Cancer cell-derived exosomal circUHRF1 induces natural killer cell exhaustion and may cause resistance to anti-PD1 therapy in hepatocellular carcinoma

Six human HCC cell lines (HepG2, HCCLM3, SMMC-7721, Huh 7, PLC/PRF/5, and Hep3B) were cultured in Dulbecco’s modified Eagle’s medium (DMEM, HyClone, Cat: SH30243) supplemented with 10% fetal bovine serum (FBS, Gibco, Cat: 10100147). The NK-92 cell line was cultured in RPMI-1640 (HyClone, Cat: SH30809) supplemented with 20% FBS and 150 IU/mL recombinant human interleukin-2 (IL-2) (Novoprotein, Shanghai, Cat: GMP-C013). The K562 cell line was cultured in RPMI-1640 supplemented with 10% FBS. All of the above cell lines were cultured at 37 °C in a 5% CO₂ incubator.

The tissue samples used in this study were collected as described in Additional file 1: Supplementary Materials and Methods.

Exosomes from the serum of HCC patients and culture medium of HCC cells were isolated using ExoQuick Exosome Precipitation Solution (SBI System Biosciences, Cat: EXOQ5A-1) according to the manufacturer’s instructions. Then, the exosomes were examined by transmission electron microscopy as described previously [15‐17].

qRT-PCR and western blotting analyses were performed as described previously and in Additional file 1: Supplementary Materials and Methods [18]. The primers and antibodies used in this study are listed in Additional file 2: Supplementary Tables 1 and 2.

Immunohistochemistry was performed, and the intensity of the positive staining was measured as described in our previous study [18]. The anti-NKG2D antibody used in this study is listed in Additional file 2: Supplementary Table 2.

RIP assays were performed using a Magna RIP RNA Binding Protein Immunoprecipitation Kit (Millipore, Cat: 17–770) according to the manufacturer’s instructions. The anti-Argonaute 2 (AGO2) and IgG antibodies used in this study are listed in Additional file 2: Supplementary Table 2.

Biotin-labeled circUHRF1 and negative control probes (Supplementary Table 3) were synthesized by GenePharma (Shanghai, China). The circRIP assay was performed as described previously [18]. The sequence of the circUHRF1 probe is listed in Additional file 2: Supplementary Table 3.

The wild-type TIM-3 3′ UTR and circUHRF1 sequences were cloned into a pLG3 plasmid. The mutant TIM-3 3′ UTR and circUHRF1 pLG3 plasmids were generated using a mutagenesis kit (Qiagen, USA, Cat: 200521) according to the manufacturer’s instructions. Healthy donor-derived NK cells and NK-92 cells were seeded into 96-well plates and cotransfected with a luciferase reporter vector and miR-449c-5p mimics or the negative control using the Lipofectamine 2000 transfection reagent (Thermo Fisher, Cat: 11668–019). After 48 h, the firefly and Renilla luciferase activities were quantified with the Dual-Luciferase Reporter Assay System (Promega, USA, Cat: E1910).

NK cells (CD16⁺/CD56⁺) (Miltenyi, Cat: 130–092-660) and CD8⁺ T cells (Miltenyi, Cat: 130–045-201) were obtained from peripheral blood mononuclear cells (PBMCs) of healthy donors by positive selection using magnetic cell separation according to the manufacturer’s instructions.

The concentrations of IFN-γ and TNF-α produced by NK cells and CD8⁺ T cells were measured by IFN-γ and TNF-α ELISA kits (eBioscience, USA, Cat: KHC4021 and Cat: BMS223HS) in accordance with the manufacturer’s guidelines.

The pulldown assay was performed as described previously [19, 20]. In brief, to generate probe-coated beads, the biotinylated circUHRF1 probe, biotinylated circANRIL probe, and biotinylated negative control (NC) probe (GenePharma, China) were incubated with M-280 streptavidin magnetic beads (Invitrogen, USA, Cat: 11205D) at room temperature for 2 h. Then, approximately 1 × 10⁷ NK-92 cells were harvested, lysed, sonicated and incubated with probe-coated beads at 4 °C overnight. Subsequently, the RNA complexes bound to the beads were eluted and extracted with an RNeasy Mini Kit (QIAGEN, USA, Cat: 74104) and analyzed by qRT-PCR assay.

The in vivo metastasis assays are described in our previous studies [21] and in the Supplementary Materials and Methods.

All lentiviral vectors used in this study were purchased from Genomeditech (Shanghai, China). The circUHRF1-overexpression and circUHRF1-shRNA lentiviral vectors were transfected into NK-92 cells, and the blank lentiviral vectors were used as negative controls. The circUHRF1 shRNA target sequences used in this study are listed in Additional file 2: Supplementary Table 4.

Male NOD/SCID mice aged 4–6 weeks were maintained according to the stated guidelines. HCCLM3 cells (5 × 10⁶) with or without reduced circUHRF1 were injected into the mouse right flank to generate subcutaneous tumors. When, the tumor reached a volume of approximately 100 mm³, NK cells (1 × 10⁶) were resuspended in phosphate-buffered saline and injected intravenously through the tail vein. The mice were randomly assigned to four groups. Then, the mice received tail vein injection of Opdivo or its isotype control at 100 μg per dose three times a week for 2 weeks. Animals were euthanized when tumors reached a maximum of 1000 mm³ (n = 6). The day that the mice received the first therapy was considered day 1. Tumor volume was calculated as (length x width²)/2.

Statistical analyses were performed with SPSS software (19.0; SPSS, Inc., Chicago, IL). Values are presented as the mean ± standard deviation (SD). The statistical analyses are described in detail in the Supplementary Materials and Methods. P < 0.05 was considered statistically significant.

UHRF1 is frequently upregulated and plays a critical pathological role in HCC progression [14]. Furthermore, circRNAs were found to be produced from the back-splicing of precursor mRNAs (pre-mRNAs) and are considered byproducts of pre-mRNA splicing in cancer cells [22‐24]. Usually, the levels of circRNAs are in accordance with their respective mRNA levels [25]. Therefore, we examined the levels of 14 circular RNAs derived from UHRF1 by qRT-PCR (circular RNA sequencing data were obtained from the online database circBase). Among them, hsa_circ_0048677 (circUHRF1) was the most abundant circRNA and was upregulated in HCC tissues (Fig. 1a and b). Then, we examined the expression of circUHRF1 by qRT-PCR in 240 HCC tissues and matched adjacent nontumor tissues. CircUHRF1 expression in HCC tissues was significantly higher than that in adjacent nontumor tissues (tumor/adjacent nontumor tissues ≥2; 136/240) (Fig. 1c and d). Next, we explored the relationship between circUHRF1 expression and the clinicopathological characteristics of 240 HCC patients, and the results are listed in Table 1. The results showed that HCC patients with circUHRF1^high expression had larger tumor size (P = 0.001) (Fig. 1e), lower NK cell proportions in blood (P = 0.005) (Fig. 1f), and more microvascular invasion (P = 0.020) (Fig. 1g) than those with circUHRF1^low expression. Furthermore, we analyzed the prognostic potential of circUHRF1 expression in HCC patients. Importantly, Kaplan-Meier survival analysis indicated that patients with circUHRF1^high expression levels had a poor clinical prognosis (Fig. 1h and i). Multivariate analysis identified circUHRF1 expression as an independent predictor of overall survival (OS) and postoperative recurrence (Tables 2 and 3). In addition, we also analyzed the clinical data in the hepatitis B-positive patients (216/240) and found that the findings in hepatitis B-positive patients alone were consistent with those in overall patient population, which might be due to the large proportion of hepatitis B-positive patients in this cohort. Taken together, our data suggest that upregulated circUHRF1 expression in HCC cells is correlated with poor outcome in patients and indicate that increased circUHRF1 expression is likely involved in the progression of HCC.

Previous studies have reported that circRNA can be transferred from cell to cell via exosomes [26, 27]. Thus, we hypothesized that circUHRF1 could be transferred from HCC cells to other cells. To verify this hypothesis, exosomes from the supernatants of HCC cell lines were isolated and evaluated by TEM and western blotting (Fig. 2a and b). Then, qRT-PCR analyses in six HCC cell lines and paired exosomes showed that circUHRF1 was markedly upregulated in the HCCLM3 and SMMC-7721 cell lines compared with the other cell lines (Fig. 2c). Interestingly, circUHRF1 expression in exosomes was consistent with the parental cell expression levels (Fig. 2d).

Because exosomal circUHRF1 was detected in HCC-derived exosomes, we explored whether circUHRF1 was present in HCC patient plasma. Importantly, circUHRF1 was present in plasma exosomal RNA. Plasma exosomal circUHRF1 levels were increased in HCC patients compared with 20 healthy donors (Fig. 2e). Furthermore, plasma exosomal circUHRF1 levels were reduced after tumor resection and increased in patients with tumor relapse, indicating that plasma exosomal circUHRF1 was mainly produced by HCC cells (Fig. 2e). Interestingly, exosomal circUHRF1 levels were significantly increased in the plasma of patients with evidence of immune evasion mechanisms, which had a decreased NK cell proportion in peripheral circulation (Fig. 2f). These data suggested that plasma exosomal circUHRF1 could serve as a critical molecular determinant of NK cell-related immune evasion.

To determine whether HCC-derived exosomal circUHRF1 induces alterations in allogeneic NK cells, NK cells from 6 healthy donors were cocultured with HCCLM3 and SMMC-7721 cells at a 1:1 ratio for 24 h (Fig. 3a). Compared with NK cells cultured alone, NK cells cocultured with SMMC-7721 and HCCLM3 cells demonstrated significant impairment of IFN-γ and TNF-α secretion when subsequently cocultured with K562 cells (Fig. 3a-c). In addition, NK cells cocultured with SMMC-7721 and HCCLM3 cells demonstrated significantly increased circUHRF1 expression (Fig. 3d), which was impaired by the exosome inhibitor GW4869 (Fig. 3e). NK cells cocultured with SMMC-7721 and HCCLM3 cells prior to treatment with an exosome inhibitor exhibited attenuated impairment of IFN-γ and TNF-α secretion when subsequently cocultured with K562 cells (Fig. 3f and g). To further investigate whether the inhibitory effects of HCC cells on NK cell function were mediated in an exosomal circUHRF1-dependent manner, circUHRF1 knockdown cells were established using an shRNA lentiviral vector (which targeted the back-splice site of circUHRF1) in SMMC-7721 and HCCLM3 cells (Additional file 3: Supplementary Fig. 1a). To further verify that circUHRF1 shRNA had no effects on other UHRF1 splicing products, qRT-PCR was carried out to detect UHRF1 mRNA expression. The expression of UHRF1 mRNA showed no significant change after transfection with circUHRF1 shRNA, indicating the specificity of circUHRF1 silencing (Additional file 3: Supplementary Fig. 1b). Our results also showed that exosomal circUHRF1 expression was significantly decreased in circUHRF1 knockdown cells (Fig. 3h). Interestingly, the reduction in exosomal circUHRF1 levels effectively attenuated HCC-derived exosome-induced impairment of IFN-γ and TNF-α secretion in NK cells (Fig. 3i and j). These results collectively suggest that HCC-derived exosomal circUHRF1 impaired the functions of NK cells.

Previous studies have reported that circRNAs mainly function as miRNA sponges to bind miRNAs and then promote miRNA-targeted gene expression. We next investigated whether circUHRF1 could bind to miRNAs in human NK-92 cells. Through StarBase v3.0 analysis, 14 miRNAs were predicted as possible targets of circUHRF1. To verify the critical miRNAs that may interact with circUHRF1 in NK-92 cells, we used a circUHRF1-specific probe to carry out circRNA in vivo precipitation (circRIP) and detected the 14 predicted miRNAs in complex with circUHRF1 using qRT-PCR. Our results showed a significant enrichment of circUHRF1 and miR-449c-5p compared with the negative controls, while the other miRNAs had no enrichment or only slight enrichment (Fig. 4a), indicating that miR-449c-5p is a circUHRF1-interacting miRNA in NK-92 cells. Given that circRNAs act as miRNA sponges [18, 28, 29], we wondered whether circUHRF1 can interact with certain miRNAs in NK cells. Next, we conducted RIP with an anti-AGO2 antibody in NK-92 cells. The results showed that circUHRF1 and miR-449c-5p, but not circANRIL (a circular RNA reported to not bind to AGO2 [30]), were significantly enriched by the anti-AGO2 antibody (Fig. 4b). This result indicated that circUHRF1 may act as a binding platform for miR-449c-5p and AGO2. To further verify this result, a biotinylated miR-449c-5p pulldown assay was performed, and the results showed significant enrichment of circUHRF1 compared with the negative control in NK-92 cells (Fig. 4c). Next, we carried out a luciferase assay using miR-449c-5p mimics cotransfected with a pGL3 luciferase plasmid expressing the wild-type circUHRF1 sequence or an miR-449c-5p binding site-mutant circUHRF1 sequence into NK-92 cells and normal NK cells (derived from a healthy donor) (Fig. 4d). Compared with the negative control, miR-449c-5p mimics significantly reduced the luciferase reporter activity of the wild-type but not the miR-449c-5p-binding site-mutant pGL3 luciferase plasmid (Fig. 4e; Additional file 3: Supplementary Fig. 2). Moreover, the level of circUHRF1 was significantly decreased after overexpression of miR-449c-5p, and the level of miR-449c-5p was significantly decreased after overexpression of circUHRF1 in NK-92 cells (Additional file 3: Supplementary Fig. 3a and b; Fig. 3f and g). These findings suggest that circUHRF1 and miR-449c-5p may target each other. All of the above results confirmed that circUHRF1 and miR-449c-5p affected each other in NK cells.

Therefore, we speculated that circUHRF1 may inhibit miR-449c-5p activity to promote functional impairment of NK cells. To test this hypothesis, we ectopically overexpressed circUHRF1 together with miR-449c-5p in NK-92 cells to evaluate the regulation of miR-449c-5p by circUHRF1 (Additional file 3: Supplementary Fig. 4a). Although forced miR-449c-5p expression alone showed significant tumor-suppressive activity (promoting IFN-γ and TNF-α secretion), increased circUHRF1 expression dramatically inhibited the miR-449c-5p tumor-suppressive function (Additional file 3: Supplementary Fig. 4b), highlighting the critical ability of HCC-derived exosomal circUHRF1 to inhibit the function of NK cells via the miR-449c-5p-related pathway.

As our data showed that circUHRF1 effectively inhibited the function of miR-449c-5p to suppress NK cell secretion of IFN-γ and TNF-α, we speculated that circUHRF1 may be responsible for upregulating the expression levels of miR-449c-5p targets by acting as a miR-449c-5p sponge and enhancing immune evasion in HCC patients. To verify this hypothesis, we first analyzed a panel of predicted miR-449c-5p target genes in NK cells, which included TIM-3 (Fig. 5a). Next, we carried out a luciferase reporter assay. The pGL3 luciferase plasmid was created based on firefly luciferase expressing the wild-type 3′ UTR of TIM-3 or 3′ UTR TIM-3 sequence with a mutated miR-449c-5p-binding site. NK-92 cells and healthy donor NK cells were cotransfected with the pGL3 luciferase plasmid along with negative control or miR-449c-5p mimics and assayed for luciferase activity. Compared with the negative control, miR-449c-5p mimics significantly reduced the luciferase reporter activity of wild-type TIM-3 (Fig. 5b). Interestingly, we verified that TIM-3 expression was significantly decreased after forced expression of miR-449c-5p (Fig. 5c and d).

In light of the critical role of TIM-3-related pathways in immune evasion, we hypothesized that circUHRF1 may be an important contributor to HCC immune evasion via its competitive circRNA activity. Our results showed that TIM-3 mRNA is a target of miR-449c-5p in NK cells. Therefore, continuous upregulation of exosomal circUHRF1 sequesters miR-449c-5p in NK cells and thereby leads to immune evasion in HCC. To further validate our hypothesis, we detected TIM-3 expression in NK-92 cells transfected with miR-449c-5p mimic alone or together with circUHRF1. The results showed that forced miR-449c-5p expression led to decreased mRNA and protein expression levels of TIM-3, while circUHRF1 reversed this inhibitory effect of miR-449c-5p on TIM-3 expression (Fig. 5b and c).

To further confirm our in vitro results that circUHRF1 impeded NK cell function through regulation of the miR-449c-5p/TIM-3 axis, we investigated the correlation between circUHRF1 and miR-449c-5p/TIM-3 expression in another 40 HCC patient peripheral NK cells. We found that circUHRF1 expression was negatively correlated with miR-449c-5p expression in peripheral NK cells (R² = 0.3065; P = 0.0002; Fig. 5e). As expected, we also found that circUHRF1 expression positively correlated with TIM-3 mRNA expression in peripheral NK cells (R² = 0.4650; P < 0.0001; Fig. 5f). Notably, we observed that miR-449c-5p expression was negatively associated with the expression of TIM-3 mRNA in peripheral NK cells (R² = 0.1166; P = 0.0310; Fig. 5g), indicating that the upregulation of exosomal circUHRF1 abrogates the tumor-suppressive effect of miR-449c-5p on TIM-3 in NK cells.

Previous studies have reported that both NK cells and CD8⁺ T cells infiltrating tumors have increased TIM-3 expression [31‐33]. Therefore, we performed experiments to explore whether HCC-derived exosomal circUHRF1 inhibits the function of CD8⁺ T cells. Regrettably, compared with CD8⁺ T cells cultured alone, CD8⁺ T cells cocultured with SMMC-7721 and HCCLM3 cells demonstrated no significant impairment of IFN-γ and TNF-α secretion when subsequently cocultured with K562 cells (Supplementary Fig.5a and b). In addition, CD8⁺ T cells cocultured with SMMC-7721 and HCCLM3 cells demonstrated significantly increased circUHRF1 expression (Supplementary Fig.5c). By determining miR-449c-5p expression in CD8⁺ T and NK cells, we found that the difference in the abundance of miR-449c-5p was obvious, and was higher in NK cells than in CD8⁺ T cells (Supplementary Fig.5d). These results might be indicate that the circUHRF1/miR-449c-5p/TIM-3 axis is specific to NK cells and does not function in CD8⁺ T cells.

To further explore the function of circUHRF1 in HCC, we used the pLO5-ciR-circUHRF1 lentivirus vector to upregulate the expression of circUHRF1 in PLC/PRF/5 cells. qRT-PCR results showed that the circUHRF1 levels in PLC/PRF/5 cells transfected with pLO5-ciR-circUHRF1 were significantly higher than those in the cells transfected with pLO5-ciR-Mock (Fig. 6a). Importantly, exosomes derived from HCC cells transfected with pLO5-ciR-circUHRF1 had higher levels of circUHRF1 than those derived from pLO5-ciR-Mock-transfected HCC cells (Fig. 6b). Next, we generated a NOD/SCID mouse model (injected with NK cells derived from healthy donors) of tumor lung metastasis using PLC/PRF/5 cells. The results showed that the number of metastatic tumor nodules induced by PLC/PRF/5 cells transfected with pLO5-ciR-circUHRF1 was significantly increased compared with that induced by HCC cells transfected with pLO5-ciR-Mock (Fig. 6c). Importantly, the mice with tumor lung metastases derived from PLC/PRF/5 cells with forced circUHRF1 expression showed an increased circUHRF1 level in serum exosomes (Fig. 6d). Additionally, the number of NKG2D-positive cells in metastatic tumor nodules induced by HCC cells transfected with pLO5-ciR-circUHRF1 was significantly decreased compared with that in tumors induced by HCC cells transfected with pLO5-ciR-Mock (Fig. 6e). These results indicate that circUHRF1 promotes HCC progression in an exosome- and NK cell-dependent manner.

Because forced circUHRF1 expression suppresses NK cell function in HCC, we next evaluated whether circUHRF1 overexpression can further impede the antitumor effect of anti-PD1 therapy (Opdivo). Therefore, we analyzed retrospective data from 30 HCC patients with recurrent distant metastasis (including lung metastasis) receiving anti-PD1 immunotherapy who had undergone liver resection 2–36 months before immunotherapy. After six therapy cycles, the efficacy was assessed using CT. Based on RECIST1.1 analysis, the results showed that there were 3 patients with partial response (PR), 7 patients with stable disease (SD), and 20 patients with progressive disease (PD) (Additional file 3: Supplementary Fig. 6). Then, circUHRF1 expression levels were measured, and the results indicated that the circUHRF1 expression levels in the PD group were much higher than those in the SD and PR groups (Fig. 7a). To further explore the relationship between circUHRF1 and immune evasion, we examined the expression of NKG2D in tissues from 30 HCC patients. The number of NKG2D-positive cells in HCC patients resistant to anti-PD1 therapy was significantly reduced compared with that in HCC patients susceptible to anti-PD1 therapy (Fig. 7b). A scatter plot analysis revealed a negative correlation between circUHRF1 expression and NKG2D-positive cell number in HCC tissues (R² = 0.1452; P = 0.0377) (Fig. 7c). To further assess the effects of circUHRF1 on resistance to anti-PD1 therapy, we established a xenograft model by subcutaneous implantation of circUHRF1-knockdown HCCLM3 cells or the respective negative control cells. The results indicated that the implantation of circUHRF1 knockdown cells resulted in sensitivity to anti-PD1 treatment and an increase in the overall survival rate (Fig. 7d-f). Taken together, these findings indicate that forced circUHRF1 expression might impede the response of HCC to anti-PD1 treatment and that targeting circUHRF1 might be a promising and effective method to recover the sensitivity of HCC to anti-PD1 therapy.