features associated with MVI. These findings deepen our understanding of the underlying mechanisms driving MVI and provide valuable insights for improving clinical diagnosis and developing more effective treatment strategies.

INTRODUCTION

HCC is a prevalent malignancy with substantial global cancer-related mortality.¹ While surgical resection, liver transplantation, and ablation are considered potentially curative treatments, a significant number of patients are deemed inoperable or experience recurrence due to local invasion or metastatic disease.² Microvascular invasion (MVI), the presence of tumor cells within vascular lumens detectable only through histopathological examination, is a critical factor associated with early recurrence and prognosis in HCC.³ MVI manifests as small thrombi formed by malignant cells in the portal and hepatic venous systems, serving as potential sources of intrahepatic and distant metastasis, including posttransplant recurrence.⁴ Therefore, understanding the mechanism underlying MVI holds substantial clinical implications for HCC treatment and management.

Vascular invasion in HCC is a complex biological process involving many factors, including the interplay of HCC with the microenvironment and host states (immune, endocrine, and metabolic).⁵ Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool for characterizing the tumor microenvironment (TME) and dissecting intratumor heterogeneity.^6,7 Several pioneering scRNA-seq studies have predominantly focused on investigating the diversity and heterogeneity of TME in primary, relapsed, or metastatic HCC.^8–13 However, a comprehensive characterization of the multicellular ecosystem and molecular features between MVI+ (MVI present) and MVI- (MVI absent) patients with HCC at the single-cell resolution level remains lacking.

In this study, we present a comprehensive analysis of the multicellular ecosystem comprising immune, stromal, and malignant compartments to dissect tumor heterogeneity in MVI+ and MVI- HCC. Our findings reveal the dominant immunosuppressive cellular community in the TME of MVI+ HCC and highlight the potential therapeutic targets involving the interactions between malignant cells and triggering receptor expressed on myeloid cells 2 (TREM2)+ macrophages in patients with HCC with MVI.

METHODS

Patient and tissue specimen collection

This study was approved by the Research Ethics Committee of the First Affiliated Hospital of Harbin Medical University (201909). Inclusion criteria for patients comprised the absence of apparent metastases, a single tumor between 3 and 5 cm in size, and no prior therapeutic intervention. Written informed consent was obtained from all participants. Ten independent surgically-resected specimens were collected from 5 patients with HCC, including HCC tissues and adjacent non-tumor tissues. Detailed clinical information for patients was listed in Supplemental Figure 1A, http://links.lww.com/HEP/I90. The collected samples were rapidly stored in sCelLive^TM Tissue Preservation Solution (Singleron) after isolation at low temperature, followed by washing with normal saline to remove blood and contaminants.

Library preparation and sequencing

A full description of tissue dissociation, library preparation and single-cell sequencing can be found in the Supplemental Materials and Methods, http://links.lww.com/HEP/I90.

Processing and analysis of scRNA-seq data

A full description of processing and analysis of scRNA-seq data, including preprocessing and quality control of scRNA-seq data, cell type determination and marker gene signature identification, doublets removal, collection and calculation of functional gene module signatures and scores, pathway analysis, Single-Cell rEgulatory Network Inference and Clustering analysis, cell developmental trajectory, cell-cell communication analysis, copy number variation analysis from scRNA-seq, infiltration estimation of cell subsets and infiltration estimation of the CLAMT community, can be found in the Supplemental Materials and Methods, http://links.lww.com/HEP/I90.

Multiplex immunofluorescence

Multiplex immunofluorescence staining assay was performed using Novo-Light®TSA 4-color kit (H-D110041-50T, WiSee Biotechnology), according to the manufacturer’s instruction. A full description of multiplex immunofluorescence and primary antibodies used can be found in the Supplemental Materials and Methods, http://links.lww.com/HEP/I90.

Public scRNA-seq, bulk transcriptome, and spatial transcriptome datasets

The public scRNA-seq, bulk transcriptome and spatial transcriptome datasets can be found in the Supplemental Materials and Methods, http://links.lww.com/HEP/I90.

Statistical analysis

Two-group comparisons of continuous variables, such as gene expression and pathway scores, were performed using the Wilcoxon rank-sum test, while one-way ANOVA was used for comparisons involving multiple groups. Patients in the bulk HCC cohort were categorized based on a defined cutoff using the survdiff() function. The cumulative survival time was estimated using Kaplan-Meier survival curves, and differences in survival were assessed using the log-rank test. Statistical significance was denoted as p < 0.05, p < 0.01, p < 0.001, and p < 0.0001 using *, **, ***, ****, respectively.

RESULTS

Landscapes of cell composition of MVI+ HCC tumor revealed by scRNA-seq analysis

To obtain a comprehensive understanding of the cellular composition and diversity in MVI+ HCC tumors, we conducted scRNA-seq analysis on our discovery cohort, which comprised 3 MVI+ tumor samples, 2 MVI- tumor samples, and 5 paired adjacent non-tumor (NT) samples from 5 patients with HCC (Figure 1A). We used 2 approaches to validate the accuracy of our data: (1) t-SNE plots, both at the sample and patient levels, distinctly segregated tumor and adjacent NT samples, along with distinguishing patients who are MVI− from patients who are MVI+. This confirmed the inherent intrinsic transcriptomic heterogeneity within MVI− and MVI+ HCC samples (Supplemental Figure 1B, http://links.lww.com/HEP/I90); (2) We conducted correlation analysis between the average expression of each gene in our single-cell data and high-quality bulk data from the The Cancer Genome Atlas Liver Hepatocellular Carcinoma (TCGA-LIHC) cohort, which demonstrated high correlations in both tumor and NT samples (Supplemental Figure 1C, http://links.lww.com/HEP/I90). Following data quality control and filtering, we obtained single-cell transcriptomes for 46,789 cells with median gene and unique molecular identifier counts of 1700 and 5806, including 12,704 cells from MVI+ tumors, 9975 cells from MVI− tumors and 24,110 cells from NT (Supplemental Figure 1D, http://links.lww.com/HEP/I90).

FIGURE 1.

Cell composition landscapes of MVI+ HCC tumors revealed by scRNA-seq analysis. (A) Experimental workflow showing the collection and processing of fresh HCC tissues and matched non-tumor live tissues for scRNA-seq (created by BioRender.com). (B) UMAP visualization showing the major cell types. Colors indicate different cell types. (C) UMAP visualization showing the expression of canonical cell type marker genes. (D) Heatmap showing the top 20 upregulated genes (ranked by log2 fold change) for each cell type, with representative genes marked (left). GO enrichment analyses of the top 20 upregulated genes for each cell type, highlighting representative GO terms (right). (E) UMAP visualization showing all cells, with colors representing different patients (top) and tissue origins (bottom). (F) UMAP split into individuals with 4 tissue origins. Colors indicate different cell types. (G) A stacked plot presenting the proportion of each cell type in the 4 tissue origins. Colors represent different cell types—TT, NY. Abbreviations: GO, Gene Ontology; NT, nontumor tissue; scRNA-seq, single-cell RNA sequencing; TT, tumor tissue; UMAP, Uniform Manifold Approximation and Projection.

Seven major cell types were identified according to well-known canonical marker genes, including 7,419 T/NK, T cell and NK cell cells (CD3E and killer cell lectin like receptor D1), 3,982 B cells (CD79A), 18,164 myeloid cells (lysozyme), 3,258 endothelial cells (platelet and endothelial cell adhesion molecule 1), 1,008 mesenchymal cells (Actin Alpha 2), 1,506 hepatic progenitor cells [hepatic progenitor cells (HPCs), epithelial cell adhesion molecule] and 11,452 malignant cells (Transferrin) (Figure 1B, C). We further confirmed the robustness of cell type annotation by deciphering the upregulated genes and unique biological features for each cell type (Figure 1D, Supplemental Table S1 and S2, http://links.lww.com/HEP/I91, S3, http://links.lww.com/HEP/I92, http://links.lww.com/HEP/I216). The patient-specific analysis demonstrated that immune and stromal components clustered by cell types, with varying proportions in tumors and NTs, whereas malignant cells showed significant intertumoral heterogeneity and patient-specific clusters (Figure 1E).

To illustrate the diverse cellular ecosystems, we partitioned the Uniform Manifold Approximation and Projection plot into 4 distinct groups according to tissue origin (Figure 1F). Comparative analysis showed that tumor ecosystems exhibited enrichment in myeloid cells and deficiency in T/NK cells and B cells, consistent with prior single-cell studies on HCC,^8,11,14 T/NK cells and mesenchymal cells were prevalent in MVI+ tumors but scarcely present in MVI− tumors (Figure 1G), indicating the intricate and heterogeneous cellular composition implicated in MVI formation in HCC.

Identification of T cell subsets nurturing the immunosuppressive TME in MVI+ tumors

Given the dramatic variation in T/NK cell composition between MVI+ and MVI− tumors, we extracted and re-clustered T/NK cells into distinct subsets. We identified 5 T cell subsets, 3 NK cell subsets, and 1 cycling subset with notably elevated S score and G2M score (Figure 2A-C, Supplemental Table S3, http://links.lww.com/HEP/I216). Using OR analysis to determine the distribution preference of each subset across different tissues (Figure 2D), we found that canonical antitumor immune-related participants (TNF Tem, T naïve, and NK subsets) were enriched in NT tissues, while cycling and IL7R Tem cells displayed enrichment in MVI+ tumors. Treg cells were significantly distributed in both MVI+ and MVI− tumors. We further divided cycling cells into cycling T and cycling NK, and observed that these 2 subsets exhibited different enrichment patterns, with cycling T cells being more prevalent in MVI+ tumor tissues, while cycling NK cells were notably enriched in MVI+ NT tissues (Figure 2E-H). Gene Set Enrichment Analysis demonstrated significant enrichment of proliferation-related pathways in cycling T cells, while inflammation-related pathways were downregulated, implying the reduced immediate effector function of cycling T cells (Figure 2I, J). These findings suggested that cycling T cells may represent potential reactive subsets and play a crucial role in shaping the immunosuppressive microenvironment of MVI+ HCC. We retrospectively analyzed T/NK cells in our validation cohort and validated the presence and enrichment of cycling T cells in MVI+ tumors (Supplemental Figure 2A-C, http://links.lww.com/HEP/I90, Supplemental Table S4, http://links.lww.com/HEP/I93). Multiplex immunofluorescence staining further confirmed the increased abundance of cycling T cells (CD3+Ki67+ cells) in MVI+ HCC samples compared to MVI− HCC samples (Figure 2K).

FIGURE 2.

Characterization of heterogeneous T cell populations in MVI+ and MVI- HCC. (A) UMAP visualization showing the T/NK subsets. Colors indicate different cell subsets. (B) Dot plot showing the expression of marker genes of each T/NK subset. Color represents normalized expression levels and dot size indicates the fraction of expressing cells. (C) Violin plots showing the scores of gene signatures for each T/NK subset. Color represents the mean signature score. (D) Heatmap showing the ORs of T/NK subsets occurring in each tissue. *OR > 1.5 indicates significant enrichment of the subset in the corresponding tissue; (E) UMAP visualization showing the cycling subsets. Colors represent different cycling subsets. (F) UMAP visualization showing the expression of upregulated genes in cycling subsets. (G) UMAP split into individuals with four tissue origins. Colors represent different cycling subsets. (H) Stacked plot showing the proportion of tissue origins in cycling subsets. Colors represent different tissue origins. (I) GSEA enrichment plot showing the upregulated pathways (top) and downregulated pathways (bottom) of cycling T compared to other T/NK subsets. (J) Violin plots showing the expression of marker genes for cycling T. Colors represent different T/NK subsets. (K) Representative multiplex immunofluorescence staining images indicating CD3+Ki67+cells in MVI− HCC and MVI+ HCC samples. Scale bar, 50 μm. (L) Scatter plot exhibiting gene expression log2 fold change between MVI+ and MVI− T/ NK cells. Color and dot size represent the value of log2 fold change. The genes validated in the validation cohort were labeled. Abbreviations: GSEA, Gene Set Enrichment Analysis; MVI+, MVI present; MVI−, MVI absent; T/NK, T cell and NK cell; Tex, exhausted T cell; Tem, effector memory T cell; UMAP, Uniform Manifold Approximation and Projection.

To investigate the exhausted states of CD8+T cells in HCC, we quantitatively compared the exhaustion phenotypes, including precursor exhaustion, terminal exhaustion, and cytotoxicity, of 4 T cell subsets with high CD8A expression (Supplemental Figure 2D, http://links.lww.com/HEP/I90). Notably, we observed that cycling T cells were biased towards a terminally exhausted state, with the highest terminally exhausted score and significantly lower precursor exhausted score compared to other subsets. Moreover, cycling T cells exhibited weak cytotoxicity, indicating their location in a dynamic terminal state with T cell dysfunction. Additionally, we identified cycling T cells as a potential neoantigen-reactive subset with a highly enriched tumor-specific signature score, similar to the proliferative CD8+exhausted T cell observed in ovarian cancer (Supplemental Figure 2D, http://links.lww.com/HEP/I90).^15,16 Specifically, cycling T cells demonstrated a lower precursor exhausted score, lower cytotoxicity score, and the highest terminal exhaustion score in the validation cohort (Supplemental Figure 2E, http://links.lww.com/HEP/I90).

To provide further evidence for the terminal state of cycling T cells, we inferred cell transitions in CD8+T cells using Monocle2. The cytotoxicity-exhaustion trajectory path showed that TNF Tem was at the beginning with the highest cytotoxicity, whereas cycling T cells were at a terminal state (Supplemental Figure 2F, http://links.lww.com/HEP/I90). The early stages of the trajectory path were predominantly characterized by the expression of chemokines and inflammation-related genes, and the exhaustion-related signatures (Programmed cell death protein 1, Cytotoxic T-Lymphocyte Associated Protein 4, hepatitis A virus cellular receptor 2, and thymocyte selection associated high mobility group box) and proliferation genes were significantly elevated along the cytotoxicity-exhaustion path (Supplemental Figure 2G, http://links.lww.com/HEP/I90). Moreover, we identified molecular features globally upregulated in T/NK cells in MVI+ tumors compared to MVI− tumors. We validated these features in the validation cohort to ensure their robustness. A set of 11 genes was found to overlap in the discovery cohort and validation cohort, including immune regulatory factors (thioredoxin interacting protein, CD52 molecule, galectin 1), tumor-promoting genes (S100A11, S100A6), and the hallmark of memory T cell (IL7R) (Figure 2L). Finally, we assessed the infiltration abundance of cycling T cells within our bulk HCC cohort to explore its clinical significance. Cycling T was significantly associated with poor prognosis in HCC (p < 0.0001, HR = 2.18). Moreover, compared to MVI− HCC, MVI+ HCC exhibits a higher infiltration abundance of cycling T (Supplemental Figure 2I, http://links.lww.com/HEP/I90).

Diversity of B cells in MVI+ and MVI-tumors

We conducted a comprehensive investigation of B cells within HCC ecosystems, revealing the presence of 8 distinct subsets. One belonged to B cells among these subsets, while the remaining 7 were identified as plasma cell subsets. The differentiation between these subsets was based on their varying expression patterns of immunoglobulin isotype genes, providing a detailed understanding of the remarkable heterogeneity exhibited by B/plasma cells in HCC (Supplemental Figure 3A, B, Supplemental Table S3, http://links.lww.com/HEP/I216, http://links.lww.com/HEP/I90). By conducting a joint analysis of their tissue distribution and pathway activities (Supplemental Figure 3C, D, http://links.lww.com/HEP/I90), we observed that 2 plasma cell subsets, IgL plasma cells and IgG2 plasma cells, were significantly enriched in MVI+ tumors. Notably, these subsets displayed activation of pathways associated with epithelial-mesenchymal transition, coagulation, and angiogenesis. These observations suggest that these specific plasma cell subsets may potentially promote MVI in HCC.

Myeloid-derived cell components in MVI+ and MVI− tumors

Among the macrophage population, we identified 4 distinct subsets (Figure 3A-C, Supplemental Figure 4A, Supplemental Table S3, http://links.lww.com/HEP/I90). Two subsets, namely AP Macro and macrophage receptor with collagenous structure (MARCO) Macro, were found to be enriched in NT samples (Figure 3D). AP Macro exhibited high expression levels of genes associated with antigen presentation, such as major histocompatibility complex, class II, DR beta 6, major histocompatibility complex, class II, DP beta 1, and major histocompatibility complex, class II, DR beta 1, while MARCO Macro was marked by MARCO expression (Figure 3C). On the other hand, IL1B Macro, associated with inflammatory activity, demonstrated comparable enrichment in MVI− tumors, and upregulation of inflammatory factors like IL1B, C-C motif chemokine ligand 3, and C-X-C motif chemokine ligand 2 (Figure 3C, D). The TREM2 Macro subset, sharing similarities with lipid-associated macrophages, showed a preferential enrichment in MVI+ tumors as evidenced by increased the expression of TREM2, apolipoprotein E, and apolipoprotein C1.^11,17

FIGURE 3.

Characterization of myeloid-derived cells in MVI+ and MVI- tumors. (A) UMAP visualization showing the myeloid cell components. Colors indicate different cell components. (B) UMAP visualization showing the macrophage subsets. Colors represent different cell subsets. (Macro, macrophage). (C) Dot plot showing the expression of marker genes of each macrophage subset. Color represents normalized expression levels and dot size indicates the fraction of expressing cells. (AP, Antigen presentation). (D) Heatmap showing the ORs of macrophage subsets occurring in each tissue. *OR > 1.5 significant enrichment of the subset in the corresponding tissue. (E) Violin plots showing the scores of gene signatures for each macrophage subset. Color represents the mean signature score. (F) GSEA enrichment plot showing the upregulated pathways (left) and downregulated pathways (right) of TREM2 macrophages compared to other macrophage subsets. (G) Heatmap showing the GSVA activity of metabolic pathways for each macrophage subset. Color represents the mean GSVA activity score. (H) Representative multiplex immunofluorescence staining images indicating C1QA+TREM2+cells in MVI− HCC and MVI+ HCC samples. Scale bar, 50 μm. (I) UMAP visualization showing the DC subsets. Colors represent different cell subsets. (J) Heatmap showing the ORs of DC subsets occurring in each tissue. *OR > 1.5 indicates that the subset is significantly enriched in the corresponding tissue. (K) Dot plot showing the expression of marker genes of each DC subset. Color represents normalized expression levels and dot size indicates the fraction of expressing cells. (L) Violin plots showing the scores of gene signatures for each DC subset. Color represents the mean signature score. (M) Heatmap showing the expression of MHC class I and class II molecules for each DC subset. Color represents mean expression. (N) Violin plot showing the expression of CD274 for each DC subset. Colors represent different cell subsets. (O) Scatter plot showing the correlation between the proportions of LAMP3 DC and cycling T cells, Treg cells and other T/NK subsets, respectively. (P) Scatter plot exhibiting gene expression log2 fold change between MVI+ and MVI− myeloid cells. Color and dot size represent the value of log2 fold change. The genes validated in the validation cohort were labeled. Abbreviation: AP, Antigen presentation; DC, dendritic cells; GSEA, Gene Set Enrichment Analysis; GSVA, Gene Set Variation Analysis; LAMP3, lysosomal associated membrane protein 3; MHC, major histocompatibility complex; MVI+, MVI present; MVI−, MVI absent; TREM2, triggering receptor expressed on myeloid cells 2; UMAP, Uniform Manifold Approximation and Projection.

To gain insights into the macrophage subset phenotypes in MVI of HCC, we examined M1 and M2 polarization using established gene signatures.¹⁸ We observed a distinct activation state among MVI+ and MVI− tumors, with TREM2 Macro exhibiting a high M2 score (anti-inflammatory) and IL1B Macro showing a dominant M1 phenotype (inflammatory) (Figure 3E). Gene Set Enrichment Analysis revealed significant upregulation of pathways related to “Coagulation,” “Adipogenesis” and “KRAS signaling up,” indicating that high metabolism and pro-metastatic characteristics were critical identifiers of this macrophage subset (Figure 3F). In contrast, inflammation-related pathways were downregulated (Figure 3F). Furthermore, a comparative analysis of cancer-associated metabolic pathways revealed marked activation within TREM2 Macro, implying metabolic dysregulation possibly contributing to tumor progression (Figure 3G). Lastly, multiplex immunofluorescence staining demonstrated the enrichment of TREM2 Macro in MVI+ HCC (Figure 3H).

In addition to conventional dendritic cells (DC) subsets of CD1C DC and CLEC9A DC, we identified 2 nonclassical DC subsets, cycling DC and LAMP3 DC, predominantly present in MVI+ tumors (Figure 3I-K, Supplemental Table S3, http://links.lww.com/HEP/I216). LAMP3 DC exhibited a high capacity for activation and migration from the tumor to the vascular region. This is consistent with its high tolerogenic score, indicating multifunctional and immunosuppressive properties (Figure 3L). In contrast, cycling DC demonstrated higher scores for S and G2M, but lower scores for antigen presentation (Figure 3L). Major histocompatibility complex molecules, essential for antigen presentation, were depleted in cycling DC (Figure 3M), indicating a state of increased proliferation and dysfunction. Remarkably, LAMP3 DCs exhibited high CD274 (PD-L1) expression, which has been shown to recruit T cell subsets to the tumor region and regulate the complex TME (Figure 3N).^11,19 We observed a positive correlation between the proportions of LAMP3 DC and 2 immunosuppressive T cell subsets (cycling T and Treg), but an inverse trend with other T cell subsets (Figure 3O), indicating that LAMP3 DCs play a crucial role in the formation of an immunosuppressive microenvironment in MVI+ tumors. The enrichment of LAMP3 DCs in MVI+ HCC was further confirmed using multiplex immunofluorescence staining (Supplemental Figure 4B, http://links.lww.com/HEP/I90).

We identified and validated 9 genes that exhibited significant upregulation in myeloid cells in MVI+ tumors, indicating their potential as distinctive features and promising therapeutic targets for MVI+ tumors (Figure 3P). To validate the MVI+ tumor-enriched myeloid subsets identified above, we analyzed myeloid cells in the validation cohort and clustered them into 5 distinct myeloid subsets based on their differentially expressed genes and TREM2 Macro and LAMP3 DC subsets were identified using specific markers (Supplemental Figure 4C, D, http://links.lww.com/HEP/I90, Supplemental Table S4 http://links.lww.com/HEP/I93). Similarly, these 2 subsets in the validation cohort exhibited higher infiltration levels within MVI+ tumors (Supplemental Figure 4D, http://links.lww.com/HEP/I90). In the bulk HCC cohort, we assessed the infiltration abundance of MVI+ HCC-enriched myeloid subsets (TREM2 Macro, Cycling DC, and LAMP3 DC). We validated that these subsets displayed higher infiltration in MVI+ HCC in the TCGA-LIHC cohort, and TREM2 Macro, Cycling DC and LAMP3 DC were related to poor prognosis (Supplemental Figure 4E, F, http://links.lww.com/HEP/I90). Although we detected 3 monocyte subsets (Supplemental Figure 4G-I, http://links.lww.com/HEP/I90), no subset showed significant enrichment in MVI+ tumors.

Dissection of stromal components and regulators in MVI+ tumors

We re-clustered stromal components into 2 mesenchymal subsets and 8 endothelial subsets (Supplemental Table S3). Among the endothelial subsets, 2 arterial endothelial cell subsets were identified that simultaneously expressed arterial markers hes related family bHLH transcription factor with YRPW motif 1 and C-X-C motif chemokine ligand 12, but were separated in the Uniform Manifold Approximation and Projection plot. These 2 subsets were named arterial i EC and arterial ii EC (Figure 4A, B, Supplemental Figure 5A, http://links.lww.com/HEP/I90). We calculated the correlation coefficients among the stromal subsets and visualized them by hierarchical clustering (Figure 4C), revealing transcriptional similarity between arterial i endothelial cell (EC) and arterial ii EC. However, these 2 arterial subsets showed distinct tissue enrichment, with arterial ii EC being specific to NT, while arterial i EC being predominantly enriched in MVI+ tumors, suggesting that arterial EC may have undergone phenotypic changes from adjacent tissue to MVI+ HCC (Figure 4D). To elucidate further the transcriptional heterogeneity and functional states of arterial EC, we performed DE analysis between arterial i EC and arterial ii EC, and observed enrichment of genes involved in tumor development and metabolism in arterial i EC. In contrast, genes upregulated in arterial ii EC were primarily associated with the production and secretion of hormones or proteins for homeostasis regulation (Figure 4E, F). Trajectory analysis revealed that these two arterial EC subsets were activated in the terminal phase of endothelial cell polarization and exhibited 2 distinct polarization directions (Figure 4G). This suggests that despite sharing the endothelial lineage, the accumulation of arterial i EC might contribute to tumor progression. Single-Cell rEgulatory Network Inference and Clustering analysis identified differentially activated and expressed transcription factors along these trajectories, with BCL2 associated X being activated in arterial i EC polarization and FOXC1 being activated in arterial ii EC, indicating that these two transcription factors may be involved in polarization processes of different arterial EC subsets in adjacent NT and tumor tissues (see Methods, Figure 4H). Survival analysis further revealed an inverse correlation between these two transcription factors and survival rates (Supplemental Figure 5B, C, http://links.lww.com/HEP/I90).

FIGURE 4.

Characterization of stromal components and regulators in MVI+ tumors. (A) UMAP visualization showing the stromal subsets. Colors represent different cell subsets. (B) Dot plot showing the expression of marker genes for each stromal subset. Color represents normalized expression levels, and dot size indicates the fraction of expressing cells. (C) Heatmap showing the relationship between different stromal subsets. Color represents correlation coefficients. (D) Heatmap showing the ORs of stromal subsets occurring in each tissue. *OR > 1.5 indicates that the subset is significantly enriched in the corresponding tissue. (E) Heatmap showing the DEGs between arterial i EC and arterial ii EC. Color represents correlation coefficients. Additional annotation includes cell subset and tissue. (F) KEGG enrichment analysis based on the DEGs presenting the top 10 enriched pathways of arterial i EC and arterial ii EC. Color represents arterial i EC and arterial ii EC. Color represents arterial i EC and arterial ii EC. (G) Monocle 2 trajectory analysis of arterial i EC, arterial ii EC, and other ECs. Colors represent different cell subsets. (H) Heatmap showing the TF activity among arterial i EC and arterial ii EC. Two specific TFs were circled. (I) Heatmap showing the GSVA activity of 50 cancer-associated hallmark pathways for each stromal subset. Color represents the mean GSVA activity score. Additional annotation includes the process category of each pathway. (J) Violin plot showing the score of growth factor for each stromal subset. Color represents the mean score. Abbreviations: DEG, differentially expressed genes; GSVA, Gene Set Variation Analysis; mCAF, myo-cancer-associated fibroblasts; MVI+, MVI present; TEC, tumor-associated endothelial cell; TF, transcription factors; UMAP, Uniform Manifold Approximation and Projection; VCMC, vascular smooth muscle cells.

Gene Set Variation Analysis analysis of cancer-associated hallmark pathways demonstrated that myo-cancer-associated fibroblasts (mCAFs) and arterial i EC were preferentially enriched in MVI+ tumors and exhibited a high abundance of pathways related to tumor development, including adipogenesis, angiogenesis, and epithelial-mesenchymal transition (Figure 4I). This suggests that these stromal cells have the potential to promote the formation of tumor blood vessels and metastasis in HCC. MCAFs and arterial i EC also highly activated growth factor pathways (Figure 4J). We confirmed the enrichment of mCAFs and arterial i EC through multiplex immunofluorescence staining (Supplemental Figure 5D, E, http://links.lww.com/HEP/I90). Validation in the TCGA-LIHC cohort demonstrated the enriched presence of mCAFs and arterial i EC in MVI+ HCC, consistent with unfavorable survival outcomes, and arterial ii EC correlated with improved survival rates (Supplemental Figure 5F, G, http://links.lww.com/HEP/I90).

The heterogeneity of malignant cells between MVI− and MVI+ HCC

To investigate the role of malignant cells in the TME and their transcriptional heterogeneity in MVI− and MVI+ tumors, we extracted HPCs and malignant cells and re-clustered malignant cells into MVI+ malignant and MVI− malignant cells based on their tissue origin (Figure 5A). We compared the copy number variations of malignant cells and nonmalignant HPCs as a reference. Copy number variation inference of 5 patients revealed extensive intratumor heterogeneity. We observed notable insertions in chromosome 7 in MVI+ malignant cells, and MVI− malignant cells predominantly exhibited chromosome 1 insertions (Supplemental Figure 6A, http://links.lww.com/HEP/I90). Cancer hallmark enrichment analysis showed that MVI+ malignant cells exhibited significant upregulation of genes associated with proliferation-associated pathways, (Figure 5B). We estimated the proportion of cells in different cell cycle phases. As the tumor progressed, the proportion of proliferative cells (S and G2M phases) gradually increased, indicating their increased proliferative activity (Figure 5C). Additionally, we confirmed the high proliferation rate of MVI+ malignant cells in the validation cohort (Supplemental Figure 6B, http://links.lww.com/HEP/I90). MVI− malignant cells were enriched in the “TNF signaling through NF-κB” pathway, which is associated with pro-inflammatory responses (Figure 5B). These cells also exhibited activation of typical pro-inflammatory chemokines, including C-C motif chemokine ligand 3, CCL4, and C-X-C motif chemokine ligand 2, as well as significantly elevated inflammatory response activity (Figure 5D), and these inflammatory characteristics were also confirmed in the validation cohort (Supplemental Figure 6C, D, http://links.lww.com/HEP/I90). We calculated the inflammation and proliferation scores in the TCGA-LIHC cohort and contrasted them between MVI− and MVI+ HCC, revealing a consistent trend that was consistent with the single-cell cohort findings (Supplemental Figure 6E, http://links.lww.com/HEP/I90). Our results indicated the distinct tumor ecosystems and characteristics of MVI+ and MVI−malignant cells.

FIGURE 5.

The heterogeneity of malignant cells between MVI− and MVI+ HCC. (A) UMAP visualization showing HPC and malignant cells. Colors represent different cell subsets. (HPC, hepatic progenitor cell). (B) Bar plot showing the GSVA activity of top 5 cancer-associated hallmark pathways for MVI+ and MVI− malignant cells. Colors represent different cell subsets. (C) Stacked plot shows the proportion of cells in G1, S and G2M phases in HPC and malignant subsets. Color represents different cell cycle phases. (D) Violin plots showing the expression of inflammatory factors and inflammatory response score for MVI+ and MVI− malignant cells. Color represents different cell subsets. (E) Monocle 2 trajectory analysis of HPC and malignant subsets. Colors represent different cell subsets. Additional annotation includes the density of cell subsets and stem score along the trajectory. (F) Heatmap showing the dynamic DEGs and their enriched pathways along the trajectory. These DEGs were divided into four main clusters. (G) Workflow for identifying key markers of MVI+ HCC. (H) Network plot showing the ranked correlation between malignant cells and other cell subsets. Color and dot size indicate the correlation coefficients. (I) Bubble plot showing the MDK-mediated interaction from malignant cell to TREM2 Macro in the discovery and validation cohorts, respectively. Statistically significant ligand-receptor pairs (p < 0.05) were selected. Color represents different tissues, and dot color indicates the interaction probability. (J) Scatter plot showing the correlation between the proportion of TREM2 Macro and the expression levels of MDK from all 10 samples in the discovery cohort. (K) Sketch map illustrating the MDK-mediated interaction from malignant cell to TREM2 Macro (created by BioRender.com). Abbreviations: DEG, differentially expressed genes; GSVA, Gene Set Variation Analysis; MDK, midkine; MVI+, MVI present; MVI−, MVI absent; T/NK, T cell and NK cell; TREM2: triggering receptor expressed on myeloid cells 2; UMAP, Uniform Manifold Approximation and Projection.

Using Monocle, we investigated the dynamic cell transitions occurring in tumor states. The trajectory analysis revealed a transitional relationship where MVI− malignant cells originated from HPCs and further transitioned to MVI+ malignant cells, representing the stepwise progression of HCC (Figure 5E). Additionally, we observed a gradual decrease in stemness scores along this trajectory, indicating a shift from a stem-like to a differentiated state (Figure 5E). To understand the transcriptional programs driving the formation of MVI, we analyzed the differential gene expression profiles along the trajectory (Figure 5F). Initially, highly expressed genes involved in TNF-alpha signaling through NF-κB, apoptosis, and immune-related pathways played crucial roles in tissue homeostasis and development. The intermediate state was marked by increased metabolic activity, followed by the enrichment of coagulation and angiogenesis, suggesting the acquisition of metastatic properties as the tumors progressed (Figure 5F). midkine (MDK) and stathmin 1 were identified as risky genes associated with unfavorable prognosis in MVI+ tumors through a multi-step analysis strategy (Figure 5G). Immunofluorescence staining verified elevated expression of MDK and stathmin 1 in MVI+ HCC compared to MVI− HCC (Supplemental Figure 7A, http://links.lww.com/HEP/I90). Correlation analysis among cell subsets showed that TREM2 Macro had the strongest correlation with malignant cells, as well as in the validation cohort (Figure 5H, Supplemental Figure 7B, http://links.lww.com/HEP/I90). Therefore, we identified putative cell-cell interactions between these two cell types using CellChat. MVI+ HCC exhibited upregulated receptor-ligand pairs mediating interactions between malignant cells and TREM2 Macro. Notably, we observed that MDK played a prominent role in facilitating cellular crosstalk between these two cell types (Figure 5I). Among the MVI+ HCC-upregulated interaction pairs (compared to MVI− HCC), we found that the MDK-nucleolin (NCL) pair exhibited the strongest interaction strength between malignant cells and TREM2 Macro in both the discovery and validation cohorts (Figure 5I). We further validated a robust positive correlation between the average expression of MDK and the ratio of TREM2 Macro in each sample in both the discovery and validation cohorts (Figure 5J, Supplemental Figure 7C, http://links.lww.com/HEP/I90). Within the MVI− HCC of the TCGA-LIHC cohort, MDK expression did not correlate with TREM2 macro infiltration levels. However, significant differences emerged within the MVI+HCC, where high MDK expression levels correlated with elevated TREM2 macro levels (Supplemental Figure 7D, http://links.lww.com/HEP/I90). These findings collectively suggested that MDK-mediated crosstalk between malignant cells and TREM2 macrophages exhibited MVI+ HCC-specific implications. Additionally, we confirmed a significant positive association between MDK and NCL in the bulk HCC cohort (Supplemental Figure 7E, http://links.lww.com/HEP/I90). Taken together, these findings provide mechanistic insight into how malignant cells recruit TREM2 macrophages to promote tumor progression in MVI+ tumors through the MDK-mediated axis (Figure 5K).

Identification of a unique multicellular community rendering the TME in MVI+ tumors

To gain a comprehensive understanding of the associations among different immune and stromal subsets in HCC, we focused our analyses on the co-enrichment programs of cells and clustered them into multiple groups based on corresponding patterns (Figure 6A). Among these groups, we identified a stable cellular community called CLAMT, consisting of cycling T, LAMP3 DC, arterial i EC, mCAF, and Treg. This community was characterized by high aggregation and was predominantly enriched in MVI+ HCC. Notably, CLAMT encompassed subsets contributing to immunosuppressive and pro-tumorigenic effects, thus holding a pivotal role in shaping the TME of MVI+ HCC. We estimated the infiltration level of the CLAMT community for each patient with HCC in the bulk HCC cohort and observed a significant association between high CLAMT community infiltration and poorer clinical prognosis (Figure 6B). We also confirmed the significant infiltration of CLAMT in MVI+ HCC in TCGA-LIHC cohort (Figure 6C).

FIGURE 6.

Identification of a unique cellular community rendering the TME in MVI+ tumors. (A) Heatmap showing the correlation coefficients between proportions of immune cell and stromal subsets. p-value<0.1 is used as the cutoff value. (B) Kaplan-Meier plot of overall survival between high CLAMT patients and low CLAMT patients. Statistical analysis was performed by the log-rank test. (C) The violin plot illustrated the differences in CLAMT scores between patients with MVI− HCC and MVI+ HCC. (D) Chord diagram showing the interaction between major cell types in MVI+ HCC (left) and MVI− HCC (right). (E) Scatter plot showing the specific ligands of MVI+ and MVI− HCC (top) and their enriched pathways (bottom). Color represents different tissues, and dot color indicates the interaction probability. Abbreviations: MVI+, MVI present; MVI−, MVI absent; TME, tumor microenvironment.

Given the broad co-enrichment communities examined across cell populations, we constructed cellular communication networks and observed that malignant and mesenchymal cells in MVI+ HCC exhibited more extensive interactions with other components compared to MVI− HCC (Figure 6D). Furthermore, we identified specific ligands prominently expressed by cells from MVI+ and MVI− HCC. In particular, ligands expressed in MVI+ HCC-derived cells were related to tumor-promoting and metastasis pathways and were primarily distributed in stromal components. On the other hand, ligands expressed in MVI− HCC-derived cells were enriched in cytokine and inflammation pathways (Figure 6E, Supplemental Figure 8A, http://links.lww.com/HEP/I90). These findings indicate distinct interactive patterns between MVI+ and MVI− HCC.

In our comparative analysis of signaling strength in MVI+ and MVI− HCC, we identified the top 5 preferential subsets in each subtype (Supplemental Figure 8B, http://links.lww.com/HEP/I90). Among these, IL1B Macro with an M1 pro-inflammatory phenotype and classical antigen presenting cell (CD1C DC and CLEC9A DC) were highly preferred in MVI− HCC, while arterial i EC and mCAF, 2 members of the CLAMT community, exhibited increased outcoming strength in MVI+ HCC.

To demonstrate the intimate crosstalk within the CLAMT community, we conducted a comprehensive analysis of the cellular interaction output by arterial i EC and mCAF to other members of the CLAMT community. Our results revealed strong communication between arterial i EC and mCAF with cycling T cells in the T/ NK cell subset. Additionally, they showed close interactions with LAMP3 DC in the myeloid cell subset regarding interaction strength (Supplemental Figure 8C, http://links.lww.com/HEP/I90). We further investigated the detailed reciprocal ligand-receptor pairs involved in these interactions. Specifically, in MVI+ HCC, the MIF-CD74 axis was specifically detected in stromal-immune interactions (Figure 7A, B), known for its pro-metastatic functions.⁸ Arterial i EC and mCAF exhibited stronger interactions with LAMP3 DC, cycling T cells, and Treg cells through the FN1-CD44 molecule (CD44) axis (Figure 7A), associated with hepatic fibrogenesis and malignancy.²⁰ Moreover, ECM-related interactions involving collagen (collagen type I alpha2-CD44, COL4A1-CD44, COL4A2-CD44) and laminin (LAMC1-CD44, LAMA4-CD44) secreted from mCAF and interacting with receptors expressed on LAMP3 DC, cycling T cells, and Treg cells were observed, resulting in immune cell activation and migration in the TME, particularly enriched in MVI+ HCC (Figure 7A, B). Furthermore, we observed diverse regulation from LAMP3 DC to cycling T cells and Treg cells compared to other T/NK subsets. Co-inhibitory signals (nectin cell adhesion molecule 2 - T cell immunoreceptor with Ig and ITIM domains, CD86-Cytotoxic T-Lymphocyte Associated Protein 4, and B and T lymphocyte associated - TNF receptor superfamily member 14) and chemokine-mediated signals (CXCL9-CXCR3) were activated between LAMP3 DC and T/NK subsets, especially in Treg cells, playing a central role in immune recruitment and shaping the immunosuppressive environment in MVI+ HCC (Figure 7C). LAMP3 DC was also able to activate cycling T cells through the CD80/CD86-CD28 and ALCAM-CD6 axis (Figure 7C). These findings provide crucial insights and clinical implications for the development of MVI+ HCC.

FIGURE 7.

Differences in molecular interaction between MVI+ and MVI− HCC-derived cells. (A) Bubble plot showing the interaction from arterial i EC and mCAF to DC subsets in MVI+ and MVI- HCC. Statistically significant ligand-receptor pairs (p<0.05) were selected. Color represents different tissues and dot size indicates the interaction probability. (B) Bubble plot showing the interaction from arterial i EC and mCAF to T/ NK subsets in MVI+ and MVI- HCC. Statistically significant ligand-receptor pairs (p<0.05) were selected. Color represents different tissues and dot color indicates the interaction probability. (C) Bubble plot showing the interaction from LAMP3 DC to T/ NK subsets in MVI+ and MVI- HCC. Statistically significant ligand-receptor pairs (p<0.05) were selected. Color represents different tissues and dot color indicates the interaction probability. Abbreviations: DC, dendritic cells; mCAF, myo-cancer-associated fibroblasts; MVI+, MVI present; MVI−, MVI absent; T/NK, T cell and NK cell.

Lastly, to validate the spatial co-localization of CLAMT community members, we performed spatial transcriptomics analysis on HCC tissues. Across multiple spots in spatial transcriptomics data obtained from 2 patients, we assessed the infiltration scores of each member of the CLAMT and depicted their spatial infiltration patterns. We observed a consistent co-localization pattern of these five cell types in the same regions within the TME of both studied patients (Supplemental Figure 9, http://links.lww.com/HEP/I90). This observation suggested that these cell subsets are collectively recruited, thereby forming an interactive cellular community within the HCC microenvironment. This collaborative presence, in turn, is likely to exert a synergistic effect on tumor progression.

DISCUSSION

Despite significant advancements in HCC treatment and postoperative monitoring technologies, the overall survival rate of patients with HCC remains below 20%. MVI, recognized as an essential marker of HCC invasion and malignancy, has consistently been associated with HCC recurrence and intrahepatic metastasis.²¹ In this study, we comprehensively characterize the immune, stromal, and tumor components within MVI+ HCC at the single-cell resolution. We identify the distinct molecular features and specific cellular communities associated with MVI+ HCC compared to MVI− HCC. To ensure the robustness of our findings, we used a multi-faceted approach, including scRNA-seq, bulk transcriptomics, spatial transcriptomics, and multiple immunofluorescence labeling.

In this study, we investigated distinct T cell subsets within MVI+ HCC and uncovered an immunosuppressive microenvironment characterized by an elevated presence of cycling T cells and Treg cells. Our findings support the existence of cytotoxicity-exhaustion transitions within CD8+ T cell subsets, with cycling T cells exhibiting notably high terminally exhausted scores, which is consistent with the concept of dysfunctional CD8+ T cells exhibiting extensive proliferative signals. The enhanced proliferative capacity of cycling T cells might represent an adaptive response to antigen stimulation within the TME, potentially compensating for their weakened antitumor responses.^22,23 Furthermore, the persistent exposure to neoantigens likely contributes to the accumulation of cycling T cells and regulator of G protein signaling 1+ exhausted T cell. Cycling T cells showed poor cytotoxicity, and it has been reported that the dysfunction of terminally exhausted T cells may be irreversible.^24,25 Moreover, the enrichment of IL7R Tem cells in MVI+ HCC probably represents a reactive mechanism to enhance rapid recall ability against identical antigens. However, this response appears insufficient to prevent further HCC progression.

Recent studies have highlighted the immunosuppressive and angiogenic roles of TREM2 Macro, particularly in HCC.^17,26 We identified TREM2 Macro as a critical mediator of tumor progression through MDK-related communications, especially the MDK-NCL axis. Other studies have indicated that tumor cells confer a malignant phenotype to other cells through the MDK-NCL axis and alter the TME to promote angiogenesis and cancer metastasis.^27–29 This suggests that malignant cells directly recruit TREM2+ macrophages to the TME, contributing to a more aggressive tumor phenotype. Therefore, targeting the MDK-NCL axis emerges as a promising therapeutic avenue.

We observed that arterial ECs undergo significant phenotypic transformations from NT tissue to MVI+ HCC, displaying extensive tumor-promoting features that indicate their adaptation to the TME, which appears to be driven by the activation of key transcription factors. Among the identified transcription factors, BCL2 associated X promotes apoptosis,³⁰ and FOXC1 is related to EC metabolism.³¹ The presence of these factors highlights their potential role as key regulators of arterial EC polarization, and further investigation is warranted to unravel the molecular mechanisms underlying this process. Importantly, we also found that most mesenchymal cells in HCC were myofibroblasts, which are well-known drivers of invasion and metastasis in the TME. These mCAF cells contribute to tumor progression by producing the ECM surrounding the tumor, facilitating tumor cell proliferation, and modulating angiogenesis.^32,33

Malignant cells from MVI+ HCC exhibited higher proliferative capacity than those from MVI− HCC. In contrast, malignant cells from MVI− HCC were found to shape an inflammatory TME by secreting classical inflammatory factors. This suggests a distinct pro-inflammatory phenotype associated with MVI− HCC. These findings highlight the inter-tumor heterogeneity of HCC, emphasizing the need for tailored therapeutic approaches that consider the specific characteristics of each subtype. Additionally, we identified an MVI+ HCC-specific cellular community, termed CLAMT, associated with poor prognosis in patients with HCC. The subsets in CLAMT exhibited strong correlations with each other in the infiltration abundance and spatial patterns and work together to promote immunosuppressive and tumorigenic processes in MVI+ HCC.

Nevertheless, certain limitations of our study should be acknowledged. Firstly, the sample size of our study was relatively small, including 3 MVI+ and 2 MVI− patients with HCC, which may limit the solid conclusions of our analysis. To overcome this limitation, we incorporated publicly available bulk, single-cell, and spatial transcriptomic datasets coupled with multiplex immunofluorescence assays for validation. However, further large-scale studies are still required to confirm and extend our findings. Secondly, the multicellular community was inferred from transcriptomic data, and although their spatial co-location has been validated using published spatial transcriptomic data, high-dimensional multiplex in situ profiling is needed. Finally, in vitro and in vivo studies would be beneficial for elucidating the underlying molecular mechanisms of MVI formation.

In conclusion, our study comprehensively characterizes the cellular ecosystem and molecular characteristics underlying MVI in HCC. By elucidating the distinct features of MVI+ HCC compared to MVI− HCC, our findings provide valuable insights for improving clinical diagnosis and developing more effective treatment strategies.

Supplementary Material

hep-79-1293-s001.docx ^{(8.5MB, docx)}

hep-79-1293-s002.xlsx ^{(683.8KB, xlsx)}

hep-79-1293-s003.xlsx ^{(13.9KB, xlsx)}

hep-79-1293-s004.xlsx ^{(1.3MB, xlsx)}

hep-79-1293-s005.xlsx ^{(13.9KB, xlsx)}