PD-1 is an inhibitory receptor on T cells; engagement by its ligands PD-L1 or PD-L2 (often expressed on tumor cells and antigen-presenting cells) transmits an inhibitory signal that dampens T-cell receptor signaling. In HCC, PD-1 is highly expressed on tumor-infiltrating T cells and PD-L1 is upregulated on a subset of tumor cells and macrophages, which contributes to an immunosuppressive milieu (15). This PD-1/PD-L1 interaction leads to T-cell exhaustion and facilitates tumor escape. Clinically, PD-L1 upregulation in HCC has been associated with aggressive clinicopathological features and a worse prognosis (16). Conversely, the presence of PD-1/PD-L1 also signifies an inflamed TME, and HCCs with high PD-L1 may demonstrate an improved response to PD-1 blockade therapy (as evidenced by higher response rates in PD-L1-positive tumors than those PD-L1-negative tumors in a meta-analysis) (2).

CTLA-4 is another co-inhibitory receptor, which predominantly acts at the priming phase of T-cell activation. CTLA-4 is expressed on activated T cells and regulatory T cells (Tregs), and competes with the costimulatory receptor CD28 for binding B7-1/B7-2 ligands on antigen-presenting cells (17). In HCC, CTLA-4 activity on intratumoral Tregs serves a key role in suppressing antitumor immunity by removing CD80/86 from dendritic cells, thereby impairing antigen presentation (18). This results in reduced T-cell activation in tumor-draining lymph nodes. High intratumoral Treg populations that often express CTLA-4 have been associated with tumor progression and worse outcomes in HCC (19,20). Blockading CTLA-4 can both release the ‘brakes’ on effector T cells and deplete Tregs; in preclinical HCC models, anti-CTLA-4 treatment restored CD8⁺ T-cell function and synergized with anti-PD-1 therapy (18,21), which supports the concept of dual checkpoint inhibition.

Beyond PD-1 and CTLA-4, HCC tumors express additional checkpoint molecules that contribute to immune evasion.

LAG-3 is an inhibitory receptor on T cells that binds MHC class II and other ligands; it often co-expresses with PD-1 on exhausted T cells. HCC-infiltrating CD8⁺ and CD4⁺ T cells can express LAG-3, particularly in advanced disease (8). Although high LAG-3 in the TME has been associated with T-cell dysfunction and shorter survival times in untreated patients, paradoxically, the presence of LAG-3⁺ T cells alongside CD8⁺ T cells may predict improved responsiveness to immunotherapy (22). Recent studies reported that patients with HCC with higher baseline LAG-3 and CD8⁺ infiltration had prolonged progression-free survival (PFS) and overall survival (OS) times on use of immune checkpoint inhibitors (ICIs) (4,22,23). This suggests upregulation of LAG-3 indicates an immune-activated, albeit exhausted, TME that can be reinvigorated by therapy.

TIM-3 is another co-inhibitory receptor, expressed on exhausted CD8⁺ T cells, some CD4⁺ T helper cells, Tregs and innate cells, such as natural killer (NK) cells and macrophages (24,25); its ligands include galectin-9 and phosphatidylserine. TIM-3 is frequently upregulated in HCC-infiltrating lymphocytes and tumor-associated macrophages (TAMs), particularly in patients with chronic hepatitis backgrounds (26,27). Activation of TIM-3 on T cells drives loss of effector function [for example, reduced interferon (IFN)-γ secretion] and promotes T-cell apoptosis, thereby enabling immune escape (28). Dual blockade of TIM-3 and PD-1 in HCC mouse models restored CD4⁺ and CD8⁺ T-cell activity and curtailed tumor growth (15), underscoring the role of TIM-3 in resistance to single-agent PD-1 therapy.

TIGIT is an inhibitory receptor on T cells and NK cells that competes with the costimulatory receptor CD226 for the ligand CD155 (PVR) on tumor cells or dendritic cells (29,30). In HCC, TIGIT expression is noted on tumor-infiltrating T cells and a subset of NK cells. Co-expression of TIGIT and TIM-3 defines an ‘exhausted’ NK cell phenotype in HCC that exhibits impaired cytotoxicity (27). A previous study in hepatitis B-related HCC reported that TIGIT⁺TIM-3⁺ NK cells were enriched in tumors and associated with advanced disease and worse patient outcomes (27). Thus, TIGIT contributes to suppression of innate immune surveillance in HCC. Blocking TIGIT can rejuvenate NK cell function and enhance T-cell responses in preclinical models (13,27), which makes it a potential target in combination with other ICIs.

Several other inhibitory pathways are active in HCC and are under investigation (Table I) (6,8,15,16,18,22,27,31,32). BTLA is a checkpoint receptor on lymphocytes that binds to herpes virus entry mediator (HVEM). BTLA is upregulated on a subset of PD-1⁺ CD4⁺ T cells in HCC, marking a highly dysfunctional T-cell population (8). BTLA⁺PD-1⁺ T cells produce little cytokine and associate with more advanced tumor stages (8). HVEM, the ligand often expressed on tumor cells and myeloid cells, has been overexpressed in HCC and associated with a worse prognosis (33). The BTLA-HVEM interaction thus potentially dampens antitumor T-cell activity in the liver, and disrupting this axis could release those BTLA⁺PD-1⁺ T cells (8,33).

VISTA is a B7-family ligand/receptor predominantly on myeloid cells and some T cells (34). VISTA can deliver inhibitory signals to T cells, particularly in low pH TMEs and has been described as a potential checkpoint that may operate when PD-1/PD-L1 is blocked (12). In HCC, VISTA is expressed both on immune infiltrates and occasionally on tumor cells. A previous study reported that tumor-cell VISTA expression in HCC was associated with a favorable immune milieu and longer OS (12). This indicates that VISTA may have context-dependent functions; in specific HCC subgroups, elevated VISTA levels could signify an active antitumor immune response, potentially serving a counter-regulatory role. Caution is warranted, since in those cases blanket VISTA blockade might theoretically remove a restraining influence on an already active immune response (12). However, overall, VISTA primarily acts as an immunosuppressive factor and therapies targeting VISTA are in early-phase trials (12,34).

B7-H3 (CD276) is an inhibitory immune checkpoint ligand highly expressed in several solid tumors, including HCC (5). In total, >80-90% of HCC tumors aberrantly express B7-H3 on tumor cells or in the stroma (10). High tumor B7-H3 associates with aggressive clinicopathological features, such as vascular invasion, metastasis and advanced stage, and with worse patient survival (5). Mechanistically, HCC cells can upregulate B7-H3 in response to cytokines, such as IFN-γ, as a feedback mechanism to inhibit cytotoxic T cells (5,10). B7-H3 on tumor cells directly impairs infiltrating T-cell proliferation and interferon production (5); it can also promote epithelial-mesenchymal transition and invasive behavior of HCC cells (10). In datasets of patients with HCC, those with high B7-H3 and low CD8⁺ T-cell infiltration have the worst outcomes (10). Blocking B7-H3 signaling, in co-culture experiments, markedly enhanced T-cell killing of HCC cells (10). In mouse HCC models, anti-B7-H3 antibodies reduced tumor growth and prolonged survival time (10). These data establish B7-H3 as a potent immunosuppressive pathway in liver cancer.

Another B7 family ligand, B7-H4 (VTCN1), is expressed on some HCC tumors and particularly on TAMs (7). B7-H4 inhibits T-cell immunity; in HCC, elevated B7-H4 in tumor tissues has been associated with advanced stage and early recurrence (6). A functional study demonstrated that knocking down B7-H4 in HCC cell lines can restore CD8⁺ T-cell activity (increasing perforin, granzymes and IFN-γ) and induce tumor cell apoptosis (6). In vivo, loss of B7-H4 led to slower HCC tumor growth in mice (6). Thus, B7-H4 appears to contribute to HCC progression by both direct tumor-promoting effects and immune evasion.

While not a T-cell checkpoint per se, the ‘don't eat me’ signal CD47-signal regulatory protein (SIRP)α axis is an important innate immune checkpoint relevant to HCC. CD47 is a protein frequently upregulated on HCC cells, which delivers a ‘do not phagocytose’ signal to macrophages via SIRPα binding (11,35). High CD47 helps tumors escape macrophage-mediated clearance. In experimental HCC models, blocking CD47 (for example, with a CD47 antibody or a bispecific that also targets a tumor antigen) promotes macrophage engulfment of tumor cells (32). Targeting CD47 can thereby stimulate both innate immunity and downstream T-cell responses against HCC. In addition, Kupffer cells (KCs), the liver-resident macrophages, are key orchestrators of immunosuppression in HCC. In the tumor milieu, KCs upregulate multiple inhibitory immune checkpoints that attenuate antitumor immunity. Notably, KCs express high levels of PD-L1, which engage PD-1 on effector T cells and induce their exhaustion (36). KCs (as TAMs) also display checkpoint molecules, such as TIM-3 and VISTA, on their surfaces (36,37). TIM-3, together with its ligand galectin-9, contributes to local T-cell dysfunction and has been associated with HCC progression (38). VISTA, an Ig family checkpoint upregulated on KCs, similarly delivers inhibitory signals that promote immune tolerance in the TME (37). In addition, HCC cells exploit the macrophage checkpoint axis CD47-SIRPα: Tumor cell CD47 (‘don't-eat-me’ signal) binds SIRPα on KCs to block phagocytosis, a mechanism reinforced by TAM-derived IL-6, which further elevates tumor CD47 and is associated with low patient survival (9). Through these checkpoint pathways, KCs blunt cytotoxic T-cell activity and foster an immunosuppressive microenvironment that enables HCC to evade immune surveillance (39,40).

Generally, HCC tumors that exhibit an immunosuppressive phenotype, characterized by high expression of checkpoints, such as PD-L1, B7-H3 or an abundance of CTLA-4⁺ Tregs, are associated with more aggressive disease and worse patient survival (5). For example, tumors with high PD-L1 expression have been associated with higher rates of vascular invasion and early recurrence, which translates to shorter OS times in some cohorts (16). High intratumoral B7-H3 has been validated as an independent predictor of reduced recurrence-free survival time after resection (5). On the other hand, the presence of tumor-infiltrating lymphocytes (TILs), even if exhausted, can signify an immunologically active TME, which can be a favorable prognostic factor if harnessed by immunotherapy (41,42). The density of CD8⁺ T cells alongside inhibitory receptors is a nuanced indicator. An ‘immune-high’ HCC (several TILs expressing PD-1 and LAG-3, among others) might portend a worse outcome without treatment due to ongoing immune suppression, but such tumors are also the ones more likely to respond if treated with ICIs (22). Indeed, an immunohistochemical study demonstrated that patients with HCC with high baseline LAG-3⁺CD8⁺ T-cell infiltration had prolonged survival times when treated with checkpoint inhibitors, compared with those patients with low LAG-3 or CD8 (22). This underscores the fact that checkpoint biomarkers can have dual implications: Adverse prognostic markers in the absence of therapy, yet positive predictive markers for immunotherapy benefit.

Unlike in other cancer types, to the best of our knowledge, no single biomarker reliably predicts which patients with HCC will respond to ICIs. PD-L1 expression has been explored in this regard. Meta-analyses reported that patients with HCC with PD-L1⁺ tumors exhibited improved response to anti-PD-1/PD-L1 therapy compared with those with PD-L1⁻ tumors. For example, the objective response rate (ORR) was ~26% for PD-L1⁺ tumors and 18% for PD-L1⁻ tumors (43). In a pooled analysis of 1,330 ICI-treated patients with HCC, PD-L1 positivity was associated with improved response (ORR, ~1.8-fold) than those patients with PD-L1-negative tumors (43). However, some patients with PD-L1⁻ HCC still respond and not all patients with PD-L1⁺ HCC benefit, so PD-L1 alone is an imperfect predictor. Tumor mutational burden (TMB) is generally low in HCC and has not demonstrated a strong association with ICI outcomes (23,44). Conversely, composite measures of the immune microenvironment are being investigated; for instance, gene expression profiles indicating an ‘inflamed’ tumor (high IFN-γ signature and T cell-inflamed score) have been associated with improved immunotherapy response (45). As aforementioned, multiplex immunohistochemistry of multiple markers (for example, CD8, PD-L1 and LAG-3) may provide a more robust predictive tool compared with any single marker (22). The ratio of effector T cells to immunosuppressive cells is another important variable: HCCs with high CD8⁺ T cells compared with forkhead box P3 (FOXP3)⁺ Tregs, or high M1 macrophages compared with M2 macrophages, generally demonstrate enhanced performance. If an HCC is ‘cold’ (low TILs) or excluded (T cells at the margin but not infiltrating) or response rates to current ICIs are low, these cases might require combination strategies to induce immune infiltration (46).

The unique background of chronic liver disease in patients with HCC means that immune-related adverse events (irAEs) from checkpoint therapy require careful monitoring (47,48). Under physiological conditions, the liver is an immunotolerant organ, rich in innate immune cells and Tregs to prevent excessive inflammation to gut-derived antigens. Checkpoint blockade can disrupt this balance. For example, immune-mediated hepatitis is a known toxicity of PD-1 or CTLA-4 inhibitors, which in cirrhotic patients could precipitate liver decompensation (47,49). Thus far, trials in HCC have largely enrolled patients with well-compensated Child-Pugh A liver function and ICIs have exhibited a generally manageable safety profile in that population (3). Reactivation of hepatitis B virus (HBV) is another concern when treating chronic HBV-associated HCC with ICIs, due to immune reconstitution. Prophylactic antiviral therapy is recommended during ICI treatment in patients with HBV to mitigate this risk (50,51). From a clinical standpoint, these considerations imply that integrating checkpoint therapy in HCC requires multidisciplinary care, with hepatologists managing the underlying liver disease to ensure patients are optimized for immunotherapy.

Anti-PD-1 antibodies were the first ICIs tested in HCC. Nivolumab (anti-PD-1) exhibited notable activity in the phase I/II CheckMate-040 trial, which led to an accelerated Food and Drug Administration (FDA) approval in 2017 for sorafenib-experienced HCC based on an ORR of ~15% and durable responses in certain patients (3). Pembrolizumab (anti-PD-1) similarly demonstrated a ~17% response rate in the phase II KEYNOTE-224 trial in advanced HCC, with certain patients achieving long-lasting tumor regression (57). However, when tested in larger phase III trials as a single agent in advanced HCC, PD-1 inhibitors yielded mixed results. CheckMate-459, which compared first-line nivolumab vs. sorafenib in 743 patients, did not meet its primary endpoint of markedly improving OS time (3). Nivolumab did demonstrate a numerically higher median OS time (16.4 months) compared with sorafenib (14.7 months), with a hazard ratio (HR) of 0.85; however, this difference was not statistically significant (P=0.075) (3). Notably, a subset of patients treated with nivolumab had prolonged survival beyond 2 years (with a 2-year survival rate of ~29 vs. 21% for sorafenib), which reflects the tail of the curve often seen with immunotherapy (58).

Similarly, pembrolizumab as second-line therapy was investigated in KEYNOTE-240 (global trial) and KEYNOTE-394 (Asia-Pacific trial) (52). KEYNOTE-240 did not achieve statistical significance for OS improvement (median OS time, ~13.9 vs. 10.6 months, P=0.023 which was just above the α boundary), being essentially a negative trial despite a positive trend. By contrast, KEYNOTE-394 (which enrolled only Asian patients) met its endpoint: Pembrolizumab markedly prolonged OS time compared with the placebo group (median OS time, 14.6 vs. 13.0 months; HR, 0.79; P=0.018) (52). Pembrolizumab also improved objective response rate (ORR, 12.7 vs. 1.3%; P<0.0001) and PFS (2.6 vs. 2.3 months; HR, 0.74; P=0.0032) in this study (52). These nuanced results led to regulatory approvals in some regions (for example, FDA approval for second-line pembrolizumab in 2023, considering the totality of evidence) (59). Overall, PD-1 monotherapy induces a complete response (CR) in 1–4% and partial responses in ~15% of advanced HCC cases, which is a minority, but is often durable when the responses occur (57). Patients who do respond may experience long-term remission, as evidenced by certain 3–5-year survivors found in trials (57).

To increase the proportion of patient benefit, combinations of ICIs that target complementary pathways have been explored. The leading example is PD-1 + CTLA-4 inhibition. In the CheckMate-040 trial, a cohort received nivolumab + ipilimumab (anti-CTLA-4) after sorafenib failure. This non-randomized study reported an ORR of 32% and a CR in 8% of patients, with some responses notably durable (median duration of response >2 years in the highest dose ipilimumab arm) (54). At the 5-year follow-up, the nivolumab + ipilimumab combination demonstrated a 34% ORR and a median OS time of ~22 months in sorafenib-experienced patients (54). These results led to an accelerated FDA approval of nivolumab + ipilimumab for second-line HCC in 2020 (60). Toxicity was higher with dual therapy (grade ≥3 immune-mediated AEs in ~37% of patients, manageable with immunosuppression) (54).

Building on these results, the phase III HIMALAYA trial investigated a modified approach in the first-line setting: A single priming dose of tremelimumab (CTLA-4 inhibitor) combined with durvalumab (PD-L1 inhibitor), termed the ‘STRIDE’ regimen, was compared with sorafenib. HIMALAYA demonstrated a notable survival benefit for the combination, which reported a median OS rime of 16.4 vs. 13.8 months with sorafenib (HR, 0.78; P=0.0035) (56). At 3 years, 31% of patients on the durvalumab + tremelimumab arm were alive, compared with 20% of patients on the sorafenib arm (56). Notably, durvalumab alone (PD-L1 monotherapy) was non-inferior to sorafenib in terms of OS, which confirmed the contribution of PD-L1 blockade; but only the addition of the one-time CTLA-4 dose resulted in improved survival (56). The intermittent CTLA-4 dosing in the STRIDE regimen aimed to balance efficacy and safety, and the rate of severe irAEs was relatively low and manageable. Based on the HIMALAYA trial, durvalumab + tremelimumab was approved (for example, by FDA and European Medicines Agency) as a first-line option for advanced HCC in 2022 (60). These findings establish dual-checkpoint inhibition as a viable strategy in HCC, particularly for patients who may not be candidates for anti-angiogenic therapy.

HCC is an angiogenesis-driven tumor (VEGF is often highly expressed) and there is crosstalk between angiogenic factors and the immune microenvironment. VEGF not only promotes blood vessel formation but also has immunosuppressive effects [for example, VEGF can recruit Tregs and myeloid-derived suppressor cells (MDSCs) and impede dendritic cell maturation]. This provided a notable rationale to combine anti-VEGF agents with ICIs to achieve a synergistic effect; normalization of tumor vasculature and reversal of VEGF-mediated immunosuppression might enhance T-cell infiltration and function, thereby augmenting immunotherapy efficacy (61). The landmark IMbrave150 trial exemplified this approach by combining atezolizumab (anti-PD-L1) with bevacizumab (anti-VEGF) in comparison to sorafenib in untreated advanced HCC (51). IMbrave150 was practice-changing as the combination markedly improved both OS and PFS times. Updated results reported a median OS time of 19.2 months with atezolizumab + bevacizumab vs. 13.4 months with sorafenib (HR, 0.66; PFS time, 6.9 vs. 4.3 months; P<0.001) (53).

The ORR, according to the Response Evaluation Criteria In Solid Tumors, was ~30% with atezolizumab + bevacizumab (including some CRs, ~8%), which was more than the 11% reported with sorafenib (53). Responses tended to be durable and quality-of-life outcomes favored the combination therapy. Atezolizumab + bevacizumab thus became the novel standard first-line therapy for advanced HCC (approved in 2020) (61), representing the first regimen to improve survival times over sorafenib in >10 years. Clinicians must screen for varices in patients (since bevacizumab can increase bleeding risk in cirrhosis), but with proper management, the regimen has been broadly adopted. Following this success, other ICI + antiangiogenic combos have been evaluated. For example, pembrolizumab + lenvatinib (a multikinase inhibitor targeting VEGFR among others) in the phase III LEAP-002 trial. The LEAP-002 trial did not demonstrate a statistically significant OS benefit for pembrolizumab + lenvatinib vs. lenvatinib alone (median OS time, 21.2 vs. 19.0 months; HR, 0.84; P=0.023 > threshold), despite encouraging response rates (~26 vs. 17%) and a modest PFS improvement (55). This outcome underscores that not all combinations will automatically outperform the tyrosine kinase inhibitor (TKI) alone, possibly due to trial design or the high efficacy of the control arm (lenvatinib is a potent agent on its own). Nonetheless, lenvatinib + pembrolizumab exhibited a high disease control rate (~88%) and a subset of patients achieved long-term remission (55). The ongoing study includes triplet combinations (PD-1 + CTLA-4 + anti-VEGF) to enhance the efficacy (62).

Researchers are exploring combinations of ICIs with locoregional therapies [such as transarterial chemoembolization (TACE), radiofrequency ablation or radiotherapy]. The rationale is that these therapies can cause immunogenic cell death and tumor antigen release, potentially turning ‘cold’ tumors ‘hot’. Small studies have demonstrated that performing TACE or radiotherapy can upregulate PD-L1 and increase T-cell infiltration in HCC, which provides a window to introduce ICIs. Clinical trials such as combining tremelimumab with ablation [for example, a trial by Sangro et al (47)] demonstrated enhanced systemic T-cell responses and some pathological tumor necrosis, which indicated possible synergy. Similarly, neoadjuvant immunotherapy (ICI administered before surgical resection) is under investigation to eliminate micrometastases and induce immune memory to prevent recurrence. Early-phase trials of neoadjuvant nivolumab or combination ICIs in resectable HCC have reported pathological response rates in a proportion of patients, and those who achieve major tumor necrosis tend to have delayed recurrence. To date, adjuvant immunotherapy for HCC remains experimental (the phase III CheckMate-9DX trial of nivolumab vs. placebo after curative resection or ablation did not meet its endpoint, reported in 2023). However, these approaches remain to be refined in future research (63–65).

In deploying checkpoint inhibitors for HCC, patient selection and sequence of therapies are key considerations. With multiple first-line options available (atezolizumab + bevacizumab, durvalumab + tremelimumab, lenvatinib and sorafenib), decisions are individualized based on factors such as vascular invasion, bleeding risk and autoimmune conditions. The successes of atezolizumab + bevacizumab and durvalumab + tremelimumab have positioned immunotherapy-based combinations as preferred front-line treatments for eligible patients. Those who progress on first-line ICIs may still benefit from second-line treatments, including TKIs (regorafenib, cabozantinib and others) or even a different immunotherapy approach (trials are exploring switching to a PD-1 + CTLA-4 combination after failing PD-1 + VEGF, for instance) (66–68). The landscape is rapidly evolving, with numerous ongoing trials of novel combinations (Table III) (10,12,15,22,27,32,56,68–70).

Several HCCs have a paucity of T-cell infiltration, rendering PD-1/PD-L1 blockade ineffective. One key tumor-intrinsic cause is aberrant Wnt/β-catenin signaling. β-catenin activation in HCC associates with an ‘immune excluded’ phenotype; T cells are kept at the margins of the tumor and have been associated with resistance to anti-PD-1 therapy (74). HCCs with β-catenin mutations lack TILs and do not respond to PD-1 inhibitors (74). Similarly, tumors with low baseline inflammation or low PD-L1 expression tend to be intrinsically resistant, as there are few effector T cells for ICIs to ‘unleash’. Furthermore, low-to-moderate TMB in HCC means fewer neoantigens to attract T cells, which can limit ICI efficacy from the start (75,76).

HCC arises in a setting of chronic inflammation [HBV or HCV, non-alcoholic steatohepatitis (NASH) and alcohol] that can pre-condition the immune system to be exhausted or tolerant. Chronic antigen exposure (for example, from viral hepatitis) drives T cells into an extremely exhausted state, characterized by co-expression of multiple inhibitory receptors, such as PD-1, LAG-3 and TIM-3, along with epigenetic changes locking in dysfunction (77,78). Such ‘terminally’ exhausted T cells cannot be fully reinvigorated by PD-1 monotherapy alone. For example, in chronic HCV infection, CD8⁺ T cells exhibit an exhaustion phenotype that associates with viral epitope variations and may be hard to reverse. Similarly, HCC in a NASH cirrhosis background demonstrated worse response to ICIs (79). The proposed mechanism is that NASH causes accumulation of dysfunctional, senescent CD8⁺ T cells in the liver that promote tissue damage but have impaired antitumor activity (79). This illustrates how immune milieu differences (NASH vs. viral) can engender primary resistance.

HCC tumors develop amid cirrhotic tissue that is inherently immunotolerant. The resident cells of the liver (Kupffer, stellate and sinusoidal endothelium cells) release immunosuppressive cytokines, such as IL-10, TGF-β and VEGF, which blunt antitumor immunity (80). The tumor itself further skews the microenvironment: TAMs in HCC often adopt an M2-like, pro-tumoral phenotype that supports tumor growth and suppresses T-cell function (81). MDSCs are abundantly recruited; they can secrete arginase and consume key nutrients (for example, L-arginine), starving T cells, as well as produce immunosuppressive factors that inhibit T-cell proliferation (82). High levels of FOXP3⁺ CTLA-4⁺ Tregs in HCC potently suppress effector T cells and impair dendritic cell function, which creates a local tolerance (83). An overwhelming immunosuppressive milieu can cause primary resistance. Checkpoint blockade simply cannot tip the scales if the baseline state is too inhibitory. In such cases, monotherapy may fail to provoke any tumor immunity.

Even when ICIs initially activate an antitumor response, HCC can evolve escape mechanisms under immune pressure. One common adaptation is the upregulation of alternative checkpoints on T cells or tumor cells. For instance, HCCs that progress after PD-1 blockade have been found to exhibit increased expression of other inhibitory receptors, such as VISTA, TIM-3 or LAG-3, on residual TILs (4,84,85). By engaging these alternate ‘brakes’, the tumor regains immune suppression despite PD-1/PD-L1 being blocked. Concurrently, the tumor may undergo immunoediting, namely, outgrowth of cancer cell clones that are less immunogenic. This includes loss of tumor antigens that were targeted by T cells and downregulation of antigen-presentation machinery (86). Nonetheless, selection for tumor cells with β2-microglobulin (B2M) mutations (disabling MHC class I expression) or loss of IFN-γ pathway signaling has been observed in other cancer types as an acquired resistance mechanism (87), and potentially serves a role in HCC as well. The net result is that the immune system no longer recognizes or effectively attacks the tumor. Furthermore, the cirrhotic liver can become more fibrotic or hypoxic over time, which fosters further exclusion of immune cells and upregulating pathways, such as adenosine, that suppress immunity (88,89). In sum, the dynamic and heterogeneous nature of HCC, coupled with an ever-evolving cirrhosis microenvironment, means the tumor can find ways to escape immune pressure at multiple levels.

Collectively, ICI resistance in HCC is multi-factorial. Tumor-intrinsic pathways (Wnt/β-catenin and loss of MHC), a highly immunosuppressive microenvironment (TAMs, MDSCs, Tregs and cytokines) and exhausted or senescent T cells all contribute to poor or short-lived responses. Recognizing these mechanisms is key, as they inform rational combination strategies to potentially overcome resistance in the future.

Even with combination regimens, a large fraction of patients with HCC do not respond to ICIs. In front-line trials, only 20–30% of patients achieve tumor responses with the best ICI-based therapies (53). The remaining patients have either stable disease or progressive disease, which indicates intrinsic (primary) resistance to checkpoint blockade. The reasons for primary resistance in HCC are multifactorial. Certain tumors have an immune ‘desert’ phenotype, a paucity of T cells in the tumor parenchyma, often due to exclusion by a fibrous stroma or an immunosuppressive myeloid cell infiltrate (90,91). These tumors lack sufficient effector cells for ICIs to act upon. Other tumors may have oncogenic pathways driving immune evasion that are not readily overcome by blocking PD-1/CTLA-4 alone (for example, β-catenin/Wnt pathway activation in HCC is associated with low T-cell infiltration and ICI resistance) (92,93). Furthermore, chronic antigen exposure from viral hepatitis can lead to T-cell exhaustion that is enhanced and possibly irreversible, marked by co-expression of multiple inhibitory receptors (PD-1, LAG-3, TIM-3 and others) along with epigenetic reprogramming (94). Reinvigorating such ‘terminally’ exhausted T cells with PD-1 antibodies alone might be insufficient unless additional pathways are targeted or novel T cells are recruited.

Even patients who initially respond to checkpoint inhibitors can develop progression months or years later (acquired resistance). Tumors may upregulate alternative checkpoints as escape mechanisms; for instance, some HCCs progressing on anti-PD-1 therapy have been found to have increased VISTA or TIM-3 expression on residual immune cells (12). There may also be a selection of tumor cell clones that are less immunogenic (loss of antigen presentation machinery, such as B2M and loss of neoantigens among others) (95,96). HCC is dynamic and heterogeneous, particularly against the ever-evolving background of cirrhosis. As a result, overcoming resistance will potentially require combination or sequential therapies.

HCC arises in an immunosuppressive microenvironment shaped by chronic inflammation. The liver contains numerous innate immune cells (Kupffer, liver sinusoidal endothelial and stellate cells) that can secrete cytokines, such as IL-10, TGF-β and VEGF, all of which inhibit effective antitumor immunity (97). Tumor-associated macrophages (TAMs) in HCC often exhibit an M2-like phenotype that promotes tumor growth and suppresses T cells. MDSCs are abundant in advanced HCC and associate with poor ICI responses; they act by consuming nutrients involved in arginase activity and releasing immunosuppressive factors (98–100). High densities of Tregs in HCC often exceed the number of CD8+ T cells and present another barrier; these FOXP3+CTLA-4+ Tregs not only dampen T-cell responses but also directly inhibit dendritic cells (18). While CTLA-4 inhibitors can deplete Tregs to some extent, an incomplete reduction of these suppressors might limit efficacy. Checkpoint therapy is essentially trying to tip the balance in favor of immunity within a milieu stacked against it. If the baseline immune suppression is overwhelming, monotherapy may fail. This reality has driven the exploration of multi-faceted combination treatments to target different components of the microenvironment.

The lack of robust predictive biomarkers for HCC immunotherapy is a practical challenge. Unlike melanoma or lung cancer, where high TMB or PD-L1 IHC can guide therapy to a degree, in HCC there is no established biomarker required for use of ICIs. The consequence is that some patients may be exposed to potential toxicity and cost without deriving benefit. Research is ongoing to identify biomarkers, circulating immune cells or cytokines, tumor gene signatures and gut microbiome profiles that could predict benefit or identify early on whether a patient is responding (101,102). For instance, peripheral blood analysis demonstrating expansion of certain T-cell clones or induction of IFN-γ response early in therapy might signal a response, whereas rising VEGF or IL-8 levels might herald resistance (103). Standardization and validation of such biomarkers is still needed.

As aforementioned, several trials have included patients with relatively preserved liver function (Child-Pugh A) (104). However, in the real world, a number of patients with HCC have Child-Pugh B cirrhosis or worse and it is uncertain how safe or effective ICIs are in that population. There is concern that patients with worse liver function might be less able to withstand immune-related hepatitis or even the mild inflammation that comes with activating the immune system. Indeed, immune-mediated liver injury (of any grade) occurs in 11.4% of patients with HCC on ICIs, a markedly higher incidence compared with that of other solid tumors (105). Thus, vigilant monitoring is imperative. Liver function tests, bilirubin level and clinical status, such as the presence of ascites or encephalopathy, should be checked at baseline and frequently during therapy. In HBV-associated HCC, ICIs can trigger hepatitis B reactivation; current guidelines recommend prophylactic antivirals during ICI treatment to prevent flares. Notably, the median survival time after an irAE was recorded as only ~3 months in this cirrhotic population, underscoring how serious irAEs can be if not promptly managed (106). Multidisciplinary care with hepatology input is advisable to optimize the liver status of the patient (for example, managing varices and encephalopathy) before and during immunotherapy. Additionally, these patients have been underrepresented in trials, so evidence is scant. Certain retrospective series and small studies (107,108) suggest anti-PD-1 can be used with caution in selecting patients with Child-Pugh B, but efficacy might be attenuated if the immune system is compromised by end-stage cirrhosis. This presents a challenge of how to extend immunotherapy to those with notable liver dysfunction or whether to prioritize liver-directed therapies in that setting.

When irAEs do occur, management should be tailored to severity, based on published research (109,110): i) Grade 1 (mild). Continue ICI with close monitoring. Symptomatic treatment if needed, but no immunosuppressive therapy is required (111,112). For example, mild transaminase elevations [alanine transaminase (ALT)/aspartate transaminase <3× upper limit of normal (ULN)] or low-grade rash can be observed while continuing therapy under careful surveillance. ii) Grade 2 (moderate). Hold ICI temporarily and initiate moderate-dose corticosteroids (for example, prednisone 0.5–1 mg/kg) (109). Resume immunotherapy only after symptoms regress to Grade 1 or better. For instance, Grade 2 immune hepatitis (ALT 3–5× ULN) would warrant withholding the drug and starting steroids, then re-challenging once LFTs normalize (109). Low-dose steroids may also be used for less severe endocrinopathies (with hormone replacement as needed). iii) Grade 3 (severe). Permanently discontinue the ICI and promptly start high-dose corticosteroids (1–2 mg/kg prednisone or methylprednisolone) (109). Taper steroids slowly (over 4–6 weeks) to prevent rebound. If no improvement within 48–72 h, escalate immunosuppression. For hepatitis, add mycophenolate mofetil instead of infliximab (109,113). TNFα inhibitors are generally avoided in liver toxicity due to infection risk. For colitis, infliximab can be used if steroids fail to control diarrhea (110). Life-threatening irAEs (for example, myocarditis and severe pneumonitis) may require intensive care and specialized interventions. iv) Grade 4 (life-threatening). Permanently discontinue immunotherapy and treat as an emergency. High-dose IV steroids (for example, methylprednisolone 1–2 mg/kg) are mandatory, often alongside additional immunosuppressants (114). Supportive care in the Intensive Care Unit is indicated for organ failure.

The etiological differences (HBV vs. HCV vs. NASH) not only impact response, as discussed, but also raise challenges in concurrent management. For instance, checkpoint blockade can trigger flares of hepatitis in patients with HBV if not vigilantly managed with antivirals (115). In HCV-related HCC, on the other hand, some studies noted that patients treated with antivirals demonstrated an improved response to immunotherapy, which suggests active viremia might be immunosuppressive (although data are not conclusive) (116,117). Autoimmune disorders are relative contraindications to ICIs, relevant since some patients with HCC have coexisting autoimmune hepatitis or primary biliary cholangitis; for those patients, current checkpoint therapies might not be feasible, which represents an unmet need for alternative immunotherapies with less autoimmune risk.

Checkpoint inhibitors and their combinations are expensive, which can limit access in low-resource settings where HCC burden is high (for example, Eastern Asia and sub-Saharan Africa with high HBV prevalence) (118). Furthermore, administering immunotherapy requires infrastructure for safe delivery and management of side effects, including multidisciplinary teams. The global challenge is to make these advances equitable and applicable where they are needed the most, perhaps through biomarker-driven use (to avoid treatment for those patients who are unlikely to benefit) or development of cost-effective biosimilars.

In summary, the current limitations in HCC immunotherapy revolve around inadequate efficacy for a sizable subset of patients (due to primary or acquired resistance mechanisms) and managing the immunosuppressive tumor milieu and patient condition. Overcoming these challenges will potentially require personalized combination approaches and improved biomarker-led patient selection. It also underscores why ongoing research into novel checkpoint targets and immunomodulatory strategies is key to break through the ceiling of response rates and tackle resistance. The next section discusses such emerging strategies and future directions, aiming to address these gaps.

Using two ICIs can broaden immune activation and prevent escape via alternative pathways. The phase III HIMALAYA trial demonstrated that adding a single priming dose of tremelimumab (anti-CTLA-4) to durvalumab (anti-PD-L1) improved OS compared with sorafenib, forming the STRIDE regimen (119). Similarly, nivolumab + ipilimumab achieved ~30% response in a phase II trial, which surpassed the rate for nivolumab alone but with greater toxicity (120). More recent targets, such as LAG-3, TIM-3 and TIGIT, are being co-blocked with PD-1 to reverse T-cell exhaustion (13,15,22,26).

Ablation, radiation and TACE may enhance tumor antigen release and immune visibility. A pilot study combining tremelimumab with ablation reported objective responses and CD8⁺ T-cell infiltration (121). Trials are exploring ICIs + SBRT or TACE to convert immune ‘cold’ tumors to ‘hot’ ones, occasionally inducing systemic abscopal effects (122).

VEGF fosters immune evasion; therefore, the IMbrave150 trial established atezolizumab + bevacizumab as a first-line HCC therapy by improving T-cell infiltration and survival (122). Additional approaches target MDSCs and Tregs using colony-stimulating factor-1 receptor (CSF1R) inhibitors, low-dose cyclophosphamide or CC chemokine receptor (CCR)4 antagonists, such as mogamulizumab (NCT02705105) (123). TGF-β blockade (for example, galunisertib) and IL-8/CXCR1/2 inhibition are also under study for improving ICI responses (124,125).

IDO depletes tryptophan and impairs T cells; IDO inhibitors, such as BMS-986205, combined with nivolumab have demonstrated early potential in HCC (69). Similarly, high adenosine suppresses T and NK cells in hypoxic tumors. Anti-CD73 antibodies (for example, oleclumab) and A2A receptor antagonists are being trialed to restore immune activity in adenosine-rich HCC environments (126).

Chimeric antigen receptor (CAR)-T cells targeting glypican-3 (GPC3) have induced tumor responses in a phase I trial, although their function is limited by liver TME barriers (127). For example, dense fibrotic stroma restricts CAR T infiltration, tumor-associated macrophages secrete IL-10 and TGF-β, and dampen CAR T function, myeloid-derived suppressor cells limit CAR-T proliferation by consuming nutrients and releasing suppressive cytokines, and Tregs inhibit effector CAR T activity. Engineered CARs capable of checkpoint blockade are in development. TIL therapy and TCR-engineered T cells targeting alpha-fetoprotein (AFP) or viral antigens are also in early clinical stages (128). These cell-based therapies could bypass resistance by delivering active effector cells, particularly when combined with ICIs.

Bispecific antibodies targeting GPC3/CD3 or GPC3/CD47 aim to recruit T cells and macrophages simultaneously (32). Oncolytic viruses [for example, pexastimogene devacirepvec (Pexa-Vec)] may inflame the tumor and synergize with ICIs (122). Personalized vaccines targeting tumor or viral antigens could prime immunity. Adaptive trials are exploring real-time therapy switches or multi-agent escalation based on early resistance signals (for example, VEGF rise) (61).

In conclusion, overcoming ICI resistance in HCC requires multifaceted approaches, such as modulating the TME, enhancing immune priming and adding cellular therapies. As combinations are refined, biomarker-driven personalization (for example, for Wnt-active or Treg-rich tumors) may guide treatment selection and improve durable response rates.

Due to the complex inhibitory networks in HCC, novel checkpoint targets beyond PD-1 and CTLA-4 are in clinical development. One of the most advanced checkpoint targets is LAG-3 inhibition. Relatlimab, an anti-LAG-3 antibody, demonstrated success in melanoma (in combination with nivolumab) (129) and is now being evaluated in HCC. A phase II trial of relatlimab + nivolumab in advanced HCC (after TKI therapy) has been conducted (130) and a larger phase III (RELATIVITY-073) is ongoing to evaluate this combination vs. nivolumab alone. By blocking LAG-3, the aim is to reinvigorate T cells that are not fully rescued by PD-1 blockade alone. Similarly, TIM-3 blockers (such as sabatolimab or cobolimab) are being combined with PD-1 inhibitors in early-phase trials in solid tumors. Preclinical evidence supports co-blockade of TIM-3 and PD-1 to prevent or overcome adaptive resistance (15,28). If safety is manageable, these could move into HCC-specific studies.

TIGIT inhibitors (for example, tiragolumab and vibostolimab) have garnered interest after demonstrating efficacy in other cancer types, such as lung cancer. While no phase III data exists in HCC yet, TIGIT blockade is expected to enhance NK and T-cell activity; trials in GI cancer types, including HCC, are anticipated. Blocking TIGIT can notably counteract the NK cell exhaustion observed in HBV-related HCC (27), which provides a rationale for combining anti-TIGIT with PD-1 or even with NK cell therapies. VISTA targeting is another novel approach; for instance, a humanized anti-VISTA antibody (CI-8993) is undergoing investigation in a phase I trial for advanced solid tumors (12). If VISTA contributes to resistance (particularly post anti-PD-1 therapy upregulation (12), anti-VISTA could be administered as an add-on in refractory cases. However, caution is necessary in HCC due to the potential protective associations between VISTA activity and tumor cells (12); patient selection or combination with other ICIs may be key to combat resistance.

Future strategies are targeting not just T cells but also immunosuppressive stromal components. For example, anti-CD47 agents (such as magrolimab) aim to release macrophages against HCC. Magrolimab is undergoing investigation in trials for various malignancies and could be combined with ICIs to produce a coordinated innate and adaptive immune attack (131). There is also a first-in-class bispecific antibody that targets GPC3 (an HCC oncofetal antigen) and CD47 simultaneously, designed to direct macrophages specifically to eat HCC cells (32), such bispecifics could be useful in liver tumors expressing GPC3 and CD47.

Some trials are evaluating low-dose cyclophosphamide or CSF1R inhibitors to deplete Tregs or macrophages in combination with ICIs. Others are investigating drugs such as CCR4 inhibitors to block Treg recruitment. The indoleamine-2,3-dioxygenase 1 (IDO1) pathway is another immunosuppressive mechanism where the enzyme IDO1 depletes tryptophan and produces kynurenine, which paralyzes T cells.

Although use of an IDO inhibitor (epacadostat) failed in melanoma, HCC studies continue; for instance, a phase I/II trial of IDO inhibitor BMS-986205 + nivolumab in first-line HCC is underway (69), examining if IDO blockade can enhance anti-PD-1 efficacy in liver cancer. Similarly, the adenosine pathway (CD39/CD73 and A2A adenosine receptor) is being targeted. High adenosine levels in the HCC microenvironment inhibit T cells and NK cells (70). Antibodies, such as oleclumab (anti-CD73), and small-molecule adenosine receptor antagonists are in trials in other cancer types and could be applied to HCC. These could be particularly relevant for HCCs where hypoxia and adenosine production (via CD73 on tumor or stromal cells) contribute to immunosuppression (70).

Another future direction is adoptive cell therapy that can bypass some of the mechanisms of checkpoint resistance. HCC-specific CAR-T cells are in early development; for instance, CAR-T cells targeting (GPC3 (an antigen on HCC) have demonstrated evidence of tumor shrinkage in a phase I trial (132), although efficacy was limited by trafficking and the immunosuppressive microenvironment. Currently, CAR-T cells are being combined with anti-PD-1 blockade (133) or engineered to secrete checkpoint inhibitors such as PD-1-blocking single-chain variable fragments (134). CAR-T cells targeting B7-H3 are also being investigated in pediatric liver tumors and could be extended to HCC (a phase I trial of B7-H3 CAR T-cells in HCC is registered) (10).

Another approach involves TIL therapy, which expands the T cells of the patient extracted from their tumor and reinfuses them after depleting suppressive cells. Checkpoint inhibitors can be combined with TILs to maintain their activity in vivo. TCR-engineered T cells targeting HCC-associated antigens, such as AFP, or viral antigens in HBV-related HCC are also being investigated. These cell therapies may work optimally when combined with checkpoint inhibitors; for example, using CAR-T cells to deliver targeted tumor killing while PD-1 antibodies maintain their sustained activity.

Therapeutic cancer vaccines for HCC (peptide or dendritic cell vaccines targeting antigens such as GPC3, AFP or neoantigens) are being investigated in trials and, although they are not very effective alone, cancer vaccines for HCC could be useful to prime T-cell responses that are then sustained by checkpoint inhibition (135–137). Similarly, oncolytic viruses designed to selectively infect and lyse HCC cells can release tumor antigens and stimulate inflammation in the tumor, potentially synergizing with ICIs. A current example is an oncolytic vaccinia virus (Pexa-Vec), which was tested with cytokine therapy and could be combined with PD-1 blockade in future studies (138).

Future clinical trials are increasingly incorporating biomarker endpoints to personalize therapy. For instance, trials may stratify patients by an immune signature; those with high Treg and macrophage genes might be routed to a combination that includes a macrophage-reprogramming agent, while those with high exhaustion markers may receive PD-1 + LAG-3 and those with β-catenin mutations (immune cold) may need a priming intervention first (74,98,139). Real-time monitoring of blood for circulating tumor DNA and T-cell receptor clonality may guide whether to continue an ICI or add a second agent. Furthermore, as more targets, such as LAG-3 and TIGIT, receive approval in other cancer types, it will become feasible to test multi-checkpoint blockade in HCC in a biomarker-selected manner (for example, adding anti-TIGIT if an increase in TIGIT⁺ cells is observed on a fresh biopsy at progression) (140).

The insight that NASH-associated HCC might not respond well to current ICIs suggests a need for tailored approaches. For NASH-HCC, strategies to modulate the metabolic/inflammatory milieu of the liver (such as farnesoid X receptor agonists, anti-IL-17 or CCR2/CCR5 inhibitors to reduce steatohepatitis) could potentially be combined with ICIs to improve efficacy. For viral HCC, combining ICIs with therapeutic antivirals or even therapeutic vaccines for HBV/HCV might further enhance immune clearance of both virus and tumor, and is an area of ongoing research (141,142).

Efforts are also underway to design ICIs with fewer side effects. Examples include localized delivery (such as drug-eluting beads with immunotherapy delivered to the liver tumor), bispecific checkpoint antibodies that preferentially activate T cells only when bound to tumor cells and engineered cytokines (IL-2/IL-15 variants) that boost antitumor immunity without broad toxicity (143–145).

Future trials may test sequencing strategies; for instance, providing a short induction of anti-CTLA-4 to deplete Tregs, followed by long-term anti-PD-1 vs. concurrent administration. The optimal timing with respect to other treatments is also under study. Using ICIs in the perioperative setting to reduce recurrence risk or as a bridge to transplant (with careful patient selection due to rejection risks) is being examined. There is also interest in retreatment; for example, whether the same or a different ICI could be reintroduced for a patient who responded to immunotherapy, relapsed and subsequently demonstrated progression. A case report suggested that some previously responsive patients can regain disease control with a second course of ICIs, particularly if combined with a novel agent to overcome the acquired resistance mechanism (146).

Future directions for checkpoint immunotherapy in HCC focus on expanding global access through trials in resource-limited settings and developing cost-effective strategies, including shorter treatment durations and early identification of non-responders. Long-term follow-up studies are necessary to determine if immunotherapy can be safely discontinued in complete responders, as suggested by melanoma data (87,129), although the underlying cirrhosis in HCC may harbor microscopic tumors. The field is evolving toward multitargeted immunomodulation using combination therapies tailored by biomarker profiles and administered at optimal disease stages, including earlier intervention for tumor downstaging. These advances aim to convert immunotherapy-resistant HCC cases into responsive ones, potentially achieving long-term remissions or cures. Success depends on the selection of appropriate targets and combinations for individual patients at the right time, with continued clinical trials and translational research, which are key to transform HCC from a historically difficult-to-treat disease into one that can be effectively managed or potentially cured.