IFNα Potentiates Anti–PD-1 Efficacy by Remodeling Glucose Metabolism in the Hepatocellular Carcinoma Microenvironment

Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related death worldwide (1). In the past few years, immune-checkpoint blockade (ICB) has revolutionized the approach to cancer therapy, and ICB has become a second-line option for patients with advanced HCC who fail sorafenib. Current ICB agents for HCC treatment mainly consist of anti–CTLA-4 and anti–PD-1/PD-L1 antibodies (2–4). However, objective response rates (ORR) were found to be approximately 15% in an Asian cohort of HCC patients treated with a single agent, such as nivolumab, pembrolizumab, or camrelizumab (2, 3, 5), suggesting that the complex immunosuppressive microenvironment of HCC leads to evasion of ICB via unknown mechanisms. Elucidating the mechanisms underlying the immunosuppressive microenvironment and subsequent remodeling to guide rational combination therapy remains a major challenge for therapeutic intervention in patients with HCC. Thus, the development of novel strategies to overcome intrinsic or adaptive resistance to immunotherapy in HCC has become an urgent need.

Responses to cancer immunotherapy are largely dampened by the immunosuppressive tumor microenvironment (TME; ref. 6). Due to increased glucose consumption in tumor cells, the TME becomes glucose-deprived in some tumor types (7). Glucose deprivation and hypoxia, induced by metabolically dysregulated cancer cells, restrict antitumor immunity by impeding the metabolic fitness of tumor-infiltrating cytotoxic CD8⁺ T cells (8). Hypoxia-inducible factor 1α (HIF1α) is the key regulator of the response to hypoxia in normal and cancer tissues; moreover, it is well-known that HIF1α regulates glycolysis and is a target for cancer therapy (9). HIF1α was shown to increase PI3K/AKT/mTOR activity in a mouse model, and it mediates glycolytic metabolic alterations that drive cancer progression and resistance to therapy (10). PD-L1 has recently been identified as a direct target of HIF1α (11), and elevated expression levels of HIF1α and PD-L1 have been associated with a poor prognosis in HCC (12).

Furthermore, metabolic pathways specifically regulate T-cell development, differentiation, and function in the TME (13). Glucose competition, amino acid competition, oxygen competition, and lactic acid secretion promote the formation of an immunosuppressive phenotype in the TME (13). The hypoxic TME also affects T-cell function, in which HIF1α drives CD8⁺ T-cell migration and effector function (14). Metabolically fit T cells contribute to relatively strong antitumor immunity by destroying tumor cells (15). Infiltrating immunomodulatory cells with different metabolic preferences also facilitate the establishment of an immunosuppressive TME (16). Blockade of PD-L1 under hypoxic conditions enhances myeloid-derived suppressor cell (MDSC)–mediated suppression of T cells present in tumors to construct an immunosuppressive TME (11). Therefore, simultaneous blockade of PD-L1 and inhibition of HIF1α may represent a novel approach for cancer immunotherapy. T-cell metabolism can be altered to perform different cellular functions in the TME (17). Although the metabolic interplay between tumor and immune cells plays an important role in immune response regulation (18), how metabolic changes in the tumor niche affect the development and maintenance of a favorable immune response and how the metabolic function of tumor or immune cells contributes to the outcome of ICB remain unclear.

IFNα is a cytokine belonging to the type I IFN family that exerts multiple biological effects, including antiviral and antitumor activities, in patients with defined types of cancer or viral diseases (19). IFNα has been widely applied to treat patients with chronic hepatitis B infection (20), but it is also a potential drug for combination with novel strategies for cancer therapy. Indeed, IFNα is considered a potential agent for tumor treatment, considering that it has been shown to directly block cell-cycle progression and induce apoptosis, which in turn inhibits tumor expansion by inducing the expression of tumor suppressor proteins and antitumor IFN-stimulated gene products (21–23). IFNα increases the surface expression of class I and II MHC molecules on tumor cells to enhance immune recognition (24). IFNα administration in combination with PD-1 blockade effectively augments antitumor immunity in tumor-bearing mice (25). The combination of an anti–PD-L1 antibody and IFNα has been reported to improve tumor targeting and antigen presentation as well as antitumor responses in a mouse model (26). Thus, a recent clinical trial identified IFNα plus an anti–PD-1 antibody as a promising treatment strategy in melanoma (27), indicating the potential value of IFNα-based combination ICB. Despite such results, the mechanism by which IFNα synergizes with ICB and whether the combination of IFNα with ICB can benefit HCC are still poorly understood.

Therefore, we aimed to investigate the mechanism by which IFNα improves the anti–PD-1 antibody–induced immune response in HCC. Here, we show for the first time that combination therapy comprising IFNα and an anti–PD-1 antibody could strongly inhibit HCC progression in both HCC mouse models and HCC patients. IFNα and anti–PD-1 combination treatment was found to impair glycolysis and glucose uptake in HCC cells and enhance glycolysis in tumor-infiltrating CTLs, which was attenuated after anti–PD-1 antibody administration. Our data reveal that combination therapy with IFNα and an anti–PD-1 antibody is a promising ICB strategy for HCC.

Although both PD-1 inhibitors and IFNα have been evaluated separately in HCC patients in clinical trials and showed promising activity, the combined effect of these two agents remains elusive in HCC. To this end, we observed the therapeutic efficacy of 15 patients with unresectable HCC who were treated with anti–PD-1–based immunotherapy combined with IFNα. Unexpectedly, we found that the ORR was 40.0% (6 of 15) and the disease control rate (DCR) was 80.0% (12 of 15), and only 20.0% of the patients (3 of 15) still progressed on therapy (Fig. 1A and B). Above all, no treatment-related deaths occurred. The clinical characteristics of all patients are listed in Supplementary Table S1. Here, we report on two patients with unresectable HCC who had previously failed PD-1 blockade but later showed a remarkable clinical response after treatment with an anti–PD-1 antibody in combination with IFNα.

Patient #1 was a 67-year-old female with unresectable HCC who underwent transarterial chemoembolization (TACE). Thereafter, the patient received PD-1 blockade therapy followed by TACE. However, she did not have a significant clinical response to the treatment and eventually progressed on PD-1 blockade. Abdominal MRI was performed, and her liver tumors had typical enhancement. The largest tumor (2.3 × 4.1 cm) was in segment VIII adjacent to branches of the hepatic portal veins. Unexpectedly, MRI scans taken about 7 weeks after the IFNα and anti–PD-1 combination treatment showed notable tumor shrinkage and necrosis in this patient, and a subsequent scan after 3 months of treatments confirmed the patient's remarkable response with minimal tumor activity (Fig. 1C). In addition, no further progression was noticeable on serial imaging. Notably, patient #1 had no other toxic effects or immune-related adverse events.

Patient #2 was a 34-year-old man with giant HCC. Previously, he had undergone multiple cycles of TACE as well as two hepatic resections, followed by administration of PD-1 blockade therapy. He appeared well on follow-up with no evidence of recurrence for 1 year. Two years later, he presented with dyspnea, and a chest CT scan showed >10 lung metastatic nodules, which were confirmed as metastatic HCC by thoracoscopic biopsy. Surprisingly, after 2-month IFNα and anti–PD-1 combination treatment, the restaging chest CT was performed and showed the complete eradication of numerous lung metastases in cross-sections taken at several levels and shrinkage of the largest nodule (1.4 cm) without evidence of newly developed metastatic disease (Fig. 1D).

Additionally, we analyzed the immune cells in the peripheral blood collected from patients who received an anti–PD-1–based immunotherapy combined with IFNα compared with those from pretreatment samples. In the analysis of the main immune cell types, we detected enrichment of CD8⁺ T cells but not CD4⁺ T cells in the peripheral blood using flow cytometry (Fig. 1E and F). We also noticed that there was no significant change of naïve CD8⁺ T cells, central memory CD8⁺ T cells, effector CD8⁺ T cells, and effector memory CD8⁺ T cells between pretreatment and posttreatment (Supplementary Fig. S1A–S1D). Collectively, our results demonstrate that the combination of IFNα and an anti–PD-1–based immunotherapy exhibits an acceptable toxicity profile with promising evidence of enhanced clinical benefit in patients with unresectable HCC.

To explore the mechanism underlying the improved efficacy of the combination therapy, we established orthotopic syngeneic mouse models of HCC with Hepa1-6 or H22 cells. We first assessed the effects of the combination therapy on immunocompetent orthotopic HCC mouse models (Fig. 2A; see Methods for the dosing schedule). There were no changes in body weight among the tested mice, suggesting the gross safety of the combination therapy (Supplementary Fig. S2A and S2B). In the Hepa1-6 model, compared with monotherapy with IFNα, an anti–PD-1 antibody, or an IgG control, the combination therapy resulted in significant Hepa1-6 tumor regression and tumor size reduction detected by an ultrasound imaging platform (Fig. 2B). Similar phenotypes were also observed in the spontaneous HCC model driven by c-Myc and immunocompetent H22 orthotopic tumor–bearing model mice (Fig. 2C; Supplementary Fig. S2C): The combination therapy led to greater tumor regression than either IFNα or anti–PD-1 treatment alone. Importantly, the overall survival (OS) of several HCC models including the c-Myc-driven spontaneous HCC model, the Hepa1-6 tumor–bearing mouse model, and the H22 tumor–bearing mouse model that received the combination therapy was significantly improved compared with that of corresponding mice that received either treatment alone (Fig. 2D and E; Supplementary Fig. S2D). IHC analysis further confirmed the spontaneous induction of HCC, as evidenced by universal nuclear staining of hepatocytic-specific transcription factor HNF4α but not biliary-specific transcription factor CK-19 (Supplementary Fig. S2E). We also observed that the combination therapy resulted in severe tumor necrosis and complete elimination of metastatic lesions in the lungs (Fig. 2F and G; Supplementary Fig. S2F–S2H).

To test the extent to which the TME contributed to the antitumor effect of IFNα, immunodeficient NOD CRISPR Prkdc II2rg (NCG) mice with established Hepa1-6 tumors were treated with IFNα (Supplementary Fig. S2I). We found that IFNα induced a minimal delay in tumor growth in the NCG mice, as indicated by the tumor size compared with that of vehicle control–treated mice, suggesting that the observed antitumor effect of IFNα largely depended on immune cells. To clarify whether IFNα directly acts on immune cells, we compared the expression pattern of IFNAR1, the specific receptor of IFNα, between HCC cells and CD8⁺ T cells derived from orthotopic models. Public single-cell sequencing data based on Ma and colleagues (28) indicated that tumor parenchymal cells show higher expression of IFNAR1 than CD8⁺ T cells (Supplementary Fig. S2J and S2K). Similarly, we found that HCC cells exhibited drastically higher expression of IFNAR1 in these orthotopic models compared with tumor-infiltrating CD8⁺ T cells (Supplementary Fig. S2L). Importantly, we also constructed Ifnar1-knockdown (KD) HCC cells by short hairpin RNA (shRNA)–mediated downregulation of Ifnar1. Orthotopic models were subsequently established with Ifnar1-KD Hepa1-6 cells and received the indicated treatment (Supplementary Fig. S2M). We found that knocking down Ifnar1 in HCC cells abolished the therapeutic potential of IFNα. Notably, IFNα treatment failed to synergize with anti–PD-1 antibody after silencing Ifnar1 expression (Supplementary Fig. S2M), highlighting that IFNα targeted tumor cells and tumor-intrinsic IFNAR1 was essential for IFNα-based therapy. Taken together, our data demonstrate that IFNα enhances the efficacy of anti–PD-1 therapy through the intact immune system.

Given the apparent dependency of the combination treatment on immune cell infiltration, we wanted to investigate the infiltrating immune cells in the TME. To conduct a high-dimensional characterization of the TME immune landscape of Hepa1-6 syngeneic mouse tumors from the four treatment groups, we first assessed the expression of 42 surface and intracellular immune markers on CD45⁺ cells from disaggregated Hepa1-6 tumors using time-of-flight mass cytometry (CyTOF). Our metal-labeled antibody panel was devised to identify the major tumor-infiltrating immune cell populations (Supplementary Table S2) and comprised cell lineage markers, function-related markers (homing and activation receptors), and cytokine-related markers. With this approach, we were able to divide CD45⁺ cells within TME into lymphoid and myeloid compartments (Fig. 2H; Supplementary Fig. S2N) and analyzed the expression of the corresponding lineage markers in each group (Supplementary Fig. S2O). However, after manually gating the percentages of each major immune lineage in each individual sample from the different treatment groups, we found that the tumor-infiltrating lymphocyte (TIL) density as well as the other major lineages showed no significant differences among the four groups, suggesting that the potential subgroup of immune cells might contribute to the antitumor effect of combination therapy (Fig. 2I). To explore the differentiation states of TILs within tumors derived from mouse models receiving different treatments, we further applied flow cytometry analysis to evaluate the exhausted states. We found anti–PD-1 treatment decreased the proportion of progenitor-exhausted CD8⁺ T cells (defined as TCF⁺TIM3⁻) compared with the vehicle group, while IFNα treatment exerted no significant effect on progenitor exhausted CD8⁺ T cells. Importantly, combination anti–PD-1 antibody and IFNα resulted in a decrease of progenitor-exhausted CD8⁺ T cells to the greatest extent (Supplementary Fig. S2P). Interestingly, although terminally exhausted CD8⁺ T cells (defined as TCF⁻TIM3⁺) were decreased after anti–PD-1 treatment, IFNα exerted more powerful induction effect on decreasing terminally exhausted CD8⁺ T cells than anti–PD-1 did. However, combination therapy showed no obvious synergic effects on reducing terminally exhausted CD8⁺ T cells (Supplementary Fig. S2Q). Thus, we found that IFNα treatment showed moderate effect on ameliorating the exhausted state of CD8⁺ T cells, and more importantly, IFNα potentiated the effect of anti–PD-1 antibody on reversing exhausted effector T cells in HCC. In summary, our results further validated the clinical efficacy of PD-1 blockade/IFNα combination therapy in immunocompetent orthotopic and spontaneous HCC model mice.

Given that cancer immunotherapy depends mainly on T-cell infiltration in the tumor immune microenvironment (TIME; ref. 29), we depleted CD4⁺ T cells, CD8⁺ T cells, or both to further assess which T-cell populations contribute to the improved anticancer effect of IFNα combined with an anti–PD-1 antibody (Fig. 3A). To deplete CD4⁺ T cells and CD8⁺ T cells in the Hepa1-6 tumor mouse model, we treated mice with anti-CD4– and anti-CD8–neutralizing antibodies or isotype control IgG. We found that depletion of CD8⁺ T cells or both CD4⁺ T cells and CD8⁺ T cells completely blocked the tumor regression induced by the combination therapy, whereas depletion of CD4⁺ T cells alone, similar to control treatment, had no such effect (Fig. 3B; Supplementary Fig. S3A and S3B). These results demonstrate that the combination of IFNα and an anti–PD-1 antibody eliminates tumor cells in a CD8⁺ T cell–dependent manner; thus, we focused on characterizing the role of CD8⁺ T cells in further analyses.

Several studies have discovered dynamic changes in CD8⁺ T-cell subpopulations during the antitumor immune response (30, 31). To further identify the specific subpopulations of CD8⁺ T cells that mediate the potent antitumor effect of the combination therapy, we computationally identified lymphoid cells in the CD45⁺ tumor-infiltrating cell population and reutilized the CyTOF-based systems approach to analyze 27 markers specific to CTL functions. This method yielded 21 distinct lymphoid subpopulations (including four distinct CD8⁺ T-cell clusters) broadly defined by the markers (Fig. 3C and D). Moreover, this method identified a significant subpopulation shift in CD8⁺ T cells within Hepa1-6 tumors among the four treatment groups. Notably, we found markedly elevated infiltration of the CD27⁺CD8⁺ T-cell subset in tumor tissues from mice that received the combination IFNα and anti–PD-1 therapy compared with those from mice on other treatments (Fig. 3E; Supplementary Fig. S3C), which was further confirmed by multicolor immunofluorescence (Fig. 3F). Furthermore, we adopted the HCC patient-derived organoid (PDO) and autologous intratumoral immune cell coculture model based on previously published methods (32, 33), which served as an in vitro platform to mimic the development of checkpoint inhibition blockade (Supplementary Fig. S3D). Similar to our in vivo results, IFNα was able to increase CD27 on CD8⁺ T cells, while the anti–PD-1 antibody failed to increase CD27. Combination IFNα and anti–PD-1 led to the increase of CD27 to the highest extent among all groups (Supplementary Fig. S3E and S3F).

We hypothesized that CD27 expression on CD8⁺ T cells might correlate with T-cell antitumor activities. We knocked down Cd27 in CD8⁺ T cells via Cd27-specific siRNA. After activation by anti-CD3 and anti-CD28 antibodies, CD8⁺ T cells derived from Hepa1-6 tumors with Cd27 KD exhibited attenuated effector function, as indicated by reduced secretion of proinflammatory cytokines such as TNFα, IFNγ, and GZMB compared with control CD8⁺ T cells transduced with nontargeting siRNA (Fig. 3G). Our findings were further supported by analysis of single-cell sequencing data. We generated integrated single-cell RNA sequencing (scRNA-seq) data of HCC TILs merged from two public HCC data sets (GSE140228 and GSE125449; refs. 28, 34) using Seurat's canonical correlation analysis (CCA; ref. 35). In accordance with our previous findings, the integrated single-cell data showed a differential CD27 expression profile among CD8⁺ T-cell subclusters (Fig. 3H), and the expression of genes important for T-cell cytotoxicity and proliferation (IFNG, ICOS, GZMA, GZMK, and STMN1) was increased in CD27^hiCD8⁺ T cells (Fig. 3I and J). Surprisingly, we noted that the expression of PDCD1 (encoding PD-1) was also elevated in CD27^hiCD8⁺ T cells (Fig. 3I and J). To confirm this finding, we separated tumor-infiltrating CD8⁺ T cells from patients with HCC and found that PD-1 expression in CD27^hiCD8⁺ T cells was significantly higher than in CD27^loCD8⁺ T cells (Fig. 3K and L). These findings provided a rationale for anti–PD-1 antibody–based immunotherapy. Additionally, the expression of genes involved in the effector memory CD8⁺ T-cell signature was elevated in CD27^hi T cells (Supplementary Fig. S3G). Furthermore, using tissue microarrays from our in-house HCC cohort, we found that CD27 was negatively correlated with FOXP3 and CD206, markers of regulatory T cells (Treg) and immunosuppressive macrophages, respectively (Fig. 3M).

Overall, we speculated that the CD27⁺CD8⁺ population is the subpopulation of CD8⁺ T cells that mediates the observed antitumor effects of the combination of IFNα and anti–PD-1 and that CD27 is the key factor in the effector function of CD8⁺ T cells.

To further elucidate the mechanism by which IFNα enhances the efficacy of anti–PD-1, RNA sequencing (RNA-seq) of tumor samples from the Hepa1-6 model was carried out to identify the transcriptional alterations after anti–PD-1 treatment with or without IFNα administration. Considering tumor purity, we selected samples composed of at least 80% tumor nuclei, which was validated by pathologists to reflect the alteration in malignant cells according to The Cancer Genome Atlas (TCGA) tissue sample requirements (available at https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga/studied-cancers; ref. 36). Gene set enrichment analysis (GSEA) showed that HIF1α signaling was positively enriched in the anti–PD-1 group compared with vehicle + IgG as the control group (P = 0.039; Fig. 4A). More importantly, HIF1α signaling was significantly inhibited following combination therapy administration (P < 0.001; Fig. 4B). To verify these findings, we separated malignant as well as CD8⁺ T cells from tumor tissues of all four groups via flow cytometry sorting and conducted immunoblotting analysis. We observed higher expression of HIF1α and its downstream targets, such as PKM2, LDHA, and HK1, in the surviving Hepa1-6 and H22 cells derived from C57BL/6J or BALB/c mice, respectively, that received anti–PD-1 compared with control-treated mice. IFNα treatment dramatically repressed the expression levels of these markers (Fig. 4C; Supplementary Fig. S4A). In contrast, the expression of HIF1α and its downstream targets in CD8⁺ T cells derived from both C57BL/6J and BALB/c mouse models was diminished by PD-1 blockade but could be successfully rescued by IFNα, indicating that the combination treatment maintained infiltrating CD8⁺ T cells in a highly glycolytic state (Fig. 4C; Supplementary Fig. S4A). Moreover, Hepa1-6 cells that survived anti–PD-1 treatment showed the highest proliferating cell nuclear antigen (PCNA) expression, accompanied by the lowest cleaved caspase-3 expression. In contrast, CD8⁺ T cells derived from residual foci in anti–PD-1 antibody–treated C57BL/6J mice showed the lowest expression of PCNA but the highest expression of cleaved caspase-3 (Supplementary Fig. S4B). As expected, the combination treatment reversed the expression patterns of PCNA and cleaved caspase-3. Glycolysis is a feature of both cancer cell progression and T-cell activation (17). Previous reports have suggested that microenvironmental glucose competition may be a driver of immunosuppression by depleting necessities required to maintain effector T-cell function, whereas blockade of glycolysis in malignant cells can boost the efficiency of immunotherapy and metabolic remodeling between tumor cells and T cells (37). By analyzing the fluorescence signal of CFSE and the percentage of IFNγ⁺ cells, we confirmed that increased glucose uptake drives CD8⁺ T-cell proliferation and activation (Supplementary Fig. S4C and S4D).

Therefore, in light of the necessity of HIF1α signaling in regulating tumoral glycolysis, we hypothesized that IFNα may also enhance anti–PD-1 therapeutic efficiency in HCC by regulating glycolysis. To verify this hypothesis, we first evaluated the glycolytic capacities of tumors and CD8⁺ T cells via a Seahorse extracellular acidification rate (ECAR) assay, which is a standard way to measure glycolysis in cells (38). The results showed that ECAR was significantly elevated in HCC cells derived from the groups that received anti–PD-1 treatment alone compared with those from the control groups in both C57BL/6J and BALB/c mouse models; however, ECAR was dramatically decreased after the addition of IFNα. Importantly, ECAR was the lowest in HCC cells derived from mice treated with the combination therapy when compared with either treatment alone or vehicle control treatment (Fig. 4D; Supplementary Fig. S4E). In contrast, ECAR was significantly decreased in CD8⁺ T cells derived from mice treated with anti–PD-1 treatment alone but increased after IFNα single reagent and was highest in CD8⁺ T cells from mice treated with anti–PD-1 and IFNα in combination in both C57BL/6J and BALB/c mouse models (Fig. 4E; Supplementary Fig. S4F). Accordingly, the microenvironmental glucose concentration was significantly decreased after anti–PD-1 treatment, but this effect was dramatically reversed by IFNα (Fig. 4F; Supplementary Fig. S4G). The proportions of IFNγ⁺ and CD27⁺ cells derived from both mouse models were also highest when mice were treated with both anti–PD-1 and IFNα (Fig. 4G and H; Supplementary Fig. S4H and S4I). These results indicated that HCC cells with enhanced HIF1α signaling activity and glycolytic potential survive; however, IFNα addition could restrain glycolysis in HCC cells, thus fostering activation of CD8⁺ T cells. To further exclude a direct effect of IFNα on CD8⁺ T cells, we also treated CD8⁺ T cells separated from the spleens of C57BL/6J mice with 500 IU/mL of IFNα, followed by functional assays. Flow cytometry assays demonstrated that IFNα treatment alone failed to increase CD27⁺ proportion (Supplementary Fig. S4J), proliferation (Supplementary Fig. S4K), or IFNγ secretion of CD8⁺ T cells (Supplementary Fig. S4L). Taken together, these data indicated that IFNα could modulate glucose competition between tumor cells and CD8⁺ T cells but not directly act on T cells, which might rely on HIF1α signaling in tumor cells.

Next, we aimed to determine whether HIF1α signaling drives the metabolic restriction of effector CD8⁺ T cells and impairment of anti–PD-1 treatment efficacy. First, we established an in vitro coculture system with syngeneic peripheral blood mononuclear cells (PBMC) derived from C57BL/6J or BALB/c mice and HCC cells in the presence of KC7F2 (a specific HIF1α inhibitor; ref. 39) or not (Fig. 4I). Immunoblot assays confirmed that HIF1α expression was extensively inhibited by KC7F2 in Hepa1-6 and H22 cells isolated from C57BL/6J or BALB/c mouse models, respectively (Supplementary Fig. S4M). Anti–PD-1 treatment resulted in a dramatically increased ECAR in both Hepa1-6 and H22 cells, and the effect was ameliorated by the HIF1α antagonist (Fig. 4J; Supplementary Fig. S4N). The ECAR of cocultured CD8⁺ T cells decreased afterward and was lowest when anti–PD-1 was used alone. In contrast, the combination of the HIF1α antagonist and anti–PD-1 antibody sustained the ECAR of cocultured CD8⁺ T cells at the highest level, which was observed in both coculture models (Fig. 4K; Supplementary Fig. S4O). Further experiments indicated that CD8⁺ T cells cocultured with HIF1α antagonist–pretreated Hepa1-6 or H22 cells supplemented with anti–PD-1 had the highest proportions of CD27⁺ and IFNγ⁺ cells (Fig. 4L and M; Supplementary Fig. S4P and S4Q). To further verify these in vitro results, we established Hif1α-KD Hepa1-6 cells to generate orthotopic HCC models in C57BL/6J mice and treated the mice with anti–PD-1. Tumor volume was assessed by b-mode ultrasound tomography (Fig. 4N). The results indicated that Hif1α KD exerted synergistic effects with anti–PD-1 treatment to inhibit tumor growth (Fig. 4O). Moreover, silencing Hif1α expression in HCC cells established a high-glucose microenvironment and exerted synergistic effects with anti–PD-1 treatment, as evidenced by this approach producing the highest microenvironmental glucose concentration (Fig. 4P). In concordance with the ex vivo glucose results, infiltrating CD8⁺ T cells from the Hif1α-KD group showed a significantly higher proportion of CD27⁺ cells after anti–PD-1 treatment (Fig. 4Q).

Subsequent in vitro cytotoxicity experiments were performed to verify whether IFNα promotes cytotoxic activities at the cellular level. In many studies, a CFSE killing assay has been widely used to detect the in vitro killing activity of cytotoxic T cells (40, 41). OT-I splenocytes were cocultured with EL4 cells in vitro, followed by different concentrations of IFNα treatment. Results showed that CD27⁺CD8⁺ T cells were also significantly increased in a dose-dependent manner in response to IFNα treatment (Supplementary Fig. S5A). In addition, combination treatment with IFNα and anti–PD-1 significantly improved T-cell cytotoxic activity (Supplementary Fig. S5B). The effects of the glucose concentration on the killing capacity of CD8⁺ T cells were also verified. Compared with those in the low-glucose (5 mmol/L) group, OT-I CTLs in the high-glucose (25 mmol/L) group showed a stronger killing ability with obvious concentration-dependent effects (Supplementary Fig. S5C).

Together, our accumulated results suggest that IFNα implementation can effectively enhance anti–PD-1 treatment efficacy in HCC by inhibiting HIF1α signaling in tumor cells while enhancing HIF1α signaling and cytotoxic activity in CD8⁺ T cells.

To elucidate the molecular mechanism by which HIF1α signaling is regulated in this process, we analyzed the differentially expressed genes in tumor samples from orthotopic HCC mice treated either with the anti–PD-1 antibody alone or with the combination of the anti–PD-1 antibody and IFNα. Venn diagram analysis identified seven overlapping transcripts with a negative fold change, which were ranked by P value, suggesting the expressions of these genes were significantly downregulated upon combination therapy with the anti–PD-1 antibody and IFNα compared with anti–PD-1 antibody monotherapy. Interestingly, FosB, which conventionally interacts with c-Jun to promote the DNA-binding activity of the AP-1 complex (42), showed the most significant difference in expression (Fig. 5A). Immunoblot assays confirmed that FosB expression was lower in Hepa1-6 or H22 derived from C57BL/6J and BALB/c mouse models treated with anti–PD-1 and IFNα than in those derived from mice that received anti–PD-1 single reagent (Fig. 5B; Supplementary Fig. S5D). As a classic oncogene, FosB is an important component of the transcription factor AP-1, which has been reported to be significantly implicated in accelerating HCC progression (43, 44). Considering that AP-1 is reported to interact with HIF1α and enhance HIF1α target transcription, we first validated the role of FosB in regulating glycolytic enzymes in HCC cells. We found that silencing FosB expression greatly decreased the expression of PKM2 and LDHA, two major rate-limiting enzymes involved in glycolysis, in Hepa1-6 and H22 cells transfected with FosB shRNA compared with those transfected with control shRNA (Fig. 5C; Supplementary Fig. S5E). Furthermore, coimmunoprecipitation (co-IP) assays demonstrated that FosB could physically interact with HIF1α (Fig. 5D and E) in Hepa1-6 cells. Next, the effect of FosB on HIF1α transcriptional targets was assessed by chromatin immunoprecipitation (ChIP)-qPCR assays. The results indicated that silencing FosB expression drastically decreased the occupation frequency of HIF1α on the promoters of its glycolysis-associated targets, such as Pfkl, Hk1, Pkm2, and Ldha; however, this inhibition effect was abolished when IFNα was applied (Fig. 5F–I). More­over, cotransfection of FosB and HIF1α dramatically boosted glycolysis-related enzyme expression as well as the ECAR, whereas HIF1α transfection alone produced only moderate alterations (Supplementary Fig. S5F and S5G).

Our above data imply that FosB might trigger downstream glycolytic targets of HIF1α transcription by enhancing HIF1α binding affinity for its target genes.

Considering that FosB mRNA expression was downregulated after IFNα treatment according to the RNA-seq results, IFNα might hinder FosB transcription. Interestingly, ChIP-PCR assays demonstrated reduced occupation of the FosB promoter by Pol-II after IFNα treatment (Fig. 5J). Moreover, decreased levels of H3K9ac in the promoter region of FosB were also observed (Fig. 5K), indicating that IFNα treatment could result in chromatin condensation in the FosB promoter region and inhibition of transcription initiation complex binding. A previous study revealed that IRF1, a key regulator of IFNα signaling, can counteract the pro-oncogenic function of FosB (45), suggesting a potential role for IRF1 in IFNα-regulated FosB expression. To verify this, we first evaluated IRF1 expression in HCC tissues from mice that received distinct treatments as described above by conducting immunoblot analysis. The results showed that IRF1 expression was greatly increased in HCC tissues after IFNα treatment, and the highest level was observed with anti–PD-1 treatment combined with IFNα administration (Fig. 5L). We also treated HCC cells with IFNα in vitro and found that IRF1 expression could be induced in a time-dependent manner, whereas FosB expression showed gradual downregulation (Supplementary Fig. S5H and S5I). Importantly, IFNα treatment induced increasing nuclear accumulation of IRF1, a critical intracellular translocation process for IRF1 function, in a dose-dependent manner (Fig. 5M; Supplementary Fig. S5J). In addition, IRF1 depletion abolished the inhibitory effects of IFNα on glycolytic enzyme expression (Fig. 5N). Together, these data imply that IFNα inhibits FosB expression by inducing IRF1-dependent transcriptional suppression.

Previous studies have revealed the vital role of mTOR signaling in sensing glucose in the microenvironment and modulating glycolysis in T cells (46). As glycolysis has been found to be a critical biological process for T-cell activation and antitumor cytotoxic functions (47), we hypothesized that the high-glucose microenvironment caused by the IFNα-induced inhibition of glycolysis in tumor cells may trigger mTOR signaling activation in CD8⁺ T cells to render them more cytotoxic toward HCC cells. Indeed, CD8⁺ T cells cultured in high-glucose medium exhibited higher levels of p-mTOR, HIF1α, PKM2, and LDHA than those cultured in low-glucose medium, which could be reversed with rapamycin, a specific inhibitor of mTOR signaling (Fig. 6A), indicating the essential role of mTOR signaling in regulating CD27 expression. We further deployed HCC PDO models and confirmed that the proportion of the CD27⁺CD8⁺ population among total CD8⁺ T cells was decreased when TILs were pretreated with rapamycin compared with the control group (Supplementary Fig. S6A). Collectively, these data suggest that mTOR sig­naling might contribute to increased CD27⁺CD8⁺ T cells resulting from IFNα and anti–PD-1 antibody combination therapy.

CD27 expression was greatly enhanced by IFNα and anti–PD-1 combination therapy, accompanied with dramatically increased p-mTOR, HIF1α, PKM2, and LDHA expression (Fig. 6B), further strengthening the necessity of mTOR sig­naling in CD27 expression in mouse models. We proceeded to examine the potential mechanism underlying how mTOR signaling regulated CD27 expression. To further identify upstream transcription factors that regulate Cd27 expression, we mapped our integrated single-cell data with potential transcription factors of CD27 in human leukocytes predicted by the Signaling Pathways Project (SPP) database (48). A total of 2,490 differential transcripts were identified between CD27^hi and CD27^lo CD8⁺ T cells. Furthermore, 57 of these transcripts were predicted to be transcription factors by the SPP database. Venn diagram analysis identified 12 overlapping genes in CD27^hi cells, which were ranked by P value. We observed that FOXM1, a member of the Fox transcription factor family (49), exhibited the greatest statistical significance and a noticeable fold change (Fig. 6C). To further confirm the bioinformatic results, we overexpressed Foxm1 via retrovirus infection in CD8⁺ T cells, followed by flow cytometry assay. We found that Foxm1 overexpression could not only significantly increase the proportion of CD27⁺CD8⁺ T cells (Supplementary Fig. S6B) but also induce CD8⁺ T-cell activation, as evidenced by increased IFNγ⁺ and IL2⁺ CD8⁺ T proportions (Supplementary Fig. S6C and S6D). On the contrary, silencing Foxm1 expression in CD8⁺ T cells via siRNAs achieved the opposite results (Supplementary Fig. S6E–S6G). In summary, these results suggest that FOXM1 is critical for CD8⁺ T-cell activation and FOXM1 might regulate CD27 expression in CD8⁺ T cells.

Interestingly, IFNα treatment of model mice enhanced FOXM1 expression in CD8⁺ T cells, and CD8⁺ T cells derived from mice treated with both IFNα and anti–PD-1 exhibited the highest level of FOXM1 expression (Fig. 6D). Moreover, in vitro experiments indicated that high-glucose medium could promote FOXM1 expression in the vehicle-treated group. However, both rapamycin and the HIF1α inhibitor could abolish this effect (Fig. 6E and F). To verify the vital role of FOXM1 in regulating CD27, we used a lentivirus to knock down Foxm1 expression in infiltrating CD8⁺ T cells derived from mice treated with IFNα and the anti–PD-1 antibody (Fig. 6G). We found that CD27 and PCNA expression was greatly downregulated; however, the p-mTOR level showed no obvious alteration, indicating that FOXM1 is downstream of mTOR. In contrast, as determined by immunoblot analysis, overexpression of FOXM1 in infiltrating CD8⁺ T cells derived from mice treated with anti–PD-1 therapy alone resulted in dramatic elevations in CD27 and PCNA expression without an alteration in the p-mTOR level (Fig. 6G).

To validate the correlation between FOXM1 and CD27 in clinical HCC specimens, TCGA analysis of HCC was conducted, and the data showed a significant positive correlation between CD27 and FOXM1 expression (Fig. 6H). In addition, our integrated single-cell data also showed a significant positive correlation between CD27 and FOXM1 expression in CD27⁺FOXM1⁺CD8⁺ T cells in HCC (Fig. 6H). Together, these data confirmed that a high-glucose microenvironment could trigger mTOR signaling activation, resulting in FOXM1 overexpression and subsequent Cd27 transcription. The combination of IFNα and anti–PD-1 showed the most significant capacity to enhance FOXM1 expression, subsequently resulting in the highest CD27 expression in HCC.

We observed that high expression of CD8 or CD27 indicated a significantly better prognosis in patients with HCC than did corresponding low expression, as evidenced by longer progression-free survival (PFS) and OS in data from Zhongshan Hospital (Fig. 7A). Moreover, combined analysis of CD8 and CD27 was also performed with tumor microarray chips. We classified patients with HCC into three subgroups: (i) low CD8 and low CD27 (n = 31), (ii) high CD8 and high CD27 (n = 38), and (iii) low CD8 and high CD27 or high CD27 and low CD8 (n = 111). We found that OS was remarkably shorter in group 1 than in group 2 (P < 0.001; Fig. 7A). Likewise, PFS was significantly shorter in group 1 than in group 2 (P < 0.001; Fig. 7A). We also validated our findings above and demonstrated that combined analysis of CD8 and CD27 provided a relatively powerful tool to predict the prognosis of patients with HCC using TCGA data (Fig. 7B). By using multicolor immunofluorescence, we also found that CD27⁺CD8⁺ infiltrating TILs correlated with the responses of HCC patients treated with PD-1–based immunotherapy (Fig. 7C). Moreover, analysis of single-cell sequencing data for TILs in patients with melanoma (GSE120575) and bladder cancer (GSE145281) treated with an anti–PD-1 antibody revealed that CD27⁺CD8⁺ subsets were significantly enriched in responders compared with nonresponders (Fig. 7D).

Together, these data confirmed that CD27⁺CD8⁺ T cells were associated with the survival of patients with HCC. Moreover, CD27⁺CD8⁺ T cells also correlated with the survival of HCC patients treated with PD1 blockade–based immunotherapy.

Anti–PD-1 antibody–based immunotherapy has become a promising cancer treatment strategy with durable clinical benefits in a wide range of cancer types (50). However, not all patients respond to anti–PD-1 monotherapy, and patients with HCC are often unresponsive (2, 3). Here, we report that combination therapy comprising IFNα administration and PD-1 blockade resulted in significant tumor repression in murine HCC models and patients with HCC. We found that IFNα induced IRF1 expression in HCC cells, thereby inactivating FosB transcription, which further attenuated the transcriptional activity of HIF1α and subsequently led to downregulation of glycolysis-related gene expression in HCC cells. Sequentially, combination IFNα and anti–PD-1 treatment remodeled the high-glucose microenvironment, which is essential for the activation of T cells and promoted T-cell costimulatory molecule (CD27) transcription via mTOR-induced FOXM1 expression in T cells (Fig. 7E).

It has been recently reported that some conventional reagents used in clinical practice, including metformin and cyclophosphamide, are able to synergize with PD-1 blockade and improve the ICB response (51, 52). As a major advantage, repurposing conventional drugs for a new use accelerates clinical application without new drug development. IFNα has been used to treat HCC for over 20 years, but its clinical benefit is limited, largely due to severe systemic side effects resulting from the regular dose used for treatment (53). These observations indicated that optimizing the dose and formulation of IFNα is essential to improve the therapeutic effects of IFNα. In our experiments, we found that 1 × 10⁴ IU of IFNα was sufficient to induce tumor regression in immunocompetent mice but not in immunodeficient mice without a significant change in body weight. These observations highlight for the first time that a suitable dose of IFNα is sufficient to inhibit tumor progression mainly through immune modulation rather than directly inducing tumor cell apoptosis (21). These findings are striking and provide a rationale for evaluating the effect of anti–PD-1 therapy on tumors exposed to IFNα. Our clinical data further provided early encouraging evidence for IFNα-based combination therapy, highlighting the potential of IFNα as a conventional drug in new applications for HCC treatment.

A “switch” from oxidative phosphorylation to glycolysis is a hallmark of T-cell activation, and this process is critical for meeting the metabolic demands of T-cell proliferation and activation (54). PD-1 signaling reprograms and inhibits glycolysis in T cells, and restricting glycolysis in tumor cells preserves the effector functions of T cells and can augment the response rate for ICB (55, 56). Interestingly, it has also been demonstrated that increased tumor glycolysis contributes to resistance to immunotherapy (57). Indeed, there is competition for glucose in the TME that can directly regulate T-cell function and ICB efficiency (58). In the current work, we were surprised to find that PD-1 blockade was able to upregulate glycolysis-related gene expression in a HIF1α-dependent manner. Strikingly, we found a previously unrevealed function of IFNα. Namely, IFNα was able to suppress tumor glycolysis by restraining expression of HIF1α-targeted master glycolytic transcripts including PKM2 and LDHA. Intriguingly, the suppressive effects of IFNα were preferentially limited in the tumor cell, evidenced by the fact that IFNα lost its suppressive effect when Ifnar1, the specific receptor of IFNα, was silenced in HCC cells. According to the scRNA-seq data of the Human Protein Atlas project, IFNAR1 showed relatively low expression on T cells (ref. 59; http://www.proteinatlas.org), and more importantly, we found that IFNAR1 expression was also higher in HCC cells when compared with infiltrating T cells according to single-cell sequencing data. Therefore, it was rational that HCC cells responded toward IFNα priorly, resulting in the selective effects of IFNα on HCC cells over T cells in HCC. Together, we uncovered a novel mechanism by which IFNα could indirectly reprogram glucose metabolism toward glycolysis in tumor-infiltrating CD8⁺ T cells by limiting tumor-imposed glucose competition, which increased glucose uptake by tumor-infiltrating CD8⁺ T cells and restored TIL function.

CD27 has been considered an activation marker of CD8⁺ T cells. We found that tumor-infiltrating CD27^hiCD8⁺ T cells express a higher level of activation-related genes, including CD28, IL2RA, IFNG, and GZMB, than those in CD27^loCD8⁺ T cells using single-cell sequencing data of tumor-infiltrating CD8⁺ T cells from patients with HCC. Moreover, we knocked down Cd27 in activated CD8⁺ T cells and found reduced secretion of effector cytokines including IFNγ, TNFα, and GZMB, supporting that the expression of CD27 was essential for CD8⁺ T-cell activation. We identified CD27⁺CD8⁺ T cells as the most enriched subtype after combination treatment with IFNα and an anti–PD-1 antibody. The proportion of CD27⁺CD8⁺ T cells might serve as a predictive biomarker for PD-1 therapy. CD27, which binds to its ligand, CD70, plays a critical role in T-cell activation, and agonistic anti-CD27 antibodies have shown great anticancer efficacy because CD27 signaling induction can enhance the proliferation and differentiation of effector and memory T cells (60). However, the transcriptional control of CD27 is poorly understood. Here, we report that IFNα-mediated increases in glucose uptake and glycolysis in T cells increased the levels of phosphorylated mTOR, which subsequently activated expression of a key transcription factor, FOXM1, to promote Cd27 transcription. Notably, it came to our attention that the expression of PDCD1 (which encodes the PD-1 protein) was also preferentially expressed in CD27^hiCD8⁺ T cells according to both public single-cell sequencing data and human samples. These data suggested that although CD27⁺CD8⁺ T cells induced by IFNα treatment acted as major effector T cells to attack tumor cells, their effector function was, at the same time, limited by feedback-high expression of PD-1. Therefore, anti–PD-1 treatment is essential to liberate the immunologic effect of CD27⁺CD8⁺ T cells. Collectively, our results provide secondary evidence explaining the synergistic effect of IFNα and anti–PD-1 antibody combination treatment. The clinical data on tumor regression in 15 HCC patients treated with IFNα combined with anti–PD-1 indicated enrichment of CD27⁺CD8⁺ T cells, which will help establish the translational potential and clinical applicability of this work. These findings could also deepen our understanding of the mechanism influencing the outcome of PD-1 blockade in patients with HCC, the potential factors related to treatment efficacy, or the factors related to selecting the right patient population for treatment. Addressing the translational and clinical potential of this work would advance the field and has potential immediate relevance to patient care. Compared with previously published data (27), our research explored the broader applicability and/or potential limitations of this therapy.

How metabolic changes in the TIME affect the development and maintenance of a favorable immune response to ICB remains unclear. We identified the role of IFNα in enhancing anti–PD-1 efficacy by suppressing HIF1α-induced glycolysis in HCC and found that IFNα suppresses FosB in an IRF1-dependent manner. We first found that IFNα could restrain the glycolytic capacity of tumor cells to enhance microenvironmental glucose levels, resulting in CD27⁺CD8⁺ T-cell infiltration. This result suggests how the metabolic function of tumor cells contributes to disease outcomes. Mechanistically, we also found that a high-glucose microenvironment could trigger CD27 expression in CD8⁺ T cells by activating the mTOR–FOXM1 axis. The mechanism will help to identify potential biomarkers to stratify patients who may respond to IFNα and anti–PD-1 combination treatment, and understanding the molecular mechanism will be of interest. The combination of IFNα and an anti–PD-1 antibody improves immunotherapeutic efficacy in patients through multiple mechanisms. We demonstrate a novel mechanism through which IFNα can reshape the HCC TME to enhance glucose metabolism.

Although our study indicated that IFNα combined with anti–PD-1 could be of therapeutic value in HCC, additional clinical studies in larger patient cohorts (with or without hepatitis etiology), as well as additional transgenic mouse models of HCC, are needed to further validate this hypothesis. In addition, unintended side effects should be carefully and systematically evaluated. Importantly, our work provides a potential application of IFNα and anti–PD-1 combination therapy for the treatment of other additional cancer types, as tumor glycolysis is a hallmark of cancer and a critical barrier to successful immunotherapy in several types of cancer models, including melanoma, breast cancer, and lung cancer models. Our work also highlights the clinical value of tumor-infiltrating CD27⁺CD8⁺ T cells as a biomarker for stratifying patients for anti–PD-1 treatment in tumors other than HCC. However, these applications still need to be further investigated.

In conclusion, we found that the combination of IFNα and anti–PD-1–based immunotherapy exhibited promising anticancer effects in 15 patients with HCC. Our work reveals a novel mechanism of the synergistic efficacy of the combination of IFNα and an anti–PD-1 antibody in HCC. The combination therapy greatly induced CD27⁺CD8⁺ T-cell infiltration, remodeled the TIME, and subsequently regressed HCC tumors. Mechanistically, IFNα corrected tumor-imposed glucose competition in the HCC microenvironment and restored the effector function of CD8⁺ T cells by increasing CD27 transcription. The clinical applicability of these data could potentially advance HCC therapy. Our data also imply the potential for broader applicability of IFNα with other ICBs in addition to PD-1 blocking and for evaluating IFNα and anti–PD-1 combination therapy in other cancer types. In summary, our data provide the initial clinical evidence and mechanistic insight on a novel combination strategy to improve the clinical outcome of anti–PD-1 antibody–based immunotherapy in HCC.

A tissue microarray composed of HCC samples was obtained from Zhongshan Hospital, Fudan University, Shanghai, China (n = 180). Biopsies were also obtained from patients with unresectable HCC (n = 19) treated with PD-1–based immunotherapy to evaluate the correlation between CD27⁺CD8⁺ TILs and drug response by multicolor IHC analysis. Radiographic assessment of response by independent radiology review was performed. Tumor response for patients was assessed by modified RECIST (mRECIST; ref. 61). IFNα is approved by the FDA for the treatment of chronic hepatitis B. Evidence supports that IFNα directly and indirectly modulates immune responses and may add to PD-1 blockade in enhancing immune and clinical responses to solid tumors (27). In addition, several studies indeed have proven that IFNα may be an effective treatment and can prolong survival in patients with HCC (62, 63), which prompted us to observe the efficacy of PD-1 blockade (200 mg, every 2/3 week, intravenous) combined with pegylated-interferon alfa-2b (3 μg/kg, every week, subcutaneous). Peripheral blood samples were collected from patients (n = 8) before/after combination treatment to measure lymphocyte levels using flow cytometry based on the prospective clinical trial (NCT04639284). In addition, a total of 15 consecutive patients treated with IFNα combined with anti–PD-1–based immunotherapy were included in the efficacy observation. Clinicopathologic information was summarized (Supplementary Table S1). These patients were enrolled under Institutional Review Board (IRB)–approved protocols (Approval No. B2020-177R).

These studies were conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonization Guidelines for Good Clinical Practice. All patients signed written informed consent prior to having any study procedures performed.

HEK293T (RRID: CVCL_0063), Hepa1-6 (RRID: CVCL_0327), and H22 (RRID: CVCL_H613) cell lines were kindly provided by Dr. Xun Huang at Lingang Laboratory, Shanghai, China. HEK293T and Hepa1-6 cells were cultured in DMEM containing 10% FBS (Gibco; #10099141C) and 100 U/L penicillin/streptomycin (Gibco; #15140122) at 37°C and 5% CO₂ in an incubator. H22 cells were cultured in RPMI 1640 (Gibco; #61870036) supplemented with 10% FBS (Gibco; #10099141C) and 100 U/L penicillin/streptomycin (Gibco; #15140122) at 37°C and 5% CO₂ in an incubator. Both of the cell lines were routinely authenticated and were free of Mycoplasma contamination. The EL4 (RRID:CVCL_0255) cell line was purchased from the ATCC and cultured in DMEM containing 10% FBS (Gibco; #10099141C) and 100 U/L penicillin/streptomycin (Gibco; #15140122) at 37°C and 5% CO₂. Cell lines were authenticated using STR profiling and used at low passage numbers. Infiltrating CD8⁺ T cells were cultured in RPMI 1640 medium (Gibco; #61870036) supplemented with 10% FBS (Gibco; #10099141C), 1% penicillin/streptomycin (Gibco; #15140122), 5 μg/mL SIINFEKL peptide (Sangon Biotech; #T510212), and 20 ng/mL mIL2 (R&D Systems; #402-ML) at 37°C and 5% CO₂. Additionally, OVA-specific CD8⁺ CTLs were generated by incubating OT-I mouse splenocytes in low-glucose (5 mmol/L) or high-glucose (25 mmol/L) RPMI 1640 medium for 7 days. On the other hand, HCC cells derived from orthotopic models were cultured in DMEM (Gibco; #11965092) supplemented with 10% FBS (Gibco; #10099141C) and 1% penicillin/streptomycin (Gibco; #15140122) in a 37°C and 5% CO₂ incubator.

All animal care and experimental protocols performed were approved by the Institutional Animal Care and Use Committee of Zhongshan Hospital, Fudan University. To establish a Myc-induced spontaneous HCC model, 50 μg pT3-EF1a-c-myc (RRID: Addgene_92046) and 10 μg Sleeping Beauty SB100x transposase-encoding plasmid (RRID: Addgene_34879) per mouse dissolved in 2 mL of 0.9% NaCl solution were injected into C57BL/6J mice through tail-vein injection within 3 to 5 seconds. For orthotopic HCC models, 6-week-old male C57BL/6J mice (Charles River Company) were injected subcutaneously in the right flank with 5 × 10⁶ Hepa1-6 cells in 100 μL PBS and 6-week-old male BALB/c mice (Charles River Company) were injected subcutaneously in the right flank with 5 × 10⁶ H22 tumor cells in 100 μL PBS. Two weeks after cell implantation, the tumors derived from Hepa1-6 or H22 tumor cells were dissected and cut into small cubes (1 mm³) under aseptic conditions. Single cubes of equal volume were implanted into the right lobe of the liver parenchyma of 6-week-old C57BL/6J mice or BALB/c mice. Ifnar1-deficient HCC cells were used to establish an orthotopic HCC model using C57BL/6J mice. NCG mice (GemPharmatech) were also used to establish an orthotopic HCC model using Hepa1-6 cells following the aforementioned schedule. Three days after implantation, mice bearing Hepa1-6 or H22 orthotopic tumors were randomized into the indicated groups. For combination therapy, C57BL/6J or BALB/c tumor–bearing mice were treated intraperitoneally with 1 × 10⁴ IU mIFNα (Miltenyi Biotec; #130-093-130) and/or 10 mg/kg anti–PD-1 monoclonal antibody (Bio X Cell, BE0146, RRID: AB_10949053) or isotype control monoclonal antibody (Bio X Cell, BE0089, RRID: AB_1107769) following the schedule shown in Fig. 2A. Briefly, the anti–PD-1 antibody was administered 3 times per week for 4 weeks, and mIFNα was given daily for 4 weeks. For the in vivo T-cell depletion studies, the indicated mice were administered anti-CD8α (Bio X Cell, BE0061, RRID: AB_1125541), anti-CD4 (Bio X Cell, BE0003-1, RRID: AB_1107636), or both the anti-CD8α and anti-CD4 antibodies [250 μg in 100 μL PBS (intraperitoneally)] following the schedule shown in Fig. 3A. For mIFNα monotherapy studies, NCG tumor-bearing mice were injected intraperitoneally with 1 × 10⁴ IU mIFNα or 0.9% NaCl as a control. Tumor growth was monitored daily by a Preclinical Vevo 3100 micro-ultrasound imaging platform (VisualSonics) and measured every 3 to 5 days. B-mode, or brightness mode, imaging was used to acquire 3D images of an area of interest and for identification of liver tumors. Images were quantified for tumor volume using VevoLAB analysis software (VisualSonics). All mice used in this study were housed under specific pathogen–free conditions in the animal center of Zhongshan Hospital. Mouse body weight was measured daily. Mice were routinely monitored to observe the effects of tumor growth and treatment on normal activity, food and water consumption, weight gain/loss, and other features. All experiments were conducted to reduce the number of mice as much as possible and alleviate their suffering. All studies that employed anesthesia or euthanasia methods were consistent with the commonly accepted norms of veterinary best practices (e.g., CO₂). At the end of the indicated therapy, mice were sacrificed, and livers bearing tumors were dissected. Tumor volume was calculated using the formula length (mm) × width² (mm) × 0.5.

For assaying the effects of rapamycin on IFNα plus anti–PD-1 treatment in HCC, an HCC PDO and autologous intratumoral immune cell coculture model was established according to published methods (32, 33) with optimization. Briefly, primary HCC patient resection tumors were mechanically minced and dissociated into 1 mm³ fragment; diluted in Advanced DMEM/F-12 (Thermo Fisher Scientific; #12634010) with 0.1% collagenase, Type 4 (Worthington; #LS004188), 0.05% hyaluronidase (Sigma; #H3506), and 0.01% deoxyribonuclease (Sigma; #DN25); and shaken on a horizontal platform for 30 minutes at 37°C. The homogenate was filtered through a 100-μm filter and then a 40-μm filter, pelleted, and washed in 1× red blood cell lysis buffer (BD; #555899). The spheroids between 40 and 100 μm were then resuspended in the classic human liver organoid isolation medium for 3 days (64). Cells less than 40 μm were isolated for TILs by a Ficoll density method (65). TILs were cultured in ImmunoCult-XF T-cell Expansion Medium (STEMCELL; #10981) with ImmunoCult Human CD3/CD28 T-cell Activator (STEMCELL; #10971) and 30 ng/mL IL2 (R&D Systems; #202-IL-050) with or without 200 ng/mL rapamycin for 3 days. On day 3, HCC organoids and autologous activated TILs were directly cocultured at a 5:1 ratio in Matrigel (Corning; #356231) in a 24-well plate and overlayed with human liver organoid isolation medium with 1 × 10⁴ IU/mL IFNα (Miltenyi Biotec; #130-093-876) plus 100 μg/mL anti–PD-1 (Bio X Cell; cat. #BE0188, RRID: AB_10950318) after incubation in a 37°C and 5% CO₂ culture incubator for 10 minutes. After 72 hours of coculture, the specific killing efficiency of T cells and CD27⁺CD8⁺ population were evaluated by flow cytometry to explore whether rapamycin could blockade the effects of IFNα treatment. The following antibodies were used for flow cytometry staining: CD45 (PerCP cy5.5, BioLegend; cat. #982318, RRID: AB_2888786), CD8 (APC-A750, BioLegend; cat. #344722, RRID: AB_2075388), CD27 (PE, Thermo Fisher Scientific; cat. #MA5-17905, RRID: AB_2539289), PD-1 (BV421, BioLegend; cat. #329920, RRID: AB_10960742).

Freshly harvested tumors were treated with the Mouse Tumor Dissociation Kit (Miltenyi Biotec; #130-096-730) according to the manufacturer's instructions. Briefly, dissociated tumor samples were filtered through a 70-μm strainer and centrifuged with 36% Percoll reagent to remove cell debris. ACK lysis buffer was then utilized to lyse erythrocytes. The cells were resuspended in 100 μL Living Cell Staining Mix (250 nmol/L cisplatin in PBS) and incubated for 5 minutes on ice. The samples were blocked and stained for 30 minutes with a surface antibody panel developed in-house, followed by fixation overnight. For intracellular marker staining, permeabilization buffer (eBioscience) was utilized following the manufacturer's instructions. Cells were then washed, resuspended, and stored at 4°C before acquisition on a Helios instrument (Fluidigm). Antibody clones and suppliers are listed in Supplementary Table S2.

Mass cytometry data were randomized and normalized before processing as previously described (66). A doublet filtering scheme was used on mass-tagged barcodes for differing samples (67). Generally, the gating strategy was as follows: cells, live cells, singlets, and valid immune cells (CD45⁺). Separate t-distributed stochastic neighbor embedding (t-SNE) plots and heat maps were developed for all CD45⁺ cells to specifically characterize the T-cell clusters. X-shift was applied to the T-cell compartment defined in the first X-shift analysis. MHC-II, B220, CD21, IgM, IgD, CD11c, CD19, F4/80, CD115, iNOS, Arginase I, CD49b, CD49a, CD23, and CD11b were removed from the analysis to exclude markers that were not expressed on T cells and reduce the likelihood of adding noise during cluster generation.

Cultured cells or mouse tumor tissue proteins were dispersed in RIPA Lysis and Extraction Buffer (Thermo Fisher Scientific; #89901) supplemented with 1/100 Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific; #78446) for 30 minutes. All the processes were performed on ice. For mouse tumor tissue samples, the tissues were first homogenized for 15 minutes. Cell or tissue lysates were centrifuged at 12,000 rpm for 15 minutes at 4°C. Supernatants were collected and evaluated by a BCA protein concentration assay (Beyotime; #P0010). Then, 5× SDS loading buffer was added to the protein samples, and the samples were boiled at 95°C for 5 minutes. A total of 50 μg protein samples was loaded onto a 10% SDS-PAGE gel, separated by electrophoresis, and then transferred onto a polyvinylidene fluoride membrane (Millipore). After transfer, membranes were blocked with TBST containing 5% skim milk for 2 hours at room temperature. The membranes were then incubated with the indicated primary antibody specific for PKM2 (Cell Signaling Technology; cat. #4053, RRID: AB_1904096, 1:1000), LDHA (Cell Signaling Technology; cat. #3582, RRID:AB_2066887, 1:1,000), HK1 (Cell Signaling Technology; cat. #2024, RRID: AB_2116996, 1:1000), HK2 (Cell Signaling Technology; cat #2867, RRID: AB_2232946, 1:1000), FosB (Cell Signaling Technology; cat. #2251, RRID: AB_2106903, 1:1,000), β-actin (Cell Signaling Technology; cat. #4970, RRID: AB_2223172, 1:1,000), histone H3 (Cell Signaling Technology; cat. #4499, RRID: AB_10544537), IRF1 (Cell Signaling Technology; cat. #8478, RRID: AB_10949108, 1:1,000), HIF1α (Cell Signaling Technology; cat #36169, RRID: AB_2799095, 1:1,000), PCNA (Cell Signaling Technology; cat. #13110, RRID: AB_2636979, 1:1,000), CD27 (R&D Systems; cat. #MAB574, RRID: AB_2073420, 1:500), mTOR (Cell Signaling Technology; cat. #2972, RRID: AB_330978, 1:1,000), p-mTOR (Cell Signaling Technology; cat. #5536, RRID: AB_10691552, 1:1,000), FOXM1 (Proteintech; cat. #13147-1-AP, RRID: AB_2106213, 1:500), IFNAR1 (Abcam; cat. #ab124764, RRID: AB_10972855, 1:800), cleaved caspase-3 (Cell Signaling Technology; cat. #9661, RRID: AB_2341188, 1:1,000), or PFKL (Abcam; cat. #ab181064, 1:1,000) at 4°C overnight. On the second day, the membranes were washed with TBST 3 times (each for 5 minutes) and then incubated with horseradish peroxidase–conjugated secondary antibodies for 2 hours at room temperature. After that, membranes were washed with TBST 3 times (each for 10 minutes), and reactive bands were visualized using ECL substrate (Tanon).

For cytoplasmic and nuclear extract preparation, nuclear and cytoplasmic extraction reagents (Thermo Fisher Scientific; #78833) were used according to the manufacturer's instructions. β-Actin and histone H3 were used as internal references for whole-cell extracts and nuclear extracts, respectively.

For co-IP, indicated cells were lysed with NP-40 buffer (Beyotime; #P0013F) containing 1/100 protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific; #78446). Cell lysates were incubated with co-IP–grade antibodies with rotation at 4°C overnight. On the second day, antibodies were pulled down with Protein A Magnetic Beads (Cell Signaling Technology; #73778) at room temperature with rotation for 20 minutes. After washing with lysis buffer five times, the supernatant was boiled at 95°C for 10 minutes, followed by immunoblot analysis.

Total RNA was extracted from cultured cells using an RNeasy Mini Kit (Qiagen; #74106) and used as a template to generate cDNA with the QuantiTect Reverse Transcription Kit (Qiagen; #205313). Quantitative RT-PCR was performed using FastStart Universal SYBR Green Master Mix (Roche) on a LightCycler 480 384-well plate (Roche) and analyzed using LightCycler 480 software v1.5 (Roche).

For multiplex staining, we followed the Opal protocol staining method with the following reagents: an anti-CD27 antibody (Abcam; #ab214041, 1:500) with subsequent visualization using fluorescein-cyanine 3 (Cy5; 1:50), an anti-CD8 antibody (Cell Signaling Technology; cat. #98941, RRID: AB_2756376, 1:500) with visualization using Cy3 (1:50), and DAPI (1:2,000) for subsequent visualization of nuclei. All sections were coverslipped using VECTASHIELD Hardset Antifade Mounting Medium (Vector Laboratories; #H-1400). For multispectral analysis of the data, a detailed methodology described previously (68) was followed. Each of the individually stained sections was used to establish the spectral library of fluorophores required for the multispectral analysis. The slides were scanned using a Vectra slide scanner (PerkinElmer) under fluorescence conditions. For each marker, the mean fluorescence intensity per case was then determined as the basis from which positive cells could be established. Finally, the colocalization algorithm was used to determine the percentage of CD27 and CD8 colocalization.

A paraffin-embedded HCC tissue microarray was constructed, and IHC analysis was performed as described previously. Briefly, patient HCC specimens were fixed with 4% paraformaldehyde and washed with 75% ethanol. The samples were then deparaffinized and hydrated, and endogenous peroxidase activity was blocked. Antigens were retrieved in EDTA buffer by microwaving for 15 minutes, followed by incubation with 5% BSA for 30 minutes at room temperature. The antibodies used for IHC staining were anti-CD27 (Abcam; cat. #ab131254, RRID: AB_11155136, 1:100), anti-CD8α (Cell Signaling Technology; cat. #85336, RRID: AB_2800052, 1:100), anti-Foxp3 (Abcam; cat. #ab20034, RRID: AB_445284, 1:500), and anti-CD206 (Abcam; cat. #ab64693, RRID: AB_1523910, 1:100). For spontaneous HCC model staining, the antibodies were anti–HNF-4α (Abcam; cat. #ab181604, RRID: AB_2890918, 1:800) and anti–CK-19 (Abcam; cat. #ab52625, RRID: AB_2281020, 1:400). For immunostaining evaluation of CD27 and CD8, the tissue microarrays were assessed by two independent pathologists blinded to the clinical data using a light microscope (200×, Olympus). CD27⁺ or CD8⁺ cells in each 1-mm-diameter cylinder were counted and are shown as the mean value of triplicates (cells/spot).

Gene expression KD was achieved by using a shRNA-mediated silencing approach as previously described. To generate stable silenced cell lines, shRNA lentiviral vectors targeting Hif1α, FosB, Irf1, or Ifnar1 were transfected into murine HCC cell lines with corresponding control vectors. The sequences of the shRNAs were as follows: Hif1α, 5′-CCGGTATGCACTTTGTCGCTATTAACTCGAGTTAATAGCGACAAAGTGCATATTTTTG-3′; FosB-1, 5′-CCGGCAACGGTCACCGCAATCACAACTCGAGTTGTGATTGCGGTGACCGTT

GTTTTTTG-3′; FosB-2, 5′-CCGGGCCAGGGTCAACATCCGCTAACTCGAGTTAGCGGATGT

TGACCCTGGCTTTTTG-3′; Irf1, 5′-CCGGGTTGCATAGAGAGGGTCTTATCTCGAGATAAG

ACCCTCTCTATGCAACTTTTTG-3′; and Ifnar1, 5′-GATCCGAATGAGGTTGATCCGTTTATCTCGAGATAAACGGATCAACCTCATTCTTTTTT-3′. shRNA lentiviral vectors targeting Foxm1 were transfected into CD8⁺ T cells to silence FOXM1 expression with the corresponding control vectors. The shRNA sequence was as follows: shFoxm1-1: 5′-CCGGACTTCCTATTCAGTCCATTAACTCGAGTTAATG GACTGAATAGGAAGTTTTTTG-3′; shFoxm1-2: 5′-CCGGGCTATCCAGTGAAGGAATAGCCTCGAGGCTATTCCTTCACTGGATAGCTTTTTG-3′. Meanwhile, for siRNA KD, 4 × 10⁶ cells/mL CD8⁺ T cells were cultured in a 24-well plate for 12 hours before transfection. Diluting 2 μL of Lipofectamine 2000 (Thermo Fisher Scientific; #11668019) with 25 μL of Opti-MEM, and diluting 120 pmol of siRNA with 25 μL of Opti-MEM. Next, the diluted Lipofectamine 2000 and diluted siRNA were mixed and incubated for 20 minutes. Nontargeting siRNA was used as negative control. The mRNA expression level of Foxm1 was measured by qPCR after 48 hours. The siRNA sequence was as follows: sense, 5′-GCAAAUUUCCAGCCGGAAUTT-3′; antisense: 5′-AUUCCGGCUGGAAAUUUGCTT-3′. For Foxm1 overexpression, murine Foxm1 cDNA was amplified and cloned into the lentiviral vector pHBLV (Hanbio Biotechnology), a lentiviral vector encoding the ZsGreen cDNA downstream of the EF1 promoter. To prepare lentiviral particles, the lentiviral vector, psPAX2, and pMD2G helper plasmid were cotransfected into HEK293T cells according to the manufacturer's protocol (Hanbio Biotechnology). CD8⁺ T cells were infected by Foxm1 overexpression lentiviruses according to the manufacturer's protocol using Polybrene (Hanbio Biotechnology). A lentiviral vector that expressed only ZsGreen (pHBLV⁻ ZsGreen) was produced as a negative control. Positive cell clones were selected by adding puromycin to the culture medium for 7 days, and antibiotic-resistant cells were identified as successfully transfected clones, which were used for further investigations.

In total, 5 × 10⁵ indicated cells were cross-linked with 1% formaldehyde for 15 minutes at room temperature, quenched with 125 mmol/L glycine, and lysed with SDS buffer, and the nuclear extracts were sonicated. The sonicated lysates were precleared with normal IgG and then immunoprecipitated with antibodies against POL2 (Santa Cruz Biotechnology; cat. #sc-47701, RRID: AB_677353, 1:20), H3K9Ac (Cell Signaling Technology; cat. #9649, RRID: AB_823528, 1:25), or HIF1α (Cell Signaling Technology; cat. #36169, RRID: AB_2799095, 1:50) or control IgG (rabbit, Cell Signaling Technology; cat. #3900, RRID: AB_1550038, 1:100; mouse, Cell Signaling Technology; cat. #53484, RRID: AB_2799435, 1:100) overnight at 4°C. On the second day, the lysates were incubated with ChIP-grade protein G agarose beads for 2 hours at room temperature. After washing with low-salt buffer (3 times, each for 5 minutes at 4°C with rotation) and high-salt buffer (5 minutes at 4°C with rotation), the chromatin was eluted from the protein/DNA complexes for 30 minutes at 65°C, and the DNA was purified with DNA purification columns and collection tubes. The purified DNA was then analyzed by qRT-PCR. The ChIP primers used in the present study were designed as follows: forward 5′-CTGCCCATAGTGTGACCCTG-3′, reverse 5′-TGGCTAATTGCGTCACAGGA-3′ for verifying POLII occupation of the mouse FosB promoter region; forward 5′-ATGTGCTCCCTCCCTGATCT-3′, reverse 5′-GTGTGGGTTCCTACCTGCTC-3′ for verifying H3K9ac occupation of the mouse FosB promoter region; forward 5′-GGCTGTGCACCCTCAACTAT-3′, reverse 5′-TGCTTCCTGTCCCTTGGATG-3′ for verifying HIF1α occupation of the mouse Pfkl gene region; forward 5′- CAAACTGGAGAACGCGAACG-3′, reverse 5′-CTCTCTTTCCACGCCCACTT-3′ for verifying HIF1α occupation of the mouse Hk1 gene region; forward 5′-CTATCCTCGCCACCCTGTTC-3′, reverse 5′-AGGGCACCGAATTATGAGGG-3′ for verifying HIF1α occupation of the mouse Pkm2 gene region; and forward 5′-CACTCTTGCAGGGACATCGT-3′, reverse 5′-GGAACCCACGTGTAGGCTG-3′ for verifying HIF1α occupation of the mouse Ldha gene region.

For the isolation of human tumor-infiltrating T cells, fresh human HCC tumor samples were collected and were cut into small pieces. Minced sample pieces were transferred to a gentleMACS C Tube (Miltenyi Biotec; #130-093-237) containing 5 mL digestion enzyme mix prepared according to the manufacturer's instructions (Human Tumor Dissociation Kit; Miltenyi Biotec, #130-095-929). CD45⁺ cells were isolated by CD45 (TIL) human microbeads (Miltenyi Biotec; #130-118-780). Mouse CD8⁺ T cells were isolated by the EasySep Mouse CD8⁺ T-cell Isolation Kit (STEMCELL; #19853). In vitro isolation of PBMCs was done using Ficoll-Paque PLUS (GE Life; #17144002). For surface staining, the following antibodies were used for flow cytometry staining: CD45-PerCP-Cy5.5 (Thermo Fisher Scientific; cat. #45-0451-82, RRID: AB_1107002), CD45-BV421 (BioLegend; cat. #103134, RRID: AB_2562559), CD3-FITC (Thermo Fisher Scientific; cat. # 11-0037-42, RRID: AB_2016669), CD4-APC (Thermo Fisher Scientific; cat. #17-0049-42, RRID: AB_1272048), CD4-APC (Thermo Fisher Scientific; cat. #17-0041-82, RRID: AB_469320), CD8-APC-eFlour780 (Thermo Fisher Scientific; cat. #47-0081-82, RRID: AB_1272185), CD8-APC-eFlour780 (Thermo Fisher Scientific; cat. #47-0088-42, RRID: AB_1272046), CD8-PerCP-Cyanine5.5 (Thermo Fisher Scientific; cat. #45-0081-82, RRID: AB_1107004), CD27-PE (Thermo Fisher Scientific; cat. #MA5-17905, RRID: AB_2539289), CD27-APC (Thermo Fisher Scientific; cat. #17-0271-82, RRID: AB_469370), PD-1-eFlour450 (Thermo Fisher Scientific; cat. #48-9969-42, RRID: AB_2762470), CD45RA-BV421 (BioLegend; cat. #304130, RRID: AB_10965547), CCR7-APC (BioLegend; cat. #353214, RRID: AB_10917387), CD56-APC (BioLegend; cat. #362504, RRID: AB_2563913), CD16-PE (BioLegend; cat. #302008, RRID: AB_314208), PD-1-FITC (Thermo Fisher Scientific; cat. #11-9985-82, RRID: AB_465472), TIM-3-APC (Thermo Fisher Scientific; cat. #17-3109-42, RRID: AB_1963622), and TIM-3-PerCP-Cy5.5 (BioLegend; cat. #119718, RRID: AB_2571935). For intracellular cytokine staining, T cells were stimulated after the indicated treatment with a cocktail containing 50 ng/mL PMA (Sigma; #P8139), 1 μg/mL ionomycin (Sigma; #407951), and 1 μg/mL Golgi­Stop (BD; #554724) for 4 hours. Activated T cells were stained with Live/Dead Fixable Violet (Invitrogen; #L34963) at room temperature for 30 minutes, and then fixed and permeabilized (BD Biosciences; #554722). After washing with Perm/Wash buffer (BD Biosciences; cat. #554723, RRID: AB_2869011), cells were stained with the following antibodies: IFNγ-PE (Thermo Fisher Scientific; cat. #12-7311-82, RRID: AB_466193), Granzyme B-eFlour450 (Thermo Fisher Scientific; cat. #48-8898-82, RRID: AB_11149362), Granzyme B-FITC (Thermo Fisher Scientific; cat. #11-8898-82, RRID: AB_10733414), TNFα-APC (BioLegend; cat. #506308, RRID: AB_315429), IL2-APC (Thermo Fisher Scientific; cat. #17-7021-82, RRID: AB_469490), and TCF-1-Alexa Flour647 (BioLegend; cat. #655204, RRID: AB_2566620). For the OT-I T-cell cytotoxicity assay, after the indicated treatment, OT-I CTLs were collected and resuspended in PBS for surface staining. For the CFSE T-cell proliferation assay, separated CD8⁺ T cells were incubated with 1 μmol/L CFSE (eBioscience; #65-0850-84) for 30 minutes. After being neutralized with FBS and washed with PBS three times, CFSE-labeled T cells were used for further assessment. Flow cytometric data were acquired on a Beckman CytoFlex S and analyzed using FlowJo V10 software (RRID: SCR_008520).

For in vitro CD8⁺ T-cell activation, CD8⁺ T cells were magnetically enriched (BioLegend; #480035) from the spleen and lymph nodes of 6-week-old male wild-type C57BL/6J mice. Cells were plated in 96-well plates precoated with 2 μg/mL anti-mouse CD3 plus 2 μg/mL anti-mouse CD28 at a density of 5 × 10⁵/well in T-cell culture medium. Twenty-four hours later, Cd27-specific siRNA prepared with OPTI-MEM (Thermo Fisher Scientific; #31985088) and RNAiMAX Transfection Reagent (Thermo Fisher Scientific; #13778150) was added to the culture medium at a final concentration of 5 μmol/L according to the manufacturer's protocol. The cells were incubated for another 48 hours, stimulated as described above, and evaluated by flow cytometry to detect the indicated cytokines and other cell markers. Sequences that specifically target murine Cd27 were sense-5′-CCAGCAUGUUUCUUAUCUUTT-3′ and antisense-5′-AAGAUAAGAAACAUGCUGGAG-3′.

Single-cell transcriptomic data were extracted from the Gene Expression Omnibus [(GEO) RRID: SCR_005012, accession code GSE140228 and GSE125449], containing data from 19 HCC patients generated with the smart-seq2 and 10X Genomics platforms. These data sets were integrated using Seurat's CCA to remove potential batch effects. For other collected scRNA-seq data sets (accession codes GSE120575 and GES145281), we applied the same filtering steps to 10X Genomics data. Data processing and analysis were performed using Seurat (version 4.0, RRID: SCR_007322).

All cells that had low-quality unique molecular identifier sequences were removed. For the remaining cells, the similarity and variability were summarized by principal component analysis. The expression trend of cell genes was proportional to the sample distance. For visualization, the dimensionality of each data set was further reduced using either Barnes-Hut t-SNE or uniform manifold approximation and projection analysis from the Seurat toolkit with the top 30 principal components with default settings. CD8⁺ T cells were reanalyzed to identify subclusters and were contrasted using the Seurat FindMarkers function to identify conserved marker genes for these subsets.

For the glucose assay, the supernatant of tumor and T cells was collected, and the glucose concentration was assessed (GemPharmatech; ref. 58). For ECAR, mouse orthotopic tumor cells or tumor-infiltrated CD8⁺ T cells harvested at different time points after the indicated treatment were plated in nonbuffered RPMI 1640 culture medium with 25 mmol/L glucose. In some experiments, cells were pretreated with an mTOR inhibitor or a HIF1α inhibitor for another 24 hours. The real-time ECAR was measured on an XF96 extracellular flux analyzer (Seahorse Bioscience).

The TRIzol method was adopted to extract total RNA according to the manufacturer's protocol. A NanoDrop 2000 spectrophotometer (Thermo Scientific, RRID: SCR_018042) was applied for RNA purity measurement and quantification, and an Agilent 2100 bioanalyzer (Agilent Technologies, RRID: SCR_019389) was applied for RNA integrity assessment. Sequencing libraries were generated using the TruSeq Stranded mRNA LT Sample Prep Kit (Illumina) following the manufacturer's recommendations.

OVA-specific CD8⁺ CTLs were generated by incubating splenocytes from OT-I mice with 5 μg/mL SIINFEKL peptide (Sangon Biotech; #T510212) and 20 ng/mL mIL2 (R&D Systems; #402-ML) in low-glucose (5 mmol/L) or high-glucose (25 mmol/L) RPMI 1640 medium for 7 days.

Prepared OT-I CTLs were treated with IFNα at different concentrations in low-glucose (5 mmol/L) or high-glucose (25 mmol/L) RPMI 1640 medium for 48 hours. Then EL4 cells were harvested and resuspended at 1 × 10⁶ cells/mL. EL4 cells were pulsed with medium containing a vehicle control or 10 μg/mL SIINFEKL peptide for 2 hours at 37°C and then labeled with 0.5 μmol/L or 5 μmol/L CFSE (Thermo Fisher Scientific; #C34554), respectively, for 10 minutes at 37°C. EL4 cells were cocultured with OT-I CTLs at different effector:target ratios in low-glucose (5 mmol/L) or high-glucose (25 mmol/L) RPMI 1640 medium for 12 hours. After incubation, the percentage of specific killing was determined by flow cytometric analysis. Specific killing (%) = [1 − (sample ratio/negative control ratio)]. Sample ratio = [CFSE^hi/CFSE^lo] value of each sample cocultured with CD8⁺ T cells. Negative control ratio = [CFSE^hi/CFSE^lo] value of EL4 cells not cultured with CD8⁺ T cells.

Prepared libraries were sequenced on an Illumina HiSeq X Ten platform (RRID: SCR_016385). Raw sequence reads were first aligned to the mouse UniGene transcriptome with Bowtie (v1.0.0, RRID: SCR_005476) to estimate the insert fragment size and the SD needed by TopHat2 (RRID: SCR_013035) to align the reads to the genome. Then, TopHat2 (RRID: SCR_013035) was used to align the reads to the reference mouse genome (GRCm38) with the aligning parameter bowtie1 and Ensembl-annotated transcripts (version 77) as a guide reference. Uniquely mapped reads were used to quantify gene expression by Cufflinks (v2.1.1, RRID: SCR_014597). The abundances of transcripts (including mRNAs, pseudogenes, noncoding RNAs, and other predicted RNAs) were calculated and normalized in fragments per kilobase of transcript per million mapped reads, as described above, from the raw RNA-seq data and used for GSEA (Broad Institute, RRID: SCR_003199). Differential gene expression evaluation was analyzed using DESeq2 software (v1.30.0, RRID: SCR_015687; ref. 69).

R (4.0.3) software was used for data analysis, and statistical significance was determined using a t test. P < 0.05 was considered statistically significant. Pearson correlation analysis was used to describe the correlation between variables. For analysis of the TCGA data set, a Kaplan–Meier curve was constructed to compare the OS and PFS rates of two groups. The log-rank P value and hazard ratio were calculated. For single-cell analysis, comparison of two groups was performed using the nonparametric Wilcoxon rank sum test with Bonferroni correction.

The TCGA data set, including liver hepatocellular carcinoma, was downloaded from cBioPortal (ref. 70; http://www.cbioportal.org/; RRID: SCR_014555). The full RNA-seq data set has been uploaded to the GEO repository (accession code GSE179352). None of the flow-cytometric data or IHC data supporting the current study have been deposited in a public repository because of privacy and ethical restrictions but are available from the corresponding author on request.

Codes supporting the findings of this study have been deposited at https://github.com/zyangx/IFN_Sens_PD1/.

H. Sun reports personal fees from Eisai, MSD, Hengrui, TopAlliance, Beigene, AstraZeneca, Bayer, Roche, and Zelgen, and grants and personal fees from Innovent during the conduct of the study. A.X. Zhu reports personal fees from I-Mab, Lilly, Merck, Exelixis, Roche, Eisai, and Bayer outside the submitted work. Y. Xu reports grants from the National Natural Science Foundation of China, the National Science and Technology Major Project, and the Science and Technology Commission of Shanghai Municipality during the conduct of the study. J. Fan reports grants from the National Natural Science Foundation of China, the National Science and Technology Major Project, and the Science and Technology Commission of Shanghai Municipality during the conduct of the study. No disclosures were reported by the other authors.

B. Hu: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, project administration. M. Yu: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, project administration. X. Ma: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, project administration. J. Sun: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, project administration. C. Liu: Data curation, software, methodology. C. Wang: Software, validation, visualization, methodology. S. Wu: Data curation, validation, investigation, methodology. P. Fu: Formal analysis, validation, visualization, methodology. Z. Yang: Software, validation, visualization. Y. He: Software, validation, visualization. Y. Zhu: Methodology. C. Huang: Supervision, writing–review and editing. X. Yang: Supervision, writing–review and editing. Y. Shi: Supervision, writing–review and editing. S. Qiu: Supervision, writing–review and editing. H. Sun: Supervision, writing–review and editing. A.X. Zhu: Supervision, writing–review and editing. J. Zhou: Supervision, writing–review and editing. Y. Xu: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing. D. Zhu: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing. J. Fan: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Xun Huang at Lingang Laboratory, Shanghai, China, for providing cell lines. We thank Yan-Yan Nie at the Shanghai Laboratory Animal Research Center for providing technical assistance with animal studies. We thank Ruo-Yu Shi at the department of Anatomical Pathology, Singapore General Hospital, for providing pathology diagnosis. We also thank Merdobio Corporation (Shanghai, China) for providing technical assistance with PDO-TIL model construction. This research was supported by Research Projects from the Science and Technology Commission of Shanghai Municipality (grant 21JC1410100, to J. Fan; grant 22140900600, to B. Hu), the National Natural Science Foundation of China (no. 81772551, to J. Fan; no. 81802364, to B. Hu; no. 82073272, to Y. Xu; no. 82073881 and 81872895, to D. Zhu; no. 82002618, to X. Ma; no. 81871929 and 82072667, to C. Huang), and the National Science and Technology Major Project (no. 2018ZX10723204-006-002; to Y. Xu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.