Distinct immune microenvironment of venous tumor thrombus in hepatocellular carcinoma at single-cell resolution

Abstract

Background and Aims:

Portal vein tumor thrombus (PVTT) worsens the prognosis of hepatocellular carcinoma by increasing intrahepatic dissemination and inducing portal vein hypertension. However, the immune characteristics of PVTT remain unclear. Therefore, this study aims to explore the immune microenvironment in PVTT.

Approach and Results:

Time-of-flight mass cytometry revealed that macrophages and monocytes were the dominant immune cell type in PVTT, with a higher proportion than in primary tumor and blood (54.1% vs. 26.3% and 9.1%, p<0.05). The differentially enriched clustering of inhibitory and regulatory immune cells in PVTT indicated an immune-suppressive environment. According to the single-cell RNA sequencing, TAM-C5AR1 was characterized by leukocyte chemotaxis and was the most common subpopulation in PVTT (36.7%). Multiplex fluorescent immunohistochemistry staining showed that the C5aR⁺ TAM/Mφ were enriched in PVTT compared to both the primary tumor and liver and positively correlated with C5a (r=0.559, p<0.001). Notably, THP-1 (monocyte cell line) was recruited by CSQT2 (PVTT cell line) and exhibited up-regulation of CD163, CD206, and PD-L1 upon stimulation. C5aR antagonist could reverse this. C5aR⁺ TAMs could also inhibit Granzyme B in CD8⁺ T cells. High infiltration of C5aR⁺ TAMs in PVTT correlated with poor differentiation (p<0.009) and was a risk factor for overall survival (p=0.003) and for reformation of PVTT after resection (p=0.007).

Conclusions:

TAMs, especially C5aR⁺ TAMs, were enriched in PVTT. C5aR⁺ TAMs contribute to the development of PVTT and poor prognosis by reshaping the immunosuppressive environment.

INTRODUCTION

Liver cancer is the sixth most commonly diagnosed cancer and the third leading cause of cancer-related death globally.¹ Hepatocellular carcinoma (HCC) comprises 75%–85% of liver cancer cases.² HCC tends to invade vessels and forms the portal vein tumor thrombus (PVTT), which is present in 10%–40% of patients with HCC at the time of diagnosis.^3,4 Once a PVTT forms, the tumor cells keep spreading along the portal vein, leading to extensive intrahepatic or distant metastasis, resulting in a poorer clinical outcome.⁵ The molecular mechanisms underlying PVTT formation are unclear. It is crucial to explore this process to develop effective new treatments for improving the prognosis of patients with HCC with PVTT.

Tumor cells are supported by a complex and heterogeneous ecosystem called tumor microenvironment (TME).⁶ Within TME, immune cells are critical components showing a dynamic and reciprocal relationship with tumor cells. The immunological landscape of HCC is remarkably unique since the liver is an organ of both immunity and immune tolerance.⁷ As to PVTT, previous studies indicated the dysregulation of extracellular matrix and focal adhesion contribute to its formation.^8,9 However, PVTT’s immune components and their contribution are still lacking. A comprehensive dissection of the immune microenvironment would help design more effective therapeutic strategies for HCC with PVTT, especially immunotherapies.

The advances in single-cell technologies, such as single-cell RNA sequencing (scRNA-seq) and cytometry by time-of-flight (CyTOF), offer an excellent opportunity to obtain an in-depth understanding of the constitution and function of the TME in an unbiased fashion.¹⁰ Several studies have profiled the single-cell landscape of immune cells and the tumor cell heterogeneity in primary HCC.^11,12 Our recent scRNA-seq study also revealed a distinctive TME in patients with early-relapse HCC​​​​​.¹³ Characterizing the cell compositions and their interactions within a PVTT at the single-cell resolution might help elucidate pathogenic mechanisms and facilitate the development of effective therapeutic strategies for the disease.

Here, we used high-resolution and multiomics technologies, including CyTOF, scRNA-seq, bulk transcriptome RNA sequencing, and whole exome sequencing (WES) for matched pairs of the primary tumor (PT) and PVTT from patients with HCC to dissect the TME of PVTT comprehensively. We focused on immune cells primarily enriched in PVTT and identified their recruitment mechanism and contribution to PVTT formation and development.

METHODS

Patients and follow-ups

A total of patients with 249 HCC were enrolled at Zhongshan Hospital Fudan University (Shanghai, China) and the study was approved by the institution’s ethics committee (No. B2019-060R). From May 2019 to February 2020, fresh PVTT, PT, adjacent liver, and peripheral blood samples were prospectively collected from 8 patients with HCC for bulk-RNA sequencing (n=12), WES (n=12), CyTOF (n=18), and scRNA-seq (n=6) (Discovery cohort, Figure 1A). The detailed and pathological information is shown in Supplemental Table S1, http://links.lww.com/HEP/J655. Another 241 patients with HCC treated from January 2012 to December 2017 were retrospectively enrolled, including 59 patients who had available paired PVTT, PT, and adjacent liver with formalin-fixed and paraffin-embedded blocks (validation cohort 1) and 182 patients who had FFPE blocks of formalin-fixed and paraffin-embedded (validation cohort 2).

FIGURE 1.

Distinct characteristics of infiltrating immune cells in PVTT. (A) Flow chart of the study. (B) A two-dimensional t-SNE illustration of the CyTOF data gated (with color coding) based on sample types (upper image) and major immune cell types (lower image) showing the heterogeneous distribution of immune cells. (C–E) The pairwise comparison of immune cell types (C), immune cell lineages (D), and immune response (E) (Wilcoxon signed-rank test, *p<0.05). (F) The trend of naïve T cells, memory T cells, and plasma B cells from Blood to PVTT to PT. (G) An ANOVA was performed for all 91 CyTOF clusters, and the detailed differentially enriched clusters among blood, PT, and PVTT are shown (Friedman test, p<0.05). Abbreviations: CyTOF, time-of-flight mass cytometry; DC, dendritic cell; FFPE, formalin-fixed and paraffin-embedded; GRAN, granulocytes; IHC, immunohistochemistry; MDSC, myeloid-derived suppressor cell; Mono/Macro, monocytes/macrophages; NK, natural killer; PT, primary tumor; PVTT, portal vein tumor thrombus; scRNA, single-cell RNA; TAM, tumor-associated macrophage.

The inclusion criteria of all patients and postoperative surveillance were described in previous reports.¹⁴ Follow-up was terminated on March 31, 2020. The overall survival was defined as the interval between surgery and death due to any cause or the last follow-up date. The time to PVTT relapse was defined as the interval between surgery and the relapse of the tumor thrombus in the portal vein system detected by imaging examination.

Sample collection, preparation of single-cell suspensions, bulk DNA and RNA extraction and sequencing, CyTOF analysis, scRNA-seq, scRNA-seq data preprocessing, ligand-receptor interaction analysis, RNA velocity analysis, immunohistochemistry (IHC) staining and multiplex fluorescent immunohistochemistry (mIHC) staining, cell lines and culture conditions, chemotaxis assay, detection of secreted and membrane protein by flow cytometry, functional analysis of CD8⁺ T cells affected by TAMs, and statistical analysis were described in Supplemental Materials and Methods section, http://links.lww.com/HEP/J656.

RESULTS

The distinctive immune cell landscape in PVTT revealed by CyTOF analysis

Based on bulk transcriptome sequencing data, enrichment analyses of genes upregulated in PVTT (Supplemental Figure S1A, http://links.lww.com/HEP/J657) showed that the biological process of “leukocyte migration” and pathways related to the “phagosome” and “Hematopoietic cell lineage” were significantly enriched (Supplemental Figure S1B, http://links.lww.com/HEP/J657). Moreover, a Tumor IMmune Estimation Resource (TIMER) analysis indicated greater infiltration of myeloid cells, including macrophages and dendritic cells (DC), in PVTT compared to PT (Supplemental Figure S1C, http://links.lww.com/HEP/J657).

To better depict the landscape of infiltrating immune cells in PVTT, we performed large-scale mass cytometry immune profiling of 18 matched samples of blood, PT, and PVTT from 6 patients with HCC using an immune cell-centric antibody panel consisting of 42 markers (Figure 1A, Supplemental Table S2, http://links.lww.com/HEP/J655). After the t-SNE dimension reduction process and the X-shift algorithm were applied, the immune cells from 3 types of samples formed 91 clusters based on the expression of immune markers. These 91 clusters were further classified into 10 major types of immune cells consisting of B cells, CD4⁺ T cells, CD8⁺ T cells, γδT cells, natural killer T (NKT) cells, natural killer (NK) cells, DC, monocytes/macrophages (mono/macro), myeloid-derived suppressor cells (MDSC), granulocytes, and others which could not be defined (Supplemental Figure S2A, B, http://links.lww.com/HEP/J657).

The t-SNE map grouped by sample types and cell types indicated that the distribution of immune cell types in PVTT differed from that in paired PT and blood (Figure 1B). The most dominant cell type in blood was granulocytes, while that in PVTT was Mono/Macro (Supplemental Figure S2B, http://links.lww.com/HEP/J657). PVTT showed considerably higher proportions of Mono/Macro compared to both PT (54.2% vs. 26.3%, p=0.031) and blood (54.2% vs. 9.1%, p=0.031) (Figure 1C). In detail, the c46, c47, and c79 in mono/macro were PVTT-enriched clusters with proportions >70%, which were 75.2%, 95.2%, and 74.2% (Supplemental Figure S2C, http://links.lww.com/HEP/J657). PVTT also had significantly lower proportions of NK cells, B cells, and MDSCs than blood samples, and a lower proportion of B cells than PT (Figure 1C). With regard to the immune lineage, PVTT and blood were enriched with myeloid cells, while the cells in PT were mainly lymphoid cells (Figure 1D). PVTT was more enriched in innate immune cells than PT (75.0% vs. 44.6%, p=0.031) (Figure 1E). We also observed an ascending trend for innate immune cells and a descending trend for adaptive immune cells from blood to PVTT to PT (Figure 1E). In addition, the naive cells decreased, and the differentiated cells increased from blood to PVTT to PT (Figure 1F). These data implied that the immune profiles in PVTT might be in an intermediate status, skewing from the local immune response (PT) to the systemic immune response (blood).

We further delineated the differentially enriched clusters in PVTT. By comparing 91 clusters among 3 groups, 11, 4, and 10 clusters were specifically enriched in blood, PVTT, and PT, respectively (Figure 1G). The distinctive distribution of clusters indicated that the immune landscapes of blood, PVTT, and PT in patients with HCC were heterogeneous. Comparing PVTT with PT, clusters enriched in PVTT (3/91, 3.3%) were TAMs, early-stage MDSCs and neutrophils, which are myeloid cells involved in the innate immune response (Supplemental Figure S2D, http://links.lww.com/HEP/J657); clusters enriched in PT (3/91, 3.3%) were exhausted CD8⁺ T cells, central memory CD4⁺ T cells (CD4⁺ Tcm) and plasma B cells, which are lymphoid cells involved in the adaptive immune response (Supplemental Figure S2D, http://links.lww.com/HEP/J657). Comparing PVTT with blood, clusters enriched in PVTT (18/91, 19.8%) included TAMs, DC, plasma B cells, regulatory T cells (Tregs), and regulatory NK cells, indicating the immunoregulatory functions in PVTT (Supplemental Figure S2E, http://links.lww.com/HEP/J657); clusters enriched in blood (13/91, 14.3%) included naive B cells, naïve CD4⁺ T cells, effector CD8⁺ T cells, cytotoxic NK cells, early-stage MDSCs, and neutrophils, which are mainly naive/early-stage cells and cytotoxic/effector cells (Supplemental Figure S2E, http://links.lww.com/HEP/J657). The scale of differentially enriched clusters was lower in the PVTT-PT comparison than in the PVTT-blood comparison (6.6% vs. 34.1%, Supplemental Figure S2D, E, http://links.lww.com/HEP/J657), indicating that PVTT’s immune environment resembled PT more closely than blood. We also found that TAMs constituted the majority (3/4, 75%, Figure 1G) of PVTT-enriched clusters, but these cells were absent in the blood and PT-enriched clusters (Figure 1G and Supplemental Figure S2D, E, http://links.lww.com/HEP/J657), suggesting that TAM clusters play a role in reshaping the local TME of PVTT.

The immunosuppressive role and heterogeneity of TAMs in patients with PVTT

Since TAMs were distinctively enriched in PVTT, we further explored the role of TAMs in the TME of PVTT. Correlation analyses between TAMs and other immune cells were performed. As a result, seven immune cell subtypes significantly correlated with TAMs, including a negative correlation with cytotoxic NK cells (r=−0.60, p=0.001) and a positive correlation with Tregs (r=0.66, p<0.001), exhausted CD8⁺ T cells (r=0.61, p<0.001), and regulatory NK cells (r=0.59, p=0.001), further suggesting that TAMs play a critical role in constructing the immunosuppressive TME of PVTT (Figure 2A).

FIGURE 2.

Immunosuppressive role and heterogeneity of TAMs. (A) The correlation analysis of immune cell subtypes (Spearman test, blue-to-red colors represent the r scores from −1 to 1, *: p<0.05) (B) The composition of Mono/Macro clusters colored by sample showed the heterogeneity based on the sample origin. (C) The composition of each sample colored by cluster showed the heterogeneous contribution of TAM subtypes. (D) The correlation analysis of 91 CyTOF clusters in PVTT showed the heterogeneous correlation pattern of the TAM subtypes (Spearman test, p<0.05, each circle represents one cluster, the number of correlated clusters was measured by circle size, and the color of the lines between circles represents the r score, where the color of the circle describes the average $\mid$ r $\mid$ of correlated lines). Abbreviations: CTLA-4, cytotoxic T-lymphocyte associated protein 4; CyTOF, time-of-flight mass cytometry; DC, dendritic cell; HLA, human leukocyte antigen; MDSC, myeloid-derived suppressor cell; Mono/Macro, monocytes/macrophages; NK, natural killer; PT, primary tumor; PVTT, portal vein tumor thrombus; TAM, tumor-associated macrophage; Treg, regulatory T cell.

The Mono/Macro category consisted of 9 clusters, 8 of which were TAMs, demonstrating the inherent heterogeneity of TAMs. To investigate the phenotypic diversity better, we depicted the features of all clusters in Mono/Macro (CD14⁺ CD15⁻ HLA-DR⁺). First, the expression levels of lineage and functional markers varied among the 9 clusters (Figure 2B). Specifically, lineage markers (HLA-DR and CD163) separated Mono/Macro clusters into HLA-DR^low monocytes (c52), CD163^−​- M1-like TAMs (c48, c50, and c66), and CD163⁺ M2-like TAMs (c45, c46, c47, c78, and c79) (Figure 2B, C). Functional markers such as CD28,¹⁵ PDL-1,¹⁶ Tim-3,¹⁷ and CTLA-4¹⁸ further subdivided M2-like TAM subpopulations and their likely phenotypes (Figure 2B). Second, the distribution of each cluster was different. C45 and c50 ranked as the top 2 among all Mono/Macro classifications (37.8% and 25.7%), and the other clusters accounted for less than 15% (Figure 2B). The proportion of each cluster varied by sample source. In malignant tissue specimens, c45 and c50 remained in the top 2 clusters (Figure 2C). In the blood samples, the top 2 clusters were c50 and c52 (Figure 2C). The proportion of sample sources also varied by cluster. Blood was the primary source for only one cluster (c52), while PT was the primary source for 2 clusters (c45 and c66). Notably, PVTT was the primary source for the other 6 clusters by number, and the difference was statistically significant in four clusters (p=0.016 in c46, p=0.009 in c47, p=0.009 in c50, and p=0.021 in c79, Supplemental Figure S2C, http://links.lww.com/HEP/J657), indicating that PVTT contributed to cluster heterogeneity. Third, different Mono/Macro clusters contributed differently to reshaping the TME of PVTT. Correlation analyses were performed on all 91 clusters in PVTT. The results showed that TAM was the immune cell type mostly related to other clusters (Supplemental Figure S3A, http://links.lww.com/HEP/J657). The 9 Mono/Macro clusters were scattered around the correlation network (|R|>0.8, p<0.05, Figure 2D) with different related clusters (Figure 2D, Supplemental Figure S3B, http://links.lww.com/HEP/J657), implying their heterogeneous relationship in the TME interaction network. For example, c78 was negatively correlated with clusters of cytotoxic effector cells and positively associated with clusters representing immunosuppressive cells, which help reshape the immunosuppressive microenvironment (Supplemental Figure S3C, http://links.lww.com/HEP/J657).

Phenotyping and functional classification of TAM subpopulations in PVTT by scRNA-seq

To further depict the transcriptional profile of the heterogeneous TAM subpopulations infiltrated in PVTT, we collected PT and paired PVTT tissues from 3 patients with HCC for scRNA-seq analysis. PCA and a UMPA analysis identified 25 clusters (Figure 3A, Supplemental Figure S4B, http://links.lww.com/HEP/J657), which could be further classified into 15 subtypes based on the differentially expressed genes (Figure 3B, Table S3, http://links.lww.com/HEP/J655). Immune cells (78.0%) accounted for the majority of the successfully sequenced cells, followed by cancer cells (17.1%) and stromal cells (4.5%) (Figure 3C). We observed that the proportion of total immune cells was similar between PVTT and PT (77.1% vs. 79.0%, Figure 3C), the proportion of T cells was slightly lower in PVTT (fold-change: 0.86, Figure 3D), and the ratio of macrophages/monocytes and DC was higher in PVTT (fold-change: 1.94 and 1.86, respectively, Figure 3D).

FIGURE 3.

Identification of cell types in PVTT and paired HCC by scRNA-seq. (A) The clustering of scRNA-seq data identified 26 clusters. (B) The single-cell populations were projected by UMAP and nominated into 15 cell types. (C) The distribution of immune cells, HCC cells, stromal cells, and erythrocytes was similar between PVTT and PT. (D) The distribution of immune cell types showed a tendency for a higher percentage of macrophage/monocyte cells in PVTT. (E) The myeloid cell population was further grouped into 5 TAM clusters, 2 monocyte clusters, 3 DC clusters, and 2 cycling MP clusters. (F) Biological functions of TAM clusters analyzed by GO enrichment analysis. Abbreviations: DC, dendritic cell; GO, Gene Ontology; HLA, human leukocyte antigen; Mono/Macro, monocytes/macrophages; MHC, major histocompatibility complex; NK, natural killer; PT, primary tumor; PVTT, portal vein tumor thrombus; scRNA, single-cell RNA; TAM, tumor-associated macrophage.

Consistent with the findings of the CyTOF analysis, the scRNA-seq data showed that immune cells in the mononuclear phagocytic system (MP) enriched in PVTT. To further depict the characteristic of the transcriptional profiles for the enriched TAMs in PVTT, cells of myeloid lineage (Macro/Mono, cycling MP, and DC) were re-clustered into 12 myeloid subtypes including 5 TAM subpopulations, 2 monocyte subtypes, 3 DC subtypes and 2 cycling MP subtypes (Figure 3E, Supplemental Figure S4C, http://links.lww.com/HEP/J657). TAM subpopulations comprised 63.7% of the total myeloid cells (Supplemental Figure S4D, http://links.lww.com/HEP/J657).

The functional heterogeneity among different TAM subpopulations was further confirmed by Gene Ontology enrichment analysis (Figure 3F). Genes involved in proinflammatory biological processes, such as the “type I interferon signaling pathway,” “interferon-γ-mediated signaling pathway,” and “T-cell activation,” were explicitly enriched in TAM-c1-CXCL10. The TAM-c2-C5AR1 showed an immune regulatory phenotype since they harbored higher expression of genes associated with biological process, such as “leukocyte chemotaxis,” “positive regulation of cytokine production,” and “myeloid cell differentiation.” The TAM-c3-C1QA was associated with conventional “macrophage” functions in view of its enrichment in “phagocytosis” and “antigen processing and presentation.” The TAM-c4-CSTB was characterized by higher expression of genes associated with “extracellular structure organization” and “cholesterol storage,” indicating a trophic phenotype. The TAM-c5-MT1H was distinguished from other subpopulations for its involvement in the “response to metal ions.”

Using common markers, we also performed correlation analyses between CyTOF clusters and scRNA subpopulations. CyTOF clusters and scRNA subpopulations did not correspond one-to-one (Supplemental Figure S3D, http://links.lww.com/HEP/J657), demonstrating that they could not replace each other. In this context, gene expression-based exploration is not a replacement, but a valuable complement to the lineage protein marker-based information to provide a comprehensive depiction of TAM’s phenotype and function within the TME.

The distinctive enrichment of C5aR⁺ TAMs in PVTT and the mechanism underlying their recruitment

The scRNA-seq data showed that TAM-c2-C5AR1 emerged as the subpopulation with the highest proportion among the TAMs (Figure 4A, Supplemental Figure S4E, http://links.lww.com/HEP/J657). Moreover, PVTT showed a greater increase in TAM-c2-C5AR1 compared with other subtypes (Figure 4A). By performing the bulk deconvolution method using an independent data set (GSE77509), TAM-c2-C5AR1 still showed a higher proportion in PVTT than in PT (p=0.017, Supplemental Figure S4F, http://links.lww.com/HEP/J657). Apart from GO analyses (Figure 3F), our single-cell transcriptome-wide analysis also revealed that the enriched genes in TAM-c2-C5AR1 were related to TGF-β signaling (by Hallmark) and NF-κB signaling pathway (by Kyoto Encyclopedia of Genes and Genomes [KEGG]), which maintained the immunosuppressive phenotype of TAMs.¹⁹ These findings led us to hypothesize that C5aR⁺ TAMs might play a role in reshaping the immunosuppressive TME in PVTT.

FIGURE 4.

Monocytes and TAMs were recruited by PVTT through the C5a-C5aR interaction. (A) Proportions of five TAM clusters and their changes in PVTT. (B) The mIHC staining assays (left) and the comparison of C5aR⁺ TAMs proportion (middle) and density (right) among PVTT, PT, and liver tissue. (C) The screening strategy of potential ligand-receptor pairs between HCC cells and TAM-c2-C5AR1 cells. (D) The correlation analysis between C5a⁺ HCC cells and C5aR⁺ TAMs according to IHC/mIHC staining assays (Spearman test, r=0.559, p<0.001). (E) The comparison of C5a level in the conditioned medium from a PVTT cell line (CSQT2), 3 metastatic HCC cell lines (LM3, 97H, and HLF), and negative control (t test, 9 replicates). (F) The chemotaxis assays of a monocyte cell line (THP-1) in conditioned medium from CSQT2 cells with or without a C5aR antagonist (PMX-53) (t test, 7 replicates). (G) The morphological assessment of THP-1 monocytes cultured with conditioned medium from CSQT2, LM3, 97H, and HLF cells (t test, 7 replicates). (H) The comparison of CD163 and CD206 (markers of M2-like macrophages) on THP-1 monocytes when cultured with different conditioned medium (t test, 6 replicates). (I) The effect of C5a concentration on the expression of CD163 and CD206. (J) The comparison of M2-like phenotype on THP-1 co-cultured in CSQT2 with or without C5aR antagonist (t test, 6 replicates). (*p<0.05, **p<0.01). Abbreviations: CM, conditioned media; IHC, immunohistochemistry; mIHC, multiplex fluorescent immunohistochemistry; MFI, mean fluorescence intensity; PT, primary tumor; PVTT, portal vein tumor thrombus.

An RNA velocity analysis embedded on a UMAP map to infer the fates of cells in PVTT based on the scRNA-seq data from TAMs was performed. The directional flow from Mono-c1-CD14 to TAM-c2-C5AR1 implied that TAM-c2-C5AR1 originated from recruited monocytes (Supplemental Figure S5A, http://links.lww.com/HEP/J657). By performing multiplex fluorescent immunohistochemistry assays in validation cohort 1, we confirmed that more C5aR⁺ TAM/Mφ (DAPI⁺ CD68⁺ C5aR⁺) infiltrated in PVTT than in PT (803.2±171.7 vs. 382.0±57.3/mm², p=0.001) and adjacent liver tissue (803.2±171.7 vs. 155.3±25.2/mm², p<0.001) (Figure 4B).

We then sought to determine the mechanism driving the recruitment of C5aR⁺ TAMs infiltration in PVTT. The interaction analyses indicated that the cross talk between malignant cells and TAM-c2-C5AR1 cells was mediated by C5-C5AR1. Out of 150 significant interacting pairs, 21 were shared by all patients (Figure 4C). To pinpoint the key pairing for TAM recruitment, we examined the overlap of the 35 genes of 21 shared pairs (scRNA-seq, Figure 4C) and 42 PVTT-upregulated genes in the “leukocyte migration” (bulk RNA-seq, Supplemental Figure S1B, http://links.lww.com/HEP/J657). Three interacting pairs were identified: C5-C5AR1, RPS19-C5AR1, and CCL3L1-DPP4. The CCL3L1-DPP4 pair was excluded as its ligand gene (CCL3L1) was mainly expressed in TAM rather than malignant cells (Figure 4C, Supplemental Table S4, http://links.lww.com/HEP/J655); C5-C5AR1 and RPS19-C5AR1 pairs were retained as their ligand genes (C5 and RPS19) were expressed by malignant cells and the receptor gene (C5AR1) was expressed by TAM (Figure 4C, Supplemental Table S4, http://links.lww.com/HEP/J655). Further analysis showed that C5AR1 was expressed explicitly on myeloid cells, especially on TAM-c2-C5AR1 (Supplemental Figure S4G, H, http://links.lww.com/HEP/J657), and C5 was specifically expressed on malignant cells (Supplemental Figure S4I, http://links.lww.com/HEP/J657). In contrast, RPS19 was universally expressed in almost all cell types (Supplemental Figure S4J, http://links.lww.com/HEP/J657). Thus, the C5-C5AR1 interaction might be the key pair for the recruitment of TAM-c2-C5AR1 cells by malignant cells in PVTT. IHC and mIHC assays also confirmed that malignant cells stained with C5a spatially located close to C5aR⁺ TAMs, and the staining intensity of C5a was positively correlated with the density of C5aR⁺ TAMs in PVTT (n=59, r=0.559, p<0.001, Figure 4D) rather than that in PT (n=59, r=−0.016, p=0.904, Supplemental Figure S5C, http://links.lww.com/HEP/J657).

The role of PVTT malignant cells in recruiting monocytes was further confirmed by in vitro and in vivo assays. By using conditioned media (CM) from four HCC cell lines (CSQT2, originated from PVTT; HCCLM3, originated from lung metastasis; MHCC97H, originated from highly metastatic primary lesion; and HLF, originated from nondifferentiated hepatoma), we found that the concentration of C5a in the CSQT2-CM was significantly higher than that of HCCLM3-CM, MHCC97H-CM, and HLF-CM, implying that PVTT cells could secret more C5a than non-PVTT cells (p<0.05, Figure 4E). When THP-1 cells were cultured with CM, the cell migration assays also indicated that the chemotactic effects of CSQT2-CM on THP-1 could be attenuated by PMX-53, the C5aR inhibitor (p=0.023) (Figure 4F). The xenograft model with subcutaneously injected cell lines was used for in vivo validation. Flow cytometry and IHC of subcutaneous tumor both showed that the PVTT cell line could recruit more TAMs and C5aR⁺ TAMs than the HCC cell line (Supplemental Figure S6A, C, http://links.lww.com/HEP/J657).

We further explored the role of malignant cells in inducing monocyte-to-macrophage polarization after recruitment by culturing THP-1 with CM. We found that THP-1, which initially had a round shape, extended more pseudopods in the CSQT2-CM, indicating a morphological change from monocyte to macrophage²⁰ (Figure 4G). Meanwhile, THP-1 showed higher expression of CD163 and CD206, markers of M2 macrophages, when cultured in the CSQT2-CM (Figure 4H, Supplemental Figure S5D, http://links.lww.com/HEP/J657). Along with the monocyte-to-macrophage transformation, C5AR1 was upregulated in THP-1 cultured with CSQT2-CM. More importantly, we found that C5a could increase the expression of both CD163 and CD206 in THP-1 in a concentration-dependent manner (Figure 4I, Supplemental Figure S5E, http://links.lww.com/HEP/J657), and the upregulation of CD163 and CD206 induced by the CSQT2-CM could be attenuated by the C5aR antagonist (Figure 4H, J, Supplemental Figure S5F, http://links.lww.com/HEP/J657). Furthermore, the xenograft model also showed that TAMs recruited by CSQT2 had higher expression of CD163, CD206, and Arg-1 than those recruited by LM3 (Figure S6B, S6C, http://links.lww.com/HEP/J657).

The C5aR⁺ TAMs reshaped the local immune-suppressive environment and attenuated the cytotoxicity of CD8⁺ T cells in PVTT

Since C5aR⁺ TAMs exhibited an M2-like phenotype, a well-known immune-suppressive phenotype, we further explored the role of C5aR⁺ TAMs in reshaping the local immune-suppressive environment in PVTT. After culturing with tumor-infiltrating C5aR⁺/C5aR^- TAMs (Figure 5A), peripheral CD8⁺ T cells co-cultured with C5aR⁺ TAMs expressed less Granzyme B than those co-cultured with C5aR^−​ TAMs (p=0.039, Figure 5B; Patients clinical feature, Supplemental Table S5, http://links.lww.com/HEP/J655). Furthermore, mining TCGA HCC data showed that the distinctive signature of TAM-c2-C5AR1 in HCC was positively correlated with exhausted T-cell infiltration (Figure 5C). Our mIHC data also confirmed that C5aR⁺ TAMs correlated with exhausted CD8⁺ T cells in PVTT (Supplemental Figure S7A, http://links.lww.com/HEP/J657). To explore the mechanism by which C5aR⁺ TAMs inhibit the cytotoxicity of CD8⁺ T cells, we reanalyzed the scRNA-seq data for TAM-c2-C5AR1 and T-CD8-c1-GZMK. By screening for the differentially expressed genes, we found that genes encoding immunosuppressive cytokines,²¹ such as IL1B, IL-10, and PTGS2, were significantly upregulated in TAM-c2-C5AR1. An interaction analysis between TAM-c2-C5AR1 and T-CD8-c1-GZMK revealed 3 pairs, CCL4L2-VSIR, CD274-PDCD1, and PDCD1LG2-PDCD1, containing key immune checkpoint molecules.^22,23 Notably, these pairs were more significant in PVTT than in PT (Figure 5D). Meanwhile, flow cytometry analyses using primary cells isolated from tumoral tissues confirmed that C5aR⁺ TAMs have higher PD-L1 expression than C5aR^−​ TAMs (p=0.016, Figure 5E, S5G). Additionally, concentration-dependent upregulation of PD-L1 was observed in THP-1 cells following treatment with recombinant C5a protein (Figure 5F, Supplemental Figure S5E, http://links.lww.com/HEP/J657), and a C5aR antagonist could restore the expression of PD-L1 in THP-1 exposed to CSQT2-CM (p=0.001, Figure 5G). In vivo experiments also showed that PD-L1 was more expressed in TAMs recruited by CSQT2 than those recruited by LM3 (Supplemental Figure S6D, http://links.lww.com/HEP/J657).

FIGURE 5.

Function and clinical significance of C5aR⁺ TAMs. (A) The flow cytometry gating and sorting scheme for primary C5aR⁺ TAMs. (B) CD8⁺ T cells co-cultured with C5aR⁺ TAMs expressed a higher level of granzyme B than those co-cultured with C5aR^- TAMs. (C) The TCGA HCC data showed that the gene signature of exhausted T cells was positively correlated with the gene signature of TAM-c2-C5AR1. (D) The interaction pair analyses of suppressive immune checkpoint between TAM-c2-C5AR1 and T-CD8-c1-GZMK showed 3 significant ligand-receptor pairs. (E) PD-L1 was expressed more often by C5aR⁺ TAMs than C5aR^- TAMs. (F) Recombinant C5a increased the level of PD-L1 in a dose-dependent manner. (G) The increase in PD-L1 expression could be reversed by treating cells with a C5aR antagonist. (H, I) The log-rank survival analysis showed that patients with high enrichment of C5aR⁺ TAMs had a poorer overall survival (H) and shorter time to PVTT relapse (I) than those with low enrichment of C5aR⁺ TAMs. (Wilcoxon signed-rank test, *p<0.05, **p<0.01). Abbreviations: MFI, mean fluorescence intensity; PD-L1, programmed cell death 1 ligand 1; PMX-53, AcPhe(ornithine-Pro-cyclohexylamine-Trp-Arg); PVTT, portal vein tumor thrombus; TAM, tumor-associated macrophage; TCGA, The Cancer Genome Atlas Program.

A retrospective cohort of 182 patients with HCC with PVTT was enrolled to explore the clinical value of C5aR⁺ TAMs. A Kaplan-Meier analysis showed that patients with low enrichment of C5aR⁺ TAMs in PVTT survived longer than those with highly enriched C5aR⁺ TAMs (median time, 17.1 vs. 9.2 months, Figure 5H). A multivariate analysis revealed that the high enrichment of C5aR⁺ TAMs in PVTT (HR, 1.69; 95% CI: 1.18–2.42; p=0.005) was an independent risk factor for the overall survival (Table 1). Furthermore, we found that the tumor thrombus in the portal vein relapsed in 46.2% (84/182) of the overall patients after resection, and those with high enrichment of C5aR⁺ TAMs relapsed significantly earlier than those with low enrichment (median time, 9.1 mo vs. not reached, p=0.007, Figure 5I). In addition, C5aR⁺ TAMs in PVTT were associated with poor differentiation (p<0.001, Supplemental Table S6, http://links.lww.com/HEP/J655).

TABLE 1.

Univariate and multivariate analysis of patients with HCC combined with PVTT

Variable	HR (95% CI)	p
	Univariate analysis
Male	1.12 (0.63–1.98)	0.707
Age ≥60 y	0.84 (0.54–1.31)	0.447
Largest diameter>5 cm	1.53 (0.93–2.49)	0.091
Multiple tumors	0.95 (0.67–1.35)	0.764
Incomplete tumor capsule	1.28 (0.74–2.24)	0.379
Edmondson grades III-IV	1.41 (0.96–2.08)	0.082
PVTT of type III	1.82 (1.24–2.67)	0.002*
Liver cirrhosis	0.99 (0.69–1.43)	0.974
Positive HBsAg	1.02 (0.64–1.63)	0.936
Detectable HBV DNA	1.28 (0.89–1.83)	0.179
Positive HCV-Ab	0.40 (0.06–2.86)	0.360
AFP>400 μg/L	1.53 (1.06–2.22)	0.023*
TB>20.4 μmol/L	0.90 (0.44–1.84)	0.768
Albumin<35 g/L	1.94 (0.90–4.18)	0.089
ALT>50 U/L	0.89 (0.60–1.31)	0.556
γ-GT>60 U/L	1.64 (1.00–2.67)	0.049*
Child-Pugh B class	29.66 (3.57–246.40)	0.002*
Adjuvant treatment	0.84 (0.59–1.20)	0.335
Highly enriched C5aR+ TAM	1.74 (1.21–2.49)	0.002*
	Multivariate analysis
PVTT of type III	1.87 (1.27–2.76)	0.001*
AFP>400 μg/L	1.54 (1.06–2.24)	0.022*
γ-GT>60 U/L	1.64 (1.00–2.69)	0.049*
Child-Pugh B class	12.27 (1.45–104.20)	0.022*
Highly enriched C5aR+ TAM	1.69 (1.18–2.42)	0.005*

Note: Analyses were performed using the chi-squared test.

^* p<0.05.

Abbreviations: γ-GT, gamma-glutamyltransferase; AFP, alpha-fetoprotein; PVTT, portal vein tumor thrombosis; TAM, tumor-associated macrophage; TB, total bilirubin.

The immunogenic and chemoattractant features of malignant cells in PVTT

Based on the bulk WES data, we explored the immunogenic features of PVTT, including microsatellite instability (MSI), the tumor mutation burden (TMB), and neoantigen expression. None of the 12 malignant samples exhibited MSI. The numbers (2716.8 vs. 2739.7, p=0.812) and proportions (0.50% vs. 0.46%, p=0.688) of mutated microsatellite loci were similar between PT and PVTT Supplemental Figure S8A, http://links.lww.com/HEP/J657). No difference in the TMB was found between PT and PVTT (126.7/Mb vs. 110.2/Mb, p=0.331, Supplemental Figure S8B, http://links.lww.com/HEP/J657). PVTT also had a similar predicted neoantigen load to PT (105.7 vs. 117.5, p=0.454, Supplemental Figure S8C, http://links.lww.com/HEP/J657). Thus, PVTT did not show any significant differences in tumor immunogenicity compared to PT.

The expression profile of malignant cells was further explored using scRNA-seq data. We observed an enrichment of genes involved in immune response pathways (eg, “TNF-α signaling via NF-kB”) and hypoxia-related pathways (eg, “hypoxia” and “reactive oxygen species pathway”) in PVTT. In contrast, the genes upregulated in PT belonged to metabolism-related pathways (eg, “fatty acid metabolism,” “adipogenesis,” and “cholesterol homeostasis”) (Figure 6A). By comparing gene signatures, we observed that PVTT samples presented higher hypoxia, proliferation, stemness, and drug resistance signatures, while PT samples had higher metabolic function signals. However, we observed no significant difference between PVTT and PT regarding the signature of immune escape-related genes (Figure 6B). Apart from comparing malignant cells between PVTT and PT, we also used external data (TCGA-LIHC) to compare PTs with or without PVTT (vascular tumor cell type: macro vs. none). The gene enrichment analyses showed that “hypoxia,” “inflammatory response,” and “complement” were upregulated Hallmarks in PTs with PVTT.

FIGURE 6.

Characteristics of malignant cells in PVTT. (A) The differential characteristics of PVTT and PT malignant cells according to the Hallmark enrichment analyses of scRNA-seq data. (B) A comparison of gene signatures of malignant cells in PVTT and PT. (C) The expression of HABP2, a hypoxia-related gene, was upregulated in malignant cells from PVTT. (D) A schematic diagram showing how PVTT tumor cells educated C5aR⁺ TAMs and contributed to the immunosuppressive microenvironment. Abbreviations: FASP, factor VII-activating protease; PD-L1, programmed cell death 1 ligand 1; PT, primary tumor; PVTT, portal vein tumor thrombus; scRNA, single-cell RNA.

Since hypoxia was a prominent feature of malignant cells in PVTT, we explored the influence of the hypoxic microenvironment. It is worth noting that hypoxia caused by large vessel occlusion increased the level of hyaluronan binding protein 2 (HABP2, also known as factor VII-activating protease [FSAP]),²⁴ which could cleave C5 into C5a.²⁵ Consistent with previous reports, our data showed that malignant cells in PVTT had a higher HABP2 level than PT (p<2.2e−16, Figure 6C). We also found a higher expression of factor VII-activating protease in the malignant area in PVTT than in PT (p=0.001, Figure 6D). Moreover, FASP⁺ cells were positively correlated with C5aR⁺ TAMs in PVTT rather than that in PT (Supplemental Figure S7B, http://links.lww.com/HEP/J657), and FASP⁺ cells were negatively correlated with cytotoxic CD8⁺ T cells in PVTT (although the difference was marginal significant) rather than that in PT (Supplemental Figure S7C, http://links.lww.com/HEP/J657). Thus, although the gene expression of C5 was similar between PVTT and PT, the higher level of serine proteases,²⁶ especially HABP2, could degrade C5 and increase the C5a level, resulting in circulating monocytes recruitment and polarizing into C5aR⁺ TAMs and finally inhibiting cytotoxic CD8⁺ T cells in PVTT (Figure 6E).

DISCUSSION

PVTT is one of the most significant factors underlying the poor prognosis of patients with HCC.²⁷ Surgical resection is possible for some patients, but the clinical outcomes remain poor due to high rates of postoperative relapse and metastasis.²⁸ A complete understanding of the components, their spatial heterogeneities, and the interplay between tumor cells and the local microenvironment in this area will help reveal the mechanism(s) underlying PVTT formation and develop novel therapeutic approaches for this disease. Here, we comprehensively analyzed the genomic, single-cell proteomic, and single-cell transcriptomic atlas to characterize the TME and its spatial heterogeneity in patients with HCC with PVTT. Our study revealed that PVTT differed from PT in its local immune microenvironment rather than in its tumor immunogenicity. The striking enrichment of macrophages, especially C5aR⁺ TAMs, reshaped the local microenvironment of PVTT after immune cells were recruited and polarized by surrounding malignant cells, and this resulted in a local immunosuppressive TME in PVTT which permitted recurrence and metastasis in patients with HCC.

The mechanism(s) of PVTT formation has been raising attention for a long time. Previous studies focused mainly on changes in malignant cells, and despite exploring the differences in DNA,²⁹ mRNA,³⁰ miRNA,³¹ lncRNA,³² and protein, they were insufficient to demonstrate the mechanism conclusively.³³ In recent years, attention has shifted to assessing the contribution of specific immune cells in PT for their role in tumor progression.^34,35 However, the immune landscape of PVTT has not been well depicted. Our present study adds an important layer of understanding of the local TME in PVTT at the single-cell resolution.

Based on our CyTOF data, the spatial heterogeneity of immune landscapes in different sites was identified. The percentage of granulocytes decrease, while the percentages of CD4⁺ T cells and CD8⁺ T cells increased in the order from blood to PVTT to PT. The ratio of innate immune cells and naïve T cells also showed a decreasing trend. Considering these sequential changes and PVTT’s anatomic position, which is located on the boundary between blood and PT, we speculate that PVTT was in the transitional stage, skewing from PT to blood. The number and abundance of immune cell clusters enriched in the PVTT-PT comparison was lower than in the PVTT-blood comparison, which implies that PVTT’s immune environment was more similar to PT than to blood (Supplemental Figure S2D, E, http://links.lww.com/HEP/J657). Furthermore, PVTT enriched more immunoregulatory clusters than blood (c21-Treg, c74-Th22, and c33-regulatory NK, Supplemental Figure S2E, http://links.lww.com/HEP/J657) and enriched more inhibitory myeloid clusters than PT (c50-TAM, c54-early-stage MDSC, and c85-neutrophil, Supplemental Figure S2D, http://links.lww.com/HEP/J657), which indicated an immunosuppressive microenvironment in PVTT.

Consistent with the concept that TAMs are central regulators of the tumor-promoting microenvironment,³⁶ our findings confirmed that TAMs with high heterogeneity play key roles in reshaping the local microenvironment of a PVTT. We found that TAMs were the most enriched immune cells in PVTT, with nearly 2-fold enrichment compared to PT and 6-fold compared to blood. Another study also observed the gathering of macrophages in venous tumor thrombi of renal cell carcinoma by bulk RNA-seq analysis.³⁷ Our CyTOF data showed that TAM subpopulations differed in terms of their marker expression, sample distribution, and component correlation (Figure 2). Furthermore, scRNA-seq analyses revealed that the TAM subpopulations in PVTT presented various features, including proinflammatory, immune regulatory, phagocytotic and antigen-presenting, trophic, and metal ion responsive phenotypes. All of these observations indicated that PVTT-enriched TAMs were heterogeneous and played key roles in remolding the local immunosuppressive environment, contributing to PVTT formation and tumor progression.

Based on our scRNA-seq data, C5aR⁺ TAMs were identified to be the most highly enriched subtype in PVTT. The immunosuppressive functions of the C5aR⁺ TAMs were indicated by the upregulation of CD163, CD206, and PD-L1, by the attenuated CD8⁺ T-cell cytotoxicity in a co-culture system, by upregulated IL-1β,^38,39 IL-10,^40,41 and PTGS2⁴² expression, and by the interaction with CD8⁺ T cells via CD274 and PDCD1 (PD-L1-PD-1). A previous study also reported that C5aR⁺ TAMs had protumorigenic properties, including the suppression of cytotoxic CD8⁺ T cells in squamous carcinoma.⁴³ Moreover, we observed that a higher density of C5aR⁺ TAMs was related to a poorer prognosis and a higher risk of PVTT relapse. These findings indicate that C5aR⁺ TAMs created the immunosuppressive TME through cytokine interactions, further promoting the formation and progression of PVTT. As such, inhibiting/depleting C5aR⁺ TAMs either alone or in combination with a PD-1/PD-L1 inhibitor might be effective in treating PVTT.

According to the gene profile of tumor cells in PVTT, the changes in malignant cells were limited during the development of PVTT. No significant difference in tumor immunogenicity was observed between PVTT and PT due to the remarkable similarity of the mutations present.²⁹ Additionally, the immune escape signature of malignant cells was not significantly different between PVTT and PT, which implied that malignant cells could not escape immune surveillance by themselves (eg, via antigenic loss and modulation, expressing less major histocompatibility complex or co-stimulatory molecules). Instead, malignant cells relied on modulating the TME to promote PVTT formation.

The hypoxic nature of PVTT could upregulate the expression of serine proteases in the local TME, including plasma factor VII activating protease (FSAP, coded by HABP2)²⁴ which cleaved C5 into the C5a secreted by tumor cells.²⁵ Then, C5a recruited circulating monocytes that were transformed into C5aR⁺ TAMs, which promoted the formation and progression of PVTT (Figure 6D). Although the combination of atezolizumab (a PD-L1 antibody) and bevacizumab (a VEGF antibody) has been used as first-line systemic therapy for advanced HCC,⁴⁴ it showed limited efficacy in patients with HCC with PVTT. Considering the rapid progression of PVTT, its induction of liver dysfunction, and its high rate of relapse, PVTT represents a more important challenge for treatment than PT. Since C5aR⁺ TAMs play a key role in establishing the local immunosuppressive TME of PVTT, C5aR⁺ TAMs might be considered an attractive target for patients with HCC with PVTT.

The limitations of this study are as follows. First, the sample size for the CyTOF and scRNA-seq studies was small, and the data set used for PVTT single-cell omics studies needs to be expanded. Second, since no well-established PVTT model animal was available, the in vivo evidence in our findings was weak and required further exploration.

In conclusion, this study provides new information about the immune microenvironment in PVTT in terms of the distribution, phenotype, and cross talk between malignant and immune cells. Our data could also be a resource to facilitate a deeper understanding of the venous metastatic process and may be helpful in developing novel therapeutic approaches for patients with HCC with PVTT.

Supplementary Material

hep-82-0566-s001.docx ^{(37.7KB, docx)}

hep-82-0566-s002.docx ^{(51.5KB, docx)}

hep-82-0566-s003.pdf ^{(693.1KB, pdf)}