Formalin-fixed, paraffin-embedded tissue from 386 patients with pathologically confirmed hepatocellular carcinoma (HCC), who had all received resection of the tumors at the Sun Yat-sen University Cancer Center between 2006 and 2010, were enrolled as previously described [32]. All samples were anonymously coded in accordance with local ethical guidelines (as requested by the Declaration of Helsinki), with written informed consent and using a protocol approved by the Review Board of Sun Yat-sen University Cancer Center. Overall survival (OS) was defined as the interval between surgery and death or between surgery and the last observation for the surviving patients. Relapse-free survival (RFS) was defined as the interval between surgery and the first of recurrence or death, or between surgery and the last observation for patients without recurrence. Tissues were used to construct a tissue microarray (TMA) as described previously [32]. A total of 386 patients who had complete OS and RFS information were used for the survival analysis. The clinical characteristics of all the patients are summarized in Additional file 2: Table S1.

Immunostaining and image analysis were conducted according to our previous reports [32]. In brief, TMA sections were dewaxed in xylene, rehydrated through a decreasing ethanol series, and then placed in 0.3% H₂O₂ to diminish the activity of endogenous peroxidase. The sections were then heated for antigen retrieval. Following incubation with rabbit anti-human EZH2 (BD Transduction Laboratories, BD Biosciences, San Jose, CA, USA), immunostaining was performed using the EnVision Detection System (DakoCytomation, Carpinteria, CA, USA) following the manufacturer’s instructions. Sections were counter-stained with hematoxylin. Image acquisition was performed using an Eclipse advanced research microscope (Nikon, Melville, NY, USA).

For multiplex immunofluorescence staining of EZH2, PD-L1 (clone: E1L3N™; Cell Signaling Technology, Danvers, MA, USA) and CD68 (DakoCytomation), Tyramide Signal Amplification (TSA) Plus Fluorescence Kits (PerkinElmer, Foster City, CA, USA) combined with immunohistochemistry (IHC) was used. To obtain multispectral images, the stained slides were scanned using the Vectra System (PerkinElmer). The definition of PD-L1 positive expression was the same as that described previously [32]. For colocalization analysis, images were acquired using a laser confocal microscope (Olympus, Essex, UK) and analyzed using FV10-ASW Viewer software (Olympus).

The expression of EZH2 was determined by nuclear EZH2 expression on tumor cells and immunohistochemical scoring of EZH2 was analyzed using the Inform software (PerkinElmer) with the modified Histo-score (H-score), which involves assessing both the intensity of staining (graded as non-staining-0, weak-1, median-2 or strong-3) and the percentage of positive cells (Additional file 1: Figure S1). The range of possible scores was from 0 to 300, quantified by H-score. The correlation of EZH2 and PD-L1 expression was analyzed by χ ² test. The cutoff value for the H-score was set at 35 with the minimum P value to categorize the samples into EZH2 high or low groups.

The human hepatoma cell lines PLC/PRF/5, Huh7, and Hep3B used in this study were purchased from the American Type Culture Collection (Manassas, VA, USA). PLC/PRF/5 cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium, and Huh7 and Hep3B cell lines were cultured in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum at 37 °C and 5% CO₂. Hepatoma cells were treated with recombinant interferon gamma (IFNγ) (Sino Biological Inc.), DZNep (MedChemExpress, Monmouth Junction, NJ, USA), or GSK-126 (MedChemExpress) for different times and at different concentrations.

Monocytes were selected from peripheral blood mononuclear cells using anti-CD14 magnetic beads (Miltenyi Biotec, Bergisch Gladbach, Germany) as described previously [33].

Hepatoma cells were transfected with small interfering RNAs (siRNAs) using Lipofectamine® RNAiMAX Reagent (Invitrogen, Waltham, MA, USA). Reverse transfection was performed according to the manufacturer’s instruction manual. The sequences of the siRNAs are listed in Additional file 2: Table S2.

Cells were collected by 0.25% trypsin digestion, and incubated with Phycoerythrin (PE) conjugated PD-L1 or isotype antibodies (eBioscience, San Diego, CA, USA). The cells were then subjected to flow cytometry.

Total RNA was isolated from cultured cells using TRIZOL (Invitrogen). Reverse transcription and real-time PCR were then performed using 5× All-In-One RT MasterMix (Applied Biological Materials, Richmond, Canada) and a SYBR green real-time PCR kit (Toyobo, Osaka, Japan). Relative quantification was calculated according to the comparative Ct method with normalization to the expression of GAPDH (encoding glyceraldehyde-3-phosphate dehydrogenase). The primers used are listed in Additional file 2: Table S3.

Cells were washed in phosphate-buffered saline (PBS) and suspended in Radioimmunoprecipitation assay (RIPA) buffer (Pierce, Rockford, IL, USA). Supernatant protein concentrations were determined using a BCA protein assay kit (Pierce). Supernatant samples were resolved by 10% or 15% SDS–PAGE depending on the sizes of target proteins, transferred to Immobilon-P polyvinylidene fluoride (PVDF) membranes (Millipore, Billerica, MA, USA) using electroblotting, and then probed with primary antibodies. Membranes were then incubated with horseradish peroxidase-conjugated secondary antibodies. The signals from the immunoreactive proteins were detected using the ECL reagent (Millipore). The information about the antibodies is listed in Additional file 2: Table S4.

Huh7 and Hep3B cells pre-transfected with siRNAs, IFNγ, or not, were cotransfected with the pGL3-PD-L1 promoter-luc reporter or pGL3-basic control vectors. pRL-TK was used as an internal control. Cell lysates were harvested for the dual-luciferase assay, which was performed according to the manufacturer’s instructions (Promega, Madison, WI, USA). The primers used are listed in Additional file 2: Table S5.

To generate a cell line with the stable knockdown of EZH2, lentiviral plasmids carrying a short hairpin RNA (shRNA) targeting EZH2 (VectorBuilder Inc., Shenandoah, TX, USA) were transfected to 293 T cells together with plasmids PMD2.G and pSPAX2 using Lipofectamine 3000 (Invitrogen). After 48 h, culture supernatants were collected, passed through 0.45-μm filters, and mixed with fresh media (1:1) and polybrene (8 μg/ml) to infect Hep3B cells. Cells infected with shEZH2 or control vectors were designated as Hep3B-shEZH2 and Hep3B-vector stable cell lines respectively, and were established using 1 μg/ml puromycin selection. The shRNA-targeted regions in EZH2 were at nt 784–804.

ChIP was performed by using SimpleChIP® Enzymatic Chromatin IP Kit (Cell Signaling Technology). Crosslinking was performed with 1% paraformaldehyde for 10 min. Micrococcal nuclease was added to digest the DNA to lengths of approximately 150–900 bp. The digested DNA solution was sonicated using a Qsonica Q700 sonicator (Qsonica, Newtown, CT, USA) for 1 min at an amplitude of 15%. Protein-DNA complexes were precipitated using specific antibodies against H3K27me3 (Cell Signaling Technology) and IgG control (Cell Signaling Technology). ChIP-enriched chromatin was used for RT-PCR with a SYBR green real-time PCR kit (Toyobo); the data were normalized to the input. The specific primers are listed in Additional file 2: Table S6.

The sequence of the human CD274 promoter was obtained from the EPD database (https://epd.vital-it.ch/index.php). MethPrimer software (http://www.urogene.org/methprimer/) was used to predict CpG islands and design bisulfite-specific primers for amplification and sequencing. Only one CpG island was predicted on the CD274 promoter (− 2000 bp to + 500 bp). The bisulfite-specific primers sequences were as follows:

Forward primer, ATTTGTTGTTTTGGGTAGAGGTG;

In current study, we used the same batch of transcriptome profiling data previously used [32], with differences in grouping. According to the status of immune activation which was defined by the expression level of PD-L1 on the infiltrated Mφs, the HCC tissues were divided into two groups: immune-activated and immune-suppressed [32] (12 cases in each group).

Gene with fold change (FC) more than two between groups and with a Student’s t tests p value < 0.05 was defined as a differentially expressed gene (DEG). And the expression data of all DEGs were analyzed to form heatmap by Funrich 3.1.3 software. The genes upregulated in immune-activated group were further analyzed for GO term enrichment by Funrich 3.1.3 software.

Differences in the means for continuous variables were compared using Student’s t test or analysis of variance, and differences in the proportions were tested using the χ ² test. Kaplan–Meier estimates were calculated and compared using the log-rank test. Cox proportional hazard regression models were applied to evaluate the prognostic variables for OS and RFS. IBM SPSS (version 21.0; IBM Corp., Armonk, NY, USA) statistics software was used for all statistical analyses. All data were analyzed using two-tailed tests unless otherwise specified, and P <  0.05 was considered statistically significant.

To evaluate the potential role of EZH2 in regulating PD-L1, we first investigated its expression in HCC tissues (Fig. 1a). IHC staining showed that EZH2 was highly expressed on hepatoma cells in HCC tumors compared with that on parenchyma cells in the non-tumor region. We noted that a few stroma cells were also positive for EZH2. Considering that Mo/Mφs are the major PD-L1-expressing stroma cells in HCC tumors, we performed multiplex staining to analyze EZH2 expression on Mo/Mφs. However, EHZ2 was weakly detected on Mo/Mφs in HCC tumor tissues (Fig. 1b).Fig1Fig. 1

EZH2 negatively regulates the IFNγ-induced PD-L1 expression. a Representative IHC staining of EZH2 in HCC tissues. The black arrows indicate the expression of EZH2 on stroma cells, and the red arrows indicate the expression of EZH2 on TCs. b Representative pictures of multiple immunofluorescence staining showing the expression of EZH2 (green) and CD68 (red) in HCC. Scale bar, 50 μm. The white arrows indicate Mo/Mφs, and the five-pointed stars indicate TCs. c Immunoblotting analysis showing the expression of IFNγ-induced PD-L1 in hepatoma cells and monocytes. d Hep3B cells were transfected with negative control (NC) or different EZH2-targeted siRNAs for 48 h, and then treated with IFNγ for 24 h. Immunoblotting analyses were performed to detect the expression of EZH2 and PD-L1. β-actin was used as a loading control. e qPCR analysis showing that downregulation of EZH2 promoted the mRNA expression of IFNγ-induced PD-L1 in PLC/PRF/5, Huh7, and Hep3B cells. f FACS staining showing that downregulation of EZH2 promoted the expression of IFNγ-induced PD-L1 in PLC/PRF/5, Huh7, and Hep3B cells. g Downregulation of EZH2 increased the protein level of IFNγ-induced PD-L1 in PLC/PRF/5, Huh7, and Hep3B cells. The indicated hepatoma cells were transfected with EZH2-targeted or NC siRNA for 48 h, and then treated with IFNγ for an additional 24 h. Immunoblotting analysis was performed to detect the protein levels of PD-L1, EZH2, and H3K27me3. GAPDH was used as a loading control

EZH2-mediated H3K27me3 often leads to epigenetic silencing of target genes [36]; thus we first tested whether EZH2 could directly regulate the H3K27me3 levels on the CD274 (the gene encoding PD-L1) promoter to suppress IFNγ-induced PD-L1 expression. A concentration gradient of DZNep (an inhibitor of all S­adenosyl­methionine (SAM)­dependent enzymes, including EZH2) [37] and GSK126 (a selective inhibitor of EZH2 methyltransferase activity) [38] were applied to reprogram the epigenetic pathways in hepatoma cells. As expected, GSK126 significantly inhibited the level of H3K27me3 without affecting EZH2 expression, while DZNep simultaneously downregulated the expression of EZH2 (Additional file 1: Figure S3a-b). Inhibiting EZH2 by GSK126 or DZNep treatment effectively increased IFNγ-induced PD-L1 expression in hepatoma cells (Fig. 2a, Additional file 1: Figure S3c-e). ChIP-qPCR analysis revealed that H3K27me3 occupancy on the promoter of CD274 was significantly downregulated in Hep3B-shEZH2 cells compared with that of the control cells (Fig. 2b). These data suggested that H3K27me3 modification at the promoter level is involved in EZH2­mediated PD-L1 repression.Fig2Fig. 2

EZH2-mediated H3K27me3 on the CD274 promoter controls PD-L1 expression. a Hep3B cells were pretreated with GSK126, DZNep, or DMSO for 48 h, and then treated with IFNγ for an additional 24 h. Immunoblotting was performed to detect the protein expression of PD-L1, EZH2, and H3K27me3. GAPDH and H3 were used as loading controls. b An H3K27me3 ChIP assay was performed in shEZH2 Hep3B and vector control cells. H3K27me3 levels at the CD274 (PD-L1) promoter were normalized to the input. TSS, transcription start site, − 0.3, − 0.5, − 1.0, and − 1.5 kb indicate the corresponding upstream loci of the CD274 gene TSS. CXCL10 was used as a positive control. (Mean ± S.E.M.; n = 3; * P < 0.05, ** P < 0.01, Wilcoxon test). c Diagram of the CpG island distribution on − 2000 nt to + 250 nt region of the CD274 promoter predicted by MethPrimer website. d DNA methylation on the CD274 promoter. DNA methylation at CpG sites was quantified using bisulfite sequencing. Filled circle, methylated; open circle, unmethylated. e DNA methylation and gene expression data for PD-L1 from TCGA HCC tissues were analyzed on the cBioportal website. The Pearson correlation coefficient (r) is shown. f and g Effect of downregulation of EZH2 on the IFNγ-STAT1 signaling activation. Huh7 (f) or PLC/PRF/5 (g) cells were pre-transfected with EZH2-targeted siRNA or NC for 48 h, and then treated with IFNγ for 0–4 h. Immunoblotting was performed to detect the levels of pSTAT1 and EZH2. STAT1 and GAPDH were used as loading controls for pSTAT1 and EZH2 respectively

To further explore the regulatory effect of EZH2 on the CD274 promoter, we constructed a promoter luciferase reporter plasmid containing different truncated versions of the CD274 promoter without H3K27me3 modification (Fig. 3a) and examined the transcription activity of these truncated promoters using dual-luciferase reporter assays. As shown in Fig. 3b, all the designed promoters exhibited similarly enhanced luciferase activity in EZH2-silenced cells when treated with IFNγ, indicating that EZH2 might regulate CD274 transcription by affecting the activities of certain transcription factors (TF) that bind to the P1 truncated promoter. Thirty-nine TFs were predicted to bind to the P1 promoter on the PROMO website [41, 42] and 469 genes who showed co-expression with CD274 with correlation coefficients more than 0.3 were screened out from the cBioportal website (HCC, TCGA, Provisional) [43, 44]. The Venn diagram analysis identified IRF1 as the only potential candidate gene that met both screening criteria (Fig. 3c-d, Additional file 3: Table S7).Fig3Fig. 3

IRF1 is a potential transcription factor involved in the negative regulation of PD-L1 by EZH2. a Schematic diagram of a series of CD274 (PD-L1) gene promoter luciferase reporter plasmids. b After transfection with EZH2-targeted or NC siRNA overnight, Huh7 cells were co-transfected with pGL3-basic vector or the indicated CD274 promoter luciferase reporter gene plasmid and the pRL-TK plasmid for 48 h, and then treated with IFNγ for an additional 24-h. Luciferase activity was determined and normalized using the dual luciferase reporter system. (Mean ± S.E.M.; n = 3; the asterisk represents a comparison between the siEZH2 group and the corresponding control group; * P < 0.05, ** P < 0.01, *** P < 0.001, Wilcoxon test). c TFs that could potentially bind to the P1 truncated promoter were predicted using the PROMO bioinformatics software (pink circle). Genes showing the absolute values of both the Pearson and Spearman expression correlation coefficient (positively or negatively correlated) of more than 0.3 with CD274 (PD-L1 gene) in HCC tissues (TCGA, Provisional) were analyzed on the cBioportal website (blue circle). Venn diagram showing that IRF-1 was the only candidate gene in both gene sets. d Scatter gram showing the mRNA expression correlation of CD274 and IRF1 from TCGA (HCC, Provisional). Pearson and Spearman correlation coefficients and P values are shown. e Huh7 and Hep3B cells were transfected with NC or EZH2-targeted, IRF1-targeted, or both, siRNA overnight, and then co-transfected with pGL3-basic vector or the P1 luciferase reporter gene plasmid and pRL-TK plasmid for 48 h. The cells were then treated with IFNγ for an additional 24 h. Luciferase activity was determined and normalized using the dual luciferase reporter system (Mean ± S.E.M.; n = 4; * P < 0.05, ** P < 0.01, Wilcoxon test). f After transfection with NC or EZH2 siRNA targeting 3′-UTR, Huh7 and Hep3B cells were transfected with the indicated plasmids for 48 h, and then treated with IFNγ for 24 h. Luciferase activity was determined and normalized using the dual luciferase reporter system (Mean ± S.E.M.; n = 3; * P < 0.05, ** P < 0.01, Wilcoxon test). pEZH2 and pIRF-1 represent ectopic expression of EZH2 and IRF-1 respectively, and the corresponding control groups were transfected with NC siRNA and/or vector plasmids. g After transfection with NC or the indicated siRNA targeting 3′-UTR, Huh7 and Hep3B cells were transfected with the indicated plasmids for 48 h, and then treated with IFNγ for 24 h. Luciferase activity was determined and normalized using the dual luciferase reporter system (Mean ± S.E.M.; n = 3; * P < 0.05, ** P < 0.01, Wilcoxon test). pIRF-1 represent ectopic expression of IRF-1, and the corresponding control groups were transfected with NC siRNA and/or vector plasmids. h Sequence logo of IRF1 binding site frequency matrix of Homo sapiens predicted using the online software JASPAR. i Schematic representation of IRF1 binding sites in the CD274 P1 promoter region, as predicted by JASPAR. IRE, IRF1 response element. j Huh7 and Hep3B cells were transfected with NC or EZH2-targeted siRNA overnight, and then co-transfected with pGL3-basic vector or the indicated P1 with or without IREs sequence deletion luciferase reporter gene plasmid and pRL-TK plasmid for 48 h. The cells were then treated with IFNγ for an additional 24 h. Luciferase activity was determined and normalized using the dual luciferase reporter system (Mean ± S.E.M.; n = 4; NS, no significant difference; * P < 0.05, ** P < 0.01, *** P < 0.001, Wilcoxon test)

The above findings indicated the effect of EZH2 in regulating IFNγ-induced PD-L1 expression. Next, we analyzed the relationship between EZH2 and PD-L1 expression in HCC tumors. Our recent study showed that Mφ-PD-L1 expression was related to the activated tumor microenvironment [32]. Transcriptome profiling and gene ontology biological analysis confirmed that Mφ-PD-L1⁺ HCC tumor samples displayed an immune-activated microenvironment and upregulated genes that were mainly involved in the IFNγ-mediated signaling pathway (Additional file 1: Figure S6a and b, Additional file 4: Table S8). Considering that IFNγ stimulation was demonstrated to be required for EZH2-mediated PD-L1 expression in the cell experiments, we analyzed the association between EZH2 and PD-L1 in samples with different microenvironments. Statistical analyses showed a significantly negative correlation between PD-L1 and EZH2 levels on TCs in immune-activated HCC tissues, but not in the total or immune-suppressed samples (Fig. 5a). Multiple immunofluorescence staining revealed that EZH2 protein was barely detected on either PD-L1⁺ Mφs or TCs (Fig. 5b). Collectively, these results suggested that EZH2 was negatively correlated with PD-L1 expression in the immune-activated HCC tumor microenvironment.Fig5Fig. 5

Correlation between EZH2 and PD-L1 expression in HCC tissues. a The expression correlation statistics of EZH2 and PD-L1 on hepatoma cells in immune-activated (left), immune-suppressed (middle), or all ungrouped (right) HCC tissues. b Representative images of multiple immunofluorescence staining showing the expression of EZH2 (gray) and PD-L1 (red) in HCC. Scale bar, 50 μm. #1361 and #1373 indicates the ID number of the HCC samples. The white stars and arrows indicate tumor cells expressing low or high levels of EZH2, respectively, and the five-pointed stars indicate the PD-L1⁺ Mo/Mφs with EZH2 low-expression. c Cumulative OS of EZH2 was calculated using the Kaplan–Meier method and analyzed using the log-rank test (P = 0.013). Cumulative OS of PD-L1 was calculated using the Kaplan–Meier method and analyzed using the log-rank test in patients with EZH2 low expression (d P = 0.025) and high expression (e P = 0.072). f Cumulative OS of EZH2 combined with PD-L1 was calculated using the Kaplan–Meier method and analyzed using the log-rank test (P = 0.002). Cumulative RFS of PD-L1 was calculated using the Kaplan–Meier method and analyzed using the log-rank test in patients with EZH2 low expression (g P = 0.020) and high expression (h P = 0.230). (* P < 0.05, *** P < 0.001)