Hypoxia-associated circPRDM4 promotes immune escape via HIF-1α regulation of PD-L1 in hepatocellular carcinoma

Cells were cultured either under normoxic conditions (21% O₂, 5% CO₂, and 74% N₂) or a hypoxic incubator (Memmert GmbH & Co. KG, Schwabach, Germany) providing a hypoxic environment (1% O₂, 5% CO₂, and 94% N₂) for 24 h.

Total RNA was isolated using TRIzol reagent (Invitrogen). For RNase R treatment, total RNA was incubated with RNase R (3 U/μg; Epicentre Technologies) at 37℃ for 20 min before RT-qPCR detection. Reverse transcription was performed using the PrimeScript RT Master Mix (TaKaRa, Dalian, China). qPCR was performed using TB Green Premix Ex Taq (TaKaRa). The primers for RT-qPCR used in this study are listed in Additional file 1: Table S1. The 2^−ΔΔCt method was used to normalize the data.

Cells or tissue sections were fixed with 4% paraformaldehyde for 30 min, permeabilized with 0.5% Triton X-100 in PBS for 20 min, and then blocked with 5% bovine serum albumin for 1 h. After incubation with the indicated primary antibodies (CD8α, 1:400 dilution, Abcam, Waltham, MA, USA; HIF-1α, 1:400 dilution, Cell Signaling Technology, Danvers, MA, USA) at 4℃ overnight, the cells were washed three times by PBS at room temperature and incubated with the fluorescence-conjugated secondary antibodies for 30 min. Coverslips were mounted on slides using antifade mounting medium. Cell nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI). Images were captured using a fluorescence microscope (Leica Microsystems, Mannheim, Germany).

FAM-labeled FISH probes specific to circPRDM4 (5’-TACACCCTGGCTTTGCGCACAAAC-3’) were designed and synthesized by GenePharma (Shanghai, China). FISH assays were performed using a Fluorescent In Situ Hybridization Kit (GenePharma) according to the manufacturer’s protocols. Briefly, cells were seeded on coverslips, fixed with 4% paraformaldehyde, and incubated with 0.5% Triton X-100 in PBS. After blocking, cells were incubated with FISH probes overnight and washed with SSC washing buffer. Cell nuclei were counterstained using DAPI. A confocal laser scanning microscope (PerkinElmer, Waltham, MA, USA) was used to capture the images.

Subcellular fractionation of cell extracts was performed using the PARIS™ Kit (Invitrogen) following the manufacturer’s protocol. Briefly, the cells were washed using PBS, added with ice-cold cell fractionation buffer, and incubated on ice for 10 min. After the samples were centrifuged at 4℃ and 500 × g for 5 min, the supernatant containing the cytoplasmic fraction were collected for subsequent analyses. The pellet containing the nuclear fraction were added with cell disruption buffer and then collected for further experiments.

The cells were centrifuged at 1000 × g for 5 min and collected. After incubating with the PE-conjugated PD-L1 antibody (1:200 dilution; BioLegend, San Diego, CA, USA) in dark for 30 min at 4 °C, cells were resuspended in FACS washing buffer on ice, and subjected to flow cytometry analyses. Samples were obtained and recorded in a FACS Aria II Cell Sorter (BD Biosciences, CA, USA), and data were analyzed with FlowJo software (TreeStar, Ashland, OR, USA).

T cells were activated using anti-CD3 (1 μg/ml) and anti-CD28 (5 μg/ml) antibodies (BD Biosciences, CA, USA) and human recombinant IL-2 (20 ng/ml) for 72 h. HCC cells with circPRDM4 knockdown or overexpression were seeded in plates overnight and incubated with activated T cells for 48 h. The cell culture media was removed. The adherent HCC cells were stained by crystal violent.

LDH release assay was performed using a LDH Cytotoxicity Assay Kit (Beyotime, Shanghai, China) according to the manufacturer’s instructions. Briefly, HCC cells were plated in 24-well plates in a density of 4 × 10⁴ cells/well. Activated CD8⁺ T cells (2 × 10⁵ cells/well) were added at an effector-to-target ratio of 5:1 and cocultured for 16 h. The culture media were centrifuged at 400 × g for 5 min, and the supernatants were transferred into 96-well plates (120 μl/well) and incubated with LDH detection reagent (60 μl/well) for 30 min at room temperature in dark. Absorbance at 490 nm was detected using a microplate reader (Bio-Tek Elx 800; Bio-Tek Instruments, Winooski, VT, USA).

To measure the levels of TNF-α and IFN-γ produced by cells, the supernatants were collected and the concentrations of TNF-α and IFN-γ were measured by the Human TNF-α ELISA Kit (MultiSciences, Hangzhou, China) and Human IFN-γ ELISA Kit (MultiSciences) according to the manufacturer’s guidelines. Optical densities were determined using a microplate reader (Bio-Tek Elx 800; Bio-Tek Instruments) at 450 nm.

For patient-derived xenograft (PDX) mouse model, 2 × 10⁶ patient-derived primary HCC cells mixed with Matrigel (Corning Inc., Corning, NY, USA) were subcutaneously transplanted in NCG mice. Human HCC tissues were minced into small pieces, treated with DNase I (20 μg/ml) and collagenase type II (100 U/ml in HBSS) for 30 min at 37 °C. Supernatants were collected and filtered through Falcon 70 μm Cell Strainer (Corning Inc.). The T cell fraction was collected using a prepared Percoll (GE Healthcare, Uppsala, Sweden) gradient by centrifuging at 800 × g for 30 min. Human CD8⁺ T cells were sorted from tumor infiltrating lymphocytes (TILs) by flow cytometer. CD8⁺ T cells were activated using anti-CD3 (1 μg/ml) and anti-CD28 (5 μg/ml) antibodies and human recombinant IL-2 (20 ng/ml) for 72 h. Each mouse was injected with 2 × 10⁶ of CD8⁺ T cells via caudal vein to reconstitute the human immune system. At day 7 after injecting with CD8⁺ T cells, circPRDM4 plasmid, empty vector plasmid, cholesterol-conjugated si-circPRDM4, or the negative control RNAi was intratumorally injected every three days for three weeks. Tumor growth was measured regularly. At day 28, mice were sacrificed and tumors were weighed and subjected for subsequent RT-qPCR and immunofluorescence analyses. To investigate whether knockdown or overexpression of circPRDM4 affect the proliferation of HCC cells under immunodeficient conditions, we also constructed the aforementioned PDX mouse models without adding immune cells to the mice. After intratumorally injection with circPRDM4 plasmid, empty vector plasmid, cholesterol-conjugated si-circPRDM4, or the negative control RNAi every three days for three weeks, the tumor growth and weight were compared between different groups.

Animal experiments were performed in compliance with the Animal Research: Reporting In Vivo Experiments (ARRIVE) guidelines and approved by the Institutional Animal Care and Use Committee of Nanjing Medical University (No. 10791) and the Animal Welfare and Ethical Committee of Wannan Medical College (No. LLSC-2020-090).

RNA pull-down assay was performed using a Magnetic RNA–protein Pull-down Kit (Thermo Fisher Scientific). Briefly, cell lysates were incubated with biotin-labeled probes (RiboBio, Guangzhou, China) and streptavidin magnetic beads (Invitrogen). After incubation at 4℃ overnight, beads were separated magnetically and washed five times. For mass spectrometry analysis, the beads were incubated with ddH₂O at 70℃ for 5 min. For western blotting, the beads were boiled in SDS for protein elution.

RIP assays were carried out using a Magna RIP RNA-Binding Protein Immunoprecipitation Kit (EMD Millipore, Billarica, MA, USA). Briefly, the cells were washed twice with ice-cold PBS, transferred to centrifuge tubes, and collected by centrifugation at 1500 rpm for 5 min at 4℃. Cells (1 × 10⁷) were resuspended in RIP Lysis Buffer and incubated on ice for 5 min. To confirm the binding between circPRDM4 and HIF-1α, 50 μl of beads coated with anti-HIF-1α antibody (5 μg; Cell Signaling Technology) or IgG antibody (5 μg; EMD Millipore) were incubated with the cell lysates in RIP Immunoprecipitation Buffer with rotating at 4℃ overnight. The protein-RNA complexes were washed six times with cold RIP Wash Buffer. The enriched RNA was purified and then isolated using TRIzol reagent and analyzed by RT-qPCR. To determine the HIF-1α domain required for circPRDM4 binding, cells were transfected with Flag-tagged bHLH (basic helix-loop-helix), PAS (Per-ARNT-Sim), or TAD (transactivation domain) truncated variants of HIF-1α, which were generated by GeneCreate (Wuhan, China). Cells were lysed and then incubated with anti-Flag antibody (5 μg; Cell Signaling Technology) or IgG antibody (5 μg; EMD Millipore) at 4 °C overnight. The abundance of protein-bound RNA was examined using RT-qPCR.

ChIP assays were performed using the Pierce Magnetic ChIP Kit (Thermo Fisher Scientific). Briefly, cells were crosslinked with 1% formaldehyde for 10 min at room temperature. Glycine was then added and incubated with the cells for 5 min to quench the crosslinking reaction. After washing twice with PBS, the cells were lysed in lysis buffer including protease and phosphatase inhibitors. Lysates were sonicated to shear DNA into smaller and workable pieces. Chromatin was incubated with anti-HIF-1α antibody (1:100 dilution; Cell Signaling Technology) or IgG antibody (2 μg; EMD Millipore) overnight at 4 °C. The following day, protein A/G magnetic beads were added and incubated for 2 h at 4 °C. After washing steps, DNA was eluted from beads and purified. The precipitated DNA was analyzed by qPCR or processed for ChIP-seq. CD274 promoter-specific primers used for qPCR were as follows: CD274 (− 2000 to − 1500 bp): Forward: 5’-ACACGAATCCTCACATTACT-3’; Reverse: 5’-AATCATATCCTCCTAGATGGC-3’; CD274 (− 1500 to − 1000 bp):Forward: 5’-TTCGGGAACTTTGGGAAG-3’; Reverse: 5’-GCTGACACTGCCTTGATT-3’; CD274 (− 1000 to − 500 bp):Forward: 5’-ATTATGACACCATCGTCTGT-3’; Reverse: 5’-TCGTGGATTCTGTGACTTC-3’; CD274 (− 500 to 0 bp):Forward: 5’-CAGATGTTGGCTTGTTGTAA-3’; Reverse: 5’-GTATCTAGTGTTGGTGTCCTA-3’.

For ChIRP assays, 4 × 10⁷ cells were harvested and washed with PBS twice. Cells were fixed in 1% formaldehyde in PBS for 10 min at room temperature. Crosslinking was quenched with glycine for 5 min. After centrifuging at 1500 × g for 5 min at 4 ℃, cells were washed by ice-cold PBS twice. Crosslinked cell pellets were resuspended in lysis buffer and sonicated to produce chromatin fractions. circPRDM4 probe was designed at the junction site. All probes were synthesized with BiotinTEG at the 3’ end. Lysates were incubated with probes at 37 °C for 4 h. Streptavidin beads were then added to isolate probe binding complex. After washing steps, the bound DNA was quantified with RT-qPCR with CD274 promoter-specific primers or processed for sequencing.

For dot blotting, wild-type or mutant circPRDM4 linearized RNAs were dropped onto Hybond-N⁺ membrane (Cytiva, Marlborough, MA, USA) followed by ultraviolet crosslinking. RNA signals were then detected using biotin-labeled single-stranded DNA segment of CD274 promoter. Biotin signals were detected with HRP-conjugated streptavidin according to the manufacturer's instructions (Thermo Fisher Scientific).

For IP between HIF-1α and p300/CBP, cells were dissolved in RIPA, added with beads, mixed with anti-HIF-1α antibody (5 μg; Cell Signaling Technology) or IgG antibody (5 μg; EMD Millipore) at 4 °C overnight. The next day, the immunoprecipitated proteins were added with SDS-PAGE loading buffer, heated at 95 °C for 10 min, and subjected to western blotting. The dilution for anti-p300 antibody (Abcam) and anti-CBP antibody (Abcam) was 1:1000.

Comparisons between groups were assessed using two-sided Student’s t test. Correlations between groups were evaluated using the Pearson’s correlation coefficient or Spearman’s ρ when appropriate. The significance of Kaplan–Meier survival curves was estimated with the log-rank test. All statistical analyses were performed using SPSS 26.0 (IBM Corporation, Armonk, NY, USA) or GraphPad Prism 9.0 (GraphPad Software, La Jolla, CA, USA), setting statistical significance at P < 0.05. *P < 0.05, **P < 0.01, ***P < 0.001, and “ns” stands for “no significance”.

Additional methodological details are included in Additional file 2: Supplementary Materials and Methods.

To identify hypoxia-associated circRNAs in HCC, we exposed HCC cell line HCCLM3 to normoxic or hypoxic treatment, and profiled the differentially expressed circRNAs. Compared to non-hypoxic controls, a total of 103 circRNAs with a > twofold change were identified in response to hypoxia (Fig. 1A). We selected the top five most regulated circRNAs for validation. Among these, hsa_circ_0007468 showed the greatest difference between hypoxic and normoxic HCCLM3 cells (Fig. 1B). The expression levels of hsa_circ_0007468 under hypoxic conditions were further assessed in several HCC cell lines. We observed high expression levels of hsa_circ_0007468 in MHCC97H and HCCLM3 cells, and low hsa_circ_0007468 levels in Hep3B and HepG2 cells (Fig. 1C), and these cell lines were chosen for subsequent experiments.

In silico analysis predicted that hsa_circ_0007468 was derived from exon 6 to exon 8 of the PRDM4 gene. As demonstrated in Fig. 1D, Sanger sequencing verified the back-splicing junction of hsa_circ_0007468 (hereafter referred to as circPRDM4). To confirm the circular characteristics of circPRDM4, we designed divergent and convergent primers to amplify the back-spliced and linear transcripts, respectively. circPRDM4 can be amplified by divergent primers from cDNA but not gDNA (Fig. 1E). When random primers were replaced by oligo dT primers, the relative expression of circPRDM4, but not PRDM4 mRNA, became barely detected (Fig. 1F). circPRDM4 was more resistant to RNase R degradation compared with the linear mRNA of PRDM4 (Fig. 1G). After transcription inhibitor actinomycin D treatment, circPRDM4 was more stable than linear PRDM4 mRNA (Fig. 1H, I). The circPRDM4 transcript was mainly observed in the nucleus of HCC cells using FISH and subcellular fractionation assays (Fig. 1J, K). Collectively, circPRDM4 is a hypoxia-associated circRNA in HCC.

Given that hypoxia is a key positive regulator of PD-L1 expression [12, 13], we set out to explore the role of hypoxia-associated circPRDM4 in HCC patients treated with PD-1 blockade. Twenty patients with advanced HCC treated with anti-PD-1 mAb were analyzed based on retrospective data. After 4 cycles of anti-PD-1 treatment, a responder is defined as having a partial response or a complete response, and a nonresponder is defined as having a progressive disease or stable disease, according to RECIST 1.1.

Representative radiographic results are shown to annotate the tumor diameter change pre- and post-immunotherapy in responders and nonresponders (Fig. 2A). Increased expression level of circPRDM4 was found in tumor tissues from responding patients compared to those from nonresponders (Fig. 2B). The difference in tumor diameter for all included patients is demonstrated in Fig. 2C. There was a significant negative correlation between the relative increase of tumor diameter and circPRDM4 expression level (Fig. 2D). Compared to patients with lower circPRDM4 expression, decreased CD8⁺ cell infiltration was observed in patients with higher expression of circPRDM4 (Fig. 2E). Furthermore, we detected a positive correlation between circPRDM4 expression level and CD274 mRNA level in HCC tissues (Fig. 2F). We observed prolonged progression-free survival (PFS) and overall survival (OS) in the group of anti-PD-1 mAb-treated HCC patients with high circPRDM4 expression (Fig. 2G, H). Together, circPRDM4 was upregulated in responders to PD-1 blockade and associated with therapeutic efficacy.

To examine the functional role of circPRDM4 in immune escape of HCC cells under hypoxic conditions, we disrupted circPRDM4 by short hairpin-mediated RNA interference in two HCC cell lines with high circPRDM4 level. The knockdown efficiency of circPRDM4 in MHCC97H and HCCLM3 cells was validated by RT-qPCR (Fig. 3A, B). No statistical difference in PRDM4 mRNA expression level was observed between HCC cells depleting circPRDM4 and the control cells (Fig. 3C). After hypoxia treatment, circPRDM4 depletion resulted in decreased expression levels of CD274 mRNA and total PD-L1 protein (Fig. 3D, E). Additionally, the knockdown of circPRDM4 reduced cell surface PD-L1 expression in HCC cells exposed to hypoxia treatment (Fig. 3F). In T cell-mediated cancer cell-killing assay, downregulation of circPRDM4 in MHCC97H and HCCLM3 cells rendered the cells more vulnerable to lymphocytes (Fig. 3G). Additionally, we found that circPRDM4 knockdown had no significant effects on HCC cell proliferation in the absence of immune cells (Additional file 3: Fig. S1A). HCC cells transfected with circPRDM4 short hairpin RNAs (shRNAs) or control cells were incubated with activated CD8⁺ T cell under hypoxia, and the cytotoxicity was measured by LDH release assays. T cell cytotoxicity assays based on LDH release confirmed aggravated cytotoxicity in circPRDM4-silenced MHCC97H and HCCLM3 cells (Fig. 3H). After hypoxia treatment, circPRDM4 depletion alleviated tumor cell-mediated immune suppression, as determined by the elevated levels of secreted TNF-α and IFN-γ from cocultured T cells (Fig. 3I).

HCC cell lines stably overexpressing circPRDM4 were successfully constructed (Fig. 4A, B). circPRDM4 overexpression did not change the expression of PRDM4 mRNA (Fig. 4C). Under hypoxic conditions, expression levels of CD274 mRNA, total PD-L1 protein, and cell surface PD-L1 were augmented in Hep3B and HepG2 cells overexpressing circPRDM4 (Fig. 4D–F). Upregulation of circPRDM4 significantly hindered T cell-induced cell death according to the results from T cell-mediated cancer cell-killing assays (Fig. 4G). Moreover, the effects on HCC cell proliferation after circPRDM4 overexpression were comparable in the absence of immune cells (Additional file 3: Fig. S1B), suggesting that circPRDM4 mainly functioned through regulating anti-tumor immune response. LDH release assays after hypoxia treatment showed that tumor cell death induced by T cells were thwarted after circPRDM4 upregulation (Fig. 4H). circPRDM4 overexpression compromised the secretion of TNF-α and IFN-γ in response to hypoxia (Fig. 4I). Collectively, these results suggested that circPRDM4 induced PD-L1 expression and promoted CD8⁺ T cell-mediated immune escape under hypoxic conditions.

To test the effects of circPRDM4 on HCC growth and immune evasion in vivo, we constructed a humanized mouse model as illustrated in Fig. 5A. Tumor infiltrating CD8⁺ T cells from HCC patients were subsequently adoptively transferred into immune-compromised NCG mice harboring HCC PDX. We observed that circPRDM4 overexpression markedly promoted tumor volume and weight, and knockdown of circPRDM4 significantly inhibited the growth and weight of tumors (Fig. 5B–F). The overexpression and knockdown efficiencies of circPRDM4 in tumors were verified using RT-qPCR (Fig. 5G, H). Additionally, in vivo experiments confirmed that circPRDM4 overexpression could increase CD274 expression levels, whereas knockdown of circPRDM4 led to a reduced expression level of CD274 (Fig. 5I, J). CD8 immunofluorescence staining of the tumors showed fewer CD8⁺ cells in the group of circPRDM4-overexpressing tumors and more CD8⁺ cells in the circPRDM4-silenced group, as compared to the corresponding controls (Fig. 5K). We further performed the animal experiments without adding immune cells to the mice. As shown in Additional file 3: Fig. S1C–I, we found that circPRDM4 knockdown or overexpression did not exert significant effects on HCC growth in vivo. Taken together, these data indicated that circPRDM4 promoted HCC growth and facilitated immune escape in vivo.

Given that circPRDM4 was mainly located in the nucleus, we wondered whether circPRDM4 could regulate the expression of its host gene. Results from RT-qPCR and western blotting assays showed that circPRDM4 had no influence on the expression of its host gene (Figs. 3C, 4C and Additional file 4: Fig. S2A). Furthermore, no protein-coding sequence could be found in circPRDM4 according to results from CPAT and circRNADb databases (Additional file 4: Fig. S2B, C).

To further understand the underlying mechanism of circPRDM4-mediated PD-L1 upregulation under hypoxic conditions, we took a circRNA pull-down approach to examining whether circPRDM4 could bind proteins. HCC cells were exposed to hypoxia treatment, and cell extracts were obtained. By incubating HCC cell extracts with a biotinylated probe complementary to circPRDM4, we enriched circPRDM4 and co-isolated a number of its potential binding proteins. Among the top candidates identified, HIF-1α most sparked our interest (Fig. 6A). Western blotting results confirmed the binding between circPRDM4 and endogenous HIF-1α under hypoxic conditions (Fig. 6B). In response to hypoxia treatment, circPRDM4 and HIF-1α colocalized intracellularly as shown in Fig. 6C. RIP assays confirmed the binding between circPRDM4 and HIF-1α (Fig. 6D).

To delineate the structural determinants for the interaction between circPRDM4 and HIF-1α, we first used RNAfold to construct the secondary structure of circPRDM4. Three well-characterized stem loop structures of circPRDM4 were predicted (Fig. 6E). RNA pull-down assays were performed with a series of circPRDM4 truncated fragments. As shown in Fig. 6F, the P1 fragment (nt 301–355 and 1–44) was found to be associated with HIF-1α. Next, we investigated the HIF-1α domain required for circPRDM4 binding by constructing Flag-tagged bHLH, PAS, and TAD of HIF-1α protein (Fig. 6G). We performed RIP assays and the results indicated that circPRDM4 bound to the bHLH domain but not to PAS nor TAD of HIF-1α (Fig. 6H). Together, we revealed that the P1 fragment of circPRDM4 bound to the bHLH domain of HIF-1α under hypoxia.

Next, we wondered how the interaction between circPRDM4 and HIF-1α promoted PD-L1 expression. Based on the results from Fig. 6I, J and Additional file 5: Fig. S3A, B, circPRDM4 did not affect the protein, mRNA, or subcellular distribution of HIF-1α. Previous studies demonstrated that HIF-1α could transcriptionally activate PD-L1 [12, 13]. Notably, the DNA-binding ability of HIF-1α is mediated by its bHLH domain [20]. Therefore, we speculated that circPRDM4 might affect the HIF-1α-mediated transactivation of PD-L1. As shown in Fig. 6K, we performed ChIP assays and found that HIF-1α could bind to the promoter of CD274 (− 500 to 0 bp). Previous report has identified the binding site of HIF-1α between -21 and -12 bp within CD274 promoter [21]. As indicated in Fig. 6L, ChIP assays further confirmed the binding of HIF-1α to CD274 promoter sequence (from − 21 to − 12 bp). circPRDM4 knockdown significantly suppressed the binding between HIF-1α and CD274 promoter (Fig. 6M). Results from dual-luciferase reporter assays demonstrated that HIF-1α overexpression elevated the luciferase activities of CD274 promoter compared to the negative control vector, whereas circPRDM4 overexpression alone exerted no significant effects on CD274 transcription (Fig. 6N). In an effort to assess whether circPRDM4 could influence the recruitment of p300/CBP during HIF-1α binding to DNA, we performed IP of p300/CBP and HIF-1α in circPRDM4-overexpressing or control cells. No significant differences in terms of interaction between p300/CBP and HIF-1α were observed (Additional file 6: Fig. S4).

We then performed ChIRP assays and verified the specific binding between circPRDM4 and – 500 to 0 bp region of CD274 promoter (Fig. 7A). The potential binding sites of circPRDM4 to CD274 promoter were predicted by two bioinformatics software programs (IntaRNA and LncTar). A position-wise minimal energy profile that visualizes the minimal energy for each intermolecular index pair is provided in a heatmap style (Fig. 7B). The results of bioinformatics analysis indicated that the region of circPRDM4 from 281 to 291 nt may be a site with high affinity for the − 170 to − 160 bp region of CD274 promoter (Fig. 7C). We generated the mutations of circPRDM4 transcript (circPRDM4-mut) to delete their base-pairing bases, followed by a hybridization assay. We found that circPRDM4-mut abolished the interaction between circPRDM4 transcript and CD274 promoter (Fig. 7D). Substitution mutation in CD274 promoter pairing region failed to activate luciferase activity by circPRDM4 overexpression in HIF-1α-overexpressing cells (Fig. 7E). In addition, DNA in situ hybridization showed that CD274 promoter colocalized with circPRDM4 and HIF-1α (Fig. 7F). We applied the circPRDM4 truncated fragments to perform rescue experiments. We found that the binding between HIF-1α and CD274 promoter (Additional file 7: Fig. S5A) and the luciferase activities of CD274 promoter (Additional file 7: Fig. S5B) were significantly decreased in the P1-Mut group and the P3-Mut group, further confirming our findings. Moreover, we stained HIF-1α, PD-L1, and circPRDM4 in the tumor tissues of clinical samples and animal experiments. As shown in Fig. 7G, H, we found that tumor sections with higher circPRDM4 level exhibited higher PD-L1 expression level, whereas no significant difference in HIF-1α expression pattern was observed.

We also assessed the effects of circPRDM4 overexpression on the expression levels of several HIF-1 target genes other than CD274 (Additional file 8: Fig. S6). No significant increase in the expression levels of these genes was detected. To compare the genome-wide occupancy profiles of HIF-1α and circPRDM4, we performed ChIP-seq for HIF-1α and ChIRP-seq for circPRDM4 in HCCLM3 cells exposed to hypoxia treatment. The ChIP-seq signal of the binding sites of HIF-1α is shown in Fig. 8A. The signals near the peak center of each HIF-1α-bound region are plotted in Fig. 8B. For ChIRP-seq analysis, global visualization of circPRDM4 signals around gene transcriptional start site is demonstrated in Fig. 8C, 8D. A total of 56 genes were intersecting between HIF-1α ChIP-seq data and circPRDM4 ChIRP-seq data (Fig. 8E). As shown in Fig. 8F, HIF-1α was enriched in the promoter region of CD274. ChIRP-seq analysis demonstrated that circPRDM4 was enriched in CD27 promoter region (Fig. 8G). These data reveal that HIF-1α and circPRDM4 can enrich in CD274 promoter region in HCC.

In addition, we silenced HIF-1α or PD-L1 in circPRDM4-overexpressing HCC cells. As shown in Additional file 9: Fig. S7A, B, HIF-1α knockdown led to a decreased level of CD274 in HCC cells overexpressing circPRDM4. According to the results from T cell cytotoxicity assays based on LDH release, HIF-1α knockdown compromised the immune evasion of circPRDM4-overexpressing cells (Additional file 9: Fig. S7C). These results indicated that circPRDM4 promoted PD-L1 transcription and HCC immune escape mainly through HIF-1α. Moreover, PD-L1 depletion hindered the immune escape of HCC cells caused by circPRDM4 overexpression (Additional file 10: Fig. S8A, B).

Taken together, these data indicated that circPRDM4 promoted HCC immune escape mainly through the HIF-1α/PD-L1 axis. A schematic model of circPRDM4 in immune escape of HCC cells under hypoxia is shown in Fig. 8H.