Introduction
Liver cancer is the seventh most common and second most lethal cancer globally, hepatocellular carcinoma (HCC) accounts for 75–85% of liver cancer cases.1 Cirrhosis is an independent risk factor and has prognostic implication for HCC,2 regular surveillance is recommended to cirrhotic patients for the early screening of HCC. The progression of cirrhosis involves telomere dysfunction, cellular senescence, inflammatory responses, and Epithelial-mesenchymal transition. At molecular level, aberrant modulations of the Wnt, RAS, JAK/STAT, and p53 pathways were frequently observed.3 Previous studies identified differentially expressed genes (DEGs) by comparing sequencing data acquired from cirrhotic patients and HCC patients. They were enriched in pathways for cell-cycle, DNA replication, drug metabolism, and p53 signaling.4,5 It was also reported that the mutation landscape of cirrhotic patients differed from that of HCC patients, and that mutations at key loci including the TERT promoter were indicators of malignant transformation.6 The transformations at genomic and transcriptomic level can be used to predict the transition from cirrhosis to HCC. Moreover, changes in the local immune landscape were also expected during this malignant transformation, which requires further investigation.
Single-cell RNA sequencing (scRNAseq) has characterized the immune microenvironment of healthy and diseased liver. Several subtypes of T cells, B cells, and unknown subtypes of endothelial cells, Kupffer cells, and hepatocytes were identified in healthy liver samples.7 Apart from conventional tumor-associated macrophages, a unique cluster of LAMP3+ dendritic cells (DCs) were identified in immune-related ligands of HCC patients, and were proposed to regulate various lymphocytes.8 Changes in infiltrating immune cells and their regulations during the transformation from healthy liver cells to cells with chronic inflammation and eventually HCCs might determine the response to immunotherapy. Therefore, the aim of the study is to investigate the change of immune cell composition from healthy liver, cirrhotic liver to HCC, and interpret the clinical impacts of the altered immune microenvironment.
Results
Immune cell composition of cirrhotic samples resembles that of healthy but not HCC samples
We downloaded scRNAseq data that composed of the RNA expression matrices of 142600 cells extracted from 41 individuals in six studies (GSE115469, GSE136103, GSE124395, HRA000069, GSE107747, SRP073767; Table 1). The cells clustered to 20 discrete cell populations that were characterized by differential gene expression (Fig. 5A–C and Supplementary Fig. 1). These cell clusters were annotated as immune cells, hepatocytes, stromal cells (endothelial cells and fibroblasts), and mixed clusters based on the expression of gene markers (Fig. 5D, E). Considering that some of the included studies conducted flow cytometry to select immune cells, the immune cell populations are more prevalent in the integrated dataset. Among immune cells, T cells, natural killer (NK) cells, neutrophils and macrophages are relatively abundant (Fig. 5F). The cluster T cell, T cell 2, and naïve T cell account for 1.57%, 9.09% and 6.38% of the total cell count. NK and NK 2 cluster attributed to 7.41% and 5.52% of the total count respectively. 8.7% cells are macrophages and monocytes. Neutrophil, one of the most abundant types of white blood cells in the blood stream, made up 8.08% of the entire population.
Healthy liver, cirrhotic liver and HCC had varied immune cell composition (Fig. 6A–F). The composition of immune cells in cirrhotic samples was similar to that of the healthy liver samples, with T cells, and NK cells being the prevalent immune cell types. Notably, the immune cell composition of HCC samples was significantly altered compared to the healthy and cirrhotic samples (Fig. 6F). The proportion of NK cells decreased, the abundance of different T cell populations was altered, and neutrophils were evidently more enriched. Comparing blood samples and tissue samples of cirrhosis individuals, neutrophils were dominant in the blood samples as expected, accounting for 14.85% of the cells in the cirrhotic blood sample. In the HCC blood sample, the proportion of neutrophils doubled to 29.05%, indicating an elevated inflammatory response. In healthy blood samples, only B cells, plasma B cells, T cells, NK cells, DCs, macrophages, monocytes, and megakaryocytes were identified (Supplementary Fig. 2). Neutrophils were not identified, possibly due to limitations of the sequencing technique and cell degradation caused by sample handling. Therefore, we excluded healthy blood samples from our downstream analysis.
B cells were more abundant in blood samples of HCC patients compared to cirrhotic patients
We identified two B-cell populations and three T cell populations. CD19-expressing B-cell populations were identified as B cell and plasma B cells based on the genetic markers acquired from published studies,7–9,21–25 public databases,26,27 and SingleR.16 Cluster 12 is a B-cell population with strong expression of MS4A1, CD79A, and BANK1, whereas the expression of DERL3, JCHAIN, and MZB1 was limited. In contrast, cluster 18 is a subset of B cells that highly expressed DERL3, JCHAIN, SDC1, and MZB1, but without the expression of MS4A1 and BANK1. MS4A1 encodes for the B-lymphocyte surface antigen CD20, whereas JCHAIN, and MZB1 coded proteins are functionally related to immunoglobulin heavy and light chains. Given the above, we postulated that cluster 12 comprised antigen-inexperienced B cells, and cluster 18 contained differentiated plasma B cells that resided in the liver tissue.
We also compared the cell fraction of the two B-cell populations in healthy, cirrhotic, and cancerous samples (Fig. 1A, B). Both the total count (Fig. 1A left panel) and statistical comparison based on individual sample counts (Fig. 1A right panel) were displayed. The B-cell population accounted for 12.47±3.42% of total cell counts in HCC blood samples, significantly higher than that of the cirrhotic blood sample (2.04±0.10%). In fact, this B-cell population was also more enriched in the HCC blood sample than in the HCC tissue sample. No difference was observed at tissue level, showing B-cell abundance was modulated for peripheral immunity. The top DEGs were primarily involved in TH1/TH2 differentiation, DCs regulation TH1/TH2 development, and CD40L signaling (Fig. 1C). T cells interacted with the B cells by CD40 signal modulation (Fig. 2D).
Similar to B cells, plasma B cells accounted for 1.48±0.43% of total cells in HCC blood samples, significantly higher than the composition of plasma B cells in cirrhotic blood samples (Fig. 1B). The plasma B cells were not active for T cell-related cellular process. Instead, they were primarily responsible for protein translation and cotranslational protein targeting to membranes (Fig. 1E). Collectively, the observations indicate that T-helper cells, antigen-inexperienced B cells, and differentiated plasma B cells responsible for antibody secretion were more abundant in HCC blood samples.
Naïve T cells were enriched in cirrhotic tissue samples, but were depleted in HCC tissue samples
Three T cell subsets were identified by high PTPRC (protein alias CD45) and CD3 expression. Cluster 7 was likely a naïve T cell and memory T cell subset because of the increased expression of CCR7. The abundance of this T cell subset was significantly lower in HCC tissue samples (1.23±0.41%) compared with healthy (2.95±0.51%) and cirrhotic liver tissues (4.48±1.11%; Fig. 2A). In addition, naïve T cells and memory T cells were significantly more enriched in blood samples than in tissue samples. Those cells were active in protein translation (Fig. 2B).
Expression of CD3E, CD2, CD7, CD69, and CD4 was enriched in cluster 16, which was believed to be another T cell subset. This T cell population was more abundant in HCC tissue than in other tissue samples (Fig. 2C). It also accounted for 2.11±0.22% of cirrhotic blood cells, significantly higher than T cells in HCC blood samples (0.81±0.27%). The subset of T cells was involved in p73 transcription factor network, and actively interacted with NK cells (Fig. 2D). Cluster 2 was a second T cell subset (T cell 2) that strongly expressed CD3E, CD2, CD7, and CD69, the percentage of cells expressing CD8 was greater than in cluster 16. The expression of T cell 2 subset showed no clear tissue type preference (Fig. 2E). Functional enrichment analysis showed that these cells were responsible for NFAT signaling, IL5-mediated signaling, CD40/CD40L signaling, and NF-kappa B pathways (Fig. 2F). Given the above, anti-inflammatory T cells were evenly distributed in all sample types, but naïve and memory T cells were depleted in HCC tissue sample and high in blood samples from HCC patients.
Neutrophils were scarce in cirrhotic tissue samples
Neutrophils account for up to 80% of white blood cells in the circulation and around 35–50% of total lymphocytes in the liver.28 Neutrophils were significantly enriched in cirrhotic blood samples (14.39±2.75%) compared with cirrhotic tissue samples (1.50±0.15%; Fig. 3A). However, the percentages of neutrophils in HCC tissue samples (7.96±1.80%) and HCC blood samples (12.47±4.60%) were similar. In fact, HCC tissues had significantly increased neutrophil infiltration compared with both healthy (3.49±0.64%) and cirrhotic tissue samples, and neutrophil infiltration in cirrhotic tissue samples was decreased. Enrichment analysis of the DE genes revealed neutrophil activation, neutrophil mediated immunity, and TGF-beta signaling functions (Fig. 3B).
A subset of NK cells was enriched in cirrhotic tissue samples compared with peripheral blood samples
Two NK cell subsets were identified. Cluster 5 (NK) strongly expressed CD7, NKG7, XCL1, XCL2, and cluster 10 (NK2) strongly expressed CD7, GNLY, and NKG7. Neither the NK nor the NK2 clusters were differentially distributed in healthy, cirrhotic, and HCC tissue samples (Fig. 3C–E). However, the NK cluster was significantly enriched in tissue samples compared with blood samples but not in the NK2 cluster (Fig. 3E). The NK cell cluster was involved in canonical cancer signaling, NK cell mediated cytotoxicity, and cell adhesion. It actively interacted with B cells, T cells and macrophages via the LIGHT signaling pathway (Fig. 3D). In comparison, the NK2 cluster was enriched in genes involved in the G2/M checkpoint, and was believed to modulate the mixed cell cluster via HLA-CD8 signaling (Fig. 3F). The immune-active NK cell subset was enriched in tissue samples, but the quiescent NK2 cluster displayed relatively equal distribution.
Pathological tissue samples and blood samples were dominated by distinct macrophage subsets
Two populations of macrophages and monocytes were identified by CD68 expression. Cluster 6 highly expressed CD68, CSF1R, MARCO, C1QA, CD14, and C1QB, whereas cluster 15 was marked by CD68, CSF1R, and MACRO expression, with limited expression of C1QA and C1QB. Cluster 6 (macrophage 1) is a subset of macrophages mixed with a few monocytes that actively interact with T cells and B cells. The infiltration of macrophage 1 cells that actively interact with other immune cells was decreased in blood samples compared with tissue samples (Fig. 4A). In addition, the macrophage population was significantly more enriched in cirrhotic blood samples (5.11±0.18%) compared with HCC blood samples (0.97±0.60%). Enrichment analysis of DEGs pointed to T cell activation, Th1/Th2 differentiation, antigen dependent B-cell activation, TCR signaling, and MHC class II antigen presentation (Fig. 4B).
Cluster 15 (macrophage 2) was characterized by strong expression of LST1, MS4A7, CKB, CDKN1C, RHOC, COTL1, FCER1G, FCGR3A, SAT1, FTH1, and LYN. In contrast to macrophage 1, this macrophage population was more enriched in blood samples than in tissue samples (1.77±0.21% vs. 3.44±0.22% for cirrhotic samples, p<0.01; 1.05±0.43% vs. 4.66±1.52% for HCC samples, not significant; Fig. 4C). The DEGs were involved in NF-kB signaling, TNF-related signaling and toll-like receptor signaling, macrophage 2 actively interacts with T cells (Fig. 4D). Notably, toll-like receptor signaling is required for immune activation.29 Both macrophage subsets were more prevalent in cirrhotic samples than in healthy and HCC samples.
DCs were enriched in cirrhotic patients with alcoholic liver disease (ALD) and primary biliary cholangitis (PBC)
We identified a DC population and a plasmacytoid DC population based on the expression of FCER1A. Cluster 14 (DC) strongly expressed CD1C, FCER1A, NDRG2, PKIB, CD1E, and CYP2S1, and cluster 19 strongly expressed PACSIN1, DNASE1L3, CLEC4C, LRRC26, SCT, and LAMP5. Unlike other immune cell types, both the DC and plasmacytoid DC populations were not differentially distributed in any sample type (Fig. 7). The expression of genes involved in oxidative stress and the regulation of RHO GTPase was increased in the DC population. The plasmacytoid DC was associated with IL2 signaling events mediated by PIK3.
Next, we investigated the effects of virus infection, tumor staging, or etiology on immune cell infiltration (Fig. 8A). For HCC patients, we grouped the samples by HBV infection status or tumor staging, and compared immune cell infiltration between the groups. To our surprise, none of the differences between the groups of immune cell types were statistically significant (Supplementary Fig. 3). We then grouped liver cirrhosis patients by their etiologies. Interestingly, DCs were relatively more enriched in patients with ALD and PBC than in patients with nonalcoholic fatty liver disease (NAFLD) or hereditary haemochromatosis (p=0.086), but the differences were not statistically significant (Fig. 8B). Similarly, healthy individuals with or without Epstein-Barr virus (EBV) or cytomegalovirus (CMV) were compared for changes in their immune landscapes. None of the differences in immune cell types in the EBV positive/negative groups, or CMV positive/negative groups were significant (Supplementary Fig. 4). However, it should be noted that the lack of statistical significance might be the result of the small sample size.
Discussion
The liver is an immunologically active organ that harbors a diverse immune cell repertoire.30 The balance between immune-tolerance and inflammatory response is crucial to liver function. Immune deficiency may lead to the development of chronic liver disease and cancer. Interpreting the immune microenvironments of healthy and pathological liver provides insights into the change of immune status caused by the diseases, and might guide the design of clinical treatment plans in the future. By integrating and re-analyzing public scRNAseq data, we compared the immune microenvironment of healthy, cirrhotic liver, and HCC tumor, as well as in cirrhotic and HCC peripheral blood. Various types of immune cells, hepatocytes, endothelial cells, and fibroblasts were identified in healthy liver as previously described.7 T cells, NK cells, and macrophages were the most abundant immune cell populations in healthy liver. Interestingly, the abundance of each immune cell type in cirrhotic liver was similar to that of the healthy liver, but the abundance of almost all immune cell types changed greatly in HCC. That observation seems to suggest that the immune microenvironment of cirrhotic liver was not substantially altered, whereas tumorigenesis was strongly associated with the immune microenvironment of the liver. This is to our surprise considering that changes in liver resident microbiome were observed in liver cirrhosis.31
In cirrhotic liver, the infiltration of naïve T cells and a subset of macrophages that actively interacts with T cells and B cells were increased. Hepatic T cells exhibited remarkable heterogeneity.12 Consistently, we identified three T cell subsets, and naïve T cells accumulated in cirrhotic samples. That cluster of cells was positive for both CD4 and CD8, although the expression was relatively low. They displayed poor correlation with cytokine production, as previously described.32 One study reported that naïve T cells were reduced in liver cirrhosis.33 However, our study showed that naïve T cells were significantly enriched in cirrhotic tissue samples, but were decreased in HCC samples, indicating a less proinflammatory T cell repertoire in cirrhotic liver.
Two macrophage subsets were identified in this study, however, they had distinct features and cannot be classified as M1 or M2 macrophages based on the expression of gene markers. The CD14 low macrophage 2 was predicted to interact with T cells and B cells, whereas the CD14 high macrophage 1 highly expressed genes for TNF and toll-like receptor signaling. It was postulated that TNF expressing macrophages may interact with T cells and promote T-cell exhaustion, therefore impair antitumor responses and facilitating HCC development.34 That macrophage subtype was more prevalent in tissue samples than in blood samples, showing that T-cell exhaustion was more evident in local immunity than in peripheral immunity. In contrast, enrichment analysis showed that macrophage 2 up-regulated genes that promote T cell activation, B cell activation, TCR activation, and Th1/Th2 differentiation. Moreover, IL-10 expression was significantly up-regulated in macrophage 2, suggesting an anti-inflammatory role of those cells.35 Interestingly, the nontypical macrophage subset was significantly increased in cirrhotic samples compared with healthy tissue samples and their abundance was decreased in HCC samples. Given the above, the anti-inflammatory, T cell/B cell interacting macrophage subsets may play a key role in the development of liver cirrhosis.
In contrast, neutrophil infiltration was limited in cirrhotic livers. Neutrophil infiltration, which is required for the anti-microbial immune responses, was observed in varies types of liver disease. However, it was reported that neutrophils were not necessary for establishing chronic inflammation and hepatic fibrosis.36 Moreover, neutrophils may amplify liver damage. Neutrophil depletion significantly reduced liver damage in an acute liver inflammation mouse model.37 Neutrophil extracellular traps were shown to disturb hepatic blood flow, which results in mild hepatic injury.38 Whether neutrophil infiltration promotes liver cirrhosis is under debate, but a study showed that the neutrophil-to-lymphocyte ratio would predict the survival of patients with liver cirrhosis.39 The mean neutrophil-to-lymphocyte ratio was lower in the survivor group, suggesting that low neutrophil content may benefit patient survival.
In HCC tumors, neutrophil infiltration was significantly increased, which suggests a different role of neutrophil in HCC. Increased neutrophil infiltration has been observed in various cancers, and promotes or suppresses tumorigenesis depending on the local immune microenvironment.40 Neutrophils were shown to promote cancer cell adhesion to hepatic sinusoids and facilitate cancer invasion.41 In fact, functional enrichment analysis of the neutrophil cluster identified in our study confirmed their role in promoting cell adhesion. Therefore, neutrophils may be protumorigenic in HCC, and the net neutrophil complement increase may be an indicator of HCC emergence in cirrhotic patients. On the other hand, the proportion of naïve T cells was decreased. Decrease of the quiescent and immune-inactive T cell population was accompanied by the increase of CD4+ and CD8+ T cells, demonstrating a more immune-active microenvironment compared to cirrhotic and healthy liver.
In clinical practice, collecting cirrhotic liver samples for the prediction of HCC is not practical. Therefore, biomarkers by liquid biopsy should be the focus of future research. Comparing the blood samples and tissue samples, it was evident that the immune cell infiltration of the blood sample cannot represent the microenvironment of the tissue samples. The major difference between the cirrhotic and HCC blood sample was manifested by two B cell subsets, T cells, NK cells, and macrophage 1. However, the differences were not observed in tissue samples, indicating that peripheral immunity was more evidently altered, and such changes in immune cells may serve as indicators of HCC development in patients with cirrhosis. Collectively, the dynamics of the immune cell subsets can be considered as novel biomarkers to predict the transition from cirrhosis to HCC.
The main study limitation is that bioinformatics analysis has limited clinical application. More stringent data needs to be shown in terms of bioinformatics approach to be supported. Therefore, prospective clinical studies are required to further validate the conclusions. The immune landscapes of cirrhosis patients with or without HCC development should also be investigated in future studies.