Hepatic Stellate Cell and Monocyte Interaction Contributes To Poor Prognosis in Hepatocellular Carcinoma

Abstract

Hepatocellular carcinoma (HCC) patients suffer from a poor survival rate and high incidence of post-operative recurrence. The hepatic microenvironment plays a significant role in the initiation, progression and recurrence of HCC; however the causal mechanisms of these phenomena are unclear. Given the predominant underlying fibrotic and cirrhotic conditions of the liver prone to HCC and its recurrence, alterations of components of the inflammatory milieu have been suggested as factors that promote HCC development. In particular, activated hepatic stellate cells (A-HSC) that play a key role in liver fibrosis and cirrhosis, have been suggested as contributors to the HCC-prone microenvironment. Here, we have identified and validated an A-HSC-specific gene expression signature among non-tumor tissues of 319 HCC patients that is significantly and independently associated with HCC recurrence and survival. Peritumoral, rather than tumor tissue-related A-HSC-specific gene expression is associated with recurrence and poor survival. Analyses of A-HSC-specific gene signatures and further immunohistochemical validation in an additional 143 HCC patients have revealed that A-HSCs preferentially affect monocyte populations, shifting their gene expression from an inflammatory to an immunosuppressive signature. In addition, the interaction between A-HSCs and monocytes induces protumorigenic and progressive features of HCC cells by enhancing cell proliferation, migration and tumor sphere formation.

Conclusion

Our results show that A-HSCs play a significant role in promoting HCC progression via interaction with and alteration of monocyte activities within the liver microenvironment. Thus, disrupting the interactions and signaling events between the inflammatory milieu and components of the microenvironment may be useful therapeutic strategies for preventing HCC tumor relapse.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a major global health problem with increasing incidence. It accounts for 70% to 85% of the total liver cancer burden and is the third most common cause of death from cancer worldwide (1). Although about 20% of HCC patients have the option of resection with a curative intent, with a reported 5-year survival rate of 40% to 50%, a high incidence of postoperative recurrence is universal and continues to be the main cause of cancer-related mortality, even for those with early stage cancer (2). Recurrence may occur in the remaining liver in 78% to 96% of cases, as a result of either intrahepatic metastasis from the original primary tumor or de novo tumor occurrence (3). Understanding the mechanisms of tumor recurrence after curative resection is crucial for improving the outcome of HCC patients.

The liver microenvironment is a crucial contributor to HCC initiation and progression. It is well known that a deregulated microenvironment promotes tumorigenesis based on the fact that chronic inflammation is associated with high incidence of cancer (4). HCC is a typical inflammation-associated cancer, given that 80–90% of HCC cases arise from cirrhotic livers (5). It has also been observed that up to 27% of cancer-free patients with cirrhosis develop liver cancer over 12 years (6). While distinct molecular subgroups of HCC reflecting tumor biology have been identified (7–10), molecular signatures of the liver microenvironment have also been shown to be linked to tumor recurrence and HCC prognosis (11, 12). One plausible hypothesis is that the liver microenvironment may contribute to HCC recurrence through reprogramming of the inflammatory milieu.

Hepatic stellate cells (HSCs), a group of star-shaped mesenchymal cells residing in the space of Disse, are well known for storing retinols (vitamin A) and for participating in repair following liver injury (13). Activation of HSCs (referred to as A-HSCs), a key feature of liver fibrosis and cirrhosis, has been shown to be essential for the development of liver fibrosis (14, 15). Considering that liver cirrhosis is strongly associated with HCC and that a majority of HCC patients following tumor resection still have fibrosis and cirrhosis in the remnant liver, it is possible that A-HSC may contribute directly to HCC recurrence, metastasis and progression. Consistently, peritumoral A-HSCs based on the expression of seprase, osteonectin, and tenascin-C is associated with early recurrence of HCC (16). Similarly, peritumoral A-HSCs based on α smooth muscle actin(αSMA) expression is associated with time to recurrence and overall survival of HCC patients (17). It is postulated that HSCs are a component of the prometastatic liver microenvironment that contributes to the desmoplastic reaction and metastatic growth (18). Considering the therapeutic potential of targeting HSCs to prevent HCC recurrence, it is important to identify the molecular mechanisms underlying the contribution of HSCs to HCC recurrence.

In the present study, we compared the expression of A-HSC specific genes to that of immune cell-specific genes in peritumoral HCC specimens and linked them to HCC prognosis. Consistent with previous findings, we found that an A-HSC-specific gene signature is associated with poor survival and tumor recurrence of HCC patients. Notably, the A-HSC prognostic signature is preferentially associated with an activation of myeloid cell lineage, but not lymphoid cell lineage. We further found that an activated human HSC cell line LX2 reprogrammed peripheral blood mononuclear cell (PBMC)-derived myeloid cells, but not lymphoid cells, to promote HCC migration and spheroid formation in vitro. Our results support the hypothesis that A-HSCs may reprogram monocytes and change their cytokine milieu to support HCC development.

MATERIALS AND METHODS

Clinical specimens, microarray and statistical analyses

A workflow of the study design, patient cohorts and data analysis strategies are included as Suppl Figure S1. Specifically, a previously described cohort of 247 Chinese HCC patients obtained with informed consent from patients at the Liver Cancer Institute (LCI) and Zhongshan Hospital (Fudan University, Shanghai, China) with publically available Affymetrix U133A array data (NCBI GEO accession number: GSE14520), was used to evaluate the prognostic correlation of A-HSCs for HCC patients (19). A cohort of 72 Korean HCC patients with mRNA array expression data (NCBI GEO accession number: GSE39791) (20) and tissue microarrays of an additional 143 HCC patients from LCI were used for validation. The clinical characteristics of the two LCI cohorts are included in Suppl Table S1 and S2. The BRB-Array Tools (version 4.3.1) was used for survival risk prediction, class comparison and searching for survival related genes as previously described (7). Refer to the Supporting Information for details.

Peripheral blood mononuclear cell (PBMC) isolation and CD14+ monocytes purification

PBMC was isolated from healthy blood donor buffycoat (approved and provided by the NIH Department of Transfusion Medicine) by density-based centrifugation through histopaque (Sigma, St. Louis, MO). For the isolation of CD14+ monocytes, PBMC were further positively selected using CD14 Microbeads and an AutoMACS separation unit (Miltenyi Biotech, Bergisch Gladbach, Germany) according to the manufacturers’ instructions. Negative cells were collected and considered as CD14− cells. The purity of CD14+ monocytes was > 98%, and the CD14− cells contained no more than 1% CD14+ cells.

Cell Culture, FACS analyses, Migration, Proliferation, Spheroid formation, and Quantitative reverse transcription-PCR assays

The human HSC cell line LX2 was provided by S.L.Friedman (Mount Sinai School of Medicine, New York, NY). LX2 cells and hepatocellular carcinoma cell lines Huh1 and Huh7 cells were cultured in DMEM (Gibco BRL, Gaithersburg, MD) supplemented with 10% FBS and L-glutamine. For PBMC, CD14+, CD14− cell culture or any co-culture with CD14+ or CD14− cells, RPMI1640 (Gibco) supplemented with 10% FBS and L-glutamine was used. For preparation of conditioned medium (CM), 2×10⁵ LX2 cells per ml and/or 2×10⁶ CD14+ or CD14− cells per ml were cultured in RPMI1640 for 48hr, followed by conditioned medium collection. Cell debris was removed by low speed centrifugation and passed through a 0.45 μm filter (Millipore, Bedford, MA). For co-culture, 2×10⁵ LX2 cells per ml were seeded and allowed to attach 24hr before 2×10⁶ CD14+ or CD14− cells per ml were added to culture. In other experiments, EGFP expression lentivirus (Leidos Biomedical Research, Frederick, MD) were transfected into Huh1 or Huh7 respectively, after selection with puromycin (2mg/ml, 1:1000) (Mediatech, Herndon, VA). The transfectants that expressed the strongest EGFP (EGFP-CMV13-Huh1 and EGFP-EF1a-Huh7) were used in spheroid formation assays. To determine the frequency and phenotype of CD14+ and CD3+ cells, multicolor fluorescence-activated cell sorting analysis was performed. Refer to the Supporting Information for details.

RESULTS

A-HSC specific gene signature predicts survival and recurrence in HCC

Several genes specific to A-HSCs compared to quiescent HSCs have been previously described (13, 14, 21, 22). To globally search for genes that are specifically activated in human A-HSCs in liver tissue, we compared gene expression data of two independent batches of human primary hepatocytes, two independent batches of primary fibroblasts and two independent batches of primary cultured A-HSCs from the Duke-UNC-Texas-EBI ENCODE Project (GSE15805) (Figure 1A). We identified 258 overlapping genes that had a relatively high expression level in A-HSCs (Log2 intensity ≥8) and were more abundantly expressed in A-HSCs compared to primary hepatocytes or to fibroblasts (≥2.5fold) (p<0,001; based on Chi-square test with Yates correction for the probability of overlapping genes) (Figure 1A). In addition, there was a significant overlap (p<0.001) between the A-HSC signature and those genes previously reported to be linked to activated HSCs as compared to quiescent HSCs (Suppl Figure S2) (22). Among the 258 genes, 194 genes were found in the Affymetrix U133A array dataset (GSE14520; Roessler’s study) and were thus used as A-HSC-specific genes for further analyses.

Figure 1.

An activated HSC (A-HSC)-specific gene expression in non-tumor tissue, but not tumor tissue is associated with HCC prognosis. (A) Identification of an A-HSC-specific gene signature based on the Duke-UNC-Texas-EBI ENCODE project (GSE15805) along with individual GEO accession #s. (B) Kaplan-Meier survival analyses of 226 Chinese HCC cases based on survival risk prediction results of the 194-gene A-HSC-signature in non-tumor (left panel) and tumor (right panel). (C) Recurrence-free survival in 226 Chinese HCC cases described in panel B. (D) Recurrence-free survival analysis based on hierarchal clustering of peritumoral A-HSC gene expression in an independent cohort of 72 Korean HCC cases.

To determine if A-HSC-specific genes were related to patient prognosis, we performed survival risk prediction using 10-fold cross validation and 1,000-fold random permutation of the Cox-Mantel log-rank test based on the expression of A-HSC specific genes in peritumoral tissues of 247 HCC patients from the cohort used in Roessler’s study. Among them, 226 patients have survival data, and 223 patients have recurrence data. We found that the A-HSC-specific gene set assigned HCC cases into high and low risk groups (hereby referred as A-HSC high risk and A-HSC low risk), which were associated with overall survival (Log rank p<0.001, permutation p=0.009) (Figure 1B, left panel). In contrast, we found no survival association of A-HSC-specific gene set in the matched tumor tissues (Figure 1B, right panel). This analysis showed that the prognosis-related A-HSC gene expression was only applicable to peritumoral tissues, but not to tumor tissues. Further analyses revealed that the A-HSC-specific gene signature in peritumoral tissues predicted tumor recurrence (p<0.05) in a LCI cohort (Figure 1C; Roessler’s study) as well as in a Korean HCC cohort (p<0.05) (Figure 1D; NCBI GEO accession number: GSE39791: Lee’s study), and was mainly associated with late recurrence (Suppl Figure S3). In addition, we found that A-HSC gene signature could predict survival in a well-defined HCC cohort with metastasis-inclined microenvironment (NCBI GEO accession number: GSE5093: Budhu’s study) (Suppl Figure S4).

Because various clinical parameters have been shown to correlate with HCC prognosis, we further examined whether the A-HSC-specific gene signature was independently related to HCC prognosis by performing univariate and multivariate Cox proportional hazards regression analysis. A univariate analysis revealed that the A-HSC-specific gene signature, gender, cirrhosis, α-fetoprotein (AFP), tumor size, nodularity, vascular invasion, tumor staging as well as our previously validated HCC metastasis/early recurrence gene signature (19) were significant predictors of survival (p=0.001) (Table 1). The multivariate Cox regression model for survival, which controlled for HCC metastasis/early recurrence gene signature, gender, cirrhosis, AFP and tumor staging revealed that the A-HSC-specific gene signature was an independent predictor of survival (p<0.01).

Table 1.

Univariate and mutivariates Cox regression analysis of clinical factors associated with overall survival of 247 HCC patients (LCI cohort)

Clinical variable	Univariate analysisa Hazard ratio (95% CI)	p value	Multivariates analysisb Hazard ratio (95% CI)	p value
A-HSCs specific genes (high vs. low risk)	1.95 (1.31–2.90)	0.001	1.74 (1.15–2.64)	0.008
Metastasis signature (high vs. low risk)c	1.78 (1.21–2.61)	0.003	1.53 (1.03–2.26)	0.034
Age (>50 y vs. ≤ 50 y)	1.16 (0.79–1.70)	0.437	n.a.d
Gender (male vs. female)	2.18 (1.06–4.49)	0.034	2.07 (1.00–4.31)	0.051
HBV (AVR-CC vs. CC)e	1.08 (0.69–1.68)	0.733	n.a.
HBsAg (positive vs. negative)	1.49 (0.65–3.39)	0.346	n.a.
HBeAg (positive vs. negative)	1.08 (0.69–1.68)	0.739	n.a.
Cirrhosis (yes vs. no)	2.85 (1.05–7.73)	0.040	2.11 (0.76–5.81)	0.150
ALT (numeric)	1.00 (1.00–1.00)	0.253	n.a.
Total bilirubin (numeric)	0.99 (0.96–1.02)	0.396	n.a.
Albumin (numeric)	0.96 (0.92–1.00)	0.057	n.a.
Child-Pugh score (B vs. A)	1.38 (0.79–2.43)	0.259	n.a.
AFP (numeric)	1.00 (1.00–1.00)	0.001	1.00 (1.00–1.00)	0.424
Tunor size (>3 cm vs. ≤ 3 cm)	1.85 (1.20–2.87)	0.006	n.a.
Multinodular (yes vs. no)	1.67 (1.08–2.59)	0.021	n.a.
Encapsulation (no vs. yes)	0.99 (0.64–1.51)	0.949	n.a.
Microvascular invasion (yes vs. no)	2.70 (1.47–4.97)	0.001	n.a.
BCLC staging (B + C vs. 0 + A)	3.04 (2.03–4.57)	< 0.001	2.25 (1.44–3.50)	< 0.001

Bold indicates significant p values.

Abbreviations: AVR-CC, active viral replication chronic carrier; CC, chronic carrier.

^aUnivariate analysis, Cox proportional hazards regression.

^bMutivariate analysis, Cox proportional hazards regression adjusting for gender, cirrhosis, AFP and BCLC staging.

^cThe HCC metastasis/early recurrence gene signature from Roessler et al (19).

^dn.a. not applicable

^eCC, chronic carrier, AVR-CC, active viral replication chronic carrier.

To further determine if the A-HSC signature was associated with prognosis due to a difference in tumor biology, we compared gene expression profiles of HCC tissues between A-HSC high and low risk groups. Surprisingly, the class comparison analysis revealed that there was no significant difference in HCC global gene expression profiles between these groups. We only found 26 genes that were differentially expressed in tumor tissue between the A-HSC high and low risk HCC cases (univariate p<0.001, fold changes from 0.74 to 1.24) and none of them were significant (FDR > 0.05) (Suppl Table S3). Taken together, the results above are consistent with the hypothesis that A-HSC mainly contributes to a liver microenvironment that favors HCC progression.

Among 194 A-HSC-specific genes, 37 genes were significantly correlated with HCC survival by Cox Regression analysis (parametric p<0.05, FDR <0.05) (Suppl Table S4). Hierarchical clustering of 226 peritumoral HCC samples based on these 37 genes resulted in two major groups, one predominantly enriched for the A-HSC high risk cases and other enriched for the A-HSC low risk cases (Figure 2A). In addition, these subgroups also largely corresponded to HSC activation status as determined by hierarchical clustering of common HSC activation markers, i.e., ATCA2, COL1A1, COL1A2, TIMP1 and SPARC (23, 24). The activation of HSCs was observed in a majority of high risk cases (74%, Figure 2A). We performed qRT-PCR to validate the expression of 3 genes highly up-regulated in A-HSC high risk cases, i.e., CCL2, SRGN and JAG1 that encode extracellular proteins, in 25 randomly selected peritumoral HCC specimens analyzed by gene array analyses. The array data and qRT-PCR data were highly correlative (Figure 2B). Ingenuity Pathway Analysis (IPA) analysis of the 37 genes revealed that among the significantly affected canonical pathways (p< 0.01), HSC activation was the top signaling pathway among other pathways such as those related to EMT, HMGB1, TREM1, MAPK, IL-17, TR/RXR activation, FGF signaling and HGF signaling (Figure 2C). Thus, these results reaffirmed the fidelity and enrichment of A-HSC-related genes from our initial selection.

Figure 2.

HSC-specific genes are associated with HCC survival. (A) Hierarchical clustering of 37 A-HSC-specific genes that are significantly associated with HCC survival. Each column represents an individual tissue sample. Genes were ordered by centered correlation and complete linkage. The scale represents gene expression levels from −2.0 to 2.0 in log 2 scale. Each case status is categorized by the A-HSC-gene signature described in Figure 1B, left panel, Hoshida’s signature, Budhu’s signature and 5 well established HSCs activation markers are included above the heatmap. (B) Validation of the array data by RT-PCR. The expression of survival related HSCs-specific genes CCL2, SRGN and JAG1 were validated in 25 randomly selected non-tumor tissues used in array analyses. (C) Canonical pathways of 37 HCC survival related HSCs-specific genes. Significantly enriched canonical pathways (p<0.01) are shown.

We also compared the A-HSC signature to two previously published stroma signatures, i.e., the Hoshida signature mainly associated with late recurrence (12) and the Budhu signature mainly associated with early recurrence (11). We found that these three signatures are largely distinct, with only one overlapping gene between Hoshida and A-HSC (Suppl Figure S5). Interestingly, we found that the A-HSCs signature and the Hoshida signature, but not the Budhu signature, assign similar cases (Figure 2A).

A unique profile of inflammatory cells in peritumoral regions is associated with A-HSC

Since A-HSC related gene expression in peritumoral regions, but not in tumor tissues, is associated with HCC prognosis and recurrence, we hypothesized that A-HSC may promote a recurrence-inclined liver milieu through alteration of immune cell profiles. Therefore, we performed a class comparison analysis of gene expression profiles in 226 peritumoral tissues between A-HSC high risk and low risk groups. In contrast to the minimum difference in tumor gene expression described above, we found 2,920 genes that were differentially expressed between high and low risk groups in peritumoral tissues (univariate p<0.001). Among them, 231 significant genes had >1.5fold differences (FDR < 0.05) (Figure 3A, Suppl Table S5). Pathway analyses of the 231 differentially expressed genes based on IPA revealed that 20 of 22 significantly enriched canonical pathways in the A-HSC high risk group were mainly related to inflammation/immune responses, such as granulocyte cell adhesion, maturation, antigen presentation, T helper cell differentiation, IL-17 and IL-4 signaling, etc (Figure 3B). The remaining 2 pathways were related to HSC activation, i.e., Hepatic Stellate Cell Activation and Inhibition of Matrix Metalloproteases. These results reaffirmed our initial analyses on the selection of A-HSC-specific genes and further suggested that inflammation/immune cell-related activities are main contributors of A-HSC related HCC poor prognosis.

Figure 3.

Gene expression profiles of peritumoral HCC tissues stratified by the A-HSC-signature-predicted high and low risk subgroups. (A) Hierarchical clustering of 231 differentially expressed genes (p<0.001; fold>2.5) between A-HSC-signature-predicted high and low risk subgroups. Each row represents an individual gene and each column represents an individual case. Genes were ordered by centered correlation and complete linkage. The scale represents gene expression levels from −2.0 to 2.0 in log 2 scale. (B) Pathways analysis of the genes described in panel A by IPA. Significantly enriched canonical pathways (p<0.01) are shown. (C) Immune cell gene expression profiles based on A-HSC high (H) and low risk (L) HCC subgroups defined by A-HSC-signature. Among 1,622 immune cell genes defined by IRIS, 535 genes unique for each cell type were used to calculate % of affected genes. The total number of significantly expressed genes (adjusted p<0.05) with different fold changes between H and L subgroups are indicated. ** p<0.01 and * p<0.05; from hypergeometric probability test.

To determine what types of immune components contribute to A-HSC-related prognosis, we further performed transcriptome analyses of 226 peritumoral HCC specimens based only on 1,622 immune cell-specific genes, which were identified from a compendium of microarray expression data through immune response in silico (IRIS) analyses (25) (Suppl Table S6). We found a preferential enrichment of gene activities related to myeloid cell lineage such as monocytes, but not lymphoid cell lineage, in A-HSC high risk cases (Figure 3C).

To further validate our gene array data, we performed IHC analyses in an independent HCC cohort (n=143) (Suppl Table S2) to determine the distributions of A-HSCs (αSMA+), monocytes/macrophages (CD68+) (Figure 4A) and lymphocytes (CD3+) (Suppl Figure S6). We found a significant positive correlation between the population of αSMA+ cells and CD68+ cells (Figure 4B), but not between αSMA+ cells and CD3+ cells (Suppl Figure S6), in peritumoral HCC specimens. Consistently, we found that HCC cases with either αSMA^high or CD68^high, or both in the peritumoral region, had a significantly worse prognosis than peritumoral αSMA^low CD68^low HCC cases (Figure 4C). Univariate and multivariate Cox proportional hazards regression analysis revealed similar results as that of A-HSC-specific gene signature (Suppl Table S7). These results indicate that myeloid cells, including monocytes, were the main contributor of A-HSC related HCC poor prognosis.

Figure 4.

Association of A-HSCs and monocytes with HCC prognosis. (A) Representative staining patterns of αSMA and CD68 in peritumoral HCC tissues. Magnifying objectives used to capture images are indicated. (B) Correlation between the number of αSMA+ cells and CD68+ cells are presented for 143 peritumoral HCC tissues. For each case, positive cells were counted in two randomly-selected peritumoral areas with a 20× objective. (C) Kaplan-Meier survival analysis of 143 HCC cases stratified by αSMA and CD68 expression status. Both high (n=50), both low (n=45) or either high (n=48); high groups: >10 αSMA+ cells or >20 CD68+ cells per whole viewing area at 20× objective.

Preferential effect of HSCs on monocytes but not lymphocytes in promoting HCC cell activities in vitro

We hypothesized that the involvement of monocytes in A-HSCs related to prognosis may be achieved through cell-cell interaction between A-HSCs and monocytes and/or immune regulation influenced by HSC-educated monocytes, which then influences HCC progression. We used a human HSC cell line LX2 (26), which resembles A-HSC to determine the influence of A-HSCs on immune cells. PBMCs from healthy donors were co-cultured with LX2 conditioned medium (CM). After 48hr, we noticed a significant enhancement of monocyte attachment in PBMC cultured with LX2 CM (Figure 5A). However, no significant difference in cell viability of CD14+ and CD3+ cells, the two main populations of PBMC, was observed when cultured with or without LX2 CM (Suppl Figure S7A, B). Analysis of cell surface marker expression using flow cytometry revealed that CD14 expression on monocytes increased. Nevertheless, we did not find differences in the expression of other monocyte surface markers, nor did we find differences in T cell activation markers, such as CD69, PD1 or CD107a, upon culture in LX2 CM. Hence, LX2 CM exhibited an effect only on CD14+ monocytes.

Figure 5.

Functional interactions between A-HSCs and monocytes. (A) Morphological characteristics of cultured PBMC in LX2 conditioned media (LX2-CM) for 24h. Representative phase contrast cell images from control and LX2-CM are shown in the top panels. Percent of attached cells are shown in the bottom panel. (B) Flow histograms of cell surface marker expression of PBMC co-cultured either with LX2-CM or directly with LX2 cells. Marker expression was restricted to CD14+ monocytes. Blue, LX2-CM; green, LX2 (LX2-CO); red, media with 10% FBS as a control. Representative histograms from 5 independent experiments are shown. (C) The expression levels of various monocyte surface markers as median fluorescence intensities (MFI) in control, LX2-CM or LX2-CO groups.

To further evaluate A-HSC-mediated activities, monocytes were enriched from PBMC by CD14 positive selection before in vitro culture. We found that co-culturing monocytes directly with LX2 cells further increased the expression of CD14 compared to culture with LX2 CM alone (Figure 5B–C). Furthermore, co-culture of LX2 cells with monocytes also up-regulated surface expression of CD15 and CCR2, while expression of the T-lymphocyte activation antigen CD86 (B7-2) was down-regulated in A-HSCs educated monocytes. Our results show, that co-culturing monocytes with LX2 cells further enhanced the slight differences observed when monocytes were incubated with LX2 CM only (Figure 5B–C). In contrast, co-culture of CD14− cells, which primarily constitute T cells, with LX2 cells did not induce expression of T cell activation markers on CD3+/CD4+ and CD3+/CD4− lymphocyte subsets (Suppl Figure S8). Thus, in our in vitro culture model, LX2 cell exhibited an effect only on monocytes and not on T cells; and this effect was enhanced when monocytes were cultured directly with LX2 cells.

We next sought to determine expression of monocyte related genes in peritumoral HCC samples, which may reflect activation, polarization and inflammatory properties of monocytes (27). We performed hierarchical clustering of a set of well-documented M1/M2 related genes (27) in 226 peritumoral HCC samples. This analysis revealed that a majority of M2-like genes were more abundantly expressed in A-HSC high risk cases while most of M1-like genes were more abundantly expressed in A-HSC low risk cases (Figure 6A). We then further determined the expression levels of genes corresponding to above M1/M2-like genes with available probes for qRT-PCR in CD14+ cells co-cultured with LX2 cells. We found that all of 7 M2-like genes were upregulated while 2 of 7 M1-like genes were downregulated in CD14+ cells co-cultured with LX2 cells (Figure 6B). These results are consistent with the hypothesis that A-HSC may affect monocytes by reprogramming them to an immunosuppressive phenotype.

Figure 6.

M1 or M2-like monocyte expression patterns linked to A-HSC. (A) Hierarchical clustering analysis based on expression profiles of 17 genes that are related to M1 and M2-like monocytes in non-tumor tissues of 226 HCC cases. Each column represents an individual tissue sample. Genes were ordered by Pearson centered correlation and average linkage. The scale bar represents gene expression levels from −2.0 to 2.0 in log 2 scale. Each case status is categorized by the A-HSCs signature predicted risk of survival. (B) The expression levels of M1 or M2-like genes in CD14+ monocytes co-cultured with LX2 cells as determined by qRT-PCR. The natural log value of gene expression, normalized to 18S rRNA and to CD14+ monocytes without co-culture, is presented as the mean ± standard deviation (SD) from 3 independent tests. Statistical significance was determined by the Student’s t test relative to control CD14+ cells. ***p<0.001, **p<0.01, *p<0.05

Finally, we used cell migration, proliferation and 3D spheroid formation assays to determine the impact of interactions between HSCs and monocytes on HCC cells in vitro. We found that cell migration of Huh7 cells was significantly enhanced when either were co-cultured with LX2 or LX2 CM (Figure 7A, Suppl Figure S9). This activity was further elevated when co-cultured with CD14+ cells, but not CD14− cells, isolated from PBMCs. Similar results were obtained with another HCC cell line, Huh1 (Suppl Figure S9). These results are consistent with the gene expression data indicating that LX2 cells interact preferentially with CD14+, but not CD14− cells, to alter HCC activities. Interestingly, while CD14+ cells alone had no activity, LX2 cells alone could promote Huh7 cell proliferation, which was abrogated by co-culturing with CD14+ cells (Suppl Figure S10). To further determine the effect on cancer cell self-renewal activity, we examined 3D spheroid formation of Huh7 and Huh1 cells as previously described (28). We transduced Huh1 or Huh7 cells with an EGFP expressing lentiviral construct to facilitate the identification of HCC spheroids in a co-culture system (Figure 7B). We found that CD14+ cells, but not LX2 cells, promoted spheroid formation of Huh7 cells while co-culture of LX2 and CD14+ further stimulated spheroid formation (Figure 7B). Similar data were obtained with Huh1 cells (Suppl Figure S11B). Thus, LX2 cells preferentially interacted with CD14+ monocytes to enhance HCC cell migration and 3D spheroid formation.

Figure 7.

The impacts of A-HSC-monocyte co-culture on HCC cell migration and spheroid formation. (A) Huh7 cells were cultured with 40% conditioned media derived from LX2 cells, CD14− cells, CD14+ cells, LX2-CD14− co-culture or LX2-CD14+ co-culture. Cell migration was assessed using BD Cell Migration chambers following 16 hr culture. Data are an average of triplicate experiments for each condition and are expressed as the mean ± CV. (B) GFP-expressing Huh7 cells were co-cultured under the condition of spheroid formation with LX2, CD14− or CD14+ cells alone, or together with LX2-CD14− or LX2-CD14+ co-culture for 14 days. Representative mature spheroids are shown. Data are an average of triplicate experiments from each PBMC donor. CD14+ and CD14− cells were derived from at least 3 donors and were independently evaluated. ***p<0.001, **p<0.01

DISCUSSION

Liver fibrosis is a precursor of cirrhosis and is strongly associated with HCC development. One common notion is that liver fibrosis may lead to an induction of a cellular inflammatory response accompanied by oxidative stress that inflicts cellular damage, which in turn induces hepatocyte proliferation, liver regeneration, and subsequent HCC initiation and progression (29). Recent studies indicate that in addition to inflammation-related and procarcinogenic parenchyma cell damage, various non-parenchyma cells of the liver may promote HCC development via reprogramming the liver milieu to favor HCC cell growth (4). Candidate non-parenchyma cells of the liver include lymphocytes, myeloid cells and stromal cells, such as HSCs and sinusoidal endothelial cells. While an essential role of HSCs in liver fibrosis is well established (13, 30), how HSCs contribute to HCC development remains unclear. In this study, we found that HSC-specific gene expression is associated with HCC recurrence and overall survival, and that A-HSC status is closely linked to the activity of myeloid cells, but not lymphoid cells. We developed an IHC-based method based on the status of A-HSC (αSMA+) and macrophage (CD68+) to predict HCC recurrence and overall survival. Consistently, ex-vivo analyses revealed that A-HSCs preferentially induced an activation of monocytes, but not lymphocytes, and that interaction between A-HSCs and monocytes, but not A-HSCs and lymphocytes, resulted in promotion of HCC cell migration and spheroid formation. These results support the hypothesis that A-HSCs may promote a recurrence-inclined liver milieu through alteration of monocyte activities leading to HCC progression. Disruption of A-HSCs and monocyte interaction may provide a therapeutic strategy as an adjuvant therapy for HCC to prevent tumor relapse.

Through transcriptome data mining of the Duke-UNC-EBI ENCODE project, we systematically identified 194 A-HSC-specific genes. Among them, 37 genes were significantly associated with HCC prognosis and 30 genes, including 10 that encode extracellular proteins, were more abundantly expressed in peritumoral HCC with poor prognosis. Notably, among the 10 extracellular proteins, CCL2 and TNFRSF11B have been shown to have cytokine activities. CCL2, a top-ranked gene, also known as MCP-1, encodes a monocyte chemotactic protein mainly produced by A-HSCs (31). CCL2 has been implicated as a key chemokine to promote a tumor-favorable inflammatory microenvironment that contributes to tumor progression, including HCC (32, 33). Furthermore, JAG1 and HGF are known growth factors, and that HGF is a potent mitogen for hepatocytes and HCC cells. ADAM9 encodes a peptidase, which is involved in cell-cell or cell-matrix interaction. Consistently, HSC-produced ADMD9 has been shown to promote HCC cell invasion through tumor-stromal interaction (34). It seems that HSC-associated secretory factors may collectively promote an inflammatory liver microenvironment, leading to liver fibrosis and supporting HCC development. Pharmacological inhibition of CCL2 by mNOX-E36 may be a viable approach to inhibit liver fibrosis (35). Thus, it may be feasible to exploit the concept of CCL2 blockage as an adjuvant therapy to prevent HCC recurrence post resection.

It was noteworthy that A-HSCs and HCC metastasis gene signatures were independent prognostic predictors, which suggests that A-HSCs might promote HCC relapse through a different mechanism than tumor metastatic potential. This was also consistent with our finding that there was no difference in tumor-related gene expression between A-HSC high and A-HSC low risk groups. In addition, we compared our human A-HSC signature to a well established murine A-HSC signature (22) and found a significant enrichment for overlapping genes between these two signatures. These results are encouraging since many factors could account for the lack of gene overlap such as mouse vs human, CCl4/bile duct ligation-induced fibrosis and stellate cell activation in mouse models vs fibrosis developed in HCC patients associated with viral hepatitis. These results support a good clinical relevance of this murine model (22).

Our gene expression analysis and the IHC results revealed a correlation between monocyte accumulation and HSC activation and poor survival in HCC. Indeed, a recent study by Hoechst et al showed that activated hepatic stellate cells could induce myeloid derived suppressor cells (MDSC) from CD14+ monocytes in a contact dependent manner (36). These MDSCs have been linked to immunosuppression and poor prognosis in a variety of cancers (37, 38). In HCC, these cells inhibit not only the adaptive, but also the innate immune response (39) and are also capable of inducing immunosuppressive regulatory T cells (40). In addition to the findings by Hoechst et al, we found that monocytes co-cultured with activated stellate cells changed cytokine gene expression from an inflammatory to an immunosuppressive signature, namely up-regulating immunosuppressive cytokines, like IL-10 and TGF-β1, and down-regulating inflammatory cytokines TNFα and IL-1β. The M1/M2 related gene expression patterns in A-HSCs high and low risk cases further supported that the myeloid cells enriched in A-HSCs high risk cases have an immunnosuppressive phenotype, suggesting that A-HSCs could have induced immunosuppressive myeloid cells in peritumor tissue of HCC patients.

While our in vitro experiments using co-cultured LX2 cells and CD14+ cells derived from PBMC are also consistent with this view, we should interpret these data with caution since LX2 is an immortalized HSC cell line and the in vitro culture conditions may not completely recapitulate the biological background found in vivo. Future studies to exploit 3D organoids-like culture in vitro and mouse models by genetically manipulating various liver cell lineages may help to clarify this notion.

We found an up-regulation of CCR2, the receptor for the chemokine CCL2, on monocytes co-cultured with activated hepatic stellate cells. CCR2, has been implicated in macrophage polarization to an immunosuppressive M2 phenotype (41). Along the same lines, recruitment of CCR2+ monocytes into the liver has been shown to promote liver metastasis development in colorectal cancer patients (42). These published results on CCR2+ myeloid cells, together with our finding of CCR2 up-regulation on HSC-educated monocytes provides further evidence of the ability of A-HSCs to reprogram monocytes, to a tumor-promoting phenotype. Therefore, HSC-induced immunosuppressive monocytes could account for the poorer prognosis in A-HSC predicted high risk HCC.

Our results are consistent with the hypothesis that fibrosis-related tumor recurrence may be a functional consequence of interactions among HSCs, myeloid cells and tumor cells. This hypothesis is nicely supported by several clinicohistopathological analyses of both HCC and intrahepatic cholangiocarcinoma (ICC) (16, 17, 43). Furthermore, in vivo studies show that simultaneous implantation of HSCs and HCC cells into nude mice promotes tumor growth and invasiveness (44). These results indicate that A-HSCs can induce a permissive liver microenvironment that favors HCC recurrence. Interestingly, a recent elegant study by Coulouarn et al suggests that a direct interaction between hepatocyte and stellate cells can drive hepatocarcinogenesis (45). The authors suggest that hepatocyte-HSC crosstalk can generate a permissive proangiogenic microenvironment that promotes HCC progression, and have developed a 292-gene signature corresponding to LX2 co-cultured with HepaRG cells. It should be noted that a comparison between the 194-gene A-HSC signature described in the current study and the 292-gene LX2-HepaRG signature described by Coulouarn et al revealed only 3 overlapping genes, i.e., CCL2, MGP and DCN. In addition, only two genes, i.e., RCAN2 and JAG1, are common between the 194 A-HSC genes and the 123 HepaRG-related genes altered by LX2 cells. One possibility for this difference is that HSCs may promote HCC progression through interactions with both myeloid cells and hepatocytes perhaps via different mechanisms and signaling pathways. We should also note that our findings, as suggested by our IHC results, do not rule out the existence of other, potential mechanisms, i.e., A-HSC independent mechanisms, which could also lead to the induction of M2 macrophages and negatively affect prognosis. A crosstalk among these cells may collectively contribute to HCC progression. It will be interesting in the future to determine if HSCs may have an impact on other hepatic cells, such as sinusoid endothelial cells or liver stem/progenitor cells to promote HCC progression.

List of Abbreviations

AFP

α-fetoprotein

A-HSC

activated hepatic stellate cell

HCC

hepatocellular carcinoma

HSC

hepatic stellate cell

IHC

immunohistochemisty

MDSC

myeloid derived suppressor cells

PBMC

peripheral blood mononuclear cell

αSMA

α smooth muscle actin